aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
authorLinus Torvalds <torvalds@linux-foundation.org>2009-04-03 13:07:43 -0400
committerLinus Torvalds <torvalds@linux-foundation.org>2009-04-03 13:07:43 -0400
commit3cc50ac0dbda5100684e570247782330155d35e0 (patch)
treef4b8f22d1725ebe65d2fe658d292dabacd7ed564 /fs
parentd9b9be024a6628a01d8730d1fd0b5f25658a2794 (diff)
parentb797cac7487dee6bfddeb161631c1bbc54fa3cdb (diff)
Merge git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-2.6-fscache
* git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-2.6-fscache: (41 commits) NFS: Add mount options to enable local caching on NFS NFS: Display local caching state NFS: Store pages from an NFS inode into a local cache NFS: Read pages from FS-Cache into an NFS inode NFS: nfs_readpage_async() needs to be accessible as a fallback for local caching NFS: Add read context retention for FS-Cache to call back with NFS: FS-Cache page management NFS: Add some new I/O counters for FS-Cache doing things for NFS NFS: Invalidate FsCache page flags when cache removed NFS: Use local disk inode cache NFS: Define and create inode-level cache objects NFS: Define and create superblock-level objects NFS: Define and create server-level objects NFS: Register NFS for caching and retrieve the top-level index NFS: Permit local filesystem caching to be enabled for NFS NFS: Add FS-Cache option bit and debug bit NFS: Add comment banners to some NFS functions FS-Cache: Make kAFS use FS-Cache CacheFiles: A cache that backs onto a mounted filesystem CacheFiles: Export things for CacheFiles ...
Diffstat (limited to 'fs')
-rw-r--r--fs/Kconfig7
-rw-r--r--fs/Makefile2
-rw-r--r--fs/afs/Kconfig8
-rw-r--r--fs/afs/Makefile3
-rw-r--r--fs/afs/cache.c503
-rw-r--r--fs/afs/cache.h15
-rw-r--r--fs/afs/cell.c16
-rw-r--r--fs/afs/file.c220
-rw-r--r--fs/afs/inode.c31
-rw-r--r--fs/afs/internal.h53
-rw-r--r--fs/afs/main.c27
-rw-r--r--fs/afs/mntpt.c4
-rw-r--r--fs/afs/vlocation.c25
-rw-r--r--fs/afs/volume.c14
-rw-r--r--fs/afs/write.c21
-rw-r--r--fs/cachefiles/Kconfig39
-rw-r--r--fs/cachefiles/Makefile18
-rw-r--r--fs/cachefiles/bind.c286
-rw-r--r--fs/cachefiles/daemon.c755
-rw-r--r--fs/cachefiles/interface.c449
-rw-r--r--fs/cachefiles/internal.h360
-rw-r--r--fs/cachefiles/key.c159
-rw-r--r--fs/cachefiles/main.c106
-rw-r--r--fs/cachefiles/namei.c771
-rw-r--r--fs/cachefiles/proc.c134
-rw-r--r--fs/cachefiles/rdwr.c879
-rw-r--r--fs/cachefiles/security.c116
-rw-r--r--fs/cachefiles/xattr.c291
-rw-r--r--fs/fscache/Kconfig56
-rw-r--r--fs/fscache/Makefile19
-rw-r--r--fs/fscache/cache.c415
-rw-r--r--fs/fscache/cookie.c500
-rw-r--r--fs/fscache/fsdef.c144
-rw-r--r--fs/fscache/histogram.c109
-rw-r--r--fs/fscache/internal.h380
-rw-r--r--fs/fscache/main.c124
-rw-r--r--fs/fscache/netfs.c103
-rw-r--r--fs/fscache/object.c810
-rw-r--r--fs/fscache/operation.c459
-rw-r--r--fs/fscache/page.c816
-rw-r--r--fs/fscache/proc.c68
-rw-r--r--fs/fscache/stats.c212
-rw-r--r--fs/nfs/Kconfig8
-rw-r--r--fs/nfs/Makefile1
-rw-r--r--fs/nfs/client.c14
-rw-r--r--fs/nfs/file.c38
-rw-r--r--fs/nfs/fscache-index.c337
-rw-r--r--fs/nfs/fscache.c523
-rw-r--r--fs/nfs/fscache.h220
-rw-r--r--fs/nfs/inode.c14
-rw-r--r--fs/nfs/internal.h4
-rw-r--r--fs/nfs/iostat.h18
-rw-r--r--fs/nfs/read.c27
-rw-r--r--fs/nfs/super.c45
-rw-r--r--fs/splice.c3
-rw-r--r--fs/super.c1
56 files changed, 10413 insertions, 367 deletions
diff --git a/fs/Kconfig b/fs/Kconfig
index ae3b34a2ea69..86b203fc3c56 100644
--- a/fs/Kconfig
+++ b/fs/Kconfig
@@ -66,6 +66,13 @@ config GENERIC_ACL
66 bool 66 bool
67 select FS_POSIX_ACL 67 select FS_POSIX_ACL
68 68
69menu "Caches"
70
71source "fs/fscache/Kconfig"
72source "fs/cachefiles/Kconfig"
73
74endmenu
75
69if BLOCK 76if BLOCK
70menu "CD-ROM/DVD Filesystems" 77menu "CD-ROM/DVD Filesystems"
71 78
diff --git a/fs/Makefile b/fs/Makefile
index 15f73014a208..70b2aed87133 100644
--- a/fs/Makefile
+++ b/fs/Makefile
@@ -63,6 +63,7 @@ obj-$(CONFIG_PROFILING) += dcookies.o
63obj-$(CONFIG_DLM) += dlm/ 63obj-$(CONFIG_DLM) += dlm/
64 64
65# Do not add any filesystems before this line 65# Do not add any filesystems before this line
66obj-$(CONFIG_FSCACHE) += fscache/
66obj-$(CONFIG_REISERFS_FS) += reiserfs/ 67obj-$(CONFIG_REISERFS_FS) += reiserfs/
67obj-$(CONFIG_EXT3_FS) += ext3/ # Before ext2 so root fs can be ext3 68obj-$(CONFIG_EXT3_FS) += ext3/ # Before ext2 so root fs can be ext3
68obj-$(CONFIG_EXT2_FS) += ext2/ 69obj-$(CONFIG_EXT2_FS) += ext2/
@@ -116,6 +117,7 @@ obj-$(CONFIG_AFS_FS) += afs/
116obj-$(CONFIG_BEFS_FS) += befs/ 117obj-$(CONFIG_BEFS_FS) += befs/
117obj-$(CONFIG_HOSTFS) += hostfs/ 118obj-$(CONFIG_HOSTFS) += hostfs/
118obj-$(CONFIG_HPPFS) += hppfs/ 119obj-$(CONFIG_HPPFS) += hppfs/
120obj-$(CONFIG_CACHEFILES) += cachefiles/
119obj-$(CONFIG_DEBUG_FS) += debugfs/ 121obj-$(CONFIG_DEBUG_FS) += debugfs/
120obj-$(CONFIG_OCFS2_FS) += ocfs2/ 122obj-$(CONFIG_OCFS2_FS) += ocfs2/
121obj-$(CONFIG_BTRFS_FS) += btrfs/ 123obj-$(CONFIG_BTRFS_FS) += btrfs/
diff --git a/fs/afs/Kconfig b/fs/afs/Kconfig
index e7b522fe15e1..5c4e61d3c772 100644
--- a/fs/afs/Kconfig
+++ b/fs/afs/Kconfig
@@ -19,3 +19,11 @@ config AFS_DEBUG
19 See <file:Documentation/filesystems/afs.txt> for more information. 19 See <file:Documentation/filesystems/afs.txt> for more information.
20 20
21 If unsure, say N. 21 If unsure, say N.
22
23config AFS_FSCACHE
24 bool "Provide AFS client caching support (EXPERIMENTAL)"
25 depends on EXPERIMENTAL
26 depends on AFS_FS=m && FSCACHE || AFS_FS=y && FSCACHE=y
27 help
28 Say Y here if you want AFS data to be cached locally on disk through
29 the generic filesystem cache manager
diff --git a/fs/afs/Makefile b/fs/afs/Makefile
index a66671082cfb..4f64b95d57bd 100644
--- a/fs/afs/Makefile
+++ b/fs/afs/Makefile
@@ -2,7 +2,10 @@
2# Makefile for Red Hat Linux AFS client. 2# Makefile for Red Hat Linux AFS client.
3# 3#
4 4
5afs-cache-$(CONFIG_AFS_FSCACHE) := cache.o
6
5kafs-objs := \ 7kafs-objs := \
8 $(afs-cache-y) \
6 callback.o \ 9 callback.o \
7 cell.o \ 10 cell.o \
8 cmservice.o \ 11 cmservice.o \
diff --git a/fs/afs/cache.c b/fs/afs/cache.c
index de0d7de69edc..e2b1d3f16519 100644
--- a/fs/afs/cache.c
+++ b/fs/afs/cache.c
@@ -1,6 +1,6 @@
1/* AFS caching stuff 1/* AFS caching stuff
2 * 2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved. 3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com) 4 * Written by David Howells (dhowells@redhat.com)
5 * 5 *
6 * This program is free software; you can redistribute it and/or 6 * This program is free software; you can redistribute it and/or
@@ -9,248 +9,395 @@
9 * 2 of the License, or (at your option) any later version. 9 * 2 of the License, or (at your option) any later version.
10 */ 10 */
11 11
12#ifdef AFS_CACHING_SUPPORT 12#include <linux/slab.h>
13static cachefs_match_val_t afs_cell_cache_match(void *target, 13#include <linux/sched.h>
14 const void *entry); 14#include "internal.h"
15static void afs_cell_cache_update(void *source, void *entry); 15
16 16static uint16_t afs_cell_cache_get_key(const void *cookie_netfs_data,
17struct cachefs_index_def afs_cache_cell_index_def = { 17 void *buffer, uint16_t buflen);
18 .name = "cell_ix", 18static uint16_t afs_cell_cache_get_aux(const void *cookie_netfs_data,
19 .data_size = sizeof(struct afs_cache_cell), 19 void *buffer, uint16_t buflen);
20 .keys[0] = { CACHEFS_INDEX_KEYS_ASCIIZ, 64 }, 20static enum fscache_checkaux afs_cell_cache_check_aux(void *cookie_netfs_data,
21 .match = afs_cell_cache_match, 21 const void *buffer,
22 .update = afs_cell_cache_update, 22 uint16_t buflen);
23
24static uint16_t afs_vlocation_cache_get_key(const void *cookie_netfs_data,
25 void *buffer, uint16_t buflen);
26static uint16_t afs_vlocation_cache_get_aux(const void *cookie_netfs_data,
27 void *buffer, uint16_t buflen);
28static enum fscache_checkaux afs_vlocation_cache_check_aux(
29 void *cookie_netfs_data, const void *buffer, uint16_t buflen);
30
31static uint16_t afs_volume_cache_get_key(const void *cookie_netfs_data,
32 void *buffer, uint16_t buflen);
33
34static uint16_t afs_vnode_cache_get_key(const void *cookie_netfs_data,
35 void *buffer, uint16_t buflen);
36static void afs_vnode_cache_get_attr(const void *cookie_netfs_data,
37 uint64_t *size);
38static uint16_t afs_vnode_cache_get_aux(const void *cookie_netfs_data,
39 void *buffer, uint16_t buflen);
40static enum fscache_checkaux afs_vnode_cache_check_aux(void *cookie_netfs_data,
41 const void *buffer,
42 uint16_t buflen);
43static void afs_vnode_cache_now_uncached(void *cookie_netfs_data);
44
45struct fscache_netfs afs_cache_netfs = {
46 .name = "afs",
47 .version = 0,
48};
49
50struct fscache_cookie_def afs_cell_cache_index_def = {
51 .name = "AFS.cell",
52 .type = FSCACHE_COOKIE_TYPE_INDEX,
53 .get_key = afs_cell_cache_get_key,
54 .get_aux = afs_cell_cache_get_aux,
55 .check_aux = afs_cell_cache_check_aux,
56};
57
58struct fscache_cookie_def afs_vlocation_cache_index_def = {
59 .name = "AFS.vldb",
60 .type = FSCACHE_COOKIE_TYPE_INDEX,
61 .get_key = afs_vlocation_cache_get_key,
62 .get_aux = afs_vlocation_cache_get_aux,
63 .check_aux = afs_vlocation_cache_check_aux,
64};
65
66struct fscache_cookie_def afs_volume_cache_index_def = {
67 .name = "AFS.volume",
68 .type = FSCACHE_COOKIE_TYPE_INDEX,
69 .get_key = afs_volume_cache_get_key,
70};
71
72struct fscache_cookie_def afs_vnode_cache_index_def = {
73 .name = "AFS.vnode",
74 .type = FSCACHE_COOKIE_TYPE_DATAFILE,
75 .get_key = afs_vnode_cache_get_key,
76 .get_attr = afs_vnode_cache_get_attr,
77 .get_aux = afs_vnode_cache_get_aux,
78 .check_aux = afs_vnode_cache_check_aux,
79 .now_uncached = afs_vnode_cache_now_uncached,
23}; 80};
24#endif
25 81
26/* 82/*
27 * match a cell record obtained from the cache 83 * set the key for the index entry
28 */ 84 */
29#ifdef AFS_CACHING_SUPPORT 85static uint16_t afs_cell_cache_get_key(const void *cookie_netfs_data,
30static cachefs_match_val_t afs_cell_cache_match(void *target, 86 void *buffer, uint16_t bufmax)
31 const void *entry)
32{ 87{
33 const struct afs_cache_cell *ccell = entry; 88 const struct afs_cell *cell = cookie_netfs_data;
34 struct afs_cell *cell = target; 89 uint16_t klen;
35 90
36 _enter("{%s},{%s}", ccell->name, cell->name); 91 _enter("%p,%p,%u", cell, buffer, bufmax);
37 92
38 if (strncmp(ccell->name, cell->name, sizeof(ccell->name)) == 0) { 93 klen = strlen(cell->name);
39 _leave(" = SUCCESS"); 94 if (klen > bufmax)
40 return CACHEFS_MATCH_SUCCESS; 95 return 0;
41 }
42 96
43 _leave(" = FAILED"); 97 memcpy(buffer, cell->name, klen);
44 return CACHEFS_MATCH_FAILED; 98 return klen;
45} 99}
46#endif
47 100
48/* 101/*
49 * update a cell record in the cache 102 * provide new auxilliary cache data
50 */ 103 */
51#ifdef AFS_CACHING_SUPPORT 104static uint16_t afs_cell_cache_get_aux(const void *cookie_netfs_data,
52static void afs_cell_cache_update(void *source, void *entry) 105 void *buffer, uint16_t bufmax)
53{ 106{
54 struct afs_cache_cell *ccell = entry; 107 const struct afs_cell *cell = cookie_netfs_data;
55 struct afs_cell *cell = source; 108 uint16_t dlen;
56 109
57 _enter("%p,%p", source, entry); 110 _enter("%p,%p,%u", cell, buffer, bufmax);
58 111
59 strncpy(ccell->name, cell->name, sizeof(ccell->name)); 112 dlen = cell->vl_naddrs * sizeof(cell->vl_addrs[0]);
113 dlen = min(dlen, bufmax);
114 dlen &= ~(sizeof(cell->vl_addrs[0]) - 1);
60 115
61 memcpy(ccell->vl_servers, 116 memcpy(buffer, cell->vl_addrs, dlen);
62 cell->vl_addrs, 117 return dlen;
63 min(sizeof(ccell->vl_servers), sizeof(cell->vl_addrs))); 118}
64 119
120/*
121 * check that the auxilliary data indicates that the entry is still valid
122 */
123static enum fscache_checkaux afs_cell_cache_check_aux(void *cookie_netfs_data,
124 const void *buffer,
125 uint16_t buflen)
126{
127 _leave(" = OKAY");
128 return FSCACHE_CHECKAUX_OKAY;
65} 129}
66#endif
67
68#ifdef AFS_CACHING_SUPPORT
69static cachefs_match_val_t afs_vlocation_cache_match(void *target,
70 const void *entry);
71static void afs_vlocation_cache_update(void *source, void *entry);
72
73struct cachefs_index_def afs_vlocation_cache_index_def = {
74 .name = "vldb",
75 .data_size = sizeof(struct afs_cache_vlocation),
76 .keys[0] = { CACHEFS_INDEX_KEYS_ASCIIZ, 64 },
77 .match = afs_vlocation_cache_match,
78 .update = afs_vlocation_cache_update,
79};
80#endif
81 130
131/*****************************************************************************/
82/* 132/*
83 * match a VLDB record stored in the cache 133 * set the key for the index entry
84 * - may also load target from entry
85 */ 134 */
86#ifdef AFS_CACHING_SUPPORT 135static uint16_t afs_vlocation_cache_get_key(const void *cookie_netfs_data,
87static cachefs_match_val_t afs_vlocation_cache_match(void *target, 136 void *buffer, uint16_t bufmax)
88 const void *entry)
89{ 137{
90 const struct afs_cache_vlocation *vldb = entry; 138 const struct afs_vlocation *vlocation = cookie_netfs_data;
91 struct afs_vlocation *vlocation = target; 139 uint16_t klen;
140
141 _enter("{%s},%p,%u", vlocation->vldb.name, buffer, bufmax);
142
143 klen = strnlen(vlocation->vldb.name, sizeof(vlocation->vldb.name));
144 if (klen > bufmax)
145 return 0;
92 146
93 _enter("{%s},{%s}", vlocation->vldb.name, vldb->name); 147 memcpy(buffer, vlocation->vldb.name, klen);
94 148
95 if (strncmp(vlocation->vldb.name, vldb->name, sizeof(vldb->name)) == 0 149 _leave(" = %u", klen);
96 ) { 150 return klen;
97 if (!vlocation->valid || 151}
98 vlocation->vldb.rtime == vldb->rtime 152
153/*
154 * provide new auxilliary cache data
155 */
156static uint16_t afs_vlocation_cache_get_aux(const void *cookie_netfs_data,
157 void *buffer, uint16_t bufmax)
158{
159 const struct afs_vlocation *vlocation = cookie_netfs_data;
160 uint16_t dlen;
161
162 _enter("{%s},%p,%u", vlocation->vldb.name, buffer, bufmax);
163
164 dlen = sizeof(struct afs_cache_vlocation);
165 dlen -= offsetof(struct afs_cache_vlocation, nservers);
166 if (dlen > bufmax)
167 return 0;
168
169 memcpy(buffer, (uint8_t *)&vlocation->vldb.nservers, dlen);
170
171 _leave(" = %u", dlen);
172 return dlen;
173}
174
175/*
176 * check that the auxilliary data indicates that the entry is still valid
177 */
178static
179enum fscache_checkaux afs_vlocation_cache_check_aux(void *cookie_netfs_data,
180 const void *buffer,
181 uint16_t buflen)
182{
183 const struct afs_cache_vlocation *cvldb;
184 struct afs_vlocation *vlocation = cookie_netfs_data;
185 uint16_t dlen;
186
187 _enter("{%s},%p,%u", vlocation->vldb.name, buffer, buflen);
188
189 /* check the size of the data is what we're expecting */
190 dlen = sizeof(struct afs_cache_vlocation);
191 dlen -= offsetof(struct afs_cache_vlocation, nservers);
192 if (dlen != buflen)
193 return FSCACHE_CHECKAUX_OBSOLETE;
194
195 cvldb = container_of(buffer, struct afs_cache_vlocation, nservers);
196
197 /* if what's on disk is more valid than what's in memory, then use the
198 * VL record from the cache */
199 if (!vlocation->valid || vlocation->vldb.rtime == cvldb->rtime) {
200 memcpy((uint8_t *)&vlocation->vldb.nservers, buffer, dlen);
201 vlocation->valid = 1;
202 _leave(" = SUCCESS [c->m]");
203 return FSCACHE_CHECKAUX_OKAY;
204 }
205
206 /* need to update the cache if the cached info differs */
207 if (memcmp(&vlocation->vldb, buffer, dlen) != 0) {
208 /* delete if the volume IDs for this name differ */
209 if (memcmp(&vlocation->vldb.vid, &cvldb->vid,
210 sizeof(cvldb->vid)) != 0
99 ) { 211 ) {
100 vlocation->vldb = *vldb; 212 _leave(" = OBSOLETE");
101 vlocation->valid = 1; 213 return FSCACHE_CHECKAUX_OBSOLETE;
102 _leave(" = SUCCESS [c->m]");
103 return CACHEFS_MATCH_SUCCESS;
104 } else if (memcmp(&vlocation->vldb, vldb, sizeof(*vldb)) != 0) {
105 /* delete if VIDs for this name differ */
106 if (memcmp(&vlocation->vldb.vid,
107 &vldb->vid,
108 sizeof(vldb->vid)) != 0) {
109 _leave(" = DELETE");
110 return CACHEFS_MATCH_SUCCESS_DELETE;
111 }
112
113 _leave(" = UPDATE");
114 return CACHEFS_MATCH_SUCCESS_UPDATE;
115 } else {
116 _leave(" = SUCCESS");
117 return CACHEFS_MATCH_SUCCESS;
118 } 214 }
215
216 _leave(" = UPDATE");
217 return FSCACHE_CHECKAUX_NEEDS_UPDATE;
119 } 218 }
120 219
121 _leave(" = FAILED"); 220 _leave(" = OKAY");
122 return CACHEFS_MATCH_FAILED; 221 return FSCACHE_CHECKAUX_OKAY;
123} 222}
124#endif
125 223
224/*****************************************************************************/
126/* 225/*
127 * update a VLDB record stored in the cache 226 * set the key for the volume index entry
128 */ 227 */
129#ifdef AFS_CACHING_SUPPORT 228static uint16_t afs_volume_cache_get_key(const void *cookie_netfs_data,
130static void afs_vlocation_cache_update(void *source, void *entry) 229 void *buffer, uint16_t bufmax)
131{ 230{
132 struct afs_cache_vlocation *vldb = entry; 231 const struct afs_volume *volume = cookie_netfs_data;
133 struct afs_vlocation *vlocation = source; 232 uint16_t klen;
233
234 _enter("{%u},%p,%u", volume->type, buffer, bufmax);
235
236 klen = sizeof(volume->type);
237 if (klen > bufmax)
238 return 0;
134 239
135 _enter(""); 240 memcpy(buffer, &volume->type, sizeof(volume->type));
241
242 _leave(" = %u", klen);
243 return klen;
136 244
137 *vldb = vlocation->vldb;
138} 245}
139#endif
140
141#ifdef AFS_CACHING_SUPPORT
142static cachefs_match_val_t afs_volume_cache_match(void *target,
143 const void *entry);
144static void afs_volume_cache_update(void *source, void *entry);
145
146struct cachefs_index_def afs_volume_cache_index_def = {
147 .name = "volume",
148 .data_size = sizeof(struct afs_cache_vhash),
149 .keys[0] = { CACHEFS_INDEX_KEYS_BIN, 1 },
150 .keys[1] = { CACHEFS_INDEX_KEYS_BIN, 1 },
151 .match = afs_volume_cache_match,
152 .update = afs_volume_cache_update,
153};
154#endif
155 246
247/*****************************************************************************/
156/* 248/*
157 * match a volume hash record stored in the cache 249 * set the key for the index entry
158 */ 250 */
159#ifdef AFS_CACHING_SUPPORT 251static uint16_t afs_vnode_cache_get_key(const void *cookie_netfs_data,
160static cachefs_match_val_t afs_volume_cache_match(void *target, 252 void *buffer, uint16_t bufmax)
161 const void *entry)
162{ 253{
163 const struct afs_cache_vhash *vhash = entry; 254 const struct afs_vnode *vnode = cookie_netfs_data;
164 struct afs_volume *volume = target; 255 uint16_t klen;
165 256
166 _enter("{%u},{%u}", volume->type, vhash->vtype); 257 _enter("{%x,%x,%llx},%p,%u",
258 vnode->fid.vnode, vnode->fid.unique, vnode->status.data_version,
259 buffer, bufmax);
167 260
168 if (volume->type == vhash->vtype) { 261 klen = sizeof(vnode->fid.vnode);
169 _leave(" = SUCCESS"); 262 if (klen > bufmax)
170 return CACHEFS_MATCH_SUCCESS; 263 return 0;
171 } 264
265 memcpy(buffer, &vnode->fid.vnode, sizeof(vnode->fid.vnode));
172 266
173 _leave(" = FAILED"); 267 _leave(" = %u", klen);
174 return CACHEFS_MATCH_FAILED; 268 return klen;
175} 269}
176#endif
177 270
178/* 271/*
179 * update a volume hash record stored in the cache 272 * provide updated file attributes
180 */ 273 */
181#ifdef AFS_CACHING_SUPPORT 274static void afs_vnode_cache_get_attr(const void *cookie_netfs_data,
182static void afs_volume_cache_update(void *source, void *entry) 275 uint64_t *size)
183{ 276{
184 struct afs_cache_vhash *vhash = entry; 277 const struct afs_vnode *vnode = cookie_netfs_data;
185 struct afs_volume *volume = source;
186 278
187 _enter(""); 279 _enter("{%x,%x,%llx},",
280 vnode->fid.vnode, vnode->fid.unique,
281 vnode->status.data_version);
188 282
189 vhash->vtype = volume->type; 283 *size = vnode->status.size;
190} 284}
191#endif
192
193#ifdef AFS_CACHING_SUPPORT
194static cachefs_match_val_t afs_vnode_cache_match(void *target,
195 const void *entry);
196static void afs_vnode_cache_update(void *source, void *entry);
197
198struct cachefs_index_def afs_vnode_cache_index_def = {
199 .name = "vnode",
200 .data_size = sizeof(struct afs_cache_vnode),
201 .keys[0] = { CACHEFS_INDEX_KEYS_BIN, 4 },
202 .match = afs_vnode_cache_match,
203 .update = afs_vnode_cache_update,
204};
205#endif
206 285
207/* 286/*
208 * match a vnode record stored in the cache 287 * provide new auxilliary cache data
288 */
289static uint16_t afs_vnode_cache_get_aux(const void *cookie_netfs_data,
290 void *buffer, uint16_t bufmax)
291{
292 const struct afs_vnode *vnode = cookie_netfs_data;
293 uint16_t dlen;
294
295 _enter("{%x,%x,%Lx},%p,%u",
296 vnode->fid.vnode, vnode->fid.unique, vnode->status.data_version,
297 buffer, bufmax);
298
299 dlen = sizeof(vnode->fid.unique) + sizeof(vnode->status.data_version);
300 if (dlen > bufmax)
301 return 0;
302
303 memcpy(buffer, &vnode->fid.unique, sizeof(vnode->fid.unique));
304 buffer += sizeof(vnode->fid.unique);
305 memcpy(buffer, &vnode->status.data_version,
306 sizeof(vnode->status.data_version));
307
308 _leave(" = %u", dlen);
309 return dlen;
310}
311
312/*
313 * check that the auxilliary data indicates that the entry is still valid
209 */ 314 */
210#ifdef AFS_CACHING_SUPPORT 315static enum fscache_checkaux afs_vnode_cache_check_aux(void *cookie_netfs_data,
211static cachefs_match_val_t afs_vnode_cache_match(void *target, 316 const void *buffer,
212 const void *entry) 317 uint16_t buflen)
213{ 318{
214 const struct afs_cache_vnode *cvnode = entry; 319 struct afs_vnode *vnode = cookie_netfs_data;
215 struct afs_vnode *vnode = target; 320 uint16_t dlen;
216 321
217 _enter("{%x,%x,%Lx},{%x,%x,%Lx}", 322 _enter("{%x,%x,%llx},%p,%u",
218 vnode->fid.vnode, 323 vnode->fid.vnode, vnode->fid.unique, vnode->status.data_version,
219 vnode->fid.unique, 324 buffer, buflen);
220 vnode->status.version, 325
221 cvnode->vnode_id, 326 /* check the size of the data is what we're expecting */
222 cvnode->vnode_unique, 327 dlen = sizeof(vnode->fid.unique) + sizeof(vnode->status.data_version);
223 cvnode->data_version); 328 if (dlen != buflen) {
224 329 _leave(" = OBSOLETE [len %hx != %hx]", dlen, buflen);
225 if (vnode->fid.vnode != cvnode->vnode_id) { 330 return FSCACHE_CHECKAUX_OBSOLETE;
226 _leave(" = FAILED");
227 return CACHEFS_MATCH_FAILED;
228 } 331 }
229 332
230 if (vnode->fid.unique != cvnode->vnode_unique || 333 if (memcmp(buffer,
231 vnode->status.version != cvnode->data_version) { 334 &vnode->fid.unique,
232 _leave(" = DELETE"); 335 sizeof(vnode->fid.unique)
233 return CACHEFS_MATCH_SUCCESS_DELETE; 336 ) != 0) {
337 unsigned unique;
338
339 memcpy(&unique, buffer, sizeof(unique));
340
341 _leave(" = OBSOLETE [uniq %x != %x]",
342 unique, vnode->fid.unique);
343 return FSCACHE_CHECKAUX_OBSOLETE;
344 }
345
346 if (memcmp(buffer + sizeof(vnode->fid.unique),
347 &vnode->status.data_version,
348 sizeof(vnode->status.data_version)
349 ) != 0) {
350 afs_dataversion_t version;
351
352 memcpy(&version, buffer + sizeof(vnode->fid.unique),
353 sizeof(version));
354
355 _leave(" = OBSOLETE [vers %llx != %llx]",
356 version, vnode->status.data_version);
357 return FSCACHE_CHECKAUX_OBSOLETE;
234 } 358 }
235 359
236 _leave(" = SUCCESS"); 360 _leave(" = SUCCESS");
237 return CACHEFS_MATCH_SUCCESS; 361 return FSCACHE_CHECKAUX_OKAY;
238} 362}
239#endif
240 363
241/* 364/*
242 * update a vnode record stored in the cache 365 * indication the cookie is no longer uncached
366 * - this function is called when the backing store currently caching a cookie
367 * is removed
368 * - the netfs should use this to clean up any markers indicating cached pages
369 * - this is mandatory for any object that may have data
243 */ 370 */
244#ifdef AFS_CACHING_SUPPORT 371static void afs_vnode_cache_now_uncached(void *cookie_netfs_data)
245static void afs_vnode_cache_update(void *source, void *entry)
246{ 372{
247 struct afs_cache_vnode *cvnode = entry; 373 struct afs_vnode *vnode = cookie_netfs_data;
248 struct afs_vnode *vnode = source; 374 struct pagevec pvec;
375 pgoff_t first;
376 int loop, nr_pages;
377
378 _enter("{%x,%x,%Lx}",
379 vnode->fid.vnode, vnode->fid.unique, vnode->status.data_version);
380
381 pagevec_init(&pvec, 0);
382 first = 0;
383
384 for (;;) {
385 /* grab a bunch of pages to clean */
386 nr_pages = pagevec_lookup(&pvec, vnode->vfs_inode.i_mapping,
387 first,
388 PAGEVEC_SIZE - pagevec_count(&pvec));
389 if (!nr_pages)
390 break;
249 391
250 _enter(""); 392 for (loop = 0; loop < nr_pages; loop++)
393 ClearPageFsCache(pvec.pages[loop]);
394
395 first = pvec.pages[nr_pages - 1]->index + 1;
396
397 pvec.nr = nr_pages;
398 pagevec_release(&pvec);
399 cond_resched();
400 }
251 401
252 cvnode->vnode_id = vnode->fid.vnode; 402 _leave("");
253 cvnode->vnode_unique = vnode->fid.unique;
254 cvnode->data_version = vnode->status.version;
255} 403}
256#endif
diff --git a/fs/afs/cache.h b/fs/afs/cache.h
index 36a3642cf90e..5c4f6b499e90 100644
--- a/fs/afs/cache.h
+++ b/fs/afs/cache.h
@@ -1,6 +1,6 @@
1/* AFS local cache management interface 1/* AFS local cache management interface
2 * 2 *
3 * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved. 3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com) 4 * Written by David Howells (dhowells@redhat.com)
5 * 5 *
6 * This program is free software; you can redistribute it and/or 6 * This program is free software; you can redistribute it and/or
@@ -9,15 +9,4 @@
9 * 2 of the License, or (at your option) any later version. 9 * 2 of the License, or (at your option) any later version.
10 */ 10 */
11 11
12#ifndef AFS_CACHE_H 12#include <linux/fscache.h>
13#define AFS_CACHE_H
14
15#undef AFS_CACHING_SUPPORT
16
17#include <linux/mm.h>
18#ifdef AFS_CACHING_SUPPORT
19#include <linux/cachefs.h>
20#endif
21#include "types.h"
22
23#endif /* AFS_CACHE_H */
diff --git a/fs/afs/cell.c b/fs/afs/cell.c
index 5e1df14e16b1..e19c13f059ed 100644
--- a/fs/afs/cell.c
+++ b/fs/afs/cell.c
@@ -147,12 +147,11 @@ struct afs_cell *afs_cell_create(const char *name, char *vllist)
147 if (ret < 0) 147 if (ret < 0)
148 goto error; 148 goto error;
149 149
150#ifdef AFS_CACHING_SUPPORT 150#ifdef CONFIG_AFS_FSCACHE
151 /* put it up for caching */ 151 /* put it up for caching (this never returns an error) */
152 cachefs_acquire_cookie(afs_cache_netfs.primary_index, 152 cell->cache = fscache_acquire_cookie(afs_cache_netfs.primary_index,
153 &afs_vlocation_cache_index_def, 153 &afs_cell_cache_index_def,
154 cell, 154 cell);
155 &cell->cache);
156#endif 155#endif
157 156
158 /* add to the cell lists */ 157 /* add to the cell lists */
@@ -362,10 +361,9 @@ static void afs_cell_destroy(struct afs_cell *cell)
362 list_del_init(&cell->proc_link); 361 list_del_init(&cell->proc_link);
363 up_write(&afs_proc_cells_sem); 362 up_write(&afs_proc_cells_sem);
364 363
365#ifdef AFS_CACHING_SUPPORT 364#ifdef CONFIG_AFS_FSCACHE
366 cachefs_relinquish_cookie(cell->cache, 0); 365 fscache_relinquish_cookie(cell->cache, 0);
367#endif 366#endif
368
369 key_put(cell->anonymous_key); 367 key_put(cell->anonymous_key);
370 kfree(cell); 368 kfree(cell);
371 369
diff --git a/fs/afs/file.c b/fs/afs/file.c
index a3901769a96c..7a1d942ef68d 100644
--- a/fs/afs/file.c
+++ b/fs/afs/file.c
@@ -23,6 +23,9 @@ static void afs_invalidatepage(struct page *page, unsigned long offset);
23static int afs_releasepage(struct page *page, gfp_t gfp_flags); 23static int afs_releasepage(struct page *page, gfp_t gfp_flags);
24static int afs_launder_page(struct page *page); 24static int afs_launder_page(struct page *page);
25 25
26static int afs_readpages(struct file *filp, struct address_space *mapping,
27 struct list_head *pages, unsigned nr_pages);
28
26const struct file_operations afs_file_operations = { 29const struct file_operations afs_file_operations = {
27 .open = afs_open, 30 .open = afs_open,
28 .release = afs_release, 31 .release = afs_release,
@@ -46,6 +49,7 @@ const struct inode_operations afs_file_inode_operations = {
46 49
47const struct address_space_operations afs_fs_aops = { 50const struct address_space_operations afs_fs_aops = {
48 .readpage = afs_readpage, 51 .readpage = afs_readpage,
52 .readpages = afs_readpages,
49 .set_page_dirty = afs_set_page_dirty, 53 .set_page_dirty = afs_set_page_dirty,
50 .launder_page = afs_launder_page, 54 .launder_page = afs_launder_page,
51 .releasepage = afs_releasepage, 55 .releasepage = afs_releasepage,
@@ -101,37 +105,18 @@ int afs_release(struct inode *inode, struct file *file)
101/* 105/*
102 * deal with notification that a page was read from the cache 106 * deal with notification that a page was read from the cache
103 */ 107 */
104#ifdef AFS_CACHING_SUPPORT 108static void afs_file_readpage_read_complete(struct page *page,
105static void afs_readpage_read_complete(void *cookie_data, 109 void *data,
106 struct page *page, 110 int error)
107 void *data,
108 int error)
109{ 111{
110 _enter("%p,%p,%p,%d", cookie_data, page, data, error); 112 _enter("%p,%p,%d", page, data, error);
111 113
112 if (error) 114 /* if the read completes with an error, we just unlock the page and let
113 SetPageError(page); 115 * the VM reissue the readpage */
114 else 116 if (!error)
115 SetPageUptodate(page); 117 SetPageUptodate(page);
116 unlock_page(page); 118 unlock_page(page);
117
118} 119}
119#endif
120
121/*
122 * deal with notification that a page was written to the cache
123 */
124#ifdef AFS_CACHING_SUPPORT
125static void afs_readpage_write_complete(void *cookie_data,
126 struct page *page,
127 void *data,
128 int error)
129{
130 _enter("%p,%p,%p,%d", cookie_data, page, data, error);
131
132 unlock_page(page);
133}
134#endif
135 120
136/* 121/*
137 * AFS read page from file, directory or symlink 122 * AFS read page from file, directory or symlink
@@ -161,9 +146,9 @@ static int afs_readpage(struct file *file, struct page *page)
161 if (test_bit(AFS_VNODE_DELETED, &vnode->flags)) 146 if (test_bit(AFS_VNODE_DELETED, &vnode->flags))
162 goto error; 147 goto error;
163 148
164#ifdef AFS_CACHING_SUPPORT
165 /* is it cached? */ 149 /* is it cached? */
166 ret = cachefs_read_or_alloc_page(vnode->cache, 150#ifdef CONFIG_AFS_FSCACHE
151 ret = fscache_read_or_alloc_page(vnode->cache,
167 page, 152 page,
168 afs_file_readpage_read_complete, 153 afs_file_readpage_read_complete,
169 NULL, 154 NULL,
@@ -171,20 +156,21 @@ static int afs_readpage(struct file *file, struct page *page)
171#else 156#else
172 ret = -ENOBUFS; 157 ret = -ENOBUFS;
173#endif 158#endif
174
175 switch (ret) { 159 switch (ret) {
176 /* read BIO submitted and wb-journal entry found */
177 case 1:
178 BUG(); // TODO - handle wb-journal match
179
180 /* read BIO submitted (page in cache) */ 160 /* read BIO submitted (page in cache) */
181 case 0: 161 case 0:
182 break; 162 break;
183 163
184 /* no page available in cache */ 164 /* page not yet cached */
185 case -ENOBUFS:
186 case -ENODATA: 165 case -ENODATA:
166 _debug("cache said ENODATA");
167 goto go_on;
168
169 /* page will not be cached */
170 case -ENOBUFS:
171 _debug("cache said ENOBUFS");
187 default: 172 default:
173 go_on:
188 offset = page->index << PAGE_CACHE_SHIFT; 174 offset = page->index << PAGE_CACHE_SHIFT;
189 len = min_t(size_t, i_size_read(inode) - offset, PAGE_SIZE); 175 len = min_t(size_t, i_size_read(inode) - offset, PAGE_SIZE);
190 176
@@ -198,27 +184,25 @@ static int afs_readpage(struct file *file, struct page *page)
198 set_bit(AFS_VNODE_DELETED, &vnode->flags); 184 set_bit(AFS_VNODE_DELETED, &vnode->flags);
199 ret = -ESTALE; 185 ret = -ESTALE;
200 } 186 }
201#ifdef AFS_CACHING_SUPPORT 187
202 cachefs_uncache_page(vnode->cache, page); 188#ifdef CONFIG_AFS_FSCACHE
189 fscache_uncache_page(vnode->cache, page);
203#endif 190#endif
191 BUG_ON(PageFsCache(page));
204 goto error; 192 goto error;
205 } 193 }
206 194
207 SetPageUptodate(page); 195 SetPageUptodate(page);
208 196
209#ifdef AFS_CACHING_SUPPORT 197 /* send the page to the cache */
210 if (cachefs_write_page(vnode->cache, 198#ifdef CONFIG_AFS_FSCACHE
211 page, 199 if (PageFsCache(page) &&
212 afs_file_readpage_write_complete, 200 fscache_write_page(vnode->cache, page, GFP_KERNEL) != 0) {
213 NULL, 201 fscache_uncache_page(vnode->cache, page);
214 GFP_KERNEL) != 0 202 BUG_ON(PageFsCache(page));
215 ) {
216 cachefs_uncache_page(vnode->cache, page);
217 unlock_page(page);
218 } 203 }
219#else
220 unlock_page(page);
221#endif 204#endif
205 unlock_page(page);
222 } 206 }
223 207
224 _leave(" = 0"); 208 _leave(" = 0");
@@ -232,34 +216,59 @@ error:
232} 216}
233 217
234/* 218/*
235 * invalidate part or all of a page 219 * read a set of pages
236 */ 220 */
237static void afs_invalidatepage(struct page *page, unsigned long offset) 221static int afs_readpages(struct file *file, struct address_space *mapping,
222 struct list_head *pages, unsigned nr_pages)
238{ 223{
239 int ret = 1; 224 struct afs_vnode *vnode;
225 int ret = 0;
240 226
241 _enter("{%lu},%lu", page->index, offset); 227 _enter(",{%lu},,%d", mapping->host->i_ino, nr_pages);
242 228
243 BUG_ON(!PageLocked(page)); 229 vnode = AFS_FS_I(mapping->host);
230 if (vnode->flags & AFS_VNODE_DELETED) {
231 _leave(" = -ESTALE");
232 return -ESTALE;
233 }
244 234
245 if (PagePrivate(page)) { 235 /* attempt to read as many of the pages as possible */
246 /* We release buffers only if the entire page is being 236#ifdef CONFIG_AFS_FSCACHE
247 * invalidated. 237 ret = fscache_read_or_alloc_pages(vnode->cache,
248 * The get_block cached value has been unconditionally 238 mapping,
249 * invalidated, so real IO is not possible anymore. 239 pages,
250 */ 240 &nr_pages,
251 if (offset == 0) { 241 afs_file_readpage_read_complete,
252 BUG_ON(!PageLocked(page)); 242 NULL,
253 243 mapping_gfp_mask(mapping));
254 ret = 0; 244#else
255 if (!PageWriteback(page)) 245 ret = -ENOBUFS;
256 ret = page->mapping->a_ops->releasepage(page, 246#endif
257 0); 247
258 /* possibly should BUG_ON(!ret); - neilb */ 248 switch (ret) {
259 } 249 /* all pages are being read from the cache */
250 case 0:
251 BUG_ON(!list_empty(pages));
252 BUG_ON(nr_pages != 0);
253 _leave(" = 0 [reading all]");
254 return 0;
255
256 /* there were pages that couldn't be read from the cache */
257 case -ENODATA:
258 case -ENOBUFS:
259 break;
260
261 /* other error */
262 default:
263 _leave(" = %d", ret);
264 return ret;
260 } 265 }
261 266
262 _leave(" = %d", ret); 267 /* load the missing pages from the network */
268 ret = read_cache_pages(mapping, pages, (void *) afs_readpage, file);
269
270 _leave(" = %d [netting]", ret);
271 return ret;
263} 272}
264 273
265/* 274/*
@@ -273,25 +282,82 @@ static int afs_launder_page(struct page *page)
273} 282}
274 283
275/* 284/*
276 * release a page and cleanup its private data 285 * invalidate part or all of a page
286 * - release a page and clean up its private data if offset is 0 (indicating
287 * the entire page)
288 */
289static void afs_invalidatepage(struct page *page, unsigned long offset)
290{
291 struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
292
293 _enter("{%lu},%lu", page->index, offset);
294
295 BUG_ON(!PageLocked(page));
296
297 /* we clean up only if the entire page is being invalidated */
298 if (offset == 0) {
299#ifdef CONFIG_AFS_FSCACHE
300 if (PageFsCache(page)) {
301 struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
302 fscache_wait_on_page_write(vnode->cache, page);
303 fscache_uncache_page(vnode->cache, page);
304 ClearPageFsCache(page);
305 }
306#endif
307
308 if (PagePrivate(page)) {
309 if (wb && !PageWriteback(page)) {
310 set_page_private(page, 0);
311 afs_put_writeback(wb);
312 }
313
314 if (!page_private(page))
315 ClearPagePrivate(page);
316 }
317 }
318
319 _leave("");
320}
321
322/*
323 * release a page and clean up its private state if it's not busy
324 * - return true if the page can now be released, false if not
277 */ 325 */
278static int afs_releasepage(struct page *page, gfp_t gfp_flags) 326static int afs_releasepage(struct page *page, gfp_t gfp_flags)
279{ 327{
328 struct afs_writeback *wb = (struct afs_writeback *) page_private(page);
280 struct afs_vnode *vnode = AFS_FS_I(page->mapping->host); 329 struct afs_vnode *vnode = AFS_FS_I(page->mapping->host);
281 struct afs_writeback *wb;
282 330
283 _enter("{{%x:%u}[%lu],%lx},%x", 331 _enter("{{%x:%u}[%lu],%lx},%x",
284 vnode->fid.vid, vnode->fid.vnode, page->index, page->flags, 332 vnode->fid.vid, vnode->fid.vnode, page->index, page->flags,
285 gfp_flags); 333 gfp_flags);
286 334
335 /* deny if page is being written to the cache and the caller hasn't
336 * elected to wait */
337#ifdef CONFIG_AFS_FSCACHE
338 if (PageFsCache(page)) {
339 if (fscache_check_page_write(vnode->cache, page)) {
340 if (!(gfp_flags & __GFP_WAIT)) {
341 _leave(" = F [cache busy]");
342 return 0;
343 }
344 fscache_wait_on_page_write(vnode->cache, page);
345 }
346
347 fscache_uncache_page(vnode->cache, page);
348 ClearPageFsCache(page);
349 }
350#endif
351
287 if (PagePrivate(page)) { 352 if (PagePrivate(page)) {
288 wb = (struct afs_writeback *) page_private(page); 353 if (wb) {
289 ASSERT(wb != NULL); 354 set_page_private(page, 0);
290 set_page_private(page, 0); 355 afs_put_writeback(wb);
356 }
291 ClearPagePrivate(page); 357 ClearPagePrivate(page);
292 afs_put_writeback(wb);
293 } 358 }
294 359
295 _leave(" = 0"); 360 /* indicate that the page can be released */
296 return 0; 361 _leave(" = T");
362 return 1;
297} 363}
diff --git a/fs/afs/inode.c b/fs/afs/inode.c
index bb47217f6a18..c048f0658751 100644
--- a/fs/afs/inode.c
+++ b/fs/afs/inode.c
@@ -61,6 +61,11 @@ static int afs_inode_map_status(struct afs_vnode *vnode, struct key *key)
61 return -EBADMSG; 61 return -EBADMSG;
62 } 62 }
63 63
64#ifdef CONFIG_AFS_FSCACHE
65 if (vnode->status.size != inode->i_size)
66 fscache_attr_changed(vnode->cache);
67#endif
68
64 inode->i_nlink = vnode->status.nlink; 69 inode->i_nlink = vnode->status.nlink;
65 inode->i_uid = vnode->status.owner; 70 inode->i_uid = vnode->status.owner;
66 inode->i_gid = 0; 71 inode->i_gid = 0;
@@ -149,15 +154,6 @@ struct inode *afs_iget(struct super_block *sb, struct key *key,
149 return inode; 154 return inode;
150 } 155 }
151 156
152#ifdef AFS_CACHING_SUPPORT
153 /* set up caching before reading the status, as fetch-status reads the
154 * first page of symlinks to see if they're really mntpts */
155 cachefs_acquire_cookie(vnode->volume->cache,
156 NULL,
157 vnode,
158 &vnode->cache);
159#endif
160
161 if (!status) { 157 if (!status) {
162 /* it's a remotely extant inode */ 158 /* it's a remotely extant inode */
163 set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags); 159 set_bit(AFS_VNODE_CB_BROKEN, &vnode->flags);
@@ -183,6 +179,15 @@ struct inode *afs_iget(struct super_block *sb, struct key *key,
183 } 179 }
184 } 180 }
185 181
182 /* set up caching before mapping the status, as map-status reads the
183 * first page of symlinks to see if they're really mountpoints */
184 inode->i_size = vnode->status.size;
185#ifdef CONFIG_AFS_FSCACHE
186 vnode->cache = fscache_acquire_cookie(vnode->volume->cache,
187 &afs_vnode_cache_index_def,
188 vnode);
189#endif
190
186 ret = afs_inode_map_status(vnode, key); 191 ret = afs_inode_map_status(vnode, key);
187 if (ret < 0) 192 if (ret < 0)
188 goto bad_inode; 193 goto bad_inode;
@@ -196,6 +201,10 @@ struct inode *afs_iget(struct super_block *sb, struct key *key,
196 201
197 /* failure */ 202 /* failure */
198bad_inode: 203bad_inode:
204#ifdef CONFIG_AFS_FSCACHE
205 fscache_relinquish_cookie(vnode->cache, 0);
206 vnode->cache = NULL;
207#endif
199 iget_failed(inode); 208 iget_failed(inode);
200 _leave(" = %d [bad]", ret); 209 _leave(" = %d [bad]", ret);
201 return ERR_PTR(ret); 210 return ERR_PTR(ret);
@@ -340,8 +349,8 @@ void afs_clear_inode(struct inode *inode)
340 ASSERT(list_empty(&vnode->writebacks)); 349 ASSERT(list_empty(&vnode->writebacks));
341 ASSERT(!vnode->cb_promised); 350 ASSERT(!vnode->cb_promised);
342 351
343#ifdef AFS_CACHING_SUPPORT 352#ifdef CONFIG_AFS_FSCACHE
344 cachefs_relinquish_cookie(vnode->cache, 0); 353 fscache_relinquish_cookie(vnode->cache, 0);
345 vnode->cache = NULL; 354 vnode->cache = NULL;
346#endif 355#endif
347 356
diff --git a/fs/afs/internal.h b/fs/afs/internal.h
index 67f259d99cd6..106be66dafd2 100644
--- a/fs/afs/internal.h
+++ b/fs/afs/internal.h
@@ -21,6 +21,7 @@
21 21
22#include "afs.h" 22#include "afs.h"
23#include "afs_vl.h" 23#include "afs_vl.h"
24#include "cache.h"
24 25
25#define AFS_CELL_MAX_ADDRS 15 26#define AFS_CELL_MAX_ADDRS 15
26 27
@@ -193,8 +194,8 @@ struct afs_cell {
193 struct key *anonymous_key; /* anonymous user key for this cell */ 194 struct key *anonymous_key; /* anonymous user key for this cell */
194 struct list_head proc_link; /* /proc cell list link */ 195 struct list_head proc_link; /* /proc cell list link */
195 struct proc_dir_entry *proc_dir; /* /proc dir for this cell */ 196 struct proc_dir_entry *proc_dir; /* /proc dir for this cell */
196#ifdef AFS_CACHING_SUPPORT 197#ifdef CONFIG_AFS_FSCACHE
197 struct cachefs_cookie *cache; /* caching cookie */ 198 struct fscache_cookie *cache; /* caching cookie */
198#endif 199#endif
199 200
200 /* server record management */ 201 /* server record management */
@@ -249,8 +250,8 @@ struct afs_vlocation {
249 struct list_head grave; /* link in master graveyard list */ 250 struct list_head grave; /* link in master graveyard list */
250 struct list_head update; /* link in master update list */ 251 struct list_head update; /* link in master update list */
251 struct afs_cell *cell; /* cell to which volume belongs */ 252 struct afs_cell *cell; /* cell to which volume belongs */
252#ifdef AFS_CACHING_SUPPORT 253#ifdef CONFIG_AFS_FSCACHE
253 struct cachefs_cookie *cache; /* caching cookie */ 254 struct fscache_cookie *cache; /* caching cookie */
254#endif 255#endif
255 struct afs_cache_vlocation vldb; /* volume information DB record */ 256 struct afs_cache_vlocation vldb; /* volume information DB record */
256 struct afs_volume *vols[3]; /* volume access record pointer (index by type) */ 257 struct afs_volume *vols[3]; /* volume access record pointer (index by type) */
@@ -302,8 +303,8 @@ struct afs_volume {
302 atomic_t usage; 303 atomic_t usage;
303 struct afs_cell *cell; /* cell to which belongs (unrefd ptr) */ 304 struct afs_cell *cell; /* cell to which belongs (unrefd ptr) */
304 struct afs_vlocation *vlocation; /* volume location */ 305 struct afs_vlocation *vlocation; /* volume location */
305#ifdef AFS_CACHING_SUPPORT 306#ifdef CONFIG_AFS_FSCACHE
306 struct cachefs_cookie *cache; /* caching cookie */ 307 struct fscache_cookie *cache; /* caching cookie */
307#endif 308#endif
308 afs_volid_t vid; /* volume ID */ 309 afs_volid_t vid; /* volume ID */
309 afs_voltype_t type; /* type of volume */ 310 afs_voltype_t type; /* type of volume */
@@ -333,8 +334,8 @@ struct afs_vnode {
333 struct afs_server *server; /* server currently supplying this file */ 334 struct afs_server *server; /* server currently supplying this file */
334 struct afs_fid fid; /* the file identifier for this inode */ 335 struct afs_fid fid; /* the file identifier for this inode */
335 struct afs_file_status status; /* AFS status info for this file */ 336 struct afs_file_status status; /* AFS status info for this file */
336#ifdef AFS_CACHING_SUPPORT 337#ifdef CONFIG_AFS_FSCACHE
337 struct cachefs_cookie *cache; /* caching cookie */ 338 struct fscache_cookie *cache; /* caching cookie */
338#endif 339#endif
339 struct afs_permits *permits; /* cache of permits so far obtained */ 340 struct afs_permits *permits; /* cache of permits so far obtained */
340 struct mutex permits_lock; /* lock for altering permits list */ 341 struct mutex permits_lock; /* lock for altering permits list */
@@ -428,6 +429,22 @@ struct afs_uuid {
428 429
429/*****************************************************************************/ 430/*****************************************************************************/
430/* 431/*
432 * cache.c
433 */
434#ifdef CONFIG_AFS_FSCACHE
435extern struct fscache_netfs afs_cache_netfs;
436extern struct fscache_cookie_def afs_cell_cache_index_def;
437extern struct fscache_cookie_def afs_vlocation_cache_index_def;
438extern struct fscache_cookie_def afs_volume_cache_index_def;
439extern struct fscache_cookie_def afs_vnode_cache_index_def;
440#else
441#define afs_cell_cache_index_def (*(struct fscache_cookie_def *) NULL)
442#define afs_vlocation_cache_index_def (*(struct fscache_cookie_def *) NULL)
443#define afs_volume_cache_index_def (*(struct fscache_cookie_def *) NULL)
444#define afs_vnode_cache_index_def (*(struct fscache_cookie_def *) NULL)
445#endif
446
447/*
431 * callback.c 448 * callback.c
432 */ 449 */
433extern void afs_init_callback_state(struct afs_server *); 450extern void afs_init_callback_state(struct afs_server *);
@@ -446,9 +463,6 @@ extern void afs_callback_update_kill(void);
446 */ 463 */
447extern struct rw_semaphore afs_proc_cells_sem; 464extern struct rw_semaphore afs_proc_cells_sem;
448extern struct list_head afs_proc_cells; 465extern struct list_head afs_proc_cells;
449#ifdef AFS_CACHING_SUPPORT
450extern struct cachefs_index_def afs_cache_cell_index_def;
451#endif
452 466
453#define afs_get_cell(C) do { atomic_inc(&(C)->usage); } while(0) 467#define afs_get_cell(C) do { atomic_inc(&(C)->usage); } while(0)
454extern int afs_cell_init(char *); 468extern int afs_cell_init(char *);
@@ -554,9 +568,6 @@ extern void afs_clear_inode(struct inode *);
554 * main.c 568 * main.c
555 */ 569 */
556extern struct afs_uuid afs_uuid; 570extern struct afs_uuid afs_uuid;
557#ifdef AFS_CACHING_SUPPORT
558extern struct cachefs_netfs afs_cache_netfs;
559#endif
560 571
561/* 572/*
562 * misc.c 573 * misc.c
@@ -637,10 +648,6 @@ extern int afs_get_MAC_address(u8 *, size_t);
637/* 648/*
638 * vlclient.c 649 * vlclient.c
639 */ 650 */
640#ifdef AFS_CACHING_SUPPORT
641extern struct cachefs_index_def afs_vlocation_cache_index_def;
642#endif
643
644extern int afs_vl_get_entry_by_name(struct in_addr *, struct key *, 651extern int afs_vl_get_entry_by_name(struct in_addr *, struct key *,
645 const char *, struct afs_cache_vlocation *, 652 const char *, struct afs_cache_vlocation *,
646 const struct afs_wait_mode *); 653 const struct afs_wait_mode *);
@@ -664,12 +671,6 @@ extern void afs_vlocation_purge(void);
664/* 671/*
665 * vnode.c 672 * vnode.c
666 */ 673 */
667#ifdef AFS_CACHING_SUPPORT
668extern struct cachefs_index_def afs_vnode_cache_index_def;
669#endif
670
671extern struct afs_timer_ops afs_vnode_cb_timed_out_ops;
672
673static inline struct afs_vnode *AFS_FS_I(struct inode *inode) 674static inline struct afs_vnode *AFS_FS_I(struct inode *inode)
674{ 675{
675 return container_of(inode, struct afs_vnode, vfs_inode); 676 return container_of(inode, struct afs_vnode, vfs_inode);
@@ -711,10 +712,6 @@ extern int afs_vnode_release_lock(struct afs_vnode *, struct key *);
711/* 712/*
712 * volume.c 713 * volume.c
713 */ 714 */
714#ifdef AFS_CACHING_SUPPORT
715extern struct cachefs_index_def afs_volume_cache_index_def;
716#endif
717
718#define afs_get_volume(V) do { atomic_inc(&(V)->usage); } while(0) 715#define afs_get_volume(V) do { atomic_inc(&(V)->usage); } while(0)
719 716
720extern void afs_put_volume(struct afs_volume *); 717extern void afs_put_volume(struct afs_volume *);
diff --git a/fs/afs/main.c b/fs/afs/main.c
index 2d3e5d4fb9f7..66d54d348c55 100644
--- a/fs/afs/main.c
+++ b/fs/afs/main.c
@@ -1,6 +1,6 @@
1/* AFS client file system 1/* AFS client file system
2 * 2 *
3 * Copyright (C) 2002 Red Hat, Inc. All Rights Reserved. 3 * Copyright (C) 2002,5 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com) 4 * Written by David Howells (dhowells@redhat.com)
5 * 5 *
6 * This program is free software; you can redistribute it and/or 6 * This program is free software; you can redistribute it and/or
@@ -29,18 +29,6 @@ static char *rootcell;
29module_param(rootcell, charp, 0); 29module_param(rootcell, charp, 0);
30MODULE_PARM_DESC(rootcell, "root AFS cell name and VL server IP addr list"); 30MODULE_PARM_DESC(rootcell, "root AFS cell name and VL server IP addr list");
31 31
32#ifdef AFS_CACHING_SUPPORT
33static struct cachefs_netfs_operations afs_cache_ops = {
34 .get_page_cookie = afs_cache_get_page_cookie,
35};
36
37struct cachefs_netfs afs_cache_netfs = {
38 .name = "afs",
39 .version = 0,
40 .ops = &afs_cache_ops,
41};
42#endif
43
44struct afs_uuid afs_uuid; 32struct afs_uuid afs_uuid;
45 33
46/* 34/*
@@ -104,10 +92,9 @@ static int __init afs_init(void)
104 if (ret < 0) 92 if (ret < 0)
105 return ret; 93 return ret;
106 94
107#ifdef AFS_CACHING_SUPPORT 95#ifdef CONFIG_AFS_FSCACHE
108 /* we want to be able to cache */ 96 /* we want to be able to cache */
109 ret = cachefs_register_netfs(&afs_cache_netfs, 97 ret = fscache_register_netfs(&afs_cache_netfs);
110 &afs_cache_cell_index_def);
111 if (ret < 0) 98 if (ret < 0)
112 goto error_cache; 99 goto error_cache;
113#endif 100#endif
@@ -142,8 +129,8 @@ error_fs:
142error_open_socket: 129error_open_socket:
143error_vl_update_init: 130error_vl_update_init:
144error_cell_init: 131error_cell_init:
145#ifdef AFS_CACHING_SUPPORT 132#ifdef CONFIG_AFS_FSCACHE
146 cachefs_unregister_netfs(&afs_cache_netfs); 133 fscache_unregister_netfs(&afs_cache_netfs);
147error_cache: 134error_cache:
148#endif 135#endif
149 afs_callback_update_kill(); 136 afs_callback_update_kill();
@@ -175,8 +162,8 @@ static void __exit afs_exit(void)
175 afs_vlocation_purge(); 162 afs_vlocation_purge();
176 flush_scheduled_work(); 163 flush_scheduled_work();
177 afs_cell_purge(); 164 afs_cell_purge();
178#ifdef AFS_CACHING_SUPPORT 165#ifdef CONFIG_AFS_FSCACHE
179 cachefs_unregister_netfs(&afs_cache_netfs); 166 fscache_unregister_netfs(&afs_cache_netfs);
180#endif 167#endif
181 afs_proc_cleanup(); 168 afs_proc_cleanup();
182 rcu_barrier(); 169 rcu_barrier();
diff --git a/fs/afs/mntpt.c b/fs/afs/mntpt.c
index 78db4953a800..2b9e2d03a390 100644
--- a/fs/afs/mntpt.c
+++ b/fs/afs/mntpt.c
@@ -173,9 +173,9 @@ static struct vfsmount *afs_mntpt_do_automount(struct dentry *mntpt)
173 if (PageError(page)) 173 if (PageError(page))
174 goto error; 174 goto error;
175 175
176 buf = kmap(page); 176 buf = kmap_atomic(page, KM_USER0);
177 memcpy(devname, buf, size); 177 memcpy(devname, buf, size);
178 kunmap(page); 178 kunmap_atomic(buf, KM_USER0);
179 page_cache_release(page); 179 page_cache_release(page);
180 page = NULL; 180 page = NULL;
181 181
diff --git a/fs/afs/vlocation.c b/fs/afs/vlocation.c
index 849fc3160cb5..ec2a7431e458 100644
--- a/fs/afs/vlocation.c
+++ b/fs/afs/vlocation.c
@@ -281,9 +281,8 @@ static void afs_vlocation_apply_update(struct afs_vlocation *vl,
281 281
282 vl->vldb = *vldb; 282 vl->vldb = *vldb;
283 283
284#ifdef AFS_CACHING_SUPPORT 284#ifdef CONFIG_AFS_FSCACHE
285 /* update volume entry in local cache */ 285 fscache_update_cookie(vl->cache);
286 cachefs_update_cookie(vl->cache);
287#endif 286#endif
288} 287}
289 288
@@ -304,11 +303,9 @@ static int afs_vlocation_fill_in_record(struct afs_vlocation *vl,
304 memset(&vldb, 0, sizeof(vldb)); 303 memset(&vldb, 0, sizeof(vldb));
305 304
306 /* see if we have an in-cache copy (will set vl->valid if there is) */ 305 /* see if we have an in-cache copy (will set vl->valid if there is) */
307#ifdef AFS_CACHING_SUPPORT 306#ifdef CONFIG_AFS_FSCACHE
308 cachefs_acquire_cookie(cell->cache, 307 vl->cache = fscache_acquire_cookie(vl->cell->cache,
309 &afs_volume_cache_index_def, 308 &afs_vlocation_cache_index_def, vl);
310 vlocation,
311 &vl->cache);
312#endif 309#endif
313 310
314 if (vl->valid) { 311 if (vl->valid) {
@@ -420,6 +417,11 @@ fill_in_record:
420 spin_unlock(&vl->lock); 417 spin_unlock(&vl->lock);
421 wake_up(&vl->waitq); 418 wake_up(&vl->waitq);
422 419
420 /* update volume entry in local cache */
421#ifdef CONFIG_AFS_FSCACHE
422 fscache_update_cookie(vl->cache);
423#endif
424
423 /* schedule for regular updates */ 425 /* schedule for regular updates */
424 afs_vlocation_queue_for_updates(vl); 426 afs_vlocation_queue_for_updates(vl);
425 goto success; 427 goto success;
@@ -465,7 +467,7 @@ found_in_memory:
465 spin_unlock(&vl->lock); 467 spin_unlock(&vl->lock);
466 468
467success: 469success:
468 _leave(" = %p",vl); 470 _leave(" = %p", vl);
469 return vl; 471 return vl;
470 472
471error_abandon: 473error_abandon:
@@ -523,10 +525,9 @@ static void afs_vlocation_destroy(struct afs_vlocation *vl)
523{ 525{
524 _enter("%p", vl); 526 _enter("%p", vl);
525 527
526#ifdef AFS_CACHING_SUPPORT 528#ifdef CONFIG_AFS_FSCACHE
527 cachefs_relinquish_cookie(vl->cache, 0); 529 fscache_relinquish_cookie(vl->cache, 0);
528#endif 530#endif
529
530 afs_put_cell(vl->cell); 531 afs_put_cell(vl->cell);
531 kfree(vl); 532 kfree(vl);
532} 533}
diff --git a/fs/afs/volume.c b/fs/afs/volume.c
index 8bab0e3437f9..a353e69e2391 100644
--- a/fs/afs/volume.c
+++ b/fs/afs/volume.c
@@ -124,13 +124,11 @@ struct afs_volume *afs_volume_lookup(struct afs_mount_params *params)
124 } 124 }
125 125
126 /* attach the cache and volume location */ 126 /* attach the cache and volume location */
127#ifdef AFS_CACHING_SUPPORT 127#ifdef CONFIG_AFS_FSCACHE
128 cachefs_acquire_cookie(vlocation->cache, 128 volume->cache = fscache_acquire_cookie(vlocation->cache,
129 &afs_vnode_cache_index_def, 129 &afs_volume_cache_index_def,
130 volume, 130 volume);
131 &volume->cache);
132#endif 131#endif
133
134 afs_get_vlocation(vlocation); 132 afs_get_vlocation(vlocation);
135 volume->vlocation = vlocation; 133 volume->vlocation = vlocation;
136 134
@@ -194,8 +192,8 @@ void afs_put_volume(struct afs_volume *volume)
194 up_write(&vlocation->cell->vl_sem); 192 up_write(&vlocation->cell->vl_sem);
195 193
196 /* finish cleaning up the volume */ 194 /* finish cleaning up the volume */
197#ifdef AFS_CACHING_SUPPORT 195#ifdef CONFIG_AFS_FSCACHE
198 cachefs_relinquish_cookie(volume->cache, 0); 196 fscache_relinquish_cookie(volume->cache, 0);
199#endif 197#endif
200 afs_put_vlocation(vlocation); 198 afs_put_vlocation(vlocation);
201 199
diff --git a/fs/afs/write.c b/fs/afs/write.c
index 3fb36d433621..c2e7a7ff0080 100644
--- a/fs/afs/write.c
+++ b/fs/afs/write.c
@@ -780,3 +780,24 @@ int afs_fsync(struct file *file, struct dentry *dentry, int datasync)
780 _leave(" = %d", ret); 780 _leave(" = %d", ret);
781 return ret; 781 return ret;
782} 782}
783
784/*
785 * notification that a previously read-only page is about to become writable
786 * - if it returns an error, the caller will deliver a bus error signal
787 */
788int afs_page_mkwrite(struct vm_area_struct *vma, struct page *page)
789{
790 struct afs_vnode *vnode = AFS_FS_I(vma->vm_file->f_mapping->host);
791
792 _enter("{{%x:%u}},{%lx}",
793 vnode->fid.vid, vnode->fid.vnode, page->index);
794
795 /* wait for the page to be written to the cache before we allow it to
796 * be modified */
797#ifdef CONFIG_AFS_FSCACHE
798 fscache_wait_on_page_write(vnode->cache, page);
799#endif
800
801 _leave(" = 0");
802 return 0;
803}
diff --git a/fs/cachefiles/Kconfig b/fs/cachefiles/Kconfig
new file mode 100644
index 000000000000..80e9c6167f0b
--- /dev/null
+++ b/fs/cachefiles/Kconfig
@@ -0,0 +1,39 @@
1
2config CACHEFILES
3 tristate "Filesystem caching on files"
4 depends on FSCACHE && BLOCK
5 help
6 This permits use of a mounted filesystem as a cache for other
7 filesystems - primarily networking filesystems - thus allowing fast
8 local disk to enhance the speed of slower devices.
9
10 See Documentation/filesystems/caching/cachefiles.txt for more
11 information.
12
13config CACHEFILES_DEBUG
14 bool "Debug CacheFiles"
15 depends on CACHEFILES
16 help
17 This permits debugging to be dynamically enabled in the filesystem
18 caching on files module. If this is set, the debugging output may be
19 enabled by setting bits in /sys/modules/cachefiles/parameter/debug or
20 by including a debugging specifier in /etc/cachefilesd.conf.
21
22config CACHEFILES_HISTOGRAM
23 bool "Gather latency information on CacheFiles"
24 depends on CACHEFILES && PROC_FS
25 help
26
27 This option causes latency information to be gathered on CacheFiles
28 operation and exported through file:
29
30 /proc/fs/cachefiles/histogram
31
32 The generation of this histogram adds a certain amount of overhead to
33 execution as there are a number of points at which data is gathered,
34 and on a multi-CPU system these may be on cachelines that keep
35 bouncing between CPUs. On the other hand, the histogram may be
36 useful for debugging purposes. Saying 'N' here is recommended.
37
38 See Documentation/filesystems/caching/cachefiles.txt for more
39 information.
diff --git a/fs/cachefiles/Makefile b/fs/cachefiles/Makefile
new file mode 100644
index 000000000000..32cbab0ffce3
--- /dev/null
+++ b/fs/cachefiles/Makefile
@@ -0,0 +1,18 @@
1#
2# Makefile for caching in a mounted filesystem
3#
4
5cachefiles-y := \
6 bind.o \
7 daemon.o \
8 interface.o \
9 key.o \
10 main.o \
11 namei.o \
12 rdwr.o \
13 security.o \
14 xattr.o
15
16cachefiles-$(CONFIG_CACHEFILES_HISTOGRAM) += proc.o
17
18obj-$(CONFIG_CACHEFILES) := cachefiles.o
diff --git a/fs/cachefiles/bind.c b/fs/cachefiles/bind.c
new file mode 100644
index 000000000000..3797e0077b35
--- /dev/null
+++ b/fs/cachefiles/bind.c
@@ -0,0 +1,286 @@
1/* Bind and unbind a cache from the filesystem backing it
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/sched.h>
15#include <linux/completion.h>
16#include <linux/slab.h>
17#include <linux/fs.h>
18#include <linux/file.h>
19#include <linux/namei.h>
20#include <linux/mount.h>
21#include <linux/statfs.h>
22#include <linux/ctype.h>
23#include "internal.h"
24
25static int cachefiles_daemon_add_cache(struct cachefiles_cache *caches);
26
27/*
28 * bind a directory as a cache
29 */
30int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args)
31{
32 _enter("{%u,%u,%u,%u,%u,%u},%s",
33 cache->frun_percent,
34 cache->fcull_percent,
35 cache->fstop_percent,
36 cache->brun_percent,
37 cache->bcull_percent,
38 cache->bstop_percent,
39 args);
40
41 /* start by checking things over */
42 ASSERT(cache->fstop_percent >= 0 &&
43 cache->fstop_percent < cache->fcull_percent &&
44 cache->fcull_percent < cache->frun_percent &&
45 cache->frun_percent < 100);
46
47 ASSERT(cache->bstop_percent >= 0 &&
48 cache->bstop_percent < cache->bcull_percent &&
49 cache->bcull_percent < cache->brun_percent &&
50 cache->brun_percent < 100);
51
52 if (*args) {
53 kerror("'bind' command doesn't take an argument");
54 return -EINVAL;
55 }
56
57 if (!cache->rootdirname) {
58 kerror("No cache directory specified");
59 return -EINVAL;
60 }
61
62 /* don't permit already bound caches to be re-bound */
63 if (test_bit(CACHEFILES_READY, &cache->flags)) {
64 kerror("Cache already bound");
65 return -EBUSY;
66 }
67
68 /* make sure we have copies of the tag and dirname strings */
69 if (!cache->tag) {
70 /* the tag string is released by the fops->release()
71 * function, so we don't release it on error here */
72 cache->tag = kstrdup("CacheFiles", GFP_KERNEL);
73 if (!cache->tag)
74 return -ENOMEM;
75 }
76
77 /* add the cache */
78 return cachefiles_daemon_add_cache(cache);
79}
80
81/*
82 * add a cache
83 */
84static int cachefiles_daemon_add_cache(struct cachefiles_cache *cache)
85{
86 struct cachefiles_object *fsdef;
87 struct nameidata nd;
88 struct kstatfs stats;
89 struct dentry *graveyard, *cachedir, *root;
90 const struct cred *saved_cred;
91 int ret;
92
93 _enter("");
94
95 /* we want to work under the module's security ID */
96 ret = cachefiles_get_security_ID(cache);
97 if (ret < 0)
98 return ret;
99
100 cachefiles_begin_secure(cache, &saved_cred);
101
102 /* allocate the root index object */
103 ret = -ENOMEM;
104
105 fsdef = kmem_cache_alloc(cachefiles_object_jar, GFP_KERNEL);
106 if (!fsdef)
107 goto error_root_object;
108
109 ASSERTCMP(fsdef->backer, ==, NULL);
110
111 atomic_set(&fsdef->usage, 1);
112 fsdef->type = FSCACHE_COOKIE_TYPE_INDEX;
113
114 _debug("- fsdef %p", fsdef);
115
116 /* look up the directory at the root of the cache */
117 memset(&nd, 0, sizeof(nd));
118
119 ret = path_lookup(cache->rootdirname, LOOKUP_DIRECTORY, &nd);
120 if (ret < 0)
121 goto error_open_root;
122
123 cache->mnt = mntget(nd.path.mnt);
124 root = dget(nd.path.dentry);
125 path_put(&nd.path);
126
127 /* check parameters */
128 ret = -EOPNOTSUPP;
129 if (!root->d_inode ||
130 !root->d_inode->i_op ||
131 !root->d_inode->i_op->lookup ||
132 !root->d_inode->i_op->mkdir ||
133 !root->d_inode->i_op->setxattr ||
134 !root->d_inode->i_op->getxattr ||
135 !root->d_sb ||
136 !root->d_sb->s_op ||
137 !root->d_sb->s_op->statfs ||
138 !root->d_sb->s_op->sync_fs)
139 goto error_unsupported;
140
141 ret = -EROFS;
142 if (root->d_sb->s_flags & MS_RDONLY)
143 goto error_unsupported;
144
145 /* determine the security of the on-disk cache as this governs
146 * security ID of files we create */
147 ret = cachefiles_determine_cache_security(cache, root, &saved_cred);
148 if (ret < 0)
149 goto error_unsupported;
150
151 /* get the cache size and blocksize */
152 ret = vfs_statfs(root, &stats);
153 if (ret < 0)
154 goto error_unsupported;
155
156 ret = -ERANGE;
157 if (stats.f_bsize <= 0)
158 goto error_unsupported;
159
160 ret = -EOPNOTSUPP;
161 if (stats.f_bsize > PAGE_SIZE)
162 goto error_unsupported;
163
164 cache->bsize = stats.f_bsize;
165 cache->bshift = 0;
166 if (stats.f_bsize < PAGE_SIZE)
167 cache->bshift = PAGE_SHIFT - ilog2(stats.f_bsize);
168
169 _debug("blksize %u (shift %u)",
170 cache->bsize, cache->bshift);
171
172 _debug("size %llu, avail %llu",
173 (unsigned long long) stats.f_blocks,
174 (unsigned long long) stats.f_bavail);
175
176 /* set up caching limits */
177 do_div(stats.f_files, 100);
178 cache->fstop = stats.f_files * cache->fstop_percent;
179 cache->fcull = stats.f_files * cache->fcull_percent;
180 cache->frun = stats.f_files * cache->frun_percent;
181
182 _debug("limits {%llu,%llu,%llu} files",
183 (unsigned long long) cache->frun,
184 (unsigned long long) cache->fcull,
185 (unsigned long long) cache->fstop);
186
187 stats.f_blocks >>= cache->bshift;
188 do_div(stats.f_blocks, 100);
189 cache->bstop = stats.f_blocks * cache->bstop_percent;
190 cache->bcull = stats.f_blocks * cache->bcull_percent;
191 cache->brun = stats.f_blocks * cache->brun_percent;
192
193 _debug("limits {%llu,%llu,%llu} blocks",
194 (unsigned long long) cache->brun,
195 (unsigned long long) cache->bcull,
196 (unsigned long long) cache->bstop);
197
198 /* get the cache directory and check its type */
199 cachedir = cachefiles_get_directory(cache, root, "cache");
200 if (IS_ERR(cachedir)) {
201 ret = PTR_ERR(cachedir);
202 goto error_unsupported;
203 }
204
205 fsdef->dentry = cachedir;
206 fsdef->fscache.cookie = NULL;
207
208 ret = cachefiles_check_object_type(fsdef);
209 if (ret < 0)
210 goto error_unsupported;
211
212 /* get the graveyard directory */
213 graveyard = cachefiles_get_directory(cache, root, "graveyard");
214 if (IS_ERR(graveyard)) {
215 ret = PTR_ERR(graveyard);
216 goto error_unsupported;
217 }
218
219 cache->graveyard = graveyard;
220
221 /* publish the cache */
222 fscache_init_cache(&cache->cache,
223 &cachefiles_cache_ops,
224 "%s",
225 fsdef->dentry->d_sb->s_id);
226
227 fscache_object_init(&fsdef->fscache, NULL, &cache->cache);
228
229 ret = fscache_add_cache(&cache->cache, &fsdef->fscache, cache->tag);
230 if (ret < 0)
231 goto error_add_cache;
232
233 /* done */
234 set_bit(CACHEFILES_READY, &cache->flags);
235 dput(root);
236
237 printk(KERN_INFO "CacheFiles:"
238 " File cache on %s registered\n",
239 cache->cache.identifier);
240
241 /* check how much space the cache has */
242 cachefiles_has_space(cache, 0, 0);
243 cachefiles_end_secure(cache, saved_cred);
244 return 0;
245
246error_add_cache:
247 dput(cache->graveyard);
248 cache->graveyard = NULL;
249error_unsupported:
250 mntput(cache->mnt);
251 cache->mnt = NULL;
252 dput(fsdef->dentry);
253 fsdef->dentry = NULL;
254 dput(root);
255error_open_root:
256 kmem_cache_free(cachefiles_object_jar, fsdef);
257error_root_object:
258 cachefiles_end_secure(cache, saved_cred);
259 kerror("Failed to register: %d", ret);
260 return ret;
261}
262
263/*
264 * unbind a cache on fd release
265 */
266void cachefiles_daemon_unbind(struct cachefiles_cache *cache)
267{
268 _enter("");
269
270 if (test_bit(CACHEFILES_READY, &cache->flags)) {
271 printk(KERN_INFO "CacheFiles:"
272 " File cache on %s unregistering\n",
273 cache->cache.identifier);
274
275 fscache_withdraw_cache(&cache->cache);
276 }
277
278 dput(cache->graveyard);
279 mntput(cache->mnt);
280
281 kfree(cache->rootdirname);
282 kfree(cache->secctx);
283 kfree(cache->tag);
284
285 _leave("");
286}
diff --git a/fs/cachefiles/daemon.c b/fs/cachefiles/daemon.c
new file mode 100644
index 000000000000..4618516dd994
--- /dev/null
+++ b/fs/cachefiles/daemon.c
@@ -0,0 +1,755 @@
1/* Daemon interface
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/sched.h>
15#include <linux/completion.h>
16#include <linux/slab.h>
17#include <linux/fs.h>
18#include <linux/file.h>
19#include <linux/namei.h>
20#include <linux/poll.h>
21#include <linux/mount.h>
22#include <linux/statfs.h>
23#include <linux/ctype.h>
24#include <linux/fs_struct.h>
25#include "internal.h"
26
27static int cachefiles_daemon_open(struct inode *, struct file *);
28static int cachefiles_daemon_release(struct inode *, struct file *);
29static ssize_t cachefiles_daemon_read(struct file *, char __user *, size_t,
30 loff_t *);
31static ssize_t cachefiles_daemon_write(struct file *, const char __user *,
32 size_t, loff_t *);
33static unsigned int cachefiles_daemon_poll(struct file *,
34 struct poll_table_struct *);
35static int cachefiles_daemon_frun(struct cachefiles_cache *, char *);
36static int cachefiles_daemon_fcull(struct cachefiles_cache *, char *);
37static int cachefiles_daemon_fstop(struct cachefiles_cache *, char *);
38static int cachefiles_daemon_brun(struct cachefiles_cache *, char *);
39static int cachefiles_daemon_bcull(struct cachefiles_cache *, char *);
40static int cachefiles_daemon_bstop(struct cachefiles_cache *, char *);
41static int cachefiles_daemon_cull(struct cachefiles_cache *, char *);
42static int cachefiles_daemon_debug(struct cachefiles_cache *, char *);
43static int cachefiles_daemon_dir(struct cachefiles_cache *, char *);
44static int cachefiles_daemon_inuse(struct cachefiles_cache *, char *);
45static int cachefiles_daemon_secctx(struct cachefiles_cache *, char *);
46static int cachefiles_daemon_tag(struct cachefiles_cache *, char *);
47
48static unsigned long cachefiles_open;
49
50const struct file_operations cachefiles_daemon_fops = {
51 .owner = THIS_MODULE,
52 .open = cachefiles_daemon_open,
53 .release = cachefiles_daemon_release,
54 .read = cachefiles_daemon_read,
55 .write = cachefiles_daemon_write,
56 .poll = cachefiles_daemon_poll,
57};
58
59struct cachefiles_daemon_cmd {
60 char name[8];
61 int (*handler)(struct cachefiles_cache *cache, char *args);
62};
63
64static const struct cachefiles_daemon_cmd cachefiles_daemon_cmds[] = {
65 { "bind", cachefiles_daemon_bind },
66 { "brun", cachefiles_daemon_brun },
67 { "bcull", cachefiles_daemon_bcull },
68 { "bstop", cachefiles_daemon_bstop },
69 { "cull", cachefiles_daemon_cull },
70 { "debug", cachefiles_daemon_debug },
71 { "dir", cachefiles_daemon_dir },
72 { "frun", cachefiles_daemon_frun },
73 { "fcull", cachefiles_daemon_fcull },
74 { "fstop", cachefiles_daemon_fstop },
75 { "inuse", cachefiles_daemon_inuse },
76 { "secctx", cachefiles_daemon_secctx },
77 { "tag", cachefiles_daemon_tag },
78 { "", NULL }
79};
80
81
82/*
83 * do various checks
84 */
85static int cachefiles_daemon_open(struct inode *inode, struct file *file)
86{
87 struct cachefiles_cache *cache;
88
89 _enter("");
90
91 /* only the superuser may do this */
92 if (!capable(CAP_SYS_ADMIN))
93 return -EPERM;
94
95 /* the cachefiles device may only be open once at a time */
96 if (xchg(&cachefiles_open, 1) == 1)
97 return -EBUSY;
98
99 /* allocate a cache record */
100 cache = kzalloc(sizeof(struct cachefiles_cache), GFP_KERNEL);
101 if (!cache) {
102 cachefiles_open = 0;
103 return -ENOMEM;
104 }
105
106 mutex_init(&cache->daemon_mutex);
107 cache->active_nodes = RB_ROOT;
108 rwlock_init(&cache->active_lock);
109 init_waitqueue_head(&cache->daemon_pollwq);
110
111 /* set default caching limits
112 * - limit at 1% free space and/or free files
113 * - cull below 5% free space and/or free files
114 * - cease culling above 7% free space and/or free files
115 */
116 cache->frun_percent = 7;
117 cache->fcull_percent = 5;
118 cache->fstop_percent = 1;
119 cache->brun_percent = 7;
120 cache->bcull_percent = 5;
121 cache->bstop_percent = 1;
122
123 file->private_data = cache;
124 cache->cachefilesd = file;
125 return 0;
126}
127
128/*
129 * release a cache
130 */
131static int cachefiles_daemon_release(struct inode *inode, struct file *file)
132{
133 struct cachefiles_cache *cache = file->private_data;
134
135 _enter("");
136
137 ASSERT(cache);
138
139 set_bit(CACHEFILES_DEAD, &cache->flags);
140
141 cachefiles_daemon_unbind(cache);
142
143 ASSERT(!cache->active_nodes.rb_node);
144
145 /* clean up the control file interface */
146 cache->cachefilesd = NULL;
147 file->private_data = NULL;
148 cachefiles_open = 0;
149
150 kfree(cache);
151
152 _leave("");
153 return 0;
154}
155
156/*
157 * read the cache state
158 */
159static ssize_t cachefiles_daemon_read(struct file *file, char __user *_buffer,
160 size_t buflen, loff_t *pos)
161{
162 struct cachefiles_cache *cache = file->private_data;
163 char buffer[256];
164 int n;
165
166 //_enter(",,%zu,", buflen);
167
168 if (!test_bit(CACHEFILES_READY, &cache->flags))
169 return 0;
170
171 /* check how much space the cache has */
172 cachefiles_has_space(cache, 0, 0);
173
174 /* summarise */
175 clear_bit(CACHEFILES_STATE_CHANGED, &cache->flags);
176
177 n = snprintf(buffer, sizeof(buffer),
178 "cull=%c"
179 " frun=%llx"
180 " fcull=%llx"
181 " fstop=%llx"
182 " brun=%llx"
183 " bcull=%llx"
184 " bstop=%llx",
185 test_bit(CACHEFILES_CULLING, &cache->flags) ? '1' : '0',
186 (unsigned long long) cache->frun,
187 (unsigned long long) cache->fcull,
188 (unsigned long long) cache->fstop,
189 (unsigned long long) cache->brun,
190 (unsigned long long) cache->bcull,
191 (unsigned long long) cache->bstop
192 );
193
194 if (n > buflen)
195 return -EMSGSIZE;
196
197 if (copy_to_user(_buffer, buffer, n) != 0)
198 return -EFAULT;
199
200 return n;
201}
202
203/*
204 * command the cache
205 */
206static ssize_t cachefiles_daemon_write(struct file *file,
207 const char __user *_data,
208 size_t datalen,
209 loff_t *pos)
210{
211 const struct cachefiles_daemon_cmd *cmd;
212 struct cachefiles_cache *cache = file->private_data;
213 ssize_t ret;
214 char *data, *args, *cp;
215
216 //_enter(",,%zu,", datalen);
217
218 ASSERT(cache);
219
220 if (test_bit(CACHEFILES_DEAD, &cache->flags))
221 return -EIO;
222
223 if (datalen < 0 || datalen > PAGE_SIZE - 1)
224 return -EOPNOTSUPP;
225
226 /* drag the command string into the kernel so we can parse it */
227 data = kmalloc(datalen + 1, GFP_KERNEL);
228 if (!data)
229 return -ENOMEM;
230
231 ret = -EFAULT;
232 if (copy_from_user(data, _data, datalen) != 0)
233 goto error;
234
235 data[datalen] = '\0';
236
237 ret = -EINVAL;
238 if (memchr(data, '\0', datalen))
239 goto error;
240
241 /* strip any newline */
242 cp = memchr(data, '\n', datalen);
243 if (cp) {
244 if (cp == data)
245 goto error;
246
247 *cp = '\0';
248 }
249
250 /* parse the command */
251 ret = -EOPNOTSUPP;
252
253 for (args = data; *args; args++)
254 if (isspace(*args))
255 break;
256 if (*args) {
257 if (args == data)
258 goto error;
259 *args = '\0';
260 for (args++; isspace(*args); args++)
261 continue;
262 }
263
264 /* run the appropriate command handler */
265 for (cmd = cachefiles_daemon_cmds; cmd->name[0]; cmd++)
266 if (strcmp(cmd->name, data) == 0)
267 goto found_command;
268
269error:
270 kfree(data);
271 //_leave(" = %zd", ret);
272 return ret;
273
274found_command:
275 mutex_lock(&cache->daemon_mutex);
276
277 ret = -EIO;
278 if (!test_bit(CACHEFILES_DEAD, &cache->flags))
279 ret = cmd->handler(cache, args);
280
281 mutex_unlock(&cache->daemon_mutex);
282
283 if (ret == 0)
284 ret = datalen;
285 goto error;
286}
287
288/*
289 * poll for culling state
290 * - use POLLOUT to indicate culling state
291 */
292static unsigned int cachefiles_daemon_poll(struct file *file,
293 struct poll_table_struct *poll)
294{
295 struct cachefiles_cache *cache = file->private_data;
296 unsigned int mask;
297
298 poll_wait(file, &cache->daemon_pollwq, poll);
299 mask = 0;
300
301 if (test_bit(CACHEFILES_STATE_CHANGED, &cache->flags))
302 mask |= POLLIN;
303
304 if (test_bit(CACHEFILES_CULLING, &cache->flags))
305 mask |= POLLOUT;
306
307 return mask;
308}
309
310/*
311 * give a range error for cache space constraints
312 * - can be tail-called
313 */
314static int cachefiles_daemon_range_error(struct cachefiles_cache *cache,
315 char *args)
316{
317 kerror("Free space limits must be in range"
318 " 0%%<=stop<cull<run<100%%");
319
320 return -EINVAL;
321}
322
323/*
324 * set the percentage of files at which to stop culling
325 * - command: "frun <N>%"
326 */
327static int cachefiles_daemon_frun(struct cachefiles_cache *cache, char *args)
328{
329 unsigned long frun;
330
331 _enter(",%s", args);
332
333 if (!*args)
334 return -EINVAL;
335
336 frun = simple_strtoul(args, &args, 10);
337 if (args[0] != '%' || args[1] != '\0')
338 return -EINVAL;
339
340 if (frun <= cache->fcull_percent || frun >= 100)
341 return cachefiles_daemon_range_error(cache, args);
342
343 cache->frun_percent = frun;
344 return 0;
345}
346
347/*
348 * set the percentage of files at which to start culling
349 * - command: "fcull <N>%"
350 */
351static int cachefiles_daemon_fcull(struct cachefiles_cache *cache, char *args)
352{
353 unsigned long fcull;
354
355 _enter(",%s", args);
356
357 if (!*args)
358 return -EINVAL;
359
360 fcull = simple_strtoul(args, &args, 10);
361 if (args[0] != '%' || args[1] != '\0')
362 return -EINVAL;
363
364 if (fcull <= cache->fstop_percent || fcull >= cache->frun_percent)
365 return cachefiles_daemon_range_error(cache, args);
366
367 cache->fcull_percent = fcull;
368 return 0;
369}
370
371/*
372 * set the percentage of files at which to stop allocating
373 * - command: "fstop <N>%"
374 */
375static int cachefiles_daemon_fstop(struct cachefiles_cache *cache, char *args)
376{
377 unsigned long fstop;
378
379 _enter(",%s", args);
380
381 if (!*args)
382 return -EINVAL;
383
384 fstop = simple_strtoul(args, &args, 10);
385 if (args[0] != '%' || args[1] != '\0')
386 return -EINVAL;
387
388 if (fstop < 0 || fstop >= cache->fcull_percent)
389 return cachefiles_daemon_range_error(cache, args);
390
391 cache->fstop_percent = fstop;
392 return 0;
393}
394
395/*
396 * set the percentage of blocks at which to stop culling
397 * - command: "brun <N>%"
398 */
399static int cachefiles_daemon_brun(struct cachefiles_cache *cache, char *args)
400{
401 unsigned long brun;
402
403 _enter(",%s", args);
404
405 if (!*args)
406 return -EINVAL;
407
408 brun = simple_strtoul(args, &args, 10);
409 if (args[0] != '%' || args[1] != '\0')
410 return -EINVAL;
411
412 if (brun <= cache->bcull_percent || brun >= 100)
413 return cachefiles_daemon_range_error(cache, args);
414
415 cache->brun_percent = brun;
416 return 0;
417}
418
419/*
420 * set the percentage of blocks at which to start culling
421 * - command: "bcull <N>%"
422 */
423static int cachefiles_daemon_bcull(struct cachefiles_cache *cache, char *args)
424{
425 unsigned long bcull;
426
427 _enter(",%s", args);
428
429 if (!*args)
430 return -EINVAL;
431
432 bcull = simple_strtoul(args, &args, 10);
433 if (args[0] != '%' || args[1] != '\0')
434 return -EINVAL;
435
436 if (bcull <= cache->bstop_percent || bcull >= cache->brun_percent)
437 return cachefiles_daemon_range_error(cache, args);
438
439 cache->bcull_percent = bcull;
440 return 0;
441}
442
443/*
444 * set the percentage of blocks at which to stop allocating
445 * - command: "bstop <N>%"
446 */
447static int cachefiles_daemon_bstop(struct cachefiles_cache *cache, char *args)
448{
449 unsigned long bstop;
450
451 _enter(",%s", args);
452
453 if (!*args)
454 return -EINVAL;
455
456 bstop = simple_strtoul(args, &args, 10);
457 if (args[0] != '%' || args[1] != '\0')
458 return -EINVAL;
459
460 if (bstop < 0 || bstop >= cache->bcull_percent)
461 return cachefiles_daemon_range_error(cache, args);
462
463 cache->bstop_percent = bstop;
464 return 0;
465}
466
467/*
468 * set the cache directory
469 * - command: "dir <name>"
470 */
471static int cachefiles_daemon_dir(struct cachefiles_cache *cache, char *args)
472{
473 char *dir;
474
475 _enter(",%s", args);
476
477 if (!*args) {
478 kerror("Empty directory specified");
479 return -EINVAL;
480 }
481
482 if (cache->rootdirname) {
483 kerror("Second cache directory specified");
484 return -EEXIST;
485 }
486
487 dir = kstrdup(args, GFP_KERNEL);
488 if (!dir)
489 return -ENOMEM;
490
491 cache->rootdirname = dir;
492 return 0;
493}
494
495/*
496 * set the cache security context
497 * - command: "secctx <ctx>"
498 */
499static int cachefiles_daemon_secctx(struct cachefiles_cache *cache, char *args)
500{
501 char *secctx;
502
503 _enter(",%s", args);
504
505 if (!*args) {
506 kerror("Empty security context specified");
507 return -EINVAL;
508 }
509
510 if (cache->secctx) {
511 kerror("Second security context specified");
512 return -EINVAL;
513 }
514
515 secctx = kstrdup(args, GFP_KERNEL);
516 if (!secctx)
517 return -ENOMEM;
518
519 cache->secctx = secctx;
520 return 0;
521}
522
523/*
524 * set the cache tag
525 * - command: "tag <name>"
526 */
527static int cachefiles_daemon_tag(struct cachefiles_cache *cache, char *args)
528{
529 char *tag;
530
531 _enter(",%s", args);
532
533 if (!*args) {
534 kerror("Empty tag specified");
535 return -EINVAL;
536 }
537
538 if (cache->tag)
539 return -EEXIST;
540
541 tag = kstrdup(args, GFP_KERNEL);
542 if (!tag)
543 return -ENOMEM;
544
545 cache->tag = tag;
546 return 0;
547}
548
549/*
550 * request a node in the cache be culled from the current working directory
551 * - command: "cull <name>"
552 */
553static int cachefiles_daemon_cull(struct cachefiles_cache *cache, char *args)
554{
555 struct fs_struct *fs;
556 struct dentry *dir;
557 const struct cred *saved_cred;
558 int ret;
559
560 _enter(",%s", args);
561
562 if (strchr(args, '/'))
563 goto inval;
564
565 if (!test_bit(CACHEFILES_READY, &cache->flags)) {
566 kerror("cull applied to unready cache");
567 return -EIO;
568 }
569
570 if (test_bit(CACHEFILES_DEAD, &cache->flags)) {
571 kerror("cull applied to dead cache");
572 return -EIO;
573 }
574
575 /* extract the directory dentry from the cwd */
576 fs = current->fs;
577 read_lock(&fs->lock);
578 dir = dget(fs->pwd.dentry);
579 read_unlock(&fs->lock);
580
581 if (!S_ISDIR(dir->d_inode->i_mode))
582 goto notdir;
583
584 cachefiles_begin_secure(cache, &saved_cred);
585 ret = cachefiles_cull(cache, dir, args);
586 cachefiles_end_secure(cache, saved_cred);
587
588 dput(dir);
589 _leave(" = %d", ret);
590 return ret;
591
592notdir:
593 dput(dir);
594 kerror("cull command requires dirfd to be a directory");
595 return -ENOTDIR;
596
597inval:
598 kerror("cull command requires dirfd and filename");
599 return -EINVAL;
600}
601
602/*
603 * set debugging mode
604 * - command: "debug <mask>"
605 */
606static int cachefiles_daemon_debug(struct cachefiles_cache *cache, char *args)
607{
608 unsigned long mask;
609
610 _enter(",%s", args);
611
612 mask = simple_strtoul(args, &args, 0);
613 if (args[0] != '\0')
614 goto inval;
615
616 cachefiles_debug = mask;
617 _leave(" = 0");
618 return 0;
619
620inval:
621 kerror("debug command requires mask");
622 return -EINVAL;
623}
624
625/*
626 * find out whether an object in the current working directory is in use or not
627 * - command: "inuse <name>"
628 */
629static int cachefiles_daemon_inuse(struct cachefiles_cache *cache, char *args)
630{
631 struct fs_struct *fs;
632 struct dentry *dir;
633 const struct cred *saved_cred;
634 int ret;
635
636 //_enter(",%s", args);
637
638 if (strchr(args, '/'))
639 goto inval;
640
641 if (!test_bit(CACHEFILES_READY, &cache->flags)) {
642 kerror("inuse applied to unready cache");
643 return -EIO;
644 }
645
646 if (test_bit(CACHEFILES_DEAD, &cache->flags)) {
647 kerror("inuse applied to dead cache");
648 return -EIO;
649 }
650
651 /* extract the directory dentry from the cwd */
652 fs = current->fs;
653 read_lock(&fs->lock);
654 dir = dget(fs->pwd.dentry);
655 read_unlock(&fs->lock);
656
657 if (!S_ISDIR(dir->d_inode->i_mode))
658 goto notdir;
659
660 cachefiles_begin_secure(cache, &saved_cred);
661 ret = cachefiles_check_in_use(cache, dir, args);
662 cachefiles_end_secure(cache, saved_cred);
663
664 dput(dir);
665 //_leave(" = %d", ret);
666 return ret;
667
668notdir:
669 dput(dir);
670 kerror("inuse command requires dirfd to be a directory");
671 return -ENOTDIR;
672
673inval:
674 kerror("inuse command requires dirfd and filename");
675 return -EINVAL;
676}
677
678/*
679 * see if we have space for a number of pages and/or a number of files in the
680 * cache
681 */
682int cachefiles_has_space(struct cachefiles_cache *cache,
683 unsigned fnr, unsigned bnr)
684{
685 struct kstatfs stats;
686 int ret;
687
688 //_enter("{%llu,%llu,%llu,%llu,%llu,%llu},%u,%u",
689 // (unsigned long long) cache->frun,
690 // (unsigned long long) cache->fcull,
691 // (unsigned long long) cache->fstop,
692 // (unsigned long long) cache->brun,
693 // (unsigned long long) cache->bcull,
694 // (unsigned long long) cache->bstop,
695 // fnr, bnr);
696
697 /* find out how many pages of blockdev are available */
698 memset(&stats, 0, sizeof(stats));
699
700 ret = vfs_statfs(cache->mnt->mnt_root, &stats);
701 if (ret < 0) {
702 if (ret == -EIO)
703 cachefiles_io_error(cache, "statfs failed");
704 _leave(" = %d", ret);
705 return ret;
706 }
707
708 stats.f_bavail >>= cache->bshift;
709
710 //_debug("avail %llu,%llu",
711 // (unsigned long long) stats.f_ffree,
712 // (unsigned long long) stats.f_bavail);
713
714 /* see if there is sufficient space */
715 if (stats.f_ffree > fnr)
716 stats.f_ffree -= fnr;
717 else
718 stats.f_ffree = 0;
719
720 if (stats.f_bavail > bnr)
721 stats.f_bavail -= bnr;
722 else
723 stats.f_bavail = 0;
724
725 ret = -ENOBUFS;
726 if (stats.f_ffree < cache->fstop ||
727 stats.f_bavail < cache->bstop)
728 goto begin_cull;
729
730 ret = 0;
731 if (stats.f_ffree < cache->fcull ||
732 stats.f_bavail < cache->bcull)
733 goto begin_cull;
734
735 if (test_bit(CACHEFILES_CULLING, &cache->flags) &&
736 stats.f_ffree >= cache->frun &&
737 stats.f_bavail >= cache->brun &&
738 test_and_clear_bit(CACHEFILES_CULLING, &cache->flags)
739 ) {
740 _debug("cease culling");
741 cachefiles_state_changed(cache);
742 }
743
744 //_leave(" = 0");
745 return 0;
746
747begin_cull:
748 if (!test_and_set_bit(CACHEFILES_CULLING, &cache->flags)) {
749 _debug("### CULL CACHE ###");
750 cachefiles_state_changed(cache);
751 }
752
753 _leave(" = %d", ret);
754 return ret;
755}
diff --git a/fs/cachefiles/interface.c b/fs/cachefiles/interface.c
new file mode 100644
index 000000000000..1e962348d111
--- /dev/null
+++ b/fs/cachefiles/interface.c
@@ -0,0 +1,449 @@
1/* FS-Cache interface to CacheFiles
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/mount.h>
13#include <linux/buffer_head.h>
14#include "internal.h"
15
16#define list_to_page(head) (list_entry((head)->prev, struct page, lru))
17
18struct cachefiles_lookup_data {
19 struct cachefiles_xattr *auxdata; /* auxiliary data */
20 char *key; /* key path */
21};
22
23static int cachefiles_attr_changed(struct fscache_object *_object);
24
25/*
26 * allocate an object record for a cookie lookup and prepare the lookup data
27 */
28static struct fscache_object *cachefiles_alloc_object(
29 struct fscache_cache *_cache,
30 struct fscache_cookie *cookie)
31{
32 struct cachefiles_lookup_data *lookup_data;
33 struct cachefiles_object *object;
34 struct cachefiles_cache *cache;
35 struct cachefiles_xattr *auxdata;
36 unsigned keylen, auxlen;
37 void *buffer;
38 char *key;
39
40 cache = container_of(_cache, struct cachefiles_cache, cache);
41
42 _enter("{%s},%p,", cache->cache.identifier, cookie);
43
44 lookup_data = kmalloc(sizeof(*lookup_data), GFP_KERNEL);
45 if (!lookup_data)
46 goto nomem_lookup_data;
47
48 /* create a new object record and a temporary leaf image */
49 object = kmem_cache_alloc(cachefiles_object_jar, GFP_KERNEL);
50 if (!object)
51 goto nomem_object;
52
53 ASSERTCMP(object->backer, ==, NULL);
54
55 BUG_ON(test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags));
56 atomic_set(&object->usage, 1);
57
58 fscache_object_init(&object->fscache, cookie, &cache->cache);
59
60 object->type = cookie->def->type;
61
62 /* get hold of the raw key
63 * - stick the length on the front and leave space on the back for the
64 * encoder
65 */
66 buffer = kmalloc((2 + 512) + 3, GFP_KERNEL);
67 if (!buffer)
68 goto nomem_buffer;
69
70 keylen = cookie->def->get_key(cookie->netfs_data, buffer + 2, 512);
71 ASSERTCMP(keylen, <, 512);
72
73 *(uint16_t *)buffer = keylen;
74 ((char *)buffer)[keylen + 2] = 0;
75 ((char *)buffer)[keylen + 3] = 0;
76 ((char *)buffer)[keylen + 4] = 0;
77
78 /* turn the raw key into something that can work with as a filename */
79 key = cachefiles_cook_key(buffer, keylen + 2, object->type);
80 if (!key)
81 goto nomem_key;
82
83 /* get hold of the auxiliary data and prepend the object type */
84 auxdata = buffer;
85 auxlen = 0;
86 if (cookie->def->get_aux) {
87 auxlen = cookie->def->get_aux(cookie->netfs_data,
88 auxdata->data, 511);
89 ASSERTCMP(auxlen, <, 511);
90 }
91
92 auxdata->len = auxlen + 1;
93 auxdata->type = cookie->def->type;
94
95 lookup_data->auxdata = auxdata;
96 lookup_data->key = key;
97 object->lookup_data = lookup_data;
98
99 _leave(" = %p [%p]", &object->fscache, lookup_data);
100 return &object->fscache;
101
102nomem_key:
103 kfree(buffer);
104nomem_buffer:
105 BUG_ON(test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags));
106 kmem_cache_free(cachefiles_object_jar, object);
107 fscache_object_destroyed(&cache->cache);
108nomem_object:
109 kfree(lookup_data);
110nomem_lookup_data:
111 _leave(" = -ENOMEM");
112 return ERR_PTR(-ENOMEM);
113}
114
115/*
116 * attempt to look up the nominated node in this cache
117 */
118static void cachefiles_lookup_object(struct fscache_object *_object)
119{
120 struct cachefiles_lookup_data *lookup_data;
121 struct cachefiles_object *parent, *object;
122 struct cachefiles_cache *cache;
123 const struct cred *saved_cred;
124 int ret;
125
126 _enter("{OBJ%x}", _object->debug_id);
127
128 cache = container_of(_object->cache, struct cachefiles_cache, cache);
129 parent = container_of(_object->parent,
130 struct cachefiles_object, fscache);
131 object = container_of(_object, struct cachefiles_object, fscache);
132 lookup_data = object->lookup_data;
133
134 ASSERTCMP(lookup_data, !=, NULL);
135
136 /* look up the key, creating any missing bits */
137 cachefiles_begin_secure(cache, &saved_cred);
138 ret = cachefiles_walk_to_object(parent, object,
139 lookup_data->key,
140 lookup_data->auxdata);
141 cachefiles_end_secure(cache, saved_cred);
142
143 /* polish off by setting the attributes of non-index files */
144 if (ret == 0 &&
145 object->fscache.cookie->def->type != FSCACHE_COOKIE_TYPE_INDEX)
146 cachefiles_attr_changed(&object->fscache);
147
148 if (ret < 0) {
149 printk(KERN_WARNING "CacheFiles: Lookup failed error %d\n",
150 ret);
151 fscache_object_lookup_error(&object->fscache);
152 }
153
154 _leave(" [%d]", ret);
155}
156
157/*
158 * indication of lookup completion
159 */
160static void cachefiles_lookup_complete(struct fscache_object *_object)
161{
162 struct cachefiles_object *object;
163
164 object = container_of(_object, struct cachefiles_object, fscache);
165
166 _enter("{OBJ%x,%p}", object->fscache.debug_id, object->lookup_data);
167
168 if (object->lookup_data) {
169 kfree(object->lookup_data->key);
170 kfree(object->lookup_data->auxdata);
171 kfree(object->lookup_data);
172 object->lookup_data = NULL;
173 }
174}
175
176/*
177 * increment the usage count on an inode object (may fail if unmounting)
178 */
179static
180struct fscache_object *cachefiles_grab_object(struct fscache_object *_object)
181{
182 struct cachefiles_object *object =
183 container_of(_object, struct cachefiles_object, fscache);
184
185 _enter("{OBJ%x,%d}", _object->debug_id, atomic_read(&object->usage));
186
187#ifdef CACHEFILES_DEBUG_SLAB
188 ASSERT((atomic_read(&object->usage) & 0xffff0000) != 0x6b6b0000);
189#endif
190
191 atomic_inc(&object->usage);
192 return &object->fscache;
193}
194
195/*
196 * update the auxilliary data for an object object on disk
197 */
198static void cachefiles_update_object(struct fscache_object *_object)
199{
200 struct cachefiles_object *object;
201 struct cachefiles_xattr *auxdata;
202 struct cachefiles_cache *cache;
203 struct fscache_cookie *cookie;
204 const struct cred *saved_cred;
205 unsigned auxlen;
206
207 _enter("{OBJ%x}", _object->debug_id);
208
209 object = container_of(_object, struct cachefiles_object, fscache);
210 cache = container_of(object->fscache.cache, struct cachefiles_cache,
211 cache);
212 cookie = object->fscache.cookie;
213
214 if (!cookie->def->get_aux) {
215 _leave(" [no aux]");
216 return;
217 }
218
219 auxdata = kmalloc(2 + 512 + 3, GFP_KERNEL);
220 if (!auxdata) {
221 _leave(" [nomem]");
222 return;
223 }
224
225 auxlen = cookie->def->get_aux(cookie->netfs_data, auxdata->data, 511);
226 ASSERTCMP(auxlen, <, 511);
227
228 auxdata->len = auxlen + 1;
229 auxdata->type = cookie->def->type;
230
231 cachefiles_begin_secure(cache, &saved_cred);
232 cachefiles_update_object_xattr(object, auxdata);
233 cachefiles_end_secure(cache, saved_cred);
234 kfree(auxdata);
235 _leave("");
236}
237
238/*
239 * discard the resources pinned by an object and effect retirement if
240 * requested
241 */
242static void cachefiles_drop_object(struct fscache_object *_object)
243{
244 struct cachefiles_object *object;
245 struct cachefiles_cache *cache;
246 const struct cred *saved_cred;
247
248 ASSERT(_object);
249
250 object = container_of(_object, struct cachefiles_object, fscache);
251
252 _enter("{OBJ%x,%d}",
253 object->fscache.debug_id, atomic_read(&object->usage));
254
255 cache = container_of(object->fscache.cache,
256 struct cachefiles_cache, cache);
257
258#ifdef CACHEFILES_DEBUG_SLAB
259 ASSERT((atomic_read(&object->usage) & 0xffff0000) != 0x6b6b0000);
260#endif
261
262 /* delete retired objects */
263 if (object->fscache.state == FSCACHE_OBJECT_RECYCLING &&
264 _object != cache->cache.fsdef
265 ) {
266 _debug("- retire object OBJ%x", object->fscache.debug_id);
267 cachefiles_begin_secure(cache, &saved_cred);
268 cachefiles_delete_object(cache, object);
269 cachefiles_end_secure(cache, saved_cred);
270 }
271
272 /* close the filesystem stuff attached to the object */
273 if (object->backer != object->dentry)
274 dput(object->backer);
275 object->backer = NULL;
276
277 /* note that the object is now inactive */
278 if (test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags)) {
279 write_lock(&cache->active_lock);
280 if (!test_and_clear_bit(CACHEFILES_OBJECT_ACTIVE,
281 &object->flags))
282 BUG();
283 rb_erase(&object->active_node, &cache->active_nodes);
284 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
285 write_unlock(&cache->active_lock);
286 }
287
288 dput(object->dentry);
289 object->dentry = NULL;
290
291 _leave("");
292}
293
294/*
295 * dispose of a reference to an object
296 */
297static void cachefiles_put_object(struct fscache_object *_object)
298{
299 struct cachefiles_object *object;
300 struct fscache_cache *cache;
301
302 ASSERT(_object);
303
304 object = container_of(_object, struct cachefiles_object, fscache);
305
306 _enter("{OBJ%x,%d}",
307 object->fscache.debug_id, atomic_read(&object->usage));
308
309#ifdef CACHEFILES_DEBUG_SLAB
310 ASSERT((atomic_read(&object->usage) & 0xffff0000) != 0x6b6b0000);
311#endif
312
313 ASSERTIFCMP(object->fscache.parent,
314 object->fscache.parent->n_children, >, 0);
315
316 if (atomic_dec_and_test(&object->usage)) {
317 _debug("- kill object OBJ%x", object->fscache.debug_id);
318
319 ASSERT(!test_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags));
320 ASSERTCMP(object->fscache.parent, ==, NULL);
321 ASSERTCMP(object->backer, ==, NULL);
322 ASSERTCMP(object->dentry, ==, NULL);
323 ASSERTCMP(object->fscache.n_ops, ==, 0);
324 ASSERTCMP(object->fscache.n_children, ==, 0);
325
326 if (object->lookup_data) {
327 kfree(object->lookup_data->key);
328 kfree(object->lookup_data->auxdata);
329 kfree(object->lookup_data);
330 object->lookup_data = NULL;
331 }
332
333 cache = object->fscache.cache;
334 kmem_cache_free(cachefiles_object_jar, object);
335 fscache_object_destroyed(cache);
336 }
337
338 _leave("");
339}
340
341/*
342 * sync a cache
343 */
344static void cachefiles_sync_cache(struct fscache_cache *_cache)
345{
346 struct cachefiles_cache *cache;
347 const struct cred *saved_cred;
348 int ret;
349
350 _enter("%p", _cache);
351
352 cache = container_of(_cache, struct cachefiles_cache, cache);
353
354 /* make sure all pages pinned by operations on behalf of the netfs are
355 * written to disc */
356 cachefiles_begin_secure(cache, &saved_cred);
357 ret = fsync_super(cache->mnt->mnt_sb);
358 cachefiles_end_secure(cache, saved_cred);
359
360 if (ret == -EIO)
361 cachefiles_io_error(cache,
362 "Attempt to sync backing fs superblock"
363 " returned error %d",
364 ret);
365}
366
367/*
368 * notification the attributes on an object have changed
369 * - called with reads/writes excluded by FS-Cache
370 */
371static int cachefiles_attr_changed(struct fscache_object *_object)
372{
373 struct cachefiles_object *object;
374 struct cachefiles_cache *cache;
375 const struct cred *saved_cred;
376 struct iattr newattrs;
377 uint64_t ni_size;
378 loff_t oi_size;
379 int ret;
380
381 _object->cookie->def->get_attr(_object->cookie->netfs_data, &ni_size);
382
383 _enter("{OBJ%x},[%llu]",
384 _object->debug_id, (unsigned long long) ni_size);
385
386 object = container_of(_object, struct cachefiles_object, fscache);
387 cache = container_of(object->fscache.cache,
388 struct cachefiles_cache, cache);
389
390 if (ni_size == object->i_size)
391 return 0;
392
393 if (!object->backer)
394 return -ENOBUFS;
395
396 ASSERT(S_ISREG(object->backer->d_inode->i_mode));
397
398 fscache_set_store_limit(&object->fscache, ni_size);
399
400 oi_size = i_size_read(object->backer->d_inode);
401 if (oi_size == ni_size)
402 return 0;
403
404 newattrs.ia_size = ni_size;
405 newattrs.ia_valid = ATTR_SIZE;
406
407 cachefiles_begin_secure(cache, &saved_cred);
408 mutex_lock(&object->backer->d_inode->i_mutex);
409 ret = notify_change(object->backer, &newattrs);
410 mutex_unlock(&object->backer->d_inode->i_mutex);
411 cachefiles_end_secure(cache, saved_cred);
412
413 if (ret == -EIO) {
414 fscache_set_store_limit(&object->fscache, 0);
415 cachefiles_io_error_obj(object, "Size set failed");
416 ret = -ENOBUFS;
417 }
418
419 _leave(" = %d", ret);
420 return ret;
421}
422
423/*
424 * dissociate a cache from all the pages it was backing
425 */
426static void cachefiles_dissociate_pages(struct fscache_cache *cache)
427{
428 _enter("");
429}
430
431const struct fscache_cache_ops cachefiles_cache_ops = {
432 .name = "cachefiles",
433 .alloc_object = cachefiles_alloc_object,
434 .lookup_object = cachefiles_lookup_object,
435 .lookup_complete = cachefiles_lookup_complete,
436 .grab_object = cachefiles_grab_object,
437 .update_object = cachefiles_update_object,
438 .drop_object = cachefiles_drop_object,
439 .put_object = cachefiles_put_object,
440 .sync_cache = cachefiles_sync_cache,
441 .attr_changed = cachefiles_attr_changed,
442 .read_or_alloc_page = cachefiles_read_or_alloc_page,
443 .read_or_alloc_pages = cachefiles_read_or_alloc_pages,
444 .allocate_page = cachefiles_allocate_page,
445 .allocate_pages = cachefiles_allocate_pages,
446 .write_page = cachefiles_write_page,
447 .uncache_page = cachefiles_uncache_page,
448 .dissociate_pages = cachefiles_dissociate_pages,
449};
diff --git a/fs/cachefiles/internal.h b/fs/cachefiles/internal.h
new file mode 100644
index 000000000000..19218e1463d6
--- /dev/null
+++ b/fs/cachefiles/internal.h
@@ -0,0 +1,360 @@
1/* General netfs cache on cache files internal defs
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/fscache-cache.h>
13#include <linux/timer.h>
14#include <linux/wait.h>
15#include <linux/workqueue.h>
16#include <linux/security.h>
17
18struct cachefiles_cache;
19struct cachefiles_object;
20
21extern unsigned cachefiles_debug;
22#define CACHEFILES_DEBUG_KENTER 1
23#define CACHEFILES_DEBUG_KLEAVE 2
24#define CACHEFILES_DEBUG_KDEBUG 4
25
26/*
27 * node records
28 */
29struct cachefiles_object {
30 struct fscache_object fscache; /* fscache handle */
31 struct cachefiles_lookup_data *lookup_data; /* cached lookup data */
32 struct dentry *dentry; /* the file/dir representing this object */
33 struct dentry *backer; /* backing file */
34 loff_t i_size; /* object size */
35 unsigned long flags;
36#define CACHEFILES_OBJECT_ACTIVE 0 /* T if marked active */
37 atomic_t usage; /* object usage count */
38 uint8_t type; /* object type */
39 uint8_t new; /* T if object new */
40 spinlock_t work_lock;
41 struct rb_node active_node; /* link in active tree (dentry is key) */
42};
43
44extern struct kmem_cache *cachefiles_object_jar;
45
46/*
47 * Cache files cache definition
48 */
49struct cachefiles_cache {
50 struct fscache_cache cache; /* FS-Cache record */
51 struct vfsmount *mnt; /* mountpoint holding the cache */
52 struct dentry *graveyard; /* directory into which dead objects go */
53 struct file *cachefilesd; /* manager daemon handle */
54 const struct cred *cache_cred; /* security override for accessing cache */
55 struct mutex daemon_mutex; /* command serialisation mutex */
56 wait_queue_head_t daemon_pollwq; /* poll waitqueue for daemon */
57 struct rb_root active_nodes; /* active nodes (can't be culled) */
58 rwlock_t active_lock; /* lock for active_nodes */
59 atomic_t gravecounter; /* graveyard uniquifier */
60 unsigned frun_percent; /* when to stop culling (% files) */
61 unsigned fcull_percent; /* when to start culling (% files) */
62 unsigned fstop_percent; /* when to stop allocating (% files) */
63 unsigned brun_percent; /* when to stop culling (% blocks) */
64 unsigned bcull_percent; /* when to start culling (% blocks) */
65 unsigned bstop_percent; /* when to stop allocating (% blocks) */
66 unsigned bsize; /* cache's block size */
67 unsigned bshift; /* min(ilog2(PAGE_SIZE / bsize), 0) */
68 uint64_t frun; /* when to stop culling */
69 uint64_t fcull; /* when to start culling */
70 uint64_t fstop; /* when to stop allocating */
71 sector_t brun; /* when to stop culling */
72 sector_t bcull; /* when to start culling */
73 sector_t bstop; /* when to stop allocating */
74 unsigned long flags;
75#define CACHEFILES_READY 0 /* T if cache prepared */
76#define CACHEFILES_DEAD 1 /* T if cache dead */
77#define CACHEFILES_CULLING 2 /* T if cull engaged */
78#define CACHEFILES_STATE_CHANGED 3 /* T if state changed (poll trigger) */
79 char *rootdirname; /* name of cache root directory */
80 char *secctx; /* LSM security context */
81 char *tag; /* cache binding tag */
82};
83
84/*
85 * backing file read tracking
86 */
87struct cachefiles_one_read {
88 wait_queue_t monitor; /* link into monitored waitqueue */
89 struct page *back_page; /* backing file page we're waiting for */
90 struct page *netfs_page; /* netfs page we're going to fill */
91 struct fscache_retrieval *op; /* retrieval op covering this */
92 struct list_head op_link; /* link in op's todo list */
93};
94
95/*
96 * backing file write tracking
97 */
98struct cachefiles_one_write {
99 struct page *netfs_page; /* netfs page to copy */
100 struct cachefiles_object *object;
101 struct list_head obj_link; /* link in object's lists */
102 fscache_rw_complete_t end_io_func;
103 void *context;
104};
105
106/*
107 * auxiliary data xattr buffer
108 */
109struct cachefiles_xattr {
110 uint16_t len;
111 uint8_t type;
112 uint8_t data[];
113};
114
115/*
116 * note change of state for daemon
117 */
118static inline void cachefiles_state_changed(struct cachefiles_cache *cache)
119{
120 set_bit(CACHEFILES_STATE_CHANGED, &cache->flags);
121 wake_up_all(&cache->daemon_pollwq);
122}
123
124/*
125 * cf-bind.c
126 */
127extern int cachefiles_daemon_bind(struct cachefiles_cache *cache, char *args);
128extern void cachefiles_daemon_unbind(struct cachefiles_cache *cache);
129
130/*
131 * cf-daemon.c
132 */
133extern const struct file_operations cachefiles_daemon_fops;
134
135extern int cachefiles_has_space(struct cachefiles_cache *cache,
136 unsigned fnr, unsigned bnr);
137
138/*
139 * cf-interface.c
140 */
141extern const struct fscache_cache_ops cachefiles_cache_ops;
142
143/*
144 * cf-key.c
145 */
146extern char *cachefiles_cook_key(const u8 *raw, int keylen, uint8_t type);
147
148/*
149 * cf-namei.c
150 */
151extern int cachefiles_delete_object(struct cachefiles_cache *cache,
152 struct cachefiles_object *object);
153extern int cachefiles_walk_to_object(struct cachefiles_object *parent,
154 struct cachefiles_object *object,
155 const char *key,
156 struct cachefiles_xattr *auxdata);
157extern struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
158 struct dentry *dir,
159 const char *name);
160
161extern int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
162 char *filename);
163
164extern int cachefiles_check_in_use(struct cachefiles_cache *cache,
165 struct dentry *dir, char *filename);
166
167/*
168 * cf-proc.c
169 */
170#ifdef CONFIG_CACHEFILES_HISTOGRAM
171extern atomic_t cachefiles_lookup_histogram[HZ];
172extern atomic_t cachefiles_mkdir_histogram[HZ];
173extern atomic_t cachefiles_create_histogram[HZ];
174
175extern int __init cachefiles_proc_init(void);
176extern void cachefiles_proc_cleanup(void);
177static inline
178void cachefiles_hist(atomic_t histogram[], unsigned long start_jif)
179{
180 unsigned long jif = jiffies - start_jif;
181 if (jif >= HZ)
182 jif = HZ - 1;
183 atomic_inc(&histogram[jif]);
184}
185
186#else
187#define cachefiles_proc_init() (0)
188#define cachefiles_proc_cleanup() do {} while (0)
189#define cachefiles_hist(hist, start_jif) do {} while (0)
190#endif
191
192/*
193 * cf-rdwr.c
194 */
195extern int cachefiles_read_or_alloc_page(struct fscache_retrieval *,
196 struct page *, gfp_t);
197extern int cachefiles_read_or_alloc_pages(struct fscache_retrieval *,
198 struct list_head *, unsigned *,
199 gfp_t);
200extern int cachefiles_allocate_page(struct fscache_retrieval *, struct page *,
201 gfp_t);
202extern int cachefiles_allocate_pages(struct fscache_retrieval *,
203 struct list_head *, unsigned *, gfp_t);
204extern int cachefiles_write_page(struct fscache_storage *, struct page *);
205extern void cachefiles_uncache_page(struct fscache_object *, struct page *);
206
207/*
208 * cf-security.c
209 */
210extern int cachefiles_get_security_ID(struct cachefiles_cache *cache);
211extern int cachefiles_determine_cache_security(struct cachefiles_cache *cache,
212 struct dentry *root,
213 const struct cred **_saved_cred);
214
215static inline void cachefiles_begin_secure(struct cachefiles_cache *cache,
216 const struct cred **_saved_cred)
217{
218 *_saved_cred = override_creds(cache->cache_cred);
219}
220
221static inline void cachefiles_end_secure(struct cachefiles_cache *cache,
222 const struct cred *saved_cred)
223{
224 revert_creds(saved_cred);
225}
226
227/*
228 * cf-xattr.c
229 */
230extern int cachefiles_check_object_type(struct cachefiles_object *object);
231extern int cachefiles_set_object_xattr(struct cachefiles_object *object,
232 struct cachefiles_xattr *auxdata);
233extern int cachefiles_update_object_xattr(struct cachefiles_object *object,
234 struct cachefiles_xattr *auxdata);
235extern int cachefiles_check_object_xattr(struct cachefiles_object *object,
236 struct cachefiles_xattr *auxdata);
237extern int cachefiles_remove_object_xattr(struct cachefiles_cache *cache,
238 struct dentry *dentry);
239
240
241/*
242 * error handling
243 */
244#define kerror(FMT, ...) printk(KERN_ERR "CacheFiles: "FMT"\n", ##__VA_ARGS__)
245
246#define cachefiles_io_error(___cache, FMT, ...) \
247do { \
248 kerror("I/O Error: " FMT, ##__VA_ARGS__); \
249 fscache_io_error(&(___cache)->cache); \
250 set_bit(CACHEFILES_DEAD, &(___cache)->flags); \
251} while (0)
252
253#define cachefiles_io_error_obj(object, FMT, ...) \
254do { \
255 struct cachefiles_cache *___cache; \
256 \
257 ___cache = container_of((object)->fscache.cache, \
258 struct cachefiles_cache, cache); \
259 cachefiles_io_error(___cache, FMT, ##__VA_ARGS__); \
260} while (0)
261
262
263/*
264 * debug tracing
265 */
266#define dbgprintk(FMT, ...) \
267 printk(KERN_DEBUG "[%-6.6s] "FMT"\n", current->comm, ##__VA_ARGS__)
268
269/* make sure we maintain the format strings, even when debugging is disabled */
270static inline void _dbprintk(const char *fmt, ...)
271 __attribute__((format(printf, 1, 2)));
272static inline void _dbprintk(const char *fmt, ...)
273{
274}
275
276#define kenter(FMT, ...) dbgprintk("==> %s("FMT")", __func__, ##__VA_ARGS__)
277#define kleave(FMT, ...) dbgprintk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
278#define kdebug(FMT, ...) dbgprintk(FMT, ##__VA_ARGS__)
279
280
281#if defined(__KDEBUG)
282#define _enter(FMT, ...) kenter(FMT, ##__VA_ARGS__)
283#define _leave(FMT, ...) kleave(FMT, ##__VA_ARGS__)
284#define _debug(FMT, ...) kdebug(FMT, ##__VA_ARGS__)
285
286#elif defined(CONFIG_CACHEFILES_DEBUG)
287#define _enter(FMT, ...) \
288do { \
289 if (cachefiles_debug & CACHEFILES_DEBUG_KENTER) \
290 kenter(FMT, ##__VA_ARGS__); \
291} while (0)
292
293#define _leave(FMT, ...) \
294do { \
295 if (cachefiles_debug & CACHEFILES_DEBUG_KLEAVE) \
296 kleave(FMT, ##__VA_ARGS__); \
297} while (0)
298
299#define _debug(FMT, ...) \
300do { \
301 if (cachefiles_debug & CACHEFILES_DEBUG_KDEBUG) \
302 kdebug(FMT, ##__VA_ARGS__); \
303} while (0)
304
305#else
306#define _enter(FMT, ...) _dbprintk("==> %s("FMT")", __func__, ##__VA_ARGS__)
307#define _leave(FMT, ...) _dbprintk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
308#define _debug(FMT, ...) _dbprintk(FMT, ##__VA_ARGS__)
309#endif
310
311#if 1 /* defined(__KDEBUGALL) */
312
313#define ASSERT(X) \
314do { \
315 if (unlikely(!(X))) { \
316 printk(KERN_ERR "\n"); \
317 printk(KERN_ERR "CacheFiles: Assertion failed\n"); \
318 BUG(); \
319 } \
320} while (0)
321
322#define ASSERTCMP(X, OP, Y) \
323do { \
324 if (unlikely(!((X) OP (Y)))) { \
325 printk(KERN_ERR "\n"); \
326 printk(KERN_ERR "CacheFiles: Assertion failed\n"); \
327 printk(KERN_ERR "%lx " #OP " %lx is false\n", \
328 (unsigned long)(X), (unsigned long)(Y)); \
329 BUG(); \
330 } \
331} while (0)
332
333#define ASSERTIF(C, X) \
334do { \
335 if (unlikely((C) && !(X))) { \
336 printk(KERN_ERR "\n"); \
337 printk(KERN_ERR "CacheFiles: Assertion failed\n"); \
338 BUG(); \
339 } \
340} while (0)
341
342#define ASSERTIFCMP(C, X, OP, Y) \
343do { \
344 if (unlikely((C) && !((X) OP (Y)))) { \
345 printk(KERN_ERR "\n"); \
346 printk(KERN_ERR "CacheFiles: Assertion failed\n"); \
347 printk(KERN_ERR "%lx " #OP " %lx is false\n", \
348 (unsigned long)(X), (unsigned long)(Y)); \
349 BUG(); \
350 } \
351} while (0)
352
353#else
354
355#define ASSERT(X) do {} while (0)
356#define ASSERTCMP(X, OP, Y) do {} while (0)
357#define ASSERTIF(C, X) do {} while (0)
358#define ASSERTIFCMP(C, X, OP, Y) do {} while (0)
359
360#endif
diff --git a/fs/cachefiles/key.c b/fs/cachefiles/key.c
new file mode 100644
index 000000000000..81b8b2b3a674
--- /dev/null
+++ b/fs/cachefiles/key.c
@@ -0,0 +1,159 @@
1/* Key to pathname encoder
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/slab.h>
13#include "internal.h"
14
15static const char cachefiles_charmap[64] =
16 "0123456789" /* 0 - 9 */
17 "abcdefghijklmnopqrstuvwxyz" /* 10 - 35 */
18 "ABCDEFGHIJKLMNOPQRSTUVWXYZ" /* 36 - 61 */
19 "_-" /* 62 - 63 */
20 ;
21
22static const char cachefiles_filecharmap[256] = {
23 /* we skip space and tab and control chars */
24 [33 ... 46] = 1, /* '!' -> '.' */
25 /* we skip '/' as it's significant to pathwalk */
26 [48 ... 127] = 1, /* '0' -> '~' */
27};
28
29/*
30 * turn the raw key into something cooked
31 * - the raw key should include the length in the two bytes at the front
32 * - the key may be up to 514 bytes in length (including the length word)
33 * - "base64" encode the strange keys, mapping 3 bytes of raw to four of
34 * cooked
35 * - need to cut the cooked key into 252 char lengths (189 raw bytes)
36 */
37char *cachefiles_cook_key(const u8 *raw, int keylen, uint8_t type)
38{
39 unsigned char csum, ch;
40 unsigned int acc;
41 char *key;
42 int loop, len, max, seg, mark, print;
43
44 _enter(",%d", keylen);
45
46 BUG_ON(keylen < 2 || keylen > 514);
47
48 csum = raw[0] + raw[1];
49 print = 1;
50 for (loop = 2; loop < keylen; loop++) {
51 ch = raw[loop];
52 csum += ch;
53 print &= cachefiles_filecharmap[ch];
54 }
55
56 if (print) {
57 /* if the path is usable ASCII, then we render it directly */
58 max = keylen - 2;
59 max += 2; /* two base64'd length chars on the front */
60 max += 5; /* @checksum/M */
61 max += 3 * 2; /* maximum number of segment dividers (".../M")
62 * is ((514 + 251) / 252) = 3
63 */
64 max += 1; /* NUL on end */
65 } else {
66 /* calculate the maximum length of the cooked key */
67 keylen = (keylen + 2) / 3;
68
69 max = keylen * 4;
70 max += 5; /* @checksum/M */
71 max += 3 * 2; /* maximum number of segment dividers (".../M")
72 * is ((514 + 188) / 189) = 3
73 */
74 max += 1; /* NUL on end */
75 }
76
77 max += 1; /* 2nd NUL on end */
78
79 _debug("max: %d", max);
80
81 key = kmalloc(max, GFP_KERNEL);
82 if (!key)
83 return NULL;
84
85 len = 0;
86
87 /* build the cooked key */
88 sprintf(key, "@%02x%c+", (unsigned) csum, 0);
89 len = 5;
90 mark = len - 1;
91
92 if (print) {
93 acc = *(uint16_t *) raw;
94 raw += 2;
95
96 key[len + 1] = cachefiles_charmap[acc & 63];
97 acc >>= 6;
98 key[len] = cachefiles_charmap[acc & 63];
99 len += 2;
100
101 seg = 250;
102 for (loop = keylen; loop > 0; loop--) {
103 if (seg <= 0) {
104 key[len++] = '\0';
105 mark = len;
106 key[len++] = '+';
107 seg = 252;
108 }
109
110 key[len++] = *raw++;
111 ASSERT(len < max);
112 }
113
114 switch (type) {
115 case FSCACHE_COOKIE_TYPE_INDEX: type = 'I'; break;
116 case FSCACHE_COOKIE_TYPE_DATAFILE: type = 'D'; break;
117 default: type = 'S'; break;
118 }
119 } else {
120 seg = 252;
121 for (loop = keylen; loop > 0; loop--) {
122 if (seg <= 0) {
123 key[len++] = '\0';
124 mark = len;
125 key[len++] = '+';
126 seg = 252;
127 }
128
129 acc = *raw++;
130 acc |= *raw++ << 8;
131 acc |= *raw++ << 16;
132
133 _debug("acc: %06x", acc);
134
135 key[len++] = cachefiles_charmap[acc & 63];
136 acc >>= 6;
137 key[len++] = cachefiles_charmap[acc & 63];
138 acc >>= 6;
139 key[len++] = cachefiles_charmap[acc & 63];
140 acc >>= 6;
141 key[len++] = cachefiles_charmap[acc & 63];
142
143 ASSERT(len < max);
144 }
145
146 switch (type) {
147 case FSCACHE_COOKIE_TYPE_INDEX: type = 'J'; break;
148 case FSCACHE_COOKIE_TYPE_DATAFILE: type = 'E'; break;
149 default: type = 'T'; break;
150 }
151 }
152
153 key[mark] = type;
154 key[len++] = 0;
155 key[len] = 0;
156
157 _leave(" = %p %d", key, len);
158 return key;
159}
diff --git a/fs/cachefiles/main.c b/fs/cachefiles/main.c
new file mode 100644
index 000000000000..4bfa8cf43bf5
--- /dev/null
+++ b/fs/cachefiles/main.c
@@ -0,0 +1,106 @@
1/* Network filesystem caching backend to use cache files on a premounted
2 * filesystem
3 *
4 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
5 * Written by David Howells (dhowells@redhat.com)
6 *
7 * This program is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU General Public Licence
9 * as published by the Free Software Foundation; either version
10 * 2 of the Licence, or (at your option) any later version.
11 */
12
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/completion.h>
17#include <linux/slab.h>
18#include <linux/fs.h>
19#include <linux/file.h>
20#include <linux/namei.h>
21#include <linux/mount.h>
22#include <linux/statfs.h>
23#include <linux/sysctl.h>
24#include <linux/miscdevice.h>
25#include "internal.h"
26
27unsigned cachefiles_debug;
28module_param_named(debug, cachefiles_debug, uint, S_IWUSR | S_IRUGO);
29MODULE_PARM_DESC(cachefiles_debug, "CacheFiles debugging mask");
30
31MODULE_DESCRIPTION("Mounted-filesystem based cache");
32MODULE_AUTHOR("Red Hat, Inc.");
33MODULE_LICENSE("GPL");
34
35struct kmem_cache *cachefiles_object_jar;
36
37static struct miscdevice cachefiles_dev = {
38 .minor = MISC_DYNAMIC_MINOR,
39 .name = "cachefiles",
40 .fops = &cachefiles_daemon_fops,
41};
42
43static void cachefiles_object_init_once(void *_object)
44{
45 struct cachefiles_object *object = _object;
46
47 memset(object, 0, sizeof(*object));
48 spin_lock_init(&object->work_lock);
49}
50
51/*
52 * initialise the fs caching module
53 */
54static int __init cachefiles_init(void)
55{
56 int ret;
57
58 ret = misc_register(&cachefiles_dev);
59 if (ret < 0)
60 goto error_dev;
61
62 /* create an object jar */
63 ret = -ENOMEM;
64 cachefiles_object_jar =
65 kmem_cache_create("cachefiles_object_jar",
66 sizeof(struct cachefiles_object),
67 0,
68 SLAB_HWCACHE_ALIGN,
69 cachefiles_object_init_once);
70 if (!cachefiles_object_jar) {
71 printk(KERN_NOTICE
72 "CacheFiles: Failed to allocate an object jar\n");
73 goto error_object_jar;
74 }
75
76 ret = cachefiles_proc_init();
77 if (ret < 0)
78 goto error_proc;
79
80 printk(KERN_INFO "CacheFiles: Loaded\n");
81 return 0;
82
83error_proc:
84 kmem_cache_destroy(cachefiles_object_jar);
85error_object_jar:
86 misc_deregister(&cachefiles_dev);
87error_dev:
88 kerror("failed to register: %d", ret);
89 return ret;
90}
91
92fs_initcall(cachefiles_init);
93
94/*
95 * clean up on module removal
96 */
97static void __exit cachefiles_exit(void)
98{
99 printk(KERN_INFO "CacheFiles: Unloading\n");
100
101 cachefiles_proc_cleanup();
102 kmem_cache_destroy(cachefiles_object_jar);
103 misc_deregister(&cachefiles_dev);
104}
105
106module_exit(cachefiles_exit);
diff --git a/fs/cachefiles/namei.c b/fs/cachefiles/namei.c
new file mode 100644
index 000000000000..4ce818ae39ea
--- /dev/null
+++ b/fs/cachefiles/namei.c
@@ -0,0 +1,771 @@
1/* CacheFiles path walking and related routines
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/module.h>
13#include <linux/sched.h>
14#include <linux/file.h>
15#include <linux/fs.h>
16#include <linux/fsnotify.h>
17#include <linux/quotaops.h>
18#include <linux/xattr.h>
19#include <linux/mount.h>
20#include <linux/namei.h>
21#include <linux/security.h>
22#include "internal.h"
23
24static int cachefiles_wait_bit(void *flags)
25{
26 schedule();
27 return 0;
28}
29
30/*
31 * record the fact that an object is now active
32 */
33static void cachefiles_mark_object_active(struct cachefiles_cache *cache,
34 struct cachefiles_object *object)
35{
36 struct cachefiles_object *xobject;
37 struct rb_node **_p, *_parent = NULL;
38 struct dentry *dentry;
39
40 _enter(",%p", object);
41
42try_again:
43 write_lock(&cache->active_lock);
44
45 if (test_and_set_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags))
46 BUG();
47
48 dentry = object->dentry;
49 _p = &cache->active_nodes.rb_node;
50 while (*_p) {
51 _parent = *_p;
52 xobject = rb_entry(_parent,
53 struct cachefiles_object, active_node);
54
55 ASSERT(xobject != object);
56
57 if (xobject->dentry > dentry)
58 _p = &(*_p)->rb_left;
59 else if (xobject->dentry < dentry)
60 _p = &(*_p)->rb_right;
61 else
62 goto wait_for_old_object;
63 }
64
65 rb_link_node(&object->active_node, _parent, _p);
66 rb_insert_color(&object->active_node, &cache->active_nodes);
67
68 write_unlock(&cache->active_lock);
69 _leave("");
70 return;
71
72 /* an old object from a previous incarnation is hogging the slot - we
73 * need to wait for it to be destroyed */
74wait_for_old_object:
75 if (xobject->fscache.state < FSCACHE_OBJECT_DYING) {
76 printk(KERN_ERR "\n");
77 printk(KERN_ERR "CacheFiles: Error:"
78 " Unexpected object collision\n");
79 printk(KERN_ERR "xobject: OBJ%x\n",
80 xobject->fscache.debug_id);
81 printk(KERN_ERR "xobjstate=%s\n",
82 fscache_object_states[xobject->fscache.state]);
83 printk(KERN_ERR "xobjflags=%lx\n", xobject->fscache.flags);
84 printk(KERN_ERR "xobjevent=%lx [%lx]\n",
85 xobject->fscache.events, xobject->fscache.event_mask);
86 printk(KERN_ERR "xops=%u inp=%u exc=%u\n",
87 xobject->fscache.n_ops, xobject->fscache.n_in_progress,
88 xobject->fscache.n_exclusive);
89 printk(KERN_ERR "xcookie=%p [pr=%p nd=%p fl=%lx]\n",
90 xobject->fscache.cookie,
91 xobject->fscache.cookie->parent,
92 xobject->fscache.cookie->netfs_data,
93 xobject->fscache.cookie->flags);
94 printk(KERN_ERR "xparent=%p\n",
95 xobject->fscache.parent);
96 printk(KERN_ERR "object: OBJ%x\n",
97 object->fscache.debug_id);
98 printk(KERN_ERR "cookie=%p [pr=%p nd=%p fl=%lx]\n",
99 object->fscache.cookie,
100 object->fscache.cookie->parent,
101 object->fscache.cookie->netfs_data,
102 object->fscache.cookie->flags);
103 printk(KERN_ERR "parent=%p\n",
104 object->fscache.parent);
105 BUG();
106 }
107 atomic_inc(&xobject->usage);
108 write_unlock(&cache->active_lock);
109
110 _debug(">>> wait");
111 wait_on_bit(&xobject->flags, CACHEFILES_OBJECT_ACTIVE,
112 cachefiles_wait_bit, TASK_UNINTERRUPTIBLE);
113 _debug("<<< waited");
114
115 cache->cache.ops->put_object(&xobject->fscache);
116 goto try_again;
117}
118
119/*
120 * delete an object representation from the cache
121 * - file backed objects are unlinked
122 * - directory backed objects are stuffed into the graveyard for userspace to
123 * delete
124 * - unlocks the directory mutex
125 */
126static int cachefiles_bury_object(struct cachefiles_cache *cache,
127 struct dentry *dir,
128 struct dentry *rep)
129{
130 struct dentry *grave, *trap;
131 char nbuffer[8 + 8 + 1];
132 int ret;
133
134 _enter(",'%*.*s','%*.*s'",
135 dir->d_name.len, dir->d_name.len, dir->d_name.name,
136 rep->d_name.len, rep->d_name.len, rep->d_name.name);
137
138 /* non-directories can just be unlinked */
139 if (!S_ISDIR(rep->d_inode->i_mode)) {
140 _debug("unlink stale object");
141 ret = vfs_unlink(dir->d_inode, rep);
142
143 mutex_unlock(&dir->d_inode->i_mutex);
144
145 if (ret == -EIO)
146 cachefiles_io_error(cache, "Unlink failed");
147
148 _leave(" = %d", ret);
149 return ret;
150 }
151
152 /* directories have to be moved to the graveyard */
153 _debug("move stale object to graveyard");
154 mutex_unlock(&dir->d_inode->i_mutex);
155
156try_again:
157 /* first step is to make up a grave dentry in the graveyard */
158 sprintf(nbuffer, "%08x%08x",
159 (uint32_t) get_seconds(),
160 (uint32_t) atomic_inc_return(&cache->gravecounter));
161
162 /* do the multiway lock magic */
163 trap = lock_rename(cache->graveyard, dir);
164
165 /* do some checks before getting the grave dentry */
166 if (rep->d_parent != dir) {
167 /* the entry was probably culled when we dropped the parent dir
168 * lock */
169 unlock_rename(cache->graveyard, dir);
170 _leave(" = 0 [culled?]");
171 return 0;
172 }
173
174 if (!S_ISDIR(cache->graveyard->d_inode->i_mode)) {
175 unlock_rename(cache->graveyard, dir);
176 cachefiles_io_error(cache, "Graveyard no longer a directory");
177 return -EIO;
178 }
179
180 if (trap == rep) {
181 unlock_rename(cache->graveyard, dir);
182 cachefiles_io_error(cache, "May not make directory loop");
183 return -EIO;
184 }
185
186 if (d_mountpoint(rep)) {
187 unlock_rename(cache->graveyard, dir);
188 cachefiles_io_error(cache, "Mountpoint in cache");
189 return -EIO;
190 }
191
192 grave = lookup_one_len(nbuffer, cache->graveyard, strlen(nbuffer));
193 if (IS_ERR(grave)) {
194 unlock_rename(cache->graveyard, dir);
195
196 if (PTR_ERR(grave) == -ENOMEM) {
197 _leave(" = -ENOMEM");
198 return -ENOMEM;
199 }
200
201 cachefiles_io_error(cache, "Lookup error %ld",
202 PTR_ERR(grave));
203 return -EIO;
204 }
205
206 if (grave->d_inode) {
207 unlock_rename(cache->graveyard, dir);
208 dput(grave);
209 grave = NULL;
210 cond_resched();
211 goto try_again;
212 }
213
214 if (d_mountpoint(grave)) {
215 unlock_rename(cache->graveyard, dir);
216 dput(grave);
217 cachefiles_io_error(cache, "Mountpoint in graveyard");
218 return -EIO;
219 }
220
221 /* target should not be an ancestor of source */
222 if (trap == grave) {
223 unlock_rename(cache->graveyard, dir);
224 dput(grave);
225 cachefiles_io_error(cache, "May not make directory loop");
226 return -EIO;
227 }
228
229 /* attempt the rename */
230 ret = vfs_rename(dir->d_inode, rep, cache->graveyard->d_inode, grave);
231 if (ret != 0 && ret != -ENOMEM)
232 cachefiles_io_error(cache, "Rename failed with error %d", ret);
233
234 unlock_rename(cache->graveyard, dir);
235 dput(grave);
236 _leave(" = 0");
237 return 0;
238}
239
240/*
241 * delete an object representation from the cache
242 */
243int cachefiles_delete_object(struct cachefiles_cache *cache,
244 struct cachefiles_object *object)
245{
246 struct dentry *dir;
247 int ret;
248
249 _enter(",{%p}", object->dentry);
250
251 ASSERT(object->dentry);
252 ASSERT(object->dentry->d_inode);
253 ASSERT(object->dentry->d_parent);
254
255 dir = dget_parent(object->dentry);
256
257 mutex_lock(&dir->d_inode->i_mutex);
258 ret = cachefiles_bury_object(cache, dir, object->dentry);
259
260 dput(dir);
261 _leave(" = %d", ret);
262 return ret;
263}
264
265/*
266 * walk from the parent object to the child object through the backing
267 * filesystem, creating directories as we go
268 */
269int cachefiles_walk_to_object(struct cachefiles_object *parent,
270 struct cachefiles_object *object,
271 const char *key,
272 struct cachefiles_xattr *auxdata)
273{
274 struct cachefiles_cache *cache;
275 struct dentry *dir, *next = NULL;
276 unsigned long start;
277 const char *name;
278 int ret, nlen;
279
280 _enter("{%p},,%s,", parent->dentry, key);
281
282 cache = container_of(parent->fscache.cache,
283 struct cachefiles_cache, cache);
284
285 ASSERT(parent->dentry);
286 ASSERT(parent->dentry->d_inode);
287
288 if (!(S_ISDIR(parent->dentry->d_inode->i_mode))) {
289 // TODO: convert file to dir
290 _leave("looking up in none directory");
291 return -ENOBUFS;
292 }
293
294 dir = dget(parent->dentry);
295
296advance:
297 /* attempt to transit the first directory component */
298 name = key;
299 nlen = strlen(key);
300
301 /* key ends in a double NUL */
302 key = key + nlen + 1;
303 if (!*key)
304 key = NULL;
305
306lookup_again:
307 /* search the current directory for the element name */
308 _debug("lookup '%s'", name);
309
310 mutex_lock(&dir->d_inode->i_mutex);
311
312 start = jiffies;
313 next = lookup_one_len(name, dir, nlen);
314 cachefiles_hist(cachefiles_lookup_histogram, start);
315 if (IS_ERR(next))
316 goto lookup_error;
317
318 _debug("next -> %p %s", next, next->d_inode ? "positive" : "negative");
319
320 if (!key)
321 object->new = !next->d_inode;
322
323 /* if this element of the path doesn't exist, then the lookup phase
324 * failed, and we can release any readers in the certain knowledge that
325 * there's nothing for them to actually read */
326 if (!next->d_inode)
327 fscache_object_lookup_negative(&object->fscache);
328
329 /* we need to create the object if it's negative */
330 if (key || object->type == FSCACHE_COOKIE_TYPE_INDEX) {
331 /* index objects and intervening tree levels must be subdirs */
332 if (!next->d_inode) {
333 ret = cachefiles_has_space(cache, 1, 0);
334 if (ret < 0)
335 goto create_error;
336
337 start = jiffies;
338 ret = vfs_mkdir(dir->d_inode, next, 0);
339 cachefiles_hist(cachefiles_mkdir_histogram, start);
340 if (ret < 0)
341 goto create_error;
342
343 ASSERT(next->d_inode);
344
345 _debug("mkdir -> %p{%p{ino=%lu}}",
346 next, next->d_inode, next->d_inode->i_ino);
347
348 } else if (!S_ISDIR(next->d_inode->i_mode)) {
349 kerror("inode %lu is not a directory",
350 next->d_inode->i_ino);
351 ret = -ENOBUFS;
352 goto error;
353 }
354
355 } else {
356 /* non-index objects start out life as files */
357 if (!next->d_inode) {
358 ret = cachefiles_has_space(cache, 1, 0);
359 if (ret < 0)
360 goto create_error;
361
362 start = jiffies;
363 ret = vfs_create(dir->d_inode, next, S_IFREG, NULL);
364 cachefiles_hist(cachefiles_create_histogram, start);
365 if (ret < 0)
366 goto create_error;
367
368 ASSERT(next->d_inode);
369
370 _debug("create -> %p{%p{ino=%lu}}",
371 next, next->d_inode, next->d_inode->i_ino);
372
373 } else if (!S_ISDIR(next->d_inode->i_mode) &&
374 !S_ISREG(next->d_inode->i_mode)
375 ) {
376 kerror("inode %lu is not a file or directory",
377 next->d_inode->i_ino);
378 ret = -ENOBUFS;
379 goto error;
380 }
381 }
382
383 /* process the next component */
384 if (key) {
385 _debug("advance");
386 mutex_unlock(&dir->d_inode->i_mutex);
387 dput(dir);
388 dir = next;
389 next = NULL;
390 goto advance;
391 }
392
393 /* we've found the object we were looking for */
394 object->dentry = next;
395
396 /* if we've found that the terminal object exists, then we need to
397 * check its attributes and delete it if it's out of date */
398 if (!object->new) {
399 _debug("validate '%*.*s'",
400 next->d_name.len, next->d_name.len, next->d_name.name);
401
402 ret = cachefiles_check_object_xattr(object, auxdata);
403 if (ret == -ESTALE) {
404 /* delete the object (the deleter drops the directory
405 * mutex) */
406 object->dentry = NULL;
407
408 ret = cachefiles_bury_object(cache, dir, next);
409 dput(next);
410 next = NULL;
411
412 if (ret < 0)
413 goto delete_error;
414
415 _debug("redo lookup");
416 goto lookup_again;
417 }
418 }
419
420 /* note that we're now using this object */
421 cachefiles_mark_object_active(cache, object);
422
423 mutex_unlock(&dir->d_inode->i_mutex);
424 dput(dir);
425 dir = NULL;
426
427 _debug("=== OBTAINED_OBJECT ===");
428
429 if (object->new) {
430 /* attach data to a newly constructed terminal object */
431 ret = cachefiles_set_object_xattr(object, auxdata);
432 if (ret < 0)
433 goto check_error;
434 } else {
435 /* always update the atime on an object we've just looked up
436 * (this is used to keep track of culling, and atimes are only
437 * updated by read, write and readdir but not lookup or
438 * open) */
439 touch_atime(cache->mnt, next);
440 }
441
442 /* open a file interface onto a data file */
443 if (object->type != FSCACHE_COOKIE_TYPE_INDEX) {
444 if (S_ISREG(object->dentry->d_inode->i_mode)) {
445 const struct address_space_operations *aops;
446
447 ret = -EPERM;
448 aops = object->dentry->d_inode->i_mapping->a_ops;
449 if (!aops->bmap)
450 goto check_error;
451
452 object->backer = object->dentry;
453 } else {
454 BUG(); // TODO: open file in data-class subdir
455 }
456 }
457
458 object->new = 0;
459 fscache_obtained_object(&object->fscache);
460
461 _leave(" = 0 [%lu]", object->dentry->d_inode->i_ino);
462 return 0;
463
464create_error:
465 _debug("create error %d", ret);
466 if (ret == -EIO)
467 cachefiles_io_error(cache, "Create/mkdir failed");
468 goto error;
469
470check_error:
471 _debug("check error %d", ret);
472 write_lock(&cache->active_lock);
473 rb_erase(&object->active_node, &cache->active_nodes);
474 clear_bit(CACHEFILES_OBJECT_ACTIVE, &object->flags);
475 wake_up_bit(&object->flags, CACHEFILES_OBJECT_ACTIVE);
476 write_unlock(&cache->active_lock);
477
478 dput(object->dentry);
479 object->dentry = NULL;
480 goto error_out;
481
482delete_error:
483 _debug("delete error %d", ret);
484 goto error_out2;
485
486lookup_error:
487 _debug("lookup error %ld", PTR_ERR(next));
488 ret = PTR_ERR(next);
489 if (ret == -EIO)
490 cachefiles_io_error(cache, "Lookup failed");
491 next = NULL;
492error:
493 mutex_unlock(&dir->d_inode->i_mutex);
494 dput(next);
495error_out2:
496 dput(dir);
497error_out:
498 if (ret == -ENOSPC)
499 ret = -ENOBUFS;
500
501 _leave(" = error %d", -ret);
502 return ret;
503}
504
505/*
506 * get a subdirectory
507 */
508struct dentry *cachefiles_get_directory(struct cachefiles_cache *cache,
509 struct dentry *dir,
510 const char *dirname)
511{
512 struct dentry *subdir;
513 unsigned long start;
514 int ret;
515
516 _enter(",,%s", dirname);
517
518 /* search the current directory for the element name */
519 mutex_lock(&dir->d_inode->i_mutex);
520
521 start = jiffies;
522 subdir = lookup_one_len(dirname, dir, strlen(dirname));
523 cachefiles_hist(cachefiles_lookup_histogram, start);
524 if (IS_ERR(subdir)) {
525 if (PTR_ERR(subdir) == -ENOMEM)
526 goto nomem_d_alloc;
527 goto lookup_error;
528 }
529
530 _debug("subdir -> %p %s",
531 subdir, subdir->d_inode ? "positive" : "negative");
532
533 /* we need to create the subdir if it doesn't exist yet */
534 if (!subdir->d_inode) {
535 ret = cachefiles_has_space(cache, 1, 0);
536 if (ret < 0)
537 goto mkdir_error;
538
539 _debug("attempt mkdir");
540
541 ret = vfs_mkdir(dir->d_inode, subdir, 0700);
542 if (ret < 0)
543 goto mkdir_error;
544
545 ASSERT(subdir->d_inode);
546
547 _debug("mkdir -> %p{%p{ino=%lu}}",
548 subdir,
549 subdir->d_inode,
550 subdir->d_inode->i_ino);
551 }
552
553 mutex_unlock(&dir->d_inode->i_mutex);
554
555 /* we need to make sure the subdir is a directory */
556 ASSERT(subdir->d_inode);
557
558 if (!S_ISDIR(subdir->d_inode->i_mode)) {
559 kerror("%s is not a directory", dirname);
560 ret = -EIO;
561 goto check_error;
562 }
563
564 ret = -EPERM;
565 if (!subdir->d_inode->i_op ||
566 !subdir->d_inode->i_op->setxattr ||
567 !subdir->d_inode->i_op->getxattr ||
568 !subdir->d_inode->i_op->lookup ||
569 !subdir->d_inode->i_op->mkdir ||
570 !subdir->d_inode->i_op->create ||
571 !subdir->d_inode->i_op->rename ||
572 !subdir->d_inode->i_op->rmdir ||
573 !subdir->d_inode->i_op->unlink)
574 goto check_error;
575
576 _leave(" = [%lu]", subdir->d_inode->i_ino);
577 return subdir;
578
579check_error:
580 dput(subdir);
581 _leave(" = %d [check]", ret);
582 return ERR_PTR(ret);
583
584mkdir_error:
585 mutex_unlock(&dir->d_inode->i_mutex);
586 dput(subdir);
587 kerror("mkdir %s failed with error %d", dirname, ret);
588 return ERR_PTR(ret);
589
590lookup_error:
591 mutex_unlock(&dir->d_inode->i_mutex);
592 ret = PTR_ERR(subdir);
593 kerror("Lookup %s failed with error %d", dirname, ret);
594 return ERR_PTR(ret);
595
596nomem_d_alloc:
597 mutex_unlock(&dir->d_inode->i_mutex);
598 _leave(" = -ENOMEM");
599 return ERR_PTR(-ENOMEM);
600}
601
602/*
603 * find out if an object is in use or not
604 * - if finds object and it's not in use:
605 * - returns a pointer to the object and a reference on it
606 * - returns with the directory locked
607 */
608static struct dentry *cachefiles_check_active(struct cachefiles_cache *cache,
609 struct dentry *dir,
610 char *filename)
611{
612 struct cachefiles_object *object;
613 struct rb_node *_n;
614 struct dentry *victim;
615 unsigned long start;
616 int ret;
617
618 //_enter(",%*.*s/,%s",
619 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
620
621 /* look up the victim */
622 mutex_lock_nested(&dir->d_inode->i_mutex, 1);
623
624 start = jiffies;
625 victim = lookup_one_len(filename, dir, strlen(filename));
626 cachefiles_hist(cachefiles_lookup_histogram, start);
627 if (IS_ERR(victim))
628 goto lookup_error;
629
630 //_debug("victim -> %p %s",
631 // victim, victim->d_inode ? "positive" : "negative");
632
633 /* if the object is no longer there then we probably retired the object
634 * at the netfs's request whilst the cull was in progress
635 */
636 if (!victim->d_inode) {
637 mutex_unlock(&dir->d_inode->i_mutex);
638 dput(victim);
639 _leave(" = -ENOENT [absent]");
640 return ERR_PTR(-ENOENT);
641 }
642
643 /* check to see if we're using this object */
644 read_lock(&cache->active_lock);
645
646 _n = cache->active_nodes.rb_node;
647
648 while (_n) {
649 object = rb_entry(_n, struct cachefiles_object, active_node);
650
651 if (object->dentry > victim)
652 _n = _n->rb_left;
653 else if (object->dentry < victim)
654 _n = _n->rb_right;
655 else
656 goto object_in_use;
657 }
658
659 read_unlock(&cache->active_lock);
660
661 //_leave(" = %p", victim);
662 return victim;
663
664object_in_use:
665 read_unlock(&cache->active_lock);
666 mutex_unlock(&dir->d_inode->i_mutex);
667 dput(victim);
668 //_leave(" = -EBUSY [in use]");
669 return ERR_PTR(-EBUSY);
670
671lookup_error:
672 mutex_unlock(&dir->d_inode->i_mutex);
673 ret = PTR_ERR(victim);
674 if (ret == -ENOENT) {
675 /* file or dir now absent - probably retired by netfs */
676 _leave(" = -ESTALE [absent]");
677 return ERR_PTR(-ESTALE);
678 }
679
680 if (ret == -EIO) {
681 cachefiles_io_error(cache, "Lookup failed");
682 } else if (ret != -ENOMEM) {
683 kerror("Internal error: %d", ret);
684 ret = -EIO;
685 }
686
687 _leave(" = %d", ret);
688 return ERR_PTR(ret);
689}
690
691/*
692 * cull an object if it's not in use
693 * - called only by cache manager daemon
694 */
695int cachefiles_cull(struct cachefiles_cache *cache, struct dentry *dir,
696 char *filename)
697{
698 struct dentry *victim;
699 int ret;
700
701 _enter(",%*.*s/,%s",
702 dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
703
704 victim = cachefiles_check_active(cache, dir, filename);
705 if (IS_ERR(victim))
706 return PTR_ERR(victim);
707
708 _debug("victim -> %p %s",
709 victim, victim->d_inode ? "positive" : "negative");
710
711 /* okay... the victim is not being used so we can cull it
712 * - start by marking it as stale
713 */
714 _debug("victim is cullable");
715
716 ret = cachefiles_remove_object_xattr(cache, victim);
717 if (ret < 0)
718 goto error_unlock;
719
720 /* actually remove the victim (drops the dir mutex) */
721 _debug("bury");
722
723 ret = cachefiles_bury_object(cache, dir, victim);
724 if (ret < 0)
725 goto error;
726
727 dput(victim);
728 _leave(" = 0");
729 return 0;
730
731error_unlock:
732 mutex_unlock(&dir->d_inode->i_mutex);
733error:
734 dput(victim);
735 if (ret == -ENOENT) {
736 /* file or dir now absent - probably retired by netfs */
737 _leave(" = -ESTALE [absent]");
738 return -ESTALE;
739 }
740
741 if (ret != -ENOMEM) {
742 kerror("Internal error: %d", ret);
743 ret = -EIO;
744 }
745
746 _leave(" = %d", ret);
747 return ret;
748}
749
750/*
751 * find out if an object is in use or not
752 * - called only by cache manager daemon
753 * - returns -EBUSY or 0 to indicate whether an object is in use or not
754 */
755int cachefiles_check_in_use(struct cachefiles_cache *cache, struct dentry *dir,
756 char *filename)
757{
758 struct dentry *victim;
759
760 //_enter(",%*.*s/,%s",
761 // dir->d_name.len, dir->d_name.len, dir->d_name.name, filename);
762
763 victim = cachefiles_check_active(cache, dir, filename);
764 if (IS_ERR(victim))
765 return PTR_ERR(victim);
766
767 mutex_unlock(&dir->d_inode->i_mutex);
768 dput(victim);
769 //_leave(" = 0");
770 return 0;
771}
diff --git a/fs/cachefiles/proc.c b/fs/cachefiles/proc.c
new file mode 100644
index 000000000000..eccd33941199
--- /dev/null
+++ b/fs/cachefiles/proc.c
@@ -0,0 +1,134 @@
1/* CacheFiles statistics
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/module.h>
13#include <linux/proc_fs.h>
14#include <linux/seq_file.h>
15#include "internal.h"
16
17atomic_t cachefiles_lookup_histogram[HZ];
18atomic_t cachefiles_mkdir_histogram[HZ];
19atomic_t cachefiles_create_histogram[HZ];
20
21/*
22 * display the latency histogram
23 */
24static int cachefiles_histogram_show(struct seq_file *m, void *v)
25{
26 unsigned long index;
27 unsigned x, y, z, t;
28
29 switch ((unsigned long) v) {
30 case 1:
31 seq_puts(m, "JIFS SECS LOOKUPS MKDIRS CREATES\n");
32 return 0;
33 case 2:
34 seq_puts(m, "===== ===== ========= ========= =========\n");
35 return 0;
36 default:
37 index = (unsigned long) v - 3;
38 x = atomic_read(&cachefiles_lookup_histogram[index]);
39 y = atomic_read(&cachefiles_mkdir_histogram[index]);
40 z = atomic_read(&cachefiles_create_histogram[index]);
41 if (x == 0 && y == 0 && z == 0)
42 return 0;
43
44 t = (index * 1000) / HZ;
45
46 seq_printf(m, "%4lu 0.%03u %9u %9u %9u\n", index, t, x, y, z);
47 return 0;
48 }
49}
50
51/*
52 * set up the iterator to start reading from the first line
53 */
54static void *cachefiles_histogram_start(struct seq_file *m, loff_t *_pos)
55{
56 if ((unsigned long long)*_pos >= HZ + 2)
57 return NULL;
58 if (*_pos == 0)
59 *_pos = 1;
60 return (void *)(unsigned long) *_pos;
61}
62
63/*
64 * move to the next line
65 */
66static void *cachefiles_histogram_next(struct seq_file *m, void *v, loff_t *pos)
67{
68 (*pos)++;
69 return (unsigned long long)*pos > HZ + 2 ?
70 NULL : (void *)(unsigned long) *pos;
71}
72
73/*
74 * clean up after reading
75 */
76static void cachefiles_histogram_stop(struct seq_file *m, void *v)
77{
78}
79
80static const struct seq_operations cachefiles_histogram_ops = {
81 .start = cachefiles_histogram_start,
82 .stop = cachefiles_histogram_stop,
83 .next = cachefiles_histogram_next,
84 .show = cachefiles_histogram_show,
85};
86
87/*
88 * open "/proc/fs/cachefiles/XXX" which provide statistics summaries
89 */
90static int cachefiles_histogram_open(struct inode *inode, struct file *file)
91{
92 return seq_open(file, &cachefiles_histogram_ops);
93}
94
95static const struct file_operations cachefiles_histogram_fops = {
96 .owner = THIS_MODULE,
97 .open = cachefiles_histogram_open,
98 .read = seq_read,
99 .llseek = seq_lseek,
100 .release = seq_release,
101};
102
103/*
104 * initialise the /proc/fs/cachefiles/ directory
105 */
106int __init cachefiles_proc_init(void)
107{
108 _enter("");
109
110 if (!proc_mkdir("fs/cachefiles", NULL))
111 goto error_dir;
112
113 if (!proc_create("fs/cachefiles/histogram", S_IFREG | 0444, NULL,
114 &cachefiles_histogram_fops))
115 goto error_histogram;
116
117 _leave(" = 0");
118 return 0;
119
120error_histogram:
121 remove_proc_entry("fs/cachefiles", NULL);
122error_dir:
123 _leave(" = -ENOMEM");
124 return -ENOMEM;
125}
126
127/*
128 * clean up the /proc/fs/cachefiles/ directory
129 */
130void cachefiles_proc_cleanup(void)
131{
132 remove_proc_entry("fs/cachefiles/histogram", NULL);
133 remove_proc_entry("fs/cachefiles", NULL);
134}
diff --git a/fs/cachefiles/rdwr.c b/fs/cachefiles/rdwr.c
new file mode 100644
index 000000000000..a69787e7dd96
--- /dev/null
+++ b/fs/cachefiles/rdwr.c
@@ -0,0 +1,879 @@
1/* Storage object read/write
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/mount.h>
13#include <linux/file.h>
14#include "internal.h"
15
16/*
17 * detect wake up events generated by the unlocking of pages in which we're
18 * interested
19 * - we use this to detect read completion of backing pages
20 * - the caller holds the waitqueue lock
21 */
22static int cachefiles_read_waiter(wait_queue_t *wait, unsigned mode,
23 int sync, void *_key)
24{
25 struct cachefiles_one_read *monitor =
26 container_of(wait, struct cachefiles_one_read, monitor);
27 struct cachefiles_object *object;
28 struct wait_bit_key *key = _key;
29 struct page *page = wait->private;
30
31 ASSERT(key);
32
33 _enter("{%lu},%u,%d,{%p,%u}",
34 monitor->netfs_page->index, mode, sync,
35 key->flags, key->bit_nr);
36
37 if (key->flags != &page->flags ||
38 key->bit_nr != PG_locked)
39 return 0;
40
41 _debug("--- monitor %p %lx ---", page, page->flags);
42
43 if (!PageUptodate(page) && !PageError(page))
44 dump_stack();
45
46 /* remove from the waitqueue */
47 list_del(&wait->task_list);
48
49 /* move onto the action list and queue for FS-Cache thread pool */
50 ASSERT(monitor->op);
51
52 object = container_of(monitor->op->op.object,
53 struct cachefiles_object, fscache);
54
55 spin_lock(&object->work_lock);
56 list_add_tail(&monitor->op_link, &monitor->op->to_do);
57 spin_unlock(&object->work_lock);
58
59 fscache_enqueue_retrieval(monitor->op);
60 return 0;
61}
62
63/*
64 * copy data from backing pages to netfs pages to complete a read operation
65 * - driven by FS-Cache's thread pool
66 */
67static void cachefiles_read_copier(struct fscache_operation *_op)
68{
69 struct cachefiles_one_read *monitor;
70 struct cachefiles_object *object;
71 struct fscache_retrieval *op;
72 struct pagevec pagevec;
73 int error, max;
74
75 op = container_of(_op, struct fscache_retrieval, op);
76 object = container_of(op->op.object,
77 struct cachefiles_object, fscache);
78
79 _enter("{ino=%lu}", object->backer->d_inode->i_ino);
80
81 pagevec_init(&pagevec, 0);
82
83 max = 8;
84 spin_lock_irq(&object->work_lock);
85
86 while (!list_empty(&op->to_do)) {
87 monitor = list_entry(op->to_do.next,
88 struct cachefiles_one_read, op_link);
89 list_del(&monitor->op_link);
90
91 spin_unlock_irq(&object->work_lock);
92
93 _debug("- copy {%lu}", monitor->back_page->index);
94
95 error = -EIO;
96 if (PageUptodate(monitor->back_page)) {
97 copy_highpage(monitor->netfs_page, monitor->back_page);
98
99 pagevec_add(&pagevec, monitor->netfs_page);
100 fscache_mark_pages_cached(monitor->op, &pagevec);
101 error = 0;
102 }
103
104 if (error)
105 cachefiles_io_error_obj(
106 object,
107 "Readpage failed on backing file %lx",
108 (unsigned long) monitor->back_page->flags);
109
110 page_cache_release(monitor->back_page);
111
112 fscache_end_io(op, monitor->netfs_page, error);
113 page_cache_release(monitor->netfs_page);
114 fscache_put_retrieval(op);
115 kfree(monitor);
116
117 /* let the thread pool have some air occasionally */
118 max--;
119 if (max < 0 || need_resched()) {
120 if (!list_empty(&op->to_do))
121 fscache_enqueue_retrieval(op);
122 _leave(" [maxed out]");
123 return;
124 }
125
126 spin_lock_irq(&object->work_lock);
127 }
128
129 spin_unlock_irq(&object->work_lock);
130 _leave("");
131}
132
133/*
134 * read the corresponding page to the given set from the backing file
135 * - an uncertain page is simply discarded, to be tried again another time
136 */
137static int cachefiles_read_backing_file_one(struct cachefiles_object *object,
138 struct fscache_retrieval *op,
139 struct page *netpage,
140 struct pagevec *pagevec)
141{
142 struct cachefiles_one_read *monitor;
143 struct address_space *bmapping;
144 struct page *newpage, *backpage;
145 int ret;
146
147 _enter("");
148
149 pagevec_reinit(pagevec);
150
151 _debug("read back %p{%lu,%d}",
152 netpage, netpage->index, page_count(netpage));
153
154 monitor = kzalloc(sizeof(*monitor), GFP_KERNEL);
155 if (!monitor)
156 goto nomem;
157
158 monitor->netfs_page = netpage;
159 monitor->op = fscache_get_retrieval(op);
160
161 init_waitqueue_func_entry(&monitor->monitor, cachefiles_read_waiter);
162
163 /* attempt to get hold of the backing page */
164 bmapping = object->backer->d_inode->i_mapping;
165 newpage = NULL;
166
167 for (;;) {
168 backpage = find_get_page(bmapping, netpage->index);
169 if (backpage)
170 goto backing_page_already_present;
171
172 if (!newpage) {
173 newpage = page_cache_alloc_cold(bmapping);
174 if (!newpage)
175 goto nomem_monitor;
176 }
177
178 ret = add_to_page_cache(newpage, bmapping,
179 netpage->index, GFP_KERNEL);
180 if (ret == 0)
181 goto installed_new_backing_page;
182 if (ret != -EEXIST)
183 goto nomem_page;
184 }
185
186 /* we've installed a new backing page, so now we need to add it
187 * to the LRU list and start it reading */
188installed_new_backing_page:
189 _debug("- new %p", newpage);
190
191 backpage = newpage;
192 newpage = NULL;
193
194 page_cache_get(backpage);
195 pagevec_add(pagevec, backpage);
196 __pagevec_lru_add_file(pagevec);
197
198read_backing_page:
199 ret = bmapping->a_ops->readpage(NULL, backpage);
200 if (ret < 0)
201 goto read_error;
202
203 /* set the monitor to transfer the data across */
204monitor_backing_page:
205 _debug("- monitor add");
206
207 /* install the monitor */
208 page_cache_get(monitor->netfs_page);
209 page_cache_get(backpage);
210 monitor->back_page = backpage;
211 monitor->monitor.private = backpage;
212 add_page_wait_queue(backpage, &monitor->monitor);
213 monitor = NULL;
214
215 /* but the page may have been read before the monitor was installed, so
216 * the monitor may miss the event - so we have to ensure that we do get
217 * one in such a case */
218 if (trylock_page(backpage)) {
219 _debug("jumpstart %p {%lx}", backpage, backpage->flags);
220 unlock_page(backpage);
221 }
222 goto success;
223
224 /* if the backing page is already present, it can be in one of
225 * three states: read in progress, read failed or read okay */
226backing_page_already_present:
227 _debug("- present");
228
229 if (newpage) {
230 page_cache_release(newpage);
231 newpage = NULL;
232 }
233
234 if (PageError(backpage))
235 goto io_error;
236
237 if (PageUptodate(backpage))
238 goto backing_page_already_uptodate;
239
240 if (!trylock_page(backpage))
241 goto monitor_backing_page;
242 _debug("read %p {%lx}", backpage, backpage->flags);
243 goto read_backing_page;
244
245 /* the backing page is already up to date, attach the netfs
246 * page to the pagecache and LRU and copy the data across */
247backing_page_already_uptodate:
248 _debug("- uptodate");
249
250 pagevec_add(pagevec, netpage);
251 fscache_mark_pages_cached(op, pagevec);
252
253 copy_highpage(netpage, backpage);
254 fscache_end_io(op, netpage, 0);
255
256success:
257 _debug("success");
258 ret = 0;
259
260out:
261 if (backpage)
262 page_cache_release(backpage);
263 if (monitor) {
264 fscache_put_retrieval(monitor->op);
265 kfree(monitor);
266 }
267 _leave(" = %d", ret);
268 return ret;
269
270read_error:
271 _debug("read error %d", ret);
272 if (ret == -ENOMEM)
273 goto out;
274io_error:
275 cachefiles_io_error_obj(object, "Page read error on backing file");
276 ret = -ENOBUFS;
277 goto out;
278
279nomem_page:
280 page_cache_release(newpage);
281nomem_monitor:
282 fscache_put_retrieval(monitor->op);
283 kfree(monitor);
284nomem:
285 _leave(" = -ENOMEM");
286 return -ENOMEM;
287}
288
289/*
290 * read a page from the cache or allocate a block in which to store it
291 * - cache withdrawal is prevented by the caller
292 * - returns -EINTR if interrupted
293 * - returns -ENOMEM if ran out of memory
294 * - returns -ENOBUFS if no buffers can be made available
295 * - returns -ENOBUFS if page is beyond EOF
296 * - if the page is backed by a block in the cache:
297 * - a read will be started which will call the callback on completion
298 * - 0 will be returned
299 * - else if the page is unbacked:
300 * - the metadata will be retained
301 * - -ENODATA will be returned
302 */
303int cachefiles_read_or_alloc_page(struct fscache_retrieval *op,
304 struct page *page,
305 gfp_t gfp)
306{
307 struct cachefiles_object *object;
308 struct cachefiles_cache *cache;
309 struct pagevec pagevec;
310 struct inode *inode;
311 sector_t block0, block;
312 unsigned shift;
313 int ret;
314
315 object = container_of(op->op.object,
316 struct cachefiles_object, fscache);
317 cache = container_of(object->fscache.cache,
318 struct cachefiles_cache, cache);
319
320 _enter("{%p},{%lx},,,", object, page->index);
321
322 if (!object->backer)
323 return -ENOBUFS;
324
325 inode = object->backer->d_inode;
326 ASSERT(S_ISREG(inode->i_mode));
327 ASSERT(inode->i_mapping->a_ops->bmap);
328 ASSERT(inode->i_mapping->a_ops->readpages);
329
330 /* calculate the shift required to use bmap */
331 if (inode->i_sb->s_blocksize > PAGE_SIZE)
332 return -ENOBUFS;
333
334 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
335
336 op->op.flags = FSCACHE_OP_FAST;
337 op->op.processor = cachefiles_read_copier;
338
339 pagevec_init(&pagevec, 0);
340
341 /* we assume the absence or presence of the first block is a good
342 * enough indication for the page as a whole
343 * - TODO: don't use bmap() for this as it is _not_ actually good
344 * enough for this as it doesn't indicate errors, but it's all we've
345 * got for the moment
346 */
347 block0 = page->index;
348 block0 <<= shift;
349
350 block = inode->i_mapping->a_ops->bmap(inode->i_mapping, block0);
351 _debug("%llx -> %llx",
352 (unsigned long long) block0,
353 (unsigned long long) block);
354
355 if (block) {
356 /* submit the apparently valid page to the backing fs to be
357 * read from disk */
358 ret = cachefiles_read_backing_file_one(object, op, page,
359 &pagevec);
360 } else if (cachefiles_has_space(cache, 0, 1) == 0) {
361 /* there's space in the cache we can use */
362 pagevec_add(&pagevec, page);
363 fscache_mark_pages_cached(op, &pagevec);
364 ret = -ENODATA;
365 } else {
366 ret = -ENOBUFS;
367 }
368
369 _leave(" = %d", ret);
370 return ret;
371}
372
373/*
374 * read the corresponding pages to the given set from the backing file
375 * - any uncertain pages are simply discarded, to be tried again another time
376 */
377static int cachefiles_read_backing_file(struct cachefiles_object *object,
378 struct fscache_retrieval *op,
379 struct list_head *list,
380 struct pagevec *mark_pvec)
381{
382 struct cachefiles_one_read *monitor = NULL;
383 struct address_space *bmapping = object->backer->d_inode->i_mapping;
384 struct pagevec lru_pvec;
385 struct page *newpage = NULL, *netpage, *_n, *backpage = NULL;
386 int ret = 0;
387
388 _enter("");
389
390 pagevec_init(&lru_pvec, 0);
391
392 list_for_each_entry_safe(netpage, _n, list, lru) {
393 list_del(&netpage->lru);
394
395 _debug("read back %p{%lu,%d}",
396 netpage, netpage->index, page_count(netpage));
397
398 if (!monitor) {
399 monitor = kzalloc(sizeof(*monitor), GFP_KERNEL);
400 if (!monitor)
401 goto nomem;
402
403 monitor->op = fscache_get_retrieval(op);
404 init_waitqueue_func_entry(&monitor->monitor,
405 cachefiles_read_waiter);
406 }
407
408 for (;;) {
409 backpage = find_get_page(bmapping, netpage->index);
410 if (backpage)
411 goto backing_page_already_present;
412
413 if (!newpage) {
414 newpage = page_cache_alloc_cold(bmapping);
415 if (!newpage)
416 goto nomem;
417 }
418
419 ret = add_to_page_cache(newpage, bmapping,
420 netpage->index, GFP_KERNEL);
421 if (ret == 0)
422 goto installed_new_backing_page;
423 if (ret != -EEXIST)
424 goto nomem;
425 }
426
427 /* we've installed a new backing page, so now we need to add it
428 * to the LRU list and start it reading */
429 installed_new_backing_page:
430 _debug("- new %p", newpage);
431
432 backpage = newpage;
433 newpage = NULL;
434
435 page_cache_get(backpage);
436 if (!pagevec_add(&lru_pvec, backpage))
437 __pagevec_lru_add_file(&lru_pvec);
438
439 reread_backing_page:
440 ret = bmapping->a_ops->readpage(NULL, backpage);
441 if (ret < 0)
442 goto read_error;
443
444 /* add the netfs page to the pagecache and LRU, and set the
445 * monitor to transfer the data across */
446 monitor_backing_page:
447 _debug("- monitor add");
448
449 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
450 GFP_KERNEL);
451 if (ret < 0) {
452 if (ret == -EEXIST) {
453 page_cache_release(netpage);
454 continue;
455 }
456 goto nomem;
457 }
458
459 page_cache_get(netpage);
460 if (!pagevec_add(&lru_pvec, netpage))
461 __pagevec_lru_add_file(&lru_pvec);
462
463 /* install a monitor */
464 page_cache_get(netpage);
465 monitor->netfs_page = netpage;
466
467 page_cache_get(backpage);
468 monitor->back_page = backpage;
469 monitor->monitor.private = backpage;
470 add_page_wait_queue(backpage, &monitor->monitor);
471 monitor = NULL;
472
473 /* but the page may have been read before the monitor was
474 * installed, so the monitor may miss the event - so we have to
475 * ensure that we do get one in such a case */
476 if (trylock_page(backpage)) {
477 _debug("2unlock %p {%lx}", backpage, backpage->flags);
478 unlock_page(backpage);
479 }
480
481 page_cache_release(backpage);
482 backpage = NULL;
483
484 page_cache_release(netpage);
485 netpage = NULL;
486 continue;
487
488 /* if the backing page is already present, it can be in one of
489 * three states: read in progress, read failed or read okay */
490 backing_page_already_present:
491 _debug("- present %p", backpage);
492
493 if (PageError(backpage))
494 goto io_error;
495
496 if (PageUptodate(backpage))
497 goto backing_page_already_uptodate;
498
499 _debug("- not ready %p{%lx}", backpage, backpage->flags);
500
501 if (!trylock_page(backpage))
502 goto monitor_backing_page;
503
504 if (PageError(backpage)) {
505 _debug("error %lx", backpage->flags);
506 unlock_page(backpage);
507 goto io_error;
508 }
509
510 if (PageUptodate(backpage))
511 goto backing_page_already_uptodate_unlock;
512
513 /* we've locked a page that's neither up to date nor erroneous,
514 * so we need to attempt to read it again */
515 goto reread_backing_page;
516
517 /* the backing page is already up to date, attach the netfs
518 * page to the pagecache and LRU and copy the data across */
519 backing_page_already_uptodate_unlock:
520 _debug("uptodate %lx", backpage->flags);
521 unlock_page(backpage);
522 backing_page_already_uptodate:
523 _debug("- uptodate");
524
525 ret = add_to_page_cache(netpage, op->mapping, netpage->index,
526 GFP_KERNEL);
527 if (ret < 0) {
528 if (ret == -EEXIST) {
529 page_cache_release(netpage);
530 continue;
531 }
532 goto nomem;
533 }
534
535 copy_highpage(netpage, backpage);
536
537 page_cache_release(backpage);
538 backpage = NULL;
539
540 if (!pagevec_add(mark_pvec, netpage))
541 fscache_mark_pages_cached(op, mark_pvec);
542
543 page_cache_get(netpage);
544 if (!pagevec_add(&lru_pvec, netpage))
545 __pagevec_lru_add_file(&lru_pvec);
546
547 fscache_end_io(op, netpage, 0);
548 page_cache_release(netpage);
549 netpage = NULL;
550 continue;
551 }
552
553 netpage = NULL;
554
555 _debug("out");
556
557out:
558 /* tidy up */
559 pagevec_lru_add_file(&lru_pvec);
560
561 if (newpage)
562 page_cache_release(newpage);
563 if (netpage)
564 page_cache_release(netpage);
565 if (backpage)
566 page_cache_release(backpage);
567 if (monitor) {
568 fscache_put_retrieval(op);
569 kfree(monitor);
570 }
571
572 list_for_each_entry_safe(netpage, _n, list, lru) {
573 list_del(&netpage->lru);
574 page_cache_release(netpage);
575 }
576
577 _leave(" = %d", ret);
578 return ret;
579
580nomem:
581 _debug("nomem");
582 ret = -ENOMEM;
583 goto out;
584
585read_error:
586 _debug("read error %d", ret);
587 if (ret == -ENOMEM)
588 goto out;
589io_error:
590 cachefiles_io_error_obj(object, "Page read error on backing file");
591 ret = -ENOBUFS;
592 goto out;
593}
594
595/*
596 * read a list of pages from the cache or allocate blocks in which to store
597 * them
598 */
599int cachefiles_read_or_alloc_pages(struct fscache_retrieval *op,
600 struct list_head *pages,
601 unsigned *nr_pages,
602 gfp_t gfp)
603{
604 struct cachefiles_object *object;
605 struct cachefiles_cache *cache;
606 struct list_head backpages;
607 struct pagevec pagevec;
608 struct inode *inode;
609 struct page *page, *_n;
610 unsigned shift, nrbackpages;
611 int ret, ret2, space;
612
613 object = container_of(op->op.object,
614 struct cachefiles_object, fscache);
615 cache = container_of(object->fscache.cache,
616 struct cachefiles_cache, cache);
617
618 _enter("{OBJ%x,%d},,%d,,",
619 object->fscache.debug_id, atomic_read(&op->op.usage),
620 *nr_pages);
621
622 if (!object->backer)
623 return -ENOBUFS;
624
625 space = 1;
626 if (cachefiles_has_space(cache, 0, *nr_pages) < 0)
627 space = 0;
628
629 inode = object->backer->d_inode;
630 ASSERT(S_ISREG(inode->i_mode));
631 ASSERT(inode->i_mapping->a_ops->bmap);
632 ASSERT(inode->i_mapping->a_ops->readpages);
633
634 /* calculate the shift required to use bmap */
635 if (inode->i_sb->s_blocksize > PAGE_SIZE)
636 return -ENOBUFS;
637
638 shift = PAGE_SHIFT - inode->i_sb->s_blocksize_bits;
639
640 pagevec_init(&pagevec, 0);
641
642 op->op.flags = FSCACHE_OP_FAST;
643 op->op.processor = cachefiles_read_copier;
644
645 INIT_LIST_HEAD(&backpages);
646 nrbackpages = 0;
647
648 ret = space ? -ENODATA : -ENOBUFS;
649 list_for_each_entry_safe(page, _n, pages, lru) {
650 sector_t block0, block;
651
652 /* we assume the absence or presence of the first block is a
653 * good enough indication for the page as a whole
654 * - TODO: don't use bmap() for this as it is _not_ actually
655 * good enough for this as it doesn't indicate errors, but
656 * it's all we've got for the moment
657 */
658 block0 = page->index;
659 block0 <<= shift;
660
661 block = inode->i_mapping->a_ops->bmap(inode->i_mapping,
662 block0);
663 _debug("%llx -> %llx",
664 (unsigned long long) block0,
665 (unsigned long long) block);
666
667 if (block) {
668 /* we have data - add it to the list to give to the
669 * backing fs */
670 list_move(&page->lru, &backpages);
671 (*nr_pages)--;
672 nrbackpages++;
673 } else if (space && pagevec_add(&pagevec, page) == 0) {
674 fscache_mark_pages_cached(op, &pagevec);
675 ret = -ENODATA;
676 }
677 }
678
679 if (pagevec_count(&pagevec) > 0)
680 fscache_mark_pages_cached(op, &pagevec);
681
682 if (list_empty(pages))
683 ret = 0;
684
685 /* submit the apparently valid pages to the backing fs to be read from
686 * disk */
687 if (nrbackpages > 0) {
688 ret2 = cachefiles_read_backing_file(object, op, &backpages,
689 &pagevec);
690 if (ret2 == -ENOMEM || ret2 == -EINTR)
691 ret = ret2;
692 }
693
694 if (pagevec_count(&pagevec) > 0)
695 fscache_mark_pages_cached(op, &pagevec);
696
697 _leave(" = %d [nr=%u%s]",
698 ret, *nr_pages, list_empty(pages) ? " empty" : "");
699 return ret;
700}
701
702/*
703 * allocate a block in the cache in which to store a page
704 * - cache withdrawal is prevented by the caller
705 * - returns -EINTR if interrupted
706 * - returns -ENOMEM if ran out of memory
707 * - returns -ENOBUFS if no buffers can be made available
708 * - returns -ENOBUFS if page is beyond EOF
709 * - otherwise:
710 * - the metadata will be retained
711 * - 0 will be returned
712 */
713int cachefiles_allocate_page(struct fscache_retrieval *op,
714 struct page *page,
715 gfp_t gfp)
716{
717 struct cachefiles_object *object;
718 struct cachefiles_cache *cache;
719 struct pagevec pagevec;
720 int ret;
721
722 object = container_of(op->op.object,
723 struct cachefiles_object, fscache);
724 cache = container_of(object->fscache.cache,
725 struct cachefiles_cache, cache);
726
727 _enter("%p,{%lx},", object, page->index);
728
729 ret = cachefiles_has_space(cache, 0, 1);
730 if (ret == 0) {
731 pagevec_init(&pagevec, 0);
732 pagevec_add(&pagevec, page);
733 fscache_mark_pages_cached(op, &pagevec);
734 } else {
735 ret = -ENOBUFS;
736 }
737
738 _leave(" = %d", ret);
739 return ret;
740}
741
742/*
743 * allocate blocks in the cache in which to store a set of pages
744 * - cache withdrawal is prevented by the caller
745 * - returns -EINTR if interrupted
746 * - returns -ENOMEM if ran out of memory
747 * - returns -ENOBUFS if some buffers couldn't be made available
748 * - returns -ENOBUFS if some pages are beyond EOF
749 * - otherwise:
750 * - -ENODATA will be returned
751 * - metadata will be retained for any page marked
752 */
753int cachefiles_allocate_pages(struct fscache_retrieval *op,
754 struct list_head *pages,
755 unsigned *nr_pages,
756 gfp_t gfp)
757{
758 struct cachefiles_object *object;
759 struct cachefiles_cache *cache;
760 struct pagevec pagevec;
761 struct page *page;
762 int ret;
763
764 object = container_of(op->op.object,
765 struct cachefiles_object, fscache);
766 cache = container_of(object->fscache.cache,
767 struct cachefiles_cache, cache);
768
769 _enter("%p,,,%d,", object, *nr_pages);
770
771 ret = cachefiles_has_space(cache, 0, *nr_pages);
772 if (ret == 0) {
773 pagevec_init(&pagevec, 0);
774
775 list_for_each_entry(page, pages, lru) {
776 if (pagevec_add(&pagevec, page) == 0)
777 fscache_mark_pages_cached(op, &pagevec);
778 }
779
780 if (pagevec_count(&pagevec) > 0)
781 fscache_mark_pages_cached(op, &pagevec);
782 ret = -ENODATA;
783 } else {
784 ret = -ENOBUFS;
785 }
786
787 _leave(" = %d", ret);
788 return ret;
789}
790
791/*
792 * request a page be stored in the cache
793 * - cache withdrawal is prevented by the caller
794 * - this request may be ignored if there's no cache block available, in which
795 * case -ENOBUFS will be returned
796 * - if the op is in progress, 0 will be returned
797 */
798int cachefiles_write_page(struct fscache_storage *op, struct page *page)
799{
800 struct cachefiles_object *object;
801 struct cachefiles_cache *cache;
802 mm_segment_t old_fs;
803 struct file *file;
804 loff_t pos;
805 void *data;
806 int ret;
807
808 ASSERT(op != NULL);
809 ASSERT(page != NULL);
810
811 object = container_of(op->op.object,
812 struct cachefiles_object, fscache);
813
814 _enter("%p,%p{%lx},,,", object, page, page->index);
815
816 if (!object->backer) {
817 _leave(" = -ENOBUFS");
818 return -ENOBUFS;
819 }
820
821 ASSERT(S_ISREG(object->backer->d_inode->i_mode));
822
823 cache = container_of(object->fscache.cache,
824 struct cachefiles_cache, cache);
825
826 /* write the page to the backing filesystem and let it store it in its
827 * own time */
828 dget(object->backer);
829 mntget(cache->mnt);
830 file = dentry_open(object->backer, cache->mnt, O_RDWR,
831 cache->cache_cred);
832 if (IS_ERR(file)) {
833 ret = PTR_ERR(file);
834 } else {
835 ret = -EIO;
836 if (file->f_op->write) {
837 pos = (loff_t) page->index << PAGE_SHIFT;
838 data = kmap(page);
839 old_fs = get_fs();
840 set_fs(KERNEL_DS);
841 ret = file->f_op->write(
842 file, (const void __user *) data, PAGE_SIZE,
843 &pos);
844 set_fs(old_fs);
845 kunmap(page);
846 if (ret != PAGE_SIZE)
847 ret = -EIO;
848 }
849 fput(file);
850 }
851
852 if (ret < 0) {
853 if (ret == -EIO)
854 cachefiles_io_error_obj(
855 object, "Write page to backing file failed");
856 ret = -ENOBUFS;
857 }
858
859 _leave(" = %d", ret);
860 return ret;
861}
862
863/*
864 * detach a backing block from a page
865 * - cache withdrawal is prevented by the caller
866 */
867void cachefiles_uncache_page(struct fscache_object *_object, struct page *page)
868{
869 struct cachefiles_object *object;
870 struct cachefiles_cache *cache;
871
872 object = container_of(_object, struct cachefiles_object, fscache);
873 cache = container_of(object->fscache.cache,
874 struct cachefiles_cache, cache);
875
876 _enter("%p,{%lu}", object, page->index);
877
878 spin_unlock(&object->fscache.cookie->lock);
879}
diff --git a/fs/cachefiles/security.c b/fs/cachefiles/security.c
new file mode 100644
index 000000000000..b5808cdb2232
--- /dev/null
+++ b/fs/cachefiles/security.c
@@ -0,0 +1,116 @@
1/* CacheFiles security management
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/fs.h>
13#include <linux/cred.h>
14#include "internal.h"
15
16/*
17 * determine the security context within which we access the cache from within
18 * the kernel
19 */
20int cachefiles_get_security_ID(struct cachefiles_cache *cache)
21{
22 struct cred *new;
23 int ret;
24
25 _enter("{%s}", cache->secctx);
26
27 new = prepare_kernel_cred(current);
28 if (!new) {
29 ret = -ENOMEM;
30 goto error;
31 }
32
33 if (cache->secctx) {
34 ret = set_security_override_from_ctx(new, cache->secctx);
35 if (ret < 0) {
36 put_cred(new);
37 printk(KERN_ERR "CacheFiles:"
38 " Security denies permission to nominate"
39 " security context: error %d\n",
40 ret);
41 goto error;
42 }
43 }
44
45 cache->cache_cred = new;
46 ret = 0;
47error:
48 _leave(" = %d", ret);
49 return ret;
50}
51
52/*
53 * see if mkdir and create can be performed in the root directory
54 */
55static int cachefiles_check_cache_dir(struct cachefiles_cache *cache,
56 struct dentry *root)
57{
58 int ret;
59
60 ret = security_inode_mkdir(root->d_inode, root, 0);
61 if (ret < 0) {
62 printk(KERN_ERR "CacheFiles:"
63 " Security denies permission to make dirs: error %d",
64 ret);
65 return ret;
66 }
67
68 ret = security_inode_create(root->d_inode, root, 0);
69 if (ret < 0)
70 printk(KERN_ERR "CacheFiles:"
71 " Security denies permission to create files: error %d",
72 ret);
73
74 return ret;
75}
76
77/*
78 * check the security details of the on-disk cache
79 * - must be called with security override in force
80 */
81int cachefiles_determine_cache_security(struct cachefiles_cache *cache,
82 struct dentry *root,
83 const struct cred **_saved_cred)
84{
85 struct cred *new;
86 int ret;
87
88 _enter("");
89
90 /* duplicate the cache creds for COW (the override is currently in
91 * force, so we can use prepare_creds() to do this) */
92 new = prepare_creds();
93 if (!new)
94 return -ENOMEM;
95
96 cachefiles_end_secure(cache, *_saved_cred);
97
98 /* use the cache root dir's security context as the basis with
99 * which create files */
100 ret = set_create_files_as(new, root->d_inode);
101 if (ret < 0) {
102 _leave(" = %d [cfa]", ret);
103 return ret;
104 }
105
106 put_cred(cache->cache_cred);
107 cache->cache_cred = new;
108
109 cachefiles_begin_secure(cache, _saved_cred);
110 ret = cachefiles_check_cache_dir(cache, root);
111
112 if (ret == -EOPNOTSUPP)
113 ret = 0;
114 _leave(" = %d", ret);
115 return ret;
116}
diff --git a/fs/cachefiles/xattr.c b/fs/cachefiles/xattr.c
new file mode 100644
index 000000000000..f3e7a0bf068b
--- /dev/null
+++ b/fs/cachefiles/xattr.c
@@ -0,0 +1,291 @@
1/* CacheFiles extended attribute management
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/module.h>
13#include <linux/sched.h>
14#include <linux/file.h>
15#include <linux/fs.h>
16#include <linux/fsnotify.h>
17#include <linux/quotaops.h>
18#include <linux/xattr.h>
19#include "internal.h"
20
21static const char cachefiles_xattr_cache[] =
22 XATTR_USER_PREFIX "CacheFiles.cache";
23
24/*
25 * check the type label on an object
26 * - done using xattrs
27 */
28int cachefiles_check_object_type(struct cachefiles_object *object)
29{
30 struct dentry *dentry = object->dentry;
31 char type[3], xtype[3];
32 int ret;
33
34 ASSERT(dentry);
35 ASSERT(dentry->d_inode);
36
37 if (!object->fscache.cookie)
38 strcpy(type, "C3");
39 else
40 snprintf(type, 3, "%02x", object->fscache.cookie->def->type);
41
42 _enter("%p{%s}", object, type);
43
44 /* attempt to install a type label directly */
45 ret = vfs_setxattr(dentry, cachefiles_xattr_cache, type, 2,
46 XATTR_CREATE);
47 if (ret == 0) {
48 _debug("SET"); /* we succeeded */
49 goto error;
50 }
51
52 if (ret != -EEXIST) {
53 kerror("Can't set xattr on %*.*s [%lu] (err %d)",
54 dentry->d_name.len, dentry->d_name.len,
55 dentry->d_name.name, dentry->d_inode->i_ino,
56 -ret);
57 goto error;
58 }
59
60 /* read the current type label */
61 ret = vfs_getxattr(dentry, cachefiles_xattr_cache, xtype, 3);
62 if (ret < 0) {
63 if (ret == -ERANGE)
64 goto bad_type_length;
65
66 kerror("Can't read xattr on %*.*s [%lu] (err %d)",
67 dentry->d_name.len, dentry->d_name.len,
68 dentry->d_name.name, dentry->d_inode->i_ino,
69 -ret);
70 goto error;
71 }
72
73 /* check the type is what we're expecting */
74 if (ret != 2)
75 goto bad_type_length;
76
77 if (xtype[0] != type[0] || xtype[1] != type[1])
78 goto bad_type;
79
80 ret = 0;
81
82error:
83 _leave(" = %d", ret);
84 return ret;
85
86bad_type_length:
87 kerror("Cache object %lu type xattr length incorrect",
88 dentry->d_inode->i_ino);
89 ret = -EIO;
90 goto error;
91
92bad_type:
93 xtype[2] = 0;
94 kerror("Cache object %*.*s [%lu] type %s not %s",
95 dentry->d_name.len, dentry->d_name.len,
96 dentry->d_name.name, dentry->d_inode->i_ino,
97 xtype, type);
98 ret = -EIO;
99 goto error;
100}
101
102/*
103 * set the state xattr on a cache file
104 */
105int cachefiles_set_object_xattr(struct cachefiles_object *object,
106 struct cachefiles_xattr *auxdata)
107{
108 struct dentry *dentry = object->dentry;
109 int ret;
110
111 ASSERT(object->fscache.cookie);
112 ASSERT(dentry);
113
114 _enter("%p,#%d", object, auxdata->len);
115
116 /* attempt to install the cache metadata directly */
117 _debug("SET %s #%u", object->fscache.cookie->def->name, auxdata->len);
118
119 ret = vfs_setxattr(dentry, cachefiles_xattr_cache,
120 &auxdata->type, auxdata->len,
121 XATTR_CREATE);
122 if (ret < 0 && ret != -ENOMEM)
123 cachefiles_io_error_obj(
124 object,
125 "Failed to set xattr with error %d", ret);
126
127 _leave(" = %d", ret);
128 return ret;
129}
130
131/*
132 * update the state xattr on a cache file
133 */
134int cachefiles_update_object_xattr(struct cachefiles_object *object,
135 struct cachefiles_xattr *auxdata)
136{
137 struct dentry *dentry = object->dentry;
138 int ret;
139
140 ASSERT(object->fscache.cookie);
141 ASSERT(dentry);
142
143 _enter("%p,#%d", object, auxdata->len);
144
145 /* attempt to install the cache metadata directly */
146 _debug("SET %s #%u", object->fscache.cookie->def->name, auxdata->len);
147
148 ret = vfs_setxattr(dentry, cachefiles_xattr_cache,
149 &auxdata->type, auxdata->len,
150 XATTR_REPLACE);
151 if (ret < 0 && ret != -ENOMEM)
152 cachefiles_io_error_obj(
153 object,
154 "Failed to update xattr with error %d", ret);
155
156 _leave(" = %d", ret);
157 return ret;
158}
159
160/*
161 * check the state xattr on a cache file
162 * - return -ESTALE if the object should be deleted
163 */
164int cachefiles_check_object_xattr(struct cachefiles_object *object,
165 struct cachefiles_xattr *auxdata)
166{
167 struct cachefiles_xattr *auxbuf;
168 struct dentry *dentry = object->dentry;
169 int ret;
170
171 _enter("%p,#%d", object, auxdata->len);
172
173 ASSERT(dentry);
174 ASSERT(dentry->d_inode);
175
176 auxbuf = kmalloc(sizeof(struct cachefiles_xattr) + 512, GFP_KERNEL);
177 if (!auxbuf) {
178 _leave(" = -ENOMEM");
179 return -ENOMEM;
180 }
181
182 /* read the current type label */
183 ret = vfs_getxattr(dentry, cachefiles_xattr_cache,
184 &auxbuf->type, 512 + 1);
185 if (ret < 0) {
186 if (ret == -ENODATA)
187 goto stale; /* no attribute - power went off
188 * mid-cull? */
189
190 if (ret == -ERANGE)
191 goto bad_type_length;
192
193 cachefiles_io_error_obj(object,
194 "Can't read xattr on %lu (err %d)",
195 dentry->d_inode->i_ino, -ret);
196 goto error;
197 }
198
199 /* check the on-disk object */
200 if (ret < 1)
201 goto bad_type_length;
202
203 if (auxbuf->type != auxdata->type)
204 goto stale;
205
206 auxbuf->len = ret;
207
208 /* consult the netfs */
209 if (object->fscache.cookie->def->check_aux) {
210 enum fscache_checkaux result;
211 unsigned int dlen;
212
213 dlen = auxbuf->len - 1;
214
215 _debug("checkaux %s #%u",
216 object->fscache.cookie->def->name, dlen);
217
218 result = fscache_check_aux(&object->fscache,
219 &auxbuf->data, dlen);
220
221 switch (result) {
222 /* entry okay as is */
223 case FSCACHE_CHECKAUX_OKAY:
224 goto okay;
225
226 /* entry requires update */
227 case FSCACHE_CHECKAUX_NEEDS_UPDATE:
228 break;
229
230 /* entry requires deletion */
231 case FSCACHE_CHECKAUX_OBSOLETE:
232 goto stale;
233
234 default:
235 BUG();
236 }
237
238 /* update the current label */
239 ret = vfs_setxattr(dentry, cachefiles_xattr_cache,
240 &auxdata->type, auxdata->len,
241 XATTR_REPLACE);
242 if (ret < 0) {
243 cachefiles_io_error_obj(object,
244 "Can't update xattr on %lu"
245 " (error %d)",
246 dentry->d_inode->i_ino, -ret);
247 goto error;
248 }
249 }
250
251okay:
252 ret = 0;
253
254error:
255 kfree(auxbuf);
256 _leave(" = %d", ret);
257 return ret;
258
259bad_type_length:
260 kerror("Cache object %lu xattr length incorrect",
261 dentry->d_inode->i_ino);
262 ret = -EIO;
263 goto error;
264
265stale:
266 ret = -ESTALE;
267 goto error;
268}
269
270/*
271 * remove the object's xattr to mark it stale
272 */
273int cachefiles_remove_object_xattr(struct cachefiles_cache *cache,
274 struct dentry *dentry)
275{
276 int ret;
277
278 ret = vfs_removexattr(dentry, cachefiles_xattr_cache);
279 if (ret < 0) {
280 if (ret == -ENOENT || ret == -ENODATA)
281 ret = 0;
282 else if (ret != -ENOMEM)
283 cachefiles_io_error(cache,
284 "Can't remove xattr from %lu"
285 " (error %d)",
286 dentry->d_inode->i_ino, -ret);
287 }
288
289 _leave(" = %d", ret);
290 return ret;
291}
diff --git a/fs/fscache/Kconfig b/fs/fscache/Kconfig
new file mode 100644
index 000000000000..9bbb8ce7bea0
--- /dev/null
+++ b/fs/fscache/Kconfig
@@ -0,0 +1,56 @@
1
2config FSCACHE
3 tristate "General filesystem local caching manager"
4 depends on EXPERIMENTAL
5 select SLOW_WORK
6 help
7 This option enables a generic filesystem caching manager that can be
8 used by various network and other filesystems to cache data locally.
9 Different sorts of caches can be plugged in, depending on the
10 resources available.
11
12 See Documentation/filesystems/caching/fscache.txt for more information.
13
14config FSCACHE_STATS
15 bool "Gather statistical information on local caching"
16 depends on FSCACHE && PROC_FS
17 help
18 This option causes statistical information to be gathered on local
19 caching and exported through file:
20
21 /proc/fs/fscache/stats
22
23 The gathering of statistics adds a certain amount of overhead to
24 execution as there are a quite a few stats gathered, and on a
25 multi-CPU system these may be on cachelines that keep bouncing
26 between CPUs. On the other hand, the stats are very useful for
27 debugging purposes. Saying 'Y' here is recommended.
28
29 See Documentation/filesystems/caching/fscache.txt for more information.
30
31config FSCACHE_HISTOGRAM
32 bool "Gather latency information on local caching"
33 depends on FSCACHE && PROC_FS
34 help
35 This option causes latency information to be gathered on local
36 caching and exported through file:
37
38 /proc/fs/fscache/histogram
39
40 The generation of this histogram adds a certain amount of overhead to
41 execution as there are a number of points at which data is gathered,
42 and on a multi-CPU system these may be on cachelines that keep
43 bouncing between CPUs. On the other hand, the histogram may be
44 useful for debugging purposes. Saying 'N' here is recommended.
45
46 See Documentation/filesystems/caching/fscache.txt for more information.
47
48config FSCACHE_DEBUG
49 bool "Debug FS-Cache"
50 depends on FSCACHE
51 help
52 This permits debugging to be dynamically enabled in the local caching
53 management module. If this is set, the debugging output may be
54 enabled by setting bits in /sys/modules/fscache/parameter/debug.
55
56 See Documentation/filesystems/caching/fscache.txt for more information.
diff --git a/fs/fscache/Makefile b/fs/fscache/Makefile
new file mode 100644
index 000000000000..91571b95aacc
--- /dev/null
+++ b/fs/fscache/Makefile
@@ -0,0 +1,19 @@
1#
2# Makefile for general filesystem caching code
3#
4
5fscache-y := \
6 cache.o \
7 cookie.o \
8 fsdef.o \
9 main.o \
10 netfs.o \
11 object.o \
12 operation.o \
13 page.o
14
15fscache-$(CONFIG_PROC_FS) += proc.o
16fscache-$(CONFIG_FSCACHE_STATS) += stats.o
17fscache-$(CONFIG_FSCACHE_HISTOGRAM) += histogram.o
18
19obj-$(CONFIG_FSCACHE) := fscache.o
diff --git a/fs/fscache/cache.c b/fs/fscache/cache.c
new file mode 100644
index 000000000000..e21985bbb1fb
--- /dev/null
+++ b/fs/fscache/cache.c
@@ -0,0 +1,415 @@
1/* FS-Cache cache handling
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL CACHE
13#include <linux/module.h>
14#include <linux/slab.h>
15#include "internal.h"
16
17LIST_HEAD(fscache_cache_list);
18DECLARE_RWSEM(fscache_addremove_sem);
19DECLARE_WAIT_QUEUE_HEAD(fscache_cache_cleared_wq);
20EXPORT_SYMBOL(fscache_cache_cleared_wq);
21
22static LIST_HEAD(fscache_cache_tag_list);
23
24/*
25 * look up a cache tag
26 */
27struct fscache_cache_tag *__fscache_lookup_cache_tag(const char *name)
28{
29 struct fscache_cache_tag *tag, *xtag;
30
31 /* firstly check for the existence of the tag under read lock */
32 down_read(&fscache_addremove_sem);
33
34 list_for_each_entry(tag, &fscache_cache_tag_list, link) {
35 if (strcmp(tag->name, name) == 0) {
36 atomic_inc(&tag->usage);
37 up_read(&fscache_addremove_sem);
38 return tag;
39 }
40 }
41
42 up_read(&fscache_addremove_sem);
43
44 /* the tag does not exist - create a candidate */
45 xtag = kzalloc(sizeof(*xtag) + strlen(name) + 1, GFP_KERNEL);
46 if (!xtag)
47 /* return a dummy tag if out of memory */
48 return ERR_PTR(-ENOMEM);
49
50 atomic_set(&xtag->usage, 1);
51 strcpy(xtag->name, name);
52
53 /* write lock, search again and add if still not present */
54 down_write(&fscache_addremove_sem);
55
56 list_for_each_entry(tag, &fscache_cache_tag_list, link) {
57 if (strcmp(tag->name, name) == 0) {
58 atomic_inc(&tag->usage);
59 up_write(&fscache_addremove_sem);
60 kfree(xtag);
61 return tag;
62 }
63 }
64
65 list_add_tail(&xtag->link, &fscache_cache_tag_list);
66 up_write(&fscache_addremove_sem);
67 return xtag;
68}
69
70/*
71 * release a reference to a cache tag
72 */
73void __fscache_release_cache_tag(struct fscache_cache_tag *tag)
74{
75 if (tag != ERR_PTR(-ENOMEM)) {
76 down_write(&fscache_addremove_sem);
77
78 if (atomic_dec_and_test(&tag->usage))
79 list_del_init(&tag->link);
80 else
81 tag = NULL;
82
83 up_write(&fscache_addremove_sem);
84
85 kfree(tag);
86 }
87}
88
89/*
90 * select a cache in which to store an object
91 * - the cache addremove semaphore must be at least read-locked by the caller
92 * - the object will never be an index
93 */
94struct fscache_cache *fscache_select_cache_for_object(
95 struct fscache_cookie *cookie)
96{
97 struct fscache_cache_tag *tag;
98 struct fscache_object *object;
99 struct fscache_cache *cache;
100
101 _enter("");
102
103 if (list_empty(&fscache_cache_list)) {
104 _leave(" = NULL [no cache]");
105 return NULL;
106 }
107
108 /* we check the parent to determine the cache to use */
109 spin_lock(&cookie->lock);
110
111 /* the first in the parent's backing list should be the preferred
112 * cache */
113 if (!hlist_empty(&cookie->backing_objects)) {
114 object = hlist_entry(cookie->backing_objects.first,
115 struct fscache_object, cookie_link);
116
117 cache = object->cache;
118 if (object->state >= FSCACHE_OBJECT_DYING ||
119 test_bit(FSCACHE_IOERROR, &cache->flags))
120 cache = NULL;
121
122 spin_unlock(&cookie->lock);
123 _leave(" = %p [parent]", cache);
124 return cache;
125 }
126
127 /* the parent is unbacked */
128 if (cookie->def->type != FSCACHE_COOKIE_TYPE_INDEX) {
129 /* cookie not an index and is unbacked */
130 spin_unlock(&cookie->lock);
131 _leave(" = NULL [cookie ub,ni]");
132 return NULL;
133 }
134
135 spin_unlock(&cookie->lock);
136
137 if (!cookie->def->select_cache)
138 goto no_preference;
139
140 /* ask the netfs for its preference */
141 tag = cookie->def->select_cache(cookie->parent->netfs_data,
142 cookie->netfs_data);
143 if (!tag)
144 goto no_preference;
145
146 if (tag == ERR_PTR(-ENOMEM)) {
147 _leave(" = NULL [nomem tag]");
148 return NULL;
149 }
150
151 if (!tag->cache) {
152 _leave(" = NULL [unbacked tag]");
153 return NULL;
154 }
155
156 if (test_bit(FSCACHE_IOERROR, &tag->cache->flags))
157 return NULL;
158
159 _leave(" = %p [specific]", tag->cache);
160 return tag->cache;
161
162no_preference:
163 /* netfs has no preference - just select first cache */
164 cache = list_entry(fscache_cache_list.next,
165 struct fscache_cache, link);
166 _leave(" = %p [first]", cache);
167 return cache;
168}
169
170/**
171 * fscache_init_cache - Initialise a cache record
172 * @cache: The cache record to be initialised
173 * @ops: The cache operations to be installed in that record
174 * @idfmt: Format string to define identifier
175 * @...: sprintf-style arguments
176 *
177 * Initialise a record of a cache and fill in the name.
178 *
179 * See Documentation/filesystems/caching/backend-api.txt for a complete
180 * description.
181 */
182void fscache_init_cache(struct fscache_cache *cache,
183 const struct fscache_cache_ops *ops,
184 const char *idfmt,
185 ...)
186{
187 va_list va;
188
189 memset(cache, 0, sizeof(*cache));
190
191 cache->ops = ops;
192
193 va_start(va, idfmt);
194 vsnprintf(cache->identifier, sizeof(cache->identifier), idfmt, va);
195 va_end(va);
196
197 INIT_WORK(&cache->op_gc, fscache_operation_gc);
198 INIT_LIST_HEAD(&cache->link);
199 INIT_LIST_HEAD(&cache->object_list);
200 INIT_LIST_HEAD(&cache->op_gc_list);
201 spin_lock_init(&cache->object_list_lock);
202 spin_lock_init(&cache->op_gc_list_lock);
203}
204EXPORT_SYMBOL(fscache_init_cache);
205
206/**
207 * fscache_add_cache - Declare a cache as being open for business
208 * @cache: The record describing the cache
209 * @ifsdef: The record of the cache object describing the top-level index
210 * @tagname: The tag describing this cache
211 *
212 * Add a cache to the system, making it available for netfs's to use.
213 *
214 * See Documentation/filesystems/caching/backend-api.txt for a complete
215 * description.
216 */
217int fscache_add_cache(struct fscache_cache *cache,
218 struct fscache_object *ifsdef,
219 const char *tagname)
220{
221 struct fscache_cache_tag *tag;
222
223 BUG_ON(!cache->ops);
224 BUG_ON(!ifsdef);
225
226 cache->flags = 0;
227 ifsdef->event_mask = ULONG_MAX & ~(1 << FSCACHE_OBJECT_EV_CLEARED);
228 ifsdef->state = FSCACHE_OBJECT_ACTIVE;
229
230 if (!tagname)
231 tagname = cache->identifier;
232
233 BUG_ON(!tagname[0]);
234
235 _enter("{%s.%s},,%s", cache->ops->name, cache->identifier, tagname);
236
237 /* we use the cache tag to uniquely identify caches */
238 tag = __fscache_lookup_cache_tag(tagname);
239 if (IS_ERR(tag))
240 goto nomem;
241
242 if (test_and_set_bit(FSCACHE_TAG_RESERVED, &tag->flags))
243 goto tag_in_use;
244
245 cache->kobj = kobject_create_and_add(tagname, fscache_root);
246 if (!cache->kobj)
247 goto error;
248
249 ifsdef->cookie = &fscache_fsdef_index;
250 ifsdef->cache = cache;
251 cache->fsdef = ifsdef;
252
253 down_write(&fscache_addremove_sem);
254
255 tag->cache = cache;
256 cache->tag = tag;
257
258 /* add the cache to the list */
259 list_add(&cache->link, &fscache_cache_list);
260
261 /* add the cache's netfs definition index object to the cache's
262 * list */
263 spin_lock(&cache->object_list_lock);
264 list_add_tail(&ifsdef->cache_link, &cache->object_list);
265 spin_unlock(&cache->object_list_lock);
266
267 /* add the cache's netfs definition index object to the top level index
268 * cookie as a known backing object */
269 spin_lock(&fscache_fsdef_index.lock);
270
271 hlist_add_head(&ifsdef->cookie_link,
272 &fscache_fsdef_index.backing_objects);
273
274 atomic_inc(&fscache_fsdef_index.usage);
275
276 /* done */
277 spin_unlock(&fscache_fsdef_index.lock);
278 up_write(&fscache_addremove_sem);
279
280 printk(KERN_NOTICE "FS-Cache: Cache \"%s\" added (type %s)\n",
281 cache->tag->name, cache->ops->name);
282 kobject_uevent(cache->kobj, KOBJ_ADD);
283
284 _leave(" = 0 [%s]", cache->identifier);
285 return 0;
286
287tag_in_use:
288 printk(KERN_ERR "FS-Cache: Cache tag '%s' already in use\n", tagname);
289 __fscache_release_cache_tag(tag);
290 _leave(" = -EXIST");
291 return -EEXIST;
292
293error:
294 __fscache_release_cache_tag(tag);
295 _leave(" = -EINVAL");
296 return -EINVAL;
297
298nomem:
299 _leave(" = -ENOMEM");
300 return -ENOMEM;
301}
302EXPORT_SYMBOL(fscache_add_cache);
303
304/**
305 * fscache_io_error - Note a cache I/O error
306 * @cache: The record describing the cache
307 *
308 * Note that an I/O error occurred in a cache and that it should no longer be
309 * used for anything. This also reports the error into the kernel log.
310 *
311 * See Documentation/filesystems/caching/backend-api.txt for a complete
312 * description.
313 */
314void fscache_io_error(struct fscache_cache *cache)
315{
316 set_bit(FSCACHE_IOERROR, &cache->flags);
317
318 printk(KERN_ERR "FS-Cache: Cache %s stopped due to I/O error\n",
319 cache->ops->name);
320}
321EXPORT_SYMBOL(fscache_io_error);
322
323/*
324 * request withdrawal of all the objects in a cache
325 * - all the objects being withdrawn are moved onto the supplied list
326 */
327static void fscache_withdraw_all_objects(struct fscache_cache *cache,
328 struct list_head *dying_objects)
329{
330 struct fscache_object *object;
331
332 spin_lock(&cache->object_list_lock);
333
334 while (!list_empty(&cache->object_list)) {
335 object = list_entry(cache->object_list.next,
336 struct fscache_object, cache_link);
337 list_move_tail(&object->cache_link, dying_objects);
338
339 _debug("withdraw %p", object->cookie);
340
341 spin_lock(&object->lock);
342 spin_unlock(&cache->object_list_lock);
343 fscache_raise_event(object, FSCACHE_OBJECT_EV_WITHDRAW);
344 spin_unlock(&object->lock);
345
346 cond_resched();
347 spin_lock(&cache->object_list_lock);
348 }
349
350 spin_unlock(&cache->object_list_lock);
351}
352
353/**
354 * fscache_withdraw_cache - Withdraw a cache from the active service
355 * @cache: The record describing the cache
356 *
357 * Withdraw a cache from service, unbinding all its cache objects from the
358 * netfs cookies they're currently representing.
359 *
360 * See Documentation/filesystems/caching/backend-api.txt for a complete
361 * description.
362 */
363void fscache_withdraw_cache(struct fscache_cache *cache)
364{
365 LIST_HEAD(dying_objects);
366
367 _enter("");
368
369 printk(KERN_NOTICE "FS-Cache: Withdrawing cache \"%s\"\n",
370 cache->tag->name);
371
372 /* make the cache unavailable for cookie acquisition */
373 if (test_and_set_bit(FSCACHE_CACHE_WITHDRAWN, &cache->flags))
374 BUG();
375
376 down_write(&fscache_addremove_sem);
377 list_del_init(&cache->link);
378 cache->tag->cache = NULL;
379 up_write(&fscache_addremove_sem);
380
381 /* make sure all pages pinned by operations on behalf of the netfs are
382 * written to disk */
383 cache->ops->sync_cache(cache);
384
385 /* dissociate all the netfs pages backed by this cache from the block
386 * mappings in the cache */
387 cache->ops->dissociate_pages(cache);
388
389 /* we now have to destroy all the active objects pertaining to this
390 * cache - which we do by passing them off to thread pool to be
391 * disposed of */
392 _debug("destroy");
393
394 fscache_withdraw_all_objects(cache, &dying_objects);
395
396 /* wait for all extant objects to finish their outstanding operations
397 * and go away */
398 _debug("wait for finish");
399 wait_event(fscache_cache_cleared_wq,
400 atomic_read(&cache->object_count) == 0);
401 _debug("wait for clearance");
402 wait_event(fscache_cache_cleared_wq,
403 list_empty(&cache->object_list));
404 _debug("cleared");
405 ASSERT(list_empty(&dying_objects));
406
407 kobject_put(cache->kobj);
408
409 clear_bit(FSCACHE_TAG_RESERVED, &cache->tag->flags);
410 fscache_release_cache_tag(cache->tag);
411 cache->tag = NULL;
412
413 _leave("");
414}
415EXPORT_SYMBOL(fscache_withdraw_cache);
diff --git a/fs/fscache/cookie.c b/fs/fscache/cookie.c
new file mode 100644
index 000000000000..72fd18f6c71f
--- /dev/null
+++ b/fs/fscache/cookie.c
@@ -0,0 +1,500 @@
1/* netfs cookie management
2 *
3 * Copyright (C) 2004-2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * See Documentation/filesystems/caching/netfs-api.txt for more information on
12 * the netfs API.
13 */
14
15#define FSCACHE_DEBUG_LEVEL COOKIE
16#include <linux/module.h>
17#include <linux/slab.h>
18#include "internal.h"
19
20struct kmem_cache *fscache_cookie_jar;
21
22static atomic_t fscache_object_debug_id = ATOMIC_INIT(0);
23
24static int fscache_acquire_non_index_cookie(struct fscache_cookie *cookie);
25static int fscache_alloc_object(struct fscache_cache *cache,
26 struct fscache_cookie *cookie);
27static int fscache_attach_object(struct fscache_cookie *cookie,
28 struct fscache_object *object);
29
30/*
31 * initialise an cookie jar slab element prior to any use
32 */
33void fscache_cookie_init_once(void *_cookie)
34{
35 struct fscache_cookie *cookie = _cookie;
36
37 memset(cookie, 0, sizeof(*cookie));
38 spin_lock_init(&cookie->lock);
39 INIT_HLIST_HEAD(&cookie->backing_objects);
40}
41
42/*
43 * request a cookie to represent an object (index, datafile, xattr, etc)
44 * - parent specifies the parent object
45 * - the top level index cookie for each netfs is stored in the fscache_netfs
46 * struct upon registration
47 * - def points to the definition
48 * - the netfs_data will be passed to the functions pointed to in *def
49 * - all attached caches will be searched to see if they contain this object
50 * - index objects aren't stored on disk until there's a dependent file that
51 * needs storing
52 * - other objects are stored in a selected cache immediately, and all the
53 * indices forming the path to it are instantiated if necessary
54 * - we never let on to the netfs about errors
55 * - we may set a negative cookie pointer, but that's okay
56 */
57struct fscache_cookie *__fscache_acquire_cookie(
58 struct fscache_cookie *parent,
59 const struct fscache_cookie_def *def,
60 void *netfs_data)
61{
62 struct fscache_cookie *cookie;
63
64 BUG_ON(!def);
65
66 _enter("{%s},{%s},%p",
67 parent ? (char *) parent->def->name : "<no-parent>",
68 def->name, netfs_data);
69
70 fscache_stat(&fscache_n_acquires);
71
72 /* if there's no parent cookie, then we don't create one here either */
73 if (!parent) {
74 fscache_stat(&fscache_n_acquires_null);
75 _leave(" [no parent]");
76 return NULL;
77 }
78
79 /* validate the definition */
80 BUG_ON(!def->get_key);
81 BUG_ON(!def->name[0]);
82
83 BUG_ON(def->type == FSCACHE_COOKIE_TYPE_INDEX &&
84 parent->def->type != FSCACHE_COOKIE_TYPE_INDEX);
85
86 /* allocate and initialise a cookie */
87 cookie = kmem_cache_alloc(fscache_cookie_jar, GFP_KERNEL);
88 if (!cookie) {
89 fscache_stat(&fscache_n_acquires_oom);
90 _leave(" [ENOMEM]");
91 return NULL;
92 }
93
94 atomic_set(&cookie->usage, 1);
95 atomic_set(&cookie->n_children, 0);
96
97 atomic_inc(&parent->usage);
98 atomic_inc(&parent->n_children);
99
100 cookie->def = def;
101 cookie->parent = parent;
102 cookie->netfs_data = netfs_data;
103 cookie->flags = 0;
104
105 INIT_RADIX_TREE(&cookie->stores, GFP_NOFS);
106
107 switch (cookie->def->type) {
108 case FSCACHE_COOKIE_TYPE_INDEX:
109 fscache_stat(&fscache_n_cookie_index);
110 break;
111 case FSCACHE_COOKIE_TYPE_DATAFILE:
112 fscache_stat(&fscache_n_cookie_data);
113 break;
114 default:
115 fscache_stat(&fscache_n_cookie_special);
116 break;
117 }
118
119 /* if the object is an index then we need do nothing more here - we
120 * create indices on disk when we need them as an index may exist in
121 * multiple caches */
122 if (cookie->def->type != FSCACHE_COOKIE_TYPE_INDEX) {
123 if (fscache_acquire_non_index_cookie(cookie) < 0) {
124 atomic_dec(&parent->n_children);
125 __fscache_cookie_put(cookie);
126 fscache_stat(&fscache_n_acquires_nobufs);
127 _leave(" = NULL");
128 return NULL;
129 }
130 }
131
132 fscache_stat(&fscache_n_acquires_ok);
133 _leave(" = %p", cookie);
134 return cookie;
135}
136EXPORT_SYMBOL(__fscache_acquire_cookie);
137
138/*
139 * acquire a non-index cookie
140 * - this must make sure the index chain is instantiated and instantiate the
141 * object representation too
142 */
143static int fscache_acquire_non_index_cookie(struct fscache_cookie *cookie)
144{
145 struct fscache_object *object;
146 struct fscache_cache *cache;
147 uint64_t i_size;
148 int ret;
149
150 _enter("");
151
152 cookie->flags = 1 << FSCACHE_COOKIE_UNAVAILABLE;
153
154 /* now we need to see whether the backing objects for this cookie yet
155 * exist, if not there'll be nothing to search */
156 down_read(&fscache_addremove_sem);
157
158 if (list_empty(&fscache_cache_list)) {
159 up_read(&fscache_addremove_sem);
160 _leave(" = 0 [no caches]");
161 return 0;
162 }
163
164 /* select a cache in which to store the object */
165 cache = fscache_select_cache_for_object(cookie->parent);
166 if (!cache) {
167 up_read(&fscache_addremove_sem);
168 fscache_stat(&fscache_n_acquires_no_cache);
169 _leave(" = -ENOMEDIUM [no cache]");
170 return -ENOMEDIUM;
171 }
172
173 _debug("cache %s", cache->tag->name);
174
175 cookie->flags =
176 (1 << FSCACHE_COOKIE_LOOKING_UP) |
177 (1 << FSCACHE_COOKIE_CREATING) |
178 (1 << FSCACHE_COOKIE_NO_DATA_YET);
179
180 /* ask the cache to allocate objects for this cookie and its parent
181 * chain */
182 ret = fscache_alloc_object(cache, cookie);
183 if (ret < 0) {
184 up_read(&fscache_addremove_sem);
185 _leave(" = %d", ret);
186 return ret;
187 }
188
189 /* pass on how big the object we're caching is supposed to be */
190 cookie->def->get_attr(cookie->netfs_data, &i_size);
191
192 spin_lock(&cookie->lock);
193 if (hlist_empty(&cookie->backing_objects)) {
194 spin_unlock(&cookie->lock);
195 goto unavailable;
196 }
197
198 object = hlist_entry(cookie->backing_objects.first,
199 struct fscache_object, cookie_link);
200
201 fscache_set_store_limit(object, i_size);
202
203 /* initiate the process of looking up all the objects in the chain
204 * (done by fscache_initialise_object()) */
205 fscache_enqueue_object(object);
206
207 spin_unlock(&cookie->lock);
208
209 /* we may be required to wait for lookup to complete at this point */
210 if (!fscache_defer_lookup) {
211 _debug("non-deferred lookup %p", &cookie->flags);
212 wait_on_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP,
213 fscache_wait_bit, TASK_UNINTERRUPTIBLE);
214 _debug("complete");
215 if (test_bit(FSCACHE_COOKIE_UNAVAILABLE, &cookie->flags))
216 goto unavailable;
217 }
218
219 up_read(&fscache_addremove_sem);
220 _leave(" = 0 [deferred]");
221 return 0;
222
223unavailable:
224 up_read(&fscache_addremove_sem);
225 _leave(" = -ENOBUFS");
226 return -ENOBUFS;
227}
228
229/*
230 * recursively allocate cache object records for a cookie/cache combination
231 * - caller must be holding the addremove sem
232 */
233static int fscache_alloc_object(struct fscache_cache *cache,
234 struct fscache_cookie *cookie)
235{
236 struct fscache_object *object;
237 struct hlist_node *_n;
238 int ret;
239
240 _enter("%p,%p{%s}", cache, cookie, cookie->def->name);
241
242 spin_lock(&cookie->lock);
243 hlist_for_each_entry(object, _n, &cookie->backing_objects,
244 cookie_link) {
245 if (object->cache == cache)
246 goto object_already_extant;
247 }
248 spin_unlock(&cookie->lock);
249
250 /* ask the cache to allocate an object (we may end up with duplicate
251 * objects at this stage, but we sort that out later) */
252 object = cache->ops->alloc_object(cache, cookie);
253 if (IS_ERR(object)) {
254 fscache_stat(&fscache_n_object_no_alloc);
255 ret = PTR_ERR(object);
256 goto error;
257 }
258
259 fscache_stat(&fscache_n_object_alloc);
260
261 object->debug_id = atomic_inc_return(&fscache_object_debug_id);
262
263 _debug("ALLOC OBJ%x: %s {%lx}",
264 object->debug_id, cookie->def->name, object->events);
265
266 ret = fscache_alloc_object(cache, cookie->parent);
267 if (ret < 0)
268 goto error_put;
269
270 /* only attach if we managed to allocate all we needed, otherwise
271 * discard the object we just allocated and instead use the one
272 * attached to the cookie */
273 if (fscache_attach_object(cookie, object) < 0)
274 cache->ops->put_object(object);
275
276 _leave(" = 0");
277 return 0;
278
279object_already_extant:
280 ret = -ENOBUFS;
281 if (object->state >= FSCACHE_OBJECT_DYING) {
282 spin_unlock(&cookie->lock);
283 goto error;
284 }
285 spin_unlock(&cookie->lock);
286 _leave(" = 0 [found]");
287 return 0;
288
289error_put:
290 cache->ops->put_object(object);
291error:
292 _leave(" = %d", ret);
293 return ret;
294}
295
296/*
297 * attach a cache object to a cookie
298 */
299static int fscache_attach_object(struct fscache_cookie *cookie,
300 struct fscache_object *object)
301{
302 struct fscache_object *p;
303 struct fscache_cache *cache = object->cache;
304 struct hlist_node *_n;
305 int ret;
306
307 _enter("{%s},{OBJ%x}", cookie->def->name, object->debug_id);
308
309 spin_lock(&cookie->lock);
310
311 /* there may be multiple initial creations of this object, but we only
312 * want one */
313 ret = -EEXIST;
314 hlist_for_each_entry(p, _n, &cookie->backing_objects, cookie_link) {
315 if (p->cache == object->cache) {
316 if (p->state >= FSCACHE_OBJECT_DYING)
317 ret = -ENOBUFS;
318 goto cant_attach_object;
319 }
320 }
321
322 /* pin the parent object */
323 spin_lock_nested(&cookie->parent->lock, 1);
324 hlist_for_each_entry(p, _n, &cookie->parent->backing_objects,
325 cookie_link) {
326 if (p->cache == object->cache) {
327 if (p->state >= FSCACHE_OBJECT_DYING) {
328 ret = -ENOBUFS;
329 spin_unlock(&cookie->parent->lock);
330 goto cant_attach_object;
331 }
332 object->parent = p;
333 spin_lock(&p->lock);
334 p->n_children++;
335 spin_unlock(&p->lock);
336 break;
337 }
338 }
339 spin_unlock(&cookie->parent->lock);
340
341 /* attach to the cache's object list */
342 if (list_empty(&object->cache_link)) {
343 spin_lock(&cache->object_list_lock);
344 list_add(&object->cache_link, &cache->object_list);
345 spin_unlock(&cache->object_list_lock);
346 }
347
348 /* attach to the cookie */
349 object->cookie = cookie;
350 atomic_inc(&cookie->usage);
351 hlist_add_head(&object->cookie_link, &cookie->backing_objects);
352 ret = 0;
353
354cant_attach_object:
355 spin_unlock(&cookie->lock);
356 _leave(" = %d", ret);
357 return ret;
358}
359
360/*
361 * update the index entries backing a cookie
362 */
363void __fscache_update_cookie(struct fscache_cookie *cookie)
364{
365 struct fscache_object *object;
366 struct hlist_node *_p;
367
368 fscache_stat(&fscache_n_updates);
369
370 if (!cookie) {
371 fscache_stat(&fscache_n_updates_null);
372 _leave(" [no cookie]");
373 return;
374 }
375
376 _enter("{%s}", cookie->def->name);
377
378 BUG_ON(!cookie->def->get_aux);
379
380 spin_lock(&cookie->lock);
381
382 /* update the index entry on disk in each cache backing this cookie */
383 hlist_for_each_entry(object, _p,
384 &cookie->backing_objects, cookie_link) {
385 fscache_raise_event(object, FSCACHE_OBJECT_EV_UPDATE);
386 }
387
388 spin_unlock(&cookie->lock);
389 _leave("");
390}
391EXPORT_SYMBOL(__fscache_update_cookie);
392
393/*
394 * release a cookie back to the cache
395 * - the object will be marked as recyclable on disk if retire is true
396 * - all dependents of this cookie must have already been unregistered
397 * (indices/files/pages)
398 */
399void __fscache_relinquish_cookie(struct fscache_cookie *cookie, int retire)
400{
401 struct fscache_cache *cache;
402 struct fscache_object *object;
403 unsigned long event;
404
405 fscache_stat(&fscache_n_relinquishes);
406
407 if (!cookie) {
408 fscache_stat(&fscache_n_relinquishes_null);
409 _leave(" [no cookie]");
410 return;
411 }
412
413 _enter("%p{%s,%p},%d",
414 cookie, cookie->def->name, cookie->netfs_data, retire);
415
416 if (atomic_read(&cookie->n_children) != 0) {
417 printk(KERN_ERR "FS-Cache: Cookie '%s' still has children\n",
418 cookie->def->name);
419 BUG();
420 }
421
422 /* wait for the cookie to finish being instantiated (or to fail) */
423 if (test_bit(FSCACHE_COOKIE_CREATING, &cookie->flags)) {
424 fscache_stat(&fscache_n_relinquishes_waitcrt);
425 wait_on_bit(&cookie->flags, FSCACHE_COOKIE_CREATING,
426 fscache_wait_bit, TASK_UNINTERRUPTIBLE);
427 }
428
429 event = retire ? FSCACHE_OBJECT_EV_RETIRE : FSCACHE_OBJECT_EV_RELEASE;
430
431 /* detach pointers back to the netfs */
432 spin_lock(&cookie->lock);
433
434 cookie->netfs_data = NULL;
435 cookie->def = NULL;
436
437 /* break links with all the active objects */
438 while (!hlist_empty(&cookie->backing_objects)) {
439 object = hlist_entry(cookie->backing_objects.first,
440 struct fscache_object,
441 cookie_link);
442
443 _debug("RELEASE OBJ%x", object->debug_id);
444
445 /* detach each cache object from the object cookie */
446 spin_lock(&object->lock);
447 hlist_del_init(&object->cookie_link);
448
449 cache = object->cache;
450 object->cookie = NULL;
451 fscache_raise_event(object, event);
452 spin_unlock(&object->lock);
453
454 if (atomic_dec_and_test(&cookie->usage))
455 /* the cookie refcount shouldn't be reduced to 0 yet */
456 BUG();
457 }
458
459 spin_unlock(&cookie->lock);
460
461 if (cookie->parent) {
462 ASSERTCMP(atomic_read(&cookie->parent->usage), >, 0);
463 ASSERTCMP(atomic_read(&cookie->parent->n_children), >, 0);
464 atomic_dec(&cookie->parent->n_children);
465 }
466
467 /* finally dispose of the cookie */
468 ASSERTCMP(atomic_read(&cookie->usage), >, 0);
469 fscache_cookie_put(cookie);
470
471 _leave("");
472}
473EXPORT_SYMBOL(__fscache_relinquish_cookie);
474
475/*
476 * destroy a cookie
477 */
478void __fscache_cookie_put(struct fscache_cookie *cookie)
479{
480 struct fscache_cookie *parent;
481
482 _enter("%p", cookie);
483
484 for (;;) {
485 _debug("FREE COOKIE %p", cookie);
486 parent = cookie->parent;
487 BUG_ON(!hlist_empty(&cookie->backing_objects));
488 kmem_cache_free(fscache_cookie_jar, cookie);
489
490 if (!parent)
491 break;
492
493 cookie = parent;
494 BUG_ON(atomic_read(&cookie->usage) <= 0);
495 if (!atomic_dec_and_test(&cookie->usage))
496 break;
497 }
498
499 _leave("");
500}
diff --git a/fs/fscache/fsdef.c b/fs/fscache/fsdef.c
new file mode 100644
index 000000000000..f5b4baee7352
--- /dev/null
+++ b/fs/fscache/fsdef.c
@@ -0,0 +1,144 @@
1/* Filesystem index definition
2 *
3 * Copyright (C) 2004-2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL CACHE
13#include <linux/module.h>
14#include "internal.h"
15
16static uint16_t fscache_fsdef_netfs_get_key(const void *cookie_netfs_data,
17 void *buffer, uint16_t bufmax);
18
19static uint16_t fscache_fsdef_netfs_get_aux(const void *cookie_netfs_data,
20 void *buffer, uint16_t bufmax);
21
22static
23enum fscache_checkaux fscache_fsdef_netfs_check_aux(void *cookie_netfs_data,
24 const void *data,
25 uint16_t datalen);
26
27/*
28 * The root index is owned by FS-Cache itself.
29 *
30 * When a netfs requests caching facilities, FS-Cache will, if one doesn't
31 * already exist, create an entry in the root index with the key being the name
32 * of the netfs ("AFS" for example), and the auxiliary data holding the index
33 * structure version supplied by the netfs:
34 *
35 * FSDEF
36 * |
37 * +-----------+
38 * | |
39 * NFS AFS
40 * [v=1] [v=1]
41 *
42 * If an entry with the appropriate name does already exist, the version is
43 * compared. If the version is different, the entire subtree from that entry
44 * will be discarded and a new entry created.
45 *
46 * The new entry will be an index, and a cookie referring to it will be passed
47 * to the netfs. This is then the root handle by which the netfs accesses the
48 * cache. It can create whatever objects it likes in that index, including
49 * further indices.
50 */
51static struct fscache_cookie_def fscache_fsdef_index_def = {
52 .name = ".FS-Cache",
53 .type = FSCACHE_COOKIE_TYPE_INDEX,
54};
55
56struct fscache_cookie fscache_fsdef_index = {
57 .usage = ATOMIC_INIT(1),
58 .lock = __SPIN_LOCK_UNLOCKED(fscache_fsdef_index.lock),
59 .backing_objects = HLIST_HEAD_INIT,
60 .def = &fscache_fsdef_index_def,
61};
62EXPORT_SYMBOL(fscache_fsdef_index);
63
64/*
65 * Definition of an entry in the root index. Each entry is an index, keyed to
66 * a specific netfs and only applicable to a particular version of the index
67 * structure used by that netfs.
68 */
69struct fscache_cookie_def fscache_fsdef_netfs_def = {
70 .name = "FSDEF.netfs",
71 .type = FSCACHE_COOKIE_TYPE_INDEX,
72 .get_key = fscache_fsdef_netfs_get_key,
73 .get_aux = fscache_fsdef_netfs_get_aux,
74 .check_aux = fscache_fsdef_netfs_check_aux,
75};
76
77/*
78 * get the key data for an FSDEF index record - this is the name of the netfs
79 * for which this entry is created
80 */
81static uint16_t fscache_fsdef_netfs_get_key(const void *cookie_netfs_data,
82 void *buffer, uint16_t bufmax)
83{
84 const struct fscache_netfs *netfs = cookie_netfs_data;
85 unsigned klen;
86
87 _enter("{%s.%u},", netfs->name, netfs->version);
88
89 klen = strlen(netfs->name);
90 if (klen > bufmax)
91 return 0;
92
93 memcpy(buffer, netfs->name, klen);
94 return klen;
95}
96
97/*
98 * get the auxiliary data for an FSDEF index record - this is the index
99 * structure version number of the netfs for which this version is created
100 */
101static uint16_t fscache_fsdef_netfs_get_aux(const void *cookie_netfs_data,
102 void *buffer, uint16_t bufmax)
103{
104 const struct fscache_netfs *netfs = cookie_netfs_data;
105 unsigned dlen;
106
107 _enter("{%s.%u},", netfs->name, netfs->version);
108
109 dlen = sizeof(uint32_t);
110 if (dlen > bufmax)
111 return 0;
112
113 memcpy(buffer, &netfs->version, dlen);
114 return dlen;
115}
116
117/*
118 * check that the index structure version number stored in the auxiliary data
119 * matches the one the netfs gave us
120 */
121static enum fscache_checkaux fscache_fsdef_netfs_check_aux(
122 void *cookie_netfs_data,
123 const void *data,
124 uint16_t datalen)
125{
126 struct fscache_netfs *netfs = cookie_netfs_data;
127 uint32_t version;
128
129 _enter("{%s},,%hu", netfs->name, datalen);
130
131 if (datalen != sizeof(version)) {
132 _leave(" = OBSOLETE [dl=%d v=%zu]", datalen, sizeof(version));
133 return FSCACHE_CHECKAUX_OBSOLETE;
134 }
135
136 memcpy(&version, data, sizeof(version));
137 if (version != netfs->version) {
138 _leave(" = OBSOLETE [ver=%x net=%x]", version, netfs->version);
139 return FSCACHE_CHECKAUX_OBSOLETE;
140 }
141
142 _leave(" = OKAY");
143 return FSCACHE_CHECKAUX_OKAY;
144}
diff --git a/fs/fscache/histogram.c b/fs/fscache/histogram.c
new file mode 100644
index 000000000000..bad496748a59
--- /dev/null
+++ b/fs/fscache/histogram.c
@@ -0,0 +1,109 @@
1/* FS-Cache latency histogram
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL THREAD
13#include <linux/module.h>
14#include <linux/proc_fs.h>
15#include <linux/seq_file.h>
16#include "internal.h"
17
18atomic_t fscache_obj_instantiate_histogram[HZ];
19atomic_t fscache_objs_histogram[HZ];
20atomic_t fscache_ops_histogram[HZ];
21atomic_t fscache_retrieval_delay_histogram[HZ];
22atomic_t fscache_retrieval_histogram[HZ];
23
24/*
25 * display the time-taken histogram
26 */
27static int fscache_histogram_show(struct seq_file *m, void *v)
28{
29 unsigned long index;
30 unsigned n[5], t;
31
32 switch ((unsigned long) v) {
33 case 1:
34 seq_puts(m, "JIFS SECS OBJ INST OP RUNS OBJ RUNS "
35 " RETRV DLY RETRIEVLS\n");
36 return 0;
37 case 2:
38 seq_puts(m, "===== ===== ========= ========= ========="
39 " ========= =========\n");
40 return 0;
41 default:
42 index = (unsigned long) v - 3;
43 n[0] = atomic_read(&fscache_obj_instantiate_histogram[index]);
44 n[1] = atomic_read(&fscache_ops_histogram[index]);
45 n[2] = atomic_read(&fscache_objs_histogram[index]);
46 n[3] = atomic_read(&fscache_retrieval_delay_histogram[index]);
47 n[4] = atomic_read(&fscache_retrieval_histogram[index]);
48 if (!(n[0] | n[1] | n[2] | n[3] | n[4]))
49 return 0;
50
51 t = (index * 1000) / HZ;
52
53 seq_printf(m, "%4lu 0.%03u %9u %9u %9u %9u %9u\n",
54 index, t, n[0], n[1], n[2], n[3], n[4]);
55 return 0;
56 }
57}
58
59/*
60 * set up the iterator to start reading from the first line
61 */
62static void *fscache_histogram_start(struct seq_file *m, loff_t *_pos)
63{
64 if ((unsigned long long)*_pos >= HZ + 2)
65 return NULL;
66 if (*_pos == 0)
67 *_pos = 1;
68 return (void *)(unsigned long) *_pos;
69}
70
71/*
72 * move to the next line
73 */
74static void *fscache_histogram_next(struct seq_file *m, void *v, loff_t *pos)
75{
76 (*pos)++;
77 return (unsigned long long)*pos > HZ + 2 ?
78 NULL : (void *)(unsigned long) *pos;
79}
80
81/*
82 * clean up after reading
83 */
84static void fscache_histogram_stop(struct seq_file *m, void *v)
85{
86}
87
88static const struct seq_operations fscache_histogram_ops = {
89 .start = fscache_histogram_start,
90 .stop = fscache_histogram_stop,
91 .next = fscache_histogram_next,
92 .show = fscache_histogram_show,
93};
94
95/*
96 * open "/proc/fs/fscache/histogram" to provide latency data
97 */
98static int fscache_histogram_open(struct inode *inode, struct file *file)
99{
100 return seq_open(file, &fscache_histogram_ops);
101}
102
103const struct file_operations fscache_histogram_fops = {
104 .owner = THIS_MODULE,
105 .open = fscache_histogram_open,
106 .read = seq_read,
107 .llseek = seq_lseek,
108 .release = seq_release,
109};
diff --git a/fs/fscache/internal.h b/fs/fscache/internal.h
new file mode 100644
index 000000000000..e0cbd16f6dc9
--- /dev/null
+++ b/fs/fscache/internal.h
@@ -0,0 +1,380 @@
1/* Internal definitions for FS-Cache
2 *
3 * Copyright (C) 2004-2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12/*
13 * Lock order, in the order in which multiple locks should be obtained:
14 * - fscache_addremove_sem
15 * - cookie->lock
16 * - cookie->parent->lock
17 * - cache->object_list_lock
18 * - object->lock
19 * - object->parent->lock
20 * - fscache_thread_lock
21 *
22 */
23
24#include <linux/fscache-cache.h>
25#include <linux/sched.h>
26
27#define FSCACHE_MIN_THREADS 4
28#define FSCACHE_MAX_THREADS 32
29
30/*
31 * fsc-cache.c
32 */
33extern struct list_head fscache_cache_list;
34extern struct rw_semaphore fscache_addremove_sem;
35
36extern struct fscache_cache *fscache_select_cache_for_object(
37 struct fscache_cookie *);
38
39/*
40 * fsc-cookie.c
41 */
42extern struct kmem_cache *fscache_cookie_jar;
43
44extern void fscache_cookie_init_once(void *);
45extern void __fscache_cookie_put(struct fscache_cookie *);
46
47/*
48 * fsc-fsdef.c
49 */
50extern struct fscache_cookie fscache_fsdef_index;
51extern struct fscache_cookie_def fscache_fsdef_netfs_def;
52
53/*
54 * fsc-histogram.c
55 */
56#ifdef CONFIG_FSCACHE_HISTOGRAM
57extern atomic_t fscache_obj_instantiate_histogram[HZ];
58extern atomic_t fscache_objs_histogram[HZ];
59extern atomic_t fscache_ops_histogram[HZ];
60extern atomic_t fscache_retrieval_delay_histogram[HZ];
61extern atomic_t fscache_retrieval_histogram[HZ];
62
63static inline void fscache_hist(atomic_t histogram[], unsigned long start_jif)
64{
65 unsigned long jif = jiffies - start_jif;
66 if (jif >= HZ)
67 jif = HZ - 1;
68 atomic_inc(&histogram[jif]);
69}
70
71extern const struct file_operations fscache_histogram_fops;
72
73#else
74#define fscache_hist(hist, start_jif) do {} while (0)
75#endif
76
77/*
78 * fsc-main.c
79 */
80extern unsigned fscache_defer_lookup;
81extern unsigned fscache_defer_create;
82extern unsigned fscache_debug;
83extern struct kobject *fscache_root;
84
85extern int fscache_wait_bit(void *);
86extern int fscache_wait_bit_interruptible(void *);
87
88/*
89 * fsc-object.c
90 */
91extern void fscache_withdrawing_object(struct fscache_cache *,
92 struct fscache_object *);
93extern void fscache_enqueue_object(struct fscache_object *);
94
95/*
96 * fsc-operation.c
97 */
98extern int fscache_submit_exclusive_op(struct fscache_object *,
99 struct fscache_operation *);
100extern int fscache_submit_op(struct fscache_object *,
101 struct fscache_operation *);
102extern void fscache_abort_object(struct fscache_object *);
103extern void fscache_start_operations(struct fscache_object *);
104extern void fscache_operation_gc(struct work_struct *);
105
106/*
107 * fsc-proc.c
108 */
109#ifdef CONFIG_PROC_FS
110extern int __init fscache_proc_init(void);
111extern void fscache_proc_cleanup(void);
112#else
113#define fscache_proc_init() (0)
114#define fscache_proc_cleanup() do {} while (0)
115#endif
116
117/*
118 * fsc-stats.c
119 */
120#ifdef CONFIG_FSCACHE_STATS
121extern atomic_t fscache_n_ops_processed[FSCACHE_MAX_THREADS];
122extern atomic_t fscache_n_objs_processed[FSCACHE_MAX_THREADS];
123
124extern atomic_t fscache_n_op_pend;
125extern atomic_t fscache_n_op_run;
126extern atomic_t fscache_n_op_enqueue;
127extern atomic_t fscache_n_op_deferred_release;
128extern atomic_t fscache_n_op_release;
129extern atomic_t fscache_n_op_gc;
130
131extern atomic_t fscache_n_attr_changed;
132extern atomic_t fscache_n_attr_changed_ok;
133extern atomic_t fscache_n_attr_changed_nobufs;
134extern atomic_t fscache_n_attr_changed_nomem;
135extern atomic_t fscache_n_attr_changed_calls;
136
137extern atomic_t fscache_n_allocs;
138extern atomic_t fscache_n_allocs_ok;
139extern atomic_t fscache_n_allocs_wait;
140extern atomic_t fscache_n_allocs_nobufs;
141extern atomic_t fscache_n_alloc_ops;
142extern atomic_t fscache_n_alloc_op_waits;
143
144extern atomic_t fscache_n_retrievals;
145extern atomic_t fscache_n_retrievals_ok;
146extern atomic_t fscache_n_retrievals_wait;
147extern atomic_t fscache_n_retrievals_nodata;
148extern atomic_t fscache_n_retrievals_nobufs;
149extern atomic_t fscache_n_retrievals_intr;
150extern atomic_t fscache_n_retrievals_nomem;
151extern atomic_t fscache_n_retrieval_ops;
152extern atomic_t fscache_n_retrieval_op_waits;
153
154extern atomic_t fscache_n_stores;
155extern atomic_t fscache_n_stores_ok;
156extern atomic_t fscache_n_stores_again;
157extern atomic_t fscache_n_stores_nobufs;
158extern atomic_t fscache_n_stores_oom;
159extern atomic_t fscache_n_store_ops;
160extern atomic_t fscache_n_store_calls;
161
162extern atomic_t fscache_n_marks;
163extern atomic_t fscache_n_uncaches;
164
165extern atomic_t fscache_n_acquires;
166extern atomic_t fscache_n_acquires_null;
167extern atomic_t fscache_n_acquires_no_cache;
168extern atomic_t fscache_n_acquires_ok;
169extern atomic_t fscache_n_acquires_nobufs;
170extern atomic_t fscache_n_acquires_oom;
171
172extern atomic_t fscache_n_updates;
173extern atomic_t fscache_n_updates_null;
174extern atomic_t fscache_n_updates_run;
175
176extern atomic_t fscache_n_relinquishes;
177extern atomic_t fscache_n_relinquishes_null;
178extern atomic_t fscache_n_relinquishes_waitcrt;
179
180extern atomic_t fscache_n_cookie_index;
181extern atomic_t fscache_n_cookie_data;
182extern atomic_t fscache_n_cookie_special;
183
184extern atomic_t fscache_n_object_alloc;
185extern atomic_t fscache_n_object_no_alloc;
186extern atomic_t fscache_n_object_lookups;
187extern atomic_t fscache_n_object_lookups_negative;
188extern atomic_t fscache_n_object_lookups_positive;
189extern atomic_t fscache_n_object_created;
190extern atomic_t fscache_n_object_avail;
191extern atomic_t fscache_n_object_dead;
192
193extern atomic_t fscache_n_checkaux_none;
194extern atomic_t fscache_n_checkaux_okay;
195extern atomic_t fscache_n_checkaux_update;
196extern atomic_t fscache_n_checkaux_obsolete;
197
198static inline void fscache_stat(atomic_t *stat)
199{
200 atomic_inc(stat);
201}
202
203extern const struct file_operations fscache_stats_fops;
204#else
205
206#define fscache_stat(stat) do {} while (0)
207#endif
208
209/*
210 * raise an event on an object
211 * - if the event is not masked for that object, then the object is
212 * queued for attention by the thread pool.
213 */
214static inline void fscache_raise_event(struct fscache_object *object,
215 unsigned event)
216{
217 if (!test_and_set_bit(event, &object->events) &&
218 test_bit(event, &object->event_mask))
219 fscache_enqueue_object(object);
220}
221
222/*
223 * drop a reference to a cookie
224 */
225static inline void fscache_cookie_put(struct fscache_cookie *cookie)
226{
227 BUG_ON(atomic_read(&cookie->usage) <= 0);
228 if (atomic_dec_and_test(&cookie->usage))
229 __fscache_cookie_put(cookie);
230}
231
232/*
233 * get an extra reference to a netfs retrieval context
234 */
235static inline
236void *fscache_get_context(struct fscache_cookie *cookie, void *context)
237{
238 if (cookie->def->get_context)
239 cookie->def->get_context(cookie->netfs_data, context);
240 return context;
241}
242
243/*
244 * release a reference to a netfs retrieval context
245 */
246static inline
247void fscache_put_context(struct fscache_cookie *cookie, void *context)
248{
249 if (cookie->def->put_context)
250 cookie->def->put_context(cookie->netfs_data, context);
251}
252
253/*****************************************************************************/
254/*
255 * debug tracing
256 */
257#define dbgprintk(FMT, ...) \
258 printk(KERN_DEBUG "[%-6.6s] "FMT"\n", current->comm, ##__VA_ARGS__)
259
260/* make sure we maintain the format strings, even when debugging is disabled */
261static inline __attribute__((format(printf, 1, 2)))
262void _dbprintk(const char *fmt, ...)
263{
264}
265
266#define kenter(FMT, ...) dbgprintk("==> %s("FMT")", __func__, ##__VA_ARGS__)
267#define kleave(FMT, ...) dbgprintk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
268#define kdebug(FMT, ...) dbgprintk(FMT, ##__VA_ARGS__)
269
270#define kjournal(FMT, ...) _dbprintk(FMT, ##__VA_ARGS__)
271
272#ifdef __KDEBUG
273#define _enter(FMT, ...) kenter(FMT, ##__VA_ARGS__)
274#define _leave(FMT, ...) kleave(FMT, ##__VA_ARGS__)
275#define _debug(FMT, ...) kdebug(FMT, ##__VA_ARGS__)
276
277#elif defined(CONFIG_FSCACHE_DEBUG)
278#define _enter(FMT, ...) \
279do { \
280 if (__do_kdebug(ENTER)) \
281 kenter(FMT, ##__VA_ARGS__); \
282} while (0)
283
284#define _leave(FMT, ...) \
285do { \
286 if (__do_kdebug(LEAVE)) \
287 kleave(FMT, ##__VA_ARGS__); \
288} while (0)
289
290#define _debug(FMT, ...) \
291do { \
292 if (__do_kdebug(DEBUG)) \
293 kdebug(FMT, ##__VA_ARGS__); \
294} while (0)
295
296#else
297#define _enter(FMT, ...) _dbprintk("==> %s("FMT")", __func__, ##__VA_ARGS__)
298#define _leave(FMT, ...) _dbprintk("<== %s()"FMT"", __func__, ##__VA_ARGS__)
299#define _debug(FMT, ...) _dbprintk(FMT, ##__VA_ARGS__)
300#endif
301
302/*
303 * determine whether a particular optional debugging point should be logged
304 * - we need to go through three steps to persuade cpp to correctly join the
305 * shorthand in FSCACHE_DEBUG_LEVEL with its prefix
306 */
307#define ____do_kdebug(LEVEL, POINT) \
308 unlikely((fscache_debug & \
309 (FSCACHE_POINT_##POINT << (FSCACHE_DEBUG_ ## LEVEL * 3))))
310#define ___do_kdebug(LEVEL, POINT) \
311 ____do_kdebug(LEVEL, POINT)
312#define __do_kdebug(POINT) \
313 ___do_kdebug(FSCACHE_DEBUG_LEVEL, POINT)
314
315#define FSCACHE_DEBUG_CACHE 0
316#define FSCACHE_DEBUG_COOKIE 1
317#define FSCACHE_DEBUG_PAGE 2
318#define FSCACHE_DEBUG_OPERATION 3
319
320#define FSCACHE_POINT_ENTER 1
321#define FSCACHE_POINT_LEAVE 2
322#define FSCACHE_POINT_DEBUG 4
323
324#ifndef FSCACHE_DEBUG_LEVEL
325#define FSCACHE_DEBUG_LEVEL CACHE
326#endif
327
328/*
329 * assertions
330 */
331#if 1 /* defined(__KDEBUGALL) */
332
333#define ASSERT(X) \
334do { \
335 if (unlikely(!(X))) { \
336 printk(KERN_ERR "\n"); \
337 printk(KERN_ERR "FS-Cache: Assertion failed\n"); \
338 BUG(); \
339 } \
340} while (0)
341
342#define ASSERTCMP(X, OP, Y) \
343do { \
344 if (unlikely(!((X) OP (Y)))) { \
345 printk(KERN_ERR "\n"); \
346 printk(KERN_ERR "FS-Cache: Assertion failed\n"); \
347 printk(KERN_ERR "%lx " #OP " %lx is false\n", \
348 (unsigned long)(X), (unsigned long)(Y)); \
349 BUG(); \
350 } \
351} while (0)
352
353#define ASSERTIF(C, X) \
354do { \
355 if (unlikely((C) && !(X))) { \
356 printk(KERN_ERR "\n"); \
357 printk(KERN_ERR "FS-Cache: Assertion failed\n"); \
358 BUG(); \
359 } \
360} while (0)
361
362#define ASSERTIFCMP(C, X, OP, Y) \
363do { \
364 if (unlikely((C) && !((X) OP (Y)))) { \
365 printk(KERN_ERR "\n"); \
366 printk(KERN_ERR "FS-Cache: Assertion failed\n"); \
367 printk(KERN_ERR "%lx " #OP " %lx is false\n", \
368 (unsigned long)(X), (unsigned long)(Y)); \
369 BUG(); \
370 } \
371} while (0)
372
373#else
374
375#define ASSERT(X) do {} while (0)
376#define ASSERTCMP(X, OP, Y) do {} while (0)
377#define ASSERTIF(C, X) do {} while (0)
378#define ASSERTIFCMP(C, X, OP, Y) do {} while (0)
379
380#endif /* assert or not */
diff --git a/fs/fscache/main.c b/fs/fscache/main.c
new file mode 100644
index 000000000000..4de41b597499
--- /dev/null
+++ b/fs/fscache/main.c
@@ -0,0 +1,124 @@
1/* General filesystem local caching manager
2 *
3 * Copyright (C) 2004-2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL CACHE
13#include <linux/module.h>
14#include <linux/init.h>
15#include <linux/sched.h>
16#include <linux/completion.h>
17#include <linux/slab.h>
18#include "internal.h"
19
20MODULE_DESCRIPTION("FS Cache Manager");
21MODULE_AUTHOR("Red Hat, Inc.");
22MODULE_LICENSE("GPL");
23
24unsigned fscache_defer_lookup = 1;
25module_param_named(defer_lookup, fscache_defer_lookup, uint,
26 S_IWUSR | S_IRUGO);
27MODULE_PARM_DESC(fscache_defer_lookup,
28 "Defer cookie lookup to background thread");
29
30unsigned fscache_defer_create = 1;
31module_param_named(defer_create, fscache_defer_create, uint,
32 S_IWUSR | S_IRUGO);
33MODULE_PARM_DESC(fscache_defer_create,
34 "Defer cookie creation to background thread");
35
36unsigned fscache_debug;
37module_param_named(debug, fscache_debug, uint,
38 S_IWUSR | S_IRUGO);
39MODULE_PARM_DESC(fscache_debug,
40 "FS-Cache debugging mask");
41
42struct kobject *fscache_root;
43
44/*
45 * initialise the fs caching module
46 */
47static int __init fscache_init(void)
48{
49 int ret;
50
51 ret = slow_work_register_user();
52 if (ret < 0)
53 goto error_slow_work;
54
55 ret = fscache_proc_init();
56 if (ret < 0)
57 goto error_proc;
58
59 fscache_cookie_jar = kmem_cache_create("fscache_cookie_jar",
60 sizeof(struct fscache_cookie),
61 0,
62 0,
63 fscache_cookie_init_once);
64 if (!fscache_cookie_jar) {
65 printk(KERN_NOTICE
66 "FS-Cache: Failed to allocate a cookie jar\n");
67 ret = -ENOMEM;
68 goto error_cookie_jar;
69 }
70
71 fscache_root = kobject_create_and_add("fscache", kernel_kobj);
72 if (!fscache_root)
73 goto error_kobj;
74
75 printk(KERN_NOTICE "FS-Cache: Loaded\n");
76 return 0;
77
78error_kobj:
79 kmem_cache_destroy(fscache_cookie_jar);
80error_cookie_jar:
81 fscache_proc_cleanup();
82error_proc:
83 slow_work_unregister_user();
84error_slow_work:
85 return ret;
86}
87
88fs_initcall(fscache_init);
89
90/*
91 * clean up on module removal
92 */
93static void __exit fscache_exit(void)
94{
95 _enter("");
96
97 kobject_put(fscache_root);
98 kmem_cache_destroy(fscache_cookie_jar);
99 fscache_proc_cleanup();
100 slow_work_unregister_user();
101 printk(KERN_NOTICE "FS-Cache: Unloaded\n");
102}
103
104module_exit(fscache_exit);
105
106/*
107 * wait_on_bit() sleep function for uninterruptible waiting
108 */
109int fscache_wait_bit(void *flags)
110{
111 schedule();
112 return 0;
113}
114EXPORT_SYMBOL(fscache_wait_bit);
115
116/*
117 * wait_on_bit() sleep function for interruptible waiting
118 */
119int fscache_wait_bit_interruptible(void *flags)
120{
121 schedule();
122 return signal_pending(current);
123}
124EXPORT_SYMBOL(fscache_wait_bit_interruptible);
diff --git a/fs/fscache/netfs.c b/fs/fscache/netfs.c
new file mode 100644
index 000000000000..e028b8eb1c40
--- /dev/null
+++ b/fs/fscache/netfs.c
@@ -0,0 +1,103 @@
1/* FS-Cache netfs (client) registration
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL COOKIE
13#include <linux/module.h>
14#include <linux/slab.h>
15#include "internal.h"
16
17static LIST_HEAD(fscache_netfs_list);
18
19/*
20 * register a network filesystem for caching
21 */
22int __fscache_register_netfs(struct fscache_netfs *netfs)
23{
24 struct fscache_netfs *ptr;
25 int ret;
26
27 _enter("{%s}", netfs->name);
28
29 INIT_LIST_HEAD(&netfs->link);
30
31 /* allocate a cookie for the primary index */
32 netfs->primary_index =
33 kmem_cache_zalloc(fscache_cookie_jar, GFP_KERNEL);
34
35 if (!netfs->primary_index) {
36 _leave(" = -ENOMEM");
37 return -ENOMEM;
38 }
39
40 /* initialise the primary index cookie */
41 atomic_set(&netfs->primary_index->usage, 1);
42 atomic_set(&netfs->primary_index->n_children, 0);
43
44 netfs->primary_index->def = &fscache_fsdef_netfs_def;
45 netfs->primary_index->parent = &fscache_fsdef_index;
46 netfs->primary_index->netfs_data = netfs;
47
48 atomic_inc(&netfs->primary_index->parent->usage);
49 atomic_inc(&netfs->primary_index->parent->n_children);
50
51 spin_lock_init(&netfs->primary_index->lock);
52 INIT_HLIST_HEAD(&netfs->primary_index->backing_objects);
53
54 /* check the netfs type is not already present */
55 down_write(&fscache_addremove_sem);
56
57 ret = -EEXIST;
58 list_for_each_entry(ptr, &fscache_netfs_list, link) {
59 if (strcmp(ptr->name, netfs->name) == 0)
60 goto already_registered;
61 }
62
63 list_add(&netfs->link, &fscache_netfs_list);
64 ret = 0;
65
66 printk(KERN_NOTICE "FS-Cache: Netfs '%s' registered for caching\n",
67 netfs->name);
68
69already_registered:
70 up_write(&fscache_addremove_sem);
71
72 if (ret < 0) {
73 netfs->primary_index->parent = NULL;
74 __fscache_cookie_put(netfs->primary_index);
75 netfs->primary_index = NULL;
76 }
77
78 _leave(" = %d", ret);
79 return ret;
80}
81EXPORT_SYMBOL(__fscache_register_netfs);
82
83/*
84 * unregister a network filesystem from the cache
85 * - all cookies must have been released first
86 */
87void __fscache_unregister_netfs(struct fscache_netfs *netfs)
88{
89 _enter("{%s.%u}", netfs->name, netfs->version);
90
91 down_write(&fscache_addremove_sem);
92
93 list_del(&netfs->link);
94 fscache_relinquish_cookie(netfs->primary_index, 0);
95
96 up_write(&fscache_addremove_sem);
97
98 printk(KERN_NOTICE "FS-Cache: Netfs '%s' unregistered from caching\n",
99 netfs->name);
100
101 _leave("");
102}
103EXPORT_SYMBOL(__fscache_unregister_netfs);
diff --git a/fs/fscache/object.c b/fs/fscache/object.c
new file mode 100644
index 000000000000..392a41b1b79d
--- /dev/null
+++ b/fs/fscache/object.c
@@ -0,0 +1,810 @@
1/* FS-Cache object state machine handler
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * See Documentation/filesystems/caching/object.txt for a description of the
12 * object state machine and the in-kernel representations.
13 */
14
15#define FSCACHE_DEBUG_LEVEL COOKIE
16#include <linux/module.h>
17#include "internal.h"
18
19const char *fscache_object_states[] = {
20 [FSCACHE_OBJECT_INIT] = "OBJECT_INIT",
21 [FSCACHE_OBJECT_LOOKING_UP] = "OBJECT_LOOKING_UP",
22 [FSCACHE_OBJECT_CREATING] = "OBJECT_CREATING",
23 [FSCACHE_OBJECT_AVAILABLE] = "OBJECT_AVAILABLE",
24 [FSCACHE_OBJECT_ACTIVE] = "OBJECT_ACTIVE",
25 [FSCACHE_OBJECT_UPDATING] = "OBJECT_UPDATING",
26 [FSCACHE_OBJECT_DYING] = "OBJECT_DYING",
27 [FSCACHE_OBJECT_LC_DYING] = "OBJECT_LC_DYING",
28 [FSCACHE_OBJECT_ABORT_INIT] = "OBJECT_ABORT_INIT",
29 [FSCACHE_OBJECT_RELEASING] = "OBJECT_RELEASING",
30 [FSCACHE_OBJECT_RECYCLING] = "OBJECT_RECYCLING",
31 [FSCACHE_OBJECT_WITHDRAWING] = "OBJECT_WITHDRAWING",
32 [FSCACHE_OBJECT_DEAD] = "OBJECT_DEAD",
33};
34EXPORT_SYMBOL(fscache_object_states);
35
36static void fscache_object_slow_work_put_ref(struct slow_work *);
37static int fscache_object_slow_work_get_ref(struct slow_work *);
38static void fscache_object_slow_work_execute(struct slow_work *);
39static void fscache_initialise_object(struct fscache_object *);
40static void fscache_lookup_object(struct fscache_object *);
41static void fscache_object_available(struct fscache_object *);
42static void fscache_release_object(struct fscache_object *);
43static void fscache_withdraw_object(struct fscache_object *);
44static void fscache_enqueue_dependents(struct fscache_object *);
45static void fscache_dequeue_object(struct fscache_object *);
46
47const struct slow_work_ops fscache_object_slow_work_ops = {
48 .get_ref = fscache_object_slow_work_get_ref,
49 .put_ref = fscache_object_slow_work_put_ref,
50 .execute = fscache_object_slow_work_execute,
51};
52EXPORT_SYMBOL(fscache_object_slow_work_ops);
53
54/*
55 * we need to notify the parent when an op completes that we had outstanding
56 * upon it
57 */
58static inline void fscache_done_parent_op(struct fscache_object *object)
59{
60 struct fscache_object *parent = object->parent;
61
62 _enter("OBJ%x {OBJ%x,%x}",
63 object->debug_id, parent->debug_id, parent->n_ops);
64
65 spin_lock_nested(&parent->lock, 1);
66 parent->n_ops--;
67 parent->n_obj_ops--;
68 if (parent->n_ops == 0)
69 fscache_raise_event(parent, FSCACHE_OBJECT_EV_CLEARED);
70 spin_unlock(&parent->lock);
71}
72
73/*
74 * process events that have been sent to an object's state machine
75 * - initiates parent lookup
76 * - does object lookup
77 * - does object creation
78 * - does object recycling and retirement
79 * - does object withdrawal
80 */
81static void fscache_object_state_machine(struct fscache_object *object)
82{
83 enum fscache_object_state new_state;
84
85 ASSERT(object != NULL);
86
87 _enter("{OBJ%x,%s,%lx}",
88 object->debug_id, fscache_object_states[object->state],
89 object->events);
90
91 switch (object->state) {
92 /* wait for the parent object to become ready */
93 case FSCACHE_OBJECT_INIT:
94 object->event_mask =
95 ULONG_MAX & ~(1 << FSCACHE_OBJECT_EV_CLEARED);
96 fscache_initialise_object(object);
97 goto done;
98
99 /* look up the object metadata on disk */
100 case FSCACHE_OBJECT_LOOKING_UP:
101 fscache_lookup_object(object);
102 goto lookup_transit;
103
104 /* create the object metadata on disk */
105 case FSCACHE_OBJECT_CREATING:
106 fscache_lookup_object(object);
107 goto lookup_transit;
108
109 /* handle an object becoming available; start pending
110 * operations and queue dependent operations for processing */
111 case FSCACHE_OBJECT_AVAILABLE:
112 fscache_object_available(object);
113 goto active_transit;
114
115 /* normal running state */
116 case FSCACHE_OBJECT_ACTIVE:
117 goto active_transit;
118
119 /* update the object metadata on disk */
120 case FSCACHE_OBJECT_UPDATING:
121 clear_bit(FSCACHE_OBJECT_EV_UPDATE, &object->events);
122 fscache_stat(&fscache_n_updates_run);
123 object->cache->ops->update_object(object);
124 goto active_transit;
125
126 /* handle an object dying during lookup or creation */
127 case FSCACHE_OBJECT_LC_DYING:
128 object->event_mask &= ~(1 << FSCACHE_OBJECT_EV_UPDATE);
129 object->cache->ops->lookup_complete(object);
130
131 spin_lock(&object->lock);
132 object->state = FSCACHE_OBJECT_DYING;
133 if (test_and_clear_bit(FSCACHE_COOKIE_CREATING,
134 &object->cookie->flags))
135 wake_up_bit(&object->cookie->flags,
136 FSCACHE_COOKIE_CREATING);
137 spin_unlock(&object->lock);
138
139 fscache_done_parent_op(object);
140
141 /* wait for completion of all active operations on this object
142 * and the death of all child objects of this object */
143 case FSCACHE_OBJECT_DYING:
144 dying:
145 clear_bit(FSCACHE_OBJECT_EV_CLEARED, &object->events);
146 spin_lock(&object->lock);
147 _debug("dying OBJ%x {%d,%d}",
148 object->debug_id, object->n_ops, object->n_children);
149 if (object->n_ops == 0 && object->n_children == 0) {
150 object->event_mask &=
151 ~(1 << FSCACHE_OBJECT_EV_CLEARED);
152 object->event_mask |=
153 (1 << FSCACHE_OBJECT_EV_WITHDRAW) |
154 (1 << FSCACHE_OBJECT_EV_RETIRE) |
155 (1 << FSCACHE_OBJECT_EV_RELEASE) |
156 (1 << FSCACHE_OBJECT_EV_ERROR);
157 } else {
158 object->event_mask &=
159 ~((1 << FSCACHE_OBJECT_EV_WITHDRAW) |
160 (1 << FSCACHE_OBJECT_EV_RETIRE) |
161 (1 << FSCACHE_OBJECT_EV_RELEASE) |
162 (1 << FSCACHE_OBJECT_EV_ERROR));
163 object->event_mask |=
164 1 << FSCACHE_OBJECT_EV_CLEARED;
165 }
166 spin_unlock(&object->lock);
167 fscache_enqueue_dependents(object);
168 goto terminal_transit;
169
170 /* handle an abort during initialisation */
171 case FSCACHE_OBJECT_ABORT_INIT:
172 _debug("handle abort init %lx", object->events);
173 object->event_mask &= ~(1 << FSCACHE_OBJECT_EV_UPDATE);
174
175 spin_lock(&object->lock);
176 fscache_dequeue_object(object);
177
178 object->state = FSCACHE_OBJECT_DYING;
179 if (test_and_clear_bit(FSCACHE_COOKIE_CREATING,
180 &object->cookie->flags))
181 wake_up_bit(&object->cookie->flags,
182 FSCACHE_COOKIE_CREATING);
183 spin_unlock(&object->lock);
184 goto dying;
185
186 /* handle the netfs releasing an object and possibly marking it
187 * obsolete too */
188 case FSCACHE_OBJECT_RELEASING:
189 case FSCACHE_OBJECT_RECYCLING:
190 object->event_mask &=
191 ~((1 << FSCACHE_OBJECT_EV_WITHDRAW) |
192 (1 << FSCACHE_OBJECT_EV_RETIRE) |
193 (1 << FSCACHE_OBJECT_EV_RELEASE) |
194 (1 << FSCACHE_OBJECT_EV_ERROR));
195 fscache_release_object(object);
196 spin_lock(&object->lock);
197 object->state = FSCACHE_OBJECT_DEAD;
198 spin_unlock(&object->lock);
199 fscache_stat(&fscache_n_object_dead);
200 goto terminal_transit;
201
202 /* handle the parent cache of this object being withdrawn from
203 * active service */
204 case FSCACHE_OBJECT_WITHDRAWING:
205 object->event_mask &=
206 ~((1 << FSCACHE_OBJECT_EV_WITHDRAW) |
207 (1 << FSCACHE_OBJECT_EV_RETIRE) |
208 (1 << FSCACHE_OBJECT_EV_RELEASE) |
209 (1 << FSCACHE_OBJECT_EV_ERROR));
210 fscache_withdraw_object(object);
211 spin_lock(&object->lock);
212 object->state = FSCACHE_OBJECT_DEAD;
213 spin_unlock(&object->lock);
214 fscache_stat(&fscache_n_object_dead);
215 goto terminal_transit;
216
217 /* complain about the object being woken up once it is
218 * deceased */
219 case FSCACHE_OBJECT_DEAD:
220 printk(KERN_ERR "FS-Cache:"
221 " Unexpected event in dead state %lx\n",
222 object->events & object->event_mask);
223 BUG();
224
225 default:
226 printk(KERN_ERR "FS-Cache: Unknown object state %u\n",
227 object->state);
228 BUG();
229 }
230
231 /* determine the transition from a lookup state */
232lookup_transit:
233 switch (fls(object->events & object->event_mask) - 1) {
234 case FSCACHE_OBJECT_EV_WITHDRAW:
235 case FSCACHE_OBJECT_EV_RETIRE:
236 case FSCACHE_OBJECT_EV_RELEASE:
237 case FSCACHE_OBJECT_EV_ERROR:
238 new_state = FSCACHE_OBJECT_LC_DYING;
239 goto change_state;
240 case FSCACHE_OBJECT_EV_REQUEUE:
241 goto done;
242 case -1:
243 goto done; /* sleep until event */
244 default:
245 goto unsupported_event;
246 }
247
248 /* determine the transition from an active state */
249active_transit:
250 switch (fls(object->events & object->event_mask) - 1) {
251 case FSCACHE_OBJECT_EV_WITHDRAW:
252 case FSCACHE_OBJECT_EV_RETIRE:
253 case FSCACHE_OBJECT_EV_RELEASE:
254 case FSCACHE_OBJECT_EV_ERROR:
255 new_state = FSCACHE_OBJECT_DYING;
256 goto change_state;
257 case FSCACHE_OBJECT_EV_UPDATE:
258 new_state = FSCACHE_OBJECT_UPDATING;
259 goto change_state;
260 case -1:
261 new_state = FSCACHE_OBJECT_ACTIVE;
262 goto change_state; /* sleep until event */
263 default:
264 goto unsupported_event;
265 }
266
267 /* determine the transition from a terminal state */
268terminal_transit:
269 switch (fls(object->events & object->event_mask) - 1) {
270 case FSCACHE_OBJECT_EV_WITHDRAW:
271 new_state = FSCACHE_OBJECT_WITHDRAWING;
272 goto change_state;
273 case FSCACHE_OBJECT_EV_RETIRE:
274 new_state = FSCACHE_OBJECT_RECYCLING;
275 goto change_state;
276 case FSCACHE_OBJECT_EV_RELEASE:
277 new_state = FSCACHE_OBJECT_RELEASING;
278 goto change_state;
279 case FSCACHE_OBJECT_EV_ERROR:
280 new_state = FSCACHE_OBJECT_WITHDRAWING;
281 goto change_state;
282 case FSCACHE_OBJECT_EV_CLEARED:
283 new_state = FSCACHE_OBJECT_DYING;
284 goto change_state;
285 case -1:
286 goto done; /* sleep until event */
287 default:
288 goto unsupported_event;
289 }
290
291change_state:
292 spin_lock(&object->lock);
293 object->state = new_state;
294 spin_unlock(&object->lock);
295
296done:
297 _leave(" [->%s]", fscache_object_states[object->state]);
298 return;
299
300unsupported_event:
301 printk(KERN_ERR "FS-Cache:"
302 " Unsupported event %lx [mask %lx] in state %s\n",
303 object->events, object->event_mask,
304 fscache_object_states[object->state]);
305 BUG();
306}
307
308/*
309 * execute an object
310 */
311static void fscache_object_slow_work_execute(struct slow_work *work)
312{
313 struct fscache_object *object =
314 container_of(work, struct fscache_object, work);
315 unsigned long start;
316
317 _enter("{OBJ%x}", object->debug_id);
318
319 clear_bit(FSCACHE_OBJECT_EV_REQUEUE, &object->events);
320
321 start = jiffies;
322 fscache_object_state_machine(object);
323 fscache_hist(fscache_objs_histogram, start);
324 if (object->events & object->event_mask)
325 fscache_enqueue_object(object);
326}
327
328/*
329 * initialise an object
330 * - check the specified object's parent to see if we can make use of it
331 * immediately to do a creation
332 * - we may need to start the process of creating a parent and we need to wait
333 * for the parent's lookup and creation to complete if it's not there yet
334 * - an object's cookie is pinned until we clear FSCACHE_COOKIE_CREATING on the
335 * leaf-most cookies of the object and all its children
336 */
337static void fscache_initialise_object(struct fscache_object *object)
338{
339 struct fscache_object *parent;
340
341 _enter("");
342 ASSERT(object->cookie != NULL);
343 ASSERT(object->cookie->parent != NULL);
344 ASSERT(list_empty(&object->work.link));
345
346 if (object->events & ((1 << FSCACHE_OBJECT_EV_ERROR) |
347 (1 << FSCACHE_OBJECT_EV_RELEASE) |
348 (1 << FSCACHE_OBJECT_EV_RETIRE) |
349 (1 << FSCACHE_OBJECT_EV_WITHDRAW))) {
350 _debug("abort init %lx", object->events);
351 spin_lock(&object->lock);
352 object->state = FSCACHE_OBJECT_ABORT_INIT;
353 spin_unlock(&object->lock);
354 return;
355 }
356
357 spin_lock(&object->cookie->lock);
358 spin_lock_nested(&object->cookie->parent->lock, 1);
359
360 parent = object->parent;
361 if (!parent) {
362 _debug("no parent");
363 set_bit(FSCACHE_OBJECT_EV_WITHDRAW, &object->events);
364 } else {
365 spin_lock(&object->lock);
366 spin_lock_nested(&parent->lock, 1);
367 _debug("parent %s", fscache_object_states[parent->state]);
368
369 if (parent->state >= FSCACHE_OBJECT_DYING) {
370 _debug("bad parent");
371 set_bit(FSCACHE_OBJECT_EV_WITHDRAW, &object->events);
372 } else if (parent->state < FSCACHE_OBJECT_AVAILABLE) {
373 _debug("wait");
374
375 /* we may get woken up in this state by child objects
376 * binding on to us, so we need to make sure we don't
377 * add ourself to the list multiple times */
378 if (list_empty(&object->dep_link)) {
379 object->cache->ops->grab_object(object);
380 list_add(&object->dep_link,
381 &parent->dependents);
382
383 /* fscache_acquire_non_index_cookie() uses this
384 * to wake the chain up */
385 if (parent->state == FSCACHE_OBJECT_INIT)
386 fscache_enqueue_object(parent);
387 }
388 } else {
389 _debug("go");
390 parent->n_ops++;
391 parent->n_obj_ops++;
392 object->lookup_jif = jiffies;
393 object->state = FSCACHE_OBJECT_LOOKING_UP;
394 set_bit(FSCACHE_OBJECT_EV_REQUEUE, &object->events);
395 }
396
397 spin_unlock(&parent->lock);
398 spin_unlock(&object->lock);
399 }
400
401 spin_unlock(&object->cookie->parent->lock);
402 spin_unlock(&object->cookie->lock);
403 _leave("");
404}
405
406/*
407 * look an object up in the cache from which it was allocated
408 * - we hold an "access lock" on the parent object, so the parent object cannot
409 * be withdrawn by either party till we've finished
410 * - an object's cookie is pinned until we clear FSCACHE_COOKIE_CREATING on the
411 * leaf-most cookies of the object and all its children
412 */
413static void fscache_lookup_object(struct fscache_object *object)
414{
415 struct fscache_cookie *cookie = object->cookie;
416 struct fscache_object *parent;
417
418 _enter("");
419
420 parent = object->parent;
421 ASSERT(parent != NULL);
422 ASSERTCMP(parent->n_ops, >, 0);
423 ASSERTCMP(parent->n_obj_ops, >, 0);
424
425 /* make sure the parent is still available */
426 ASSERTCMP(parent->state, >=, FSCACHE_OBJECT_AVAILABLE);
427
428 if (parent->state >= FSCACHE_OBJECT_DYING ||
429 test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
430 _debug("unavailable");
431 set_bit(FSCACHE_OBJECT_EV_WITHDRAW, &object->events);
432 _leave("");
433 return;
434 }
435
436 _debug("LOOKUP \"%s/%s\" in \"%s\"",
437 parent->cookie->def->name, cookie->def->name,
438 object->cache->tag->name);
439
440 fscache_stat(&fscache_n_object_lookups);
441 object->cache->ops->lookup_object(object);
442
443 if (test_bit(FSCACHE_OBJECT_EV_ERROR, &object->events))
444 set_bit(FSCACHE_COOKIE_UNAVAILABLE, &cookie->flags);
445
446 _leave("");
447}
448
449/**
450 * fscache_object_lookup_negative - Note negative cookie lookup
451 * @object: Object pointing to cookie to mark
452 *
453 * Note negative lookup, permitting those waiting to read data from an already
454 * existing backing object to continue as there's no data for them to read.
455 */
456void fscache_object_lookup_negative(struct fscache_object *object)
457{
458 struct fscache_cookie *cookie = object->cookie;
459
460 _enter("{OBJ%x,%s}",
461 object->debug_id, fscache_object_states[object->state]);
462
463 spin_lock(&object->lock);
464 if (object->state == FSCACHE_OBJECT_LOOKING_UP) {
465 fscache_stat(&fscache_n_object_lookups_negative);
466
467 /* transit here to allow write requests to begin stacking up
468 * and read requests to begin returning ENODATA */
469 object->state = FSCACHE_OBJECT_CREATING;
470 spin_unlock(&object->lock);
471
472 set_bit(FSCACHE_COOKIE_PENDING_FILL, &cookie->flags);
473 set_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
474
475 _debug("wake up lookup %p", &cookie->flags);
476 smp_mb__before_clear_bit();
477 clear_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags);
478 smp_mb__after_clear_bit();
479 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP);
480 set_bit(FSCACHE_OBJECT_EV_REQUEUE, &object->events);
481 } else {
482 ASSERTCMP(object->state, ==, FSCACHE_OBJECT_CREATING);
483 spin_unlock(&object->lock);
484 }
485
486 _leave("");
487}
488EXPORT_SYMBOL(fscache_object_lookup_negative);
489
490/**
491 * fscache_obtained_object - Note successful object lookup or creation
492 * @object: Object pointing to cookie to mark
493 *
494 * Note successful lookup and/or creation, permitting those waiting to write
495 * data to a backing object to continue.
496 *
497 * Note that after calling this, an object's cookie may be relinquished by the
498 * netfs, and so must be accessed with object lock held.
499 */
500void fscache_obtained_object(struct fscache_object *object)
501{
502 struct fscache_cookie *cookie = object->cookie;
503
504 _enter("{OBJ%x,%s}",
505 object->debug_id, fscache_object_states[object->state]);
506
507 /* if we were still looking up, then we must have a positive lookup
508 * result, in which case there may be data available */
509 spin_lock(&object->lock);
510 if (object->state == FSCACHE_OBJECT_LOOKING_UP) {
511 fscache_stat(&fscache_n_object_lookups_positive);
512
513 clear_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
514
515 object->state = FSCACHE_OBJECT_AVAILABLE;
516 spin_unlock(&object->lock);
517
518 smp_mb__before_clear_bit();
519 clear_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags);
520 smp_mb__after_clear_bit();
521 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP);
522 set_bit(FSCACHE_OBJECT_EV_REQUEUE, &object->events);
523 } else {
524 ASSERTCMP(object->state, ==, FSCACHE_OBJECT_CREATING);
525 fscache_stat(&fscache_n_object_created);
526
527 object->state = FSCACHE_OBJECT_AVAILABLE;
528 spin_unlock(&object->lock);
529 set_bit(FSCACHE_OBJECT_EV_REQUEUE, &object->events);
530 smp_wmb();
531 }
532
533 if (test_and_clear_bit(FSCACHE_COOKIE_CREATING, &cookie->flags))
534 wake_up_bit(&cookie->flags, FSCACHE_COOKIE_CREATING);
535
536 _leave("");
537}
538EXPORT_SYMBOL(fscache_obtained_object);
539
540/*
541 * handle an object that has just become available
542 */
543static void fscache_object_available(struct fscache_object *object)
544{
545 _enter("{OBJ%x}", object->debug_id);
546
547 spin_lock(&object->lock);
548
549 if (test_and_clear_bit(FSCACHE_COOKIE_CREATING, &object->cookie->flags))
550 wake_up_bit(&object->cookie->flags, FSCACHE_COOKIE_CREATING);
551
552 fscache_done_parent_op(object);
553 if (object->n_in_progress == 0) {
554 if (object->n_ops > 0) {
555 ASSERTCMP(object->n_ops, >=, object->n_obj_ops);
556 ASSERTIF(object->n_ops > object->n_obj_ops,
557 !list_empty(&object->pending_ops));
558 fscache_start_operations(object);
559 } else {
560 ASSERT(list_empty(&object->pending_ops));
561 }
562 }
563 spin_unlock(&object->lock);
564
565 object->cache->ops->lookup_complete(object);
566 fscache_enqueue_dependents(object);
567
568 fscache_hist(fscache_obj_instantiate_histogram, object->lookup_jif);
569 fscache_stat(&fscache_n_object_avail);
570
571 _leave("");
572}
573
574/*
575 * drop an object's attachments
576 */
577static void fscache_drop_object(struct fscache_object *object)
578{
579 struct fscache_object *parent = object->parent;
580 struct fscache_cache *cache = object->cache;
581
582 _enter("{OBJ%x,%d}", object->debug_id, object->n_children);
583
584 spin_lock(&cache->object_list_lock);
585 list_del_init(&object->cache_link);
586 spin_unlock(&cache->object_list_lock);
587
588 cache->ops->drop_object(object);
589
590 if (parent) {
591 _debug("release parent OBJ%x {%d}",
592 parent->debug_id, parent->n_children);
593
594 spin_lock(&parent->lock);
595 parent->n_children--;
596 if (parent->n_children == 0)
597 fscache_raise_event(parent, FSCACHE_OBJECT_EV_CLEARED);
598 spin_unlock(&parent->lock);
599 object->parent = NULL;
600 }
601
602 /* this just shifts the object release to the slow work processor */
603 object->cache->ops->put_object(object);
604
605 _leave("");
606}
607
608/*
609 * release or recycle an object that the netfs has discarded
610 */
611static void fscache_release_object(struct fscache_object *object)
612{
613 _enter("");
614
615 fscache_drop_object(object);
616}
617
618/*
619 * withdraw an object from active service
620 */
621static void fscache_withdraw_object(struct fscache_object *object)
622{
623 struct fscache_cookie *cookie;
624 bool detached;
625
626 _enter("");
627
628 spin_lock(&object->lock);
629 cookie = object->cookie;
630 if (cookie) {
631 /* need to get the cookie lock before the object lock, starting
632 * from the object pointer */
633 atomic_inc(&cookie->usage);
634 spin_unlock(&object->lock);
635
636 detached = false;
637 spin_lock(&cookie->lock);
638 spin_lock(&object->lock);
639
640 if (object->cookie == cookie) {
641 hlist_del_init(&object->cookie_link);
642 object->cookie = NULL;
643 detached = true;
644 }
645 spin_unlock(&cookie->lock);
646 fscache_cookie_put(cookie);
647 if (detached)
648 fscache_cookie_put(cookie);
649 }
650
651 spin_unlock(&object->lock);
652
653 fscache_drop_object(object);
654}
655
656/*
657 * withdraw an object from active service at the behest of the cache
658 * - need break the links to a cached object cookie
659 * - called under two situations:
660 * (1) recycler decides to reclaim an in-use object
661 * (2) a cache is unmounted
662 * - have to take care as the cookie can be being relinquished by the netfs
663 * simultaneously
664 * - the object is pinned by the caller holding a refcount on it
665 */
666void fscache_withdrawing_object(struct fscache_cache *cache,
667 struct fscache_object *object)
668{
669 bool enqueue = false;
670
671 _enter(",OBJ%x", object->debug_id);
672
673 spin_lock(&object->lock);
674 if (object->state < FSCACHE_OBJECT_WITHDRAWING) {
675 object->state = FSCACHE_OBJECT_WITHDRAWING;
676 enqueue = true;
677 }
678 spin_unlock(&object->lock);
679
680 if (enqueue)
681 fscache_enqueue_object(object);
682
683 _leave("");
684}
685
686/*
687 * allow the slow work item processor to get a ref on an object
688 */
689static int fscache_object_slow_work_get_ref(struct slow_work *work)
690{
691 struct fscache_object *object =
692 container_of(work, struct fscache_object, work);
693
694 return object->cache->ops->grab_object(object) ? 0 : -EAGAIN;
695}
696
697/*
698 * allow the slow work item processor to discard a ref on a work item
699 */
700static void fscache_object_slow_work_put_ref(struct slow_work *work)
701{
702 struct fscache_object *object =
703 container_of(work, struct fscache_object, work);
704
705 return object->cache->ops->put_object(object);
706}
707
708/*
709 * enqueue an object for metadata-type processing
710 */
711void fscache_enqueue_object(struct fscache_object *object)
712{
713 _enter("{OBJ%x}", object->debug_id);
714
715 slow_work_enqueue(&object->work);
716}
717
718/*
719 * enqueue the dependents of an object for metadata-type processing
720 * - the caller must hold the object's lock
721 * - this may cause an already locked object to wind up being processed again
722 */
723static void fscache_enqueue_dependents(struct fscache_object *object)
724{
725 struct fscache_object *dep;
726
727 _enter("{OBJ%x}", object->debug_id);
728
729 if (list_empty(&object->dependents))
730 return;
731
732 spin_lock(&object->lock);
733
734 while (!list_empty(&object->dependents)) {
735 dep = list_entry(object->dependents.next,
736 struct fscache_object, dep_link);
737 list_del_init(&dep->dep_link);
738
739
740 /* sort onto appropriate lists */
741 fscache_enqueue_object(dep);
742 dep->cache->ops->put_object(dep);
743
744 if (!list_empty(&object->dependents))
745 cond_resched_lock(&object->lock);
746 }
747
748 spin_unlock(&object->lock);
749}
750
751/*
752 * remove an object from whatever queue it's waiting on
753 * - the caller must hold object->lock
754 */
755void fscache_dequeue_object(struct fscache_object *object)
756{
757 _enter("{OBJ%x}", object->debug_id);
758
759 if (!list_empty(&object->dep_link)) {
760 spin_lock(&object->parent->lock);
761 list_del_init(&object->dep_link);
762 spin_unlock(&object->parent->lock);
763 }
764
765 _leave("");
766}
767
768/**
769 * fscache_check_aux - Ask the netfs whether an object on disk is still valid
770 * @object: The object to ask about
771 * @data: The auxiliary data for the object
772 * @datalen: The size of the auxiliary data
773 *
774 * This function consults the netfs about the coherency state of an object
775 */
776enum fscache_checkaux fscache_check_aux(struct fscache_object *object,
777 const void *data, uint16_t datalen)
778{
779 enum fscache_checkaux result;
780
781 if (!object->cookie->def->check_aux) {
782 fscache_stat(&fscache_n_checkaux_none);
783 return FSCACHE_CHECKAUX_OKAY;
784 }
785
786 result = object->cookie->def->check_aux(object->cookie->netfs_data,
787 data, datalen);
788 switch (result) {
789 /* entry okay as is */
790 case FSCACHE_CHECKAUX_OKAY:
791 fscache_stat(&fscache_n_checkaux_okay);
792 break;
793
794 /* entry requires update */
795 case FSCACHE_CHECKAUX_NEEDS_UPDATE:
796 fscache_stat(&fscache_n_checkaux_update);
797 break;
798
799 /* entry requires deletion */
800 case FSCACHE_CHECKAUX_OBSOLETE:
801 fscache_stat(&fscache_n_checkaux_obsolete);
802 break;
803
804 default:
805 BUG();
806 }
807
808 return result;
809}
810EXPORT_SYMBOL(fscache_check_aux);
diff --git a/fs/fscache/operation.c b/fs/fscache/operation.c
new file mode 100644
index 000000000000..e7f8d53b8b6b
--- /dev/null
+++ b/fs/fscache/operation.c
@@ -0,0 +1,459 @@
1/* FS-Cache worker operation management routines
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 *
11 * See Documentation/filesystems/caching/operations.txt
12 */
13
14#define FSCACHE_DEBUG_LEVEL OPERATION
15#include <linux/module.h>
16#include "internal.h"
17
18atomic_t fscache_op_debug_id;
19EXPORT_SYMBOL(fscache_op_debug_id);
20
21/**
22 * fscache_enqueue_operation - Enqueue an operation for processing
23 * @op: The operation to enqueue
24 *
25 * Enqueue an operation for processing by the FS-Cache thread pool.
26 *
27 * This will get its own ref on the object.
28 */
29void fscache_enqueue_operation(struct fscache_operation *op)
30{
31 _enter("{OBJ%x OP%x,%u}",
32 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
33
34 ASSERT(op->processor != NULL);
35 ASSERTCMP(op->object->state, >=, FSCACHE_OBJECT_AVAILABLE);
36 ASSERTCMP(atomic_read(&op->usage), >, 0);
37
38 if (list_empty(&op->pend_link)) {
39 switch (op->flags & FSCACHE_OP_TYPE) {
40 case FSCACHE_OP_FAST:
41 _debug("queue fast");
42 atomic_inc(&op->usage);
43 if (!schedule_work(&op->fast_work))
44 fscache_put_operation(op);
45 break;
46 case FSCACHE_OP_SLOW:
47 _debug("queue slow");
48 slow_work_enqueue(&op->slow_work);
49 break;
50 case FSCACHE_OP_MYTHREAD:
51 _debug("queue for caller's attention");
52 break;
53 default:
54 printk(KERN_ERR "FS-Cache: Unexpected op type %lx",
55 op->flags);
56 BUG();
57 break;
58 }
59 fscache_stat(&fscache_n_op_enqueue);
60 }
61}
62EXPORT_SYMBOL(fscache_enqueue_operation);
63
64/*
65 * start an op running
66 */
67static void fscache_run_op(struct fscache_object *object,
68 struct fscache_operation *op)
69{
70 object->n_in_progress++;
71 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
72 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
73 if (op->processor)
74 fscache_enqueue_operation(op);
75 fscache_stat(&fscache_n_op_run);
76}
77
78/*
79 * submit an exclusive operation for an object
80 * - other ops are excluded from running simultaneously with this one
81 * - this gets any extra refs it needs on an op
82 */
83int fscache_submit_exclusive_op(struct fscache_object *object,
84 struct fscache_operation *op)
85{
86 int ret;
87
88 _enter("{OBJ%x OP%x},", object->debug_id, op->debug_id);
89
90 spin_lock(&object->lock);
91 ASSERTCMP(object->n_ops, >=, object->n_in_progress);
92 ASSERTCMP(object->n_ops, >=, object->n_exclusive);
93
94 ret = -ENOBUFS;
95 if (fscache_object_is_active(object)) {
96 op->object = object;
97 object->n_ops++;
98 object->n_exclusive++; /* reads and writes must wait */
99
100 if (object->n_ops > 0) {
101 atomic_inc(&op->usage);
102 list_add_tail(&op->pend_link, &object->pending_ops);
103 fscache_stat(&fscache_n_op_pend);
104 } else if (!list_empty(&object->pending_ops)) {
105 atomic_inc(&op->usage);
106 list_add_tail(&op->pend_link, &object->pending_ops);
107 fscache_stat(&fscache_n_op_pend);
108 fscache_start_operations(object);
109 } else {
110 ASSERTCMP(object->n_in_progress, ==, 0);
111 fscache_run_op(object, op);
112 }
113
114 /* need to issue a new write op after this */
115 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
116 ret = 0;
117 } else if (object->state == FSCACHE_OBJECT_CREATING) {
118 op->object = object;
119 object->n_ops++;
120 object->n_exclusive++; /* reads and writes must wait */
121 atomic_inc(&op->usage);
122 list_add_tail(&op->pend_link, &object->pending_ops);
123 fscache_stat(&fscache_n_op_pend);
124 ret = 0;
125 } else {
126 /* not allowed to submit ops in any other state */
127 BUG();
128 }
129
130 spin_unlock(&object->lock);
131 return ret;
132}
133
134/*
135 * report an unexpected submission
136 */
137static void fscache_report_unexpected_submission(struct fscache_object *object,
138 struct fscache_operation *op,
139 unsigned long ostate)
140{
141 static bool once_only;
142 struct fscache_operation *p;
143 unsigned n;
144
145 if (once_only)
146 return;
147 once_only = true;
148
149 kdebug("unexpected submission OP%x [OBJ%x %s]",
150 op->debug_id, object->debug_id,
151 fscache_object_states[object->state]);
152 kdebug("objstate=%s [%s]",
153 fscache_object_states[object->state],
154 fscache_object_states[ostate]);
155 kdebug("objflags=%lx", object->flags);
156 kdebug("objevent=%lx [%lx]", object->events, object->event_mask);
157 kdebug("ops=%u inp=%u exc=%u",
158 object->n_ops, object->n_in_progress, object->n_exclusive);
159
160 if (!list_empty(&object->pending_ops)) {
161 n = 0;
162 list_for_each_entry(p, &object->pending_ops, pend_link) {
163 ASSERTCMP(p->object, ==, object);
164 kdebug("%p %p", op->processor, op->release);
165 n++;
166 }
167
168 kdebug("n=%u", n);
169 }
170
171 dump_stack();
172}
173
174/*
175 * submit an operation for an object
176 * - objects may be submitted only in the following states:
177 * - during object creation (write ops may be submitted)
178 * - whilst the object is active
179 * - after an I/O error incurred in one of the two above states (op rejected)
180 * - this gets any extra refs it needs on an op
181 */
182int fscache_submit_op(struct fscache_object *object,
183 struct fscache_operation *op)
184{
185 unsigned long ostate;
186 int ret;
187
188 _enter("{OBJ%x OP%x},{%u}",
189 object->debug_id, op->debug_id, atomic_read(&op->usage));
190
191 ASSERTCMP(atomic_read(&op->usage), >, 0);
192
193 spin_lock(&object->lock);
194 ASSERTCMP(object->n_ops, >=, object->n_in_progress);
195 ASSERTCMP(object->n_ops, >=, object->n_exclusive);
196
197 ostate = object->state;
198 smp_rmb();
199
200 if (fscache_object_is_active(object)) {
201 op->object = object;
202 object->n_ops++;
203
204 if (object->n_exclusive > 0) {
205 atomic_inc(&op->usage);
206 list_add_tail(&op->pend_link, &object->pending_ops);
207 fscache_stat(&fscache_n_op_pend);
208 } else if (!list_empty(&object->pending_ops)) {
209 atomic_inc(&op->usage);
210 list_add_tail(&op->pend_link, &object->pending_ops);
211 fscache_stat(&fscache_n_op_pend);
212 fscache_start_operations(object);
213 } else {
214 ASSERTCMP(object->n_exclusive, ==, 0);
215 fscache_run_op(object, op);
216 }
217 ret = 0;
218 } else if (object->state == FSCACHE_OBJECT_CREATING) {
219 op->object = object;
220 object->n_ops++;
221 atomic_inc(&op->usage);
222 list_add_tail(&op->pend_link, &object->pending_ops);
223 fscache_stat(&fscache_n_op_pend);
224 ret = 0;
225 } else if (!test_bit(FSCACHE_IOERROR, &object->cache->flags)) {
226 fscache_report_unexpected_submission(object, op, ostate);
227 ASSERT(!fscache_object_is_active(object));
228 ret = -ENOBUFS;
229 } else {
230 ret = -ENOBUFS;
231 }
232
233 spin_unlock(&object->lock);
234 return ret;
235}
236
237/*
238 * queue an object for withdrawal on error, aborting all following asynchronous
239 * operations
240 */
241void fscache_abort_object(struct fscache_object *object)
242{
243 _enter("{OBJ%x}", object->debug_id);
244
245 fscache_raise_event(object, FSCACHE_OBJECT_EV_ERROR);
246}
247
248/*
249 * jump start the operation processing on an object
250 * - caller must hold object->lock
251 */
252void fscache_start_operations(struct fscache_object *object)
253{
254 struct fscache_operation *op;
255 bool stop = false;
256
257 while (!list_empty(&object->pending_ops) && !stop) {
258 op = list_entry(object->pending_ops.next,
259 struct fscache_operation, pend_link);
260
261 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
262 if (object->n_in_progress > 0)
263 break;
264 stop = true;
265 }
266 list_del_init(&op->pend_link);
267 object->n_in_progress++;
268
269 if (test_and_clear_bit(FSCACHE_OP_WAITING, &op->flags))
270 wake_up_bit(&op->flags, FSCACHE_OP_WAITING);
271 if (op->processor)
272 fscache_enqueue_operation(op);
273
274 /* the pending queue was holding a ref on the object */
275 fscache_put_operation(op);
276 }
277
278 ASSERTCMP(object->n_in_progress, <=, object->n_ops);
279
280 _debug("woke %d ops on OBJ%x",
281 object->n_in_progress, object->debug_id);
282}
283
284/*
285 * release an operation
286 * - queues pending ops if this is the last in-progress op
287 */
288void fscache_put_operation(struct fscache_operation *op)
289{
290 struct fscache_object *object;
291 struct fscache_cache *cache;
292
293 _enter("{OBJ%x OP%x,%d}",
294 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
295
296 ASSERTCMP(atomic_read(&op->usage), >, 0);
297
298 if (!atomic_dec_and_test(&op->usage))
299 return;
300
301 _debug("PUT OP");
302 if (test_and_set_bit(FSCACHE_OP_DEAD, &op->flags))
303 BUG();
304
305 fscache_stat(&fscache_n_op_release);
306
307 if (op->release) {
308 op->release(op);
309 op->release = NULL;
310 }
311
312 object = op->object;
313
314 /* now... we may get called with the object spinlock held, so we
315 * complete the cleanup here only if we can immediately acquire the
316 * lock, and defer it otherwise */
317 if (!spin_trylock(&object->lock)) {
318 _debug("defer put");
319 fscache_stat(&fscache_n_op_deferred_release);
320
321 cache = object->cache;
322 spin_lock(&cache->op_gc_list_lock);
323 list_add_tail(&op->pend_link, &cache->op_gc_list);
324 spin_unlock(&cache->op_gc_list_lock);
325 schedule_work(&cache->op_gc);
326 _leave(" [defer]");
327 return;
328 }
329
330 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
331 ASSERTCMP(object->n_exclusive, >, 0);
332 object->n_exclusive--;
333 }
334
335 ASSERTCMP(object->n_in_progress, >, 0);
336 object->n_in_progress--;
337 if (object->n_in_progress == 0)
338 fscache_start_operations(object);
339
340 ASSERTCMP(object->n_ops, >, 0);
341 object->n_ops--;
342 if (object->n_ops == 0)
343 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
344
345 spin_unlock(&object->lock);
346
347 kfree(op);
348 _leave(" [done]");
349}
350EXPORT_SYMBOL(fscache_put_operation);
351
352/*
353 * garbage collect operations that have had their release deferred
354 */
355void fscache_operation_gc(struct work_struct *work)
356{
357 struct fscache_operation *op;
358 struct fscache_object *object;
359 struct fscache_cache *cache =
360 container_of(work, struct fscache_cache, op_gc);
361 int count = 0;
362
363 _enter("");
364
365 do {
366 spin_lock(&cache->op_gc_list_lock);
367 if (list_empty(&cache->op_gc_list)) {
368 spin_unlock(&cache->op_gc_list_lock);
369 break;
370 }
371
372 op = list_entry(cache->op_gc_list.next,
373 struct fscache_operation, pend_link);
374 list_del(&op->pend_link);
375 spin_unlock(&cache->op_gc_list_lock);
376
377 object = op->object;
378
379 _debug("GC DEFERRED REL OBJ%x OP%x",
380 object->debug_id, op->debug_id);
381 fscache_stat(&fscache_n_op_gc);
382
383 ASSERTCMP(atomic_read(&op->usage), ==, 0);
384
385 spin_lock(&object->lock);
386 if (test_bit(FSCACHE_OP_EXCLUSIVE, &op->flags)) {
387 ASSERTCMP(object->n_exclusive, >, 0);
388 object->n_exclusive--;
389 }
390
391 ASSERTCMP(object->n_in_progress, >, 0);
392 object->n_in_progress--;
393 if (object->n_in_progress == 0)
394 fscache_start_operations(object);
395
396 ASSERTCMP(object->n_ops, >, 0);
397 object->n_ops--;
398 if (object->n_ops == 0)
399 fscache_raise_event(object, FSCACHE_OBJECT_EV_CLEARED);
400
401 spin_unlock(&object->lock);
402
403 } while (count++ < 20);
404
405 if (!list_empty(&cache->op_gc_list))
406 schedule_work(&cache->op_gc);
407
408 _leave("");
409}
410
411/*
412 * allow the slow work item processor to get a ref on an operation
413 */
414static int fscache_op_get_ref(struct slow_work *work)
415{
416 struct fscache_operation *op =
417 container_of(work, struct fscache_operation, slow_work);
418
419 atomic_inc(&op->usage);
420 return 0;
421}
422
423/*
424 * allow the slow work item processor to discard a ref on an operation
425 */
426static void fscache_op_put_ref(struct slow_work *work)
427{
428 struct fscache_operation *op =
429 container_of(work, struct fscache_operation, slow_work);
430
431 fscache_put_operation(op);
432}
433
434/*
435 * execute an operation using the slow thread pool to provide processing context
436 * - the caller holds a ref to this object, so we don't need to hold one
437 */
438static void fscache_op_execute(struct slow_work *work)
439{
440 struct fscache_operation *op =
441 container_of(work, struct fscache_operation, slow_work);
442 unsigned long start;
443
444 _enter("{OBJ%x OP%x,%d}",
445 op->object->debug_id, op->debug_id, atomic_read(&op->usage));
446
447 ASSERT(op->processor != NULL);
448 start = jiffies;
449 op->processor(op);
450 fscache_hist(fscache_ops_histogram, start);
451
452 _leave("");
453}
454
455const struct slow_work_ops fscache_op_slow_work_ops = {
456 .get_ref = fscache_op_get_ref,
457 .put_ref = fscache_op_put_ref,
458 .execute = fscache_op_execute,
459};
diff --git a/fs/fscache/page.c b/fs/fscache/page.c
new file mode 100644
index 000000000000..2568e0eb644f
--- /dev/null
+++ b/fs/fscache/page.c
@@ -0,0 +1,816 @@
1/* Cache page management and data I/O routines
2 *
3 * Copyright (C) 2004-2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL PAGE
13#include <linux/module.h>
14#include <linux/fscache-cache.h>
15#include <linux/buffer_head.h>
16#include <linux/pagevec.h>
17#include "internal.h"
18
19/*
20 * check to see if a page is being written to the cache
21 */
22bool __fscache_check_page_write(struct fscache_cookie *cookie, struct page *page)
23{
24 void *val;
25
26 rcu_read_lock();
27 val = radix_tree_lookup(&cookie->stores, page->index);
28 rcu_read_unlock();
29
30 return val != NULL;
31}
32EXPORT_SYMBOL(__fscache_check_page_write);
33
34/*
35 * wait for a page to finish being written to the cache
36 */
37void __fscache_wait_on_page_write(struct fscache_cookie *cookie, struct page *page)
38{
39 wait_queue_head_t *wq = bit_waitqueue(&cookie->flags, 0);
40
41 wait_event(*wq, !__fscache_check_page_write(cookie, page));
42}
43EXPORT_SYMBOL(__fscache_wait_on_page_write);
44
45/*
46 * note that a page has finished being written to the cache
47 */
48static void fscache_end_page_write(struct fscache_cookie *cookie, struct page *page)
49{
50 struct page *xpage;
51
52 spin_lock(&cookie->lock);
53 xpage = radix_tree_delete(&cookie->stores, page->index);
54 spin_unlock(&cookie->lock);
55 ASSERT(xpage != NULL);
56
57 wake_up_bit(&cookie->flags, 0);
58}
59
60/*
61 * actually apply the changed attributes to a cache object
62 */
63static void fscache_attr_changed_op(struct fscache_operation *op)
64{
65 struct fscache_object *object = op->object;
66
67 _enter("{OBJ%x OP%x}", object->debug_id, op->debug_id);
68
69 fscache_stat(&fscache_n_attr_changed_calls);
70
71 if (fscache_object_is_active(object) &&
72 object->cache->ops->attr_changed(object) < 0)
73 fscache_abort_object(object);
74
75 _leave("");
76}
77
78/*
79 * notification that the attributes on an object have changed
80 */
81int __fscache_attr_changed(struct fscache_cookie *cookie)
82{
83 struct fscache_operation *op;
84 struct fscache_object *object;
85
86 _enter("%p", cookie);
87
88 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
89
90 fscache_stat(&fscache_n_attr_changed);
91
92 op = kzalloc(sizeof(*op), GFP_KERNEL);
93 if (!op) {
94 fscache_stat(&fscache_n_attr_changed_nomem);
95 _leave(" = -ENOMEM");
96 return -ENOMEM;
97 }
98
99 fscache_operation_init(op, NULL);
100 fscache_operation_init_slow(op, fscache_attr_changed_op);
101 op->flags = FSCACHE_OP_SLOW | (1 << FSCACHE_OP_EXCLUSIVE);
102
103 spin_lock(&cookie->lock);
104
105 if (hlist_empty(&cookie->backing_objects))
106 goto nobufs;
107 object = hlist_entry(cookie->backing_objects.first,
108 struct fscache_object, cookie_link);
109
110 if (fscache_submit_exclusive_op(object, op) < 0)
111 goto nobufs;
112 spin_unlock(&cookie->lock);
113 fscache_stat(&fscache_n_attr_changed_ok);
114 fscache_put_operation(op);
115 _leave(" = 0");
116 return 0;
117
118nobufs:
119 spin_unlock(&cookie->lock);
120 kfree(op);
121 fscache_stat(&fscache_n_attr_changed_nobufs);
122 _leave(" = %d", -ENOBUFS);
123 return -ENOBUFS;
124}
125EXPORT_SYMBOL(__fscache_attr_changed);
126
127/*
128 * handle secondary execution given to a retrieval op on behalf of the
129 * cache
130 */
131static void fscache_retrieval_work(struct work_struct *work)
132{
133 struct fscache_retrieval *op =
134 container_of(work, struct fscache_retrieval, op.fast_work);
135 unsigned long start;
136
137 _enter("{OP%x}", op->op.debug_id);
138
139 start = jiffies;
140 op->op.processor(&op->op);
141 fscache_hist(fscache_ops_histogram, start);
142 fscache_put_operation(&op->op);
143}
144
145/*
146 * release a retrieval op reference
147 */
148static void fscache_release_retrieval_op(struct fscache_operation *_op)
149{
150 struct fscache_retrieval *op =
151 container_of(_op, struct fscache_retrieval, op);
152
153 _enter("{OP%x}", op->op.debug_id);
154
155 fscache_hist(fscache_retrieval_histogram, op->start_time);
156 if (op->context)
157 fscache_put_context(op->op.object->cookie, op->context);
158
159 _leave("");
160}
161
162/*
163 * allocate a retrieval op
164 */
165static struct fscache_retrieval *fscache_alloc_retrieval(
166 struct address_space *mapping,
167 fscache_rw_complete_t end_io_func,
168 void *context)
169{
170 struct fscache_retrieval *op;
171
172 /* allocate a retrieval operation and attempt to submit it */
173 op = kzalloc(sizeof(*op), GFP_NOIO);
174 if (!op) {
175 fscache_stat(&fscache_n_retrievals_nomem);
176 return NULL;
177 }
178
179 fscache_operation_init(&op->op, fscache_release_retrieval_op);
180 op->op.flags = FSCACHE_OP_MYTHREAD | (1 << FSCACHE_OP_WAITING);
181 op->mapping = mapping;
182 op->end_io_func = end_io_func;
183 op->context = context;
184 op->start_time = jiffies;
185 INIT_WORK(&op->op.fast_work, fscache_retrieval_work);
186 INIT_LIST_HEAD(&op->to_do);
187 return op;
188}
189
190/*
191 * wait for a deferred lookup to complete
192 */
193static int fscache_wait_for_deferred_lookup(struct fscache_cookie *cookie)
194{
195 unsigned long jif;
196
197 _enter("");
198
199 if (!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags)) {
200 _leave(" = 0 [imm]");
201 return 0;
202 }
203
204 fscache_stat(&fscache_n_retrievals_wait);
205
206 jif = jiffies;
207 if (wait_on_bit(&cookie->flags, FSCACHE_COOKIE_LOOKING_UP,
208 fscache_wait_bit_interruptible,
209 TASK_INTERRUPTIBLE) != 0) {
210 fscache_stat(&fscache_n_retrievals_intr);
211 _leave(" = -ERESTARTSYS");
212 return -ERESTARTSYS;
213 }
214
215 ASSERT(!test_bit(FSCACHE_COOKIE_LOOKING_UP, &cookie->flags));
216
217 smp_rmb();
218 fscache_hist(fscache_retrieval_delay_histogram, jif);
219 _leave(" = 0 [dly]");
220 return 0;
221}
222
223/*
224 * read a page from the cache or allocate a block in which to store it
225 * - we return:
226 * -ENOMEM - out of memory, nothing done
227 * -ERESTARTSYS - interrupted
228 * -ENOBUFS - no backing object available in which to cache the block
229 * -ENODATA - no data available in the backing object for this block
230 * 0 - dispatched a read - it'll call end_io_func() when finished
231 */
232int __fscache_read_or_alloc_page(struct fscache_cookie *cookie,
233 struct page *page,
234 fscache_rw_complete_t end_io_func,
235 void *context,
236 gfp_t gfp)
237{
238 struct fscache_retrieval *op;
239 struct fscache_object *object;
240 int ret;
241
242 _enter("%p,%p,,,", cookie, page);
243
244 fscache_stat(&fscache_n_retrievals);
245
246 if (hlist_empty(&cookie->backing_objects))
247 goto nobufs;
248
249 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
250 ASSERTCMP(page, !=, NULL);
251
252 if (fscache_wait_for_deferred_lookup(cookie) < 0)
253 return -ERESTARTSYS;
254
255 op = fscache_alloc_retrieval(page->mapping, end_io_func, context);
256 if (!op) {
257 _leave(" = -ENOMEM");
258 return -ENOMEM;
259 }
260
261 spin_lock(&cookie->lock);
262
263 if (hlist_empty(&cookie->backing_objects))
264 goto nobufs_unlock;
265 object = hlist_entry(cookie->backing_objects.first,
266 struct fscache_object, cookie_link);
267
268 ASSERTCMP(object->state, >, FSCACHE_OBJECT_LOOKING_UP);
269
270 if (fscache_submit_op(object, &op->op) < 0)
271 goto nobufs_unlock;
272 spin_unlock(&cookie->lock);
273
274 fscache_stat(&fscache_n_retrieval_ops);
275
276 /* pin the netfs read context in case we need to do the actual netfs
277 * read because we've encountered a cache read failure */
278 fscache_get_context(object->cookie, op->context);
279
280 /* we wait for the operation to become active, and then process it
281 * *here*, in this thread, and not in the thread pool */
282 if (test_bit(FSCACHE_OP_WAITING, &op->op.flags)) {
283 _debug(">>> WT");
284 fscache_stat(&fscache_n_retrieval_op_waits);
285 wait_on_bit(&op->op.flags, FSCACHE_OP_WAITING,
286 fscache_wait_bit, TASK_UNINTERRUPTIBLE);
287 _debug("<<< GO");
288 }
289
290 /* ask the cache to honour the operation */
291 if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags)) {
292 ret = object->cache->ops->allocate_page(op, page, gfp);
293 if (ret == 0)
294 ret = -ENODATA;
295 } else {
296 ret = object->cache->ops->read_or_alloc_page(op, page, gfp);
297 }
298
299 if (ret == -ENOMEM)
300 fscache_stat(&fscache_n_retrievals_nomem);
301 else if (ret == -ERESTARTSYS)
302 fscache_stat(&fscache_n_retrievals_intr);
303 else if (ret == -ENODATA)
304 fscache_stat(&fscache_n_retrievals_nodata);
305 else if (ret < 0)
306 fscache_stat(&fscache_n_retrievals_nobufs);
307 else
308 fscache_stat(&fscache_n_retrievals_ok);
309
310 fscache_put_retrieval(op);
311 _leave(" = %d", ret);
312 return ret;
313
314nobufs_unlock:
315 spin_unlock(&cookie->lock);
316 kfree(op);
317nobufs:
318 fscache_stat(&fscache_n_retrievals_nobufs);
319 _leave(" = -ENOBUFS");
320 return -ENOBUFS;
321}
322EXPORT_SYMBOL(__fscache_read_or_alloc_page);
323
324/*
325 * read a list of page from the cache or allocate a block in which to store
326 * them
327 * - we return:
328 * -ENOMEM - out of memory, some pages may be being read
329 * -ERESTARTSYS - interrupted, some pages may be being read
330 * -ENOBUFS - no backing object or space available in which to cache any
331 * pages not being read
332 * -ENODATA - no data available in the backing object for some or all of
333 * the pages
334 * 0 - dispatched a read on all pages
335 *
336 * end_io_func() will be called for each page read from the cache as it is
337 * finishes being read
338 *
339 * any pages for which a read is dispatched will be removed from pages and
340 * nr_pages
341 */
342int __fscache_read_or_alloc_pages(struct fscache_cookie *cookie,
343 struct address_space *mapping,
344 struct list_head *pages,
345 unsigned *nr_pages,
346 fscache_rw_complete_t end_io_func,
347 void *context,
348 gfp_t gfp)
349{
350 fscache_pages_retrieval_func_t func;
351 struct fscache_retrieval *op;
352 struct fscache_object *object;
353 int ret;
354
355 _enter("%p,,%d,,,", cookie, *nr_pages);
356
357 fscache_stat(&fscache_n_retrievals);
358
359 if (hlist_empty(&cookie->backing_objects))
360 goto nobufs;
361
362 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
363 ASSERTCMP(*nr_pages, >, 0);
364 ASSERT(!list_empty(pages));
365
366 if (fscache_wait_for_deferred_lookup(cookie) < 0)
367 return -ERESTARTSYS;
368
369 op = fscache_alloc_retrieval(mapping, end_io_func, context);
370 if (!op)
371 return -ENOMEM;
372
373 spin_lock(&cookie->lock);
374
375 if (hlist_empty(&cookie->backing_objects))
376 goto nobufs_unlock;
377 object = hlist_entry(cookie->backing_objects.first,
378 struct fscache_object, cookie_link);
379
380 if (fscache_submit_op(object, &op->op) < 0)
381 goto nobufs_unlock;
382 spin_unlock(&cookie->lock);
383
384 fscache_stat(&fscache_n_retrieval_ops);
385
386 /* pin the netfs read context in case we need to do the actual netfs
387 * read because we've encountered a cache read failure */
388 fscache_get_context(object->cookie, op->context);
389
390 /* we wait for the operation to become active, and then process it
391 * *here*, in this thread, and not in the thread pool */
392 if (test_bit(FSCACHE_OP_WAITING, &op->op.flags)) {
393 _debug(">>> WT");
394 fscache_stat(&fscache_n_retrieval_op_waits);
395 wait_on_bit(&op->op.flags, FSCACHE_OP_WAITING,
396 fscache_wait_bit, TASK_UNINTERRUPTIBLE);
397 _debug("<<< GO");
398 }
399
400 /* ask the cache to honour the operation */
401 if (test_bit(FSCACHE_COOKIE_NO_DATA_YET, &object->cookie->flags))
402 func = object->cache->ops->allocate_pages;
403 else
404 func = object->cache->ops->read_or_alloc_pages;
405 ret = func(op, pages, nr_pages, gfp);
406
407 if (ret == -ENOMEM)
408 fscache_stat(&fscache_n_retrievals_nomem);
409 else if (ret == -ERESTARTSYS)
410 fscache_stat(&fscache_n_retrievals_intr);
411 else if (ret == -ENODATA)
412 fscache_stat(&fscache_n_retrievals_nodata);
413 else if (ret < 0)
414 fscache_stat(&fscache_n_retrievals_nobufs);
415 else
416 fscache_stat(&fscache_n_retrievals_ok);
417
418 fscache_put_retrieval(op);
419 _leave(" = %d", ret);
420 return ret;
421
422nobufs_unlock:
423 spin_unlock(&cookie->lock);
424 kfree(op);
425nobufs:
426 fscache_stat(&fscache_n_retrievals_nobufs);
427 _leave(" = -ENOBUFS");
428 return -ENOBUFS;
429}
430EXPORT_SYMBOL(__fscache_read_or_alloc_pages);
431
432/*
433 * allocate a block in the cache on which to store a page
434 * - we return:
435 * -ENOMEM - out of memory, nothing done
436 * -ERESTARTSYS - interrupted
437 * -ENOBUFS - no backing object available in which to cache the block
438 * 0 - block allocated
439 */
440int __fscache_alloc_page(struct fscache_cookie *cookie,
441 struct page *page,
442 gfp_t gfp)
443{
444 struct fscache_retrieval *op;
445 struct fscache_object *object;
446 int ret;
447
448 _enter("%p,%p,,,", cookie, page);
449
450 fscache_stat(&fscache_n_allocs);
451
452 if (hlist_empty(&cookie->backing_objects))
453 goto nobufs;
454
455 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
456 ASSERTCMP(page, !=, NULL);
457
458 if (fscache_wait_for_deferred_lookup(cookie) < 0)
459 return -ERESTARTSYS;
460
461 op = fscache_alloc_retrieval(page->mapping, NULL, NULL);
462 if (!op)
463 return -ENOMEM;
464
465 spin_lock(&cookie->lock);
466
467 if (hlist_empty(&cookie->backing_objects))
468 goto nobufs_unlock;
469 object = hlist_entry(cookie->backing_objects.first,
470 struct fscache_object, cookie_link);
471
472 if (fscache_submit_op(object, &op->op) < 0)
473 goto nobufs_unlock;
474 spin_unlock(&cookie->lock);
475
476 fscache_stat(&fscache_n_alloc_ops);
477
478 if (test_bit(FSCACHE_OP_WAITING, &op->op.flags)) {
479 _debug(">>> WT");
480 fscache_stat(&fscache_n_alloc_op_waits);
481 wait_on_bit(&op->op.flags, FSCACHE_OP_WAITING,
482 fscache_wait_bit, TASK_UNINTERRUPTIBLE);
483 _debug("<<< GO");
484 }
485
486 /* ask the cache to honour the operation */
487 ret = object->cache->ops->allocate_page(op, page, gfp);
488
489 if (ret < 0)
490 fscache_stat(&fscache_n_allocs_nobufs);
491 else
492 fscache_stat(&fscache_n_allocs_ok);
493
494 fscache_put_retrieval(op);
495 _leave(" = %d", ret);
496 return ret;
497
498nobufs_unlock:
499 spin_unlock(&cookie->lock);
500 kfree(op);
501nobufs:
502 fscache_stat(&fscache_n_allocs_nobufs);
503 _leave(" = -ENOBUFS");
504 return -ENOBUFS;
505}
506EXPORT_SYMBOL(__fscache_alloc_page);
507
508/*
509 * release a write op reference
510 */
511static void fscache_release_write_op(struct fscache_operation *_op)
512{
513 _enter("{OP%x}", _op->debug_id);
514}
515
516/*
517 * perform the background storage of a page into the cache
518 */
519static void fscache_write_op(struct fscache_operation *_op)
520{
521 struct fscache_storage *op =
522 container_of(_op, struct fscache_storage, op);
523 struct fscache_object *object = op->op.object;
524 struct fscache_cookie *cookie = object->cookie;
525 struct page *page;
526 unsigned n;
527 void *results[1];
528 int ret;
529
530 _enter("{OP%x,%d}", op->op.debug_id, atomic_read(&op->op.usage));
531
532 spin_lock(&cookie->lock);
533 spin_lock(&object->lock);
534
535 if (!fscache_object_is_active(object)) {
536 spin_unlock(&object->lock);
537 spin_unlock(&cookie->lock);
538 _leave("");
539 return;
540 }
541
542 fscache_stat(&fscache_n_store_calls);
543
544 /* find a page to store */
545 page = NULL;
546 n = radix_tree_gang_lookup_tag(&cookie->stores, results, 0, 1,
547 FSCACHE_COOKIE_PENDING_TAG);
548 if (n != 1)
549 goto superseded;
550 page = results[0];
551 _debug("gang %d [%lx]", n, page->index);
552 if (page->index > op->store_limit)
553 goto superseded;
554
555 radix_tree_tag_clear(&cookie->stores, page->index,
556 FSCACHE_COOKIE_PENDING_TAG);
557
558 spin_unlock(&object->lock);
559 spin_unlock(&cookie->lock);
560
561 if (page) {
562 ret = object->cache->ops->write_page(op, page);
563 fscache_end_page_write(cookie, page);
564 page_cache_release(page);
565 if (ret < 0)
566 fscache_abort_object(object);
567 else
568 fscache_enqueue_operation(&op->op);
569 }
570
571 _leave("");
572 return;
573
574superseded:
575 /* this writer is going away and there aren't any more things to
576 * write */
577 _debug("cease");
578 clear_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags);
579 spin_unlock(&object->lock);
580 spin_unlock(&cookie->lock);
581 _leave("");
582}
583
584/*
585 * request a page be stored in the cache
586 * - returns:
587 * -ENOMEM - out of memory, nothing done
588 * -ENOBUFS - no backing object available in which to cache the page
589 * 0 - dispatched a write - it'll call end_io_func() when finished
590 *
591 * if the cookie still has a backing object at this point, that object can be
592 * in one of a few states with respect to storage processing:
593 *
594 * (1) negative lookup, object not yet created (FSCACHE_COOKIE_CREATING is
595 * set)
596 *
597 * (a) no writes yet (set FSCACHE_COOKIE_PENDING_FILL and queue deferred
598 * fill op)
599 *
600 * (b) writes deferred till post-creation (mark page for writing and
601 * return immediately)
602 *
603 * (2) negative lookup, object created, initial fill being made from netfs
604 * (FSCACHE_COOKIE_INITIAL_FILL is set)
605 *
606 * (a) fill point not yet reached this page (mark page for writing and
607 * return)
608 *
609 * (b) fill point passed this page (queue op to store this page)
610 *
611 * (3) object extant (queue op to store this page)
612 *
613 * any other state is invalid
614 */
615int __fscache_write_page(struct fscache_cookie *cookie,
616 struct page *page,
617 gfp_t gfp)
618{
619 struct fscache_storage *op;
620 struct fscache_object *object;
621 int ret;
622
623 _enter("%p,%x,", cookie, (u32) page->flags);
624
625 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
626 ASSERT(PageFsCache(page));
627
628 fscache_stat(&fscache_n_stores);
629
630 op = kzalloc(sizeof(*op), GFP_NOIO);
631 if (!op)
632 goto nomem;
633
634 fscache_operation_init(&op->op, fscache_release_write_op);
635 fscache_operation_init_slow(&op->op, fscache_write_op);
636 op->op.flags = FSCACHE_OP_SLOW | (1 << FSCACHE_OP_WAITING);
637
638 ret = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
639 if (ret < 0)
640 goto nomem_free;
641
642 ret = -ENOBUFS;
643 spin_lock(&cookie->lock);
644
645 if (hlist_empty(&cookie->backing_objects))
646 goto nobufs;
647 object = hlist_entry(cookie->backing_objects.first,
648 struct fscache_object, cookie_link);
649 if (test_bit(FSCACHE_IOERROR, &object->cache->flags))
650 goto nobufs;
651
652 /* add the page to the pending-storage radix tree on the backing
653 * object */
654 spin_lock(&object->lock);
655
656 _debug("store limit %llx", (unsigned long long) object->store_limit);
657
658 ret = radix_tree_insert(&cookie->stores, page->index, page);
659 if (ret < 0) {
660 if (ret == -EEXIST)
661 goto already_queued;
662 _debug("insert failed %d", ret);
663 goto nobufs_unlock_obj;
664 }
665
666 radix_tree_tag_set(&cookie->stores, page->index,
667 FSCACHE_COOKIE_PENDING_TAG);
668 page_cache_get(page);
669
670 /* we only want one writer at a time, but we do need to queue new
671 * writers after exclusive ops */
672 if (test_and_set_bit(FSCACHE_OBJECT_PENDING_WRITE, &object->flags))
673 goto already_pending;
674
675 spin_unlock(&object->lock);
676
677 op->op.debug_id = atomic_inc_return(&fscache_op_debug_id);
678 op->store_limit = object->store_limit;
679
680 if (fscache_submit_op(object, &op->op) < 0)
681 goto submit_failed;
682
683 spin_unlock(&cookie->lock);
684 radix_tree_preload_end();
685 fscache_stat(&fscache_n_store_ops);
686 fscache_stat(&fscache_n_stores_ok);
687
688 /* the slow work queue now carries its own ref on the object */
689 fscache_put_operation(&op->op);
690 _leave(" = 0");
691 return 0;
692
693already_queued:
694 fscache_stat(&fscache_n_stores_again);
695already_pending:
696 spin_unlock(&object->lock);
697 spin_unlock(&cookie->lock);
698 radix_tree_preload_end();
699 kfree(op);
700 fscache_stat(&fscache_n_stores_ok);
701 _leave(" = 0");
702 return 0;
703
704submit_failed:
705 radix_tree_delete(&cookie->stores, page->index);
706 page_cache_release(page);
707 ret = -ENOBUFS;
708 goto nobufs;
709
710nobufs_unlock_obj:
711 spin_unlock(&object->lock);
712nobufs:
713 spin_unlock(&cookie->lock);
714 radix_tree_preload_end();
715 kfree(op);
716 fscache_stat(&fscache_n_stores_nobufs);
717 _leave(" = -ENOBUFS");
718 return -ENOBUFS;
719
720nomem_free:
721 kfree(op);
722nomem:
723 fscache_stat(&fscache_n_stores_oom);
724 _leave(" = -ENOMEM");
725 return -ENOMEM;
726}
727EXPORT_SYMBOL(__fscache_write_page);
728
729/*
730 * remove a page from the cache
731 */
732void __fscache_uncache_page(struct fscache_cookie *cookie, struct page *page)
733{
734 struct fscache_object *object;
735
736 _enter(",%p", page);
737
738 ASSERTCMP(cookie->def->type, !=, FSCACHE_COOKIE_TYPE_INDEX);
739 ASSERTCMP(page, !=, NULL);
740
741 fscache_stat(&fscache_n_uncaches);
742
743 /* cache withdrawal may beat us to it */
744 if (!PageFsCache(page))
745 goto done;
746
747 /* get the object */
748 spin_lock(&cookie->lock);
749
750 if (hlist_empty(&cookie->backing_objects)) {
751 ClearPageFsCache(page);
752 goto done_unlock;
753 }
754
755 object = hlist_entry(cookie->backing_objects.first,
756 struct fscache_object, cookie_link);
757
758 /* there might now be stuff on disk we could read */
759 clear_bit(FSCACHE_COOKIE_NO_DATA_YET, &cookie->flags);
760
761 /* only invoke the cache backend if we managed to mark the page
762 * uncached here; this deals with synchronisation vs withdrawal */
763 if (TestClearPageFsCache(page) &&
764 object->cache->ops->uncache_page) {
765 /* the cache backend releases the cookie lock */
766 object->cache->ops->uncache_page(object, page);
767 goto done;
768 }
769
770done_unlock:
771 spin_unlock(&cookie->lock);
772done:
773 _leave("");
774}
775EXPORT_SYMBOL(__fscache_uncache_page);
776
777/**
778 * fscache_mark_pages_cached - Mark pages as being cached
779 * @op: The retrieval op pages are being marked for
780 * @pagevec: The pages to be marked
781 *
782 * Mark a bunch of netfs pages as being cached. After this is called,
783 * the netfs must call fscache_uncache_page() to remove the mark.
784 */
785void fscache_mark_pages_cached(struct fscache_retrieval *op,
786 struct pagevec *pagevec)
787{
788 struct fscache_cookie *cookie = op->op.object->cookie;
789 unsigned long loop;
790
791#ifdef CONFIG_FSCACHE_STATS
792 atomic_add(pagevec->nr, &fscache_n_marks);
793#endif
794
795 for (loop = 0; loop < pagevec->nr; loop++) {
796 struct page *page = pagevec->pages[loop];
797
798 _debug("- mark %p{%lx}", page, page->index);
799 if (TestSetPageFsCache(page)) {
800 static bool once_only;
801 if (!once_only) {
802 once_only = true;
803 printk(KERN_WARNING "FS-Cache:"
804 " Cookie type %s marked page %lx"
805 " multiple times\n",
806 cookie->def->name, page->index);
807 }
808 }
809 }
810
811 if (cookie->def->mark_pages_cached)
812 cookie->def->mark_pages_cached(cookie->netfs_data,
813 op->mapping, pagevec);
814 pagevec_reinit(pagevec);
815}
816EXPORT_SYMBOL(fscache_mark_pages_cached);
diff --git a/fs/fscache/proc.c b/fs/fscache/proc.c
new file mode 100644
index 000000000000..beeab44bc31a
--- /dev/null
+++ b/fs/fscache/proc.c
@@ -0,0 +1,68 @@
1/* FS-Cache statistics viewing interface
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL OPERATION
13#include <linux/module.h>
14#include <linux/proc_fs.h>
15#include <linux/seq_file.h>
16#include "internal.h"
17
18/*
19 * initialise the /proc/fs/fscache/ directory
20 */
21int __init fscache_proc_init(void)
22{
23 _enter("");
24
25 if (!proc_mkdir("fs/fscache", NULL))
26 goto error_dir;
27
28#ifdef CONFIG_FSCACHE_STATS
29 if (!proc_create("fs/fscache/stats", S_IFREG | 0444, NULL,
30 &fscache_stats_fops))
31 goto error_stats;
32#endif
33
34#ifdef CONFIG_FSCACHE_HISTOGRAM
35 if (!proc_create("fs/fscache/histogram", S_IFREG | 0444, NULL,
36 &fscache_histogram_fops))
37 goto error_histogram;
38#endif
39
40 _leave(" = 0");
41 return 0;
42
43#ifdef CONFIG_FSCACHE_HISTOGRAM
44error_histogram:
45#endif
46#ifdef CONFIG_FSCACHE_STATS
47 remove_proc_entry("fs/fscache/stats", NULL);
48error_stats:
49#endif
50 remove_proc_entry("fs/fscache", NULL);
51error_dir:
52 _leave(" = -ENOMEM");
53 return -ENOMEM;
54}
55
56/*
57 * clean up the /proc/fs/fscache/ directory
58 */
59void fscache_proc_cleanup(void)
60{
61#ifdef CONFIG_FSCACHE_HISTOGRAM
62 remove_proc_entry("fs/fscache/histogram", NULL);
63#endif
64#ifdef CONFIG_FSCACHE_STATS
65 remove_proc_entry("fs/fscache/stats", NULL);
66#endif
67 remove_proc_entry("fs/fscache", NULL);
68}
diff --git a/fs/fscache/stats.c b/fs/fscache/stats.c
new file mode 100644
index 000000000000..65deb99e756b
--- /dev/null
+++ b/fs/fscache/stats.c
@@ -0,0 +1,212 @@
1/* FS-Cache statistics
2 *
3 * Copyright (C) 2007 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
10 */
11
12#define FSCACHE_DEBUG_LEVEL THREAD
13#include <linux/module.h>
14#include <linux/proc_fs.h>
15#include <linux/seq_file.h>
16#include "internal.h"
17
18/*
19 * operation counters
20 */
21atomic_t fscache_n_op_pend;
22atomic_t fscache_n_op_run;
23atomic_t fscache_n_op_enqueue;
24atomic_t fscache_n_op_requeue;
25atomic_t fscache_n_op_deferred_release;
26atomic_t fscache_n_op_release;
27atomic_t fscache_n_op_gc;
28
29atomic_t fscache_n_attr_changed;
30atomic_t fscache_n_attr_changed_ok;
31atomic_t fscache_n_attr_changed_nobufs;
32atomic_t fscache_n_attr_changed_nomem;
33atomic_t fscache_n_attr_changed_calls;
34
35atomic_t fscache_n_allocs;
36atomic_t fscache_n_allocs_ok;
37atomic_t fscache_n_allocs_wait;
38atomic_t fscache_n_allocs_nobufs;
39atomic_t fscache_n_alloc_ops;
40atomic_t fscache_n_alloc_op_waits;
41
42atomic_t fscache_n_retrievals;
43atomic_t fscache_n_retrievals_ok;
44atomic_t fscache_n_retrievals_wait;
45atomic_t fscache_n_retrievals_nodata;
46atomic_t fscache_n_retrievals_nobufs;
47atomic_t fscache_n_retrievals_intr;
48atomic_t fscache_n_retrievals_nomem;
49atomic_t fscache_n_retrieval_ops;
50atomic_t fscache_n_retrieval_op_waits;
51
52atomic_t fscache_n_stores;
53atomic_t fscache_n_stores_ok;
54atomic_t fscache_n_stores_again;
55atomic_t fscache_n_stores_nobufs;
56atomic_t fscache_n_stores_oom;
57atomic_t fscache_n_store_ops;
58atomic_t fscache_n_store_calls;
59
60atomic_t fscache_n_marks;
61atomic_t fscache_n_uncaches;
62
63atomic_t fscache_n_acquires;
64atomic_t fscache_n_acquires_null;
65atomic_t fscache_n_acquires_no_cache;
66atomic_t fscache_n_acquires_ok;
67atomic_t fscache_n_acquires_nobufs;
68atomic_t fscache_n_acquires_oom;
69
70atomic_t fscache_n_updates;
71atomic_t fscache_n_updates_null;
72atomic_t fscache_n_updates_run;
73
74atomic_t fscache_n_relinquishes;
75atomic_t fscache_n_relinquishes_null;
76atomic_t fscache_n_relinquishes_waitcrt;
77
78atomic_t fscache_n_cookie_index;
79atomic_t fscache_n_cookie_data;
80atomic_t fscache_n_cookie_special;
81
82atomic_t fscache_n_object_alloc;
83atomic_t fscache_n_object_no_alloc;
84atomic_t fscache_n_object_lookups;
85atomic_t fscache_n_object_lookups_negative;
86atomic_t fscache_n_object_lookups_positive;
87atomic_t fscache_n_object_created;
88atomic_t fscache_n_object_avail;
89atomic_t fscache_n_object_dead;
90
91atomic_t fscache_n_checkaux_none;
92atomic_t fscache_n_checkaux_okay;
93atomic_t fscache_n_checkaux_update;
94atomic_t fscache_n_checkaux_obsolete;
95
96/*
97 * display the general statistics
98 */
99static int fscache_stats_show(struct seq_file *m, void *v)
100{
101 seq_puts(m, "FS-Cache statistics\n");
102
103 seq_printf(m, "Cookies: idx=%u dat=%u spc=%u\n",
104 atomic_read(&fscache_n_cookie_index),
105 atomic_read(&fscache_n_cookie_data),
106 atomic_read(&fscache_n_cookie_special));
107
108 seq_printf(m, "Objects: alc=%u nal=%u avl=%u ded=%u\n",
109 atomic_read(&fscache_n_object_alloc),
110 atomic_read(&fscache_n_object_no_alloc),
111 atomic_read(&fscache_n_object_avail),
112 atomic_read(&fscache_n_object_dead));
113 seq_printf(m, "ChkAux : non=%u ok=%u upd=%u obs=%u\n",
114 atomic_read(&fscache_n_checkaux_none),
115 atomic_read(&fscache_n_checkaux_okay),
116 atomic_read(&fscache_n_checkaux_update),
117 atomic_read(&fscache_n_checkaux_obsolete));
118
119 seq_printf(m, "Pages : mrk=%u unc=%u\n",
120 atomic_read(&fscache_n_marks),
121 atomic_read(&fscache_n_uncaches));
122
123 seq_printf(m, "Acquire: n=%u nul=%u noc=%u ok=%u nbf=%u"
124 " oom=%u\n",
125 atomic_read(&fscache_n_acquires),
126 atomic_read(&fscache_n_acquires_null),
127 atomic_read(&fscache_n_acquires_no_cache),
128 atomic_read(&fscache_n_acquires_ok),
129 atomic_read(&fscache_n_acquires_nobufs),
130 atomic_read(&fscache_n_acquires_oom));
131
132 seq_printf(m, "Lookups: n=%u neg=%u pos=%u crt=%u\n",
133 atomic_read(&fscache_n_object_lookups),
134 atomic_read(&fscache_n_object_lookups_negative),
135 atomic_read(&fscache_n_object_lookups_positive),
136 atomic_read(&fscache_n_object_created));
137
138 seq_printf(m, "Updates: n=%u nul=%u run=%u\n",
139 atomic_read(&fscache_n_updates),
140 atomic_read(&fscache_n_updates_null),
141 atomic_read(&fscache_n_updates_run));
142
143 seq_printf(m, "Relinqs: n=%u nul=%u wcr=%u\n",
144 atomic_read(&fscache_n_relinquishes),
145 atomic_read(&fscache_n_relinquishes_null),
146 atomic_read(&fscache_n_relinquishes_waitcrt));
147
148 seq_printf(m, "AttrChg: n=%u ok=%u nbf=%u oom=%u run=%u\n",
149 atomic_read(&fscache_n_attr_changed),
150 atomic_read(&fscache_n_attr_changed_ok),
151 atomic_read(&fscache_n_attr_changed_nobufs),
152 atomic_read(&fscache_n_attr_changed_nomem),
153 atomic_read(&fscache_n_attr_changed_calls));
154
155 seq_printf(m, "Allocs : n=%u ok=%u wt=%u nbf=%u\n",
156 atomic_read(&fscache_n_allocs),
157 atomic_read(&fscache_n_allocs_ok),
158 atomic_read(&fscache_n_allocs_wait),
159 atomic_read(&fscache_n_allocs_nobufs));
160 seq_printf(m, "Allocs : ops=%u owt=%u\n",
161 atomic_read(&fscache_n_alloc_ops),
162 atomic_read(&fscache_n_alloc_op_waits));
163
164 seq_printf(m, "Retrvls: n=%u ok=%u wt=%u nod=%u nbf=%u"
165 " int=%u oom=%u\n",
166 atomic_read(&fscache_n_retrievals),
167 atomic_read(&fscache_n_retrievals_ok),
168 atomic_read(&fscache_n_retrievals_wait),
169 atomic_read(&fscache_n_retrievals_nodata),
170 atomic_read(&fscache_n_retrievals_nobufs),
171 atomic_read(&fscache_n_retrievals_intr),
172 atomic_read(&fscache_n_retrievals_nomem));
173 seq_printf(m, "Retrvls: ops=%u owt=%u\n",
174 atomic_read(&fscache_n_retrieval_ops),
175 atomic_read(&fscache_n_retrieval_op_waits));
176
177 seq_printf(m, "Stores : n=%u ok=%u agn=%u nbf=%u oom=%u\n",
178 atomic_read(&fscache_n_stores),
179 atomic_read(&fscache_n_stores_ok),
180 atomic_read(&fscache_n_stores_again),
181 atomic_read(&fscache_n_stores_nobufs),
182 atomic_read(&fscache_n_stores_oom));
183 seq_printf(m, "Stores : ops=%u run=%u\n",
184 atomic_read(&fscache_n_store_ops),
185 atomic_read(&fscache_n_store_calls));
186
187 seq_printf(m, "Ops : pend=%u run=%u enq=%u\n",
188 atomic_read(&fscache_n_op_pend),
189 atomic_read(&fscache_n_op_run),
190 atomic_read(&fscache_n_op_enqueue));
191 seq_printf(m, "Ops : dfr=%u rel=%u gc=%u\n",
192 atomic_read(&fscache_n_op_deferred_release),
193 atomic_read(&fscache_n_op_release),
194 atomic_read(&fscache_n_op_gc));
195 return 0;
196}
197
198/*
199 * open "/proc/fs/fscache/stats" allowing provision of a statistical summary
200 */
201static int fscache_stats_open(struct inode *inode, struct file *file)
202{
203 return single_open(file, fscache_stats_show, NULL);
204}
205
206const struct file_operations fscache_stats_fops = {
207 .owner = THIS_MODULE,
208 .open = fscache_stats_open,
209 .read = seq_read,
210 .llseek = seq_lseek,
211 .release = seq_release,
212};
diff --git a/fs/nfs/Kconfig b/fs/nfs/Kconfig
index 36fe20d6eba2..e67f3ec07736 100644
--- a/fs/nfs/Kconfig
+++ b/fs/nfs/Kconfig
@@ -84,3 +84,11 @@ config ROOT_NFS
84 <file:Documentation/filesystems/nfsroot.txt>. 84 <file:Documentation/filesystems/nfsroot.txt>.
85 85
86 Most people say N here. 86 Most people say N here.
87
88config NFS_FSCACHE
89 bool "Provide NFS client caching support (EXPERIMENTAL)"
90 depends on EXPERIMENTAL
91 depends on NFS_FS=m && FSCACHE || NFS_FS=y && FSCACHE=y
92 help
93 Say Y here if you want NFS data to be cached locally on disc through
94 the general filesystem cache manager
diff --git a/fs/nfs/Makefile b/fs/nfs/Makefile
index ac6170c594a3..845159814de2 100644
--- a/fs/nfs/Makefile
+++ b/fs/nfs/Makefile
@@ -15,3 +15,4 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \
15 callback.o callback_xdr.o callback_proc.o \ 15 callback.o callback_xdr.o callback_proc.o \
16 nfs4namespace.o 16 nfs4namespace.o
17nfs-$(CONFIG_SYSCTL) += sysctl.o 17nfs-$(CONFIG_SYSCTL) += sysctl.o
18nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
diff --git a/fs/nfs/client.c b/fs/nfs/client.c
index aba38017bdef..75c9cd2aa119 100644
--- a/fs/nfs/client.c
+++ b/fs/nfs/client.c
@@ -45,6 +45,7 @@
45#include "delegation.h" 45#include "delegation.h"
46#include "iostat.h" 46#include "iostat.h"
47#include "internal.h" 47#include "internal.h"
48#include "fscache.h"
48 49
49#define NFSDBG_FACILITY NFSDBG_CLIENT 50#define NFSDBG_FACILITY NFSDBG_CLIENT
50 51
@@ -154,6 +155,8 @@ static struct nfs_client *nfs_alloc_client(const struct nfs_client_initdata *cl_
154 if (!IS_ERR(cred)) 155 if (!IS_ERR(cred))
155 clp->cl_machine_cred = cred; 156 clp->cl_machine_cred = cred;
156 157
158 nfs_fscache_get_client_cookie(clp);
159
157 return clp; 160 return clp;
158 161
159error_3: 162error_3:
@@ -187,6 +190,8 @@ static void nfs_free_client(struct nfs_client *clp)
187 190
188 nfs4_shutdown_client(clp); 191 nfs4_shutdown_client(clp);
189 192
193 nfs_fscache_release_client_cookie(clp);
194
190 /* -EIO all pending I/O */ 195 /* -EIO all pending I/O */
191 if (!IS_ERR(clp->cl_rpcclient)) 196 if (!IS_ERR(clp->cl_rpcclient))
192 rpc_shutdown_client(clp->cl_rpcclient); 197 rpc_shutdown_client(clp->cl_rpcclient);
@@ -760,6 +765,7 @@ static int nfs_init_server(struct nfs_server *server,
760 765
761 /* Initialise the client representation from the mount data */ 766 /* Initialise the client representation from the mount data */
762 server->flags = data->flags; 767 server->flags = data->flags;
768 server->options = data->options;
763 769
764 if (data->rsize) 770 if (data->rsize)
765 server->rsize = nfs_block_size(data->rsize, NULL); 771 server->rsize = nfs_block_size(data->rsize, NULL);
@@ -1148,6 +1154,7 @@ static int nfs4_init_server(struct nfs_server *server,
1148 /* Initialise the client representation from the mount data */ 1154 /* Initialise the client representation from the mount data */
1149 server->flags = data->flags; 1155 server->flags = data->flags;
1150 server->caps |= NFS_CAP_ATOMIC_OPEN; 1156 server->caps |= NFS_CAP_ATOMIC_OPEN;
1157 server->options = data->options;
1151 1158
1152 /* Get a client record */ 1159 /* Get a client record */
1153 error = nfs4_set_client(server, 1160 error = nfs4_set_client(server,
@@ -1559,7 +1566,7 @@ static int nfs_volume_list_show(struct seq_file *m, void *v)
1559 1566
1560 /* display header on line 1 */ 1567 /* display header on line 1 */
1561 if (v == &nfs_volume_list) { 1568 if (v == &nfs_volume_list) {
1562 seq_puts(m, "NV SERVER PORT DEV FSID\n"); 1569 seq_puts(m, "NV SERVER PORT DEV FSID FSC\n");
1563 return 0; 1570 return 0;
1564 } 1571 }
1565 /* display one transport per line on subsequent lines */ 1572 /* display one transport per line on subsequent lines */
@@ -1573,12 +1580,13 @@ static int nfs_volume_list_show(struct seq_file *m, void *v)
1573 (unsigned long long) server->fsid.major, 1580 (unsigned long long) server->fsid.major,
1574 (unsigned long long) server->fsid.minor); 1581 (unsigned long long) server->fsid.minor);
1575 1582
1576 seq_printf(m, "v%u %s %s %-7s %-17s\n", 1583 seq_printf(m, "v%u %s %s %-7s %-17s %s\n",
1577 clp->rpc_ops->version, 1584 clp->rpc_ops->version,
1578 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_ADDR), 1585 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_ADDR),
1579 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_PORT), 1586 rpc_peeraddr2str(clp->cl_rpcclient, RPC_DISPLAY_HEX_PORT),
1580 dev, 1587 dev,
1581 fsid); 1588 fsid,
1589 nfs_server_fscache_state(server));
1582 1590
1583 return 0; 1591 return 0;
1584} 1592}
diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index 0abf3f331f56..3523b895eb4b 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -35,6 +35,7 @@
35#include "delegation.h" 35#include "delegation.h"
36#include "internal.h" 36#include "internal.h"
37#include "iostat.h" 37#include "iostat.h"
38#include "fscache.h"
38 39
39#define NFSDBG_FACILITY NFSDBG_FILE 40#define NFSDBG_FACILITY NFSDBG_FILE
40 41
@@ -409,6 +410,13 @@ static int nfs_write_end(struct file *file, struct address_space *mapping,
409 return copied; 410 return copied;
410} 411}
411 412
413/*
414 * Partially or wholly invalidate a page
415 * - Release the private state associated with a page if undergoing complete
416 * page invalidation
417 * - Called if either PG_private or PG_fscache is set on the page
418 * - Caller holds page lock
419 */
412static void nfs_invalidate_page(struct page *page, unsigned long offset) 420static void nfs_invalidate_page(struct page *page, unsigned long offset)
413{ 421{
414 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset); 422 dfprintk(PAGECACHE, "NFS: invalidate_page(%p, %lu)\n", page, offset);
@@ -417,23 +425,43 @@ static void nfs_invalidate_page(struct page *page, unsigned long offset)
417 return; 425 return;
418 /* Cancel any unstarted writes on this page */ 426 /* Cancel any unstarted writes on this page */
419 nfs_wb_page_cancel(page->mapping->host, page); 427 nfs_wb_page_cancel(page->mapping->host, page);
428
429 nfs_fscache_invalidate_page(page, page->mapping->host);
420} 430}
421 431
432/*
433 * Attempt to release the private state associated with a page
434 * - Called if either PG_private or PG_fscache is set on the page
435 * - Caller holds page lock
436 * - Return true (may release page) or false (may not)
437 */
422static int nfs_release_page(struct page *page, gfp_t gfp) 438static int nfs_release_page(struct page *page, gfp_t gfp)
423{ 439{
424 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page); 440 dfprintk(PAGECACHE, "NFS: release_page(%p)\n", page);
425 441
426 /* If PagePrivate() is set, then the page is not freeable */ 442 /* If PagePrivate() is set, then the page is not freeable */
427 return 0; 443 if (PagePrivate(page))
444 return 0;
445 return nfs_fscache_release_page(page, gfp);
428} 446}
429 447
448/*
449 * Attempt to clear the private state associated with a page when an error
450 * occurs that requires the cached contents of an inode to be written back or
451 * destroyed
452 * - Called if either PG_private or fscache is set on the page
453 * - Caller holds page lock
454 * - Return 0 if successful, -error otherwise
455 */
430static int nfs_launder_page(struct page *page) 456static int nfs_launder_page(struct page *page)
431{ 457{
432 struct inode *inode = page->mapping->host; 458 struct inode *inode = page->mapping->host;
459 struct nfs_inode *nfsi = NFS_I(inode);
433 460
434 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n", 461 dfprintk(PAGECACHE, "NFS: launder_page(%ld, %llu)\n",
435 inode->i_ino, (long long)page_offset(page)); 462 inode->i_ino, (long long)page_offset(page));
436 463
464 nfs_fscache_wait_on_page_write(nfsi, page);
437 return nfs_wb_page(inode, page); 465 return nfs_wb_page(inode, page);
438} 466}
439 467
@@ -451,6 +479,11 @@ const struct address_space_operations nfs_file_aops = {
451 .launder_page = nfs_launder_page, 479 .launder_page = nfs_launder_page,
452}; 480};
453 481
482/*
483 * Notification that a PTE pointing to an NFS page is about to be made
484 * writable, implying that someone is about to modify the page through a
485 * shared-writable mapping
486 */
454static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf) 487static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
455{ 488{
456 struct page *page = vmf->page; 489 struct page *page = vmf->page;
@@ -465,6 +498,9 @@ static int nfs_vm_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
465 filp->f_mapping->host->i_ino, 498 filp->f_mapping->host->i_ino,
466 (long long)page_offset(page)); 499 (long long)page_offset(page));
467 500
501 /* make sure the cache has finished storing the page */
502 nfs_fscache_wait_on_page_write(NFS_I(dentry->d_inode), page);
503
468 lock_page(page); 504 lock_page(page);
469 mapping = page->mapping; 505 mapping = page->mapping;
470 if (mapping != dentry->d_inode->i_mapping) 506 if (mapping != dentry->d_inode->i_mapping)
diff --git a/fs/nfs/fscache-index.c b/fs/nfs/fscache-index.c
new file mode 100644
index 000000000000..5b1006480bc2
--- /dev/null
+++ b/fs/nfs/fscache-index.c
@@ -0,0 +1,337 @@
1/* NFS FS-Cache index structure definition
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/sched.h>
15#include <linux/mm.h>
16#include <linux/nfs_fs.h>
17#include <linux/nfs_fs_sb.h>
18#include <linux/in6.h>
19
20#include "internal.h"
21#include "fscache.h"
22
23#define NFSDBG_FACILITY NFSDBG_FSCACHE
24
25/*
26 * Define the NFS filesystem for FS-Cache. Upon registration FS-Cache sticks
27 * the cookie for the top-level index object for NFS into here. The top-level
28 * index can than have other cache objects inserted into it.
29 */
30struct fscache_netfs nfs_fscache_netfs = {
31 .name = "nfs",
32 .version = 0,
33};
34
35/*
36 * Register NFS for caching
37 */
38int nfs_fscache_register(void)
39{
40 return fscache_register_netfs(&nfs_fscache_netfs);
41}
42
43/*
44 * Unregister NFS for caching
45 */
46void nfs_fscache_unregister(void)
47{
48 fscache_unregister_netfs(&nfs_fscache_netfs);
49}
50
51/*
52 * Layout of the key for an NFS server cache object.
53 */
54struct nfs_server_key {
55 uint16_t nfsversion; /* NFS protocol version */
56 uint16_t family; /* address family */
57 uint16_t port; /* IP port */
58 union {
59 struct in_addr ipv4_addr; /* IPv4 address */
60 struct in6_addr ipv6_addr; /* IPv6 address */
61 } addr[0];
62};
63
64/*
65 * Generate a key to describe a server in the main NFS index
66 * - We return the length of the key, or 0 if we can't generate one
67 */
68static uint16_t nfs_server_get_key(const void *cookie_netfs_data,
69 void *buffer, uint16_t bufmax)
70{
71 const struct nfs_client *clp = cookie_netfs_data;
72 const struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *) &clp->cl_addr;
73 const struct sockaddr_in *sin = (struct sockaddr_in *) &clp->cl_addr;
74 struct nfs_server_key *key = buffer;
75 uint16_t len = sizeof(struct nfs_server_key);
76
77 key->nfsversion = clp->rpc_ops->version;
78 key->family = clp->cl_addr.ss_family;
79
80 memset(key, 0, len);
81
82 switch (clp->cl_addr.ss_family) {
83 case AF_INET:
84 key->port = sin->sin_port;
85 key->addr[0].ipv4_addr = sin->sin_addr;
86 len += sizeof(key->addr[0].ipv4_addr);
87 break;
88
89 case AF_INET6:
90 key->port = sin6->sin6_port;
91 key->addr[0].ipv6_addr = sin6->sin6_addr;
92 len += sizeof(key->addr[0].ipv6_addr);
93 break;
94
95 default:
96 printk(KERN_WARNING "NFS: Unknown network family '%d'\n",
97 clp->cl_addr.ss_family);
98 len = 0;
99 break;
100 }
101
102 return len;
103}
104
105/*
106 * Define the server object for FS-Cache. This is used to describe a server
107 * object to fscache_acquire_cookie(). It is keyed by the NFS protocol and
108 * server address parameters.
109 */
110const struct fscache_cookie_def nfs_fscache_server_index_def = {
111 .name = "NFS.server",
112 .type = FSCACHE_COOKIE_TYPE_INDEX,
113 .get_key = nfs_server_get_key,
114};
115
116/*
117 * Generate a key to describe a superblock key in the main NFS index
118 */
119static uint16_t nfs_super_get_key(const void *cookie_netfs_data,
120 void *buffer, uint16_t bufmax)
121{
122 const struct nfs_fscache_key *key;
123 const struct nfs_server *nfss = cookie_netfs_data;
124 uint16_t len;
125
126 key = nfss->fscache_key;
127 len = sizeof(key->key) + key->key.uniq_len;
128 if (len > bufmax) {
129 len = 0;
130 } else {
131 memcpy(buffer, &key->key, sizeof(key->key));
132 memcpy(buffer + sizeof(key->key),
133 key->key.uniquifier, key->key.uniq_len);
134 }
135
136 return len;
137}
138
139/*
140 * Define the superblock object for FS-Cache. This is used to describe a
141 * superblock object to fscache_acquire_cookie(). It is keyed by all the NFS
142 * parameters that might cause a separate superblock.
143 */
144const struct fscache_cookie_def nfs_fscache_super_index_def = {
145 .name = "NFS.super",
146 .type = FSCACHE_COOKIE_TYPE_INDEX,
147 .get_key = nfs_super_get_key,
148};
149
150/*
151 * Definition of the auxiliary data attached to NFS inode storage objects
152 * within the cache.
153 *
154 * The contents of this struct are recorded in the on-disk local cache in the
155 * auxiliary data attached to the data storage object backing an inode. This
156 * permits coherency to be managed when a new inode binds to an already extant
157 * cache object.
158 */
159struct nfs_fscache_inode_auxdata {
160 struct timespec mtime;
161 struct timespec ctime;
162 loff_t size;
163 u64 change_attr;
164};
165
166/*
167 * Generate a key to describe an NFS inode in an NFS server's index
168 */
169static uint16_t nfs_fscache_inode_get_key(const void *cookie_netfs_data,
170 void *buffer, uint16_t bufmax)
171{
172 const struct nfs_inode *nfsi = cookie_netfs_data;
173 uint16_t nsize;
174
175 /* use the inode's NFS filehandle as the key */
176 nsize = nfsi->fh.size;
177 memcpy(buffer, nfsi->fh.data, nsize);
178 return nsize;
179}
180
181/*
182 * Get certain file attributes from the netfs data
183 * - This function can be absent for an index
184 * - Not permitted to return an error
185 * - The netfs data from the cookie being used as the source is presented
186 */
187static void nfs_fscache_inode_get_attr(const void *cookie_netfs_data,
188 uint64_t *size)
189{
190 const struct nfs_inode *nfsi = cookie_netfs_data;
191
192 *size = nfsi->vfs_inode.i_size;
193}
194
195/*
196 * Get the auxiliary data from netfs data
197 * - This function can be absent if the index carries no state data
198 * - Should store the auxiliary data in the buffer
199 * - Should return the amount of amount stored
200 * - Not permitted to return an error
201 * - The netfs data from the cookie being used as the source is presented
202 */
203static uint16_t nfs_fscache_inode_get_aux(const void *cookie_netfs_data,
204 void *buffer, uint16_t bufmax)
205{
206 struct nfs_fscache_inode_auxdata auxdata;
207 const struct nfs_inode *nfsi = cookie_netfs_data;
208
209 memset(&auxdata, 0, sizeof(auxdata));
210 auxdata.size = nfsi->vfs_inode.i_size;
211 auxdata.mtime = nfsi->vfs_inode.i_mtime;
212 auxdata.ctime = nfsi->vfs_inode.i_ctime;
213
214 if (NFS_SERVER(&nfsi->vfs_inode)->nfs_client->rpc_ops->version == 4)
215 auxdata.change_attr = nfsi->change_attr;
216
217 if (bufmax > sizeof(auxdata))
218 bufmax = sizeof(auxdata);
219
220 memcpy(buffer, &auxdata, bufmax);
221 return bufmax;
222}
223
224/*
225 * Consult the netfs about the state of an object
226 * - This function can be absent if the index carries no state data
227 * - The netfs data from the cookie being used as the target is
228 * presented, as is the auxiliary data
229 */
230static
231enum fscache_checkaux nfs_fscache_inode_check_aux(void *cookie_netfs_data,
232 const void *data,
233 uint16_t datalen)
234{
235 struct nfs_fscache_inode_auxdata auxdata;
236 struct nfs_inode *nfsi = cookie_netfs_data;
237
238 if (datalen != sizeof(auxdata))
239 return FSCACHE_CHECKAUX_OBSOLETE;
240
241 memset(&auxdata, 0, sizeof(auxdata));
242 auxdata.size = nfsi->vfs_inode.i_size;
243 auxdata.mtime = nfsi->vfs_inode.i_mtime;
244 auxdata.ctime = nfsi->vfs_inode.i_ctime;
245
246 if (NFS_SERVER(&nfsi->vfs_inode)->nfs_client->rpc_ops->version == 4)
247 auxdata.change_attr = nfsi->change_attr;
248
249 if (memcmp(data, &auxdata, datalen) != 0)
250 return FSCACHE_CHECKAUX_OBSOLETE;
251
252 return FSCACHE_CHECKAUX_OKAY;
253}
254
255/*
256 * Indication from FS-Cache that the cookie is no longer cached
257 * - This function is called when the backing store currently caching a cookie
258 * is removed
259 * - The netfs should use this to clean up any markers indicating cached pages
260 * - This is mandatory for any object that may have data
261 */
262static void nfs_fscache_inode_now_uncached(void *cookie_netfs_data)
263{
264 struct nfs_inode *nfsi = cookie_netfs_data;
265 struct pagevec pvec;
266 pgoff_t first;
267 int loop, nr_pages;
268
269 pagevec_init(&pvec, 0);
270 first = 0;
271
272 dprintk("NFS: nfs_inode_now_uncached: nfs_inode 0x%p\n", nfsi);
273
274 for (;;) {
275 /* grab a bunch of pages to unmark */
276 nr_pages = pagevec_lookup(&pvec,
277 nfsi->vfs_inode.i_mapping,
278 first,
279 PAGEVEC_SIZE - pagevec_count(&pvec));
280 if (!nr_pages)
281 break;
282
283 for (loop = 0; loop < nr_pages; loop++)
284 ClearPageFsCache(pvec.pages[loop]);
285
286 first = pvec.pages[nr_pages - 1]->index + 1;
287
288 pvec.nr = nr_pages;
289 pagevec_release(&pvec);
290 cond_resched();
291 }
292}
293
294/*
295 * Get an extra reference on a read context.
296 * - This function can be absent if the completion function doesn't require a
297 * context.
298 * - The read context is passed back to NFS in the event that a data read on the
299 * cache fails with EIO - in which case the server must be contacted to
300 * retrieve the data, which requires the read context for security.
301 */
302static void nfs_fh_get_context(void *cookie_netfs_data, void *context)
303{
304 get_nfs_open_context(context);
305}
306
307/*
308 * Release an extra reference on a read context.
309 * - This function can be absent if the completion function doesn't require a
310 * context.
311 */
312static void nfs_fh_put_context(void *cookie_netfs_data, void *context)
313{
314 if (context)
315 put_nfs_open_context(context);
316}
317
318/*
319 * Define the inode object for FS-Cache. This is used to describe an inode
320 * object to fscache_acquire_cookie(). It is keyed by the NFS file handle for
321 * an inode.
322 *
323 * Coherency is managed by comparing the copies of i_size, i_mtime and i_ctime
324 * held in the cache auxiliary data for the data storage object with those in
325 * the inode struct in memory.
326 */
327const struct fscache_cookie_def nfs_fscache_inode_object_def = {
328 .name = "NFS.fh",
329 .type = FSCACHE_COOKIE_TYPE_DATAFILE,
330 .get_key = nfs_fscache_inode_get_key,
331 .get_attr = nfs_fscache_inode_get_attr,
332 .get_aux = nfs_fscache_inode_get_aux,
333 .check_aux = nfs_fscache_inode_check_aux,
334 .now_uncached = nfs_fscache_inode_now_uncached,
335 .get_context = nfs_fh_get_context,
336 .put_context = nfs_fh_put_context,
337};
diff --git a/fs/nfs/fscache.c b/fs/nfs/fscache.c
new file mode 100644
index 000000000000..379be678cb7e
--- /dev/null
+++ b/fs/nfs/fscache.c
@@ -0,0 +1,523 @@
1/* NFS filesystem cache interface
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/sched.h>
15#include <linux/mm.h>
16#include <linux/nfs_fs.h>
17#include <linux/nfs_fs_sb.h>
18#include <linux/in6.h>
19#include <linux/seq_file.h>
20
21#include "internal.h"
22#include "iostat.h"
23#include "fscache.h"
24
25#define NFSDBG_FACILITY NFSDBG_FSCACHE
26
27static struct rb_root nfs_fscache_keys = RB_ROOT;
28static DEFINE_SPINLOCK(nfs_fscache_keys_lock);
29
30/*
31 * Get the per-client index cookie for an NFS client if the appropriate mount
32 * flag was set
33 * - We always try and get an index cookie for the client, but get filehandle
34 * cookies on a per-superblock basis, depending on the mount flags
35 */
36void nfs_fscache_get_client_cookie(struct nfs_client *clp)
37{
38 /* create a cache index for looking up filehandles */
39 clp->fscache = fscache_acquire_cookie(nfs_fscache_netfs.primary_index,
40 &nfs_fscache_server_index_def,
41 clp);
42 dfprintk(FSCACHE, "NFS: get client cookie (0x%p/0x%p)\n",
43 clp, clp->fscache);
44}
45
46/*
47 * Dispose of a per-client cookie
48 */
49void nfs_fscache_release_client_cookie(struct nfs_client *clp)
50{
51 dfprintk(FSCACHE, "NFS: releasing client cookie (0x%p/0x%p)\n",
52 clp, clp->fscache);
53
54 fscache_relinquish_cookie(clp->fscache, 0);
55 clp->fscache = NULL;
56}
57
58/*
59 * Get the cache cookie for an NFS superblock. We have to handle
60 * uniquification here because the cache doesn't do it for us.
61 */
62void nfs_fscache_get_super_cookie(struct super_block *sb,
63 struct nfs_parsed_mount_data *data)
64{
65 struct nfs_fscache_key *key, *xkey;
66 struct nfs_server *nfss = NFS_SB(sb);
67 struct rb_node **p, *parent;
68 const char *uniq = data->fscache_uniq ?: "";
69 int diff, ulen;
70
71 ulen = strlen(uniq);
72 key = kzalloc(sizeof(*key) + ulen, GFP_KERNEL);
73 if (!key)
74 return;
75
76 key->nfs_client = nfss->nfs_client;
77 key->key.super.s_flags = sb->s_flags & NFS_MS_MASK;
78 key->key.nfs_server.flags = nfss->flags;
79 key->key.nfs_server.rsize = nfss->rsize;
80 key->key.nfs_server.wsize = nfss->wsize;
81 key->key.nfs_server.acregmin = nfss->acregmin;
82 key->key.nfs_server.acregmax = nfss->acregmax;
83 key->key.nfs_server.acdirmin = nfss->acdirmin;
84 key->key.nfs_server.acdirmax = nfss->acdirmax;
85 key->key.nfs_server.fsid = nfss->fsid;
86 key->key.rpc_auth.au_flavor = nfss->client->cl_auth->au_flavor;
87
88 key->key.uniq_len = ulen;
89 memcpy(key->key.uniquifier, uniq, ulen);
90
91 spin_lock(&nfs_fscache_keys_lock);
92 p = &nfs_fscache_keys.rb_node;
93 parent = NULL;
94 while (*p) {
95 parent = *p;
96 xkey = rb_entry(parent, struct nfs_fscache_key, node);
97
98 if (key->nfs_client < xkey->nfs_client)
99 goto go_left;
100 if (key->nfs_client > xkey->nfs_client)
101 goto go_right;
102
103 diff = memcmp(&key->key, &xkey->key, sizeof(key->key));
104 if (diff < 0)
105 goto go_left;
106 if (diff > 0)
107 goto go_right;
108
109 if (key->key.uniq_len == 0)
110 goto non_unique;
111 diff = memcmp(key->key.uniquifier,
112 xkey->key.uniquifier,
113 key->key.uniq_len);
114 if (diff < 0)
115 goto go_left;
116 if (diff > 0)
117 goto go_right;
118 goto non_unique;
119
120 go_left:
121 p = &(*p)->rb_left;
122 continue;
123 go_right:
124 p = &(*p)->rb_right;
125 }
126
127 rb_link_node(&key->node, parent, p);
128 rb_insert_color(&key->node, &nfs_fscache_keys);
129 spin_unlock(&nfs_fscache_keys_lock);
130 nfss->fscache_key = key;
131
132 /* create a cache index for looking up filehandles */
133 nfss->fscache = fscache_acquire_cookie(nfss->nfs_client->fscache,
134 &nfs_fscache_super_index_def,
135 nfss);
136 dfprintk(FSCACHE, "NFS: get superblock cookie (0x%p/0x%p)\n",
137 nfss, nfss->fscache);
138 return;
139
140non_unique:
141 spin_unlock(&nfs_fscache_keys_lock);
142 kfree(key);
143 nfss->fscache_key = NULL;
144 nfss->fscache = NULL;
145 printk(KERN_WARNING "NFS:"
146 " Cache request denied due to non-unique superblock keys\n");
147}
148
149/*
150 * release a per-superblock cookie
151 */
152void nfs_fscache_release_super_cookie(struct super_block *sb)
153{
154 struct nfs_server *nfss = NFS_SB(sb);
155
156 dfprintk(FSCACHE, "NFS: releasing superblock cookie (0x%p/0x%p)\n",
157 nfss, nfss->fscache);
158
159 fscache_relinquish_cookie(nfss->fscache, 0);
160 nfss->fscache = NULL;
161
162 if (nfss->fscache_key) {
163 spin_lock(&nfs_fscache_keys_lock);
164 rb_erase(&nfss->fscache_key->node, &nfs_fscache_keys);
165 spin_unlock(&nfs_fscache_keys_lock);
166 kfree(nfss->fscache_key);
167 nfss->fscache_key = NULL;
168 }
169}
170
171/*
172 * Initialise the per-inode cache cookie pointer for an NFS inode.
173 */
174void nfs_fscache_init_inode_cookie(struct inode *inode)
175{
176 NFS_I(inode)->fscache = NULL;
177 if (S_ISREG(inode->i_mode))
178 set_bit(NFS_INO_FSCACHE, &NFS_I(inode)->flags);
179}
180
181/*
182 * Get the per-inode cache cookie for an NFS inode.
183 */
184static void nfs_fscache_enable_inode_cookie(struct inode *inode)
185{
186 struct super_block *sb = inode->i_sb;
187 struct nfs_inode *nfsi = NFS_I(inode);
188
189 if (nfsi->fscache || !NFS_FSCACHE(inode))
190 return;
191
192 if ((NFS_SB(sb)->options & NFS_OPTION_FSCACHE)) {
193 nfsi->fscache = fscache_acquire_cookie(
194 NFS_SB(sb)->fscache,
195 &nfs_fscache_inode_object_def,
196 nfsi);
197
198 dfprintk(FSCACHE, "NFS: get FH cookie (0x%p/0x%p/0x%p)\n",
199 sb, nfsi, nfsi->fscache);
200 }
201}
202
203/*
204 * Release a per-inode cookie.
205 */
206void nfs_fscache_release_inode_cookie(struct inode *inode)
207{
208 struct nfs_inode *nfsi = NFS_I(inode);
209
210 dfprintk(FSCACHE, "NFS: clear cookie (0x%p/0x%p)\n",
211 nfsi, nfsi->fscache);
212
213 fscache_relinquish_cookie(nfsi->fscache, 0);
214 nfsi->fscache = NULL;
215}
216
217/*
218 * Retire a per-inode cookie, destroying the data attached to it.
219 */
220void nfs_fscache_zap_inode_cookie(struct inode *inode)
221{
222 struct nfs_inode *nfsi = NFS_I(inode);
223
224 dfprintk(FSCACHE, "NFS: zapping cookie (0x%p/0x%p)\n",
225 nfsi, nfsi->fscache);
226
227 fscache_relinquish_cookie(nfsi->fscache, 1);
228 nfsi->fscache = NULL;
229}
230
231/*
232 * Turn off the cache with regard to a per-inode cookie if opened for writing,
233 * invalidating all the pages in the page cache relating to the associated
234 * inode to clear the per-page caching.
235 */
236static void nfs_fscache_disable_inode_cookie(struct inode *inode)
237{
238 clear_bit(NFS_INO_FSCACHE, &NFS_I(inode)->flags);
239
240 if (NFS_I(inode)->fscache) {
241 dfprintk(FSCACHE,
242 "NFS: nfsi 0x%p turning cache off\n", NFS_I(inode));
243
244 /* Need to invalidate any mapped pages that were read in before
245 * turning off the cache.
246 */
247 if (inode->i_mapping && inode->i_mapping->nrpages)
248 invalidate_inode_pages2(inode->i_mapping);
249
250 nfs_fscache_zap_inode_cookie(inode);
251 }
252}
253
254/*
255 * wait_on_bit() sleep function for uninterruptible waiting
256 */
257static int nfs_fscache_wait_bit(void *flags)
258{
259 schedule();
260 return 0;
261}
262
263/*
264 * Lock against someone else trying to also acquire or relinquish a cookie
265 */
266static inline void nfs_fscache_inode_lock(struct inode *inode)
267{
268 struct nfs_inode *nfsi = NFS_I(inode);
269
270 while (test_and_set_bit(NFS_INO_FSCACHE_LOCK, &nfsi->flags))
271 wait_on_bit(&nfsi->flags, NFS_INO_FSCACHE_LOCK,
272 nfs_fscache_wait_bit, TASK_UNINTERRUPTIBLE);
273}
274
275/*
276 * Unlock cookie management lock
277 */
278static inline void nfs_fscache_inode_unlock(struct inode *inode)
279{
280 struct nfs_inode *nfsi = NFS_I(inode);
281
282 smp_mb__before_clear_bit();
283 clear_bit(NFS_INO_FSCACHE_LOCK, &nfsi->flags);
284 smp_mb__after_clear_bit();
285 wake_up_bit(&nfsi->flags, NFS_INO_FSCACHE_LOCK);
286}
287
288/*
289 * Decide if we should enable or disable local caching for this inode.
290 * - For now, with NFS, only regular files that are open read-only will be able
291 * to use the cache.
292 * - May be invoked multiple times in parallel by parallel nfs_open() functions.
293 */
294void nfs_fscache_set_inode_cookie(struct inode *inode, struct file *filp)
295{
296 if (NFS_FSCACHE(inode)) {
297 nfs_fscache_inode_lock(inode);
298 if ((filp->f_flags & O_ACCMODE) != O_RDONLY)
299 nfs_fscache_disable_inode_cookie(inode);
300 else
301 nfs_fscache_enable_inode_cookie(inode);
302 nfs_fscache_inode_unlock(inode);
303 }
304}
305
306/*
307 * Replace a per-inode cookie due to revalidation detecting a file having
308 * changed on the server.
309 */
310void nfs_fscache_reset_inode_cookie(struct inode *inode)
311{
312 struct nfs_inode *nfsi = NFS_I(inode);
313 struct nfs_server *nfss = NFS_SERVER(inode);
314 struct fscache_cookie *old = nfsi->fscache;
315
316 nfs_fscache_inode_lock(inode);
317 if (nfsi->fscache) {
318 /* retire the current fscache cache and get a new one */
319 fscache_relinquish_cookie(nfsi->fscache, 1);
320
321 nfsi->fscache = fscache_acquire_cookie(
322 nfss->nfs_client->fscache,
323 &nfs_fscache_inode_object_def,
324 nfsi);
325
326 dfprintk(FSCACHE,
327 "NFS: revalidation new cookie (0x%p/0x%p/0x%p/0x%p)\n",
328 nfss, nfsi, old, nfsi->fscache);
329 }
330 nfs_fscache_inode_unlock(inode);
331}
332
333/*
334 * Release the caching state associated with a page, if the page isn't busy
335 * interacting with the cache.
336 * - Returns true (can release page) or false (page busy).
337 */
338int nfs_fscache_release_page(struct page *page, gfp_t gfp)
339{
340 struct nfs_inode *nfsi = NFS_I(page->mapping->host);
341 struct fscache_cookie *cookie = nfsi->fscache;
342
343 BUG_ON(!cookie);
344
345 if (fscache_check_page_write(cookie, page)) {
346 if (!(gfp & __GFP_WAIT))
347 return 0;
348 fscache_wait_on_page_write(cookie, page);
349 }
350
351 if (PageFsCache(page)) {
352 dfprintk(FSCACHE, "NFS: fscache releasepage (0x%p/0x%p/0x%p)\n",
353 cookie, page, nfsi);
354
355 fscache_uncache_page(cookie, page);
356 nfs_add_fscache_stats(page->mapping->host,
357 NFSIOS_FSCACHE_PAGES_UNCACHED, 1);
358 }
359
360 return 1;
361}
362
363/*
364 * Release the caching state associated with a page if undergoing complete page
365 * invalidation.
366 */
367void __nfs_fscache_invalidate_page(struct page *page, struct inode *inode)
368{
369 struct nfs_inode *nfsi = NFS_I(inode);
370 struct fscache_cookie *cookie = nfsi->fscache;
371
372 BUG_ON(!cookie);
373
374 dfprintk(FSCACHE, "NFS: fscache invalidatepage (0x%p/0x%p/0x%p)\n",
375 cookie, page, nfsi);
376
377 fscache_wait_on_page_write(cookie, page);
378
379 BUG_ON(!PageLocked(page));
380 fscache_uncache_page(cookie, page);
381 nfs_add_fscache_stats(page->mapping->host,
382 NFSIOS_FSCACHE_PAGES_UNCACHED, 1);
383}
384
385/*
386 * Handle completion of a page being read from the cache.
387 * - Called in process (keventd) context.
388 */
389static void nfs_readpage_from_fscache_complete(struct page *page,
390 void *context,
391 int error)
392{
393 dfprintk(FSCACHE,
394 "NFS: readpage_from_fscache_complete (0x%p/0x%p/%d)\n",
395 page, context, error);
396
397 /* if the read completes with an error, we just unlock the page and let
398 * the VM reissue the readpage */
399 if (!error) {
400 SetPageUptodate(page);
401 unlock_page(page);
402 } else {
403 error = nfs_readpage_async(context, page->mapping->host, page);
404 if (error)
405 unlock_page(page);
406 }
407}
408
409/*
410 * Retrieve a page from fscache
411 */
412int __nfs_readpage_from_fscache(struct nfs_open_context *ctx,
413 struct inode *inode, struct page *page)
414{
415 int ret;
416
417 dfprintk(FSCACHE,
418 "NFS: readpage_from_fscache(fsc:%p/p:%p(i:%lx f:%lx)/0x%p)\n",
419 NFS_I(inode)->fscache, page, page->index, page->flags, inode);
420
421 ret = fscache_read_or_alloc_page(NFS_I(inode)->fscache,
422 page,
423 nfs_readpage_from_fscache_complete,
424 ctx,
425 GFP_KERNEL);
426
427 switch (ret) {
428 case 0: /* read BIO submitted (page in fscache) */
429 dfprintk(FSCACHE,
430 "NFS: readpage_from_fscache: BIO submitted\n");
431 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK, 1);
432 return ret;
433
434 case -ENOBUFS: /* inode not in cache */
435 case -ENODATA: /* page not in cache */
436 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL, 1);
437 dfprintk(FSCACHE,
438 "NFS: readpage_from_fscache %d\n", ret);
439 return 1;
440
441 default:
442 dfprintk(FSCACHE, "NFS: readpage_from_fscache %d\n", ret);
443 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL, 1);
444 }
445 return ret;
446}
447
448/*
449 * Retrieve a set of pages from fscache
450 */
451int __nfs_readpages_from_fscache(struct nfs_open_context *ctx,
452 struct inode *inode,
453 struct address_space *mapping,
454 struct list_head *pages,
455 unsigned *nr_pages)
456{
457 int ret, npages = *nr_pages;
458
459 dfprintk(FSCACHE, "NFS: nfs_getpages_from_fscache (0x%p/%u/0x%p)\n",
460 NFS_I(inode)->fscache, npages, inode);
461
462 ret = fscache_read_or_alloc_pages(NFS_I(inode)->fscache,
463 mapping, pages, nr_pages,
464 nfs_readpage_from_fscache_complete,
465 ctx,
466 mapping_gfp_mask(mapping));
467 if (*nr_pages < npages)
468 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_OK,
469 npages);
470 if (*nr_pages > 0)
471 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_READ_FAIL,
472 *nr_pages);
473
474 switch (ret) {
475 case 0: /* read submitted to the cache for all pages */
476 BUG_ON(!list_empty(pages));
477 BUG_ON(*nr_pages != 0);
478 dfprintk(FSCACHE,
479 "NFS: nfs_getpages_from_fscache: submitted\n");
480
481 return ret;
482
483 case -ENOBUFS: /* some pages aren't cached and can't be */
484 case -ENODATA: /* some pages aren't cached */
485 dfprintk(FSCACHE,
486 "NFS: nfs_getpages_from_fscache: no page: %d\n", ret);
487 return 1;
488
489 default:
490 dfprintk(FSCACHE,
491 "NFS: nfs_getpages_from_fscache: ret %d\n", ret);
492 }
493
494 return ret;
495}
496
497/*
498 * Store a newly fetched page in fscache
499 * - PG_fscache must be set on the page
500 */
501void __nfs_readpage_to_fscache(struct inode *inode, struct page *page, int sync)
502{
503 int ret;
504
505 dfprintk(FSCACHE,
506 "NFS: readpage_to_fscache(fsc:%p/p:%p(i:%lx f:%lx)/%d)\n",
507 NFS_I(inode)->fscache, page, page->index, page->flags, sync);
508
509 ret = fscache_write_page(NFS_I(inode)->fscache, page, GFP_KERNEL);
510 dfprintk(FSCACHE,
511 "NFS: readpage_to_fscache: p:%p(i:%lu f:%lx) ret %d\n",
512 page, page->index, page->flags, ret);
513
514 if (ret != 0) {
515 fscache_uncache_page(NFS_I(inode)->fscache, page);
516 nfs_add_fscache_stats(inode,
517 NFSIOS_FSCACHE_PAGES_WRITTEN_FAIL, 1);
518 nfs_add_fscache_stats(inode, NFSIOS_FSCACHE_PAGES_UNCACHED, 1);
519 } else {
520 nfs_add_fscache_stats(inode,
521 NFSIOS_FSCACHE_PAGES_WRITTEN_OK, 1);
522 }
523}
diff --git a/fs/nfs/fscache.h b/fs/nfs/fscache.h
new file mode 100644
index 000000000000..6e809bb0ff08
--- /dev/null
+++ b/fs/nfs/fscache.h
@@ -0,0 +1,220 @@
1/* NFS filesystem cache interface definitions
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 */
11
12#ifndef _NFS_FSCACHE_H
13#define _NFS_FSCACHE_H
14
15#include <linux/nfs_fs.h>
16#include <linux/nfs_mount.h>
17#include <linux/nfs4_mount.h>
18#include <linux/fscache.h>
19
20#ifdef CONFIG_NFS_FSCACHE
21
22/*
23 * set of NFS FS-Cache objects that form a superblock key
24 */
25struct nfs_fscache_key {
26 struct rb_node node;
27 struct nfs_client *nfs_client; /* the server */
28
29 /* the elements of the unique key - as used by nfs_compare_super() and
30 * nfs_compare_mount_options() to distinguish superblocks */
31 struct {
32 struct {
33 unsigned long s_flags; /* various flags
34 * (& NFS_MS_MASK) */
35 } super;
36
37 struct {
38 struct nfs_fsid fsid;
39 int flags;
40 unsigned int rsize; /* read size */
41 unsigned int wsize; /* write size */
42 unsigned int acregmin; /* attr cache timeouts */
43 unsigned int acregmax;
44 unsigned int acdirmin;
45 unsigned int acdirmax;
46 } nfs_server;
47
48 struct {
49 rpc_authflavor_t au_flavor;
50 } rpc_auth;
51
52 /* uniquifier - can be used if nfs_server.flags includes
53 * NFS_MOUNT_UNSHARED */
54 u8 uniq_len;
55 char uniquifier[0];
56 } key;
57};
58
59/*
60 * fscache-index.c
61 */
62extern struct fscache_netfs nfs_fscache_netfs;
63extern const struct fscache_cookie_def nfs_fscache_server_index_def;
64extern const struct fscache_cookie_def nfs_fscache_super_index_def;
65extern const struct fscache_cookie_def nfs_fscache_inode_object_def;
66
67extern int nfs_fscache_register(void);
68extern void nfs_fscache_unregister(void);
69
70/*
71 * fscache.c
72 */
73extern void nfs_fscache_get_client_cookie(struct nfs_client *);
74extern void nfs_fscache_release_client_cookie(struct nfs_client *);
75
76extern void nfs_fscache_get_super_cookie(struct super_block *,
77 struct nfs_parsed_mount_data *);
78extern void nfs_fscache_release_super_cookie(struct super_block *);
79
80extern void nfs_fscache_init_inode_cookie(struct inode *);
81extern void nfs_fscache_release_inode_cookie(struct inode *);
82extern void nfs_fscache_zap_inode_cookie(struct inode *);
83extern void nfs_fscache_set_inode_cookie(struct inode *, struct file *);
84extern void nfs_fscache_reset_inode_cookie(struct inode *);
85
86extern void __nfs_fscache_invalidate_page(struct page *, struct inode *);
87extern int nfs_fscache_release_page(struct page *, gfp_t);
88
89extern int __nfs_readpage_from_fscache(struct nfs_open_context *,
90 struct inode *, struct page *);
91extern int __nfs_readpages_from_fscache(struct nfs_open_context *,
92 struct inode *, struct address_space *,
93 struct list_head *, unsigned *);
94extern void __nfs_readpage_to_fscache(struct inode *, struct page *, int);
95
96/*
97 * wait for a page to complete writing to the cache
98 */
99static inline void nfs_fscache_wait_on_page_write(struct nfs_inode *nfsi,
100 struct page *page)
101{
102 if (PageFsCache(page))
103 fscache_wait_on_page_write(nfsi->fscache, page);
104}
105
106/*
107 * release the caching state associated with a page if undergoing complete page
108 * invalidation
109 */
110static inline void nfs_fscache_invalidate_page(struct page *page,
111 struct inode *inode)
112{
113 if (PageFsCache(page))
114 __nfs_fscache_invalidate_page(page, inode);
115}
116
117/*
118 * Retrieve a page from an inode data storage object.
119 */
120static inline int nfs_readpage_from_fscache(struct nfs_open_context *ctx,
121 struct inode *inode,
122 struct page *page)
123{
124 if (NFS_I(inode)->fscache)
125 return __nfs_readpage_from_fscache(ctx, inode, page);
126 return -ENOBUFS;
127}
128
129/*
130 * Retrieve a set of pages from an inode data storage object.
131 */
132static inline int nfs_readpages_from_fscache(struct nfs_open_context *ctx,
133 struct inode *inode,
134 struct address_space *mapping,
135 struct list_head *pages,
136 unsigned *nr_pages)
137{
138 if (NFS_I(inode)->fscache)
139 return __nfs_readpages_from_fscache(ctx, inode, mapping, pages,
140 nr_pages);
141 return -ENOBUFS;
142}
143
144/*
145 * Store a page newly fetched from the server in an inode data storage object
146 * in the cache.
147 */
148static inline void nfs_readpage_to_fscache(struct inode *inode,
149 struct page *page,
150 int sync)
151{
152 if (PageFsCache(page))
153 __nfs_readpage_to_fscache(inode, page, sync);
154}
155
156/*
157 * indicate the client caching state as readable text
158 */
159static inline const char *nfs_server_fscache_state(struct nfs_server *server)
160{
161 if (server->fscache && (server->options & NFS_OPTION_FSCACHE))
162 return "yes";
163 return "no ";
164}
165
166
167#else /* CONFIG_NFS_FSCACHE */
168static inline int nfs_fscache_register(void) { return 0; }
169static inline void nfs_fscache_unregister(void) {}
170
171static inline void nfs_fscache_get_client_cookie(struct nfs_client *clp) {}
172static inline void nfs_fscache_release_client_cookie(struct nfs_client *clp) {}
173
174static inline void nfs_fscache_get_super_cookie(
175 struct super_block *sb,
176 struct nfs_parsed_mount_data *data)
177{
178}
179static inline void nfs_fscache_release_super_cookie(struct super_block *sb) {}
180
181static inline void nfs_fscache_init_inode_cookie(struct inode *inode) {}
182static inline void nfs_fscache_release_inode_cookie(struct inode *inode) {}
183static inline void nfs_fscache_zap_inode_cookie(struct inode *inode) {}
184static inline void nfs_fscache_set_inode_cookie(struct inode *inode,
185 struct file *filp) {}
186static inline void nfs_fscache_reset_inode_cookie(struct inode *inode) {}
187
188static inline int nfs_fscache_release_page(struct page *page, gfp_t gfp)
189{
190 return 1; /* True: may release page */
191}
192static inline void nfs_fscache_invalidate_page(struct page *page,
193 struct inode *inode) {}
194static inline void nfs_fscache_wait_on_page_write(struct nfs_inode *nfsi,
195 struct page *page) {}
196
197static inline int nfs_readpage_from_fscache(struct nfs_open_context *ctx,
198 struct inode *inode,
199 struct page *page)
200{
201 return -ENOBUFS;
202}
203static inline int nfs_readpages_from_fscache(struct nfs_open_context *ctx,
204 struct inode *inode,
205 struct address_space *mapping,
206 struct list_head *pages,
207 unsigned *nr_pages)
208{
209 return -ENOBUFS;
210}
211static inline void nfs_readpage_to_fscache(struct inode *inode,
212 struct page *page, int sync) {}
213
214static inline const char *nfs_server_fscache_state(struct nfs_server *server)
215{
216 return "no ";
217}
218
219#endif /* CONFIG_NFS_FSCACHE */
220#endif /* _NFS_FSCACHE_H */
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index a834d1d850b7..64f87194d390 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -46,6 +46,7 @@
46#include "delegation.h" 46#include "delegation.h"
47#include "iostat.h" 47#include "iostat.h"
48#include "internal.h" 48#include "internal.h"
49#include "fscache.h"
49 50
50#define NFSDBG_FACILITY NFSDBG_VFS 51#define NFSDBG_FACILITY NFSDBG_VFS
51 52
@@ -121,6 +122,7 @@ void nfs_clear_inode(struct inode *inode)
121 BUG_ON(!list_empty(&NFS_I(inode)->open_files)); 122 BUG_ON(!list_empty(&NFS_I(inode)->open_files));
122 nfs_zap_acl_cache(inode); 123 nfs_zap_acl_cache(inode);
123 nfs_access_zap_cache(inode); 124 nfs_access_zap_cache(inode);
125 nfs_fscache_release_inode_cookie(inode);
124} 126}
125 127
126/** 128/**
@@ -355,6 +357,8 @@ nfs_fhget(struct super_block *sb, struct nfs_fh *fh, struct nfs_fattr *fattr)
355 nfsi->attrtimeo_timestamp = now; 357 nfsi->attrtimeo_timestamp = now;
356 nfsi->access_cache = RB_ROOT; 358 nfsi->access_cache = RB_ROOT;
357 359
360 nfs_fscache_init_inode_cookie(inode);
361
358 unlock_new_inode(inode); 362 unlock_new_inode(inode);
359 } else 363 } else
360 nfs_refresh_inode(inode, fattr); 364 nfs_refresh_inode(inode, fattr);
@@ -686,6 +690,7 @@ int nfs_open(struct inode *inode, struct file *filp)
686 ctx->mode = filp->f_mode; 690 ctx->mode = filp->f_mode;
687 nfs_file_set_open_context(filp, ctx); 691 nfs_file_set_open_context(filp, ctx);
688 put_nfs_open_context(ctx); 692 put_nfs_open_context(ctx);
693 nfs_fscache_set_inode_cookie(inode, filp);
689 return 0; 694 return 0;
690} 695}
691 696
@@ -786,6 +791,7 @@ static int nfs_invalidate_mapping_nolock(struct inode *inode, struct address_spa
786 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf)); 791 memset(nfsi->cookieverf, 0, sizeof(nfsi->cookieverf));
787 spin_unlock(&inode->i_lock); 792 spin_unlock(&inode->i_lock);
788 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE); 793 nfs_inc_stats(inode, NFSIOS_DATAINVALIDATE);
794 nfs_fscache_reset_inode_cookie(inode);
789 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n", 795 dfprintk(PAGECACHE, "NFS: (%s/%Ld) data cache invalidated\n",
790 inode->i_sb->s_id, (long long)NFS_FILEID(inode)); 796 inode->i_sb->s_id, (long long)NFS_FILEID(inode));
791 return 0; 797 return 0;
@@ -1030,6 +1036,7 @@ int nfs_refresh_inode(struct inode *inode, struct nfs_fattr *fattr)
1030 spin_lock(&inode->i_lock); 1036 spin_lock(&inode->i_lock);
1031 status = nfs_refresh_inode_locked(inode, fattr); 1037 status = nfs_refresh_inode_locked(inode, fattr);
1032 spin_unlock(&inode->i_lock); 1038 spin_unlock(&inode->i_lock);
1039
1033 return status; 1040 return status;
1034} 1041}
1035 1042
@@ -1436,6 +1443,10 @@ static int __init init_nfs_fs(void)
1436{ 1443{
1437 int err; 1444 int err;
1438 1445
1446 err = nfs_fscache_register();
1447 if (err < 0)
1448 goto out7;
1449
1439 err = nfsiod_start(); 1450 err = nfsiod_start();
1440 if (err) 1451 if (err)
1441 goto out6; 1452 goto out6;
@@ -1488,6 +1499,8 @@ out4:
1488out5: 1499out5:
1489 nfsiod_stop(); 1500 nfsiod_stop();
1490out6: 1501out6:
1502 nfs_fscache_unregister();
1503out7:
1491 return err; 1504 return err;
1492} 1505}
1493 1506
@@ -1498,6 +1511,7 @@ static void __exit exit_nfs_fs(void)
1498 nfs_destroy_readpagecache(); 1511 nfs_destroy_readpagecache();
1499 nfs_destroy_inodecache(); 1512 nfs_destroy_inodecache();
1500 nfs_destroy_nfspagecache(); 1513 nfs_destroy_nfspagecache();
1514 nfs_fscache_unregister();
1501#ifdef CONFIG_PROC_FS 1515#ifdef CONFIG_PROC_FS
1502 rpc_proc_unregister("nfs"); 1516 rpc_proc_unregister("nfs");
1503#endif 1517#endif
diff --git a/fs/nfs/internal.h b/fs/nfs/internal.h
index 2041f68ff1cc..e4d6a8348adf 100644
--- a/fs/nfs/internal.h
+++ b/fs/nfs/internal.h
@@ -5,6 +5,8 @@
5#include <linux/mount.h> 5#include <linux/mount.h>
6#include <linux/security.h> 6#include <linux/security.h>
7 7
8#define NFS_MS_MASK (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_SYNCHRONOUS)
9
8struct nfs_string; 10struct nfs_string;
9 11
10/* Maximum number of readahead requests 12/* Maximum number of readahead requests
@@ -37,10 +39,12 @@ struct nfs_parsed_mount_data {
37 int acregmin, acregmax, 39 int acregmin, acregmax,
38 acdirmin, acdirmax; 40 acdirmin, acdirmax;
39 int namlen; 41 int namlen;
42 unsigned int options;
40 unsigned int bsize; 43 unsigned int bsize;
41 unsigned int auth_flavor_len; 44 unsigned int auth_flavor_len;
42 rpc_authflavor_t auth_flavors[1]; 45 rpc_authflavor_t auth_flavors[1];
43 char *client_address; 46 char *client_address;
47 char *fscache_uniq;
44 48
45 struct { 49 struct {
46 struct sockaddr_storage address; 50 struct sockaddr_storage address;
diff --git a/fs/nfs/iostat.h b/fs/nfs/iostat.h
index a36952810032..a2ab2529b5ca 100644
--- a/fs/nfs/iostat.h
+++ b/fs/nfs/iostat.h
@@ -16,6 +16,9 @@
16 16
17struct nfs_iostats { 17struct nfs_iostats {
18 unsigned long long bytes[__NFSIOS_BYTESMAX]; 18 unsigned long long bytes[__NFSIOS_BYTESMAX];
19#ifdef CONFIG_NFS_FSCACHE
20 unsigned long long fscache[__NFSIOS_FSCACHEMAX];
21#endif
19 unsigned long events[__NFSIOS_COUNTSMAX]; 22 unsigned long events[__NFSIOS_COUNTSMAX];
20} ____cacheline_aligned; 23} ____cacheline_aligned;
21 24
@@ -57,6 +60,21 @@ static inline void nfs_add_stats(const struct inode *inode,
57 nfs_add_server_stats(NFS_SERVER(inode), stat, addend); 60 nfs_add_server_stats(NFS_SERVER(inode), stat, addend);
58} 61}
59 62
63#ifdef CONFIG_NFS_FSCACHE
64static inline void nfs_add_fscache_stats(struct inode *inode,
65 enum nfs_stat_fscachecounters stat,
66 unsigned long addend)
67{
68 struct nfs_iostats *iostats;
69 int cpu;
70
71 cpu = get_cpu();
72 iostats = per_cpu_ptr(NFS_SERVER(inode)->io_stats, cpu);
73 iostats->fscache[stat] += addend;
74 put_cpu_no_resched();
75}
76#endif
77
60static inline struct nfs_iostats *nfs_alloc_iostats(void) 78static inline struct nfs_iostats *nfs_alloc_iostats(void)
61{ 79{
62 return alloc_percpu(struct nfs_iostats); 80 return alloc_percpu(struct nfs_iostats);
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index f856004bb7fa..4ace3c50a8eb 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -24,6 +24,7 @@
24 24
25#include "internal.h" 25#include "internal.h"
26#include "iostat.h" 26#include "iostat.h"
27#include "fscache.h"
27 28
28#define NFSDBG_FACILITY NFSDBG_PAGECACHE 29#define NFSDBG_FACILITY NFSDBG_PAGECACHE
29 30
@@ -111,8 +112,8 @@ static void nfs_readpage_truncate_uninitialised_page(struct nfs_read_data *data)
111 } 112 }
112} 113}
113 114
114static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode, 115int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
115 struct page *page) 116 struct page *page)
116{ 117{
117 LIST_HEAD(one_request); 118 LIST_HEAD(one_request);
118 struct nfs_page *new; 119 struct nfs_page *new;
@@ -139,6 +140,11 @@ static int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
139 140
140static void nfs_readpage_release(struct nfs_page *req) 141static void nfs_readpage_release(struct nfs_page *req)
141{ 142{
143 struct inode *d_inode = req->wb_context->path.dentry->d_inode;
144
145 if (PageUptodate(req->wb_page))
146 nfs_readpage_to_fscache(d_inode, req->wb_page, 0);
147
142 unlock_page(req->wb_page); 148 unlock_page(req->wb_page);
143 149
144 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n", 150 dprintk("NFS: read done (%s/%Ld %d@%Ld)\n",
@@ -510,8 +516,15 @@ int nfs_readpage(struct file *file, struct page *page)
510 } else 516 } else
511 ctx = get_nfs_open_context(nfs_file_open_context(file)); 517 ctx = get_nfs_open_context(nfs_file_open_context(file));
512 518
519 if (!IS_SYNC(inode)) {
520 error = nfs_readpage_from_fscache(ctx, inode, page);
521 if (error == 0)
522 goto out;
523 }
524
513 error = nfs_readpage_async(ctx, inode, page); 525 error = nfs_readpage_async(ctx, inode, page);
514 526
527out:
515 put_nfs_open_context(ctx); 528 put_nfs_open_context(ctx);
516 return error; 529 return error;
517out_unlock: 530out_unlock:
@@ -584,6 +597,15 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,
584 return -EBADF; 597 return -EBADF;
585 } else 598 } else
586 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp)); 599 desc.ctx = get_nfs_open_context(nfs_file_open_context(filp));
600
601 /* attempt to read as many of the pages as possible from the cache
602 * - this returns -ENOBUFS immediately if the cookie is negative
603 */
604 ret = nfs_readpages_from_fscache(desc.ctx, inode, mapping,
605 pages, &nr_pages);
606 if (ret == 0)
607 goto read_complete; /* all pages were read */
608
587 if (rsize < PAGE_CACHE_SIZE) 609 if (rsize < PAGE_CACHE_SIZE)
588 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); 610 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
589 else 611 else
@@ -594,6 +616,7 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,
594 nfs_pageio_complete(&pgio); 616 nfs_pageio_complete(&pgio);
595 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 617 npages = (pgio.pg_bytes_written + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
596 nfs_add_stats(inode, NFSIOS_READPAGES, npages); 618 nfs_add_stats(inode, NFSIOS_READPAGES, npages);
619read_complete:
597 put_nfs_open_context(desc.ctx); 620 put_nfs_open_context(desc.ctx);
598out: 621out:
599 return ret; 622 return ret;
diff --git a/fs/nfs/super.c b/fs/nfs/super.c
index 0942fcbbad3c..82eaadbff408 100644
--- a/fs/nfs/super.c
+++ b/fs/nfs/super.c
@@ -60,6 +60,7 @@
60#include "delegation.h" 60#include "delegation.h"
61#include "iostat.h" 61#include "iostat.h"
62#include "internal.h" 62#include "internal.h"
63#include "fscache.h"
63 64
64#define NFSDBG_FACILITY NFSDBG_VFS 65#define NFSDBG_FACILITY NFSDBG_VFS
65 66
@@ -76,6 +77,7 @@ enum {
76 Opt_rdirplus, Opt_nordirplus, 77 Opt_rdirplus, Opt_nordirplus,
77 Opt_sharecache, Opt_nosharecache, 78 Opt_sharecache, Opt_nosharecache,
78 Opt_resvport, Opt_noresvport, 79 Opt_resvport, Opt_noresvport,
80 Opt_fscache, Opt_nofscache,
79 81
80 /* Mount options that take integer arguments */ 82 /* Mount options that take integer arguments */
81 Opt_port, 83 Opt_port,
@@ -93,6 +95,7 @@ enum {
93 Opt_sec, Opt_proto, Opt_mountproto, Opt_mounthost, 95 Opt_sec, Opt_proto, Opt_mountproto, Opt_mounthost,
94 Opt_addr, Opt_mountaddr, Opt_clientaddr, 96 Opt_addr, Opt_mountaddr, Opt_clientaddr,
95 Opt_lookupcache, 97 Opt_lookupcache,
98 Opt_fscache_uniq,
96 99
97 /* Special mount options */ 100 /* Special mount options */
98 Opt_userspace, Opt_deprecated, Opt_sloppy, 101 Opt_userspace, Opt_deprecated, Opt_sloppy,
@@ -132,6 +135,9 @@ static const match_table_t nfs_mount_option_tokens = {
132 { Opt_nosharecache, "nosharecache" }, 135 { Opt_nosharecache, "nosharecache" },
133 { Opt_resvport, "resvport" }, 136 { Opt_resvport, "resvport" },
134 { Opt_noresvport, "noresvport" }, 137 { Opt_noresvport, "noresvport" },
138 { Opt_fscache, "fsc" },
139 { Opt_fscache_uniq, "fsc=%s" },
140 { Opt_nofscache, "nofsc" },
135 141
136 { Opt_port, "port=%u" }, 142 { Opt_port, "port=%u" },
137 { Opt_rsize, "rsize=%u" }, 143 { Opt_rsize, "rsize=%u" },
@@ -563,6 +569,8 @@ static void nfs_show_mount_options(struct seq_file *m, struct nfs_server *nfss,
563 if (clp->rpc_ops->version == 4) 569 if (clp->rpc_ops->version == 4)
564 seq_printf(m, ",clientaddr=%s", clp->cl_ipaddr); 570 seq_printf(m, ",clientaddr=%s", clp->cl_ipaddr);
565#endif 571#endif
572 if (nfss->options & NFS_OPTION_FSCACHE)
573 seq_printf(m, ",fsc");
566} 574}
567 575
568/* 576/*
@@ -641,6 +649,10 @@ static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
641 totals.events[i] += stats->events[i]; 649 totals.events[i] += stats->events[i];
642 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 650 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
643 totals.bytes[i] += stats->bytes[i]; 651 totals.bytes[i] += stats->bytes[i];
652#ifdef CONFIG_NFS_FSCACHE
653 for (i = 0; i < __NFSIOS_FSCACHEMAX; i++)
654 totals.fscache[i] += stats->fscache[i];
655#endif
644 656
645 preempt_enable(); 657 preempt_enable();
646 } 658 }
@@ -651,6 +663,13 @@ static int nfs_show_stats(struct seq_file *m, struct vfsmount *mnt)
651 seq_printf(m, "\n\tbytes:\t"); 663 seq_printf(m, "\n\tbytes:\t");
652 for (i = 0; i < __NFSIOS_BYTESMAX; i++) 664 for (i = 0; i < __NFSIOS_BYTESMAX; i++)
653 seq_printf(m, "%Lu ", totals.bytes[i]); 665 seq_printf(m, "%Lu ", totals.bytes[i]);
666#ifdef CONFIG_NFS_FSCACHE
667 if (nfss->options & NFS_OPTION_FSCACHE) {
668 seq_printf(m, "\n\tfsc:\t");
669 for (i = 0; i < __NFSIOS_FSCACHEMAX; i++)
670 seq_printf(m, "%Lu ", totals.bytes[i]);
671 }
672#endif
654 seq_printf(m, "\n"); 673 seq_printf(m, "\n");
655 674
656 rpc_print_iostats(m, nfss->client); 675 rpc_print_iostats(m, nfss->client);
@@ -1044,6 +1063,24 @@ static int nfs_parse_mount_options(char *raw,
1044 case Opt_noresvport: 1063 case Opt_noresvport:
1045 mnt->flags |= NFS_MOUNT_NORESVPORT; 1064 mnt->flags |= NFS_MOUNT_NORESVPORT;
1046 break; 1065 break;
1066 case Opt_fscache:
1067 mnt->options |= NFS_OPTION_FSCACHE;
1068 kfree(mnt->fscache_uniq);
1069 mnt->fscache_uniq = NULL;
1070 break;
1071 case Opt_nofscache:
1072 mnt->options &= ~NFS_OPTION_FSCACHE;
1073 kfree(mnt->fscache_uniq);
1074 mnt->fscache_uniq = NULL;
1075 break;
1076 case Opt_fscache_uniq:
1077 string = match_strdup(args);
1078 if (!string)
1079 goto out_nomem;
1080 kfree(mnt->fscache_uniq);
1081 mnt->fscache_uniq = string;
1082 mnt->options |= NFS_OPTION_FSCACHE;
1083 break;
1047 1084
1048 /* 1085 /*
1049 * options that take numeric values 1086 * options that take numeric values
@@ -1870,8 +1907,6 @@ static void nfs_clone_super(struct super_block *sb,
1870 nfs_initialise_sb(sb); 1907 nfs_initialise_sb(sb);
1871} 1908}
1872 1909
1873#define NFS_MS_MASK (MS_RDONLY|MS_NOSUID|MS_NODEV|MS_NOEXEC|MS_SYNCHRONOUS)
1874
1875static int nfs_compare_mount_options(const struct super_block *s, const struct nfs_server *b, int flags) 1910static int nfs_compare_mount_options(const struct super_block *s, const struct nfs_server *b, int flags)
1876{ 1911{
1877 const struct nfs_server *a = s->s_fs_info; 1912 const struct nfs_server *a = s->s_fs_info;
@@ -2036,6 +2071,7 @@ static int nfs_get_sb(struct file_system_type *fs_type,
2036 if (!s->s_root) { 2071 if (!s->s_root) {
2037 /* initial superblock/root creation */ 2072 /* initial superblock/root creation */
2038 nfs_fill_super(s, data); 2073 nfs_fill_super(s, data);
2074 nfs_fscache_get_super_cookie(s, data);
2039 } 2075 }
2040 2076
2041 mntroot = nfs_get_root(s, mntfh); 2077 mntroot = nfs_get_root(s, mntfh);
@@ -2056,6 +2092,7 @@ static int nfs_get_sb(struct file_system_type *fs_type,
2056out: 2092out:
2057 kfree(data->nfs_server.hostname); 2093 kfree(data->nfs_server.hostname);
2058 kfree(data->mount_server.hostname); 2094 kfree(data->mount_server.hostname);
2095 kfree(data->fscache_uniq);
2059 security_free_mnt_opts(&data->lsm_opts); 2096 security_free_mnt_opts(&data->lsm_opts);
2060out_free_fh: 2097out_free_fh:
2061 kfree(mntfh); 2098 kfree(mntfh);
@@ -2083,6 +2120,7 @@ static void nfs_kill_super(struct super_block *s)
2083 2120
2084 bdi_unregister(&server->backing_dev_info); 2121 bdi_unregister(&server->backing_dev_info);
2085 kill_anon_super(s); 2122 kill_anon_super(s);
2123 nfs_fscache_release_super_cookie(s);
2086 nfs_free_server(server); 2124 nfs_free_server(server);
2087} 2125}
2088 2126
@@ -2390,6 +2428,7 @@ static int nfs4_get_sb(struct file_system_type *fs_type,
2390 if (!s->s_root) { 2428 if (!s->s_root) {
2391 /* initial superblock/root creation */ 2429 /* initial superblock/root creation */
2392 nfs4_fill_super(s); 2430 nfs4_fill_super(s);
2431 nfs_fscache_get_super_cookie(s, data);
2393 } 2432 }
2394 2433
2395 mntroot = nfs4_get_root(s, mntfh); 2434 mntroot = nfs4_get_root(s, mntfh);
@@ -2411,6 +2450,7 @@ out:
2411 kfree(data->client_address); 2450 kfree(data->client_address);
2412 kfree(data->nfs_server.export_path); 2451 kfree(data->nfs_server.export_path);
2413 kfree(data->nfs_server.hostname); 2452 kfree(data->nfs_server.hostname);
2453 kfree(data->fscache_uniq);
2414 security_free_mnt_opts(&data->lsm_opts); 2454 security_free_mnt_opts(&data->lsm_opts);
2415out_free_fh: 2455out_free_fh:
2416 kfree(mntfh); 2456 kfree(mntfh);
@@ -2437,6 +2477,7 @@ static void nfs4_kill_super(struct super_block *sb)
2437 kill_anon_super(sb); 2477 kill_anon_super(sb);
2438 2478
2439 nfs4_renewd_prepare_shutdown(server); 2479 nfs4_renewd_prepare_shutdown(server);
2480 nfs_fscache_release_super_cookie(sb);
2440 nfs_free_server(server); 2481 nfs_free_server(server);
2441} 2482}
2442 2483
diff --git a/fs/splice.c b/fs/splice.c
index 4ed0ba44a966..dd727d43e5b7 100644
--- a/fs/splice.c
+++ b/fs/splice.c
@@ -59,7 +59,8 @@ static int page_cache_pipe_buf_steal(struct pipe_inode_info *pipe,
59 */ 59 */
60 wait_on_page_writeback(page); 60 wait_on_page_writeback(page);
61 61
62 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL)) 62 if (page_has_private(page) &&
63 !try_to_release_page(page, GFP_KERNEL))
63 goto out_unlock; 64 goto out_unlock;
64 65
65 /* 66 /*
diff --git a/fs/super.c b/fs/super.c
index 2ba481518ba7..77cb4ec919b9 100644
--- a/fs/super.c
+++ b/fs/super.c
@@ -287,6 +287,7 @@ int fsync_super(struct super_block *sb)
287 __fsync_super(sb); 287 __fsync_super(sb);
288 return sync_blockdev(sb->s_bdev); 288 return sync_blockdev(sb->s_bdev);
289} 289}
290EXPORT_SYMBOL_GPL(fsync_super);
290 291
291/** 292/**
292 * generic_shutdown_super - common helper for ->kill_sb() 293 * generic_shutdown_super - common helper for ->kill_sb()
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-23 07:27:29 -0500