aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
Diffstat (limited to 'fs')
-rw-r--r--fs/nfs/Kconfig8
-rw-r--r--fs/nfs/Makefile4
-rw-r--r--fs/nfs/callback_proc.c8
-rw-r--r--fs/nfs/client.c10
-rw-r--r--fs/nfs/file.c5
-rw-r--r--fs/nfs/inode.c3
-rw-r--r--fs/nfs/nfs4filelayout.c280
-rw-r--r--fs/nfs/nfs4filelayout.h94
-rw-r--r--fs/nfs/nfs4filelayoutdev.c448
-rw-r--r--fs/nfs/nfs4proc.c218
-rw-r--r--fs/nfs/nfs4state.c2
-rw-r--r--fs/nfs/nfs4xdr.c360
-rw-r--r--fs/nfs/pnfs.c783
-rw-r--r--fs/nfs/pnfs.h189
-rw-r--r--fs/nfs/read.c3
-rw-r--r--fs/nfsd/nfs4callback.c1
16 files changed, 2374 insertions, 42 deletions
diff --git a/fs/nfs/Kconfig b/fs/nfs/Kconfig
index 5c55c26af165..fd667652c502 100644
--- a/fs/nfs/Kconfig
+++ b/fs/nfs/Kconfig
@@ -77,13 +77,17 @@ config NFS_V4
77 77
78config NFS_V4_1 78config NFS_V4_1
79 bool "NFS client support for NFSv4.1 (EXPERIMENTAL)" 79 bool "NFS client support for NFSv4.1 (EXPERIMENTAL)"
80 depends on NFS_V4 && EXPERIMENTAL 80 depends on NFS_FS && NFS_V4 && EXPERIMENTAL
81 select PNFS_FILE_LAYOUT
81 help 82 help
82 This option enables support for minor version 1 of the NFSv4 protocol 83 This option enables support for minor version 1 of the NFSv4 protocol
83 (draft-ietf-nfsv4-minorversion1) in the kernel's NFS client. 84 (RFC 5661) in the kernel's NFS client.
84 85
85 If unsure, say N. 86 If unsure, say N.
86 87
88config PNFS_FILE_LAYOUT
89 tristate
90
87config ROOT_NFS 91config ROOT_NFS
88 bool "Root file system on NFS" 92 bool "Root file system on NFS"
89 depends on NFS_FS=y && IP_PNP 93 depends on NFS_FS=y && IP_PNP
diff --git a/fs/nfs/Makefile b/fs/nfs/Makefile
index da7fda639eac..4776ff9e3814 100644
--- a/fs/nfs/Makefile
+++ b/fs/nfs/Makefile
@@ -15,5 +15,9 @@ nfs-$(CONFIG_NFS_V4) += nfs4proc.o nfs4xdr.o nfs4state.o nfs4renewd.o \
15 delegation.o idmap.o \ 15 delegation.o idmap.o \
16 callback.o callback_xdr.o callback_proc.o \ 16 callback.o callback_xdr.o callback_proc.o \
17 nfs4namespace.o 17 nfs4namespace.o
18nfs-$(CONFIG_NFS_V4_1) += pnfs.o
18nfs-$(CONFIG_SYSCTL) += sysctl.o 19nfs-$(CONFIG_SYSCTL) += sysctl.o
19nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o 20nfs-$(CONFIG_NFS_FSCACHE) += fscache.o fscache-index.o
21
22obj-$(CONFIG_PNFS_FILE_LAYOUT) += nfs_layout_nfsv41_files.o
23nfs_layout_nfsv41_files-y := nfs4filelayout.o nfs4filelayoutdev.o
diff --git a/fs/nfs/callback_proc.c b/fs/nfs/callback_proc.c
index 930d10fecdaf..2950fca0c61b 100644
--- a/fs/nfs/callback_proc.c
+++ b/fs/nfs/callback_proc.c
@@ -118,11 +118,11 @@ int nfs41_validate_delegation_stateid(struct nfs_delegation *delegation, const n
118 if (delegation == NULL) 118 if (delegation == NULL)
119 return 0; 119 return 0;
120 120
121 /* seqid is 4-bytes long */ 121 if (stateid->stateid.seqid != 0)
122 if (((u32 *) &stateid->data)[0] != 0)
123 return 0; 122 return 0;
124 if (memcmp(&delegation->stateid.data[4], &stateid->data[4], 123 if (memcmp(&delegation->stateid.stateid.other,
125 sizeof(stateid->data)-4)) 124 &stateid->stateid.other,
125 NFS4_STATEID_OTHER_SIZE))
126 return 0; 126 return 0;
127 127
128 return 1; 128 return 1;
diff --git a/fs/nfs/client.c b/fs/nfs/client.c
index a882785eba41..fd6f0a70021b 100644
--- a/fs/nfs/client.c
+++ b/fs/nfs/client.c
@@ -48,6 +48,7 @@
48#include "iostat.h" 48#include "iostat.h"
49#include "internal.h" 49#include "internal.h"
50#include "fscache.h" 50#include "fscache.h"
51#include "pnfs.h"
51 52
52#define NFSDBG_FACILITY NFSDBG_CLIENT 53#define NFSDBG_FACILITY NFSDBG_CLIENT
53 54
@@ -155,7 +156,9 @@ static struct nfs_client *nfs_alloc_client(const struct nfs_client_initdata *cl_
155 cred = rpc_lookup_machine_cred(); 156 cred = rpc_lookup_machine_cred();
156 if (!IS_ERR(cred)) 157 if (!IS_ERR(cred))
157 clp->cl_machine_cred = cred; 158 clp->cl_machine_cred = cred;
158 159#if defined(CONFIG_NFS_V4_1)
160 INIT_LIST_HEAD(&clp->cl_layouts);
161#endif
159 nfs_fscache_get_client_cookie(clp); 162 nfs_fscache_get_client_cookie(clp);
160 163
161 return clp; 164 return clp;
@@ -252,6 +255,7 @@ void nfs_put_client(struct nfs_client *clp)
252 nfs_free_client(clp); 255 nfs_free_client(clp);
253 } 256 }
254} 257}
258EXPORT_SYMBOL_GPL(nfs_put_client);
255 259
256#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE) 260#if defined(CONFIG_IPV6) || defined(CONFIG_IPV6_MODULE)
257/* 261/*
@@ -900,6 +904,8 @@ static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *
900 if (server->wsize > NFS_MAX_FILE_IO_SIZE) 904 if (server->wsize > NFS_MAX_FILE_IO_SIZE)
901 server->wsize = NFS_MAX_FILE_IO_SIZE; 905 server->wsize = NFS_MAX_FILE_IO_SIZE;
902 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 906 server->wpages = (server->wsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
907 set_pnfs_layoutdriver(server, fsinfo->layouttype);
908
903 server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL); 909 server->wtmult = nfs_block_bits(fsinfo->wtmult, NULL);
904 910
905 server->dtsize = nfs_block_size(fsinfo->dtpref, NULL); 911 server->dtsize = nfs_block_size(fsinfo->dtpref, NULL);
@@ -939,6 +945,7 @@ static int nfs_probe_fsinfo(struct nfs_server *server, struct nfs_fh *mntfh, str
939 } 945 }
940 946
941 fsinfo.fattr = fattr; 947 fsinfo.fattr = fattr;
948 fsinfo.layouttype = 0;
942 error = clp->rpc_ops->fsinfo(server, mntfh, &fsinfo); 949 error = clp->rpc_ops->fsinfo(server, mntfh, &fsinfo);
943 if (error < 0) 950 if (error < 0)
944 goto out_error; 951 goto out_error;
@@ -1021,6 +1028,7 @@ void nfs_free_server(struct nfs_server *server)
1021{ 1028{
1022 dprintk("--> nfs_free_server()\n"); 1029 dprintk("--> nfs_free_server()\n");
1023 1030
1031 unset_pnfs_layoutdriver(server);
1024 spin_lock(&nfs_client_lock); 1032 spin_lock(&nfs_client_lock);
1025 list_del(&server->client_link); 1033 list_del(&server->client_link);
1026 list_del(&server->master_link); 1034 list_del(&server->master_link);
diff --git a/fs/nfs/file.c b/fs/nfs/file.c
index e18c31e08a28..e756075637b0 100644
--- a/fs/nfs/file.c
+++ b/fs/nfs/file.c
@@ -36,6 +36,7 @@
36#include "internal.h" 36#include "internal.h"
37#include "iostat.h" 37#include "iostat.h"
38#include "fscache.h" 38#include "fscache.h"
39#include "pnfs.h"
39 40
40#define NFSDBG_FACILITY NFSDBG_FILE 41#define NFSDBG_FACILITY NFSDBG_FILE
41 42
@@ -386,6 +387,10 @@ static int nfs_write_begin(struct file *file, struct address_space *mapping,
386 file->f_path.dentry->d_name.name, 387 file->f_path.dentry->d_name.name,
387 mapping->host->i_ino, len, (long long) pos); 388 mapping->host->i_ino, len, (long long) pos);
388 389
390 pnfs_update_layout(mapping->host,
391 nfs_file_open_context(file),
392 IOMODE_RW);
393
389start: 394start:
390 /* 395 /*
391 * Prevent starvation issues if someone is doing a consistency 396 * Prevent starvation issues if someone is doing a consistency
diff --git a/fs/nfs/inode.c b/fs/nfs/inode.c
index 6eec28656415..314f57164602 100644
--- a/fs/nfs/inode.c
+++ b/fs/nfs/inode.c
@@ -48,6 +48,7 @@
48#include "internal.h" 48#include "internal.h"
49#include "fscache.h" 49#include "fscache.h"
50#include "dns_resolve.h" 50#include "dns_resolve.h"
51#include "pnfs.h"
51 52
52#define NFSDBG_FACILITY NFSDBG_VFS 53#define NFSDBG_FACILITY NFSDBG_VFS
53 54
@@ -1410,6 +1411,7 @@ void nfs4_evict_inode(struct inode *inode)
1410{ 1411{
1411 truncate_inode_pages(&inode->i_data, 0); 1412 truncate_inode_pages(&inode->i_data, 0);
1412 end_writeback(inode); 1413 end_writeback(inode);
1414 pnfs_destroy_layout(NFS_I(inode));
1413 /* If we are holding a delegation, return it! */ 1415 /* If we are holding a delegation, return it! */
1414 nfs_inode_return_delegation_noreclaim(inode); 1416 nfs_inode_return_delegation_noreclaim(inode);
1415 /* First call standard NFS clear_inode() code */ 1417 /* First call standard NFS clear_inode() code */
@@ -1447,6 +1449,7 @@ static inline void nfs4_init_once(struct nfs_inode *nfsi)
1447 nfsi->delegation = NULL; 1449 nfsi->delegation = NULL;
1448 nfsi->delegation_state = 0; 1450 nfsi->delegation_state = 0;
1449 init_rwsem(&nfsi->rwsem); 1451 init_rwsem(&nfsi->rwsem);
1452 nfsi->layout = NULL;
1450#endif 1453#endif
1451} 1454}
1452 1455
diff --git a/fs/nfs/nfs4filelayout.c b/fs/nfs/nfs4filelayout.c
new file mode 100644
index 000000000000..2e92f0d8d654
--- /dev/null
+++ b/fs/nfs/nfs4filelayout.c
@@ -0,0 +1,280 @@
1/*
2 * Module for the pnfs nfs4 file layout driver.
3 * Defines all I/O and Policy interface operations, plus code
4 * to register itself with the pNFS client.
5 *
6 * Copyright (c) 2002
7 * The Regents of the University of Michigan
8 * All Rights Reserved
9 *
10 * Dean Hildebrand <dhildebz@umich.edu>
11 *
12 * Permission is granted to use, copy, create derivative works, and
13 * redistribute this software and such derivative works for any purpose,
14 * so long as the name of the University of Michigan is not used in
15 * any advertising or publicity pertaining to the use or distribution
16 * of this software without specific, written prior authorization. If
17 * the above copyright notice or any other identification of the
18 * University of Michigan is included in any copy of any portion of
19 * this software, then the disclaimer below must also be included.
20 *
21 * This software is provided as is, without representation or warranty
22 * of any kind either express or implied, including without limitation
23 * the implied warranties of merchantability, fitness for a particular
24 * purpose, or noninfringement. The Regents of the University of
25 * Michigan shall not be liable for any damages, including special,
26 * indirect, incidental, or consequential damages, with respect to any
27 * claim arising out of or in connection with the use of the software,
28 * even if it has been or is hereafter advised of the possibility of
29 * such damages.
30 */
31
32#include <linux/nfs_fs.h>
33
34#include "internal.h"
35#include "nfs4filelayout.h"
36
37#define NFSDBG_FACILITY NFSDBG_PNFS_LD
38
39MODULE_LICENSE("GPL");
40MODULE_AUTHOR("Dean Hildebrand <dhildebz@umich.edu>");
41MODULE_DESCRIPTION("The NFSv4 file layout driver");
42
43static int
44filelayout_set_layoutdriver(struct nfs_server *nfss)
45{
46 int status = pnfs_alloc_init_deviceid_cache(nfss->nfs_client,
47 nfs4_fl_free_deviceid_callback);
48 if (status) {
49 printk(KERN_WARNING "%s: deviceid cache could not be "
50 "initialized\n", __func__);
51 return status;
52 }
53 dprintk("%s: deviceid cache has been initialized successfully\n",
54 __func__);
55 return 0;
56}
57
58/* Clear out the layout by destroying its device list */
59static int
60filelayout_clear_layoutdriver(struct nfs_server *nfss)
61{
62 dprintk("--> %s\n", __func__);
63
64 if (nfss->nfs_client->cl_devid_cache)
65 pnfs_put_deviceid_cache(nfss->nfs_client);
66 return 0;
67}
68
69/*
70 * filelayout_check_layout()
71 *
72 * Make sure layout segment parameters are sane WRT the device.
73 * At this point no generic layer initialization of the lseg has occurred,
74 * and nothing has been added to the layout_hdr cache.
75 *
76 */
77static int
78filelayout_check_layout(struct pnfs_layout_hdr *lo,
79 struct nfs4_filelayout_segment *fl,
80 struct nfs4_layoutget_res *lgr,
81 struct nfs4_deviceid *id)
82{
83 struct nfs4_file_layout_dsaddr *dsaddr;
84 int status = -EINVAL;
85 struct nfs_server *nfss = NFS_SERVER(lo->inode);
86
87 dprintk("--> %s\n", __func__);
88
89 if (fl->pattern_offset > lgr->range.offset) {
90 dprintk("%s pattern_offset %lld to large\n",
91 __func__, fl->pattern_offset);
92 goto out;
93 }
94
95 if (fl->stripe_unit % PAGE_SIZE) {
96 dprintk("%s Stripe unit (%u) not page aligned\n",
97 __func__, fl->stripe_unit);
98 goto out;
99 }
100
101 /* find and reference the deviceid */
102 dsaddr = nfs4_fl_find_get_deviceid(nfss->nfs_client, id);
103 if (dsaddr == NULL) {
104 dsaddr = get_device_info(lo->inode, id);
105 if (dsaddr == NULL)
106 goto out;
107 }
108 fl->dsaddr = dsaddr;
109
110 if (fl->first_stripe_index < 0 ||
111 fl->first_stripe_index >= dsaddr->stripe_count) {
112 dprintk("%s Bad first_stripe_index %d\n",
113 __func__, fl->first_stripe_index);
114 goto out_put;
115 }
116
117 if ((fl->stripe_type == STRIPE_SPARSE &&
118 fl->num_fh > 1 && fl->num_fh != dsaddr->ds_num) ||
119 (fl->stripe_type == STRIPE_DENSE &&
120 fl->num_fh != dsaddr->stripe_count)) {
121 dprintk("%s num_fh %u not valid for given packing\n",
122 __func__, fl->num_fh);
123 goto out_put;
124 }
125
126 if (fl->stripe_unit % nfss->rsize || fl->stripe_unit % nfss->wsize) {
127 dprintk("%s Stripe unit (%u) not aligned with rsize %u "
128 "wsize %u\n", __func__, fl->stripe_unit, nfss->rsize,
129 nfss->wsize);
130 }
131
132 status = 0;
133out:
134 dprintk("--> %s returns %d\n", __func__, status);
135 return status;
136out_put:
137 pnfs_put_deviceid(nfss->nfs_client->cl_devid_cache, &dsaddr->deviceid);
138 goto out;
139}
140
141static void filelayout_free_fh_array(struct nfs4_filelayout_segment *fl)
142{
143 int i;
144
145 for (i = 0; i < fl->num_fh; i++) {
146 if (!fl->fh_array[i])
147 break;
148 kfree(fl->fh_array[i]);
149 }
150 kfree(fl->fh_array);
151 fl->fh_array = NULL;
152}
153
154static void
155_filelayout_free_lseg(struct nfs4_filelayout_segment *fl)
156{
157 filelayout_free_fh_array(fl);
158 kfree(fl);
159}
160
161static int
162filelayout_decode_layout(struct pnfs_layout_hdr *flo,
163 struct nfs4_filelayout_segment *fl,
164 struct nfs4_layoutget_res *lgr,
165 struct nfs4_deviceid *id)
166{
167 uint32_t *p = (uint32_t *)lgr->layout.buf;
168 uint32_t nfl_util;
169 int i;
170
171 dprintk("%s: set_layout_map Begin\n", __func__);
172
173 memcpy(id, p, sizeof(*id));
174 p += XDR_QUADLEN(NFS4_DEVICEID4_SIZE);
175 print_deviceid(id);
176
177 nfl_util = be32_to_cpup(p++);
178 if (nfl_util & NFL4_UFLG_COMMIT_THRU_MDS)
179 fl->commit_through_mds = 1;
180 if (nfl_util & NFL4_UFLG_DENSE)
181 fl->stripe_type = STRIPE_DENSE;
182 else
183 fl->stripe_type = STRIPE_SPARSE;
184 fl->stripe_unit = nfl_util & ~NFL4_UFLG_MASK;
185
186 fl->first_stripe_index = be32_to_cpup(p++);
187 p = xdr_decode_hyper(p, &fl->pattern_offset);
188 fl->num_fh = be32_to_cpup(p++);
189
190 dprintk("%s: nfl_util 0x%X num_fh %u fsi %u po %llu\n",
191 __func__, nfl_util, fl->num_fh, fl->first_stripe_index,
192 fl->pattern_offset);
193
194 fl->fh_array = kzalloc(fl->num_fh * sizeof(struct nfs_fh *),
195 GFP_KERNEL);
196 if (!fl->fh_array)
197 return -ENOMEM;
198
199 for (i = 0; i < fl->num_fh; i++) {
200 /* Do we want to use a mempool here? */
201 fl->fh_array[i] = kmalloc(sizeof(struct nfs_fh), GFP_KERNEL);
202 if (!fl->fh_array[i]) {
203 filelayout_free_fh_array(fl);
204 return -ENOMEM;
205 }
206 fl->fh_array[i]->size = be32_to_cpup(p++);
207 if (sizeof(struct nfs_fh) < fl->fh_array[i]->size) {
208 printk(KERN_ERR "Too big fh %d received %d\n",
209 i, fl->fh_array[i]->size);
210 filelayout_free_fh_array(fl);
211 return -EIO;
212 }
213 memcpy(fl->fh_array[i]->data, p, fl->fh_array[i]->size);
214 p += XDR_QUADLEN(fl->fh_array[i]->size);
215 dprintk("DEBUG: %s: fh len %d\n", __func__,
216 fl->fh_array[i]->size);
217 }
218
219 return 0;
220}
221
222static struct pnfs_layout_segment *
223filelayout_alloc_lseg(struct pnfs_layout_hdr *layoutid,
224 struct nfs4_layoutget_res *lgr)
225{
226 struct nfs4_filelayout_segment *fl;
227 int rc;
228 struct nfs4_deviceid id;
229
230 dprintk("--> %s\n", __func__);
231 fl = kzalloc(sizeof(*fl), GFP_KERNEL);
232 if (!fl)
233 return NULL;
234
235 rc = filelayout_decode_layout(layoutid, fl, lgr, &id);
236 if (rc != 0 || filelayout_check_layout(layoutid, fl, lgr, &id)) {
237 _filelayout_free_lseg(fl);
238 return NULL;
239 }
240 return &fl->generic_hdr;
241}
242
243static void
244filelayout_free_lseg(struct pnfs_layout_segment *lseg)
245{
246 struct nfs_server *nfss = NFS_SERVER(lseg->layout->inode);
247 struct nfs4_filelayout_segment *fl = FILELAYOUT_LSEG(lseg);
248
249 dprintk("--> %s\n", __func__);
250 pnfs_put_deviceid(nfss->nfs_client->cl_devid_cache,
251 &fl->dsaddr->deviceid);
252 _filelayout_free_lseg(fl);
253}
254
255static struct pnfs_layoutdriver_type filelayout_type = {
256 .id = LAYOUT_NFSV4_1_FILES,
257 .name = "LAYOUT_NFSV4_1_FILES",
258 .owner = THIS_MODULE,
259 .set_layoutdriver = filelayout_set_layoutdriver,
260 .clear_layoutdriver = filelayout_clear_layoutdriver,
261 .alloc_lseg = filelayout_alloc_lseg,
262 .free_lseg = filelayout_free_lseg,
263};
264
265static int __init nfs4filelayout_init(void)
266{
267 printk(KERN_INFO "%s: NFSv4 File Layout Driver Registering...\n",
268 __func__);
269 return pnfs_register_layoutdriver(&filelayout_type);
270}
271
272static void __exit nfs4filelayout_exit(void)
273{
274 printk(KERN_INFO "%s: NFSv4 File Layout Driver Unregistering...\n",
275 __func__);
276 pnfs_unregister_layoutdriver(&filelayout_type);
277}
278
279module_init(nfs4filelayout_init);
280module_exit(nfs4filelayout_exit);
diff --git a/fs/nfs/nfs4filelayout.h b/fs/nfs/nfs4filelayout.h
new file mode 100644
index 000000000000..bbf60dd2ab9d
--- /dev/null
+++ b/fs/nfs/nfs4filelayout.h
@@ -0,0 +1,94 @@
1/*
2 * NFSv4 file layout driver data structures.
3 *
4 * Copyright (c) 2002
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#ifndef FS_NFS_NFS4FILELAYOUT_H
31#define FS_NFS_NFS4FILELAYOUT_H
32
33#include "pnfs.h"
34
35/*
36 * Field testing shows we need to support upto 4096 stripe indices.
37 * We store each index as a u8 (u32 on the wire) to keep the memory footprint
38 * reasonable. This in turn means we support a maximum of 256
39 * RFC 5661 multipath_list4 structures.
40 */
41#define NFS4_PNFS_MAX_STRIPE_CNT 4096
42#define NFS4_PNFS_MAX_MULTI_CNT 256 /* 256 fit into a u8 stripe_index */
43
44enum stripetype4 {
45 STRIPE_SPARSE = 1,
46 STRIPE_DENSE = 2
47};
48
49/* Individual ip address */
50struct nfs4_pnfs_ds {
51 struct list_head ds_node; /* nfs4_pnfs_dev_hlist dev_dslist */
52 u32 ds_ip_addr;
53 u32 ds_port;
54 struct nfs_client *ds_clp;
55 atomic_t ds_count;
56};
57
58struct nfs4_file_layout_dsaddr {
59 struct pnfs_deviceid_node deviceid;
60 u32 stripe_count;
61 u8 *stripe_indices;
62 u32 ds_num;
63 struct nfs4_pnfs_ds *ds_list[1];
64};
65
66struct nfs4_filelayout_segment {
67 struct pnfs_layout_segment generic_hdr;
68 u32 stripe_type;
69 u32 commit_through_mds;
70 u32 stripe_unit;
71 u32 first_stripe_index;
72 u64 pattern_offset;
73 struct nfs4_file_layout_dsaddr *dsaddr; /* Point to GETDEVINFO data */
74 unsigned int num_fh;
75 struct nfs_fh **fh_array;
76};
77
78static inline struct nfs4_filelayout_segment *
79FILELAYOUT_LSEG(struct pnfs_layout_segment *lseg)
80{
81 return container_of(lseg,
82 struct nfs4_filelayout_segment,
83 generic_hdr);
84}
85
86extern void nfs4_fl_free_deviceid_callback(struct pnfs_deviceid_node *);
87extern void print_ds(struct nfs4_pnfs_ds *ds);
88extern void print_deviceid(struct nfs4_deviceid *dev_id);
89extern struct nfs4_file_layout_dsaddr *
90nfs4_fl_find_get_deviceid(struct nfs_client *, struct nfs4_deviceid *dev_id);
91struct nfs4_file_layout_dsaddr *
92get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id);
93
94#endif /* FS_NFS_NFS4FILELAYOUT_H */
diff --git a/fs/nfs/nfs4filelayoutdev.c b/fs/nfs/nfs4filelayoutdev.c
new file mode 100644
index 000000000000..51fe64ace55a
--- /dev/null
+++ b/fs/nfs/nfs4filelayoutdev.c
@@ -0,0 +1,448 @@
1/*
2 * Device operations for the pnfs nfs4 file layout driver.
3 *
4 * Copyright (c) 2002
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 * Garth Goodson <Garth.Goodson@netapp.com>
10 *
11 * Permission is granted to use, copy, create derivative works, and
12 * redistribute this software and such derivative works for any purpose,
13 * so long as the name of the University of Michigan is not used in
14 * any advertising or publicity pertaining to the use or distribution
15 * of this software without specific, written prior authorization. If
16 * the above copyright notice or any other identification of the
17 * University of Michigan is included in any copy of any portion of
18 * this software, then the disclaimer below must also be included.
19 *
20 * This software is provided as is, without representation or warranty
21 * of any kind either express or implied, including without limitation
22 * the implied warranties of merchantability, fitness for a particular
23 * purpose, or noninfringement. The Regents of the University of
24 * Michigan shall not be liable for any damages, including special,
25 * indirect, incidental, or consequential damages, with respect to any
26 * claim arising out of or in connection with the use of the software,
27 * even if it has been or is hereafter advised of the possibility of
28 * such damages.
29 */
30
31#include <linux/nfs_fs.h>
32#include <linux/vmalloc.h>
33
34#include "internal.h"
35#include "nfs4filelayout.h"
36
37#define NFSDBG_FACILITY NFSDBG_PNFS_LD
38
39/*
40 * Data server cache
41 *
42 * Data servers can be mapped to different device ids.
43 * nfs4_pnfs_ds reference counting
44 * - set to 1 on allocation
45 * - incremented when a device id maps a data server already in the cache.
46 * - decremented when deviceid is removed from the cache.
47 */
48DEFINE_SPINLOCK(nfs4_ds_cache_lock);
49static LIST_HEAD(nfs4_data_server_cache);
50
51/* Debug routines */
52void
53print_ds(struct nfs4_pnfs_ds *ds)
54{
55 if (ds == NULL) {
56 printk("%s NULL device\n", __func__);
57 return;
58 }
59 printk(" ip_addr %x port %hu\n"
60 " ref count %d\n"
61 " client %p\n"
62 " cl_exchange_flags %x\n",
63 ntohl(ds->ds_ip_addr), ntohs(ds->ds_port),
64 atomic_read(&ds->ds_count), ds->ds_clp,
65 ds->ds_clp ? ds->ds_clp->cl_exchange_flags : 0);
66}
67
68void
69print_ds_list(struct nfs4_file_layout_dsaddr *dsaddr)
70{
71 int i;
72
73 ifdebug(FACILITY) {
74 printk("%s dsaddr->ds_num %d\n", __func__,
75 dsaddr->ds_num);
76 for (i = 0; i < dsaddr->ds_num; i++)
77 print_ds(dsaddr->ds_list[i]);
78 }
79}
80
81void print_deviceid(struct nfs4_deviceid *id)
82{
83 u32 *p = (u32 *)id;
84
85 dprintk("%s: device id= [%x%x%x%x]\n", __func__,
86 p[0], p[1], p[2], p[3]);
87}
88
89/* nfs4_ds_cache_lock is held */
90static struct nfs4_pnfs_ds *
91_data_server_lookup_locked(u32 ip_addr, u32 port)
92{
93 struct nfs4_pnfs_ds *ds;
94
95 dprintk("_data_server_lookup: ip_addr=%x port=%hu\n",
96 ntohl(ip_addr), ntohs(port));
97
98 list_for_each_entry(ds, &nfs4_data_server_cache, ds_node) {
99 if (ds->ds_ip_addr == ip_addr &&
100 ds->ds_port == port) {
101 return ds;
102 }
103 }
104 return NULL;
105}
106
107static void
108destroy_ds(struct nfs4_pnfs_ds *ds)
109{
110 dprintk("--> %s\n", __func__);
111 ifdebug(FACILITY)
112 print_ds(ds);
113
114 if (ds->ds_clp)
115 nfs_put_client(ds->ds_clp);
116 kfree(ds);
117}
118
119static void
120nfs4_fl_free_deviceid(struct nfs4_file_layout_dsaddr *dsaddr)
121{
122 struct nfs4_pnfs_ds *ds;
123 int i;
124
125 print_deviceid(&dsaddr->deviceid.de_id);
126
127 for (i = 0; i < dsaddr->ds_num; i++) {
128 ds = dsaddr->ds_list[i];
129 if (ds != NULL) {
130 if (atomic_dec_and_lock(&ds->ds_count,
131 &nfs4_ds_cache_lock)) {
132 list_del_init(&ds->ds_node);
133 spin_unlock(&nfs4_ds_cache_lock);
134 destroy_ds(ds);
135 }
136 }
137 }
138 kfree(dsaddr->stripe_indices);
139 kfree(dsaddr);
140}
141
142void
143nfs4_fl_free_deviceid_callback(struct pnfs_deviceid_node *device)
144{
145 struct nfs4_file_layout_dsaddr *dsaddr =
146 container_of(device, struct nfs4_file_layout_dsaddr, deviceid);
147
148 nfs4_fl_free_deviceid(dsaddr);
149}
150
151static struct nfs4_pnfs_ds *
152nfs4_pnfs_ds_add(struct inode *inode, u32 ip_addr, u32 port)
153{
154 struct nfs4_pnfs_ds *tmp_ds, *ds;
155
156 ds = kzalloc(sizeof(*tmp_ds), GFP_KERNEL);
157 if (!ds)
158 goto out;
159
160 spin_lock(&nfs4_ds_cache_lock);
161 tmp_ds = _data_server_lookup_locked(ip_addr, port);
162 if (tmp_ds == NULL) {
163 ds->ds_ip_addr = ip_addr;
164 ds->ds_port = port;
165 atomic_set(&ds->ds_count, 1);
166 INIT_LIST_HEAD(&ds->ds_node);
167 ds->ds_clp = NULL;
168 list_add(&ds->ds_node, &nfs4_data_server_cache);
169 dprintk("%s add new data server ip 0x%x\n", __func__,
170 ds->ds_ip_addr);
171 } else {
172 kfree(ds);
173 atomic_inc(&tmp_ds->ds_count);
174 dprintk("%s data server found ip 0x%x, inc'ed ds_count to %d\n",
175 __func__, tmp_ds->ds_ip_addr,
176 atomic_read(&tmp_ds->ds_count));
177 ds = tmp_ds;
178 }
179 spin_unlock(&nfs4_ds_cache_lock);
180out:
181 return ds;
182}
183
184/*
185 * Currently only support ipv4, and one multi-path address.
186 */
187static struct nfs4_pnfs_ds *
188decode_and_add_ds(__be32 **pp, struct inode *inode)
189{
190 struct nfs4_pnfs_ds *ds = NULL;
191 char *buf;
192 const char *ipend, *pstr;
193 u32 ip_addr, port;
194 int nlen, rlen, i;
195 int tmp[2];
196 __be32 *r_netid, *r_addr, *p = *pp;
197
198 /* r_netid */
199 nlen = be32_to_cpup(p++);
200 r_netid = p;
201 p += XDR_QUADLEN(nlen);
202
203 /* r_addr */
204 rlen = be32_to_cpup(p++);
205 r_addr = p;
206 p += XDR_QUADLEN(rlen);
207 *pp = p;
208
209 /* Check that netid is "tcp" */
210 if (nlen != 3 || memcmp((char *)r_netid, "tcp", 3)) {
211 dprintk("%s: ERROR: non ipv4 TCP r_netid\n", __func__);
212 goto out_err;
213 }
214
215 /* ipv6 length plus port is legal */
216 if (rlen > INET6_ADDRSTRLEN + 8) {
217 dprintk("%s Invalid address, length %d\n", __func__,
218 rlen);
219 goto out_err;
220 }
221 buf = kmalloc(rlen + 1, GFP_KERNEL);
222 buf[rlen] = '\0';
223 memcpy(buf, r_addr, rlen);
224
225 /* replace the port dots with dashes for the in4_pton() delimiter*/
226 for (i = 0; i < 2; i++) {
227 char *res = strrchr(buf, '.');
228 *res = '-';
229 }
230
231 /* Currently only support ipv4 address */
232 if (in4_pton(buf, rlen, (u8 *)&ip_addr, '-', &ipend) == 0) {
233 dprintk("%s: Only ipv4 addresses supported\n", __func__);
234 goto out_free;
235 }
236
237 /* port */
238 pstr = ipend;
239 sscanf(pstr, "-%d-%d", &tmp[0], &tmp[1]);
240 port = htons((tmp[0] << 8) | (tmp[1]));
241
242 ds = nfs4_pnfs_ds_add(inode, ip_addr, port);
243 dprintk("%s Decoded address and port %s\n", __func__, buf);
244out_free:
245 kfree(buf);
246out_err:
247 return ds;
248}
249
250/* Decode opaque device data and return the result */
251static struct nfs4_file_layout_dsaddr*
252decode_device(struct inode *ino, struct pnfs_device *pdev)
253{
254 int i, dummy;
255 u32 cnt, num;
256 u8 *indexp;
257 __be32 *p = (__be32 *)pdev->area, *indicesp;
258 struct nfs4_file_layout_dsaddr *dsaddr;
259
260 /* Get the stripe count (number of stripe index) */
261 cnt = be32_to_cpup(p++);
262 dprintk("%s stripe count %d\n", __func__, cnt);
263 if (cnt > NFS4_PNFS_MAX_STRIPE_CNT) {
264 printk(KERN_WARNING "%s: stripe count %d greater than "
265 "supported maximum %d\n", __func__,
266 cnt, NFS4_PNFS_MAX_STRIPE_CNT);
267 goto out_err;
268 }
269
270 /* Check the multipath list count */
271 indicesp = p;
272 p += XDR_QUADLEN(cnt << 2);
273 num = be32_to_cpup(p++);
274 dprintk("%s ds_num %u\n", __func__, num);
275 if (num > NFS4_PNFS_MAX_MULTI_CNT) {
276 printk(KERN_WARNING "%s: multipath count %d greater than "
277 "supported maximum %d\n", __func__,
278 num, NFS4_PNFS_MAX_MULTI_CNT);
279 goto out_err;
280 }
281 dsaddr = kzalloc(sizeof(*dsaddr) +
282 (sizeof(struct nfs4_pnfs_ds *) * (num - 1)),
283 GFP_KERNEL);
284 if (!dsaddr)
285 goto out_err;
286
287 dsaddr->stripe_indices = kzalloc(sizeof(u8) * cnt, GFP_KERNEL);
288 if (!dsaddr->stripe_indices)
289 goto out_err_free;
290
291 dsaddr->stripe_count = cnt;
292 dsaddr->ds_num = num;
293
294 memcpy(&dsaddr->deviceid.de_id, &pdev->dev_id, sizeof(pdev->dev_id));
295
296 /* Go back an read stripe indices */
297 p = indicesp;
298 indexp = &dsaddr->stripe_indices[0];
299 for (i = 0; i < dsaddr->stripe_count; i++) {
300 *indexp = be32_to_cpup(p++);
301 if (*indexp >= num)
302 goto out_err_free;
303 indexp++;
304 }
305 /* Skip already read multipath list count */
306 p++;
307
308 for (i = 0; i < dsaddr->ds_num; i++) {
309 int j;
310
311 dummy = be32_to_cpup(p++); /* multipath count */
312 if (dummy > 1) {
313 printk(KERN_WARNING
314 "%s: Multipath count %d not supported, "
315 "skipping all greater than 1\n", __func__,
316 dummy);
317 }
318 for (j = 0; j < dummy; j++) {
319 if (j == 0) {
320 dsaddr->ds_list[i] = decode_and_add_ds(&p, ino);
321 if (dsaddr->ds_list[i] == NULL)
322 goto out_err_free;
323 } else {
324 u32 len;
325 /* skip extra multipath */
326 len = be32_to_cpup(p++);
327 p += XDR_QUADLEN(len);
328 len = be32_to_cpup(p++);
329 p += XDR_QUADLEN(len);
330 continue;
331 }
332 }
333 }
334 return dsaddr;
335
336out_err_free:
337 nfs4_fl_free_deviceid(dsaddr);
338out_err:
339 dprintk("%s ERROR: returning NULL\n", __func__);
340 return NULL;
341}
342
343/*
344 * Decode the opaque device specified in 'dev'
345 * and add it to the list of available devices.
346 * If the deviceid is already cached, nfs4_add_deviceid will return
347 * a pointer to the cached struct and throw away the new.
348 */
349static struct nfs4_file_layout_dsaddr*
350decode_and_add_device(struct inode *inode, struct pnfs_device *dev)
351{
352 struct nfs4_file_layout_dsaddr *dsaddr;
353 struct pnfs_deviceid_node *d;
354
355 dsaddr = decode_device(inode, dev);
356 if (!dsaddr) {
357 printk(KERN_WARNING "%s: Could not decode or add device\n",
358 __func__);
359 return NULL;
360 }
361
362 d = pnfs_add_deviceid(NFS_SERVER(inode)->nfs_client->cl_devid_cache,
363 &dsaddr->deviceid);
364
365 return container_of(d, struct nfs4_file_layout_dsaddr, deviceid);
366}
367
368/*
369 * Retrieve the information for dev_id, add it to the list
370 * of available devices, and return it.
371 */
372struct nfs4_file_layout_dsaddr *
373get_device_info(struct inode *inode, struct nfs4_deviceid *dev_id)
374{
375 struct pnfs_device *pdev = NULL;
376 u32 max_resp_sz;
377 int max_pages;
378 struct page **pages = NULL;
379 struct nfs4_file_layout_dsaddr *dsaddr = NULL;
380 int rc, i;
381 struct nfs_server *server = NFS_SERVER(inode);
382
383 /*
384 * Use the session max response size as the basis for setting
385 * GETDEVICEINFO's maxcount
386 */
387 max_resp_sz = server->nfs_client->cl_session->fc_attrs.max_resp_sz;
388 max_pages = max_resp_sz >> PAGE_SHIFT;
389 dprintk("%s inode %p max_resp_sz %u max_pages %d\n",
390 __func__, inode, max_resp_sz, max_pages);
391
392 pdev = kzalloc(sizeof(struct pnfs_device), GFP_KERNEL);
393 if (pdev == NULL)
394 return NULL;
395
396 pages = kzalloc(max_pages * sizeof(struct page *), GFP_KERNEL);
397 if (pages == NULL) {
398 kfree(pdev);
399 return NULL;
400 }
401 for (i = 0; i < max_pages; i++) {
402 pages[i] = alloc_page(GFP_KERNEL);
403 if (!pages[i])
404 goto out_free;
405 }
406
407 /* set pdev->area */
408 pdev->area = vmap(pages, max_pages, VM_MAP, PAGE_KERNEL);
409 if (!pdev->area)
410 goto out_free;
411
412 memcpy(&pdev->dev_id, dev_id, sizeof(*dev_id));
413 pdev->layout_type = LAYOUT_NFSV4_1_FILES;
414 pdev->pages = pages;
415 pdev->pgbase = 0;
416 pdev->pglen = PAGE_SIZE * max_pages;
417 pdev->mincount = 0;
418
419 rc = nfs4_proc_getdeviceinfo(server, pdev);
420 dprintk("%s getdevice info returns %d\n", __func__, rc);
421 if (rc)
422 goto out_free;
423
424 /*
425 * Found new device, need to decode it and then add it to the
426 * list of known devices for this mountpoint.
427 */
428 dsaddr = decode_and_add_device(inode, pdev);
429out_free:
430 if (pdev->area != NULL)
431 vunmap(pdev->area);
432 for (i = 0; i < max_pages; i++)
433 __free_page(pages[i]);
434 kfree(pages);
435 kfree(pdev);
436 dprintk("<-- %s dsaddr %p\n", __func__, dsaddr);
437 return dsaddr;
438}
439
440struct nfs4_file_layout_dsaddr *
441nfs4_fl_find_get_deviceid(struct nfs_client *clp, struct nfs4_deviceid *id)
442{
443 struct pnfs_deviceid_node *d;
444
445 d = pnfs_find_get_deviceid(clp->cl_devid_cache, id);
446 return (d == NULL) ? NULL :
447 container_of(d, struct nfs4_file_layout_dsaddr, deviceid);
448}
diff --git a/fs/nfs/nfs4proc.c b/fs/nfs/nfs4proc.c
index e87fe612ca18..32c8758c99fd 100644
--- a/fs/nfs/nfs4proc.c
+++ b/fs/nfs/nfs4proc.c
@@ -55,6 +55,7 @@
55#include "internal.h" 55#include "internal.h"
56#include "iostat.h" 56#include "iostat.h"
57#include "callback.h" 57#include "callback.h"
58#include "pnfs.h"
58 59
59#define NFSDBG_FACILITY NFSDBG_PROC 60#define NFSDBG_FACILITY NFSDBG_PROC
60 61
@@ -130,6 +131,7 @@ const u32 nfs4_fsinfo_bitmap[2] = { FATTR4_WORD0_MAXFILESIZE
130 | FATTR4_WORD0_MAXWRITE 131 | FATTR4_WORD0_MAXWRITE
131 | FATTR4_WORD0_LEASE_TIME, 132 | FATTR4_WORD0_LEASE_TIME,
132 FATTR4_WORD1_TIME_DELTA 133 FATTR4_WORD1_TIME_DELTA
134 | FATTR4_WORD1_FS_LAYOUT_TYPES
133}; 135};
134 136
135const u32 nfs4_fs_locations_bitmap[2] = { 137const u32 nfs4_fs_locations_bitmap[2] = {
@@ -4840,49 +4842,56 @@ static void nfs4_init_channel_attrs(struct nfs41_create_session_args *args)
4840 args->bc_attrs.max_reqs); 4842 args->bc_attrs.max_reqs);
4841} 4843}
4842 4844
4843static int _verify_channel_attr(char *chan, char *attr_name, u32 sent, u32 rcvd) 4845static int nfs4_verify_fore_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
4844{ 4846{
4845 if (rcvd <= sent) 4847 struct nfs4_channel_attrs *sent = &args->fc_attrs;
4846 return 0; 4848 struct nfs4_channel_attrs *rcvd = &session->fc_attrs;
4847 printk(KERN_WARNING "%s: Session INVALID: %s channel %s increased. " 4849
4848 "sent=%u rcvd=%u\n", __func__, chan, attr_name, sent, rcvd); 4850 if (rcvd->headerpadsz > sent->headerpadsz)
4849 return -EINVAL; 4851 return -EINVAL;
4852 if (rcvd->max_resp_sz > sent->max_resp_sz)
4853 return -EINVAL;
4854 /*
4855 * Our requested max_ops is the minimum we need; we're not
4856 * prepared to break up compounds into smaller pieces than that.
4857 * So, no point even trying to continue if the server won't
4858 * cooperate:
4859 */
4860 if (rcvd->max_ops < sent->max_ops)
4861 return -EINVAL;
4862 if (rcvd->max_reqs == 0)
4863 return -EINVAL;
4864 return 0;
4850} 4865}
4851 4866
4852#define _verify_fore_channel_attr(_name_) \ 4867static int nfs4_verify_back_channel_attrs(struct nfs41_create_session_args *args, struct nfs4_session *session)
4853 _verify_channel_attr("fore", #_name_, \ 4868{
4854 args->fc_attrs._name_, \ 4869 struct nfs4_channel_attrs *sent = &args->bc_attrs;
4855 session->fc_attrs._name_) 4870 struct nfs4_channel_attrs *rcvd = &session->bc_attrs;
4856 4871
4857#define _verify_back_channel_attr(_name_) \ 4872 if (rcvd->max_rqst_sz > sent->max_rqst_sz)
4858 _verify_channel_attr("back", #_name_, \ 4873 return -EINVAL;
4859 args->bc_attrs._name_, \ 4874 if (rcvd->max_resp_sz < sent->max_resp_sz)
4860 session->bc_attrs._name_) 4875 return -EINVAL;
4876 if (rcvd->max_resp_sz_cached > sent->max_resp_sz_cached)
4877 return -EINVAL;
4878 /* These would render the backchannel useless: */
4879 if (rcvd->max_ops == 0)
4880 return -EINVAL;
4881 if (rcvd->max_reqs == 0)
4882 return -EINVAL;
4883 return 0;
4884}
4861 4885
4862/*
4863 * The server is not allowed to increase the fore channel header pad size,
4864 * maximum response size, or maximum number of operations.
4865 *
4866 * The back channel attributes are only negotiatied down: We send what the
4867 * (back channel) server insists upon.
4868 */
4869static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args, 4886static int nfs4_verify_channel_attrs(struct nfs41_create_session_args *args,
4870 struct nfs4_session *session) 4887 struct nfs4_session *session)
4871{ 4888{
4872 int ret = 0; 4889 int ret;
4873
4874 ret |= _verify_fore_channel_attr(headerpadsz);
4875 ret |= _verify_fore_channel_attr(max_resp_sz);
4876 ret |= _verify_fore_channel_attr(max_ops);
4877
4878 ret |= _verify_back_channel_attr(headerpadsz);
4879 ret |= _verify_back_channel_attr(max_rqst_sz);
4880 ret |= _verify_back_channel_attr(max_resp_sz);
4881 ret |= _verify_back_channel_attr(max_resp_sz_cached);
4882 ret |= _verify_back_channel_attr(max_ops);
4883 ret |= _verify_back_channel_attr(max_reqs);
4884 4890
4885 return ret; 4891 ret = nfs4_verify_fore_channel_attrs(args, session);
4892 if (ret)
4893 return ret;
4894 return nfs4_verify_back_channel_attrs(args, session);
4886} 4895}
4887 4896
4888static int _nfs4_proc_create_session(struct nfs_client *clp) 4897static int _nfs4_proc_create_session(struct nfs_client *clp)
@@ -5255,6 +5264,147 @@ out:
5255 dprintk("<-- %s status=%d\n", __func__, status); 5264 dprintk("<-- %s status=%d\n", __func__, status);
5256 return status; 5265 return status;
5257} 5266}
5267
5268static void
5269nfs4_layoutget_prepare(struct rpc_task *task, void *calldata)
5270{
5271 struct nfs4_layoutget *lgp = calldata;
5272 struct inode *ino = lgp->args.inode;
5273 struct nfs_server *server = NFS_SERVER(ino);
5274
5275 dprintk("--> %s\n", __func__);
5276 if (nfs4_setup_sequence(server, &lgp->args.seq_args,
5277 &lgp->res.seq_res, 0, task))
5278 return;
5279 rpc_call_start(task);
5280}
5281
5282static void nfs4_layoutget_done(struct rpc_task *task, void *calldata)
5283{
5284 struct nfs4_layoutget *lgp = calldata;
5285 struct nfs_server *server = NFS_SERVER(lgp->args.inode);
5286
5287 dprintk("--> %s\n", __func__);
5288
5289 if (!nfs4_sequence_done(task, &lgp->res.seq_res))
5290 return;
5291
5292 switch (task->tk_status) {
5293 case 0:
5294 break;
5295 case -NFS4ERR_LAYOUTTRYLATER:
5296 case -NFS4ERR_RECALLCONFLICT:
5297 task->tk_status = -NFS4ERR_DELAY;
5298 /* Fall through */
5299 default:
5300 if (nfs4_async_handle_error(task, server, NULL) == -EAGAIN) {
5301 rpc_restart_call_prepare(task);
5302 return;
5303 }
5304 }
5305 lgp->status = task->tk_status;
5306 dprintk("<-- %s\n", __func__);
5307}
5308
5309static void nfs4_layoutget_release(void *calldata)
5310{
5311 struct nfs4_layoutget *lgp = calldata;
5312
5313 dprintk("--> %s\n", __func__);
5314 put_layout_hdr(lgp->args.inode);
5315 if (lgp->res.layout.buf != NULL)
5316 free_page((unsigned long) lgp->res.layout.buf);
5317 put_nfs_open_context(lgp->args.ctx);
5318 kfree(calldata);
5319 dprintk("<-- %s\n", __func__);
5320}
5321
5322static const struct rpc_call_ops nfs4_layoutget_call_ops = {
5323 .rpc_call_prepare = nfs4_layoutget_prepare,
5324 .rpc_call_done = nfs4_layoutget_done,
5325 .rpc_release = nfs4_layoutget_release,
5326};
5327
5328int nfs4_proc_layoutget(struct nfs4_layoutget *lgp)
5329{
5330 struct nfs_server *server = NFS_SERVER(lgp->args.inode);
5331 struct rpc_task *task;
5332 struct rpc_message msg = {
5333 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_LAYOUTGET],
5334 .rpc_argp = &lgp->args,
5335 .rpc_resp = &lgp->res,
5336 };
5337 struct rpc_task_setup task_setup_data = {
5338 .rpc_client = server->client,
5339 .rpc_message = &msg,
5340 .callback_ops = &nfs4_layoutget_call_ops,
5341 .callback_data = lgp,
5342 .flags = RPC_TASK_ASYNC,
5343 };
5344 int status = 0;
5345
5346 dprintk("--> %s\n", __func__);
5347
5348 lgp->res.layout.buf = (void *)__get_free_page(GFP_NOFS);
5349 if (lgp->res.layout.buf == NULL) {
5350 nfs4_layoutget_release(lgp);
5351 return -ENOMEM;
5352 }
5353
5354 lgp->res.seq_res.sr_slot = NULL;
5355 task = rpc_run_task(&task_setup_data);
5356 if (IS_ERR(task))
5357 return PTR_ERR(task);
5358 status = nfs4_wait_for_completion_rpc_task(task);
5359 if (status != 0)
5360 goto out;
5361 status = lgp->status;
5362 if (status != 0)
5363 goto out;
5364 status = pnfs_layout_process(lgp);
5365out:
5366 rpc_put_task(task);
5367 dprintk("<-- %s status=%d\n", __func__, status);
5368 return status;
5369}
5370
5371static int
5372_nfs4_proc_getdeviceinfo(struct nfs_server *server, struct pnfs_device *pdev)
5373{
5374 struct nfs4_getdeviceinfo_args args = {
5375 .pdev = pdev,
5376 };
5377 struct nfs4_getdeviceinfo_res res = {
5378 .pdev = pdev,
5379 };
5380 struct rpc_message msg = {
5381 .rpc_proc = &nfs4_procedures[NFSPROC4_CLNT_GETDEVICEINFO],
5382 .rpc_argp = &args,
5383 .rpc_resp = &res,
5384 };
5385 int status;
5386
5387 dprintk("--> %s\n", __func__);
5388 status = nfs4_call_sync(server, &msg, &args, &res, 0);
5389 dprintk("<-- %s status=%d\n", __func__, status);
5390
5391 return status;
5392}
5393
5394int nfs4_proc_getdeviceinfo(struct nfs_server *server, struct pnfs_device *pdev)
5395{
5396 struct nfs4_exception exception = { };
5397 int err;
5398
5399 do {
5400 err = nfs4_handle_exception(server,
5401 _nfs4_proc_getdeviceinfo(server, pdev),
5402 &exception);
5403 } while (exception.retry);
5404 return err;
5405}
5406EXPORT_SYMBOL_GPL(nfs4_proc_getdeviceinfo);
5407
5258#endif /* CONFIG_NFS_V4_1 */ 5408#endif /* CONFIG_NFS_V4_1 */
5259 5409
5260struct nfs4_state_recovery_ops nfs40_reboot_recovery_ops = { 5410struct nfs4_state_recovery_ops nfs40_reboot_recovery_ops = {
diff --git a/fs/nfs/nfs4state.c b/fs/nfs/nfs4state.c
index aa0b02a610c4..f575a3126737 100644
--- a/fs/nfs/nfs4state.c
+++ b/fs/nfs/nfs4state.c
@@ -54,6 +54,7 @@
54#include "callback.h" 54#include "callback.h"
55#include "delegation.h" 55#include "delegation.h"
56#include "internal.h" 56#include "internal.h"
57#include "pnfs.h"
57 58
58#define OPENOWNER_POOL_SIZE 8 59#define OPENOWNER_POOL_SIZE 8
59 60
@@ -1475,6 +1476,7 @@ static void nfs4_state_manager(struct nfs_client *clp)
1475 } 1476 }
1476 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state); 1477 clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state);
1477 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state); 1478 set_bit(NFS4CLNT_RECLAIM_REBOOT, &clp->cl_state);
1479 pnfs_destroy_all_layouts(clp);
1478 } 1480 }
1479 1481
1480 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) { 1482 if (test_and_clear_bit(NFS4CLNT_CHECK_LEASE, &clp->cl_state)) {
diff --git a/fs/nfs/nfs4xdr.c b/fs/nfs/nfs4xdr.c
index bd2101d918c8..f313c4cce7e4 100644
--- a/fs/nfs/nfs4xdr.c
+++ b/fs/nfs/nfs4xdr.c
@@ -52,6 +52,7 @@
52#include <linux/nfs_idmap.h> 52#include <linux/nfs_idmap.h>
53#include "nfs4_fs.h" 53#include "nfs4_fs.h"
54#include "internal.h" 54#include "internal.h"
55#include "pnfs.h"
55 56
56#define NFSDBG_FACILITY NFSDBG_XDR 57#define NFSDBG_FACILITY NFSDBG_XDR
57 58
@@ -310,6 +311,19 @@ static int nfs4_stat_to_errno(int);
310 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + 5) 311 XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + 5)
311#define encode_reclaim_complete_maxsz (op_encode_hdr_maxsz + 4) 312#define encode_reclaim_complete_maxsz (op_encode_hdr_maxsz + 4)
312#define decode_reclaim_complete_maxsz (op_decode_hdr_maxsz + 4) 313#define decode_reclaim_complete_maxsz (op_decode_hdr_maxsz + 4)
314#define encode_getdeviceinfo_maxsz (op_encode_hdr_maxsz + 4 + \
315 XDR_QUADLEN(NFS4_DEVICEID4_SIZE))
316#define decode_getdeviceinfo_maxsz (op_decode_hdr_maxsz + \
317 1 /* layout type */ + \
318 1 /* opaque devaddr4 length */ + \
319 /* devaddr4 payload is read into page */ \
320 1 /* notification bitmap length */ + \
321 1 /* notification bitmap */)
322#define encode_layoutget_maxsz (op_encode_hdr_maxsz + 10 + \
323 encode_stateid_maxsz)
324#define decode_layoutget_maxsz (op_decode_hdr_maxsz + 8 + \
325 decode_stateid_maxsz + \
326 XDR_QUADLEN(PNFS_LAYOUT_MAXSIZE))
313#else /* CONFIG_NFS_V4_1 */ 327#else /* CONFIG_NFS_V4_1 */
314#define encode_sequence_maxsz 0 328#define encode_sequence_maxsz 0
315#define decode_sequence_maxsz 0 329#define decode_sequence_maxsz 0
@@ -699,6 +713,20 @@ static int nfs4_stat_to_errno(int);
699#define NFS4_dec_reclaim_complete_sz (compound_decode_hdr_maxsz + \ 713#define NFS4_dec_reclaim_complete_sz (compound_decode_hdr_maxsz + \
700 decode_sequence_maxsz + \ 714 decode_sequence_maxsz + \
701 decode_reclaim_complete_maxsz) 715 decode_reclaim_complete_maxsz)
716#define NFS4_enc_getdeviceinfo_sz (compound_encode_hdr_maxsz + \
717 encode_sequence_maxsz +\
718 encode_getdeviceinfo_maxsz)
719#define NFS4_dec_getdeviceinfo_sz (compound_decode_hdr_maxsz + \
720 decode_sequence_maxsz + \
721 decode_getdeviceinfo_maxsz)
722#define NFS4_enc_layoutget_sz (compound_encode_hdr_maxsz + \
723 encode_sequence_maxsz + \
724 encode_putfh_maxsz + \
725 encode_layoutget_maxsz)
726#define NFS4_dec_layoutget_sz (compound_decode_hdr_maxsz + \
727 decode_sequence_maxsz + \
728 decode_putfh_maxsz + \
729 decode_layoutget_maxsz)
702 730
703const u32 nfs41_maxwrite_overhead = ((RPC_MAX_HEADER_WITH_AUTH + 731const u32 nfs41_maxwrite_overhead = ((RPC_MAX_HEADER_WITH_AUTH +
704 compound_encode_hdr_maxsz + 732 compound_encode_hdr_maxsz +
@@ -1737,6 +1765,58 @@ static void encode_sequence(struct xdr_stream *xdr,
1737#endif /* CONFIG_NFS_V4_1 */ 1765#endif /* CONFIG_NFS_V4_1 */
1738} 1766}
1739 1767
1768#ifdef CONFIG_NFS_V4_1
1769static void
1770encode_getdeviceinfo(struct xdr_stream *xdr,
1771 const struct nfs4_getdeviceinfo_args *args,
1772 struct compound_hdr *hdr)
1773{
1774 __be32 *p;
1775
1776 p = reserve_space(xdr, 16 + NFS4_DEVICEID4_SIZE);
1777 *p++ = cpu_to_be32(OP_GETDEVICEINFO);
1778 p = xdr_encode_opaque_fixed(p, args->pdev->dev_id.data,
1779 NFS4_DEVICEID4_SIZE);
1780 *p++ = cpu_to_be32(args->pdev->layout_type);
1781 *p++ = cpu_to_be32(args->pdev->pglen); /* gdia_maxcount */
1782 *p++ = cpu_to_be32(0); /* bitmap length 0 */
1783 hdr->nops++;
1784 hdr->replen += decode_getdeviceinfo_maxsz;
1785}
1786
1787static void
1788encode_layoutget(struct xdr_stream *xdr,
1789 const struct nfs4_layoutget_args *args,
1790 struct compound_hdr *hdr)
1791{
1792 nfs4_stateid stateid;
1793 __be32 *p;
1794
1795 p = reserve_space(xdr, 44 + NFS4_STATEID_SIZE);
1796 *p++ = cpu_to_be32(OP_LAYOUTGET);
1797 *p++ = cpu_to_be32(0); /* Signal layout available */
1798 *p++ = cpu_to_be32(args->type);
1799 *p++ = cpu_to_be32(args->range.iomode);
1800 p = xdr_encode_hyper(p, args->range.offset);
1801 p = xdr_encode_hyper(p, args->range.length);
1802 p = xdr_encode_hyper(p, args->minlength);
1803 pnfs_get_layout_stateid(&stateid, NFS_I(args->inode)->layout,
1804 args->ctx->state);
1805 p = xdr_encode_opaque_fixed(p, &stateid.data, NFS4_STATEID_SIZE);
1806 *p = cpu_to_be32(args->maxcount);
1807
1808 dprintk("%s: 1st type:0x%x iomode:%d off:%lu len:%lu mc:%d\n",
1809 __func__,
1810 args->type,
1811 args->range.iomode,
1812 (unsigned long)args->range.offset,
1813 (unsigned long)args->range.length,
1814 args->maxcount);
1815 hdr->nops++;
1816 hdr->replen += decode_layoutget_maxsz;
1817}
1818#endif /* CONFIG_NFS_V4_1 */
1819
1740/* 1820/*
1741 * END OF "GENERIC" ENCODE ROUTINES. 1821 * END OF "GENERIC" ENCODE ROUTINES.
1742 */ 1822 */
@@ -2554,6 +2634,51 @@ static int nfs4_xdr_enc_reclaim_complete(struct rpc_rqst *req, uint32_t *p,
2554 return 0; 2634 return 0;
2555} 2635}
2556 2636
2637/*
2638 * Encode GETDEVICEINFO request
2639 */
2640static int nfs4_xdr_enc_getdeviceinfo(struct rpc_rqst *req, uint32_t *p,
2641 struct nfs4_getdeviceinfo_args *args)
2642{
2643 struct xdr_stream xdr;
2644 struct compound_hdr hdr = {
2645 .minorversion = nfs4_xdr_minorversion(&args->seq_args),
2646 };
2647
2648 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
2649 encode_compound_hdr(&xdr, req, &hdr);
2650 encode_sequence(&xdr, &args->seq_args, &hdr);
2651 encode_getdeviceinfo(&xdr, args, &hdr);
2652
2653 /* set up reply kvec. Subtract notification bitmap max size (2)
2654 * so that notification bitmap is put in xdr_buf tail */
2655 xdr_inline_pages(&req->rq_rcv_buf, (hdr.replen - 2) << 2,
2656 args->pdev->pages, args->pdev->pgbase,
2657 args->pdev->pglen);
2658
2659 encode_nops(&hdr);
2660 return 0;
2661}
2662
2663/*
2664 * Encode LAYOUTGET request
2665 */
2666static int nfs4_xdr_enc_layoutget(struct rpc_rqst *req, uint32_t *p,
2667 struct nfs4_layoutget_args *args)
2668{
2669 struct xdr_stream xdr;
2670 struct compound_hdr hdr = {
2671 .minorversion = nfs4_xdr_minorversion(&args->seq_args),
2672 };
2673
2674 xdr_init_encode(&xdr, &req->rq_snd_buf, p);
2675 encode_compound_hdr(&xdr, req, &hdr);
2676 encode_sequence(&xdr, &args->seq_args, &hdr);
2677 encode_putfh(&xdr, NFS_FH(args->inode), &hdr);
2678 encode_layoutget(&xdr, args, &hdr);
2679 encode_nops(&hdr);
2680 return 0;
2681}
2557#endif /* CONFIG_NFS_V4_1 */ 2682#endif /* CONFIG_NFS_V4_1 */
2558 2683
2559static void print_overflow_msg(const char *func, const struct xdr_stream *xdr) 2684static void print_overflow_msg(const char *func, const struct xdr_stream *xdr)
@@ -3978,6 +4103,61 @@ static int decode_getfattr(struct xdr_stream *xdr, struct nfs_fattr *fattr,
3978 return decode_getfattr_generic(xdr, fattr, NULL, server, may_sleep); 4103 return decode_getfattr_generic(xdr, fattr, NULL, server, may_sleep);
3979} 4104}
3980 4105
4106/*
4107 * Decode potentially multiple layout types. Currently we only support
4108 * one layout driver per file system.
4109 */
4110static int decode_first_pnfs_layout_type(struct xdr_stream *xdr,
4111 uint32_t *layouttype)
4112{
4113 uint32_t *p;
4114 int num;
4115
4116 p = xdr_inline_decode(xdr, 4);
4117 if (unlikely(!p))
4118 goto out_overflow;
4119 num = be32_to_cpup(p);
4120
4121 /* pNFS is not supported by the underlying file system */
4122 if (num == 0) {
4123 *layouttype = 0;
4124 return 0;
4125 }
4126 if (num > 1)
4127 printk(KERN_INFO "%s: Warning: Multiple pNFS layout drivers "
4128 "per filesystem not supported\n", __func__);
4129
4130 /* Decode and set first layout type, move xdr->p past unused types */
4131 p = xdr_inline_decode(xdr, num * 4);
4132 if (unlikely(!p))
4133 goto out_overflow;
4134 *layouttype = be32_to_cpup(p);
4135 return 0;
4136out_overflow:
4137 print_overflow_msg(__func__, xdr);
4138 return -EIO;
4139}
4140
4141/*
4142 * The type of file system exported.
4143 * Note we must ensure that layouttype is set in any non-error case.
4144 */
4145static int decode_attr_pnfstype(struct xdr_stream *xdr, uint32_t *bitmap,
4146 uint32_t *layouttype)
4147{
4148 int status = 0;
4149
4150 dprintk("%s: bitmap is %x\n", __func__, bitmap[1]);
4151 if (unlikely(bitmap[1] & (FATTR4_WORD1_FS_LAYOUT_TYPES - 1U)))
4152 return -EIO;
4153 if (bitmap[1] & FATTR4_WORD1_FS_LAYOUT_TYPES) {
4154 status = decode_first_pnfs_layout_type(xdr, layouttype);
4155 bitmap[1] &= ~FATTR4_WORD1_FS_LAYOUT_TYPES;
4156 } else
4157 *layouttype = 0;
4158 return status;
4159}
4160
3981static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo) 4161static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo)
3982{ 4162{
3983 __be32 *savep; 4163 __be32 *savep;
@@ -4006,6 +4186,9 @@ static int decode_fsinfo(struct xdr_stream *xdr, struct nfs_fsinfo *fsinfo)
4006 status = decode_attr_time_delta(xdr, bitmap, &fsinfo->time_delta); 4186 status = decode_attr_time_delta(xdr, bitmap, &fsinfo->time_delta);
4007 if (status != 0) 4187 if (status != 0)
4008 goto xdr_error; 4188 goto xdr_error;
4189 status = decode_attr_pnfstype(xdr, bitmap, &fsinfo->layouttype);
4190 if (status != 0)
4191 goto xdr_error;
4009 4192
4010 status = verify_attr_len(xdr, savep, attrlen); 4193 status = verify_attr_len(xdr, savep, attrlen);
4011xdr_error: 4194xdr_error:
@@ -4772,6 +4955,134 @@ out_overflow:
4772#endif /* CONFIG_NFS_V4_1 */ 4955#endif /* CONFIG_NFS_V4_1 */
4773} 4956}
4774 4957
4958#if defined(CONFIG_NFS_V4_1)
4959
4960static int decode_getdeviceinfo(struct xdr_stream *xdr,
4961 struct pnfs_device *pdev)
4962{
4963 __be32 *p;
4964 uint32_t len, type;
4965 int status;
4966
4967 status = decode_op_hdr(xdr, OP_GETDEVICEINFO);
4968 if (status) {
4969 if (status == -ETOOSMALL) {
4970 p = xdr_inline_decode(xdr, 4);
4971 if (unlikely(!p))
4972 goto out_overflow;
4973 pdev->mincount = be32_to_cpup(p);
4974 dprintk("%s: Min count too small. mincnt = %u\n",
4975 __func__, pdev->mincount);
4976 }
4977 return status;
4978 }
4979
4980 p = xdr_inline_decode(xdr, 8);
4981 if (unlikely(!p))
4982 goto out_overflow;
4983 type = be32_to_cpup(p++);
4984 if (type != pdev->layout_type) {
4985 dprintk("%s: layout mismatch req: %u pdev: %u\n",
4986 __func__, pdev->layout_type, type);
4987 return -EINVAL;
4988 }
4989 /*
4990 * Get the length of the opaque device_addr4. xdr_read_pages places
4991 * the opaque device_addr4 in the xdr_buf->pages (pnfs_device->pages)
4992 * and places the remaining xdr data in xdr_buf->tail
4993 */
4994 pdev->mincount = be32_to_cpup(p);
4995 xdr_read_pages(xdr, pdev->mincount); /* include space for the length */
4996
4997 /* Parse notification bitmap, verifying that it is zero. */
4998 p = xdr_inline_decode(xdr, 4);
4999 if (unlikely(!p))
5000 goto out_overflow;
5001 len = be32_to_cpup(p);
5002 if (len) {
5003 int i;
5004
5005 p = xdr_inline_decode(xdr, 4 * len);
5006 if (unlikely(!p))
5007 goto out_overflow;
5008 for (i = 0; i < len; i++, p++) {
5009 if (be32_to_cpup(p)) {
5010 dprintk("%s: notifications not supported\n",
5011 __func__);
5012 return -EIO;
5013 }
5014 }
5015 }
5016 return 0;
5017out_overflow:
5018 print_overflow_msg(__func__, xdr);
5019 return -EIO;
5020}
5021
5022static int decode_layoutget(struct xdr_stream *xdr, struct rpc_rqst *req,
5023 struct nfs4_layoutget_res *res)
5024{
5025 __be32 *p;
5026 int status;
5027 u32 layout_count;
5028
5029 status = decode_op_hdr(xdr, OP_LAYOUTGET);
5030 if (status)
5031 return status;
5032 p = xdr_inline_decode(xdr, 8 + NFS4_STATEID_SIZE);
5033 if (unlikely(!p))
5034 goto out_overflow;
5035 res->return_on_close = be32_to_cpup(p++);
5036 p = xdr_decode_opaque_fixed(p, res->stateid.data, NFS4_STATEID_SIZE);
5037 layout_count = be32_to_cpup(p);
5038 if (!layout_count) {
5039 dprintk("%s: server responded with empty layout array\n",
5040 __func__);
5041 return -EINVAL;
5042 }
5043
5044 p = xdr_inline_decode(xdr, 24);
5045 if (unlikely(!p))
5046 goto out_overflow;
5047 p = xdr_decode_hyper(p, &res->range.offset);
5048 p = xdr_decode_hyper(p, &res->range.length);
5049 res->range.iomode = be32_to_cpup(p++);
5050 res->type = be32_to_cpup(p++);
5051
5052 status = decode_opaque_inline(xdr, &res->layout.len, (char **)&p);
5053 if (unlikely(status))
5054 return status;
5055
5056 dprintk("%s roff:%lu rlen:%lu riomode:%d, lo_type:0x%x, lo.len:%d\n",
5057 __func__,
5058 (unsigned long)res->range.offset,
5059 (unsigned long)res->range.length,
5060 res->range.iomode,
5061 res->type,
5062 res->layout.len);
5063
5064 /* nfs4_proc_layoutget allocated a single page */
5065 if (res->layout.len > PAGE_SIZE)
5066 return -ENOMEM;
5067 memcpy(res->layout.buf, p, res->layout.len);
5068
5069 if (layout_count > 1) {
5070 /* We only handle a length one array at the moment. Any
5071 * further entries are just ignored. Note that this means
5072 * the client may see a response that is less than the
5073 * minimum it requested.
5074 */
5075 dprintk("%s: server responded with %d layouts, dropping tail\n",
5076 __func__, layout_count);
5077 }
5078
5079 return 0;
5080out_overflow:
5081 print_overflow_msg(__func__, xdr);
5082 return -EIO;
5083}
5084#endif /* CONFIG_NFS_V4_1 */
5085
4775/* 5086/*
4776 * END OF "GENERIC" DECODE ROUTINES. 5087 * END OF "GENERIC" DECODE ROUTINES.
4777 */ 5088 */
@@ -5799,6 +6110,53 @@ static int nfs4_xdr_dec_reclaim_complete(struct rpc_rqst *rqstp, uint32_t *p,
5799 status = decode_reclaim_complete(&xdr, (void *)NULL); 6110 status = decode_reclaim_complete(&xdr, (void *)NULL);
5800 return status; 6111 return status;
5801} 6112}
6113
6114/*
6115 * Decode GETDEVINFO response
6116 */
6117static int nfs4_xdr_dec_getdeviceinfo(struct rpc_rqst *rqstp, uint32_t *p,
6118 struct nfs4_getdeviceinfo_res *res)
6119{
6120 struct xdr_stream xdr;
6121 struct compound_hdr hdr;
6122 int status;
6123
6124 xdr_init_decode(&xdr, &rqstp->rq_rcv_buf, p);
6125 status = decode_compound_hdr(&xdr, &hdr);
6126 if (status != 0)
6127 goto out;
6128 status = decode_sequence(&xdr, &res->seq_res, rqstp);
6129 if (status != 0)
6130 goto out;
6131 status = decode_getdeviceinfo(&xdr, res->pdev);
6132out:
6133 return status;
6134}
6135
6136/*
6137 * Decode LAYOUTGET response
6138 */
6139static int nfs4_xdr_dec_layoutget(struct rpc_rqst *rqstp, uint32_t *p,
6140 struct nfs4_layoutget_res *res)
6141{
6142 struct xdr_stream xdr;
6143 struct compound_hdr hdr;
6144 int status;
6145
6146 xdr_init_decode(&xdr, &rqstp->rq_rcv_buf, p);
6147 status = decode_compound_hdr(&xdr, &hdr);
6148 if (status)
6149 goto out;
6150 status = decode_sequence(&xdr, &res->seq_res, rqstp);
6151 if (status)
6152 goto out;
6153 status = decode_putfh(&xdr);
6154 if (status)
6155 goto out;
6156 status = decode_layoutget(&xdr, rqstp, res);
6157out:
6158 return status;
6159}
5802#endif /* CONFIG_NFS_V4_1 */ 6160#endif /* CONFIG_NFS_V4_1 */
5803 6161
5804__be32 *nfs4_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry, 6162__be32 *nfs4_decode_dirent(struct xdr_stream *xdr, struct nfs_entry *entry,
@@ -5990,6 +6348,8 @@ struct rpc_procinfo nfs4_procedures[] = {
5990 PROC(SEQUENCE, enc_sequence, dec_sequence), 6348 PROC(SEQUENCE, enc_sequence, dec_sequence),
5991 PROC(GET_LEASE_TIME, enc_get_lease_time, dec_get_lease_time), 6349 PROC(GET_LEASE_TIME, enc_get_lease_time, dec_get_lease_time),
5992 PROC(RECLAIM_COMPLETE, enc_reclaim_complete, dec_reclaim_complete), 6350 PROC(RECLAIM_COMPLETE, enc_reclaim_complete, dec_reclaim_complete),
6351 PROC(GETDEVICEINFO, enc_getdeviceinfo, dec_getdeviceinfo),
6352 PROC(LAYOUTGET, enc_layoutget, dec_layoutget),
5993#endif /* CONFIG_NFS_V4_1 */ 6353#endif /* CONFIG_NFS_V4_1 */
5994}; 6354};
5995 6355
diff --git a/fs/nfs/pnfs.c b/fs/nfs/pnfs.c
new file mode 100644
index 000000000000..db773428f95f
--- /dev/null
+++ b/fs/nfs/pnfs.c
@@ -0,0 +1,783 @@
1/*
2 * pNFS functions to call and manage layout drivers.
3 *
4 * Copyright (c) 2002 [year of first publication]
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#include <linux/nfs_fs.h>
31#include "internal.h"
32#include "pnfs.h"
33
34#define NFSDBG_FACILITY NFSDBG_PNFS
35
36/* Locking:
37 *
38 * pnfs_spinlock:
39 * protects pnfs_modules_tbl.
40 */
41static DEFINE_SPINLOCK(pnfs_spinlock);
42
43/*
44 * pnfs_modules_tbl holds all pnfs modules
45 */
46static LIST_HEAD(pnfs_modules_tbl);
47
48/* Return the registered pnfs layout driver module matching given id */
49static struct pnfs_layoutdriver_type *
50find_pnfs_driver_locked(u32 id)
51{
52 struct pnfs_layoutdriver_type *local;
53
54 list_for_each_entry(local, &pnfs_modules_tbl, pnfs_tblid)
55 if (local->id == id)
56 goto out;
57 local = NULL;
58out:
59 dprintk("%s: Searching for id %u, found %p\n", __func__, id, local);
60 return local;
61}
62
63static struct pnfs_layoutdriver_type *
64find_pnfs_driver(u32 id)
65{
66 struct pnfs_layoutdriver_type *local;
67
68 spin_lock(&pnfs_spinlock);
69 local = find_pnfs_driver_locked(id);
70 spin_unlock(&pnfs_spinlock);
71 return local;
72}
73
74void
75unset_pnfs_layoutdriver(struct nfs_server *nfss)
76{
77 if (nfss->pnfs_curr_ld) {
78 nfss->pnfs_curr_ld->clear_layoutdriver(nfss);
79 module_put(nfss->pnfs_curr_ld->owner);
80 }
81 nfss->pnfs_curr_ld = NULL;
82}
83
84/*
85 * Try to set the server's pnfs module to the pnfs layout type specified by id.
86 * Currently only one pNFS layout driver per filesystem is supported.
87 *
88 * @id layout type. Zero (illegal layout type) indicates pNFS not in use.
89 */
90void
91set_pnfs_layoutdriver(struct nfs_server *server, u32 id)
92{
93 struct pnfs_layoutdriver_type *ld_type = NULL;
94
95 if (id == 0)
96 goto out_no_driver;
97 if (!(server->nfs_client->cl_exchange_flags &
98 (EXCHGID4_FLAG_USE_NON_PNFS | EXCHGID4_FLAG_USE_PNFS_MDS))) {
99 printk(KERN_ERR "%s: id %u cl_exchange_flags 0x%x\n", __func__,
100 id, server->nfs_client->cl_exchange_flags);
101 goto out_no_driver;
102 }
103 ld_type = find_pnfs_driver(id);
104 if (!ld_type) {
105 request_module("%s-%u", LAYOUT_NFSV4_1_MODULE_PREFIX, id);
106 ld_type = find_pnfs_driver(id);
107 if (!ld_type) {
108 dprintk("%s: No pNFS module found for %u.\n",
109 __func__, id);
110 goto out_no_driver;
111 }
112 }
113 if (!try_module_get(ld_type->owner)) {
114 dprintk("%s: Could not grab reference on module\n", __func__);
115 goto out_no_driver;
116 }
117 server->pnfs_curr_ld = ld_type;
118 if (ld_type->set_layoutdriver(server)) {
119 printk(KERN_ERR
120 "%s: Error initializing mount point for layout driver %u.\n",
121 __func__, id);
122 module_put(ld_type->owner);
123 goto out_no_driver;
124 }
125 dprintk("%s: pNFS module for %u set\n", __func__, id);
126 return;
127
128out_no_driver:
129 dprintk("%s: Using NFSv4 I/O\n", __func__);
130 server->pnfs_curr_ld = NULL;
131}
132
133int
134pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
135{
136 int status = -EINVAL;
137 struct pnfs_layoutdriver_type *tmp;
138
139 if (ld_type->id == 0) {
140 printk(KERN_ERR "%s id 0 is reserved\n", __func__);
141 return status;
142 }
143 if (!ld_type->alloc_lseg || !ld_type->free_lseg) {
144 printk(KERN_ERR "%s Layout driver must provide "
145 "alloc_lseg and free_lseg.\n", __func__);
146 return status;
147 }
148
149 spin_lock(&pnfs_spinlock);
150 tmp = find_pnfs_driver_locked(ld_type->id);
151 if (!tmp) {
152 list_add(&ld_type->pnfs_tblid, &pnfs_modules_tbl);
153 status = 0;
154 dprintk("%s Registering id:%u name:%s\n", __func__, ld_type->id,
155 ld_type->name);
156 } else {
157 printk(KERN_ERR "%s Module with id %d already loaded!\n",
158 __func__, ld_type->id);
159 }
160 spin_unlock(&pnfs_spinlock);
161
162 return status;
163}
164EXPORT_SYMBOL_GPL(pnfs_register_layoutdriver);
165
166void
167pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *ld_type)
168{
169 dprintk("%s Deregistering id:%u\n", __func__, ld_type->id);
170 spin_lock(&pnfs_spinlock);
171 list_del(&ld_type->pnfs_tblid);
172 spin_unlock(&pnfs_spinlock);
173}
174EXPORT_SYMBOL_GPL(pnfs_unregister_layoutdriver);
175
176/*
177 * pNFS client layout cache
178 */
179
180static void
181get_layout_hdr_locked(struct pnfs_layout_hdr *lo)
182{
183 assert_spin_locked(&lo->inode->i_lock);
184 lo->refcount++;
185}
186
187static void
188put_layout_hdr_locked(struct pnfs_layout_hdr *lo)
189{
190 assert_spin_locked(&lo->inode->i_lock);
191 BUG_ON(lo->refcount == 0);
192
193 lo->refcount--;
194 if (!lo->refcount) {
195 dprintk("%s: freeing layout cache %p\n", __func__, lo);
196 BUG_ON(!list_empty(&lo->layouts));
197 NFS_I(lo->inode)->layout = NULL;
198 kfree(lo);
199 }
200}
201
202void
203put_layout_hdr(struct inode *inode)
204{
205 spin_lock(&inode->i_lock);
206 put_layout_hdr_locked(NFS_I(inode)->layout);
207 spin_unlock(&inode->i_lock);
208}
209
210static void
211init_lseg(struct pnfs_layout_hdr *lo, struct pnfs_layout_segment *lseg)
212{
213 INIT_LIST_HEAD(&lseg->fi_list);
214 kref_init(&lseg->kref);
215 lseg->layout = lo;
216}
217
218/* Called without i_lock held, as the free_lseg call may sleep */
219static void
220destroy_lseg(struct kref *kref)
221{
222 struct pnfs_layout_segment *lseg =
223 container_of(kref, struct pnfs_layout_segment, kref);
224 struct inode *ino = lseg->layout->inode;
225
226 dprintk("--> %s\n", __func__);
227 NFS_SERVER(ino)->pnfs_curr_ld->free_lseg(lseg);
228 /* Matched by get_layout_hdr_locked in pnfs_insert_layout */
229 put_layout_hdr(ino);
230}
231
232static void
233put_lseg(struct pnfs_layout_segment *lseg)
234{
235 if (!lseg)
236 return;
237
238 dprintk("%s: lseg %p ref %d\n", __func__, lseg,
239 atomic_read(&lseg->kref.refcount));
240 kref_put(&lseg->kref, destroy_lseg);
241}
242
243static void
244pnfs_clear_lseg_list(struct pnfs_layout_hdr *lo, struct list_head *tmp_list)
245{
246 struct pnfs_layout_segment *lseg, *next;
247 struct nfs_client *clp;
248
249 dprintk("%s:Begin lo %p\n", __func__, lo);
250
251 assert_spin_locked(&lo->inode->i_lock);
252 list_for_each_entry_safe(lseg, next, &lo->segs, fi_list) {
253 dprintk("%s: freeing lseg %p\n", __func__, lseg);
254 list_move(&lseg->fi_list, tmp_list);
255 }
256 clp = NFS_SERVER(lo->inode)->nfs_client;
257 spin_lock(&clp->cl_lock);
258 /* List does not take a reference, so no need for put here */
259 list_del_init(&lo->layouts);
260 spin_unlock(&clp->cl_lock);
261 write_seqlock(&lo->seqlock);
262 clear_bit(NFS_LAYOUT_STATEID_SET, &lo->state);
263 write_sequnlock(&lo->seqlock);
264
265 dprintk("%s:Return\n", __func__);
266}
267
268static void
269pnfs_free_lseg_list(struct list_head *tmp_list)
270{
271 struct pnfs_layout_segment *lseg;
272
273 while (!list_empty(tmp_list)) {
274 lseg = list_entry(tmp_list->next, struct pnfs_layout_segment,
275 fi_list);
276 dprintk("%s calling put_lseg on %p\n", __func__, lseg);
277 list_del(&lseg->fi_list);
278 put_lseg(lseg);
279 }
280}
281
282void
283pnfs_destroy_layout(struct nfs_inode *nfsi)
284{
285 struct pnfs_layout_hdr *lo;
286 LIST_HEAD(tmp_list);
287
288 spin_lock(&nfsi->vfs_inode.i_lock);
289 lo = nfsi->layout;
290 if (lo) {
291 pnfs_clear_lseg_list(lo, &tmp_list);
292 /* Matched by refcount set to 1 in alloc_init_layout_hdr */
293 put_layout_hdr_locked(lo);
294 }
295 spin_unlock(&nfsi->vfs_inode.i_lock);
296 pnfs_free_lseg_list(&tmp_list);
297}
298
299/*
300 * Called by the state manger to remove all layouts established under an
301 * expired lease.
302 */
303void
304pnfs_destroy_all_layouts(struct nfs_client *clp)
305{
306 struct pnfs_layout_hdr *lo;
307 LIST_HEAD(tmp_list);
308
309 spin_lock(&clp->cl_lock);
310 list_splice_init(&clp->cl_layouts, &tmp_list);
311 spin_unlock(&clp->cl_lock);
312
313 while (!list_empty(&tmp_list)) {
314 lo = list_entry(tmp_list.next, struct pnfs_layout_hdr,
315 layouts);
316 dprintk("%s freeing layout for inode %lu\n", __func__,
317 lo->inode->i_ino);
318 pnfs_destroy_layout(NFS_I(lo->inode));
319 }
320}
321
322/* update lo->stateid with new if is more recent
323 *
324 * lo->stateid could be the open stateid, in which case we just use what given.
325 */
326static void
327pnfs_set_layout_stateid(struct pnfs_layout_hdr *lo,
328 const nfs4_stateid *new)
329{
330 nfs4_stateid *old = &lo->stateid;
331 bool overwrite = false;
332
333 write_seqlock(&lo->seqlock);
334 if (!test_bit(NFS_LAYOUT_STATEID_SET, &lo->state) ||
335 memcmp(old->stateid.other, new->stateid.other, sizeof(new->stateid.other)))
336 overwrite = true;
337 else {
338 u32 oldseq, newseq;
339
340 oldseq = be32_to_cpu(old->stateid.seqid);
341 newseq = be32_to_cpu(new->stateid.seqid);
342 if ((int)(newseq - oldseq) > 0)
343 overwrite = true;
344 }
345 if (overwrite)
346 memcpy(&old->stateid, &new->stateid, sizeof(new->stateid));
347 write_sequnlock(&lo->seqlock);
348}
349
350static void
351pnfs_layout_from_open_stateid(struct pnfs_layout_hdr *lo,
352 struct nfs4_state *state)
353{
354 int seq;
355
356 dprintk("--> %s\n", __func__);
357 write_seqlock(&lo->seqlock);
358 do {
359 seq = read_seqbegin(&state->seqlock);
360 memcpy(lo->stateid.data, state->stateid.data,
361 sizeof(state->stateid.data));
362 } while (read_seqretry(&state->seqlock, seq));
363 set_bit(NFS_LAYOUT_STATEID_SET, &lo->state);
364 write_sequnlock(&lo->seqlock);
365 dprintk("<-- %s\n", __func__);
366}
367
368void
369pnfs_get_layout_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
370 struct nfs4_state *open_state)
371{
372 int seq;
373
374 dprintk("--> %s\n", __func__);
375 do {
376 seq = read_seqbegin(&lo->seqlock);
377 if (!test_bit(NFS_LAYOUT_STATEID_SET, &lo->state)) {
378 /* This will trigger retry of the read */
379 pnfs_layout_from_open_stateid(lo, open_state);
380 } else
381 memcpy(dst->data, lo->stateid.data,
382 sizeof(lo->stateid.data));
383 } while (read_seqretry(&lo->seqlock, seq));
384 dprintk("<-- %s\n", __func__);
385}
386
387/*
388* Get layout from server.
389* for now, assume that whole file layouts are requested.
390* arg->offset: 0
391* arg->length: all ones
392*/
393static struct pnfs_layout_segment *
394send_layoutget(struct pnfs_layout_hdr *lo,
395 struct nfs_open_context *ctx,
396 u32 iomode)
397{
398 struct inode *ino = lo->inode;
399 struct nfs_server *server = NFS_SERVER(ino);
400 struct nfs4_layoutget *lgp;
401 struct pnfs_layout_segment *lseg = NULL;
402
403 dprintk("--> %s\n", __func__);
404
405 BUG_ON(ctx == NULL);
406 lgp = kzalloc(sizeof(*lgp), GFP_KERNEL);
407 if (lgp == NULL) {
408 put_layout_hdr(lo->inode);
409 return NULL;
410 }
411 lgp->args.minlength = NFS4_MAX_UINT64;
412 lgp->args.maxcount = PNFS_LAYOUT_MAXSIZE;
413 lgp->args.range.iomode = iomode;
414 lgp->args.range.offset = 0;
415 lgp->args.range.length = NFS4_MAX_UINT64;
416 lgp->args.type = server->pnfs_curr_ld->id;
417 lgp->args.inode = ino;
418 lgp->args.ctx = get_nfs_open_context(ctx);
419 lgp->lsegpp = &lseg;
420
421 /* Synchronously retrieve layout information from server and
422 * store in lseg.
423 */
424 nfs4_proc_layoutget(lgp);
425 if (!lseg) {
426 /* remember that LAYOUTGET failed and suspend trying */
427 set_bit(lo_fail_bit(iomode), &lo->state);
428 }
429 return lseg;
430}
431
432/*
433 * Compare two layout segments for sorting into layout cache.
434 * We want to preferentially return RW over RO layouts, so ensure those
435 * are seen first.
436 */
437static s64
438cmp_layout(u32 iomode1, u32 iomode2)
439{
440 /* read > read/write */
441 return (int)(iomode2 == IOMODE_READ) - (int)(iomode1 == IOMODE_READ);
442}
443
444static void
445pnfs_insert_layout(struct pnfs_layout_hdr *lo,
446 struct pnfs_layout_segment *lseg)
447{
448 struct pnfs_layout_segment *lp;
449 int found = 0;
450
451 dprintk("%s:Begin\n", __func__);
452
453 assert_spin_locked(&lo->inode->i_lock);
454 if (list_empty(&lo->segs)) {
455 struct nfs_client *clp = NFS_SERVER(lo->inode)->nfs_client;
456
457 spin_lock(&clp->cl_lock);
458 BUG_ON(!list_empty(&lo->layouts));
459 list_add_tail(&lo->layouts, &clp->cl_layouts);
460 spin_unlock(&clp->cl_lock);
461 }
462 list_for_each_entry(lp, &lo->segs, fi_list) {
463 if (cmp_layout(lp->range.iomode, lseg->range.iomode) > 0)
464 continue;
465 list_add_tail(&lseg->fi_list, &lp->fi_list);
466 dprintk("%s: inserted lseg %p "
467 "iomode %d offset %llu length %llu before "
468 "lp %p iomode %d offset %llu length %llu\n",
469 __func__, lseg, lseg->range.iomode,
470 lseg->range.offset, lseg->range.length,
471 lp, lp->range.iomode, lp->range.offset,
472 lp->range.length);
473 found = 1;
474 break;
475 }
476 if (!found) {
477 list_add_tail(&lseg->fi_list, &lo->segs);
478 dprintk("%s: inserted lseg %p "
479 "iomode %d offset %llu length %llu at tail\n",
480 __func__, lseg, lseg->range.iomode,
481 lseg->range.offset, lseg->range.length);
482 }
483 get_layout_hdr_locked(lo);
484
485 dprintk("%s:Return\n", __func__);
486}
487
488static struct pnfs_layout_hdr *
489alloc_init_layout_hdr(struct inode *ino)
490{
491 struct pnfs_layout_hdr *lo;
492
493 lo = kzalloc(sizeof(struct pnfs_layout_hdr), GFP_KERNEL);
494 if (!lo)
495 return NULL;
496 lo->refcount = 1;
497 INIT_LIST_HEAD(&lo->layouts);
498 INIT_LIST_HEAD(&lo->segs);
499 seqlock_init(&lo->seqlock);
500 lo->inode = ino;
501 return lo;
502}
503
504static struct pnfs_layout_hdr *
505pnfs_find_alloc_layout(struct inode *ino)
506{
507 struct nfs_inode *nfsi = NFS_I(ino);
508 struct pnfs_layout_hdr *new = NULL;
509
510 dprintk("%s Begin ino=%p layout=%p\n", __func__, ino, nfsi->layout);
511
512 assert_spin_locked(&ino->i_lock);
513 if (nfsi->layout)
514 return nfsi->layout;
515
516 spin_unlock(&ino->i_lock);
517 new = alloc_init_layout_hdr(ino);
518 spin_lock(&ino->i_lock);
519
520 if (likely(nfsi->layout == NULL)) /* Won the race? */
521 nfsi->layout = new;
522 else
523 kfree(new);
524 return nfsi->layout;
525}
526
527/*
528 * iomode matching rules:
529 * iomode lseg match
530 * ----- ----- -----
531 * ANY READ true
532 * ANY RW true
533 * RW READ false
534 * RW RW true
535 * READ READ true
536 * READ RW true
537 */
538static int
539is_matching_lseg(struct pnfs_layout_segment *lseg, u32 iomode)
540{
541 return (iomode != IOMODE_RW || lseg->range.iomode == IOMODE_RW);
542}
543
544/*
545 * lookup range in layout
546 */
547static struct pnfs_layout_segment *
548pnfs_has_layout(struct pnfs_layout_hdr *lo, u32 iomode)
549{
550 struct pnfs_layout_segment *lseg, *ret = NULL;
551
552 dprintk("%s:Begin\n", __func__);
553
554 assert_spin_locked(&lo->inode->i_lock);
555 list_for_each_entry(lseg, &lo->segs, fi_list) {
556 if (is_matching_lseg(lseg, iomode)) {
557 ret = lseg;
558 break;
559 }
560 if (cmp_layout(iomode, lseg->range.iomode) > 0)
561 break;
562 }
563
564 dprintk("%s:Return lseg %p ref %d\n",
565 __func__, ret, ret ? atomic_read(&ret->kref.refcount) : 0);
566 return ret;
567}
568
569/*
570 * Layout segment is retreived from the server if not cached.
571 * The appropriate layout segment is referenced and returned to the caller.
572 */
573struct pnfs_layout_segment *
574pnfs_update_layout(struct inode *ino,
575 struct nfs_open_context *ctx,
576 enum pnfs_iomode iomode)
577{
578 struct nfs_inode *nfsi = NFS_I(ino);
579 struct pnfs_layout_hdr *lo;
580 struct pnfs_layout_segment *lseg = NULL;
581
582 if (!pnfs_enabled_sb(NFS_SERVER(ino)))
583 return NULL;
584 spin_lock(&ino->i_lock);
585 lo = pnfs_find_alloc_layout(ino);
586 if (lo == NULL) {
587 dprintk("%s ERROR: can't get pnfs_layout_hdr\n", __func__);
588 goto out_unlock;
589 }
590
591 /* Check to see if the layout for the given range already exists */
592 lseg = pnfs_has_layout(lo, iomode);
593 if (lseg) {
594 dprintk("%s: Using cached lseg %p for iomode %d)\n",
595 __func__, lseg, iomode);
596 goto out_unlock;
597 }
598
599 /* if LAYOUTGET already failed once we don't try again */
600 if (test_bit(lo_fail_bit(iomode), &nfsi->layout->state))
601 goto out_unlock;
602
603 get_layout_hdr_locked(lo); /* Matched in nfs4_layoutget_release */
604 spin_unlock(&ino->i_lock);
605
606 lseg = send_layoutget(lo, ctx, iomode);
607out:
608 dprintk("%s end, state 0x%lx lseg %p\n", __func__,
609 nfsi->layout->state, lseg);
610 return lseg;
611out_unlock:
612 spin_unlock(&ino->i_lock);
613 goto out;
614}
615
616int
617pnfs_layout_process(struct nfs4_layoutget *lgp)
618{
619 struct pnfs_layout_hdr *lo = NFS_I(lgp->args.inode)->layout;
620 struct nfs4_layoutget_res *res = &lgp->res;
621 struct pnfs_layout_segment *lseg;
622 struct inode *ino = lo->inode;
623 int status = 0;
624
625 /* Inject layout blob into I/O device driver */
626 lseg = NFS_SERVER(ino)->pnfs_curr_ld->alloc_lseg(lo, res);
627 if (!lseg || IS_ERR(lseg)) {
628 if (!lseg)
629 status = -ENOMEM;
630 else
631 status = PTR_ERR(lseg);
632 dprintk("%s: Could not allocate layout: error %d\n",
633 __func__, status);
634 goto out;
635 }
636
637 spin_lock(&ino->i_lock);
638 init_lseg(lo, lseg);
639 lseg->range = res->range;
640 *lgp->lsegpp = lseg;
641 pnfs_insert_layout(lo, lseg);
642
643 /* Done processing layoutget. Set the layout stateid */
644 pnfs_set_layout_stateid(lo, &res->stateid);
645 spin_unlock(&ino->i_lock);
646out:
647 return status;
648}
649
650/*
651 * Device ID cache. Currently supports one layout type per struct nfs_client.
652 * Add layout type to the lookup key to expand to support multiple types.
653 */
654int
655pnfs_alloc_init_deviceid_cache(struct nfs_client *clp,
656 void (*free_callback)(struct pnfs_deviceid_node *))
657{
658 struct pnfs_deviceid_cache *c;
659
660 c = kzalloc(sizeof(struct pnfs_deviceid_cache), GFP_KERNEL);
661 if (!c)
662 return -ENOMEM;
663 spin_lock(&clp->cl_lock);
664 if (clp->cl_devid_cache != NULL) {
665 atomic_inc(&clp->cl_devid_cache->dc_ref);
666 dprintk("%s [kref [%d]]\n", __func__,
667 atomic_read(&clp->cl_devid_cache->dc_ref));
668 kfree(c);
669 } else {
670 /* kzalloc initializes hlists */
671 spin_lock_init(&c->dc_lock);
672 atomic_set(&c->dc_ref, 1);
673 c->dc_free_callback = free_callback;
674 clp->cl_devid_cache = c;
675 dprintk("%s [new]\n", __func__);
676 }
677 spin_unlock(&clp->cl_lock);
678 return 0;
679}
680EXPORT_SYMBOL_GPL(pnfs_alloc_init_deviceid_cache);
681
682/*
683 * Called from pnfs_layoutdriver_type->free_lseg
684 * last layout segment reference frees deviceid
685 */
686void
687pnfs_put_deviceid(struct pnfs_deviceid_cache *c,
688 struct pnfs_deviceid_node *devid)
689{
690 struct nfs4_deviceid *id = &devid->de_id;
691 struct pnfs_deviceid_node *d;
692 struct hlist_node *n;
693 long h = nfs4_deviceid_hash(id);
694
695 dprintk("%s [%d]\n", __func__, atomic_read(&devid->de_ref));
696 if (!atomic_dec_and_lock(&devid->de_ref, &c->dc_lock))
697 return;
698
699 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[h], de_node)
700 if (!memcmp(&d->de_id, id, sizeof(*id))) {
701 hlist_del_rcu(&d->de_node);
702 spin_unlock(&c->dc_lock);
703 synchronize_rcu();
704 c->dc_free_callback(devid);
705 return;
706 }
707 spin_unlock(&c->dc_lock);
708 /* Why wasn't it found in the list? */
709 BUG();
710}
711EXPORT_SYMBOL_GPL(pnfs_put_deviceid);
712
713/* Find and reference a deviceid */
714struct pnfs_deviceid_node *
715pnfs_find_get_deviceid(struct pnfs_deviceid_cache *c, struct nfs4_deviceid *id)
716{
717 struct pnfs_deviceid_node *d;
718 struct hlist_node *n;
719 long hash = nfs4_deviceid_hash(id);
720
721 dprintk("--> %s hash %ld\n", __func__, hash);
722 rcu_read_lock();
723 hlist_for_each_entry_rcu(d, n, &c->dc_deviceids[hash], de_node) {
724 if (!memcmp(&d->de_id, id, sizeof(*id))) {
725 if (!atomic_inc_not_zero(&d->de_ref)) {
726 goto fail;
727 } else {
728 rcu_read_unlock();
729 return d;
730 }
731 }
732 }
733fail:
734 rcu_read_unlock();
735 return NULL;
736}
737EXPORT_SYMBOL_GPL(pnfs_find_get_deviceid);
738
739/*
740 * Add a deviceid to the cache.
741 * GETDEVICEINFOs for same deviceid can race. If deviceid is found, discard new
742 */
743struct pnfs_deviceid_node *
744pnfs_add_deviceid(struct pnfs_deviceid_cache *c, struct pnfs_deviceid_node *new)
745{
746 struct pnfs_deviceid_node *d;
747 long hash = nfs4_deviceid_hash(&new->de_id);
748
749 dprintk("--> %s hash %ld\n", __func__, hash);
750 spin_lock(&c->dc_lock);
751 d = pnfs_find_get_deviceid(c, &new->de_id);
752 if (d) {
753 spin_unlock(&c->dc_lock);
754 dprintk("%s [discard]\n", __func__);
755 c->dc_free_callback(new);
756 return d;
757 }
758 INIT_HLIST_NODE(&new->de_node);
759 atomic_set(&new->de_ref, 1);
760 hlist_add_head_rcu(&new->de_node, &c->dc_deviceids[hash]);
761 spin_unlock(&c->dc_lock);
762 dprintk("%s [new]\n", __func__);
763 return new;
764}
765EXPORT_SYMBOL_GPL(pnfs_add_deviceid);
766
767void
768pnfs_put_deviceid_cache(struct nfs_client *clp)
769{
770 struct pnfs_deviceid_cache *local = clp->cl_devid_cache;
771
772 dprintk("--> %s cl_devid_cache %p\n", __func__, clp->cl_devid_cache);
773 if (atomic_dec_and_lock(&local->dc_ref, &clp->cl_lock)) {
774 int i;
775 /* Verify cache is empty */
776 for (i = 0; i < NFS4_DEVICE_ID_HASH_SIZE; i++)
777 BUG_ON(!hlist_empty(&local->dc_deviceids[i]));
778 clp->cl_devid_cache = NULL;
779 spin_unlock(&clp->cl_lock);
780 kfree(local);
781 }
782}
783EXPORT_SYMBOL_GPL(pnfs_put_deviceid_cache);
diff --git a/fs/nfs/pnfs.h b/fs/nfs/pnfs.h
new file mode 100644
index 000000000000..e12367d50489
--- /dev/null
+++ b/fs/nfs/pnfs.h
@@ -0,0 +1,189 @@
1/*
2 * pNFS client data structures.
3 *
4 * Copyright (c) 2002
5 * The Regents of the University of Michigan
6 * All Rights Reserved
7 *
8 * Dean Hildebrand <dhildebz@umich.edu>
9 *
10 * Permission is granted to use, copy, create derivative works, and
11 * redistribute this software and such derivative works for any purpose,
12 * so long as the name of the University of Michigan is not used in
13 * any advertising or publicity pertaining to the use or distribution
14 * of this software without specific, written prior authorization. If
15 * the above copyright notice or any other identification of the
16 * University of Michigan is included in any copy of any portion of
17 * this software, then the disclaimer below must also be included.
18 *
19 * This software is provided as is, without representation or warranty
20 * of any kind either express or implied, including without limitation
21 * the implied warranties of merchantability, fitness for a particular
22 * purpose, or noninfringement. The Regents of the University of
23 * Michigan shall not be liable for any damages, including special,
24 * indirect, incidental, or consequential damages, with respect to any
25 * claim arising out of or in connection with the use of the software,
26 * even if it has been or is hereafter advised of the possibility of
27 * such damages.
28 */
29
30#ifndef FS_NFS_PNFS_H
31#define FS_NFS_PNFS_H
32
33struct pnfs_layout_segment {
34 struct list_head fi_list;
35 struct pnfs_layout_range range;
36 struct kref kref;
37 struct pnfs_layout_hdr *layout;
38};
39
40#ifdef CONFIG_NFS_V4_1
41
42#define LAYOUT_NFSV4_1_MODULE_PREFIX "nfs-layouttype4"
43
44enum {
45 NFS_LAYOUT_RO_FAILED = 0, /* get ro layout failed stop trying */
46 NFS_LAYOUT_RW_FAILED, /* get rw layout failed stop trying */
47 NFS_LAYOUT_STATEID_SET, /* have a valid layout stateid */
48};
49
50/* Per-layout driver specific registration structure */
51struct pnfs_layoutdriver_type {
52 struct list_head pnfs_tblid;
53 const u32 id;
54 const char *name;
55 struct module *owner;
56 int (*set_layoutdriver) (struct nfs_server *);
57 int (*clear_layoutdriver) (struct nfs_server *);
58 struct pnfs_layout_segment * (*alloc_lseg) (struct pnfs_layout_hdr *layoutid, struct nfs4_layoutget_res *lgr);
59 void (*free_lseg) (struct pnfs_layout_segment *lseg);
60};
61
62struct pnfs_layout_hdr {
63 unsigned long refcount;
64 struct list_head layouts; /* other client layouts */
65 struct list_head segs; /* layout segments list */
66 seqlock_t seqlock; /* Protects the stateid */
67 nfs4_stateid stateid;
68 unsigned long state;
69 struct inode *inode;
70};
71
72struct pnfs_device {
73 struct nfs4_deviceid dev_id;
74 unsigned int layout_type;
75 unsigned int mincount;
76 struct page **pages;
77 void *area;
78 unsigned int pgbase;
79 unsigned int pglen;
80};
81
82/*
83 * Device ID RCU cache. A device ID is unique per client ID and layout type.
84 */
85#define NFS4_DEVICE_ID_HASH_BITS 5
86#define NFS4_DEVICE_ID_HASH_SIZE (1 << NFS4_DEVICE_ID_HASH_BITS)
87#define NFS4_DEVICE_ID_HASH_MASK (NFS4_DEVICE_ID_HASH_SIZE - 1)
88
89static inline u32
90nfs4_deviceid_hash(struct nfs4_deviceid *id)
91{
92 unsigned char *cptr = (unsigned char *)id->data;
93 unsigned int nbytes = NFS4_DEVICEID4_SIZE;
94 u32 x = 0;
95
96 while (nbytes--) {
97 x *= 37;
98 x += *cptr++;
99 }
100 return x & NFS4_DEVICE_ID_HASH_MASK;
101}
102
103struct pnfs_deviceid_node {
104 struct hlist_node de_node;
105 struct nfs4_deviceid de_id;
106 atomic_t de_ref;
107};
108
109struct pnfs_deviceid_cache {
110 spinlock_t dc_lock;
111 atomic_t dc_ref;
112 void (*dc_free_callback)(struct pnfs_deviceid_node *);
113 struct hlist_head dc_deviceids[NFS4_DEVICE_ID_HASH_SIZE];
114};
115
116extern int pnfs_alloc_init_deviceid_cache(struct nfs_client *,
117 void (*free_callback)(struct pnfs_deviceid_node *));
118extern void pnfs_put_deviceid_cache(struct nfs_client *);
119extern struct pnfs_deviceid_node *pnfs_find_get_deviceid(
120 struct pnfs_deviceid_cache *,
121 struct nfs4_deviceid *);
122extern struct pnfs_deviceid_node *pnfs_add_deviceid(
123 struct pnfs_deviceid_cache *,
124 struct pnfs_deviceid_node *);
125extern void pnfs_put_deviceid(struct pnfs_deviceid_cache *c,
126 struct pnfs_deviceid_node *devid);
127
128extern int pnfs_register_layoutdriver(struct pnfs_layoutdriver_type *);
129extern void pnfs_unregister_layoutdriver(struct pnfs_layoutdriver_type *);
130
131/* nfs4proc.c */
132extern int nfs4_proc_getdeviceinfo(struct nfs_server *server,
133 struct pnfs_device *dev);
134extern int nfs4_proc_layoutget(struct nfs4_layoutget *lgp);
135
136/* pnfs.c */
137struct pnfs_layout_segment *
138pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx,
139 enum pnfs_iomode access_type);
140void set_pnfs_layoutdriver(struct nfs_server *, u32 id);
141void unset_pnfs_layoutdriver(struct nfs_server *);
142int pnfs_layout_process(struct nfs4_layoutget *lgp);
143void pnfs_destroy_layout(struct nfs_inode *);
144void pnfs_destroy_all_layouts(struct nfs_client *);
145void put_layout_hdr(struct inode *inode);
146void pnfs_get_layout_stateid(nfs4_stateid *dst, struct pnfs_layout_hdr *lo,
147 struct nfs4_state *open_state);
148
149
150static inline int lo_fail_bit(u32 iomode)
151{
152 return iomode == IOMODE_RW ?
153 NFS_LAYOUT_RW_FAILED : NFS_LAYOUT_RO_FAILED;
154}
155
156/* Return true if a layout driver is being used for this mountpoint */
157static inline int pnfs_enabled_sb(struct nfs_server *nfss)
158{
159 return nfss->pnfs_curr_ld != NULL;
160}
161
162#else /* CONFIG_NFS_V4_1 */
163
164static inline void pnfs_destroy_all_layouts(struct nfs_client *clp)
165{
166}
167
168static inline void pnfs_destroy_layout(struct nfs_inode *nfsi)
169{
170}
171
172static inline struct pnfs_layout_segment *
173pnfs_update_layout(struct inode *ino, struct nfs_open_context *ctx,
174 enum pnfs_iomode access_type)
175{
176 return NULL;
177}
178
179static inline void set_pnfs_layoutdriver(struct nfs_server *s, u32 id)
180{
181}
182
183static inline void unset_pnfs_layoutdriver(struct nfs_server *s)
184{
185}
186
187#endif /* CONFIG_NFS_V4_1 */
188
189#endif /* FS_NFS_PNFS_H */
diff --git a/fs/nfs/read.c b/fs/nfs/read.c
index 79859c81a943..e4b62c6f5a6e 100644
--- a/fs/nfs/read.c
+++ b/fs/nfs/read.c
@@ -25,6 +25,7 @@
25#include "internal.h" 25#include "internal.h"
26#include "iostat.h" 26#include "iostat.h"
27#include "fscache.h" 27#include "fscache.h"
28#include "pnfs.h"
28 29
29#define NFSDBG_FACILITY NFSDBG_PAGECACHE 30#define NFSDBG_FACILITY NFSDBG_PAGECACHE
30 31
@@ -120,6 +121,7 @@ int nfs_readpage_async(struct nfs_open_context *ctx, struct inode *inode,
120 len = nfs_page_length(page); 121 len = nfs_page_length(page);
121 if (len == 0) 122 if (len == 0)
122 return nfs_return_empty_page(page); 123 return nfs_return_empty_page(page);
124 pnfs_update_layout(inode, ctx, IOMODE_READ);
123 new = nfs_create_request(ctx, inode, page, 0, len); 125 new = nfs_create_request(ctx, inode, page, 0, len);
124 if (IS_ERR(new)) { 126 if (IS_ERR(new)) {
125 unlock_page(page); 127 unlock_page(page);
@@ -624,6 +626,7 @@ int nfs_readpages(struct file *filp, struct address_space *mapping,
624 if (ret == 0) 626 if (ret == 0)
625 goto read_complete; /* all pages were read */ 627 goto read_complete; /* all pages were read */
626 628
629 pnfs_update_layout(inode, desc.ctx, IOMODE_READ);
627 if (rsize < PAGE_CACHE_SIZE) 630 if (rsize < PAGE_CACHE_SIZE)
628 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0); 631 nfs_pageio_init(&pgio, inode, nfs_pagein_multi, rsize, 0);
629 else 632 else
diff --git a/fs/nfsd/nfs4callback.c b/fs/nfsd/nfs4callback.c
index 988cbb3a19b6..014482c4e57d 100644
--- a/fs/nfsd/nfs4callback.c
+++ b/fs/nfsd/nfs4callback.c
@@ -41,7 +41,6 @@
41 41
42#define NFSPROC4_CB_NULL 0 42#define NFSPROC4_CB_NULL 0
43#define NFSPROC4_CB_COMPOUND 1 43#define NFSPROC4_CB_COMPOUND 1
44#define NFS4_STATEID_SIZE 16
45 44
46/* Index of predefined Linux callback client operations */ 45/* Index of predefined Linux callback client operations */
47 46
generated by cgit v1.2.2 (git 2.25.0) at 2024-10-09 19:31:35 -0400