aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
Diffstat
-rw-r--r--fs/exofs/common.h39
-rw-r--r--fs/exofs/exofs.h53
-rw-r--r--fs/exofs/inode.c194
-rw-r--r--fs/exofs/ios.c575
-rw-r--r--fs/exofs/super.c121
5 files changed, 779 insertions, 203 deletions
diff --git a/fs/exofs/common.h b/fs/exofs/common.h
index b1b178e61718..f0d520312d8b 100644
--- a/fs/exofs/common.h
+++ b/fs/exofs/common.h
@@ -55,6 +55,8 @@
55/* exofs Application specific page/attribute */ 55/* exofs Application specific page/attribute */
56# define EXOFS_APAGE_FS_DATA (OSD_APAGE_APP_DEFINED_FIRST + 3) 56# define EXOFS_APAGE_FS_DATA (OSD_APAGE_APP_DEFINED_FIRST + 3)
57# 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
58 60
59/* 61/*
60 * 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
@@ -206,4 +208,41 @@ enum {
206 (((name_len) + offsetof(struct exofs_dir_entry, name) + \ 208 (((name_len) + offsetof(struct exofs_dir_entry, name) + \
207 EXOFS_DIR_ROUND) & ~EXOFS_DIR_ROUND) 209 EXOFS_DIR_ROUND) & ~EXOFS_DIR_ROUND)
208 210
211/*
212 * The on-disk (optional) layout structure.
213 * sits in an EXOFS_ATTR_INODE_FILE_LAYOUT or EXOFS_ATTR_INODE_DIR_LAYOUT
214 * attribute, attached to any inode, usually to a directory.
215 */
216
217enum exofs_inode_layout_gen_functions {
218 LAYOUT_MOVING_WINDOW = 0,
219 LAYOUT_IMPLICT = 1,
220};
221
222struct exofs_on_disk_inode_layout {
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)
243{
244 return sizeof(struct exofs_on_disk_inode_layout) +
245 max_devs * sizeof(__le32);
246}
247
209#endif /*ifndef __EXOFS_COM_H__*/ 248#endif /*ifndef __EXOFS_COM_H__*/
diff --git a/fs/exofs/exofs.h b/fs/exofs/exofs.h
index c35fd4623986..59b8bf2825c7 100644
--- a/fs/exofs/exofs.h
+++ b/fs/exofs/exofs.h
@@ -55,12 +55,28 @@
55/* 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 */
56#define _LLU(x) (unsigned long long)(x) 56#define _LLU(x) (unsigned long long)(x)
57 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
58/* 75/*
59 * our extension to the in-memory superblock 76 * our extension to the in-memory superblock
60 */ 77 */
61struct exofs_sb_info { 78struct exofs_sb_info {
62 struct exofs_fscb s_fscb; /* Written often, pre-allocate*/ 79 struct exofs_fscb s_fscb; /* Written often, pre-allocate*/
63 osd_id s_pid; /* partition ID of file system*/
64 int s_timeout; /* timeout for OSD operations */ 80 int s_timeout; /* timeout for OSD operations */
65 uint64_t s_nextid; /* highest object ID used */ 81 uint64_t s_nextid; /* highest object ID used */
66 uint32_t s_numfiles; /* number of files on fs */ 82 uint32_t s_numfiles; /* number of files on fs */
@@ -69,22 +85,27 @@ struct exofs_sb_info {
69 atomic_t s_curr_pending; /* number of pending commands */ 85 atomic_t s_curr_pending; /* number of pending commands */
70 uint8_t s_cred[OSD_CAP_LEN]; /* credential for the fscb */ 86 uint8_t s_cred[OSD_CAP_LEN]; /* credential for the fscb */
71 87
72 struct pnfs_osd_data_map data_map; /* Default raid to use */ 88 struct pnfs_osd_data_map data_map; /* Default raid to use
73 unsigned s_numdevs; /* Num of devices in array */ 89 * FIXME: Needed ?
74 struct osd_dev *s_ods[1]; /* Variable length, minimum 1 */ 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 */
75}; 96};
76 97
77/* 98/*
78 * our extension to the in-memory inode 99 * our extension to the in-memory inode
79 */ 100 */
80struct 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 */
81 unsigned long i_flags; /* various atomic flags */ 104 unsigned long i_flags; /* various atomic flags */
82 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*/
83 uint32_t i_dir_start_lookup; /* which page to start lookup */ 106 uint32_t i_dir_start_lookup; /* which page to start lookup */
84 wait_queue_head_t i_wq; /* wait queue for inode */
85 uint64_t i_commit_size; /* the object's written length */ 107 uint64_t i_commit_size; /* the object's written length */
86 uint8_t i_cred[OSD_CAP_LEN];/* all-powerful credential */ 108 uint8_t i_cred[OSD_CAP_LEN];/* all-powerful credential */
87 struct inode vfs_inode; /* normal in-memory inode */
88}; 109};
89 110
90static inline osd_id exofs_oi_objno(struct exofs_i_info *oi) 111static inline osd_id exofs_oi_objno(struct exofs_i_info *oi)
@@ -101,7 +122,7 @@ struct exofs_io_state {
101 void *private; 122 void *private;
102 exofs_io_done_fn done; 123 exofs_io_done_fn done;
103 124
104 struct exofs_sb_info *sbi; 125 struct exofs_layout *layout;
105 struct osd_obj_id obj; 126 struct osd_obj_id obj;
106 u8 *cred; 127 u8 *cred;
107 128
@@ -109,7 +130,11 @@ struct exofs_io_state {
109 loff_t offset; 130 loff_t offset;
110 unsigned long length; 131 unsigned long length;
111 void *kern_buff; 132 void *kern_buff;
112 struct bio *bio; 133
134 struct page **pages;
135 unsigned nr_pages;
136 unsigned pgbase;
137 unsigned pages_consumed;
113 138
114 /* Attributes */ 139 /* Attributes */
115 unsigned in_attr_len; 140 unsigned in_attr_len;
@@ -122,6 +147,9 @@ struct exofs_io_state {
122 struct exofs_per_dev_state { 147 struct exofs_per_dev_state {
123 struct osd_request *or; 148 struct osd_request *or;
124 struct bio *bio; 149 struct bio *bio;
150 loff_t offset;
151 unsigned length;
152 unsigned dev;
125 } per_dev[]; 153 } per_dev[];
126}; 154};
127 155
@@ -175,6 +203,12 @@ static inline struct exofs_i_info *exofs_i(struct inode *inode)
175} 203}
176 204
177/* 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/*
178 * Maximum count of links to a file 212 * Maximum count of links to a file
179 */ 213 */
180#define EXOFS_LINK_MAX 32000 214#define EXOFS_LINK_MAX 32000
@@ -189,7 +223,8 @@ void exofs_make_credential(u8 cred_a[OSD_CAP_LEN],
189int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj, 223int exofs_read_kern(struct osd_dev *od, u8 *cred, struct osd_obj_id *obj,
190 u64 offset, void *p, unsigned length); 224 u64 offset, void *p, unsigned length);
191 225
192int exofs_get_io_state(struct exofs_sb_info *sbi, struct exofs_io_state** ios); 226int exofs_get_io_state(struct exofs_layout *layout,
227 struct exofs_io_state **ios);
193void exofs_put_io_state(struct exofs_io_state *ios); 228void exofs_put_io_state(struct exofs_io_state *ios);
194 229
195int exofs_check_io(struct exofs_io_state *ios, u64 *resid); 230int exofs_check_io(struct exofs_io_state *ios, u64 *resid);
diff --git a/fs/exofs/inode.c b/fs/exofs/inode.c
index 2afbcebeda71..5514f3c2c2f4 100644
--- a/fs/exofs/inode.c
+++ b/fs/exofs/inode.c
@@ -41,16 +41,18 @@
41 41
42enum { BIO_MAX_PAGES_KMALLOC = 42enum { BIO_MAX_PAGES_KMALLOC =
43 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec), 43 (PAGE_SIZE - sizeof(struct bio)) / sizeof(struct bio_vec),
44 MAX_PAGES_KMALLOC =
45 PAGE_SIZE / sizeof(struct page *),
44}; 46};
45 47
46struct page_collect { 48struct page_collect {
47 struct exofs_sb_info *sbi; 49 struct exofs_sb_info *sbi;
48 struct request_queue *req_q;
49 struct inode *inode; 50 struct inode *inode;
50 unsigned expected_pages; 51 unsigned expected_pages;
51 struct exofs_io_state *ios; 52 struct exofs_io_state *ios;
52 53
53 struct bio *bio; 54 struct page **pages;
55 unsigned alloc_pages;
54 unsigned nr_pages; 56 unsigned nr_pages;
55 unsigned long length; 57 unsigned long length;
56 loff_t pg_first; /* keep 64bit also in 32-arches */ 58 loff_t pg_first; /* keep 64bit also in 32-arches */
@@ -62,15 +64,12 @@ static void _pcol_init(struct page_collect *pcol, unsigned expected_pages,
62 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info; 64 struct exofs_sb_info *sbi = inode->i_sb->s_fs_info;
63 65
64 pcol->sbi = sbi; 66 pcol->sbi = sbi;
65 /* Create master bios on first Q, later on cloning, each clone will be
66 * allocated on it's destination Q
67 */
68 pcol->req_q = osd_request_queue(sbi->s_ods[0]);
69 pcol->inode = inode; 67 pcol->inode = inode;
70 pcol->expected_pages = expected_pages; 68 pcol->expected_pages = expected_pages;
71 69
72 pcol->ios = NULL; 70 pcol->ios = NULL;
73 pcol->bio = NULL; 71 pcol->pages = NULL;
72 pcol->alloc_pages = 0;
74 pcol->nr_pages = 0; 73 pcol->nr_pages = 0;
75 pcol->length = 0; 74 pcol->length = 0;
76 pcol->pg_first = -1; 75 pcol->pg_first = -1;
@@ -80,7 +79,8 @@ static void _pcol_reset(struct page_collect *pcol)
80{ 79{
81 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages); 80 pcol->expected_pages -= min(pcol->nr_pages, pcol->expected_pages);
82 81
83 pcol->bio = NULL; 82 pcol->pages = NULL;
83 pcol->alloc_pages = 0;
84 pcol->nr_pages = 0; 84 pcol->nr_pages = 0;
85 pcol->length = 0; 85 pcol->length = 0;
86 pcol->pg_first = -1; 86 pcol->pg_first = -1;
@@ -90,38 +90,43 @@ static void _pcol_reset(struct page_collect *pcol)
90 * it might not end here. don't be left with nothing 90 * it might not end here. don't be left with nothing
91 */ 91 */
92 if (!pcol->expected_pages) 92 if (!pcol->expected_pages)
93 pcol->expected_pages = BIO_MAX_PAGES_KMALLOC; 93 pcol->expected_pages = MAX_PAGES_KMALLOC;
94} 94}
95 95
96static int pcol_try_alloc(struct page_collect *pcol) 96static int pcol_try_alloc(struct page_collect *pcol)
97{ 97{
98 int pages = min_t(unsigned, pcol->expected_pages, 98 unsigned pages = min_t(unsigned, pcol->expected_pages,
99 BIO_MAX_PAGES_KMALLOC); 99 MAX_PAGES_KMALLOC);
100 100
101 if (!pcol->ios) { /* First time allocate io_state */ 101 if (!pcol->ios) { /* First time allocate io_state */
102 int ret = exofs_get_io_state(pcol->sbi, &pcol->ios); 102 int ret = exofs_get_io_state(&pcol->sbi->layout, &pcol->ios);
103 103
104 if (ret) 104 if (ret)
105 return ret; 105 return ret;
106 } 106 }
107 107
108 /* TODO: easily support bio chaining */
109 pages = min_t(unsigned, pages,
110 pcol->sbi->layout.group_width * BIO_MAX_PAGES_KMALLOC);
111
108 for (; pages; pages >>= 1) { 112 for (; pages; pages >>= 1) {
109 pcol->bio = bio_kmalloc(GFP_KERNEL, pages); 113 pcol->pages = kmalloc(pages * sizeof(struct page *),
110 if (likely(pcol->bio)) 114 GFP_KERNEL);
115 if (likely(pcol->pages)) {
116 pcol->alloc_pages = pages;
111 return 0; 117 return 0;
118 }
112 } 119 }
113 120
114 EXOFS_ERR("Failed to bio_kmalloc expected_pages=%u\n", 121 EXOFS_ERR("Failed to kmalloc expected_pages=%u\n",
115 pcol->expected_pages); 122 pcol->expected_pages);
116 return -ENOMEM; 123 return -ENOMEM;
117} 124}
118 125
119static void pcol_free(struct page_collect *pcol) 126static void pcol_free(struct page_collect *pcol)
120{ 127{
121 if (pcol->bio) { 128 kfree(pcol->pages);
122 bio_put(pcol->bio); 129 pcol->pages = NULL;
123 pcol->bio = NULL;
124 }
125 130
126 if (pcol->ios) { 131 if (pcol->ios) {
127 exofs_put_io_state(pcol->ios); 132 exofs_put_io_state(pcol->ios);
@@ -132,11 +137,10 @@ static void pcol_free(struct page_collect *pcol)
132static int pcol_add_page(struct page_collect *pcol, struct page *page, 137static int pcol_add_page(struct page_collect *pcol, struct page *page,
133 unsigned len) 138 unsigned len)
134{ 139{
135 int added_len = bio_add_pc_page(pcol->req_q, pcol->bio, page, len, 0); 140 if (unlikely(pcol->nr_pages >= pcol->alloc_pages))
136 if (unlikely(len != added_len))
137 return -ENOMEM; 141 return -ENOMEM;
138 142
139 ++pcol->nr_pages; 143 pcol->pages[pcol->nr_pages++] = page;
140 pcol->length += len; 144 pcol->length += len;
141 return 0; 145 return 0;
142} 146}
@@ -181,7 +185,6 @@ static void update_write_page(struct page *page, int ret)
181 */ 185 */
182static int __readpages_done(struct page_collect *pcol, bool do_unlock) 186static int __readpages_done(struct page_collect *pcol, bool do_unlock)
183{ 187{
184 struct bio_vec *bvec;
185 int i; 188 int i;
186 u64 resid; 189 u64 resid;
187 u64 good_bytes; 190 u64 good_bytes;
@@ -193,13 +196,13 @@ static int __readpages_done(struct page_collect *pcol, bool do_unlock)
193 else 196 else
194 good_bytes = pcol->length - resid; 197 good_bytes = pcol->length - resid;
195 198
196 EXOFS_DBGMSG("readpages_done(0x%lx) good_bytes=0x%llx" 199 EXOFS_DBGMSG2("readpages_done(0x%lx) good_bytes=0x%llx"
197 " length=0x%lx nr_pages=%u\n", 200 " length=0x%lx nr_pages=%u\n",
198 pcol->inode->i_ino, _LLU(good_bytes), pcol->length, 201 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
199 pcol->nr_pages); 202 pcol->nr_pages);
200 203
201 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 204 for (i = 0; i < pcol->nr_pages; i++) {
202 struct page *page = bvec->bv_page; 205 struct page *page = pcol->pages[i];
203 struct inode *inode = page->mapping->host; 206 struct inode *inode = page->mapping->host;
204 int page_stat; 207 int page_stat;
205 208
@@ -218,11 +221,11 @@ static int __readpages_done(struct page_collect *pcol, bool do_unlock)
218 ret = update_read_page(page, page_stat); 221 ret = update_read_page(page, page_stat);
219 if (do_unlock) 222 if (do_unlock)
220 unlock_page(page); 223 unlock_page(page);
221 length += bvec->bv_len; 224 length += PAGE_SIZE;
222 } 225 }
223 226
224 pcol_free(pcol); 227 pcol_free(pcol);
225 EXOFS_DBGMSG("readpages_done END\n"); 228 EXOFS_DBGMSG2("readpages_done END\n");
226 return ret; 229 return ret;
227} 230}
228 231
@@ -238,11 +241,10 @@ static void readpages_done(struct exofs_io_state *ios, void *p)
238 241
239static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw) 242static void _unlock_pcol_pages(struct page_collect *pcol, int ret, int rw)
240{ 243{
241 struct bio_vec *bvec;
242 int i; 244 int i;
243 245
244 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 246 for (i = 0; i < pcol->nr_pages; i++) {
245 struct page *page = bvec->bv_page; 247 struct page *page = pcol->pages[i];
246 248
247 if (rw == READ) 249 if (rw == READ)
248 update_read_page(page, ret); 250 update_read_page(page, ret);
@@ -260,13 +262,14 @@ static int read_exec(struct page_collect *pcol, bool is_sync)
260 struct page_collect *pcol_copy = NULL; 262 struct page_collect *pcol_copy = NULL;
261 int ret; 263 int ret;
262 264
263 if (!pcol->bio) 265 if (!pcol->pages)
264 return 0; 266 return 0;
265 267
266 /* see comment in _readpage() about sync reads */ 268 /* see comment in _readpage() about sync reads */
267 WARN_ON(is_sync && (pcol->nr_pages != 1)); 269 WARN_ON(is_sync && (pcol->nr_pages != 1));
268 270
269 ios->bio = pcol->bio; 271 ios->pages = pcol->pages;
272 ios->nr_pages = pcol->nr_pages;
270 ios->length = pcol->length; 273 ios->length = pcol->length;
271 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT; 274 ios->offset = pcol->pg_first << PAGE_CACHE_SHIFT;
272 275
@@ -290,7 +293,7 @@ static int read_exec(struct page_collect *pcol, bool is_sync)
290 293
291 atomic_inc(&pcol->sbi->s_curr_pending); 294 atomic_inc(&pcol->sbi->s_curr_pending);
292 295
293 EXOFS_DBGMSG("read_exec obj=0x%llx start=0x%llx length=0x%lx\n", 296 EXOFS_DBGMSG2("read_exec obj=0x%llx start=0x%llx length=0x%lx\n",
294 ios->obj.id, _LLU(ios->offset), pcol->length); 297 ios->obj.id, _LLU(ios->offset), pcol->length);
295 298
296 /* pages ownership was passed to pcol_copy */ 299 /* pages ownership was passed to pcol_copy */
@@ -366,7 +369,7 @@ try_again:
366 goto try_again; 369 goto try_again;
367 } 370 }
368 371
369 if (!pcol->bio) { 372 if (!pcol->pages) {
370 ret = pcol_try_alloc(pcol); 373 ret = pcol_try_alloc(pcol);
371 if (unlikely(ret)) 374 if (unlikely(ret))
372 goto fail; 375 goto fail;
@@ -448,7 +451,6 @@ static int exofs_readpage(struct file *file, struct page *page)
448static void writepages_done(struct exofs_io_state *ios, void *p) 451static void writepages_done(struct exofs_io_state *ios, void *p)
449{ 452{
450 struct page_collect *pcol = p; 453 struct page_collect *pcol = p;
451 struct bio_vec *bvec;
452 int i; 454 int i;
453 u64 resid; 455 u64 resid;
454 u64 good_bytes; 456 u64 good_bytes;
@@ -462,13 +464,13 @@ static void writepages_done(struct exofs_io_state *ios, void *p)
462 else 464 else
463 good_bytes = pcol->length - resid; 465 good_bytes = pcol->length - resid;
464 466
465 EXOFS_DBGMSG("writepages_done(0x%lx) good_bytes=0x%llx" 467 EXOFS_DBGMSG2("writepages_done(0x%lx) good_bytes=0x%llx"
466 " length=0x%lx nr_pages=%u\n", 468 " length=0x%lx nr_pages=%u\n",
467 pcol->inode->i_ino, _LLU(good_bytes), pcol->length, 469 pcol->inode->i_ino, _LLU(good_bytes), pcol->length,
468 pcol->nr_pages); 470 pcol->nr_pages);
469 471
470 __bio_for_each_segment(bvec, pcol->bio, i, 0) { 472 for (i = 0; i < pcol->nr_pages; i++) {
471 struct page *page = bvec->bv_page; 473 struct page *page = pcol->pages[i];
472 struct inode *inode = page->mapping->host; 474 struct inode *inode = page->mapping->host;
473 int page_stat; 475 int page_stat;
474 476
@@ -485,12 +487,12 @@ static void writepages_done(struct exofs_io_state *ios, void *p)
485 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n", 487 EXOFS_DBGMSG2(" writepages_done(0x%lx, 0x%lx) status=%d\n",
486 inode->i_ino, page->index, page_stat); 488 inode->i_ino, page->index, page_stat);
487 489
488 length += bvec->bv_len; 490 length += PAGE_SIZE;
489 } 491 }
490 492
491 pcol_free(pcol); 493 pcol_free(pcol);
492 kfree(pcol); 494 kfree(pcol);
493 EXOFS_DBGMSG("writepages_done END\n"); 495 EXOFS_DBGMSG2("writepages_done END\n");
494} 496}
495 497
496static int write_exec(struct page_collect *pcol) 498static int write_exec(struct page_collect *pcol)
@@ -500,7 +502,7 @@ static int write_exec(struct page_collect *pcol)
500 struct page_collect *pcol_copy = NULL; 502 struct page_collect *pcol_copy = NULL;
501 int ret; 503 int ret;
502 504
503 if (!pcol->bio) 505 if (!pcol->pages)
504 return 0; 506 return 0;
505 507
506 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL); 508 pcol_copy = kmalloc(sizeof(*pcol_copy), GFP_KERNEL);
@@ -512,9 +514,8 @@ static int write_exec(struct page_collect *pcol)
512 514
513 *pcol_copy = *pcol; 515 *pcol_copy = *pcol;
514 516
515 pcol_copy->bio->bi_rw |= (1 << BIO_RW); /* FIXME: bio_set_dir() */ 517 ios->pages = pcol_copy->pages;
516 518 ios->nr_pages = pcol_copy->nr_pages;
517 ios->bio = pcol_copy->bio;
518 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT; 519 ios->offset = pcol_copy->pg_first << PAGE_CACHE_SHIFT;
519 ios->length = pcol_copy->length; 520 ios->length = pcol_copy->length;
520 ios->done = writepages_done; 521 ios->done = writepages_done;
@@ -527,7 +528,7 @@ static int write_exec(struct page_collect *pcol)
527 } 528 }
528 529
529 atomic_inc(&pcol->sbi->s_curr_pending); 530 atomic_inc(&pcol->sbi->s_curr_pending);
530 EXOFS_DBGMSG("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n", 531 EXOFS_DBGMSG2("write_exec(0x%lx, 0x%llx) start=0x%llx length=0x%lx\n",
531 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset), 532 pcol->inode->i_ino, pcol->pg_first, _LLU(ios->offset),
532 pcol->length); 533 pcol->length);
533 /* pages ownership was passed to pcol_copy */ 534 /* pages ownership was passed to pcol_copy */
@@ -605,7 +606,7 @@ try_again:
605 goto try_again; 606 goto try_again;
606 } 607 }
607 608
608 if (!pcol->bio) { 609 if (!pcol->pages) {
609 ret = pcol_try_alloc(pcol); 610 ret = pcol_try_alloc(pcol);
610 if (unlikely(ret)) 611 if (unlikely(ret))
611 goto fail; 612 goto fail;
@@ -616,7 +617,7 @@ try_again:
616 617
617 ret = pcol_add_page(pcol, page, len); 618 ret = pcol_add_page(pcol, page, len);
618 if (unlikely(ret)) { 619 if (unlikely(ret)) {
619 EXOFS_DBGMSG("Failed pcol_add_page " 620 EXOFS_DBGMSG2("Failed pcol_add_page "
620 "nr_pages=%u total_length=0x%lx\n", 621 "nr_pages=%u total_length=0x%lx\n",
621 pcol->nr_pages, pcol->length); 622 pcol->nr_pages, pcol->length);
622 623
@@ -663,7 +664,7 @@ static int exofs_writepages(struct address_space *mapping,
663 if (expected_pages < 32L) 664 if (expected_pages < 32L)
664 expected_pages = 32L; 665 expected_pages = 32L;
665 666
666 EXOFS_DBGMSG("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx " 667 EXOFS_DBGMSG2("inode(0x%lx) wbc->start=0x%llx wbc->end=0x%llx "
667 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n", 668 "nrpages=%lu start=0x%lx end=0x%lx expected_pages=%ld\n",
668 mapping->host->i_ino, wbc->range_start, wbc->range_end, 669 mapping->host->i_ino, wbc->range_start, wbc->range_end,
669 mapping->nrpages, start, end, expected_pages); 670 mapping->nrpages, start, end, expected_pages);
@@ -859,20 +860,33 @@ int exofs_setattr(struct dentry *dentry, struct iattr *iattr)
859 return error; 860 return error;
860} 861}
861 862
863static const struct osd_attr g_attr_inode_file_layout = ATTR_DEF(
864 EXOFS_APAGE_FS_DATA,
865 EXOFS_ATTR_INODE_FILE_LAYOUT,
866 0);
867static const struct osd_attr g_attr_inode_dir_layout = ATTR_DEF(
868 EXOFS_APAGE_FS_DATA,
869 EXOFS_ATTR_INODE_DIR_LAYOUT,
870 0);
871
862/* 872/*
863 * Read an inode from the OSD, and return it as is. We also return the size 873 * Read the Linux inode info from the OSD, and return it as is. In exofs the
864 * attribute in the 'obj_size' argument. 874 * inode info is in an application specific page/attribute of the osd-object.
865 */ 875 */
866static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi, 876static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
867 struct exofs_fcb *inode, uint64_t *obj_size) 877 struct exofs_fcb *inode)
868{ 878{
869 struct exofs_sb_info *sbi = sb->s_fs_info; 879 struct exofs_sb_info *sbi = sb->s_fs_info;
870 struct osd_attr attrs[2]; 880 struct osd_attr attrs[] = {
881 [0] = g_attr_inode_data,
882 [1] = g_attr_inode_file_layout,
883 [2] = g_attr_inode_dir_layout,
884 };
871 struct exofs_io_state *ios; 885 struct exofs_io_state *ios;
886 struct exofs_on_disk_inode_layout *layout;
872 int ret; 887 int ret;
873 888
874 *obj_size = ~0; 889 ret = exofs_get_io_state(&sbi->layout, &ios);
875 ret = exofs_get_io_state(sbi, &ios);
876 if (unlikely(ret)) { 890 if (unlikely(ret)) {
877 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__); 891 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
878 return ret; 892 return ret;
@@ -882,14 +896,25 @@ static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
882 exofs_make_credential(oi->i_cred, &ios->obj); 896 exofs_make_credential(oi->i_cred, &ios->obj);
883 ios->cred = oi->i_cred; 897 ios->cred = oi->i_cred;
884 898
885 attrs[0] = g_attr_inode_data; 899 attrs[1].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
886 attrs[1] = g_attr_logical_length; 900 attrs[2].len = exofs_on_disk_inode_layout_size(sbi->layout.s_numdevs);
901
887 ios->in_attr = attrs; 902 ios->in_attr = attrs;
888 ios->in_attr_len = ARRAY_SIZE(attrs); 903 ios->in_attr_len = ARRAY_SIZE(attrs);
889 904
890 ret = exofs_sbi_read(ios); 905 ret = exofs_sbi_read(ios);
891 if (ret) 906 if (unlikely(ret)) {
907 EXOFS_ERR("object(0x%llx) corrupted, return empty file=>%d\n",
908 _LLU(ios->obj.id), ret);
909 memset(inode, 0, sizeof(*inode));
910 inode->i_mode = 0040000 | (0777 & ~022);
911 /* If object is lost on target we might as well enable it's
912 * delete.
913 */
914 if ((ret == -ENOENT) || (ret == -EINVAL))
915 ret = 0;
892 goto out; 916 goto out;
917 }
893 918
894 ret = extract_attr_from_ios(ios, &attrs[0]); 919 ret = extract_attr_from_ios(ios, &attrs[0]);
895 if (ret) { 920 if (ret) {
@@ -901,11 +926,33 @@ static int exofs_get_inode(struct super_block *sb, struct exofs_i_info *oi,
901 926
902 ret = extract_attr_from_ios(ios, &attrs[1]); 927 ret = extract_attr_from_ios(ios, &attrs[1]);
903 if (ret) { 928 if (ret) {
904 EXOFS_ERR("%s: extract_attr of logical_length failed\n", 929 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
905 __func__); 930 goto out;
931 }
932 if (attrs[1].len) {
933 layout = attrs[1].val_ptr;
934 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
935 EXOFS_ERR("%s: unsupported files layout %d\n",
936 __func__, layout->gen_func);
937 ret = -ENOTSUPP;
938 goto out;
939 }
940 }
941
942 ret = extract_attr_from_ios(ios, &attrs[2]);
943 if (ret) {
944 EXOFS_ERR("%s: extract_attr of inode_data failed\n", __func__);
906 goto out; 945 goto out;
907 } 946 }
908 *obj_size = get_unaligned_be64(attrs[1].val_ptr); 947 if (attrs[2].len) {
948 layout = attrs[2].val_ptr;
949 if (layout->gen_func != cpu_to_le16(LAYOUT_MOVING_WINDOW)) {
950 EXOFS_ERR("%s: unsupported meta-data layout %d\n",
951 __func__, layout->gen_func);
952 ret = -ENOTSUPP;
953 goto out;
954 }
955 }
909 956
910out: 957out:
911 exofs_put_io_state(ios); 958 exofs_put_io_state(ios);
@@ -925,7 +972,6 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
925 struct exofs_i_info *oi; 972 struct exofs_i_info *oi;
926 struct exofs_fcb fcb; 973 struct exofs_fcb fcb;
927 struct inode *inode; 974 struct inode *inode;
928 uint64_t obj_size;
929 int ret; 975 int ret;
930 976
931 inode = iget_locked(sb, ino); 977 inode = iget_locked(sb, ino);
@@ -937,7 +983,7 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
937 __oi_init(oi); 983 __oi_init(oi);
938 984
939 /* read the inode from the osd */ 985 /* read the inode from the osd */
940 ret = exofs_get_inode(sb, oi, &fcb, &obj_size); 986 ret = exofs_get_inode(sb, oi, &fcb);
941 if (ret) 987 if (ret)
942 goto bad_inode; 988 goto bad_inode;
943 989
@@ -958,13 +1004,6 @@ struct inode *exofs_iget(struct super_block *sb, unsigned long ino)
958 inode->i_blkbits = EXOFS_BLKSHIFT; 1004 inode->i_blkbits = EXOFS_BLKSHIFT;
959 inode->i_generation = le32_to_cpu(fcb.i_generation); 1005 inode->i_generation = le32_to_cpu(fcb.i_generation);
960 1006
961 if ((inode->i_size != obj_size) &&
962 (!exofs_inode_is_fast_symlink(inode))) {
963 EXOFS_ERR("WARNING: Size of inode=%llu != object=%llu\n",
964 inode->i_size, _LLU(obj_size));
965 /* FIXME: call exofs_inode_recovery() */
966 }
967
968 oi->i_dir_start_lookup = 0; 1007 oi->i_dir_start_lookup = 0;
969 1008
970 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) { 1009 if ((inode->i_nlink == 0) && (inode->i_mode == 0)) {
@@ -1043,7 +1082,7 @@ static void create_done(struct exofs_io_state *ios, void *p)
1043 1082
1044 if (unlikely(ret)) { 1083 if (unlikely(ret)) {
1045 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx", 1084 EXOFS_ERR("object=0x%llx creation faild in pid=0x%llx",
1046 _LLU(exofs_oi_objno(oi)), _LLU(sbi->s_pid)); 1085 _LLU(exofs_oi_objno(oi)), _LLU(sbi->layout.s_pid));
1047 /*TODO: When FS is corrupted creation can fail, object already 1086 /*TODO: When FS is corrupted creation can fail, object already
1048 * exist. Get rid of this asynchronous creation, if exist 1087 * exist. Get rid of this asynchronous creation, if exist
1049 * increment the obj counter and try the next object. Until we 1088 * increment the obj counter and try the next object. Until we
@@ -1104,7 +1143,7 @@ struct inode *exofs_new_inode(struct inode *dir, int mode)
1104 1143
1105 mark_inode_dirty(inode); 1144 mark_inode_dirty(inode);
1106 1145
1107 ret = exofs_get_io_state(sbi, &ios); 1146 ret = exofs_get_io_state(&sbi->layout, &ios);
1108 if (unlikely(ret)) { 1147 if (unlikely(ret)) {
1109 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n"); 1148 EXOFS_ERR("exofs_new_inode: exofs_get_io_state failed\n");
1110 return ERR_PTR(ret); 1149 return ERR_PTR(ret);
@@ -1170,8 +1209,10 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1170 int ret; 1209 int ret;
1171 1210
1172 args = kzalloc(sizeof(*args), GFP_KERNEL); 1211 args = kzalloc(sizeof(*args), GFP_KERNEL);
1173 if (!args) 1212 if (!args) {
1213 EXOFS_DBGMSG("Faild kzalloc of args\n");
1174 return -ENOMEM; 1214 return -ENOMEM;
1215 }
1175 1216
1176 fcb = &args->fcb; 1217 fcb = &args->fcb;
1177 1218
@@ -1200,7 +1241,7 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1200 } else 1241 } else
1201 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data)); 1242 memcpy(fcb->i_data, oi->i_data, sizeof(fcb->i_data));
1202 1243
1203 ret = exofs_get_io_state(sbi, &ios); 1244 ret = exofs_get_io_state(&sbi->layout, &ios);
1204 if (unlikely(ret)) { 1245 if (unlikely(ret)) {
1205 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__); 1246 EXOFS_ERR("%s: exofs_get_io_state failed.\n", __func__);
1206 goto free_args; 1247 goto free_args;
@@ -1234,7 +1275,8 @@ static int exofs_update_inode(struct inode *inode, int do_sync)
1234free_args: 1275free_args:
1235 kfree(args); 1276 kfree(args);
1236out: 1277out:
1237 EXOFS_DBGMSG("ret=>%d\n", ret); 1278 EXOFS_DBGMSG("(0x%lx) do_sync=%d ret=>%d\n",
1279 inode->i_ino, do_sync, ret);
1238 return ret; 1280 return ret;
1239} 1281}
1240 1282
@@ -1283,7 +1325,7 @@ void exofs_delete_inode(struct inode *inode)
1283 1325
1284 clear_inode(inode); 1326 clear_inode(inode);
1285 1327
1286 ret = exofs_get_io_state(sbi, &ios); 1328 ret = exofs_get_io_state(&sbi->layout, &ios);
1287 if (unlikely(ret)) { 1329 if (unlikely(ret)) {
1288 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__); 1330 EXOFS_ERR("%s: exofs_get_io_state failed\n", __func__);
1289 return; 1331 return;
diff --git a/fs/exofs/ios.c b/fs/exofs/ios.c
index 5bad01fa1f9f..5293bc411d17 100644
--- a/fs/exofs/ios.c
+++ b/fs/exofs/ios.c
@@ -23,9 +23,13 @@
23 */ 23 */
24 24
25#include <scsi/scsi_device.h> 25#include <scsi/scsi_device.h>
26#include <asm/div64.h>
26 27
27#include "exofs.h" 28#include "exofs.h"
28 29
30#define EXOFS_DBGMSG2(M...) do {} while (0)
31/* #define EXOFS_DBGMSG2 EXOFS_DBGMSG */
32
29void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj) 33void exofs_make_credential(u8 cred_a[OSD_CAP_LEN], const struct osd_obj_id *obj)
30{ 34{
31 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true); 35 osd_sec_init_nosec_doall_caps(cred_a, obj, false, true);
@@ -64,21 +68,24 @@ out:
64 return ret; 68 return ret;
65} 69}
66 70
67int exofs_get_io_state(struct exofs_sb_info *sbi, struct exofs_io_state** pios) 71int exofs_get_io_state(struct exofs_layout *layout,
72 struct exofs_io_state **pios)
68{ 73{
69 struct exofs_io_state *ios; 74 struct exofs_io_state *ios;
70 75
71 /*TODO: Maybe use kmem_cach per sbi of size 76 /*TODO: Maybe use kmem_cach per sbi of size
72 * exofs_io_state_size(sbi->s_numdevs) 77 * exofs_io_state_size(layout->s_numdevs)
73 */ 78 */
74 ios = kzalloc(exofs_io_state_size(sbi->s_numdevs), GFP_KERNEL); 79 ios = kzalloc(exofs_io_state_size(layout->s_numdevs), GFP_KERNEL);
75 if (unlikely(!ios)) { 80 if (unlikely(!ios)) {
81 EXOFS_DBGMSG("Faild kzalloc bytes=%d\n",
82 exofs_io_state_size(layout->s_numdevs));
76 *pios = NULL; 83 *pios = NULL;
77 return -ENOMEM; 84 return -ENOMEM;
78 } 85 }
79 86
80 ios->sbi = sbi; 87 ios->layout = layout;
81 ios->obj.partition = sbi->s_pid; 88 ios->obj.partition = layout->s_pid;
82 *pios = ios; 89 *pios = ios;
83 return 0; 90 return 0;
84} 91}
@@ -101,6 +108,29 @@ void exofs_put_io_state(struct exofs_io_state *ios)
101 } 108 }
102} 109}
103 110
111unsigned exofs_layout_od_id(struct exofs_layout *layout,
112 osd_id obj_no, unsigned layout_index)
113{
114/* switch (layout->lay_func) {
115 case LAYOUT_MOVING_WINDOW:
116 {*/
117 unsigned dev_mod = obj_no;
118
119 return (layout_index + dev_mod * layout->mirrors_p1) %
120 layout->s_numdevs;
121/* }
122 case LAYOUT_FUNC_IMPLICT:
123 return layout->devs[layout_index];
124 }*/
125}
126
127static inline struct osd_dev *exofs_ios_od(struct exofs_io_state *ios,
128 unsigned layout_index)
129{
130 return ios->layout->s_ods[
131 exofs_layout_od_id(ios->layout, ios->obj.id, layout_index)];
132}
133
104static void _sync_done(struct exofs_io_state *ios, void *p) 134static void _sync_done(struct exofs_io_state *ios, void *p)
105{ 135{
106 struct completion *waiting = p; 136 struct completion *waiting = p;
@@ -168,6 +198,21 @@ static int exofs_io_execute(struct exofs_io_state *ios)
168 return ret; 198 return ret;
169} 199}
170 200
201static void _clear_bio(struct bio *bio)
202{
203 struct bio_vec *bv;
204 unsigned i;
205
206 __bio_for_each_segment(bv, bio, i, 0) {
207 unsigned this_count = bv->bv_len;
208
209 if (likely(PAGE_SIZE == this_count))
210 clear_highpage(bv->bv_page);
211 else
212 zero_user(bv->bv_page, bv->bv_offset, this_count);
213 }
214}
215
171int exofs_check_io(struct exofs_io_state *ios, u64 *resid) 216int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
172{ 217{
173 enum osd_err_priority acumulated_osd_err = 0; 218 enum osd_err_priority acumulated_osd_err = 0;
@@ -176,16 +221,25 @@ int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
176 221
177 for (i = 0; i < ios->numdevs; i++) { 222 for (i = 0; i < ios->numdevs; i++) {
178 struct osd_sense_info osi; 223 struct osd_sense_info osi;
179 int ret = osd_req_decode_sense(ios->per_dev[i].or, &osi); 224 struct osd_request *or = ios->per_dev[i].or;
225 int ret;
226
227 if (unlikely(!or))
228 continue;
180 229
230 ret = osd_req_decode_sense(or, &osi);
181 if (likely(!ret)) 231 if (likely(!ret))
182 continue; 232 continue;
183 233
184 if (unlikely(ret == -EFAULT)) { 234 if (OSD_ERR_PRI_CLEAR_PAGES == osi.osd_err_pri) {
185 EXOFS_DBGMSG("%s: EFAULT Need page clear\n", __func__); 235 /* start read offset passed endof file */
186 /*FIXME: All the pages in this device range should: 236 _clear_bio(ios->per_dev[i].bio);
187 * clear_highpage(page); 237 EXOFS_DBGMSG("start read offset passed end of file "
188 */ 238 "offset=0x%llx, length=0x%llx\n",
239 _LLU(ios->per_dev[i].offset),
240 _LLU(ios->per_dev[i].length));
241
242 continue; /* we recovered */
189 } 243 }
190 244
191 if (osi.osd_err_pri >= acumulated_osd_err) { 245 if (osi.osd_err_pri >= acumulated_osd_err) {
@@ -205,14 +259,259 @@ int exofs_check_io(struct exofs_io_state *ios, u64 *resid)
205 return acumulated_lin_err; 259 return acumulated_lin_err;
206} 260}
207 261
262/*
263 * L - logical offset into the file
264 *
265 * U - The number of bytes in a stripe within a group
266 *
267 * U = stripe_unit * group_width
268 *
269 * T - The number of bytes striped within a group of component objects
270 * (before advancing to the next group)
271 *
272 * T = stripe_unit * group_width * group_depth
273 *
274 * S - The number of bytes striped across all component objects
275 * before the pattern repeats
276 *
277 * S = stripe_unit * group_width * group_depth * group_count
278 *
279 * M - The "major" (i.e., across all components) stripe number
280 *
281 * M = L / S
282 *
283 * G - Counts the groups from the beginning of the major stripe
284 *
285 * G = (L - (M * S)) / T [or (L % S) / T]
286 *
287 * H - The byte offset within the group
288 *
289 * H = (L - (M * S)) % T [or (L % S) % T]
290 *
291 * N - The "minor" (i.e., across the group) stripe number
292 *
293 * N = H / U
294 *
295 * C - The component index coresponding to L
296 *
297 * C = (H - (N * U)) / stripe_unit + G * group_width
298 * [or (L % U) / stripe_unit + G * group_width]
299 *
300 * O - The component offset coresponding to L
301 *
302 * O = L % stripe_unit + N * stripe_unit + M * group_depth * stripe_unit
303 */
304struct _striping_info {
305 u64 obj_offset;
306 u64 group_length;
307 u64 total_group_length;
308 u64 Major;
309 unsigned dev;
310 unsigned unit_off;
311};
312
313static void _calc_stripe_info(struct exofs_io_state *ios, u64 file_offset,
314 struct _striping_info *si)
315{
316 u32 stripe_unit = ios->layout->stripe_unit;
317 u32 group_width = ios->layout->group_width;
318 u64 group_depth = ios->layout->group_depth;
319
320 u32 U = stripe_unit * group_width;
321 u64 T = U * group_depth;
322 u64 S = T * ios->layout->group_count;
323 u64 M = div64_u64(file_offset, S);
324
325 /*
326 G = (L - (M * S)) / T
327 H = (L - (M * S)) % T
328 */
329 u64 LmodS = file_offset - M * S;
330 u32 G = div64_u64(LmodS, T);
331 u64 H = LmodS - G * T;
332
333 u32 N = div_u64(H, U);
334
335 /* "H - (N * U)" is just "H % U" so it's bound to u32 */
336 si->dev = (u32)(H - (N * U)) / stripe_unit + G * group_width;
337 si->dev *= ios->layout->mirrors_p1;
338
339 div_u64_rem(file_offset, stripe_unit, &si->unit_off);
340
341 si->obj_offset = si->unit_off + (N * stripe_unit) +
342 (M * group_depth * stripe_unit);
343
344 si->group_length = T - H;
345 si->total_group_length = T;
346 si->Major = M;
347}
348
349static int _add_stripe_unit(struct exofs_io_state *ios, unsigned *cur_pg,
350 unsigned pgbase, struct exofs_per_dev_state *per_dev,
351 int cur_len)
352{
353 unsigned pg = *cur_pg;
354 struct request_queue *q =
355 osd_request_queue(exofs_ios_od(ios, per_dev->dev));
356
357 per_dev->length += cur_len;
358
359 if (per_dev->bio == NULL) {
360 unsigned pages_in_stripe = ios->layout->group_width *
361 (ios->layout->stripe_unit / PAGE_SIZE);
362 unsigned bio_size = (ios->nr_pages + pages_in_stripe) /
363 ios->layout->group_width;
364
365 per_dev->bio = bio_kmalloc(GFP_KERNEL, bio_size);
366 if (unlikely(!per_dev->bio)) {
367 EXOFS_DBGMSG("Faild to allocate BIO size=%u\n",
368 bio_size);
369 return -ENOMEM;
370 }
371 }
372
373 while (cur_len > 0) {
374 unsigned pglen = min_t(unsigned, PAGE_SIZE - pgbase, cur_len);
375 unsigned added_len;
376
377 BUG_ON(ios->nr_pages <= pg);
378 cur_len -= pglen;
379
380 added_len = bio_add_pc_page(q, per_dev->bio, ios->pages[pg],
381 pglen, pgbase);
382 if (unlikely(pglen != added_len))
383 return -ENOMEM;
384 pgbase = 0;
385 ++pg;
386 }
387 BUG_ON(cur_len);
388
389 *cur_pg = pg;
390 return 0;
391}
392
393static int _prepare_one_group(struct exofs_io_state *ios, u64 length,
394 struct _striping_info *si, unsigned first_comp)
395{
396 unsigned stripe_unit = ios->layout->stripe_unit;
397 unsigned mirrors_p1 = ios->layout->mirrors_p1;
398 unsigned devs_in_group = ios->layout->group_width * mirrors_p1;
399 unsigned dev = si->dev;
400 unsigned first_dev = dev - (dev % devs_in_group);
401 unsigned comp = first_comp + (dev - first_dev);
402 unsigned max_comp = ios->numdevs ? ios->numdevs - mirrors_p1 : 0;
403 unsigned cur_pg = ios->pages_consumed;
404 int ret = 0;
405
406 while (length) {
407 struct exofs_per_dev_state *per_dev = &ios->per_dev[comp];
408 unsigned cur_len, page_off = 0;
409
410 if (!per_dev->length) {
411 per_dev->dev = dev;
412 if (dev < si->dev) {
413 per_dev->offset = si->obj_offset + stripe_unit -
414 si->unit_off;
415 cur_len = stripe_unit;
416 } else if (dev == si->dev) {
417 per_dev->offset = si->obj_offset;
418 cur_len = stripe_unit - si->unit_off;
419 page_off = si->unit_off & ~PAGE_MASK;
420 BUG_ON(page_off && (page_off != ios->pgbase));
421 } else { /* dev > si->dev */
422 per_dev->offset = si->obj_offset - si->unit_off;
423 cur_len = stripe_unit;
424 }
425
426 if (max_comp < comp)
427 max_comp = comp;
428
429 dev += mirrors_p1;
430 dev = (dev % devs_in_group) + first_dev;
431 } else {
432 cur_len = stripe_unit;
433 }
434 if (cur_len >= length)
435 cur_len = length;
436
437 ret = _add_stripe_unit(ios, &cur_pg, page_off , per_dev,
438 cur_len);
439 if (unlikely(ret))
440 goto out;
441
442 comp += mirrors_p1;
443 comp = (comp % devs_in_group) + first_comp;
444
445 length -= cur_len;
446 }
447out:
448 ios->numdevs = max_comp + mirrors_p1;
449 ios->pages_consumed = cur_pg;
450 return ret;
451}
452
453static int _prepare_for_striping(struct exofs_io_state *ios)
454{
455 u64 length = ios->length;
456 struct _striping_info si;
457 unsigned devs_in_group = ios->layout->group_width *
458 ios->layout->mirrors_p1;
459 unsigned first_comp = 0;
460 int ret = 0;
461
462 _calc_stripe_info(ios, ios->offset, &si);
463
464 if (!ios->pages) {
465 if (ios->kern_buff) {
466 struct exofs_per_dev_state *per_dev = &ios->per_dev[0];
467
468 per_dev->offset = si.obj_offset;
469 per_dev->dev = si.dev;
470
471 /* no cross device without page array */
472 BUG_ON((ios->layout->group_width > 1) &&
473 (si.unit_off + ios->length >
474 ios->layout->stripe_unit));
475 }
476 ios->numdevs = ios->layout->mirrors_p1;
477 return 0;
478 }
479
480 while (length) {
481 if (length < si.group_length)
482 si.group_length = length;
483
484 ret = _prepare_one_group(ios, si.group_length, &si, first_comp);
485 if (unlikely(ret))
486 goto out;
487
488 length -= si.group_length;
489
490 si.group_length = si.total_group_length;
491 si.unit_off = 0;
492 ++si.Major;
493 si.obj_offset = si.Major * ios->layout->stripe_unit *
494 ios->layout->group_depth;
495
496 si.dev = (si.dev - (si.dev % devs_in_group)) + devs_in_group;
497 si.dev %= ios->layout->s_numdevs;
498
499 first_comp += devs_in_group;
500 first_comp %= ios->layout->s_numdevs;
501 }
502
503out:
504 return ret;
505}
506
208int exofs_sbi_create(struct exofs_io_state *ios) 507int exofs_sbi_create(struct exofs_io_state *ios)
209{ 508{
210 int i, ret; 509 int i, ret;
211 510
212 for (i = 0; i < ios->sbi->s_numdevs; i++) { 511 for (i = 0; i < ios->layout->s_numdevs; i++) {
213 struct osd_request *or; 512 struct osd_request *or;
214 513
215 or = osd_start_request(ios->sbi->s_ods[i], GFP_KERNEL); 514 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
216 if (unlikely(!or)) { 515 if (unlikely(!or)) {
217 EXOFS_ERR("%s: osd_start_request failed\n", __func__); 516 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
218 ret = -ENOMEM; 517 ret = -ENOMEM;
@@ -233,10 +532,10 @@ int exofs_sbi_remove(struct exofs_io_state *ios)
233{ 532{
234 int i, ret; 533 int i, ret;
235 534
236 for (i = 0; i < ios->sbi->s_numdevs; i++) { 535 for (i = 0; i < ios->layout->s_numdevs; i++) {
237 struct osd_request *or; 536 struct osd_request *or;
238 537
239 or = osd_start_request(ios->sbi->s_ods[i], GFP_KERNEL); 538 or = osd_start_request(exofs_ios_od(ios, i), GFP_KERNEL);
240 if (unlikely(!or)) { 539 if (unlikely(!or)) {
241 EXOFS_ERR("%s: osd_start_request failed\n", __func__); 540 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
242 ret = -ENOMEM; 541 ret = -ENOMEM;
@@ -253,51 +552,74 @@ out:
253 return ret; 552 return ret;
254} 553}
255 554
256int exofs_sbi_write(struct exofs_io_state *ios) 555static int _sbi_write_mirror(struct exofs_io_state *ios, int cur_comp)
257{ 556{
258 int i, ret; 557 struct exofs_per_dev_state *master_dev = &ios->per_dev[cur_comp];
558 unsigned dev = ios->per_dev[cur_comp].dev;
559 unsigned last_comp = cur_comp + ios->layout->mirrors_p1;
560 int ret = 0;
259 561
260 for (i = 0; i < ios->sbi->s_numdevs; i++) { 562 if (ios->pages && !master_dev->length)
563 return 0; /* Just an empty slot */
564
565 for (; cur_comp < last_comp; ++cur_comp, ++dev) {
566 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
261 struct osd_request *or; 567 struct osd_request *or;
262 568
263 or = osd_start_request(ios->sbi->s_ods[i], GFP_KERNEL); 569 or = osd_start_request(exofs_ios_od(ios, dev), GFP_KERNEL);
264 if (unlikely(!or)) { 570 if (unlikely(!or)) {
265 EXOFS_ERR("%s: osd_start_request failed\n", __func__); 571 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
266 ret = -ENOMEM; 572 ret = -ENOMEM;
267 goto out; 573 goto out;
268 } 574 }
269 ios->per_dev[i].or = or; 575 per_dev->or = or;
270 ios->numdevs++; 576 per_dev->offset = master_dev->offset;
271 577
272 if (ios->bio) { 578 if (ios->pages) {
273 struct bio *bio; 579 struct bio *bio;
274 580
275 if (i != 0) { 581 if (per_dev != master_dev) {
276 bio = bio_kmalloc(GFP_KERNEL, 582 bio = bio_kmalloc(GFP_KERNEL,
277 ios->bio->bi_max_vecs); 583 master_dev->bio->bi_max_vecs);
278 if (unlikely(!bio)) { 584 if (unlikely(!bio)) {
585 EXOFS_DBGMSG(
586 "Faild to allocate BIO size=%u\n",
587 master_dev->bio->bi_max_vecs);
279 ret = -ENOMEM; 588 ret = -ENOMEM;
280 goto out; 589 goto out;
281 } 590 }
282 591
283 __bio_clone(bio, ios->bio); 592 __bio_clone(bio, master_dev->bio);
284 bio->bi_bdev = NULL; 593 bio->bi_bdev = NULL;
285 bio->bi_next = NULL; 594 bio->bi_next = NULL;
286 ios->per_dev[i].bio = bio; 595 per_dev->length = master_dev->length;
596 per_dev->bio = bio;
597 per_dev->dev = dev;
287 } else { 598 } else {
288 bio = ios->bio; 599 bio = master_dev->bio;
600 /* FIXME: bio_set_dir() */
601 bio->bi_rw |= (1 << BIO_RW);
289 } 602 }
290 603
291 osd_req_write(or, &ios->obj, ios->offset, bio, 604 osd_req_write(or, &ios->obj, per_dev->offset, bio,
292 ios->length); 605 per_dev->length);
293/* EXOFS_DBGMSG("write sync=%d\n", sync);*/ 606 EXOFS_DBGMSG("write(0x%llx) offset=0x%llx "
607 "length=0x%llx dev=%d\n",
608 _LLU(ios->obj.id), _LLU(per_dev->offset),
609 _LLU(per_dev->length), dev);
294 } else if (ios->kern_buff) { 610 } else if (ios->kern_buff) {
295 osd_req_write_kern(or, &ios->obj, ios->offset, 611 ret = osd_req_write_kern(or, &ios->obj, per_dev->offset,
296 ios->kern_buff, ios->length); 612 ios->kern_buff, ios->length);
297/* EXOFS_DBGMSG("write_kern sync=%d\n", sync);*/ 613 if (unlikely(ret))
614 goto out;
615 EXOFS_DBGMSG2("write_kern(0x%llx) offset=0x%llx "
616 "length=0x%llx dev=%d\n",
617 _LLU(ios->obj.id), _LLU(per_dev->offset),
618 _LLU(ios->length), dev);
298 } else { 619 } else {
299 osd_req_set_attributes(or, &ios->obj); 620 osd_req_set_attributes(or, &ios->obj);
300/* EXOFS_DBGMSG("set_attributes sync=%d\n", sync);*/ 621 EXOFS_DBGMSG2("obj(0x%llx) set_attributes=%d dev=%d\n",
622 _LLU(ios->obj.id), ios->out_attr_len, dev);
301 } 623 }
302 624
303 if (ios->out_attr) 625 if (ios->out_attr)
@@ -308,54 +630,93 @@ int exofs_sbi_write(struct exofs_io_state *ios)
308 osd_req_add_get_attr_list(or, ios->in_attr, 630 osd_req_add_get_attr_list(or, ios->in_attr,
309 ios->in_attr_len); 631 ios->in_attr_len);
310 } 632 }
311 ret = exofs_io_execute(ios);
312 633
313out: 634out:
314 return ret; 635 return ret;
315} 636}
316 637
317int exofs_sbi_read(struct exofs_io_state *ios) 638int exofs_sbi_write(struct exofs_io_state *ios)
318{ 639{
319 int i, ret; 640 int i;
641 int ret;
320 642
321 for (i = 0; i < 1; i++) { 643 ret = _prepare_for_striping(ios);
322 struct osd_request *or; 644 if (unlikely(ret))
323 unsigned first_dev = (unsigned)ios->obj.id; 645 return ret;
324 646
325 first_dev %= ios->sbi->s_numdevs; 647 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
326 or = osd_start_request(ios->sbi->s_ods[first_dev], GFP_KERNEL); 648 ret = _sbi_write_mirror(ios, i);
327 if (unlikely(!or)) { 649 if (unlikely(ret))
328 EXOFS_ERR("%s: osd_start_request failed\n", __func__); 650 return ret;
329 ret = -ENOMEM; 651 }
330 goto out;
331 }
332 ios->per_dev[i].or = or;
333 ios->numdevs++;
334 652
335 if (ios->bio) { 653 ret = exofs_io_execute(ios);
336 osd_req_read(or, &ios->obj, ios->offset, ios->bio, 654 return ret;
337 ios->length); 655}
338/* EXOFS_DBGMSG("read sync=%d\n", sync);*/
339 } else if (ios->kern_buff) {
340 osd_req_read_kern(or, &ios->obj, ios->offset,
341 ios->kern_buff, ios->length);
342/* EXOFS_DBGMSG("read_kern sync=%d\n", sync);*/
343 } else {
344 osd_req_get_attributes(or, &ios->obj);
345/* EXOFS_DBGMSG("get_attributes sync=%d\n", sync);*/
346 }
347 656
348 if (ios->out_attr) 657static int _sbi_read_mirror(struct exofs_io_state *ios, unsigned cur_comp)
349 osd_req_add_set_attr_list(or, ios->out_attr, 658{
350 ios->out_attr_len); 659 struct osd_request *or;
660 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
661 unsigned first_dev = (unsigned)ios->obj.id;
351 662
352 if (ios->in_attr) 663 if (ios->pages && !per_dev->length)
353 osd_req_add_get_attr_list(or, ios->in_attr, 664 return 0; /* Just an empty slot */
354 ios->in_attr_len); 665
666 first_dev = per_dev->dev + first_dev % ios->layout->mirrors_p1;
667 or = osd_start_request(exofs_ios_od(ios, first_dev), GFP_KERNEL);
668 if (unlikely(!or)) {
669 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
670 return -ENOMEM;
355 } 671 }
356 ret = exofs_io_execute(ios); 672 per_dev->or = or;
673
674 if (ios->pages) {
675 osd_req_read(or, &ios->obj, per_dev->offset,
676 per_dev->bio, per_dev->length);
677 EXOFS_DBGMSG("read(0x%llx) offset=0x%llx length=0x%llx"
678 " dev=%d\n", _LLU(ios->obj.id),
679 _LLU(per_dev->offset), _LLU(per_dev->length),
680 first_dev);
681 } else if (ios->kern_buff) {
682 int ret = osd_req_read_kern(or, &ios->obj, per_dev->offset,
683 ios->kern_buff, ios->length);
684 EXOFS_DBGMSG2("read_kern(0x%llx) offset=0x%llx "
685 "length=0x%llx dev=%d ret=>%d\n",
686 _LLU(ios->obj.id), _LLU(per_dev->offset),
687 _LLU(ios->length), first_dev, ret);
688 if (unlikely(ret))
689 return ret;
690 } else {
691 osd_req_get_attributes(or, &ios->obj);
692 EXOFS_DBGMSG2("obj(0x%llx) get_attributes=%d dev=%d\n",
693 _LLU(ios->obj.id), ios->in_attr_len, first_dev);
694 }
695 if (ios->out_attr)
696 osd_req_add_set_attr_list(or, ios->out_attr, ios->out_attr_len);
357 697
358out: 698 if (ios->in_attr)
699 osd_req_add_get_attr_list(or, ios->in_attr, ios->in_attr_len);
700
701 return 0;
702}
703
704int exofs_sbi_read(struct exofs_io_state *ios)
705{
706 int i;
707 int ret;
708
709 ret = _prepare_for_striping(ios);
710 if (unlikely(ret))
711 return ret;
712
713 for (i = 0; i < ios->numdevs; i += ios->layout->mirrors_p1) {
714 ret = _sbi_read_mirror(ios, i);
715 if (unlikely(ret))
716 return ret;
717 }
718
719 ret = exofs_io_execute(ios);
359 return ret; 720 return ret;
360} 721}
361 722
@@ -380,42 +741,82 @@ int extract_attr_from_ios(struct exofs_io_state *ios, struct osd_attr *attr)
380 return -EIO; 741 return -EIO;
381} 742}
382 743
744static int _truncate_mirrors(struct exofs_io_state *ios, unsigned cur_comp,
745 struct osd_attr *attr)
746{
747 int last_comp = cur_comp + ios->layout->mirrors_p1;
748
749 for (; cur_comp < last_comp; ++cur_comp) {
750 struct exofs_per_dev_state *per_dev = &ios->per_dev[cur_comp];
751 struct osd_request *or;
752
753 or = osd_start_request(exofs_ios_od(ios, cur_comp), GFP_KERNEL);
754 if (unlikely(!or)) {
755 EXOFS_ERR("%s: osd_start_request failed\n", __func__);
756 return -ENOMEM;
757 }
758 per_dev->or = or;
759
760 osd_req_set_attributes(or, &ios->obj);
761 osd_req_add_set_attr_list(or, attr, 1);
762 }
763
764 return 0;
765}
766
383int exofs_oi_truncate(struct exofs_i_info *oi, u64 size) 767int exofs_oi_truncate(struct exofs_i_info *oi, u64 size)
384{ 768{
385 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info; 769 struct exofs_sb_info *sbi = oi->vfs_inode.i_sb->s_fs_info;
386 struct exofs_io_state *ios; 770 struct exofs_io_state *ios;
387 struct osd_attr attr; 771 struct exofs_trunc_attr {
388 __be64 newsize; 772 struct osd_attr attr;
773 __be64 newsize;
774 } *size_attrs;
775 struct _striping_info si;
389 int i, ret; 776 int i, ret;
390 777
391 if (exofs_get_io_state(sbi, &ios)) 778 ret = exofs_get_io_state(&sbi->layout, &ios);
392 return -ENOMEM; 779 if (unlikely(ret))
780 return ret;
781
782 size_attrs = kcalloc(ios->layout->group_width, sizeof(*size_attrs),
783 GFP_KERNEL);
784 if (unlikely(!size_attrs)) {
785 ret = -ENOMEM;
786 goto out;
787 }
393 788
394 ios->obj.id = exofs_oi_objno(oi); 789 ios->obj.id = exofs_oi_objno(oi);
395 ios->cred = oi->i_cred; 790 ios->cred = oi->i_cred;
396 791
397 newsize = cpu_to_be64(size); 792 ios->numdevs = ios->layout->s_numdevs;
398 attr = g_attr_logical_length; 793 _calc_stripe_info(ios, size, &si);
399 attr.val_ptr = &newsize;
400 794
401 for (i = 0; i < sbi->s_numdevs; i++) { 795 for (i = 0; i < ios->layout->group_width; ++i) {
402 struct osd_request *or; 796 struct exofs_trunc_attr *size_attr = &size_attrs[i];
797 u64 obj_size;
403 798
404 or = osd_start_request(sbi->s_ods[i], GFP_KERNEL); 799 if (i < si.dev)
405 if (unlikely(!or)) { 800 obj_size = si.obj_offset +
406 EXOFS_ERR("%s: osd_start_request failed\n", __func__); 801 ios->layout->stripe_unit - si.unit_off;
407 ret = -ENOMEM; 802 else if (i == si.dev)
408 goto out; 803 obj_size = si.obj_offset;
409 } 804 else /* i > si.dev */
410 ios->per_dev[i].or = or; 805 obj_size = si.obj_offset - si.unit_off;
411 ios->numdevs++;
412 806
413 osd_req_set_attributes(or, &ios->obj); 807 size_attr->newsize = cpu_to_be64(obj_size);
414 osd_req_add_set_attr_list(or, &attr, 1); 808 size_attr->attr = g_attr_logical_length;
809 size_attr->attr.val_ptr = &size_attr->newsize;
810
811 ret = _truncate_mirrors(ios, i * ios->layout->mirrors_p1,
812 &size_attr->attr);
813 if (unlikely(ret))
814 goto out;
415 } 815 }
416 ret = exofs_io_execute(ios); 816 ret = exofs_io_execute(ios);
417 817
418out: 818out:
819 kfree(size_attrs);
419 exofs_put_io_state(ios); 820 exofs_put_io_state(ios);
420 return ret; 821 return ret;
421} 822}
diff --git a/fs/exofs/super.c b/fs/exofs/super.c
index a1d1e77b12eb..6cf5e4e84d61 100644
--- a/fs/exofs/super.c
+++ b/fs/exofs/super.c
@@ -210,7 +210,7 @@ int exofs_sync_fs(struct super_block *sb, int wait)
210 sbi = sb->s_fs_info; 210 sbi = sb->s_fs_info;
211 fscb = &sbi->s_fscb; 211 fscb = &sbi->s_fscb;
212 212
213 ret = exofs_get_io_state(sbi, &ios); 213 ret = exofs_get_io_state(&sbi->layout, &ios);
214 if (ret) 214 if (ret)
215 goto out; 215 goto out;
216 216
@@ -264,12 +264,12 @@ static void _exofs_print_device(const char *msg, const char *dev_path,
264 264
265void exofs_free_sbi(struct exofs_sb_info *sbi) 265void exofs_free_sbi(struct exofs_sb_info *sbi)
266{ 266{
267 while (sbi->s_numdevs) { 267 while (sbi->layout.s_numdevs) {
268 int i = --sbi->s_numdevs; 268 int i = --sbi->layout.s_numdevs;
269 struct osd_dev *od = sbi->s_ods[i]; 269 struct osd_dev *od = sbi->layout.s_ods[i];
270 270
271 if (od) { 271 if (od) {
272 sbi->s_ods[i] = NULL; 272 sbi->layout.s_ods[i] = NULL;
273 osduld_put_device(od); 273 osduld_put_device(od);
274 } 274 }
275 } 275 }
@@ -298,7 +298,8 @@ static void exofs_put_super(struct super_block *sb)
298 msecs_to_jiffies(100)); 298 msecs_to_jiffies(100));
299 } 299 }
300 300
301 _exofs_print_device("Unmounting", NULL, sbi->s_ods[0], sbi->s_pid); 301 _exofs_print_device("Unmounting", NULL, sbi->layout.s_ods[0],
302 sbi->layout.s_pid);
302 303
303 exofs_free_sbi(sbi); 304 exofs_free_sbi(sbi);
304 sb->s_fs_info = NULL; 305 sb->s_fs_info = NULL;
@@ -307,6 +308,8 @@ static void exofs_put_super(struct super_block *sb)
307static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs, 308static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
308 struct exofs_device_table *dt) 309 struct exofs_device_table *dt)
309{ 310{
311 u64 stripe_length;
312
310 sbi->data_map.odm_num_comps = 313 sbi->data_map.odm_num_comps =
311 le32_to_cpu(dt->dt_data_map.cb_num_comps); 314 le32_to_cpu(dt->dt_data_map.cb_num_comps);
312 sbi->data_map.odm_stripe_unit = 315 sbi->data_map.odm_stripe_unit =
@@ -320,14 +323,63 @@ static int _read_and_match_data_map(struct exofs_sb_info *sbi, unsigned numdevs,
320 sbi->data_map.odm_raid_algorithm = 323 sbi->data_map.odm_raid_algorithm =
321 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm); 324 le32_to_cpu(dt->dt_data_map.cb_raid_algorithm);
322 325
323/* FIXME: Hard coded mirror only for now. if not so do not mount */ 326/* FIXME: Only raid0 for now. if not so, do not mount */
324 if ((sbi->data_map.odm_num_comps != numdevs) || 327 if (sbi->data_map.odm_num_comps != numdevs) {
325 (sbi->data_map.odm_stripe_unit != EXOFS_BLKSIZE) || 328 EXOFS_ERR("odm_num_comps(%u) != numdevs(%u)\n",
326 (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) || 329 sbi->data_map.odm_num_comps, numdevs);
327 (sbi->data_map.odm_mirror_cnt != (numdevs - 1)))
328 return -EINVAL; 330 return -EINVAL;
329 else 331 }
330 return 0; 332 if (sbi->data_map.odm_raid_algorithm != PNFS_OSD_RAID_0) {
333 EXOFS_ERR("Only RAID_0 for now\n");
334 return -EINVAL;
335 }
336 if (0 != (numdevs % (sbi->data_map.odm_mirror_cnt + 1))) {
337 EXOFS_ERR("Data Map wrong, numdevs=%d mirrors=%d\n",
338 numdevs, sbi->data_map.odm_mirror_cnt);
339 return -EINVAL;
340 }
341
342 if (0 != (sbi->data_map.odm_stripe_unit & ~PAGE_MASK)) {
343 EXOFS_ERR("Stripe Unit(0x%llx)"
344 " must be Multples of PAGE_SIZE(0x%lx)\n",
345 _LLU(sbi->data_map.odm_stripe_unit), PAGE_SIZE);
346 return -EINVAL;
347 }
348
349 sbi->layout.stripe_unit = sbi->data_map.odm_stripe_unit;
350 sbi->layout.mirrors_p1 = sbi->data_map.odm_mirror_cnt + 1;
351
352 if (sbi->data_map.odm_group_width) {
353 sbi->layout.group_width = sbi->data_map.odm_group_width;
354 sbi->layout.group_depth = sbi->data_map.odm_group_depth;
355 if (!sbi->layout.group_depth) {
356 EXOFS_ERR("group_depth == 0 && group_width != 0\n");
357 return -EINVAL;
358 }
359 sbi->layout.group_count = sbi->data_map.odm_num_comps /
360 sbi->layout.mirrors_p1 /
361 sbi->data_map.odm_group_width;
362 } else {
363 if (sbi->data_map.odm_group_depth) {
364 printk(KERN_NOTICE "Warning: group_depth ignored "
365 "group_width == 0 && group_depth == %d\n",
366 sbi->data_map.odm_group_depth);
367 sbi->data_map.odm_group_depth = 0;
368 }
369 sbi->layout.group_width = sbi->data_map.odm_num_comps /
370 sbi->layout.mirrors_p1;
371 sbi->layout.group_depth = -1;
372 sbi->layout.group_count = 1;
373 }
374
375 stripe_length = (u64)sbi->layout.group_width * sbi->layout.stripe_unit;
376 if (stripe_length >= (1ULL << 32)) {
377 EXOFS_ERR("Total Stripe length(0x%llx)"
378 " >= 32bit is not supported\n", _LLU(stripe_length));
379 return -EINVAL;
380 }
381
382 return 0;
331} 383}
332 384
333/* @odi is valid only as long as @fscb_dev is valid */ 385/* @odi is valid only as long as @fscb_dev is valid */
@@ -361,7 +413,7 @@ static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
361{ 413{
362 struct exofs_sb_info *sbi = *psbi; 414 struct exofs_sb_info *sbi = *psbi;
363 struct osd_dev *fscb_od; 415 struct osd_dev *fscb_od;
364 struct osd_obj_id obj = {.partition = sbi->s_pid, 416 struct osd_obj_id obj = {.partition = sbi->layout.s_pid,
365 .id = EXOFS_DEVTABLE_ID}; 417 .id = EXOFS_DEVTABLE_ID};
366 struct exofs_device_table *dt; 418 struct exofs_device_table *dt;
367 unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) + 419 unsigned table_bytes = table_count * sizeof(dt->dt_dev_table[0]) +
@@ -376,9 +428,9 @@ static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
376 return -ENOMEM; 428 return -ENOMEM;
377 } 429 }
378 430
379 fscb_od = sbi->s_ods[0]; 431 fscb_od = sbi->layout.s_ods[0];
380 sbi->s_ods[0] = NULL; 432 sbi->layout.s_ods[0] = NULL;
381 sbi->s_numdevs = 0; 433 sbi->layout.s_numdevs = 0;
382 ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes); 434 ret = exofs_read_kern(fscb_od, sbi->s_cred, &obj, 0, dt, table_bytes);
383 if (unlikely(ret)) { 435 if (unlikely(ret)) {
384 EXOFS_ERR("ERROR: reading device table\n"); 436 EXOFS_ERR("ERROR: reading device table\n");
@@ -397,14 +449,15 @@ static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
397 goto out; 449 goto out;
398 450
399 if (likely(numdevs > 1)) { 451 if (likely(numdevs > 1)) {
400 unsigned size = numdevs * sizeof(sbi->s_ods[0]); 452 unsigned size = numdevs * sizeof(sbi->layout.s_ods[0]);
401 453
402 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL); 454 sbi = krealloc(sbi, sizeof(*sbi) + size, GFP_KERNEL);
403 if (unlikely(!sbi)) { 455 if (unlikely(!sbi)) {
404 ret = -ENOMEM; 456 ret = -ENOMEM;
405 goto out; 457 goto out;
406 } 458 }
407 memset(&sbi->s_ods[1], 0, size - sizeof(sbi->s_ods[0])); 459 memset(&sbi->layout.s_ods[1], 0,
460 size - sizeof(sbi->layout.s_ods[0]));
408 *psbi = sbi; 461 *psbi = sbi;
409 } 462 }
410 463
@@ -427,8 +480,8 @@ static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
427 * line. We always keep them in device-table order. 480 * line. We always keep them in device-table order.
428 */ 481 */
429 if (fscb_od && osduld_device_same(fscb_od, &odi)) { 482 if (fscb_od && osduld_device_same(fscb_od, &odi)) {
430 sbi->s_ods[i] = fscb_od; 483 sbi->layout.s_ods[i] = fscb_od;
431 ++sbi->s_numdevs; 484 ++sbi->layout.s_numdevs;
432 fscb_od = NULL; 485 fscb_od = NULL;
433 continue; 486 continue;
434 } 487 }
@@ -441,8 +494,8 @@ static int exofs_read_lookup_dev_table(struct exofs_sb_info **psbi,
441 goto out; 494 goto out;
442 } 495 }
443 496
444 sbi->s_ods[i] = od; 497 sbi->layout.s_ods[i] = od;
445 ++sbi->s_numdevs; 498 ++sbi->layout.s_numdevs;
446 499
447 /* Read the fscb of the other devices to make sure the FS 500 /* Read the fscb of the other devices to make sure the FS
448 * partition is there. 501 * partition is there.
@@ -499,9 +552,15 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
499 goto free_sbi; 552 goto free_sbi;
500 } 553 }
501 554
502 sbi->s_ods[0] = od; 555 /* Default layout in case we do not have a device-table */
503 sbi->s_numdevs = 1; 556 sbi->layout.stripe_unit = PAGE_SIZE;
504 sbi->s_pid = opts->pid; 557 sbi->layout.mirrors_p1 = 1;
558 sbi->layout.group_width = 1;
559 sbi->layout.group_depth = -1;
560 sbi->layout.group_count = 1;
561 sbi->layout.s_ods[0] = od;
562 sbi->layout.s_numdevs = 1;
563 sbi->layout.s_pid = opts->pid;
505 sbi->s_timeout = opts->timeout; 564 sbi->s_timeout = opts->timeout;
506 565
507 /* fill in some other data by hand */ 566 /* fill in some other data by hand */
@@ -514,7 +573,7 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
514 sb->s_bdev = NULL; 573 sb->s_bdev = NULL;
515 sb->s_dev = 0; 574 sb->s_dev = 0;
516 575
517 obj.partition = sbi->s_pid; 576 obj.partition = sbi->layout.s_pid;
518 obj.id = EXOFS_SUPER_ID; 577 obj.id = EXOFS_SUPER_ID;
519 exofs_make_credential(sbi->s_cred, &obj); 578 exofs_make_credential(sbi->s_cred, &obj);
520 579
@@ -578,13 +637,13 @@ static int exofs_fill_super(struct super_block *sb, void *data, int silent)
578 goto free_sbi; 637 goto free_sbi;
579 } 638 }
580 639
581 _exofs_print_device("Mounting", opts->dev_name, sbi->s_ods[0], 640 _exofs_print_device("Mounting", opts->dev_name, sbi->layout.s_ods[0],
582 sbi->s_pid); 641 sbi->layout.s_pid);
583 return 0; 642 return 0;
584 643
585free_sbi: 644free_sbi:
586 EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n", 645 EXOFS_ERR("Unable to mount exofs on %s pid=0x%llx err=%d\n",
587 opts->dev_name, sbi->s_pid, ret); 646 opts->dev_name, sbi->layout.s_pid, ret);
588 exofs_free_sbi(sbi); 647 exofs_free_sbi(sbi);
589 return ret; 648 return ret;
590} 649}
@@ -627,7 +686,7 @@ static int exofs_statfs(struct dentry *dentry, struct kstatfs *buf)
627 uint8_t cred_a[OSD_CAP_LEN]; 686 uint8_t cred_a[OSD_CAP_LEN];
628 int ret; 687 int ret;
629 688
630 ret = exofs_get_io_state(sbi, &ios); 689 ret = exofs_get_io_state(&sbi->layout, &ios);
631 if (ret) { 690 if (ret) {
632 EXOFS_DBGMSG("exofs_get_io_state failed.\n"); 691 EXOFS_DBGMSG("exofs_get_io_state failed.\n");
633 return ret; 692 return ret;
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-02 03:53:52 -0500