aboutsummaryrefslogtreecommitdiffstats
path: root/fs/exofs
diff options
context:
space:
mode:
Diffstat (limited to 'fs/exofs')
-rw-r--r--fs/exofs/Kbuild2
-rw-r--r--fs/exofs/common.h112
-rw-r--r--fs/exofs/exofs.h140
-rw-r--r--fs/exofs/inode.c547
-rw-r--r--fs/exofs/ios.c823
-rw-r--r--fs/exofs/osd.c125
-rw-r--r--fs/exofs/pnfs.h45
-rw-r--r--fs/exofs/super.c417
8 files changed, 1723 insertions, 488 deletions
diff --git a/fs/exofs/Kbuild b/fs/exofs/Kbuild
index cc2d22db119c..2d0f757fda3e 100644
--- a/fs/exofs/Kbuild
+++ b/fs/exofs/Kbuild
@@ -12,5 +12,5 @@
12# Kbuild - Gets included from the Kernels Makefile and build system 12# Kbuild - Gets included from the Kernels Makefile and build system
13# 13#
14 14
15exofs-y := osd.o inode.o file.o symlink.o namei.o dir.o super.o 15exofs-y := ios.o inode.o file.o symlink.o namei.o dir.o super.o
16obj-$(CONFIG_EXOFS_FS) += exofs.o 16obj-$(CONFIG_EXOFS_FS) += exofs.o
diff --git a/fs/exofs/common.h b/fs/exofs/common.h
index c6718e4817fe..f0d520312d8b 100644
--- a/fs/exofs/common.h
+++ b/fs/exofs/common.h
@@ -49,11 +49,14 @@
49#define EXOFS_MIN_PID 0x10000 /* Smallest partition ID */ 49#define EXOFS_MIN_PID 0x10000 /* Smallest partition ID */
50#define EXOFS_OBJ_OFF 0x10000 /* offset for objects */ 50#define EXOFS_OBJ_OFF 0x10000 /* offset for objects */
51#define EXOFS_SUPER_ID 0x10000 /* object ID for on-disk superblock */ 51#define EXOFS_SUPER_ID 0x10000 /* object ID for on-disk superblock */
52#define EXOFS_DEVTABLE_ID 0x10001 /* object ID for on-disk device table */
52#define EXOFS_ROOT_ID 0x10002 /* object ID for root directory */ 53#define EXOFS_ROOT_ID 0x10002 /* object ID for root directory */
53 54
54/* exofs Application specific page/attribute */ 55/* exofs Application specific page/attribute */
55# define EXOFS_APAGE_FS_DATA (OSD_APAGE_APP_DEFINED_FIRST + 3) 56# define EXOFS_APAGE_FS_DATA (OSD_APAGE_APP_DEFINED_FIRST + 3)
56# define EXOFS_ATTR_INODE_DATA 1 57# define EXOFS_ATTR_INODE_DATA 1
58# define EXOFS_ATTR_INODE_FILE_LAYOUT 2
59# define EXOFS_ATTR_INODE_DIR_LAYOUT 3
57 60
58/* 61/*
59 * The maximum number of files we can have is limited by the size of the 62 * The maximum number of files we can have is limited by the size of the
@@ -78,17 +81,67 @@ enum {
78#define EXOFS_SUPER_MAGIC 0x5DF5 81#define EXOFS_SUPER_MAGIC 0x5DF5
79 82
80/* 83/*
81 * The file system control block - stored in an object's data (mainly, the one 84 * The file system control block - stored in object EXOFS_SUPER_ID's data.
82 * with ID EXOFS_SUPER_ID). This is where the in-memory superblock is stored 85 * This is where the in-memory superblock is stored on disk.
83 * on disk. Right now it just has a magic value, which is basically a sanity
84 * check on our ability to communicate with the object store.
85 */ 86 */
87enum {EXOFS_FSCB_VER = 1, EXOFS_DT_VER = 1};
86struct exofs_fscb { 88struct exofs_fscb {
87 __le64 s_nextid; /* Highest object ID used */ 89 __le64 s_nextid; /* Highest object ID used */
88 __le32 s_numfiles; /* Number of files on fs */ 90 __le64 s_numfiles; /* Number of files on fs */
91 __le32 s_version; /* == EXOFS_FSCB_VER */
89 __le16 s_magic; /* Magic signature */ 92 __le16 s_magic; /* Magic signature */
90 __le16 s_newfs; /* Non-zero if this is a new fs */ 93 __le16 s_newfs; /* Non-zero if this is a new fs */
91}; 94
95 /* From here on it's a static part, only written by mkexofs */
96 __le64 s_dev_table_oid; /* Resurved, not used */
97 __le64 s_dev_table_count; /* == 0 means no dev_table */
98} __packed;
99
100/*
101 * Describes the raid used in the FS. It is part of the device table.
102 * This here is taken from the pNFS-objects definition. In exofs we
103 * use one raid policy through-out the filesystem. (NOTE: the funny
104 * alignment at begining. We take care of it at exofs_device_table.
105 */
106struct exofs_dt_data_map {
107 __le32 cb_num_comps;
108 __le64 cb_stripe_unit;
109 __le32 cb_group_width;
110 __le32 cb_group_depth;
111 __le32 cb_mirror_cnt;
112 __le32 cb_raid_algorithm;
113} __packed;
114
115/*
116 * This is an osd device information descriptor. It is a single entry in
117 * the exofs device table. It describes an osd target lun which
118 * contains data belonging to this FS. (Same partition_id on all devices)
119 */
120struct exofs_dt_device_info {
121 __le32 systemid_len;
122 u8 systemid[OSD_SYSTEMID_LEN];
123 __le64 long_name_offset; /* If !0 then offset-in-file */
124 __le32 osdname_len; /* */
125 u8 osdname[44]; /* Embbeded, Ususally an asci uuid */
126} __packed;
127
128/*
129 * The EXOFS device table - stored in object EXOFS_DEVTABLE_ID's data.
130 * It contains the raid used for this multy-device FS and an array of
131 * participating devices.
132 */
133struct exofs_device_table {
134 __le32 dt_version; /* == EXOFS_DT_VER */
135 struct exofs_dt_data_map dt_data_map; /* Raid policy to use */
136
137 /* Resurved space For future use. Total includeing this:
138 * (8 * sizeof(le64))
139 */
140 __le64 __Resurved[4];
141
142 __le64 dt_num_devices; /* Array size */
143 struct exofs_dt_device_info dt_dev_table[]; /* Array of devices */
144} __packed;
92 145
93/**************************************************************************** 146/****************************************************************************
94 * inode-related things 147 * inode-related things
@@ -155,22 +208,41 @@ enum {
155 (((name_len) + offsetof(struct exofs_dir_entry, name) + \ 208 (((name_len) + offsetof(struct exofs_dir_entry, name) + \
156 EXOFS_DIR_ROUND) & ~EXOFS_DIR_ROUND) 209 EXOFS_DIR_ROUND) & ~EXOFS_DIR_ROUND)
157 210
158/************************* 211/*
159 * function declarations * 212 * The on-disk (optional) layout structure.
160 *************************/ 213 * sits in an EXOFS_ATTR_INODE_FILE_LAYOUT or EXOFS_ATTR_INODE_DIR_LAYOUT
161/* osd.c */ 214 * attribute, attached to any inode, usually to a directory.
162void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], 215 */
163 const struct osd_obj_id *obj); 216
217enum exofs_inode_layout_gen_functions {
218 LAYOUT_MOVING_WINDOW = 0,
219 LAYOUT_IMPLICT = 1,
220};
164 221
165int exofs_check_ok_resid(struct osd_request *or, u64 *in_resid, u64 *out_resid); 222struct exofs_on_disk_inode_layout {
166static inline int exofs_check_ok(struct osd_request *or) 223 __le16 gen_func; /* One of enum exofs_inode_layout_gen_functions */
224 __le16 pad;
225 union {
226 /* gen_func == LAYOUT_MOVING_WINDOW (default) */
227 struct exofs_layout_sliding_window {
228 __le32 num_devices; /* first n devices in global-table*/
229 } sliding_window __packed;
230
231 /* gen_func == LAYOUT_IMPLICT */
232 struct exofs_layout_implict_list {
233 struct exofs_dt_data_map data_map;
234 /* Variable array of size data_map.cb_num_comps. These
235 * are device indexes of the devices in the global table
236 */
237 __le32 dev_indexes[];
238 } implict __packed;
239 };
240} __packed;
241
242static inline size_t exofs_on_disk_inode_layout_size(unsigned max_devs)
167{ 243{
168 return exofs_check_ok_resid(or, NULL, NULL); 244 return sizeof(struct exofs_on_disk_inode_layout) +
245 max_devs * sizeof(__le32);
169} 246}
170int exofs_sync_op(struct osd_request *or, int timeout, u8 *cred);
171int exofs_async_op(struct osd_request *or,
172 osd_req_done_fn *async_done, void *caller_context, u8 *cred);
173
174int extract_attr_from_req(struct osd_request *or, struct osd_attr *attr);
175 247
176#endif /*ifndef __EXOFS_COM_H__*/ 248#endif /*ifndef __EXOFS_COM_H__*/
diff --git a/fs/exofs/exofs.h b/fs/exofs/exofs.h
index 5ec72e020b22..8442e353309f 100644
--- a/fs/exofs/exofs.h
+++ b/fs/exofs/exofs.h
@@ -30,13 +30,17 @@
30 * along with exofs; if not, write to the Free Software 30 * along with exofs; if not, write to the Free Software
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
32 */ 32 */
33#ifndef __EXOFS_H__
34#define __EXOFS_H__
33 35
34#include <linux/fs.h> 36#include <linux/fs.h>
35#include <linux/time.h> 37#include <linux/time.h>
36#include "common.h" 38#include "common.h"
37 39
38#ifndef __EXOFS_H__ 40/* FIXME: Remove once pnfs hits mainline
39#define __EXOFS_H__ 41 * #include <linux/exportfs/pnfs_osd_xdr.h>
42 */
43#include "pnfs.h"
40 44
41#define EXOFS_ERR(fmt, a...) printk(KERN_ERR "exofs: " fmt, ##a) 45#define EXOFS_ERR(fmt, a...) printk(KERN_ERR "exofs: " fmt, ##a)
42 46
@@ -51,34 +55,110 @@
51/* u64 has problems with printk this will cast it to unsigned long long */ 55/* u64 has problems with printk this will cast it to unsigned long long */
52#define _LLU(x) (unsigned long long)(x) 56#define _LLU(x) (unsigned long long)(x)
53 57
58struct exofs_layout {
59 osd_id s_pid; /* partition ID of file system*/
60
61 /* Our way of looking at the data_map */
62 unsigned stripe_unit;
63 unsigned mirrors_p1;
64
65 unsigned group_width;
66 u64 group_depth;
67 unsigned group_count;
68
69 enum exofs_inode_layout_gen_functions lay_func;
70
71 unsigned s_numdevs; /* Num of devices in array */
72 struct osd_dev *s_ods[0]; /* Variable length */
73};
74
54/* 75/*
55 * our extension to the in-memory superblock 76 * our extension to the in-memory superblock
56 */ 77 */
57struct exofs_sb_info { 78struct exofs_sb_info {
58 struct osd_dev *s_dev; /* returned by get_osd_dev */ 79 struct exofs_fscb s_fscb; /* Written often, pre-allocate*/
59 osd_id s_pid; /* partition ID of file system*/
60 int s_timeout; /* timeout for OSD operations */ 80 int s_timeout; /* timeout for OSD operations */
61 uint64_t s_nextid; /* highest object ID used */ 81 uint64_t s_nextid; /* highest object ID used */
62 uint32_t s_numfiles; /* number of files on fs */ 82 uint32_t s_numfiles; /* number of files on fs */
63 spinlock_t s_next_gen_lock; /* spinlock for gen # update */ 83 spinlock_t s_next_gen_lock; /* spinlock for gen # update */
64 u32 s_next_generation; /* next gen # to use */ 84 u32 s_next_generation; /* next gen # to use */
65 atomic_t s_curr_pending; /* number of pending commands */ 85 atomic_t s_curr_pending; /* number of pending commands */
66 uint8_t s_cred[OSD_CAP_LEN]; /* all-powerful credential */ 86 uint8_t s_cred[OSD_CAP_LEN]; /* credential for the fscb */
87
88 struct pnfs_osd_data_map data_map; /* Default raid to use
89 * FIXME: Needed ?
90 */
91/* struct exofs_layout dir_layout;*/ /* Default dir layout */
92 struct exofs_layout layout; /* Default files layout,
93 * contains the variable osd_dev
94 * array. Keep last */
95 struct osd_dev *_min_one_dev[1]; /* Place holder for one dev */
67}; 96};
68 97
69/* 98/*
70 * our extension to the in-memory inode 99 * our extension to the in-memory inode
71 */ 100 */
72struct exofs_i_info { 101struct exofs_i_info {
102 struct inode vfs_inode; /* normal in-memory inode */
103 wait_queue_head_t i_wq; /* wait queue for inode */
73 unsigned long i_flags; /* various atomic flags */ 104 unsigned long i_flags; /* various atomic flags */
74 uint32_t i_data[EXOFS_IDATA];/*short symlink names and device #s*/ 105 uint32_t i_data[EXOFS_IDATA];/*short symlink names and device #s*/
75 uint32_t i_dir_start_lookup; /* which page to start lookup */ 106 uint32_t i_dir_start_lookup; /* which page to start lookup */
76 wait_queue_head_t i_wq; /* wait queue for inode */
77 uint64_t i_commit_size; /* the object's written length */ 107 uint64_t i_commit_size; /* the object's written length */
78 uint8_t i_cred[OSD_CAP_LEN];/* all-powerful credential */ 108 uint8_t i_cred[OSD_CAP_LEN];/* all-powerful credential */
79 struct inode vfs_inode; /* normal in-memory inode */
80}; 109};
81 110
111static inline osd_id exofs_oi_objno(struct exofs_i_info *oi)
112{
113 return oi->vfs_inode.i_ino + EXOFS_OBJ_OFF;
114}
115
116struct exofs_io_state;
117typedef void (*exofs_io_done_fn)(struct exofs_io_state *or, void *private);
118
119struct exofs_io_state {
120 struct kref kref;
121
122 void *private;
123 exofs_io_done_fn done;
124
125 struct exofs_layout *layout;
126 struct osd_obj_id obj;
127 u8 *cred;
128
129 /* Global read/write IO*/
130 loff_t offset;
131 unsigned long length;
132 void *kern_buff;
133
134 struct page **pages;
135 unsigned nr_pages;
136 unsigned pgbase;
137 unsigned pages_consumed;
138
139 /* Attributes */
140 unsigned in_attr_len;
141 struct osd_attr *in_attr;
142 unsigned out_attr_len;
143 struct osd_attr *out_attr;
144
145 /* Variable array of size numdevs */
146 unsigned numdevs;
147 struct exofs_per_dev_state {
148 struct osd_request *or;
149 struct bio *bio;
150 loff_t offset;
151 unsigned length;
152 unsigned dev;
153 } per_dev[];
154};
155
156static inline unsigned exofs_io_state_size(unsigned numdevs)
157{
158 return sizeof(struct exofs_io_state) +
159 sizeof(struct exofs_per_dev_state) * numdevs;
160}
161
82/* 162/*
83 * our inode flags 163 * our inode flags
84 */ 164 */
@@ -123,6 +203,12 @@ static inline struct exofs_i_info *exofs_i(struct inode *inode)
123} 203}
124 204
125/* 205/*
206 * Given a layout, object_number and stripe_index return the associated global
207 * dev_index
208 */
209unsigned exofs_layout_od_id(struct exofs_layout *layout,
210 osd_id obj_no, unsigned layout_index);
211/*
126 * Maximum count of links to a file 212 * Maximum count of links to a file
127 */ 213 */
128#define EXOFS_LINK_MAX 32000 214#define EXOFS_LINK_MAX 32000
@@ -130,6 +216,43 @@ static inline struct exofs_i_info *exofs_i(struct inode *inode)
130/************************* 216/*************************
131 * function declarations * 217 * function declarations *
132 *************************/ 218 *************************/
219
220/* ios.c */
221void exofs_make_credential(u8 cred_a[OSD_CAP_LEN],
222 const struct osd_obj_id *obj);
223int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
224 u64 offset, void *p, unsigned length);
225
226int exofs_get_io_state(struct exofs_layout *layout,
227 struct exofs_io_state **ios);
228void exofs_put_io_state(struct exofs_io_state *ios);
229
230int exofs_check_io(struct exofs_io_state *ios, u64 *resid);
231
232int exofs_sbi_create(struct exofs_io_state *ios);
233int exofs_sbi_remove(struct exofs_io_state *ios);
234int exofs_sbi_write(struct exofs_io_state *ios);
235int exofs_sbi_read(struct exofs_io_state *ios);
236
237int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr);
238
239int exofs_oi_truncate(struct exofs_i_info *oi, u64 new_len);
240static inline int exofs_oi_write(struct exofs_i_info *oi,
241 struct exofs_io_state *ios)
242{
243 ios->obj.id = exofs_oi_objno(oi);
244 ios->cred = oi->i_cred;
245 return exofs_sbi_write(ios);
246}
247
248static inline int exofs_oi_read(struct exofs_i_info *oi,
249 struct exofs_io_state *ios)
250{
251 ios->obj.id = exofs_oi_objno(oi);
252 ios->cred = oi->i_cred;
253 return exofs_sbi_read(ios);
254}
255
133/* inode.c */ 256/* inode.c */
134void exofs_truncate(struct inode *inode); 257void exofs_truncate(struct inode *inode);
135int exofs_setattr(struct dentry *, struct iattr *); 258int exofs_setattr(struct dentry *, struct iattr *);
@@ -138,7 +261,7 @@ int exofs_write_begin(struct file *file, struct address_space *mapping,
138 struct page **pagep, void **fsdata); 261 struct page **pagep, void **fsdata);
139extern struct inode *exofs_iget(struct super_block *, unsigned long); 262extern struct inode *exofs_iget(struct super_block *, unsigned long);
140struct inode *exofs_new_inode(struct inode *, int); 263struct inode *exofs_new_inode(struct inode *, int);
141extern int exofs_write_inode(struct inode *, int); 264extern int exofs_write_inode(struct inode *, struct writeback_control *wbc);
142extern void exofs_delete_inode(struct inode *); 265extern void exofs_delete_inode(struct inode *);
143 266
144/* dir.c: */ 267/* dir.c: */
@@ -169,6 +292,7 @@ extern const struct file_operations exofs_file_operations;
169 292
170/* inode.c */ 293/* inode.c */
171extern const struct address_space_operations exofs_aops; 294extern const struct address_space_operations exofs_aops;
295extern const struct osd_attr g_attr_logical_length;
172 296
173/* namei.c */ 297/* namei.c */
174extern const struct inode_operations exofs_dir_inode_operations; 298extern const struct inode_operations exofs_dir_inode_operations;
diff --git a/fs/exofs/inode.c b/fs/exofs/inode.c
index 6c10f7476699..76d2a79ef93e 100644
--- a/fs/exofs/inode.c
+++ b/fs/exofs/inode.c
@@ -31,94 +31,117 @@
31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA 31 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
32 */ 32 */
33 33
34#include <linux/slab.h>
34#include <linux/writeback.h> 35#include <linux/writeback.h>
35#include <linux/buffer_head.h> 36#include <linux/buffer_head.h>
36#include <scsi/scsi_device.h> 37#include <scsi/scsi_device.h>
37 38
38#include "exofs.h" 39#include "exofs.h"
39 40
40#ifdef CONFIG_EXOFS_DEBUG 41#define EXOFS_DBGMSG2(M...) do {} while (0)
41# define EXOFS_DEBUG_OBJ_ISIZE 1 42
42#endif 43enum { BIO_MAX_PAGES_KMALLOC =
44 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
45 MAX_PAGES_KMALLOC =
46 PAGE_SIZE / sizeof(struct page *),
47};
43 48
44struct page_collect { 49struct page_collect {
45 struct exofs_sb_info *sbi; 50 struct exofs_sb_info *sbi;
46 struct request_queue *req_q;
47 struct inode *inode; 51 struct inode *inode;
48 unsigned expected_pages; 52 unsigned expected_pages;
53 struct exofs_io_state *ios;
49 54
50 struct bio *bio; 55 struct page **pages;
56 unsigned alloc_pages;
51 unsigned nr_pages; 57 unsigned nr_pages;
52 unsigned long length; 58 unsigned long length;
53 loff_t pg_first; /* keep 64bit also in 32-arches */ 59 loff_t pg_first; /* keep 64bit also in 32-arches */
54}; 60};
55 61
56static void _pcol_init(struct page_collect *pcol, unsigned expected_pages, 62static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
57 struct inode *inode) 63 struct inode *inode)
58{ 64{
59 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; 65 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
60 66
61 pcol->sbi = sbi; 67 pcol->sbi = sbi;
62 pcol->req_q = osd_request_queue(sbi->s_dev);
63 pcol->inode = inode; 68 pcol->inode = inode;
64 pcol->expected_pages = expected_pages; 69 pcol->expected_pages = expected_pages;
65 70
66 pcol->bio = NULL; 71 pcol->ios = NULL;
72 pcol->pages = NULL;
73 pcol->alloc_pages = 0;
67 pcol->nr_pages = 0; 74 pcol->nr_pages = 0;
68 pcol->length = 0; 75 pcol->length = 0;
69 pcol->pg_first = -1; 76 pcol->pg_first = -1;
70
71 EXOFS_DBGMSG("_pcol_init ino=0x%lx expected_pages=%u\n", inode->i_ino,
72 expected_pages);
73} 77}
74 78
75static void _pcol_reset(struct page_collect *pcol) 79static void _pcol_reset(struct page_collect *pcol)
76{ 80{
77 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages); 81 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
78 82
79 pcol->bio = NULL; 83 pcol->pages = NULL;
84 pcol->alloc_pages = 0;
80 pcol->nr_pages = 0; 85 pcol->nr_pages = 0;
81 pcol->length = 0; 86 pcol->length = 0;
82 pcol->pg_first = -1; 87 pcol->pg_first = -1;
83 EXOFS_DBGMSG("_pcol_reset ino=0x%lx expected_pages=%u\n", 88 pcol->ios = NULL;
84 pcol->inode->i_ino, pcol->expected_pages);
85 89
86 /* this is probably the end of the loop but in writes 90 /* this is probably the end of the loop but in writes
87 * it might not end here. don't be left with nothing 91 * it might not end here. don't be left with nothing
88 */ 92 */
89 if (!pcol->expected_pages) 93 if (!pcol->expected_pages)
90 pcol->expected_pages = 128; 94 pcol->expected_pages = MAX_PAGES_KMALLOC;
91} 95}
92 96
93static int pcol_try_alloc(struct page_collect *pcol) 97static int pcol_try_alloc(struct page_collect *pcol)
94{ 98{
95 int pages = min_t(unsigned, pcol->expected_pages, BIO_MAX_PAGES); 99 unsigned pages = min_t(unsigned, pcol->expected_pages,
100 MAX_PAGES_KMALLOC);
101
102 if (!pcol->ios) { /* First time allocate io_state */
103 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
104
105 if (ret)
106 return ret;
107 }
108
109 /* TODO: easily support bio chaining */
110 pages = min_t(unsigned, pages,
111 pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
96 112
97 for (; pages; pages >>= 1) { 113 for (; pages; pages >>= 1) {
98 pcol->bio = bio_alloc(GFP_KERNEL, pages); 114 pcol->pages = kmalloc(pages * sizeof(struct page *),
99 if (likely(pcol->bio)) 115 GFP_KERNEL);
116 if (likely(pcol->pages)) {
117 pcol->alloc_pages = pages;
100 return 0; 118 return 0;
119 }
101 } 120 }
102 121
103 EXOFS_ERR("Failed to kcalloc expected_pages=%u\n", 122 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
104 pcol->expected_pages); 123 pcol->expected_pages);
105 return -ENOMEM; 124 return -ENOMEM;
106} 125}
107 126
108static void pcol_free(struct page_collect *pcol) 127static void pcol_free(struct page_collect *pcol)
109{ 128{
110 bio_put(pcol->bio); 129 kfree(pcol->pages);
111 pcol->bio = NULL; 130 pcol->pages = NULL;
131
132 if (pcol->ios) {
133 exofs_put_io_state(pcol->ios);
134 pcol->ios = NULL;
135 }
112} 136}
113 137
114static int pcol_add_page(struct page_collect *pcol, struct page *page, 138static int pcol_add_page(struct page_collect *pcol, struct page *page,
115 unsigned len) 139 unsigned len)
116{ 140{
117 int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0); 141 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
118 if (unlikely(len != added_len))
119 return -ENOMEM; 142 return -ENOMEM;
120 143
121 ++pcol->nr_pages; 144 pcol->pages[pcol->nr_pages++] = page;
122 pcol->length += len; 145 pcol->length += len;
123 return 0; 146 return 0;
124} 147}
@@ -161,32 +184,26 @@ static void update_write_page(struct page *page, int ret)
161/* Called at the end of reads, to optionally unlock pages and update their 184/* Called at the end of reads, to optionally unlock pages and update their
162 * status. 185 * status.
163 */ 186 */
164static int __readpages_done(struct osd_request *or, struct page_collect *pcol, 187static int __readpages_done(struct page_collect *pcol, bool do_unlock)
165 bool do_unlock)
166{ 188{
167 struct bio_vec *bvec;
168 int i; 189 int i;
169 u64 resid; 190 u64 resid;
170 u64 good_bytes; 191 u64 good_bytes;
171 u64 length = 0; 192 u64 length = 0;
172 int ret = exofs_check_ok_resid(or, &resid, NULL); 193 int ret = exofs_check_io(pcol->ios, &resid);
173
174 osd_end_request(or);
175 194
176 if (likely(!ret)) 195 if (likely(!ret))
177 good_bytes = pcol->length; 196 good_bytes = pcol->length;
178 else if (!resid)
179 good_bytes = 0;
180 else 197 else
181 good_bytes = pcol->length - resid; 198 good_bytes = pcol->length - resid;
182 199
183 EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx" 200 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
184 " length=0x%lx nr_pages=%u\n", 201 " length=0x%lx nr_pages=%u\n",
185 pcol->inode->i_ino, _LLU(good_bytes), pcol->length, 202 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
186 pcol->nr_pages); 203 pcol->nr_pages);
187 204
188 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 205 for (i = 0; i < pcol->nr_pages; i++) {
189 struct page *page = bvec->bv_page; 206 struct page *page = pcol->pages[i];
190 struct inode *inode = page->mapping->host; 207 struct inode *inode = page->mapping->host;
191 int page_stat; 208 int page_stat;
192 209
@@ -198,38 +215,37 @@ static int __readpages_done(struct osd_request *or, struct page_collect *pcol,
198 else 215 else
199 page_stat = ret; 216 page_stat = ret;
200 217
201 EXOFS_DBGMSG(" readpages_done(0x%lx, 0x%lx) %s\n", 218 EXOFS_DBGMSG2(" readpages_done(0x%lx, 0x%lx) %s\n",
202 inode->i_ino, page->index, 219 inode->i_ino, page->index,
203 page_stat ? "bad_bytes" : "good_bytes"); 220 page_stat ? "bad_bytes" : "good_bytes");
204 221
205 ret = update_read_page(page, page_stat); 222 ret = update_read_page(page, page_stat);
206 if (do_unlock) 223 if (do_unlock)
207 unlock_page(page); 224 unlock_page(page);
208 length += bvec->bv_len; 225 length += PAGE_SIZE;
209 } 226 }
210 227
211 pcol_free(pcol); 228 pcol_free(pcol);
212 EXOFS_DBGMSG("readpages_done END\n"); 229 EXOFS_DBGMSG2("readpages_done END\n");
213 return ret; 230 return ret;
214} 231}
215 232
216/* callback of async reads */ 233/* callback of async reads */
217static void readpages_done(struct osd_request *or, void *p) 234static void readpages_done(struct exofs_io_state *ios, void *p)
218{ 235{
219 struct page_collect *pcol = p; 236 struct page_collect *pcol = p;
220 237
221 __readpages_done(or, pcol, true); 238 __readpages_done(pcol, true);
222 atomic_dec(&pcol->sbi->s_curr_pending); 239 atomic_dec(&pcol->sbi->s_curr_pending);
223 kfree(p); 240 kfree(pcol);
224} 241}
225 242
226static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw) 243static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
227{ 244{
228 struct bio_vec *bvec;
229 int i; 245 int i;
230 246
231 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 247 for (i = 0; i < pcol->nr_pages; i++) {
232 struct page *page = bvec->bv_page; 248 struct page *page = pcol->pages[i];
233 249
234 if (rw == READ) 250 if (rw == READ)
235 update_read_page(page, ret); 251 update_read_page(page, ret);
@@ -238,36 +254,29 @@ static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
238 254
239 unlock_page(page); 255 unlock_page(page);
240 } 256 }
241 pcol_free(pcol);
242} 257}
243 258
244static int read_exec(struct page_collect *pcol, bool is_sync) 259static int read_exec(struct page_collect *pcol, bool is_sync)
245{ 260{
246 struct exofs_i_info *oi = exofs_i(pcol->inode); 261 struct exofs_i_info *oi = exofs_i(pcol->inode);
247 struct osd_obj_id obj = {pcol->sbi->s_pid, 262 struct exofs_io_state *ios = pcol->ios;
248 pcol->inode->i_ino + EXOFS_OBJ_OFF};
249 struct osd_request *or = NULL;
250 struct page_collect *pcol_copy = NULL; 263 struct page_collect *pcol_copy = NULL;
251 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
252 int ret; 264 int ret;
253 265
254 if (!pcol->bio) 266 if (!pcol->pages)
255 return 0; 267 return 0;
256 268
257 /* see comment in _readpage() about sync reads */ 269 /* see comment in _readpage() about sync reads */
258 WARN_ON(is_sync && (pcol->nr_pages != 1)); 270 WARN_ON(is_sync && (pcol->nr_pages != 1));
259 271
260 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL); 272 ios->pages = pcol->pages;
261 if (unlikely(!or)) { 273 ios->nr_pages = pcol->nr_pages;
262 ret = -ENOMEM; 274 ios->length = pcol->length;
263 goto err; 275 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
264 }
265
266 osd_req_read(or, &obj, i_start, pcol->bio, pcol->length);
267 276
268 if (is_sync) { 277 if (is_sync) {
269 exofs_sync_op(or, pcol->sbi->s_timeout, oi->i_cred); 278 exofs_oi_read(oi, pcol->ios);
270 return __readpages_done(or, pcol, false); 279 return __readpages_done(pcol, false);
271 } 280 }
272 281
273 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); 282 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
@@ -277,14 +286,16 @@ static int read_exec(struct page_collect *pcol, bool is_sync)
277 } 286 }
278 287
279 *pcol_copy = *pcol; 288 *pcol_copy = *pcol;
280 ret = exofs_async_op(or, readpages_done, pcol_copy, oi->i_cred); 289 ios->done = readpages_done;
290 ios->private = pcol_copy;
291 ret = exofs_oi_read(oi, ios);
281 if (unlikely(ret)) 292 if (unlikely(ret))
282 goto err; 293 goto err;
283 294
284 atomic_inc(&pcol->sbi->s_curr_pending); 295 atomic_inc(&pcol->sbi->s_curr_pending);
285 296
286 EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n", 297 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
287 obj.id, _LLU(i_start), pcol->length); 298 ios->obj.id, _LLU(ios->offset), pcol->length);
288 299
289 /* pages ownership was passed to pcol_copy */ 300 /* pages ownership was passed to pcol_copy */
290 _pcol_reset(pcol); 301 _pcol_reset(pcol);
@@ -293,12 +304,10 @@ static int read_exec(struct page_collect *pcol, bool is_sync)
293err: 304err:
294 if (!is_sync) 305 if (!is_sync)
295 _unlock_pcol_pages(pcol, ret, READ); 306 _unlock_pcol_pages(pcol, ret, READ);
296 else /* Pages unlocked by caller in sync mode only free bio */ 307
297 pcol_free(pcol); 308 pcol_free(pcol);
298 309
299 kfree(pcol_copy); 310 kfree(pcol_copy);
300 if (or)
301 osd_end_request(or);
302 return ret; 311 return ret;
303} 312}
304 313
@@ -361,7 +370,7 @@ try_again:
361 goto try_again; 370 goto try_again;
362 } 371 }
363 372
364 if (!pcol->bio) { 373 if (!pcol->pages) {
365 ret = pcol_try_alloc(pcol); 374 ret = pcol_try_alloc(pcol);
366 if (unlikely(ret)) 375 if (unlikely(ret))
367 goto fail; 376 goto fail;
@@ -370,12 +379,12 @@ try_again:
370 if (len != PAGE_CACHE_SIZE) 379 if (len != PAGE_CACHE_SIZE)
371 zero_user(page, len, PAGE_CACHE_SIZE - len); 380 zero_user(page, len, PAGE_CACHE_SIZE - len);
372 381
373 EXOFS_DBGMSG(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n", 382 EXOFS_DBGMSG2(" readpage_strip(0x%lx, 0x%lx) len=0x%zx\n",
374 inode->i_ino, page->index, len); 383 inode->i_ino, page->index, len);
375 384
376 ret = pcol_add_page(pcol, page, len); 385 ret = pcol_add_page(pcol, page, len);
377 if (ret) { 386 if (ret) {
378 EXOFS_DBGMSG("Failed pcol_add_page pages[i]=%p " 387 EXOFS_DBGMSG2("Failed pcol_add_page pages[i]=%p "
379 "this_len=0x%zx nr_pages=%u length=0x%lx\n", 388 "this_len=0x%zx nr_pages=%u length=0x%lx\n",
380 page, len, pcol->nr_pages, pcol->length); 389 page, len, pcol->nr_pages, pcol->length);
381 390
@@ -419,9 +428,8 @@ static int _readpage(struct page *page, bool is_sync)
419 428
420 _pcol_init(&pcol, 1, page->mapping->host); 429 _pcol_init(&pcol, 1, page->mapping->host);
421 430
422 /* readpage_strip might call read_exec(,async) inside at several places 431 /* readpage_strip might call read_exec(,is_sync==false) at several
423 * but this is safe for is_async=0 since read_exec will not do anything 432 * places but not if we have a single page.
424 * when we have a single page.
425 */ 433 */
426 ret = readpage_strip(&pcol, page); 434 ret = readpage_strip(&pcol, page);
427 if (ret) { 435 if (ret) {
@@ -440,35 +448,30 @@ static int exofs_readpage(struct file *file, struct page *page)
440 return _readpage(page, false); 448 return _readpage(page, false);
441} 449}
442 450
443/* Callback for osd_write. All writes are asynchronouse */ 451/* Callback for osd_write. All writes are asynchronous */
444static void writepages_done(struct osd_request *or, void *p) 452static void writepages_done(struct exofs_io_state *ios, void *p)
445{ 453{
446 struct page_collect *pcol = p; 454 struct page_collect *pcol = p;
447 struct bio_vec *bvec;
448 int i; 455 int i;
449 u64 resid; 456 u64 resid;
450 u64 good_bytes; 457 u64 good_bytes;
451 u64 length = 0; 458 u64 length = 0;
459 int ret = exofs_check_io(ios, &resid);
452 460
453 int ret = exofs_check_ok_resid(or, NULL, &resid);
454
455 osd_end_request(or);
456 atomic_dec(&pcol->sbi->s_curr_pending); 461 atomic_dec(&pcol->sbi->s_curr_pending);
457 462
458 if (likely(!ret)) 463 if (likely(!ret))
459 good_bytes = pcol->length; 464 good_bytes = pcol->length;
460 else if (!resid)
461 good_bytes = 0;
462 else 465 else
463 good_bytes = pcol->length - resid; 466 good_bytes = pcol->length - resid;
464 467
465 EXOFS_DBGMSG("writepages_done(0x%lx) good_bytes=0x%llx" 468 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
466 " length=0x%lx nr_pages=%u\n", 469 " length=0x%lx nr_pages=%u\n",
467 pcol->inode->i_ino, _LLU(good_bytes), pcol->length, 470 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
468 pcol->nr_pages); 471 pcol->nr_pages);
469 472
470 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 473 for (i = 0; i < pcol->nr_pages; i++) {
471 struct page *page = bvec->bv_page; 474 struct page *page = pcol->pages[i];
472 struct inode *inode = page->mapping->host; 475 struct inode *inode = page->mapping->host;
473 int page_stat; 476 int page_stat;
474 477
@@ -482,37 +485,27 @@ static void writepages_done(struct osd_request *or, void *p)
482 485
483 update_write_page(page, page_stat); 486 update_write_page(page, page_stat);
484 unlock_page(page); 487 unlock_page(page);
485 EXOFS_DBGMSG(" writepages_done(0x%lx, 0x%lx) status=%d\n", 488 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
486 inode->i_ino, page->index, page_stat); 489 inode->i_ino, page->index, page_stat);
487 490
488 length += bvec->bv_len; 491 length += PAGE_SIZE;
489 } 492 }
490 493
491 pcol_free(pcol); 494 pcol_free(pcol);
492 kfree(pcol); 495 kfree(pcol);
493 EXOFS_DBGMSG("writepages_done END\n"); 496 EXOFS_DBGMSG2("writepages_done END\n");
494} 497}
495 498
496static int write_exec(struct page_collect *pcol) 499static int write_exec(struct page_collect *pcol)
497{ 500{
498 struct exofs_i_info *oi = exofs_i(pcol->inode); 501 struct exofs_i_info *oi = exofs_i(pcol->inode);
499 struct osd_obj_id obj = {pcol->sbi->s_pid, 502 struct exofs_io_state *ios = pcol->ios;
500 pcol->inode->i_ino + EXOFS_OBJ_OFF};
501 struct osd_request *or = NULL;
502 struct page_collect *pcol_copy = NULL; 503 struct page_collect *pcol_copy = NULL;
503 loff_t i_start = pcol->pg_first << PAGE_CACHE_SHIFT;
504 int ret; 504 int ret;
505 505
506 if (!pcol->bio) 506 if (!pcol->pages)
507 return 0; 507 return 0;
508 508
509 or = osd_start_request(pcol->sbi->s_dev, GFP_KERNEL);
510 if (unlikely(!or)) {
511 EXOFS_ERR("write_exec: Faild to osd_start_request()\n");
512 ret = -ENOMEM;
513 goto err;
514 }
515
516 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); 509 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
517 if (!pcol_copy) { 510 if (!pcol_copy) {
518 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n"); 511 EXOFS_ERR("write_exec: Faild to kmalloc(pcol)\n");
@@ -522,17 +515,22 @@ static int write_exec(struct page_collect *pcol)
522 515
523 *pcol_copy = *pcol; 516 *pcol_copy = *pcol;
524 517
525 pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */ 518 ios->pages = pcol_copy->pages;
526 osd_req_write(or, &obj, i_start, pcol_copy->bio, pcol_copy->length); 519 ios->nr_pages = pcol_copy->nr_pages;
527 ret = exofs_async_op(or, writepages_done, pcol_copy, oi->i_cred); 520 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
521 ios->length = pcol_copy->length;
522 ios->done = writepages_done;
523 ios->private = pcol_copy;
524
525 ret = exofs_oi_write(oi, ios);
528 if (unlikely(ret)) { 526 if (unlikely(ret)) {
529 EXOFS_ERR("write_exec: exofs_async_op() Faild\n"); 527 EXOFS_ERR("write_exec: exofs_oi_write() Faild\n");
530 goto err; 528 goto err;
531 } 529 }
532 530
533 atomic_inc(&pcol->sbi->s_curr_pending); 531 atomic_inc(&pcol->sbi->s_curr_pending);
534 EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n", 532 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
535 pcol->inode->i_ino, pcol->pg_first, _LLU(i_start), 533 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
536 pcol->length); 534 pcol->length);
537 /* pages ownership was passed to pcol_copy */ 535 /* pages ownership was passed to pcol_copy */
538 _pcol_reset(pcol); 536 _pcol_reset(pcol);
@@ -540,9 +538,9 @@ static int write_exec(struct page_collect *pcol)
540 538
541err: 539err:
542 _unlock_pcol_pages(pcol, ret, WRITE); 540 _unlock_pcol_pages(pcol, ret, WRITE);
541 pcol_free(pcol);
543 kfree(pcol_copy); 542 kfree(pcol_copy);
544 if (or) 543
545 osd_end_request(or);
546 return ret; 544 return ret;
547} 545}
548 546
@@ -586,6 +584,9 @@ static int writepage_strip(struct page *page,
586 if (PageError(page)) 584 if (PageError(page))
587 ClearPageError(page); 585 ClearPageError(page);
588 unlock_page(page); 586 unlock_page(page);
587 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) "
588 "outside the limits\n",
589 inode->i_ino, page->index);
589 return 0; 590 return 0;
590 } 591 }
591 } 592 }
@@ -600,21 +601,24 @@ try_again:
600 ret = write_exec(pcol); 601 ret = write_exec(pcol);
601 if (unlikely(ret)) 602 if (unlikely(ret))
602 goto fail; 603 goto fail;
604
605 EXOFS_DBGMSG("writepage_strip(0x%lx, 0x%lx) Discontinuity\n",
606 inode->i_ino, page->index);
603 goto try_again; 607 goto try_again;
604 } 608 }
605 609
606 if (!pcol->bio) { 610 if (!pcol->pages) {
607 ret = pcol_try_alloc(pcol); 611 ret = pcol_try_alloc(pcol);
608 if (unlikely(ret)) 612 if (unlikely(ret))
609 goto fail; 613 goto fail;
610 } 614 }
611 615
612 EXOFS_DBGMSG(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n", 616 EXOFS_DBGMSG2(" writepage_strip(0x%lx, 0x%lx) len=0x%zx\n",
613 inode->i_ino, page->index, len); 617 inode->i_ino, page->index, len);
614 618
615 ret = pcol_add_page(pcol, page, len); 619 ret = pcol_add_page(pcol, page, len);
616 if (unlikely(ret)) { 620 if (unlikely(ret)) {
617 EXOFS_DBGMSG("Failed pcol_add_page " 621 EXOFS_DBGMSG2("Failed pcol_add_page "
618 "nr_pages=%u total_length=0x%lx\n", 622 "nr_pages=%u total_length=0x%lx\n",
619 pcol->nr_pages, pcol->length); 623 pcol->nr_pages, pcol->length);
620 624
@@ -634,6 +638,8 @@ try_again:
634 return 0; 638 return 0;
635 639
636fail: 640fail:
641 EXOFS_DBGMSG("Error: writepage_strip(0x%lx, 0x%lx)=>%d\n",
642 inode->i_ino, page->index, ret);
637 set_bit(AS_EIO, &page->mapping->flags); 643 set_bit(AS_EIO, &page->mapping->flags);
638 unlock_page(page); 644 unlock_page(page);
639 return ret; 645 return ret;
@@ -652,14 +658,17 @@ static int exofs_writepages(struct address_space *mapping,
652 wbc->range_end >> PAGE_CACHE_SHIFT; 658 wbc->range_end >> PAGE_CACHE_SHIFT;
653 659
654 if (start || end) 660 if (start || end)
655 expected_pages = min(end - start + 1, 32L); 661 expected_pages = end - start + 1;
656 else 662 else
657 expected_pages = mapping->nrpages; 663 expected_pages = mapping->nrpages;
658 664
659 EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx" 665 if (expected_pages < 32L)
660 " m->nrpages=%lu start=0x%lx end=0x%lx\n", 666 expected_pages = 32L;
667
668 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
669 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
661 mapping->host->i_ino, wbc->range_start, wbc->range_end, 670 mapping->host->i_ino, wbc->range_start, wbc->range_end,
662 mapping->nrpages, start, end); 671 mapping->nrpages, start, end, expected_pages);
663 672
664 _pcol_init(&pcol, expected_pages, mapping->host); 673 _pcol_init(&pcol, expected_pages, mapping->host);
665 674
@@ -731,13 +740,28 @@ static int exofs_write_begin_export(struct file *file,
731 fsdata); 740 fsdata);
732} 741}
733 742
743static int exofs_write_end(struct file *file, struct address_space *mapping,
744 loff_t pos, unsigned len, unsigned copied,
745 struct page *page, void *fsdata)
746{
747 struct inode *inode = mapping->host;
748 /* According to comment in simple_write_end i_mutex is held */
749 loff_t i_size = inode->i_size;
750 int ret;
751
752 ret = simple_write_end(file, mapping,pos, len, copied, page, fsdata);
753 if (i_size != inode->i_size)
754 mark_inode_dirty(inode);
755 return ret;
756}
757
734const struct address_space_operations exofs_aops = { 758const struct address_space_operations exofs_aops = {
735 .readpage = exofs_readpage, 759 .readpage = exofs_readpage,
736 .readpages = exofs_readpages, 760 .readpages = exofs_readpages,
737 .writepage = exofs_writepage, 761 .writepage = exofs_writepage,
738 .writepages = exofs_writepages, 762 .writepages = exofs_writepages,
739 .write_begin = exofs_write_begin_export, 763 .write_begin = exofs_write_begin_export,
740 .write_end = simple_write_end, 764 .write_end = exofs_write_end,
741}; 765};
742 766
743/****************************************************************************** 767/******************************************************************************
@@ -771,19 +795,28 @@ static int exofs_get_block(struct inode *inode, sector_t iblock,
771const struct osd_attr g_attr_logical_length = ATTR_DEF( 795const struct osd_attr g_attr_logical_length = ATTR_DEF(
772 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8); 796 OSD_APAGE_OBJECT_INFORMATION, OSD_ATTR_OI_LOGICAL_LENGTH, 8);
773 797
798static int _do_truncate(struct inode *inode)
799{
800 struct exofs_i_info *oi = exofs_i(inode);
801 loff_t isize = i_size_read(inode);
802 int ret;
803
804 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
805
806 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
807
808 ret = exofs_oi_truncate(oi, (u64)isize);
809 EXOFS_DBGMSG("(0x%lx) size=0x%llx\n", inode->i_ino, isize);
810 return ret;
811}
812
774/* 813/*
775 * Truncate a file to the specified size - all we have to do is set the size 814 * Truncate a file to the specified size - all we have to do is set the size
776 * attribute. We make sure the object exists first. 815 * attribute. We make sure the object exists first.
777 */ 816 */
778void exofs_truncate(struct inode *inode) 817void exofs_truncate(struct inode *inode)
779{ 818{
780 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
781 struct exofs_i_info *oi = exofs_i(inode); 819 struct exofs_i_info *oi = exofs_i(inode);
782 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF};
783 struct osd_request *or;
784 struct osd_attr attr;
785 loff_t isize = i_size_read(inode);
786 __be64 newsize;
787 int ret; 820 int ret;
788 821
789 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode) 822 if (!(S_ISREG(inode->i_mode) || S_ISDIR(inode->i_mode)
@@ -793,22 +826,6 @@ void exofs_truncate(struct inode *inode)
793 return; 826 return;
794 if (IS_APPEND(inode) || IS_IMMUTABLE(inode)) 827 if (IS_APPEND(inode) || IS_IMMUTABLE(inode))
795 return; 828 return;
796 inode->i_mtime = inode->i_ctime = CURRENT_TIME;
797
798 nobh_truncate_page(inode->i_mapping, isize, exofs_get_block);
799
800 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
801 if (unlikely(!or)) {
802 EXOFS_ERR("ERROR: exofs_truncate: osd_start_request failed\n");
803 goto fail;
804 }
805
806 osd_req_set_attributes(or, &obj);
807
808 newsize = cpu_to_be64((u64)isize);
809 attr = g_attr_logical_length;
810 attr.val_ptr = &newsize;
811 osd_req_add_set_attr_list(or, &attr, 1);
812 829
813 /* if we are about to truncate an object, and it hasn't been 830 /* if we are about to truncate an object, and it hasn't been
814 * created yet, wait 831 * created yet, wait
@@ -816,8 +833,7 @@ void exofs_truncate(struct inode *inode)
816 if (unlikely(wait_obj_created(oi))) 833 if (unlikely(wait_obj_created(oi)))
817 goto fail; 834 goto fail;
818 835
819 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred); 836 ret = _do_truncate(inode);
820 osd_end_request(or);
821 if (ret) 837 if (ret)
822 goto fail; 838 goto fail;
823 839
@@ -845,67 +861,110 @@ int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
845 return error; 861 return error;
846} 862}
847 863
864static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
865 EXOFS_APAGE_FS_DATA,
866 EXOFS_ATTR_INODE_FILE_LAYOUT,
867 0);
868static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
869 EXOFS_APAGE_FS_DATA,
870 EXOFS_ATTR_INODE_DIR_LAYOUT,
871 0);
872
848/* 873/*
849 * Read an inode from the OSD, and return it as is. We also return the size 874 * Read the Linux inode info from the OSD, and return it as is. In exofs the
850 * attribute in the 'sanity' argument if we got compiled with debugging turned 875 * inode info is in an application specific page/attribute of the osd-object.
851 * on.
852 */ 876 */
853static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi, 877static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
854 struct exofs_fcb *inode, uint64_t *sanity) 878 struct exofs_fcb *inode)
855{ 879{
856 struct exofs_sb_info *sbi = sb->s_fs_info; 880 struct exofs_sb_info *sbi = sb->s_fs_info;
857 struct osd_request *or; 881 struct osd_attr attrs[] = {
858 struct osd_attr attr; 882 [0] = g_attr_inode_data,
859 struct osd_obj_id obj = {sbi->s_pid, 883 [1] = g_attr_inode_file_layout,
860 oi->vfs_inode.i_ino + EXOFS_OBJ_OFF}; 884 [2] = g_attr_inode_dir_layout,
885 };
886 struct exofs_io_state *ios;
887 struct exofs_on_disk_inode_layout *layout;
861 int ret; 888 int ret;
862 889
863 exofs_make_credential(oi->i_cred, &obj); 890 ret = exofs_get_io_state(&sbi->layout, &ios);
864 891 if (unlikely(ret)) {
865 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 892 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
866 if (unlikely(!or)) { 893 return ret;
867 EXOFS_ERR("exofs_get_inode: osd_start_request failed.\n");
868 return -ENOMEM;
869 } 894 }
870 osd_req_get_attributes(or, &obj);
871 895
872 /* we need the inode attribute */ 896 ios->obj.id = exofs_oi_objno(oi);
873 osd_req_add_get_attr_list(or, &g_attr_inode_data, 1); 897 exofs_make_credential(oi->i_cred, &ios->obj);
898 ios->cred = oi->i_cred;
874 899
875#ifdef EXOFS_DEBUG_OBJ_ISIZE 900 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
876 /* we get the size attributes to do a sanity check */ 901 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
877 osd_req_add_get_attr_list(or, &g_attr_logical_length, 1);
878#endif
879 902
880 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred); 903 ios->in_attr = attrs;
881 if (ret) 904 ios->in_attr_len = ARRAY_SIZE(attrs);
905
906 ret = exofs_sbi_read(ios);
907 if (unlikely(ret)) {
908 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
909 _LLU(ios->obj.id), ret);
910 memset(inode, 0, sizeof(*inode));
911 inode->i_mode = 0040000 | (0777 & ~022);
912 /* If object is lost on target we might as well enable it's
913 * delete.
914 */
915 if ((ret == -ENOENT) || (ret == -EINVAL))
916 ret = 0;
882 goto out; 917 goto out;
918 }
883 919
884 attr = g_attr_inode_data; 920 ret = extract_attr_from_ios(ios, &attrs[0]);
885 ret = extract_attr_from_req(or, &attr);
886 if (ret) { 921 if (ret) {
887 EXOFS_ERR("exofs_get_inode: extract_attr_from_req failed\n"); 922 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
888 goto out; 923 goto out;
889 } 924 }
925 WARN_ON(attrs[0].len != EXOFS_INO_ATTR_SIZE);
926 memcpy(inode, attrs[0].val_ptr, EXOFS_INO_ATTR_SIZE);
890 927
891 WARN_ON(attr.len != EXOFS_INO_ATTR_SIZE); 928 ret = extract_attr_from_ios(ios, &attrs[1]);
892 memcpy(inode, attr.val_ptr, EXOFS_INO_ATTR_SIZE); 929 if (ret) {
930 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
931 goto out;
932 }
933 if (attrs[1].len) {
934 layout = attrs[1].val_ptr;
935 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
936 EXOFS_ERR("%s: unsupported files layout %d\n",
937 __func__, layout->gen_func);
938 ret = -ENOTSUPP;
939 goto out;
940 }
941 }
893 942
894#ifdef EXOFS_DEBUG_OBJ_ISIZE 943 ret = extract_attr_from_ios(ios, &attrs[2]);
895 attr = g_attr_logical_length;
896 ret = extract_attr_from_req(or, &attr);
897 if (ret) { 944 if (ret) {
898 EXOFS_ERR("ERROR: extract attr from or failed\n"); 945 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
899 goto out; 946 goto out;
900 } 947 }
901 *sanity = get_unaligned_be64(attr.val_ptr); 948 if (attrs[2].len) {
902#endif 949 layout = attrs[2].val_ptr;
950 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
951 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
952 __func__, layout->gen_func);
953 ret = -ENOTSUPP;
954 goto out;
955 }
956 }
903 957
904out: 958out:
905 osd_end_request(or); 959 exofs_put_io_state(ios);
906 return ret; 960 return ret;
907} 961}
908 962
963static void __oi_init(struct exofs_i_info *oi)
964{
965 init_waitqueue_head(&oi->i_wq);
966 oi->i_flags = 0;
967}
909/* 968/*
910 * Fill in an inode read from the OSD and set it up for use 969 * Fill in an inode read from the OSD and set it up for use
911 */ 970 */
@@ -914,7 +973,6 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
914 struct exofs_i_info *oi; 973 struct exofs_i_info *oi;
915 struct exofs_fcb fcb; 974 struct exofs_fcb fcb;
916 struct inode *inode; 975 struct inode *inode;
917 uint64_t uninitialized_var(sanity);
918 int ret; 976 int ret;
919 977
920 inode = iget_locked(sb, ino); 978 inode = iget_locked(sb, ino);
@@ -923,13 +981,13 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
923 if (!(inode->i_state & I_NEW)) 981 if (!(inode->i_state & I_NEW))
924 return inode; 982 return inode;
925 oi = exofs_i(inode); 983 oi = exofs_i(inode);
984 __oi_init(oi);
926 985
927 /* read the inode from the osd */ 986 /* read the inode from the osd */
928 ret = exofs_get_inode(sb, oi, &fcb, &sanity); 987 ret = exofs_get_inode(sb, oi, &fcb);
929 if (ret) 988 if (ret)
930 goto bad_inode; 989 goto bad_inode;
931 990
932 init_waitqueue_head(&oi->i_wq);
933 set_obj_created(oi); 991 set_obj_created(oi);
934 992
935 /* copy stuff from on-disk struct to in-memory struct */ 993 /* copy stuff from on-disk struct to in-memory struct */
@@ -947,15 +1005,6 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
947 inode->i_blkbits = EXOFS_BLKSHIFT; 1005 inode->i_blkbits = EXOFS_BLKSHIFT;
948 inode->i_generation = le32_to_cpu(fcb.i_generation); 1006 inode->i_generation = le32_to_cpu(fcb.i_generation);
949 1007
950#ifdef EXOFS_DEBUG_OBJ_ISIZE
951 if ((inode->i_size != sanity) &&
952 (!exofs_inode_is_fast_symlink(inode))) {
953 EXOFS_ERR("WARNING: Size of object from inode and "
954 "attributes differ (%lld != %llu)\n",
955 inode->i_size, _LLU(sanity));
956 }
957#endif
958
959 oi->i_dir_start_lookup = 0; 1008 oi->i_dir_start_lookup = 0;
960 1009
961 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) { 1010 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
@@ -1020,23 +1069,30 @@ int __exofs_wait_obj_created(struct exofs_i_info *oi)
1020 * set the obj_created flag so that other methods know that the object exists on 1069 * set the obj_created flag so that other methods know that the object exists on
1021 * the OSD. 1070 * the OSD.
1022 */ 1071 */
1023static void create_done(struct osd_request *or, void *p) 1072static void create_done(struct exofs_io_state *ios, void *p)
1024{ 1073{
1025 struct inode *inode = p; 1074 struct inode *inode = p;
1026 struct exofs_i_info *oi = exofs_i(inode); 1075 struct exofs_i_info *oi = exofs_i(inode);
1027 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; 1076 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
1028 int ret; 1077 int ret;
1029 1078
1030 ret = exofs_check_ok(or); 1079 ret = exofs_check_io(ios, NULL);
1031 osd_end_request(or); 1080 exofs_put_io_state(ios);
1081
1032 atomic_dec(&sbi->s_curr_pending); 1082 atomic_dec(&sbi->s_curr_pending);
1033 1083
1034 if (unlikely(ret)) { 1084 if (unlikely(ret)) {
1035 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx", 1085 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1036 _LLU(sbi->s_pid), _LLU(inode->i_ino + EXOFS_OBJ_OFF)); 1086 _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1037 make_bad_inode(inode); 1087 /*TODO: When FS is corrupted creation can fail, object already
1038 } else 1088 * exist. Get rid of this asynchronous creation, if exist
1039 set_obj_created(oi); 1089 * increment the obj counter and try the next object. Until we
1090 * succeed. All these dangling objects will be made into lost
1091 * files by chkfs.exofs
1092 */
1093 }
1094
1095 set_obj_created(oi);
1040 1096
1041 atomic_dec(&inode->i_count); 1097 atomic_dec(&inode->i_count);
1042 wake_up(&oi->i_wq); 1098 wake_up(&oi->i_wq);
@@ -1051,8 +1107,7 @@ struct inode *exofs_new_inode(struct inode *dir, int mode)
1051 struct inode *inode; 1107 struct inode *inode;
1052 struct exofs_i_info *oi; 1108 struct exofs_i_info *oi;
1053 struct exofs_sb_info *sbi; 1109 struct exofs_sb_info *sbi;
1054 struct osd_request *or; 1110 struct exofs_io_state *ios;
1055 struct osd_obj_id obj;
1056 int ret; 1111 int ret;
1057 1112
1058 sb = dir->i_sb; 1113 sb = dir->i_sb;
@@ -1061,8 +1116,8 @@ struct inode *exofs_new_inode(struct inode *dir, int mode)
1061 return ERR_PTR(-ENOMEM); 1116 return ERR_PTR(-ENOMEM);
1062 1117
1063 oi = exofs_i(inode); 1118 oi = exofs_i(inode);
1119 __oi_init(oi);
1064 1120
1065 init_waitqueue_head(&oi->i_wq);
1066 set_obj_2bcreated(oi); 1121 set_obj_2bcreated(oi);
1067 1122
1068 sbi = sb->s_fs_info; 1123 sbi = sb->s_fs_info;
@@ -1089,28 +1144,28 @@ struct inode *exofs_new_inode(struct inode *dir, int mode)
1089 1144
1090 mark_inode_dirty(inode); 1145 mark_inode_dirty(inode);
1091 1146
1092 obj.partition = sbi->s_pid; 1147 ret = exofs_get_io_state(&sbi->layout, &ios);
1093 obj.id = inode->i_ino + EXOFS_OBJ_OFF; 1148 if (unlikely(ret)) {
1094 exofs_make_credential(oi->i_cred, &obj); 1149 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1095 1150 return ERR_PTR(ret);
1096 or = osd_start_request(sbi->s_dev, GFP_KERNEL);
1097 if (unlikely(!or)) {
1098 EXOFS_ERR("exofs_new_inode: osd_start_request failed\n");
1099 return ERR_PTR(-ENOMEM);
1100 } 1151 }
1101 1152
1102 osd_req_create_object(or, &obj); 1153 ios->obj.id = exofs_oi_objno(oi);
1154 exofs_make_credential(oi->i_cred, &ios->obj);
1103 1155
1104 /* increment the refcount so that the inode will still be around when we 1156 /* increment the refcount so that the inode will still be around when we
1105 * reach the callback 1157 * reach the callback
1106 */ 1158 */
1107 atomic_inc(&inode->i_count); 1159 atomic_inc(&inode->i_count);
1108 1160
1109 ret = exofs_async_op(or, create_done, inode, oi->i_cred); 1161 ios->done = create_done;
1162 ios->private = inode;
1163 ios->cred = oi->i_cred;
1164 ret = exofs_sbi_create(ios);
1110 if (ret) { 1165 if (ret) {
1111 atomic_dec(&inode->i_count); 1166 atomic_dec(&inode->i_count);
1112 osd_end_request(or); 1167 exofs_put_io_state(ios);
1113 return ERR_PTR(-EIO); 1168 return ERR_PTR(ret);
1114 } 1169 }
1115 atomic_inc(&sbi->s_curr_pending); 1170 atomic_inc(&sbi->s_curr_pending);
1116 1171
@@ -1128,11 +1183,11 @@ struct updatei_args {
1128/* 1183/*
1129 * Callback function from exofs_update_inode(). 1184 * Callback function from exofs_update_inode().
1130 */ 1185 */
1131static void updatei_done(struct osd_request *or, void *p) 1186static void updatei_done(struct exofs_io_state *ios, void *p)
1132{ 1187{
1133 struct updatei_args *args = p; 1188 struct updatei_args *args = p;
1134 1189
1135 osd_end_request(or); 1190 exofs_put_io_state(ios);
1136 1191
1137 atomic_dec(&args->sbi->s_curr_pending); 1192 atomic_dec(&args->sbi->s_curr_pending);
1138 1193
@@ -1148,16 +1203,17 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1148 struct exofs_i_info *oi = exofs_i(inode); 1203 struct exofs_i_info *oi = exofs_i(inode);
1149 struct super_block *sb = inode->i_sb; 1204 struct super_block *sb = inode->i_sb;
1150 struct exofs_sb_info *sbi = sb->s_fs_info; 1205 struct exofs_sb_info *sbi = sb->s_fs_info;
1151 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF}; 1206 struct exofs_io_state *ios;
1152 struct osd_request *or;
1153 struct osd_attr attr; 1207 struct osd_attr attr;
1154 struct exofs_fcb *fcb; 1208 struct exofs_fcb *fcb;
1155 struct updatei_args *args; 1209 struct updatei_args *args;
1156 int ret; 1210 int ret;
1157 1211
1158 args = kzalloc(sizeof(*args), GFP_KERNEL); 1212 args = kzalloc(sizeof(*args), GFP_KERNEL);
1159 if (!args) 1213 if (!args) {
1214 EXOFS_DBGMSG("Faild kzalloc of args\n");
1160 return -ENOMEM; 1215 return -ENOMEM;
1216 }
1161 1217
1162 fcb = &args->fcb; 1218 fcb = &args->fcb;
1163 1219
@@ -1186,18 +1242,16 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1186 } else 1242 } else
1187 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data)); 1243 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1188 1244
1189 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 1245 ret = exofs_get_io_state(&sbi->layout, &ios);
1190 if (unlikely(!or)) { 1246 if (unlikely(ret)) {
1191 EXOFS_ERR("exofs_update_inode: osd_start_request failed.\n"); 1247 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1192 ret = -ENOMEM;
1193 goto free_args; 1248 goto free_args;
1194 } 1249 }
1195 1250
1196 osd_req_set_attributes(or, &obj);
1197
1198 attr = g_attr_inode_data; 1251 attr = g_attr_inode_data;
1199 attr.val_ptr = fcb; 1252 attr.val_ptr = fcb;
1200 osd_req_add_set_attr_list(or, &attr, 1); 1253 ios->out_attr_len = 1;
1254 ios->out_attr = &attr;
1201 1255
1202 if (!obj_created(oi)) { 1256 if (!obj_created(oi)) {
1203 EXOFS_DBGMSG("!obj_created\n"); 1257 EXOFS_DBGMSG("!obj_created\n");
@@ -1206,43 +1260,42 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1206 EXOFS_DBGMSG("wait_event done\n"); 1260 EXOFS_DBGMSG("wait_event done\n");
1207 } 1261 }
1208 1262
1209 if (do_sync) { 1263 if (!do_sync) {
1210 ret = exofs_sync_op(or, sbi->s_timeout, oi->i_cred);
1211 osd_end_request(or);
1212 goto free_args;
1213 } else {
1214 args->sbi = sbi; 1264 args->sbi = sbi;
1265 ios->done = updatei_done;
1266 ios->private = args;
1267 }
1215 1268
1216 ret = exofs_async_op(or, updatei_done, args, oi->i_cred); 1269 ret = exofs_oi_write(oi, ios);
1217 if (ret) { 1270 if (!do_sync && !ret) {
1218 osd_end_request(or);
1219 goto free_args;
1220 }
1221 atomic_inc(&sbi->s_curr_pending); 1271 atomic_inc(&sbi->s_curr_pending);
1222 goto out; /* deallocation in updatei_done */ 1272 goto out; /* deallocation in updatei_done */
1223 } 1273 }
1224 1274
1275 exofs_put_io_state(ios);
1225free_args: 1276free_args:
1226 kfree(args); 1277 kfree(args);
1227out: 1278out:
1228 EXOFS_DBGMSG("ret=>%d\n", ret); 1279 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1280 inode->i_ino, do_sync, ret);
1229 return ret; 1281 return ret;
1230} 1282}
1231 1283
1232int exofs_write_inode(struct inode *inode, int wait) 1284int exofs_write_inode(struct inode *inode, struct writeback_control *wbc)
1233{ 1285{
1234 return exofs_update_inode(inode, wait); 1286 return exofs_update_inode(inode, wbc->sync_mode == WB_SYNC_ALL);
1235} 1287}
1236 1288
1237/* 1289/*
1238 * Callback function from exofs_delete_inode() - don't have much cleaning up to 1290 * Callback function from exofs_delete_inode() - don't have much cleaning up to
1239 * do. 1291 * do.
1240 */ 1292 */
1241static void delete_done(struct osd_request *or, void *p) 1293static void delete_done(struct exofs_io_state *ios, void *p)
1242{ 1294{
1243 struct exofs_sb_info *sbi; 1295 struct exofs_sb_info *sbi = p;
1244 osd_end_request(or); 1296
1245 sbi = p; 1297 exofs_put_io_state(ios);
1298
1246 atomic_dec(&sbi->s_curr_pending); 1299 atomic_dec(&sbi->s_curr_pending);
1247} 1300}
1248 1301
@@ -1256,8 +1309,7 @@ void exofs_delete_inode(struct inode *inode)
1256 struct exofs_i_info *oi = exofs_i(inode); 1309 struct exofs_i_info *oi = exofs_i(inode);
1257 struct super_block *sb = inode->i_sb; 1310 struct super_block *sb = inode->i_sb;
1258 struct exofs_sb_info *sbi = sb->s_fs_info; 1311 struct exofs_sb_info *sbi = sb->s_fs_info;
1259 struct osd_obj_id obj = {sbi->s_pid, inode->i_ino + EXOFS_OBJ_OFF}; 1312 struct exofs_io_state *ios;
1260 struct osd_request *or;
1261 int ret; 1313 int ret;
1262 1314
1263 truncate_inode_pages(&inode->i_data, 0); 1315 truncate_inode_pages(&inode->i_data, 0);
@@ -1274,25 +1326,26 @@ void exofs_delete_inode(struct inode *inode)
1274 1326
1275 clear_inode(inode); 1327 clear_inode(inode);
1276 1328
1277 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 1329 ret = exofs_get_io_state(&sbi->layout, &ios);
1278 if (unlikely(!or)) { 1330 if (unlikely(ret)) {
1279 EXOFS_ERR("exofs_delete_inode: osd_start_request failed\n"); 1331 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1280 return; 1332 return;
1281 } 1333 }
1282 1334
1283 osd_req_remove_object(or, &obj);
1284
1285 /* if we are deleting an obj that hasn't been created yet, wait */ 1335 /* if we are deleting an obj that hasn't been created yet, wait */
1286 if (!obj_created(oi)) { 1336 if (!obj_created(oi)) {
1287 BUG_ON(!obj_2bcreated(oi)); 1337 BUG_ON(!obj_2bcreated(oi));
1288 wait_event(oi->i_wq, obj_created(oi)); 1338 wait_event(oi->i_wq, obj_created(oi));
1289 } 1339 }
1290 1340
1291 ret = exofs_async_op(or, delete_done, sbi, oi->i_cred); 1341 ios->obj.id = exofs_oi_objno(oi);
1342 ios->done = delete_done;
1343 ios->private = sbi;
1344 ios->cred = oi->i_cred;
1345 ret = exofs_sbi_remove(ios);
1292 if (ret) { 1346 if (ret) {
1293 EXOFS_ERR( 1347 EXOFS_ERR("%s: exofs_sbi_remove failed\n", __func__);
1294 "ERROR: @exofs_delete_inode exofs_async_op failed\n"); 1348 exofs_put_io_state(ios);
1295 osd_end_request(or);
1296 return; 1349 return;
1297 } 1350 }
1298 atomic_inc(&sbi->s_curr_pending); 1351 atomic_inc(&sbi->s_curr_pending);
diff --git a/fs/exofs/ios.c b/fs/exofs/ios.c
new file mode 100644
index 000000000000..4337cad7777b
--- /dev/null
+++ b/fs/exofs/ios.c
@@ -0,0 +1,823 @@
1/*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
6 *
7 * This file is part of exofs.
8 *
9 * exofs is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation. Since it is based on ext2, and the only
12 * valid version of GPL for the Linux kernel is version 2, the only valid
13 * version of GPL for exofs is version 2.
14 *
15 * exofs is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with exofs; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25#include <linux/slab.h>
26#include <scsi/scsi_device.h>
27#include <asm/div64.h>
28
29#include "exofs.h"
30
31#define EXOFS_DBGMSG2(M...) do {} while (0)
32/* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
33
34void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
35{
36 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
37}
38
39int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
40 u64 offset, void *p, unsigned length)
41{
42 struct osd_request *or = osd_start_request(od, GFP_KERNEL);
43/* struct osd_sense_info osi = {.key = 0};*/
44 int ret;
45
46 if (unlikely(!or)) {
47 EXOFS_DBGMSG("%s: osd_start_request failed.\n", __func__);
48 return -ENOMEM;
49 }
50 ret = osd_req_read_kern(or, obj, offset, p, length);
51 if (unlikely(ret)) {
52 EXOFS_DBGMSG("%s: osd_req_read_kern failed.\n", __func__);
53 goto out;
54 }
55
56 ret = osd_finalize_request(or, 0, cred, NULL);
57 if (unlikely(ret)) {
58 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
59 goto out;
60 }
61
62 ret = osd_execute_request(or);
63 if (unlikely(ret))
64 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
65 /* osd_req_decode_sense(or, ret); */
66
67out:
68 osd_end_request(or);
69 return ret;
70}
71
72int exofs_get_io_state(struct exofs_layout *layout,
73 struct exofs_io_state **pios)
74{
75 struct exofs_io_state *ios;
76
77 /*TODO: Maybe use kmem_cach per sbi of size
78 * exofs_io_state_size(layout->s_numdevs)
79 */
80 ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
81 if (unlikely(!ios)) {
82 EXOFS_DBGMSG("Faild kzalloc bytes=%d\n",
83 exofs_io_state_size(layout->s_numdevs));
84 *pios = NULL;
85 return -ENOMEM;
86 }
87
88 ios->layout = layout;
89 ios->obj.partition = layout->s_pid;
90 *pios = ios;
91 return 0;
92}
93
94void exofs_put_io_state(struct exofs_io_state *ios)
95{
96 if (ios) {
97 unsigned i;
98
99 for (i = 0; i < ios->numdevs; i++) {
100 struct exofs_per_dev_state *per_dev = &ios->per_dev[i];
101
102 if (per_dev->or)
103 osd_end_request(per_dev->or);
104 if (per_dev->bio)
105 bio_put(per_dev->bio);
106 }
107
108 kfree(ios);
109 }
110}
111
112unsigned exofs_layout_od_id(struct exofs_layout *layout,
113 osd_id obj_no, unsigned layout_index)
114{
115/* switch (layout->lay_func) {
116 case LAYOUT_MOVING_WINDOW:
117 {*/
118 unsigned dev_mod = obj_no;
119
120 return (layout_index + dev_mod * layout->mirrors_p1) %
121 layout->s_numdevs;
122/* }
123 case LAYOUT_FUNC_IMPLICT:
124 return layout->devs[layout_index];
125 }*/
126}
127
128static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
129 unsigned layout_index)
130{
131 return ios->layout->s_ods[
132 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
133}
134
135static void _sync_done(struct exofs_io_state *ios, void *p)
136{
137 struct completion *waiting = p;
138
139 complete(waiting);
140}
141
142static void _last_io(struct kref *kref)
143{
144 struct exofs_io_state *ios = container_of(
145 kref, struct exofs_io_state, kref);
146
147 ios->done(ios, ios->private);
148}
149
150static void _done_io(struct osd_request *or, void *p)
151{
152 struct exofs_io_state *ios = p;
153
154 kref_put(&ios->kref, _last_io);
155}
156
157static int exofs_io_execute(struct exofs_io_state *ios)
158{
159 DECLARE_COMPLETION_ONSTACK(wait);
160 bool sync = (ios->done == NULL);
161 int i, ret;
162
163 if (sync) {
164 ios->done = _sync_done;
165 ios->private = &wait;
166 }
167
168 for (i = 0; i < ios->numdevs; i++) {
169 struct osd_request *or = ios->per_dev[i].or;
170 if (unlikely(!or))
171 continue;
172
173 ret = osd_finalize_request(or, 0, ios->cred, NULL);
174 if (unlikely(ret)) {
175 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n",
176 ret);
177 return ret;
178 }
179 }
180
181 kref_init(&ios->kref);
182
183 for (i = 0; i < ios->numdevs; i++) {
184 struct osd_request *or = ios->per_dev[i].or;
185 if (unlikely(!or))
186 continue;
187
188 kref_get(&ios->kref);
189 osd_execute_request_async(or, _done_io, ios);
190 }
191
192 kref_put(&ios->kref, _last_io);
193 ret = 0;
194
195 if (sync) {
196 wait_for_completion(&wait);
197 ret = exofs_check_io(ios, NULL);
198 }
199 return ret;
200}
201
202static void _clear_bio(struct bio *bio)
203{
204 struct bio_vec *bv;
205 unsigned i;
206
207 __bio_for_each_segment(bv, bio, i, 0) {
208 unsigned this_count = bv->bv_len;
209
210 if (likely(PAGE_SIZE == this_count))
211 clear_highpage(bv->bv_page);
212 else
213 zero_user(bv->bv_page, bv->bv_offset, this_count);
214 }
215}
216
217int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
218{
219 enum osd_err_priority acumulated_osd_err = 0;
220 int acumulated_lin_err = 0;
221 int i;
222
223 for (i = 0; i < ios->numdevs; i++) {
224 struct osd_sense_info osi;
225 struct osd_request *or = ios->per_dev[i].or;
226 int ret;
227
228 if (unlikely(!or))
229 continue;
230
231 ret = osd_req_decode_sense(or, &osi);
232 if (likely(!ret))
233 continue;
234
235 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
236 /* start read offset passed endof file */
237 _clear_bio(ios->per_dev[i].bio);
238 EXOFS_DBGMSG("start read offset passed end of file "
239 "offset=0x%llx, length=0x%llx\n",
240 _LLU(ios->per_dev[i].offset),
241 _LLU(ios->per_dev[i].length));
242
243 continue; /* we recovered */
244 }
245
246 if (osi.osd_err_pri >= acumulated_osd_err) {
247 acumulated_osd_err = osi.osd_err_pri;
248 acumulated_lin_err = ret;
249 }
250 }
251
252 /* TODO: raid specific residual calculations */
253 if (resid) {
254 if (likely(!acumulated_lin_err))
255 *resid = 0;
256 else
257 *resid = ios->length;
258 }
259
260 return acumulated_lin_err;
261}
262
263/*
264 * L - logical offset into the file
265 *
266 * U - The number of bytes in a stripe within a group
267 *
268 * U = stripe_unit * group_width
269 *
270 * T - The number of bytes striped within a group of component objects
271 * (before advancing to the next group)
272 *
273 * T = stripe_unit * group_width * group_depth
274 *
275 * S - The number of bytes striped across all component objects
276 * before the pattern repeats
277 *
278 * S = stripe_unit * group_width * group_depth * group_count
279 *
280 * M - The "major" (i.e., across all components) stripe number
281 *
282 * M = L / S
283 *
284 * G - Counts the groups from the beginning of the major stripe
285 *
286 * G = (L - (M * S)) / T [or (L % S) / T]
287 *
288 * H - The byte offset within the group
289 *
290 * H = (L - (M * S)) % T [or (L % S) % T]
291 *
292 * N - The "minor" (i.e., across the group) stripe number
293 *
294 * N = H / U
295 *
296 * C - The component index coresponding to L
297 *
298 * C = (H - (N * U)) / stripe_unit + G * group_width
299 * [or (L % U) / stripe_unit + G * group_width]
300 *
301 * O - The component offset coresponding to L
302 *
303 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
304 */
305struct _striping_info {
306 u64 obj_offset;
307 u64 group_length;
308 u64 total_group_length;
309 u64 Major;
310 unsigned dev;
311 unsigned unit_off;
312};
313
314static void _calc_stripe_info(struct exofs_io_state *ios, u64 file_offset,
315 struct _striping_info *si)
316{
317 u32 stripe_unit = ios->layout->stripe_unit;
318 u32 group_width = ios->layout->group_width;
319 u64 group_depth = ios->layout->group_depth;
320
321 u32 U = stripe_unit * group_width;
322 u64 T = U * group_depth;
323 u64 S = T * ios->layout->group_count;
324 u64 M = div64_u64(file_offset, S);
325
326 /*
327 G = (L - (M * S)) / T
328 H = (L - (M * S)) % T
329 */
330 u64 LmodS = file_offset - M * S;
331 u32 G = div64_u64(LmodS, T);
332 u64 H = LmodS - G * T;
333
334 u32 N = div_u64(H, U);
335
336 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
337 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
338 si->dev *= ios->layout->mirrors_p1;
339
340 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
341
342 si->obj_offset = si->unit_off + (N * stripe_unit) +
343 (M * group_depth * stripe_unit);
344
345 si->group_length = T - H;
346 si->total_group_length = T;
347 si->Major = M;
348}
349
350static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
351 unsigned pgbase, struct exofs_per_dev_state *per_dev,
352 int cur_len)
353{
354 unsigned pg = *cur_pg;
355 struct request_queue *q =
356 osd_request_queue(exofs_ios_od(ios, per_dev->dev));
357
358 per_dev->length += cur_len;
359
360 if (per_dev->bio == NULL) {
361 unsigned pages_in_stripe = ios->layout->group_width *
362 (ios->layout->stripe_unit / PAGE_SIZE);
363 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
364 ios->layout->group_width;
365
366 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
367 if (unlikely(!per_dev->bio)) {
368 EXOFS_DBGMSG("Faild to allocate BIO size=%u\n",
369 bio_size);
370 return -ENOMEM;
371 }
372 }
373
374 while (cur_len > 0) {
375 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
376 unsigned added_len;
377
378 BUG_ON(ios->nr_pages <= pg);
379 cur_len -= pglen;
380
381 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
382 pglen, pgbase);
383 if (unlikely(pglen != added_len))
384 return -ENOMEM;
385 pgbase = 0;
386 ++pg;
387 }
388 BUG_ON(cur_len);
389
390 *cur_pg = pg;
391 return 0;
392}
393
394static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
395 struct _striping_info *si, unsigned first_comp)
396{
397 unsigned stripe_unit = ios->layout->stripe_unit;
398 unsigned mirrors_p1 = ios->layout->mirrors_p1;
399 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
400 unsigned dev = si->dev;
401 unsigned first_dev = dev - (dev % devs_in_group);
402 unsigned comp = first_comp + (dev - first_dev);
403 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
404 unsigned cur_pg = ios->pages_consumed;
405 int ret = 0;
406
407 while (length) {
408 struct exofs_per_dev_state *per_dev = &ios->per_dev[comp];
409 unsigned cur_len, page_off = 0;
410
411 if (!per_dev->length) {
412 per_dev->dev = dev;
413 if (dev < si->dev) {
414 per_dev->offset = si->obj_offset + stripe_unit -
415 si->unit_off;
416 cur_len = stripe_unit;
417 } else if (dev == si->dev) {
418 per_dev->offset = si->obj_offset;
419 cur_len = stripe_unit - si->unit_off;
420 page_off = si->unit_off & ~PAGE_MASK;
421 BUG_ON(page_off && (page_off != ios->pgbase));
422 } else { /* dev > si->dev */
423 per_dev->offset = si->obj_offset - si->unit_off;
424 cur_len = stripe_unit;
425 }
426
427 if (max_comp < comp)
428 max_comp = comp;
429
430 dev += mirrors_p1;
431 dev = (dev % devs_in_group) + first_dev;
432 } else {
433 cur_len = stripe_unit;
434 }
435 if (cur_len >= length)
436 cur_len = length;
437
438 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
439 cur_len);
440 if (unlikely(ret))
441 goto out;
442
443 comp += mirrors_p1;
444 comp = (comp % devs_in_group) + first_comp;
445
446 length -= cur_len;
447 }
448out:
449 ios->numdevs = max_comp + mirrors_p1;
450 ios->pages_consumed = cur_pg;
451 return ret;
452}
453
454static int _prepare_for_striping(struct exofs_io_state *ios)
455{
456 u64 length = ios->length;
457 struct _striping_info si;
458 unsigned devs_in_group = ios->layout->group_width *
459 ios->layout->mirrors_p1;
460 unsigned first_comp = 0;
461 int ret = 0;
462
463 _calc_stripe_info(ios, ios->offset, &si);
464
465 if (!ios->pages) {
466 if (ios->kern_buff) {
467 struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
468
469 per_dev->offset = si.obj_offset;
470 per_dev->dev = si.dev;
471
472 /* no cross device without page array */
473 BUG_ON((ios->layout->group_width > 1) &&
474 (si.unit_off + ios->length >
475 ios->layout->stripe_unit));
476 }
477 ios->numdevs = ios->layout->mirrors_p1;
478 return 0;
479 }
480
481 while (length) {
482 if (length < si.group_length)
483 si.group_length = length;
484
485 ret = _prepare_one_group(ios, si.group_length, &si, first_comp);
486 if (unlikely(ret))
487 goto out;
488
489 length -= si.group_length;
490
491 si.group_length = si.total_group_length;
492 si.unit_off = 0;
493 ++si.Major;
494 si.obj_offset = si.Major * ios->layout->stripe_unit *
495 ios->layout->group_depth;
496
497 si.dev = (si.dev - (si.dev % devs_in_group)) + devs_in_group;
498 si.dev %= ios->layout->s_numdevs;
499
500 first_comp += devs_in_group;
501 first_comp %= ios->layout->s_numdevs;
502 }
503
504out:
505 return ret;
506}
507
508int exofs_sbi_create(struct exofs_io_state *ios)
509{
510 int i, ret;
511
512 for (i = 0; i < ios->layout->s_numdevs; i++) {
513 struct osd_request *or;
514
515 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
516 if (unlikely(!or)) {
517 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
518 ret = -ENOMEM;
519 goto out;
520 }
521 ios->per_dev[i].or = or;
522 ios->numdevs++;
523
524 osd_req_create_object(or, &ios->obj);
525 }
526 ret = exofs_io_execute(ios);
527
528out:
529 return ret;
530}
531
532int exofs_sbi_remove(struct exofs_io_state *ios)
533{
534 int i, ret;
535
536 for (i = 0; i < ios->layout->s_numdevs; i++) {
537 struct osd_request *or;
538
539 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
540 if (unlikely(!or)) {
541 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
542 ret = -ENOMEM;
543 goto out;
544 }
545 ios->per_dev[i].or = or;
546 ios->numdevs++;
547
548 osd_req_remove_object(or, &ios->obj);
549 }
550 ret = exofs_io_execute(ios);
551
552out:
553 return ret;
554}
555
556static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
557{
558 struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
559 unsigned dev = ios->per_dev[cur_comp].dev;
560 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
561 int ret = 0;
562
563 if (ios->pages && !master_dev->length)
564 return 0; /* Just an empty slot */
565
566 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
567 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
568 struct osd_request *or;
569
570 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
571 if (unlikely(!or)) {
572 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
573 ret = -ENOMEM;
574 goto out;
575 }
576 per_dev->or = or;
577 per_dev->offset = master_dev->offset;
578
579 if (ios->pages) {
580 struct bio *bio;
581
582 if (per_dev != master_dev) {
583 bio = bio_kmalloc(GFP_KERNEL,
584 master_dev->bio->bi_max_vecs);
585 if (unlikely(!bio)) {
586 EXOFS_DBGMSG(
587 "Faild to allocate BIO size=%u\n",
588 master_dev->bio->bi_max_vecs);
589 ret = -ENOMEM;
590 goto out;
591 }
592
593 __bio_clone(bio, master_dev->bio);
594 bio->bi_bdev = NULL;
595 bio->bi_next = NULL;
596 per_dev->length = master_dev->length;
597 per_dev->bio = bio;
598 per_dev->dev = dev;
599 } else {
600 bio = master_dev->bio;
601 /* FIXME: bio_set_dir() */
602 bio->bi_rw |= (1 << BIO_RW);
603 }
604
605 osd_req_write(or, &ios->obj, per_dev->offset, bio,
606 per_dev->length);
607 EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
608 "length=0x%llx dev=%d\n",
609 _LLU(ios->obj.id), _LLU(per_dev->offset),
610 _LLU(per_dev->length), dev);
611 } else if (ios->kern_buff) {
612 ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
613 ios->kern_buff, ios->length);
614 if (unlikely(ret))
615 goto out;
616 EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
617 "length=0x%llx dev=%d\n",
618 _LLU(ios->obj.id), _LLU(per_dev->offset),
619 _LLU(ios->length), dev);
620 } else {
621 osd_req_set_attributes(or, &ios->obj);
622 EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
623 _LLU(ios->obj.id), ios->out_attr_len, dev);
624 }
625
626 if (ios->out_attr)
627 osd_req_add_set_attr_list(or, ios->out_attr,
628 ios->out_attr_len);
629
630 if (ios->in_attr)
631 osd_req_add_get_attr_list(or, ios->in_attr,
632 ios->in_attr_len);
633 }
634
635out:
636 return ret;
637}
638
639int exofs_sbi_write(struct exofs_io_state *ios)
640{
641 int i;
642 int ret;
643
644 ret = _prepare_for_striping(ios);
645 if (unlikely(ret))
646 return ret;
647
648 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
649 ret = _sbi_write_mirror(ios, i);
650 if (unlikely(ret))
651 return ret;
652 }
653
654 ret = exofs_io_execute(ios);
655 return ret;
656}
657
658static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
659{
660 struct osd_request *or;
661 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
662 unsigned first_dev = (unsigned)ios->obj.id;
663
664 if (ios->pages && !per_dev->length)
665 return 0; /* Just an empty slot */
666
667 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
668 or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
669 if (unlikely(!or)) {
670 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
671 return -ENOMEM;
672 }
673 per_dev->or = or;
674
675 if (ios->pages) {
676 osd_req_read(or, &ios->obj, per_dev->offset,
677 per_dev->bio, per_dev->length);
678 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
679 " dev=%d\n", _LLU(ios->obj.id),
680 _LLU(per_dev->offset), _LLU(per_dev->length),
681 first_dev);
682 } else if (ios->kern_buff) {
683 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
684 ios->kern_buff, ios->length);
685 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
686 "length=0x%llx dev=%d ret=>%d\n",
687 _LLU(ios->obj.id), _LLU(per_dev->offset),
688 _LLU(ios->length), first_dev, ret);
689 if (unlikely(ret))
690 return ret;
691 } else {
692 osd_req_get_attributes(or, &ios->obj);
693 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
694 _LLU(ios->obj.id), ios->in_attr_len, first_dev);
695 }
696 if (ios->out_attr)
697 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
698
699 if (ios->in_attr)
700 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
701
702 return 0;
703}
704
705int exofs_sbi_read(struct exofs_io_state *ios)
706{
707 int i;
708 int ret;
709
710 ret = _prepare_for_striping(ios);
711 if (unlikely(ret))
712 return ret;
713
714 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
715 ret = _sbi_read_mirror(ios, i);
716 if (unlikely(ret))
717 return ret;
718 }
719
720 ret = exofs_io_execute(ios);
721 return ret;
722}
723
724int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
725{
726 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
727 void *iter = NULL;
728 int nelem;
729
730 do {
731 nelem = 1;
732 osd_req_decode_get_attr_list(ios->per_dev[0].or,
733 &cur_attr, &nelem, &iter);
734 if ((cur_attr.attr_page == attr->attr_page) &&
735 (cur_attr.attr_id == attr->attr_id)) {
736 attr->len = cur_attr.len;
737 attr->val_ptr = cur_attr.val_ptr;
738 return 0;
739 }
740 } while (iter);
741
742 return -EIO;
743}
744
745static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
746 struct osd_attr *attr)
747{
748 int last_comp = cur_comp + ios->layout->mirrors_p1;
749
750 for (; cur_comp < last_comp; ++cur_comp) {
751 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
752 struct osd_request *or;
753
754 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
755 if (unlikely(!or)) {
756 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
757 return -ENOMEM;
758 }
759 per_dev->or = or;
760
761 osd_req_set_attributes(or, &ios->obj);
762 osd_req_add_set_attr_list(or, attr, 1);
763 }
764
765 return 0;
766}
767
768int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
769{
770 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
771 struct exofs_io_state *ios;
772 struct exofs_trunc_attr {
773 struct osd_attr attr;
774 __be64 newsize;
775 } *size_attrs;
776 struct _striping_info si;
777 int i, ret;
778
779 ret = exofs_get_io_state(&sbi->layout, &ios);
780 if (unlikely(ret))
781 return ret;
782
783 size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs),
784 GFP_KERNEL);
785 if (unlikely(!size_attrs)) {
786 ret = -ENOMEM;
787 goto out;
788 }
789
790 ios->obj.id = exofs_oi_objno(oi);
791 ios->cred = oi->i_cred;
792
793 ios->numdevs = ios->layout->s_numdevs;
794 _calc_stripe_info(ios, size, &si);
795
796 for (i = 0; i < ios->layout->group_width; ++i) {
797 struct exofs_trunc_attr *size_attr = &size_attrs[i];
798 u64 obj_size;
799
800 if (i < si.dev)
801 obj_size = si.obj_offset +
802 ios->layout->stripe_unit - si.unit_off;
803 else if (i == si.dev)
804 obj_size = si.obj_offset;
805 else /* i > si.dev */
806 obj_size = si.obj_offset - si.unit_off;
807
808 size_attr->newsize = cpu_to_be64(obj_size);
809 size_attr->attr = g_attr_logical_length;
810 size_attr->attr.val_ptr = &size_attr->newsize;
811
812 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
813 &size_attr->attr);
814 if (unlikely(ret))
815 goto out;
816 }
817 ret = exofs_io_execute(ios);
818
819out:
820 kfree(size_attrs);
821 exofs_put_io_state(ios);
822 return ret;
823}
diff --git a/fs/exofs/osd.c b/fs/exofs/osd.c
deleted file mode 100644
index 4372542df284..000000000000
--- a/fs/exofs/osd.c
+++ /dev/null
@@ -1,125 +0,0 @@
1/*
2 * Copyright (C) 2005, 2006
3 * Avishay Traeger (avishay@gmail.com)
4 * Copyright (C) 2008, 2009
5 * Boaz Harrosh <bharrosh@panasas.com>
6 *
7 * This file is part of exofs.
8 *
9 * exofs is free software; you can redistribute it and/or modify
10 * it under the terms of the GNU General Public License as published by
11 * the Free Software Foundation. Since it is based on ext2, and the only
12 * valid version of GPL for the Linux kernel is version 2, the only valid
13 * version of GPL for exofs is version 2.
14 *
15 * exofs is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with exofs; if not, write to the Free Software
22 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
23 */
24
25#include <scsi/scsi_device.h>
26#include <scsi/osd_sense.h>
27
28#include "exofs.h"
29
30int exofs_check_ok_resid(struct osd_request *or, u64 *in_resid, u64 *out_resid)
31{
32 struct osd_sense_info osi;
33 int ret = osd_req_decode_sense(or, &osi);
34
35 if (ret) { /* translate to Linux codes */
36 if (osi.additional_code == scsi_invalid_field_in_cdb) {
37 if (osi.cdb_field_offset == OSD_CFO_STARTING_BYTE)
38 ret = -EFAULT;
39 if (osi.cdb_field_offset == OSD_CFO_OBJECT_ID)
40 ret = -ENOENT;
41 else
42 ret = -EINVAL;
43 } else if (osi.additional_code == osd_quota_error)
44 ret = -ENOSPC;
45 else
46 ret = -EIO;
47 }
48
49 /* FIXME: should be include in osd_sense_info */
50 if (in_resid)
51 *in_resid = or->in.req ? or->in.req->resid_len : 0;
52
53 if (out_resid)
54 *out_resid = or->out.req ? or->out.req->resid_len : 0;
55
56 return ret;
57}
58
59void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
60{
61 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
62}
63
64/*
65 * Perform a synchronous OSD operation.
66 */
67int exofs_sync_op(struct osd_request *or, int timeout, uint8_t *credential)
68{
69 int ret;
70
71 or->timeout = timeout;
72 ret = osd_finalize_request(or, 0, credential, NULL);
73 if (ret) {
74 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
75 return ret;
76 }
77
78 ret = osd_execute_request(or);
79
80 if (ret)
81 EXOFS_DBGMSG("osd_execute_request() => %d\n", ret);
82 /* osd_req_decode_sense(or, ret); */
83 return ret;
84}
85
86/*
87 * Perform an asynchronous OSD operation.
88 */
89int exofs_async_op(struct osd_request *or, osd_req_done_fn *async_done,
90 void *caller_context, u8 *cred)
91{
92 int ret;
93
94 ret = osd_finalize_request(or, 0, cred, NULL);
95 if (ret) {
96 EXOFS_DBGMSG("Faild to osd_finalize_request() => %d\n", ret);
97 return ret;
98 }
99
100 ret = osd_execute_request_async(or, async_done, caller_context);
101
102 if (ret)
103 EXOFS_DBGMSG("osd_execute_request_async() => %d\n", ret);
104 return ret;
105}
106
107int extract_attr_from_req(struct osd_request *or, struct osd_attr *attr)
108{
109 struct osd_attr cur_attr = {.attr_page = 0}; /* start with zeros */
110 void *iter = NULL;
111 int nelem;
112
113 do {
114 nelem = 1;
115 osd_req_decode_get_attr_list(or, &cur_attr, &nelem, &iter);
116 if ((cur_attr.attr_page == attr->attr_page) &&
117 (cur_attr.attr_id == attr->attr_id)) {
118 attr->len = cur_attr.len;
119 attr->val_ptr = cur_attr.val_ptr;
120 return 0;
121 }
122 } while (iter);
123
124 return -EIO;
125}
diff --git a/fs/exofs/pnfs.h b/fs/exofs/pnfs.h
new file mode 100644
index 000000000000..c52e9888b8ab
--- /dev/null
+++ b/fs/exofs/pnfs.h
@@ -0,0 +1,45 @@
1/*
2 * Copyright (C) 2008, 2009
3 * Boaz Harrosh <bharrosh@panasas.com>
4 *
5 * This file is part of exofs.
6 *
7 * exofs is free software; you can redistribute it and/or modify it under the
8 * terms of the GNU General Public License version 2 as published by the Free
9 * Software Foundation.
10 *
11 */
12
13/* FIXME: Remove this file once pnfs hits mainline */
14
15#ifndef __EXOFS_PNFS_H__
16#define __EXOFS_PNFS_H__
17
18#if ! defined(__PNFS_OSD_XDR_H__)
19
20enum pnfs_iomode {
21 IOMODE_READ = 1,
22 IOMODE_RW = 2,
23 IOMODE_ANY = 3,
24};
25
26/* Layout Structure */
27enum pnfs_osd_raid_algorithm4 {
28 PNFS_OSD_RAID_0 = 1,
29 PNFS_OSD_RAID_4 = 2,
30 PNFS_OSD_RAID_5 = 3,
31 PNFS_OSD_RAID_PQ = 4 /* Reed-Solomon P+Q */
32};
33
34struct pnfs_osd_data_map {
35 u32 odm_num_comps;
36 u64 odm_stripe_unit;
37 u32 odm_group_width;
38 u32 odm_group_depth;
39 u32 odm_mirror_cnt;
40 u32 odm_raid_algorithm;
41};
42
43#endif /* ! defined(__PNFS_OSD_XDR_H__) */
44
45#endif /* __EXOFS_PNFS_H__ */
diff --git a/fs/exofs/super.c b/fs/exofs/super.c
index 9f500dec3b59..18e57ea1e5b4 100644
--- a/fs/exofs/super.c
+++ b/fs/exofs/super.c
@@ -37,6 +37,7 @@
37#include <linux/vfs.h> 37#include <linux/vfs.h>
38#include <linux/random.h> 38#include <linux/random.h>
39#include <linux/exportfs.h> 39#include <linux/exportfs.h>
40#include <linux/slab.h>
40 41
41#include "exofs.h" 42#include "exofs.h"
42 43
@@ -203,49 +204,45 @@ int exofs_sync_fs(struct super_block *sb, int wait)
203{ 204{
204 struct exofs_sb_info *sbi; 205 struct exofs_sb_info *sbi;
205 struct exofs_fscb *fscb; 206 struct exofs_fscb *fscb;
206 struct osd_request *or; 207 struct exofs_io_state *ios;
207 struct osd_obj_id obj;
208 int ret = -ENOMEM; 208 int ret = -ENOMEM;
209 209
210 fscb = kzalloc(sizeof(struct exofs_fscb), GFP_KERNEL);
211 if (!fscb) {
212 EXOFS_ERR("exofs_write_super: memory allocation failed.\n");
213 return -ENOMEM;
214 }
215
216 lock_super(sb); 210 lock_super(sb);
217 sbi = sb->s_fs_info; 211 sbi = sb->s_fs_info;
212 fscb = &sbi->s_fscb;
213
214 ret = exofs_get_io_state(&sbi->layout, &ios);
215 if (ret)
216 goto out;
217
218 /* Note: We only write the changing part of the fscb. .i.e upto the
219 * the fscb->s_dev_table_oid member. There is no read-modify-write
220 * here.
221 */
222 ios->length = offsetof(struct exofs_fscb, s_dev_table_oid);
223 memset(fscb, 0, ios->length);
218 fscb->s_nextid = cpu_to_le64(sbi->s_nextid); 224 fscb->s_nextid = cpu_to_le64(sbi->s_nextid);
219 fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles); 225 fscb->s_numfiles = cpu_to_le32(sbi->s_numfiles);
220 fscb->s_magic = cpu_to_le16(sb->s_magic); 226 fscb->s_magic = cpu_to_le16(sb->s_magic);
221 fscb->s_newfs = 0; 227 fscb->s_newfs = 0;
228 fscb->s_version = EXOFS_FSCB_VER;
222 229
223 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 230 ios->obj.id = EXOFS_SUPER_ID;
224 if (unlikely(!or)) { 231 ios->offset = 0;
225 EXOFS_ERR("exofs_write_super: osd_start_request failed.\n"); 232 ios->kern_buff = fscb;
226 goto out; 233 ios->cred = sbi->s_cred;
227 }
228
229 obj.partition = sbi->s_pid;
230 obj.id = EXOFS_SUPER_ID;
231 ret = osd_req_write_kern(or, &obj, 0, fscb, sizeof(*fscb));
232 if (unlikely(ret)) {
233 EXOFS_ERR("exofs_write_super: osd_req_write_kern failed.\n");
234 goto out;
235 }
236 234
237 ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred); 235 ret = exofs_sbi_write(ios);
238 if (unlikely(ret)) { 236 if (unlikely(ret)) {
239 EXOFS_ERR("exofs_write_super: exofs_sync_op failed.\n"); 237 EXOFS_ERR("%s: exofs_sbi_write failed.\n", __func__);
240 goto out; 238 goto out;
241 } 239 }
242 sb->s_dirt = 0; 240 sb->s_dirt = 0;
243 241
244out: 242out:
245 if (or) 243 EXOFS_DBGMSG("s_nextid=0x%llx ret=%d\n", _LLU(sbi->s_nextid), ret);
246 osd_end_request(or); 244 exofs_put_io_state(ios);
247 unlock_super(sb); 245 unlock_super(sb);
248 kfree(fscb);
249 return ret; 246 return ret;
250} 247}
251 248
@@ -257,6 +254,29 @@ static void exofs_write_super(struct super_block *sb)
257 sb->s_dirt = 0; 254 sb->s_dirt = 0;
258} 255}
259 256
257static void _exofs_print_device(const char *msg, const char *dev_path,
258 struct osd_dev *od, u64 pid)
259{
260 const struct osd_dev_info *odi = osduld_device_info(od);
261
262 printk(KERN_NOTICE "exofs: %s %s osd_name-%s pid-0x%llx\n",
263 msg, dev_path ?: "", odi->osdname, _LLU(pid));
264}
265
266void exofs_free_sbi(struct exofs_sb_info *sbi)
267{
268 while (sbi->layout.s_numdevs) {
269 int i = --sbi->layout.s_numdevs;
270 struct osd_dev *od = sbi->layout.s_ods[i];
271
272 if (od) {
273 sbi->layout.s_ods[i] = NULL;
274 osduld_put_device(od);
275 }
276 }
277 kfree(sbi);
278}
279
260/* 280/*
261 * This function is called when the vfs is freeing the superblock. We just 281 * This function is called when the vfs is freeing the superblock. We just
262 * need to free our own part. 282 * need to free our own part.
@@ -279,11 +299,235 @@ static void exofs_put_super(struct super_block *sb)
279 msecs_to_jiffies(100)); 299 msecs_to_jiffies(100));
280 } 300 }
281 301
282 osduld_put_device(sbi->s_dev); 302 _exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
283 kfree(sb->s_fs_info); 303 sbi->layout.s_pid);
304
305 exofs_free_sbi(sbi);
284 sb->s_fs_info = NULL; 306 sb->s_fs_info = NULL;
285} 307}
286 308
309static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
310 struct exofs_device_table *dt)
311{
312 u64 stripe_length;
313
314 sbi->data_map.odm_num_comps =
315 le32_to_cpu(dt->dt_data_map.cb_num_comps);
316 sbi->data_map.odm_stripe_unit =
317 le64_to_cpu(dt->dt_data_map.cb_stripe_unit);
318 sbi->data_map.odm_group_width =
319 le32_to_cpu(dt->dt_data_map.cb_group_width);
320 sbi->data_map.odm_group_depth =
321 le32_to_cpu(dt->dt_data_map.cb_group_depth);
322 sbi->data_map.odm_mirror_cnt =
323 le32_to_cpu(dt->dt_data_map.cb_mirror_cnt);
324 sbi->data_map.odm_raid_algorithm =
325 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
326
327/* FIXME: Only raid0 for now. if not so, do not mount */
328 if (sbi->data_map.odm_num_comps != numdevs) {
329 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
330 sbi->data_map.odm_num_comps, numdevs);
331 return -EINVAL;
332 }
333 if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
334 EXOFS_ERR("Only RAID_0 for now\n");
335 return -EINVAL;
336 }
337 if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
338 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
339 numdevs, sbi->data_map.odm_mirror_cnt);
340 return -EINVAL;
341 }
342
343 if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
344 EXOFS_ERR("Stripe Unit(0x%llx)"
345 " must be Multples of PAGE_SIZE(0x%lx)\n",
346 _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
347 return -EINVAL;
348 }
349
350 sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
351 sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
352
353 if (sbi->data_map.odm_group_width) {
354 sbi->layout.group_width = sbi->data_map.odm_group_width;
355 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
356 if (!sbi->layout.group_depth) {
357 EXOFS_ERR("group_depth == 0 && group_width != 0\n");
358 return -EINVAL;
359 }
360 sbi->layout.group_count = sbi->data_map.odm_num_comps /
361 sbi->layout.mirrors_p1 /
362 sbi->data_map.odm_group_width;
363 } else {
364 if (sbi->data_map.odm_group_depth) {
365 printk(KERN_NOTICE "Warning: group_depth ignored "
366 "group_width == 0 && group_depth == %d\n",
367 sbi->data_map.odm_group_depth);
368 sbi->data_map.odm_group_depth = 0;
369 }
370 sbi->layout.group_width = sbi->data_map.odm_num_comps /
371 sbi->layout.mirrors_p1;
372 sbi->layout.group_depth = -1;
373 sbi->layout.group_count = 1;
374 }
375
376 stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
377 if (stripe_length >= (1ULL << 32)) {
378 EXOFS_ERR("Total Stripe length(0x%llx)"
379 " >= 32bit is not supported\n", _LLU(stripe_length));
380 return -EINVAL;
381 }
382
383 return 0;
384}
385
386/* @odi is valid only as long as @fscb_dev is valid */
387static int exofs_devs_2_odi(struct exofs_dt_device_info *dt_dev,
388 struct osd_dev_info *odi)
389{
390 odi->systemid_len = le32_to_cpu(dt_dev->systemid_len);
391 memcpy(odi->systemid, dt_dev->systemid, odi->systemid_len);
392
393 odi->osdname_len = le32_to_cpu(dt_dev->osdname_len);
394 odi->osdname = dt_dev->osdname;
395
396 /* FIXME support long names. Will need a _put function */
397 if (dt_dev->long_name_offset)
398 return -EINVAL;
399
400 /* Make sure osdname is printable!
401 * mkexofs should give us space for a null-terminator else the
402 * device-table is invalid.
403 */
404 if (unlikely(odi->osdname_len >= sizeof(dt_dev->osdname)))
405 odi->osdname_len = sizeof(dt_dev->osdname) - 1;
406 dt_dev->osdname[odi->osdname_len] = 0;
407
408 /* If it's all zeros something is bad we read past end-of-obj */
409 return !(odi->systemid_len || odi->osdname_len);
410}
411
412static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
413 unsigned table_count)
414{
415 struct exofs_sb_info *sbi = *psbi;
416 struct osd_dev *fscb_od;
417 struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
418 .id = EXOFS_DEVTABLE_ID};
419 struct exofs_device_table *dt;
420 unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
421 sizeof(*dt);
422 unsigned numdevs, i;
423 int ret;
424
425 dt = kmalloc(table_bytes, GFP_KERNEL);
426 if (unlikely(!dt)) {
427 EXOFS_ERR("ERROR: allocating %x bytes for device table\n",
428 table_bytes);
429 return -ENOMEM;
430 }
431
432 fscb_od = sbi->layout.s_ods[0];
433 sbi->layout.s_ods[0] = NULL;
434 sbi->layout.s_numdevs = 0;
435 ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
436 if (unlikely(ret)) {
437 EXOFS_ERR("ERROR: reading device table\n");
438 goto out;
439 }
440
441 numdevs = le64_to_cpu(dt->dt_num_devices);
442 if (unlikely(!numdevs)) {
443 ret = -EINVAL;
444 goto out;
445 }
446 WARN_ON(table_count != numdevs);
447
448 ret = _read_and_match_data_map(sbi, numdevs, dt);
449 if (unlikely(ret))
450 goto out;
451
452 if (likely(numdevs > 1)) {
453 unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
454
455 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
456 if (unlikely(!sbi)) {
457 ret = -ENOMEM;
458 goto out;
459 }
460 memset(&sbi->layout.s_ods[1], 0,
461 size - sizeof(sbi->layout.s_ods[0]));
462 *psbi = sbi;
463 }
464
465 for (i = 0; i < numdevs; i++) {
466 struct exofs_fscb fscb;
467 struct osd_dev_info odi;
468 struct osd_dev *od;
469
470 if (exofs_devs_2_odi(&dt->dt_dev_table[i], &odi)) {
471 EXOFS_ERR("ERROR: Read all-zeros device entry\n");
472 ret = -EINVAL;
473 goto out;
474 }
475
476 printk(KERN_NOTICE "Add device[%d]: osd_name-%s\n",
477 i, odi.osdname);
478
479 /* On all devices the device table is identical. The user can
480 * specify any one of the participating devices on the command
481 * line. We always keep them in device-table order.
482 */
483 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
484 sbi->layout.s_ods[i] = fscb_od;
485 ++sbi->layout.s_numdevs;
486 fscb_od = NULL;
487 continue;
488 }
489
490 od = osduld_info_lookup(&odi);
491 if (unlikely(IS_ERR(od))) {
492 ret = PTR_ERR(od);
493 EXOFS_ERR("ERROR: device requested is not found "
494 "osd_name-%s =>%d\n", odi.osdname, ret);
495 goto out;
496 }
497
498 sbi->layout.s_ods[i] = od;
499 ++sbi->layout.s_numdevs;
500
501 /* Read the fscb of the other devices to make sure the FS
502 * partition is there.
503 */
504 ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb,
505 sizeof(fscb));
506 if (unlikely(ret)) {
507 EXOFS_ERR("ERROR: Malformed participating device "
508 "error reading fscb osd_name-%s\n",
509 odi.osdname);
510 goto out;
511 }
512
513 /* TODO: verify other information is correct and FS-uuid
514 * matches. Benny what did you say about device table
515 * generation and old devices?
516 */
517 }
518
519out:
520 kfree(dt);
521 if (unlikely(!ret && fscb_od)) {
522 EXOFS_ERR(
523 "ERROR: Bad device-table container device not present\n");
524 osduld_put_device(fscb_od);
525 ret = -EINVAL;
526 }
527
528 return ret;
529}
530
287/* 531/*
288 * Read the superblock from the OSD and fill in the fields 532 * Read the superblock from the OSD and fill in the fields
289 */ 533 */
@@ -292,25 +536,32 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
292 struct inode *root; 536 struct inode *root;
293 struct exofs_mountopt *opts = data; 537 struct exofs_mountopt *opts = data;
294 struct exofs_sb_info *sbi; /*extended info */ 538 struct exofs_sb_info *sbi; /*extended info */
539 struct osd_dev *od; /* Master device */
295 struct exofs_fscb fscb; /*on-disk superblock info */ 540 struct exofs_fscb fscb; /*on-disk superblock info */
296 struct osd_request *or = NULL;
297 struct osd_obj_id obj; 541 struct osd_obj_id obj;
542 unsigned table_count;
298 int ret; 543 int ret;
299 544
300 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL); 545 sbi = kzalloc(sizeof(*sbi), GFP_KERNEL);
301 if (!sbi) 546 if (!sbi)
302 return -ENOMEM; 547 return -ENOMEM;
303 sb->s_fs_info = sbi;
304 548
305 /* use mount options to fill superblock */ 549 /* use mount options to fill superblock */
306 sbi->s_dev = osduld_path_lookup(opts->dev_name); 550 od = osduld_path_lookup(opts->dev_name);
307 if (IS_ERR(sbi->s_dev)) { 551 if (IS_ERR(od)) {
308 ret = PTR_ERR(sbi->s_dev); 552 ret = PTR_ERR(od);
309 sbi->s_dev = NULL;
310 goto free_sbi; 553 goto free_sbi;
311 } 554 }
312 555
313 sbi->s_pid = opts->pid; 556 /* Default layout in case we do not have a device-table */
557 sbi->layout.stripe_unit = PAGE_SIZE;
558 sbi->layout.mirrors_p1 = 1;
559 sbi->layout.group_width = 1;
560 sbi->layout.group_depth = -1;
561 sbi->layout.group_count = 1;
562 sbi->layout.s_ods[0] = od;
563 sbi->layout.s_numdevs = 1;
564 sbi->layout.s_pid = opts->pid;
314 sbi->s_timeout = opts->timeout; 565 sbi->s_timeout = opts->timeout;
315 566
316 /* fill in some other data by hand */ 567 /* fill in some other data by hand */
@@ -323,35 +574,13 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
323 sb->s_bdev = NULL; 574 sb->s_bdev = NULL;
324 sb->s_dev = 0; 575 sb->s_dev = 0;
325 576
326 /* read data from on-disk superblock object */ 577 obj.partition = sbi->layout.s_pid;
327 obj.partition = sbi->s_pid;
328 obj.id = EXOFS_SUPER_ID; 578 obj.id = EXOFS_SUPER_ID;
329 exofs_make_credential(sbi->s_cred, &obj); 579 exofs_make_credential(sbi->s_cred, &obj);
330 580
331 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 581 ret = exofs_read_kern(od, sbi->s_cred, &obj, 0, &fscb, sizeof(fscb));
332 if (unlikely(!or)) { 582 if (unlikely(ret))
333 if (!silent)
334 EXOFS_ERR(
335 "exofs_fill_super: osd_start_request failed.\n");
336 ret = -ENOMEM;
337 goto free_sbi;
338 }
339 ret = osd_req_read_kern(or, &obj, 0, &fscb, sizeof(fscb));
340 if (unlikely(ret)) {
341 if (!silent)
342 EXOFS_ERR(
343 "exofs_fill_super: osd_req_read_kern failed.\n");
344 ret = -ENOMEM;
345 goto free_sbi;
346 }
347
348 ret = exofs_sync_op(or, sbi->s_timeout, sbi->s_cred);
349 if (unlikely(ret)) {
350 if (!silent)
351 EXOFS_ERR("exofs_fill_super: exofs_sync_op failed.\n");
352 ret = -EIO;
353 goto free_sbi; 583 goto free_sbi;
354 }
355 584
356 sb->s_magic = le16_to_cpu(fscb.s_magic); 585 sb->s_magic = le16_to_cpu(fscb.s_magic);
357 sbi->s_nextid = le64_to_cpu(fscb.s_nextid); 586 sbi->s_nextid = le64_to_cpu(fscb.s_nextid);
@@ -364,12 +593,26 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
364 ret = -EINVAL; 593 ret = -EINVAL;
365 goto free_sbi; 594 goto free_sbi;
366 } 595 }
596 if (le32_to_cpu(fscb.s_version) != EXOFS_FSCB_VER) {
597 EXOFS_ERR("ERROR: Bad FSCB version expected-%d got-%d\n",
598 EXOFS_FSCB_VER, le32_to_cpu(fscb.s_version));
599 ret = -EINVAL;
600 goto free_sbi;
601 }
367 602
368 /* start generation numbers from a random point */ 603 /* start generation numbers from a random point */
369 get_random_bytes(&sbi->s_next_generation, sizeof(u32)); 604 get_random_bytes(&sbi->s_next_generation, sizeof(u32));
370 spin_lock_init(&sbi->s_next_gen_lock); 605 spin_lock_init(&sbi->s_next_gen_lock);
371 606
607 table_count = le64_to_cpu(fscb.s_dev_table_count);
608 if (table_count) {
609 ret = exofs_read_lookup_dev_table(&sbi, table_count);
610 if (unlikely(ret))
611 goto free_sbi;
612 }
613
372 /* set up operation vectors */ 614 /* set up operation vectors */
615 sb->s_fs_info = sbi;
373 sb->s_op = &exofs_sops; 616 sb->s_op = &exofs_sops;
374 sb->s_export_op = &exofs_export_ops; 617 sb->s_export_op = &exofs_export_ops;
375 root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF); 618 root = exofs_iget(sb, EXOFS_ROOT_ID - EXOFS_OBJ_OFF);
@@ -395,16 +638,15 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
395 goto free_sbi; 638 goto free_sbi;
396 } 639 }
397 640
398 ret = 0; 641 _exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
399out: 642 sbi->layout.s_pid);
400 if (or) 643 return 0;
401 osd_end_request(or);
402 return ret;
403 644
404free_sbi: 645free_sbi:
405 osduld_put_device(sbi->s_dev); /* NULL safe */ 646 EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
406 kfree(sbi); 647 opts->dev_name, sbi->layout.s_pid, ret);
407 goto out; 648 exofs_free_sbi(sbi);
649 return ret;
408} 650}
409 651
410/* 652/*
@@ -433,7 +675,7 @@ static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
433{ 675{
434 struct super_block *sb = dentry->d_sb; 676 struct super_block *sb = dentry->d_sb;
435 struct exofs_sb_info *sbi = sb->s_fs_info; 677 struct exofs_sb_info *sbi = sb->s_fs_info;
436 struct osd_obj_id obj = {sbi->s_pid, 0}; 678 struct exofs_io_state *ios;
437 struct osd_attr attrs[] = { 679 struct osd_attr attrs[] = {
438 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS, 680 ATTR_DEF(OSD_APAGE_PARTITION_QUOTAS,
439 OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)), 681 OSD_ATTR_PQ_CAPACITY_QUOTA, sizeof(__be64)),
@@ -442,32 +684,33 @@ static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
442 }; 684 };
443 uint64_t capacity = ULLONG_MAX; 685 uint64_t capacity = ULLONG_MAX;
444 uint64_t used = ULLONG_MAX; 686 uint64_t used = ULLONG_MAX;
445 struct osd_request *or;
446 uint8_t cred_a[OSD_CAP_LEN]; 687 uint8_t cred_a[OSD_CAP_LEN];
447 int ret; 688 int ret;
448 689
449 /* get used/capacity attributes */ 690 ret = exofs_get_io_state(&sbi->layout, &ios);
450 exofs_make_credential(cred_a, &obj); 691 if (ret) {
451 692 EXOFS_DBGMSG("exofs_get_io_state failed.\n");
452 or = osd_start_request(sbi->s_dev, GFP_KERNEL); 693 return ret;
453 if (unlikely(!or)) {
454 EXOFS_DBGMSG("exofs_statfs: osd_start_request failed.\n");
455 return -ENOMEM;
456 } 694 }
457 695
458 osd_req_get_attributes(or, &obj); 696 exofs_make_credential(cred_a, &ios->obj);
459 osd_req_add_get_attr_list(or, attrs, ARRAY_SIZE(attrs)); 697 ios->cred = sbi->s_cred;
460 ret = exofs_sync_op(or, sbi->s_timeout, cred_a); 698 ios->in_attr = attrs;
699 ios->in_attr_len = ARRAY_SIZE(attrs);
700
701 ret = exofs_sbi_read(ios);
461 if (unlikely(ret)) 702 if (unlikely(ret))
462 goto out; 703 goto out;
463 704
464 ret = extract_attr_from_req(or, &attrs[0]); 705 ret = extract_attr_from_ios(ios, &attrs[0]);
465 if (likely(!ret)) 706 if (likely(!ret)) {
466 capacity = get_unaligned_be64(attrs[0].val_ptr); 707 capacity = get_unaligned_be64(attrs[0].val_ptr);
467 else 708 if (unlikely(!capacity))
709 capacity = ULLONG_MAX;
710 } else
468 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n"); 711 EXOFS_DBGMSG("exofs_statfs: get capacity failed.\n");
469 712
470 ret = extract_attr_from_req(or, &attrs[1]); 713 ret = extract_attr_from_ios(ios, &attrs[1]);
471 if (likely(!ret)) 714 if (likely(!ret))
472 used = get_unaligned_be64(attrs[1].val_ptr); 715 used = get_unaligned_be64(attrs[1].val_ptr);
473 else 716 else
@@ -476,15 +719,15 @@ static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
476 /* fill in the stats buffer */ 719 /* fill in the stats buffer */
477 buf->f_type = EXOFS_SUPER_MAGIC; 720 buf->f_type = EXOFS_SUPER_MAGIC;
478 buf->f_bsize = EXOFS_BLKSIZE; 721 buf->f_bsize = EXOFS_BLKSIZE;
479 buf->f_blocks = (capacity >> EXOFS_BLKSHIFT); 722 buf->f_blocks = capacity >> 9;
480 buf->f_bfree = ((capacity - used) >> EXOFS_BLKSHIFT); 723 buf->f_bfree = (capacity - used) >> 9;
481 buf->f_bavail = buf->f_bfree; 724 buf->f_bavail = buf->f_bfree;
482 buf->f_files = sbi->s_numfiles; 725 buf->f_files = sbi->s_numfiles;
483 buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles; 726 buf->f_ffree = EXOFS_MAX_ID - sbi->s_numfiles;
484 buf->f_namelen = EXOFS_NAME_LEN; 727 buf->f_namelen = EXOFS_NAME_LEN;
485 728
486out: 729out:
487 osd_end_request(or); 730 exofs_put_io_state(ios);
488 return ret; 731 return ret;
489} 732}
490 733
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-22 02:10:59 -0500