Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 1 | /* |
| 2 | * recovery.c - NILFS recovery logic |
| 3 | * |
| 4 | * Copyright (C) 2005-2008 Nippon Telegraph and Telephone Corporation. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA |
| 19 | * |
| 20 | * Written by Ryusuke Konishi <ryusuke@osrg.net> |
| 21 | */ |
| 22 | |
| 23 | #include <linux/buffer_head.h> |
| 24 | #include <linux/blkdev.h> |
| 25 | #include <linux/swap.h> |
| 26 | #include <linux/crc32.h> |
| 27 | #include "nilfs.h" |
| 28 | #include "segment.h" |
| 29 | #include "sufile.h" |
| 30 | #include "page.h" |
| 31 | #include "seglist.h" |
| 32 | #include "segbuf.h" |
| 33 | |
| 34 | /* |
| 35 | * Segment check result |
| 36 | */ |
| 37 | enum { |
| 38 | NILFS_SEG_VALID, |
| 39 | NILFS_SEG_NO_SUPER_ROOT, |
| 40 | NILFS_SEG_FAIL_IO, |
| 41 | NILFS_SEG_FAIL_MAGIC, |
| 42 | NILFS_SEG_FAIL_SEQ, |
| 43 | NILFS_SEG_FAIL_CHECKSUM_SEGSUM, |
| 44 | NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT, |
| 45 | NILFS_SEG_FAIL_CHECKSUM_FULL, |
| 46 | NILFS_SEG_FAIL_CONSISTENCY, |
| 47 | }; |
| 48 | |
| 49 | /* work structure for recovery */ |
| 50 | struct nilfs_recovery_block { |
| 51 | ino_t ino; /* Inode number of the file that this block |
| 52 | belongs to */ |
| 53 | sector_t blocknr; /* block number */ |
| 54 | __u64 vblocknr; /* virtual block number */ |
| 55 | unsigned long blkoff; /* File offset of the data block (per block) */ |
| 56 | struct list_head list; |
| 57 | }; |
| 58 | |
| 59 | |
| 60 | static int nilfs_warn_segment_error(int err) |
| 61 | { |
| 62 | switch (err) { |
| 63 | case NILFS_SEG_FAIL_IO: |
| 64 | printk(KERN_WARNING |
| 65 | "NILFS warning: I/O error on loading last segment\n"); |
| 66 | return -EIO; |
| 67 | case NILFS_SEG_FAIL_MAGIC: |
| 68 | printk(KERN_WARNING |
| 69 | "NILFS warning: Segment magic number invalid\n"); |
| 70 | break; |
| 71 | case NILFS_SEG_FAIL_SEQ: |
| 72 | printk(KERN_WARNING |
| 73 | "NILFS warning: Sequence number mismatch\n"); |
| 74 | break; |
| 75 | case NILFS_SEG_FAIL_CHECKSUM_SEGSUM: |
| 76 | printk(KERN_WARNING |
| 77 | "NILFS warning: Checksum error in segment summary\n"); |
| 78 | break; |
| 79 | case NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT: |
| 80 | printk(KERN_WARNING |
| 81 | "NILFS warning: Checksum error in super root\n"); |
| 82 | break; |
| 83 | case NILFS_SEG_FAIL_CHECKSUM_FULL: |
| 84 | printk(KERN_WARNING |
| 85 | "NILFS warning: Checksum error in segment payload\n"); |
| 86 | break; |
| 87 | case NILFS_SEG_FAIL_CONSISTENCY: |
| 88 | printk(KERN_WARNING |
| 89 | "NILFS warning: Inconsistent segment\n"); |
| 90 | break; |
| 91 | case NILFS_SEG_NO_SUPER_ROOT: |
| 92 | printk(KERN_WARNING |
| 93 | "NILFS warning: No super root in the last segment\n"); |
| 94 | break; |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 95 | } |
| 96 | return -EINVAL; |
| 97 | } |
| 98 | |
| 99 | static void store_segsum_info(struct nilfs_segsum_info *ssi, |
| 100 | struct nilfs_segment_summary *sum, |
| 101 | unsigned int blocksize) |
| 102 | { |
| 103 | ssi->flags = le16_to_cpu(sum->ss_flags); |
| 104 | ssi->seg_seq = le64_to_cpu(sum->ss_seq); |
| 105 | ssi->ctime = le64_to_cpu(sum->ss_create); |
| 106 | ssi->next = le64_to_cpu(sum->ss_next); |
| 107 | ssi->nblocks = le32_to_cpu(sum->ss_nblocks); |
| 108 | ssi->nfinfo = le32_to_cpu(sum->ss_nfinfo); |
| 109 | ssi->sumbytes = le32_to_cpu(sum->ss_sumbytes); |
| 110 | |
| 111 | ssi->nsumblk = DIV_ROUND_UP(ssi->sumbytes, blocksize); |
| 112 | ssi->nfileblk = ssi->nblocks - ssi->nsumblk - !!NILFS_SEG_HAS_SR(ssi); |
| 113 | } |
| 114 | |
| 115 | /** |
| 116 | * calc_crc_cont - check CRC of blocks continuously |
| 117 | * @sbi: nilfs_sb_info |
| 118 | * @bhs: buffer head of start block |
| 119 | * @sum: place to store result |
| 120 | * @offset: offset bytes in the first block |
| 121 | * @check_bytes: number of bytes to be checked |
| 122 | * @start: DBN of start block |
| 123 | * @nblock: number of blocks to be checked |
| 124 | */ |
| 125 | static int calc_crc_cont(struct nilfs_sb_info *sbi, struct buffer_head *bhs, |
| 126 | u32 *sum, unsigned long offset, u64 check_bytes, |
| 127 | sector_t start, unsigned long nblock) |
| 128 | { |
| 129 | unsigned long blocksize = sbi->s_super->s_blocksize; |
| 130 | unsigned long size; |
| 131 | u32 crc; |
| 132 | |
| 133 | BUG_ON(offset >= blocksize); |
| 134 | check_bytes -= offset; |
| 135 | size = min_t(u64, check_bytes, blocksize - offset); |
| 136 | crc = crc32_le(sbi->s_nilfs->ns_crc_seed, |
| 137 | (unsigned char *)bhs->b_data + offset, size); |
| 138 | if (--nblock > 0) { |
| 139 | do { |
| 140 | struct buffer_head *bh |
| 141 | = sb_bread(sbi->s_super, ++start); |
| 142 | if (!bh) |
| 143 | return -EIO; |
| 144 | check_bytes -= size; |
| 145 | size = min_t(u64, check_bytes, blocksize); |
| 146 | crc = crc32_le(crc, bh->b_data, size); |
| 147 | brelse(bh); |
| 148 | } while (--nblock > 0); |
| 149 | } |
| 150 | *sum = crc; |
| 151 | return 0; |
| 152 | } |
| 153 | |
| 154 | /** |
| 155 | * nilfs_read_super_root_block - read super root block |
| 156 | * @sb: super_block |
| 157 | * @sr_block: disk block number of the super root block |
| 158 | * @pbh: address of a buffer_head pointer to return super root buffer |
| 159 | * @check: CRC check flag |
| 160 | */ |
| 161 | int nilfs_read_super_root_block(struct super_block *sb, sector_t sr_block, |
| 162 | struct buffer_head **pbh, int check) |
| 163 | { |
| 164 | struct buffer_head *bh_sr; |
| 165 | struct nilfs_super_root *sr; |
| 166 | u32 crc; |
| 167 | int ret; |
| 168 | |
| 169 | *pbh = NULL; |
| 170 | bh_sr = sb_bread(sb, sr_block); |
| 171 | if (unlikely(!bh_sr)) { |
| 172 | ret = NILFS_SEG_FAIL_IO; |
| 173 | goto failed; |
| 174 | } |
| 175 | |
| 176 | sr = (struct nilfs_super_root *)bh_sr->b_data; |
| 177 | if (check) { |
| 178 | unsigned bytes = le16_to_cpu(sr->sr_bytes); |
| 179 | |
| 180 | if (bytes == 0 || bytes > sb->s_blocksize) { |
| 181 | ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT; |
| 182 | goto failed_bh; |
| 183 | } |
| 184 | if (calc_crc_cont(NILFS_SB(sb), bh_sr, &crc, |
| 185 | sizeof(sr->sr_sum), bytes, sr_block, 1)) { |
| 186 | ret = NILFS_SEG_FAIL_IO; |
| 187 | goto failed_bh; |
| 188 | } |
| 189 | if (crc != le32_to_cpu(sr->sr_sum)) { |
| 190 | ret = NILFS_SEG_FAIL_CHECKSUM_SUPER_ROOT; |
| 191 | goto failed_bh; |
| 192 | } |
| 193 | } |
| 194 | *pbh = bh_sr; |
| 195 | return 0; |
| 196 | |
| 197 | failed_bh: |
| 198 | brelse(bh_sr); |
| 199 | |
| 200 | failed: |
| 201 | return nilfs_warn_segment_error(ret); |
| 202 | } |
| 203 | |
| 204 | /** |
| 205 | * load_segment_summary - read segment summary of the specified partial segment |
| 206 | * @sbi: nilfs_sb_info |
| 207 | * @pseg_start: start disk block number of partial segment |
| 208 | * @seg_seq: sequence number requested |
| 209 | * @ssi: pointer to nilfs_segsum_info struct to store information |
| 210 | * @full_check: full check flag |
| 211 | * (0: only checks segment summary CRC, 1: data CRC) |
| 212 | */ |
| 213 | static int |
| 214 | load_segment_summary(struct nilfs_sb_info *sbi, sector_t pseg_start, |
| 215 | u64 seg_seq, struct nilfs_segsum_info *ssi, |
| 216 | int full_check) |
| 217 | { |
| 218 | struct buffer_head *bh_sum; |
| 219 | struct nilfs_segment_summary *sum; |
| 220 | unsigned long offset, nblock; |
| 221 | u64 check_bytes; |
| 222 | u32 crc, crc_sum; |
| 223 | int ret = NILFS_SEG_FAIL_IO; |
| 224 | |
| 225 | bh_sum = sb_bread(sbi->s_super, pseg_start); |
| 226 | if (!bh_sum) |
| 227 | goto out; |
| 228 | |
| 229 | sum = (struct nilfs_segment_summary *)bh_sum->b_data; |
| 230 | |
| 231 | /* Check consistency of segment summary */ |
| 232 | if (le32_to_cpu(sum->ss_magic) != NILFS_SEGSUM_MAGIC) { |
| 233 | ret = NILFS_SEG_FAIL_MAGIC; |
| 234 | goto failed; |
| 235 | } |
| 236 | store_segsum_info(ssi, sum, sbi->s_super->s_blocksize); |
| 237 | if (seg_seq != ssi->seg_seq) { |
| 238 | ret = NILFS_SEG_FAIL_SEQ; |
| 239 | goto failed; |
| 240 | } |
| 241 | if (full_check) { |
| 242 | offset = sizeof(sum->ss_datasum); |
| 243 | check_bytes = |
| 244 | ((u64)ssi->nblocks << sbi->s_super->s_blocksize_bits); |
| 245 | nblock = ssi->nblocks; |
| 246 | crc_sum = le32_to_cpu(sum->ss_datasum); |
| 247 | ret = NILFS_SEG_FAIL_CHECKSUM_FULL; |
| 248 | } else { /* only checks segment summary */ |
| 249 | offset = sizeof(sum->ss_datasum) + sizeof(sum->ss_sumsum); |
| 250 | check_bytes = ssi->sumbytes; |
| 251 | nblock = ssi->nsumblk; |
| 252 | crc_sum = le32_to_cpu(sum->ss_sumsum); |
| 253 | ret = NILFS_SEG_FAIL_CHECKSUM_SEGSUM; |
| 254 | } |
| 255 | |
| 256 | if (unlikely(nblock == 0 || |
| 257 | nblock > sbi->s_nilfs->ns_blocks_per_segment)) { |
| 258 | /* This limits the number of blocks read in the CRC check */ |
| 259 | ret = NILFS_SEG_FAIL_CONSISTENCY; |
| 260 | goto failed; |
| 261 | } |
| 262 | if (calc_crc_cont(sbi, bh_sum, &crc, offset, check_bytes, |
| 263 | pseg_start, nblock)) { |
| 264 | ret = NILFS_SEG_FAIL_IO; |
| 265 | goto failed; |
| 266 | } |
| 267 | if (crc == crc_sum) |
| 268 | ret = 0; |
| 269 | failed: |
| 270 | brelse(bh_sum); |
| 271 | out: |
| 272 | return ret; |
| 273 | } |
| 274 | |
| 275 | static void *segsum_get(struct super_block *sb, struct buffer_head **pbh, |
| 276 | unsigned int *offset, unsigned int bytes) |
| 277 | { |
| 278 | void *ptr; |
| 279 | sector_t blocknr; |
| 280 | |
| 281 | BUG_ON((*pbh)->b_size < *offset); |
| 282 | if (bytes > (*pbh)->b_size - *offset) { |
| 283 | blocknr = (*pbh)->b_blocknr; |
| 284 | brelse(*pbh); |
| 285 | *pbh = sb_bread(sb, blocknr + 1); |
| 286 | if (unlikely(!*pbh)) |
| 287 | return NULL; |
| 288 | *offset = 0; |
| 289 | } |
| 290 | ptr = (*pbh)->b_data + *offset; |
| 291 | *offset += bytes; |
| 292 | return ptr; |
| 293 | } |
| 294 | |
| 295 | static void segsum_skip(struct super_block *sb, struct buffer_head **pbh, |
| 296 | unsigned int *offset, unsigned int bytes, |
| 297 | unsigned long count) |
| 298 | { |
| 299 | unsigned int rest_item_in_current_block |
| 300 | = ((*pbh)->b_size - *offset) / bytes; |
| 301 | |
| 302 | if (count <= rest_item_in_current_block) { |
| 303 | *offset += bytes * count; |
| 304 | } else { |
| 305 | sector_t blocknr = (*pbh)->b_blocknr; |
| 306 | unsigned int nitem_per_block = (*pbh)->b_size / bytes; |
| 307 | unsigned int bcnt; |
| 308 | |
| 309 | count -= rest_item_in_current_block; |
| 310 | bcnt = DIV_ROUND_UP(count, nitem_per_block); |
| 311 | *offset = bytes * (count - (bcnt - 1) * nitem_per_block); |
| 312 | |
| 313 | brelse(*pbh); |
| 314 | *pbh = sb_bread(sb, blocknr + bcnt); |
| 315 | } |
| 316 | } |
| 317 | |
| 318 | static int |
| 319 | collect_blocks_from_segsum(struct nilfs_sb_info *sbi, sector_t sum_blocknr, |
| 320 | struct nilfs_segsum_info *ssi, |
| 321 | struct list_head *head) |
| 322 | { |
| 323 | struct buffer_head *bh; |
| 324 | unsigned int offset; |
| 325 | unsigned long nfinfo = ssi->nfinfo; |
| 326 | sector_t blocknr = sum_blocknr + ssi->nsumblk; |
| 327 | ino_t ino; |
| 328 | int err = -EIO; |
| 329 | |
| 330 | if (!nfinfo) |
| 331 | return 0; |
| 332 | |
| 333 | bh = sb_bread(sbi->s_super, sum_blocknr); |
| 334 | if (unlikely(!bh)) |
| 335 | goto out; |
| 336 | |
| 337 | offset = le16_to_cpu( |
| 338 | ((struct nilfs_segment_summary *)bh->b_data)->ss_bytes); |
| 339 | for (;;) { |
| 340 | unsigned long nblocks, ndatablk, nnodeblk; |
| 341 | struct nilfs_finfo *finfo; |
| 342 | |
| 343 | finfo = segsum_get(sbi->s_super, &bh, &offset, sizeof(*finfo)); |
| 344 | if (unlikely(!finfo)) |
| 345 | goto out; |
| 346 | |
| 347 | ino = le64_to_cpu(finfo->fi_ino); |
| 348 | nblocks = le32_to_cpu(finfo->fi_nblocks); |
| 349 | ndatablk = le32_to_cpu(finfo->fi_ndatablk); |
| 350 | nnodeblk = nblocks - ndatablk; |
| 351 | |
| 352 | while (ndatablk-- > 0) { |
| 353 | struct nilfs_recovery_block *rb; |
| 354 | struct nilfs_binfo_v *binfo; |
| 355 | |
| 356 | binfo = segsum_get(sbi->s_super, &bh, &offset, |
| 357 | sizeof(*binfo)); |
| 358 | if (unlikely(!binfo)) |
| 359 | goto out; |
| 360 | |
| 361 | rb = kmalloc(sizeof(*rb), GFP_NOFS); |
| 362 | if (unlikely(!rb)) { |
| 363 | err = -ENOMEM; |
| 364 | goto out; |
| 365 | } |
| 366 | rb->ino = ino; |
| 367 | rb->blocknr = blocknr++; |
| 368 | rb->vblocknr = le64_to_cpu(binfo->bi_vblocknr); |
| 369 | rb->blkoff = le64_to_cpu(binfo->bi_blkoff); |
| 370 | /* INIT_LIST_HEAD(&rb->list); */ |
| 371 | list_add_tail(&rb->list, head); |
| 372 | } |
| 373 | if (--nfinfo == 0) |
| 374 | break; |
| 375 | blocknr += nnodeblk; /* always 0 for the data sync segments */ |
| 376 | segsum_skip(sbi->s_super, &bh, &offset, sizeof(__le64), |
| 377 | nnodeblk); |
| 378 | if (unlikely(!bh)) |
| 379 | goto out; |
| 380 | } |
| 381 | err = 0; |
| 382 | out: |
| 383 | brelse(bh); /* brelse(NULL) is just ignored */ |
| 384 | return err; |
| 385 | } |
| 386 | |
| 387 | static void dispose_recovery_list(struct list_head *head) |
| 388 | { |
| 389 | while (!list_empty(head)) { |
| 390 | struct nilfs_recovery_block *rb |
| 391 | = list_entry(head->next, |
| 392 | struct nilfs_recovery_block, list); |
| 393 | list_del(&rb->list); |
| 394 | kfree(rb); |
| 395 | } |
| 396 | } |
| 397 | |
| 398 | void nilfs_dispose_segment_list(struct list_head *head) |
| 399 | { |
| 400 | while (!list_empty(head)) { |
| 401 | struct nilfs_segment_entry *ent |
| 402 | = list_entry(head->next, |
| 403 | struct nilfs_segment_entry, list); |
| 404 | list_del(&ent->list); |
| 405 | nilfs_free_segment_entry(ent); |
| 406 | } |
| 407 | } |
| 408 | |
| 409 | static int nilfs_prepare_segment_for_recovery(struct the_nilfs *nilfs, |
| 410 | struct nilfs_recovery_info *ri) |
| 411 | { |
| 412 | struct list_head *head = &ri->ri_used_segments; |
| 413 | struct nilfs_segment_entry *ent, *n; |
| 414 | struct inode *sufile = nilfs->ns_sufile; |
| 415 | __u64 segnum[4]; |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 416 | time_t mtime; |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 417 | int err; |
| 418 | int i; |
| 419 | |
| 420 | segnum[0] = nilfs->ns_segnum; |
| 421 | segnum[1] = nilfs->ns_nextnum; |
| 422 | segnum[2] = ri->ri_segnum; |
| 423 | segnum[3] = ri->ri_nextnum; |
| 424 | |
| 425 | /* |
| 426 | * Releasing the next segment of the latest super root. |
| 427 | * The next segment is invalidated by this recovery. |
| 428 | */ |
| 429 | err = nilfs_sufile_free(sufile, segnum[1]); |
| 430 | if (unlikely(err)) |
| 431 | goto failed; |
| 432 | |
| 433 | err = -ENOMEM; |
| 434 | for (i = 1; i < 4; i++) { |
| 435 | ent = nilfs_alloc_segment_entry(segnum[i]); |
| 436 | if (unlikely(!ent)) |
| 437 | goto failed; |
| 438 | list_add_tail(&ent->list, head); |
| 439 | } |
| 440 | |
| 441 | /* |
| 442 | * Collecting segments written after the latest super root. |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 443 | * These are marked dirty to avoid being reallocated in the next write. |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 444 | */ |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 445 | mtime = get_seconds(); |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 446 | list_for_each_entry_safe(ent, n, head, list) { |
| 447 | if (ent->segnum == segnum[0]) { |
| 448 | list_del(&ent->list); |
| 449 | nilfs_free_segment_entry(ent); |
| 450 | continue; |
| 451 | } |
| 452 | err = nilfs_open_segment_entry(ent, sufile); |
| 453 | if (unlikely(err)) |
| 454 | goto failed; |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 455 | if (!nilfs_segment_usage_dirty(ent->raw_su)) { |
| 456 | /* make the segment garbage */ |
| 457 | ent->raw_su->su_nblocks = cpu_to_le32(0); |
| 458 | ent->raw_su->su_lastmod = cpu_to_le32(mtime); |
| 459 | nilfs_segment_usage_set_dirty(ent->raw_su); |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 460 | } |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 461 | list_del(&ent->list); |
| 462 | nilfs_close_segment_entry(ent, sufile); |
| 463 | nilfs_free_segment_entry(ent); |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 464 | } |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 465 | |
| 466 | /* |
| 467 | * The segment having the latest super root is active, and |
| 468 | * should be deactivated on the next construction for recovery. |
| 469 | */ |
| 470 | err = -ENOMEM; |
| 471 | ent = nilfs_alloc_segment_entry(segnum[0]); |
| 472 | if (unlikely(!ent)) |
| 473 | goto failed; |
| 474 | list_add_tail(&ent->list, &ri->ri_used_segments); |
| 475 | |
| 476 | /* Allocate new segments for recovery */ |
| 477 | err = nilfs_sufile_alloc(sufile, &segnum[0]); |
| 478 | if (unlikely(err)) |
| 479 | goto failed; |
| 480 | |
| 481 | nilfs->ns_pseg_offset = 0; |
| 482 | nilfs->ns_seg_seq = ri->ri_seq + 2; |
| 483 | nilfs->ns_nextnum = nilfs->ns_segnum = segnum[0]; |
| 484 | return 0; |
| 485 | |
| 486 | failed: |
| 487 | /* No need to recover sufile because it will be destroyed on error */ |
| 488 | return err; |
| 489 | } |
| 490 | |
| 491 | static int nilfs_recovery_copy_block(struct nilfs_sb_info *sbi, |
| 492 | struct nilfs_recovery_block *rb, |
| 493 | struct page *page) |
| 494 | { |
| 495 | struct buffer_head *bh_org; |
| 496 | void *kaddr; |
| 497 | |
| 498 | bh_org = sb_bread(sbi->s_super, rb->blocknr); |
| 499 | if (unlikely(!bh_org)) |
| 500 | return -EIO; |
| 501 | |
| 502 | kaddr = kmap_atomic(page, KM_USER0); |
| 503 | memcpy(kaddr + bh_offset(bh_org), bh_org->b_data, bh_org->b_size); |
| 504 | kunmap_atomic(kaddr, KM_USER0); |
| 505 | brelse(bh_org); |
| 506 | return 0; |
| 507 | } |
| 508 | |
| 509 | static int recover_dsync_blocks(struct nilfs_sb_info *sbi, |
| 510 | struct list_head *head, |
| 511 | unsigned long *nr_salvaged_blocks) |
| 512 | { |
| 513 | struct inode *inode; |
| 514 | struct nilfs_recovery_block *rb, *n; |
| 515 | unsigned blocksize = sbi->s_super->s_blocksize; |
| 516 | struct page *page; |
| 517 | loff_t pos; |
| 518 | int err = 0, err2 = 0; |
| 519 | |
| 520 | list_for_each_entry_safe(rb, n, head, list) { |
| 521 | inode = nilfs_iget(sbi->s_super, rb->ino); |
| 522 | if (IS_ERR(inode)) { |
| 523 | err = PTR_ERR(inode); |
| 524 | inode = NULL; |
| 525 | goto failed_inode; |
| 526 | } |
| 527 | |
| 528 | pos = rb->blkoff << inode->i_blkbits; |
| 529 | page = NULL; |
| 530 | err = block_write_begin(NULL, inode->i_mapping, pos, blocksize, |
| 531 | 0, &page, NULL, nilfs_get_block); |
| 532 | if (unlikely(err)) |
| 533 | goto failed_inode; |
| 534 | |
| 535 | err = nilfs_recovery_copy_block(sbi, rb, page); |
| 536 | if (unlikely(err)) |
| 537 | goto failed_page; |
| 538 | |
| 539 | err = nilfs_set_file_dirty(sbi, inode, 1); |
| 540 | if (unlikely(err)) |
| 541 | goto failed_page; |
| 542 | |
| 543 | block_write_end(NULL, inode->i_mapping, pos, blocksize, |
| 544 | blocksize, page, NULL); |
| 545 | |
| 546 | unlock_page(page); |
| 547 | page_cache_release(page); |
| 548 | |
| 549 | (*nr_salvaged_blocks)++; |
| 550 | goto next; |
| 551 | |
| 552 | failed_page: |
| 553 | unlock_page(page); |
| 554 | page_cache_release(page); |
| 555 | |
| 556 | failed_inode: |
| 557 | printk(KERN_WARNING |
| 558 | "NILFS warning: error recovering data block " |
| 559 | "(err=%d, ino=%lu, block-offset=%llu)\n", |
| 560 | err, rb->ino, (unsigned long long)rb->blkoff); |
| 561 | if (!err2) |
| 562 | err2 = err; |
| 563 | next: |
| 564 | iput(inode); /* iput(NULL) is just ignored */ |
| 565 | list_del_init(&rb->list); |
| 566 | kfree(rb); |
| 567 | } |
| 568 | return err2; |
| 569 | } |
| 570 | |
| 571 | /** |
| 572 | * nilfs_do_roll_forward - salvage logical segments newer than the latest |
| 573 | * checkpoint |
| 574 | * @sbi: nilfs_sb_info |
| 575 | * @nilfs: the_nilfs |
| 576 | * @ri: pointer to a nilfs_recovery_info |
| 577 | */ |
| 578 | static int nilfs_do_roll_forward(struct the_nilfs *nilfs, |
| 579 | struct nilfs_sb_info *sbi, |
| 580 | struct nilfs_recovery_info *ri) |
| 581 | { |
| 582 | struct nilfs_segsum_info ssi; |
| 583 | sector_t pseg_start; |
| 584 | sector_t seg_start, seg_end; /* Starting/ending DBN of full segment */ |
| 585 | unsigned long nsalvaged_blocks = 0; |
| 586 | u64 seg_seq; |
| 587 | __u64 segnum, nextnum = 0; |
| 588 | int empty_seg = 0; |
| 589 | int err = 0, ret; |
| 590 | LIST_HEAD(dsync_blocks); /* list of data blocks to be recovered */ |
| 591 | enum { |
| 592 | RF_INIT_ST, |
| 593 | RF_DSYNC_ST, /* scanning data-sync segments */ |
| 594 | }; |
| 595 | int state = RF_INIT_ST; |
| 596 | |
| 597 | nilfs_attach_writer(nilfs, sbi); |
| 598 | pseg_start = ri->ri_lsegs_start; |
| 599 | seg_seq = ri->ri_lsegs_start_seq; |
| 600 | segnum = nilfs_get_segnum_of_block(nilfs, pseg_start); |
| 601 | nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); |
| 602 | |
| 603 | while (segnum != ri->ri_segnum || pseg_start <= ri->ri_pseg_start) { |
| 604 | |
| 605 | ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1); |
| 606 | if (ret) { |
| 607 | if (ret == NILFS_SEG_FAIL_IO) { |
| 608 | err = -EIO; |
| 609 | goto failed; |
| 610 | } |
| 611 | goto strayed; |
| 612 | } |
| 613 | if (unlikely(NILFS_SEG_HAS_SR(&ssi))) |
| 614 | goto confused; |
| 615 | |
| 616 | /* Found a valid partial segment; do recovery actions */ |
| 617 | nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next); |
| 618 | empty_seg = 0; |
| 619 | nilfs->ns_ctime = ssi.ctime; |
| 620 | if (!(ssi.flags & NILFS_SS_GC)) |
| 621 | nilfs->ns_nongc_ctime = ssi.ctime; |
| 622 | |
| 623 | switch (state) { |
| 624 | case RF_INIT_ST: |
| 625 | if (!NILFS_SEG_LOGBGN(&ssi) || !NILFS_SEG_DSYNC(&ssi)) |
| 626 | goto try_next_pseg; |
| 627 | state = RF_DSYNC_ST; |
| 628 | /* Fall through */ |
| 629 | case RF_DSYNC_ST: |
| 630 | if (!NILFS_SEG_DSYNC(&ssi)) |
| 631 | goto confused; |
| 632 | |
| 633 | err = collect_blocks_from_segsum( |
| 634 | sbi, pseg_start, &ssi, &dsync_blocks); |
| 635 | if (unlikely(err)) |
| 636 | goto failed; |
| 637 | if (NILFS_SEG_LOGEND(&ssi)) { |
| 638 | err = recover_dsync_blocks( |
| 639 | sbi, &dsync_blocks, &nsalvaged_blocks); |
| 640 | if (unlikely(err)) |
| 641 | goto failed; |
| 642 | state = RF_INIT_ST; |
| 643 | } |
| 644 | break; /* Fall through to try_next_pseg */ |
| 645 | } |
| 646 | |
| 647 | try_next_pseg: |
| 648 | if (pseg_start == ri->ri_lsegs_end) |
| 649 | break; |
| 650 | pseg_start += ssi.nblocks; |
| 651 | if (pseg_start < seg_end) |
| 652 | continue; |
| 653 | goto feed_segment; |
| 654 | |
| 655 | strayed: |
| 656 | if (pseg_start == ri->ri_lsegs_end) |
| 657 | break; |
| 658 | |
| 659 | feed_segment: |
| 660 | /* Looking to the next full segment */ |
| 661 | if (empty_seg++) |
| 662 | break; |
| 663 | seg_seq++; |
| 664 | segnum = nextnum; |
| 665 | nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); |
| 666 | pseg_start = seg_start; |
| 667 | } |
| 668 | |
| 669 | if (nsalvaged_blocks) { |
| 670 | printk(KERN_INFO "NILFS (device %s): salvaged %lu blocks\n", |
| 671 | sbi->s_super->s_id, nsalvaged_blocks); |
| 672 | ri->ri_need_recovery = NILFS_RECOVERY_ROLLFORWARD_DONE; |
| 673 | } |
| 674 | out: |
| 675 | dispose_recovery_list(&dsync_blocks); |
| 676 | nilfs_detach_writer(sbi->s_nilfs, sbi); |
| 677 | return err; |
| 678 | |
| 679 | confused: |
| 680 | err = -EINVAL; |
| 681 | failed: |
| 682 | printk(KERN_ERR |
| 683 | "NILFS (device %s): Error roll-forwarding " |
| 684 | "(err=%d, pseg block=%llu). ", |
| 685 | sbi->s_super->s_id, err, (unsigned long long)pseg_start); |
| 686 | goto out; |
| 687 | } |
| 688 | |
| 689 | static void nilfs_finish_roll_forward(struct the_nilfs *nilfs, |
| 690 | struct nilfs_sb_info *sbi, |
| 691 | struct nilfs_recovery_info *ri) |
| 692 | { |
| 693 | struct buffer_head *bh; |
| 694 | int err; |
| 695 | |
| 696 | if (nilfs_get_segnum_of_block(nilfs, ri->ri_lsegs_start) != |
| 697 | nilfs_get_segnum_of_block(nilfs, ri->ri_super_root)) |
| 698 | return; |
| 699 | |
| 700 | bh = sb_getblk(sbi->s_super, ri->ri_lsegs_start); |
| 701 | BUG_ON(!bh); |
| 702 | memset(bh->b_data, 0, bh->b_size); |
| 703 | set_buffer_dirty(bh); |
| 704 | err = sync_dirty_buffer(bh); |
| 705 | if (unlikely(err)) |
| 706 | printk(KERN_WARNING |
| 707 | "NILFS warning: buffer sync write failed during " |
| 708 | "post-cleaning of recovery.\n"); |
| 709 | brelse(bh); |
| 710 | } |
| 711 | |
| 712 | /** |
| 713 | * nilfs_recover_logical_segments - salvage logical segments written after |
| 714 | * the latest super root |
| 715 | * @nilfs: the_nilfs |
| 716 | * @sbi: nilfs_sb_info |
| 717 | * @ri: pointer to a nilfs_recovery_info struct to store search results. |
| 718 | * |
| 719 | * Return Value: On success, 0 is returned. On error, one of the following |
| 720 | * negative error code is returned. |
| 721 | * |
| 722 | * %-EINVAL - Inconsistent filesystem state. |
| 723 | * |
| 724 | * %-EIO - I/O error |
| 725 | * |
| 726 | * %-ENOSPC - No space left on device (only in a panic state). |
| 727 | * |
| 728 | * %-ERESTARTSYS - Interrupted. |
| 729 | * |
| 730 | * %-ENOMEM - Insufficient memory available. |
| 731 | */ |
| 732 | int nilfs_recover_logical_segments(struct the_nilfs *nilfs, |
| 733 | struct nilfs_sb_info *sbi, |
| 734 | struct nilfs_recovery_info *ri) |
| 735 | { |
| 736 | int err; |
| 737 | |
| 738 | if (ri->ri_lsegs_start == 0 || ri->ri_lsegs_end == 0) |
| 739 | return 0; |
| 740 | |
| 741 | err = nilfs_attach_checkpoint(sbi, ri->ri_cno); |
| 742 | if (unlikely(err)) { |
| 743 | printk(KERN_ERR |
| 744 | "NILFS: error loading the latest checkpoint.\n"); |
| 745 | return err; |
| 746 | } |
| 747 | |
| 748 | err = nilfs_do_roll_forward(nilfs, sbi, ri); |
| 749 | if (unlikely(err)) |
| 750 | goto failed; |
| 751 | |
| 752 | if (ri->ri_need_recovery == NILFS_RECOVERY_ROLLFORWARD_DONE) { |
| 753 | err = nilfs_prepare_segment_for_recovery(nilfs, ri); |
| 754 | if (unlikely(err)) { |
| 755 | printk(KERN_ERR "NILFS: Error preparing segments for " |
| 756 | "recovery.\n"); |
| 757 | goto failed; |
| 758 | } |
| 759 | |
| 760 | err = nilfs_attach_segment_constructor(sbi, ri); |
| 761 | if (unlikely(err)) |
| 762 | goto failed; |
| 763 | |
| 764 | set_nilfs_discontinued(nilfs); |
| 765 | err = nilfs_construct_segment(sbi->s_super); |
| 766 | nilfs_detach_segment_constructor(sbi); |
| 767 | |
| 768 | if (unlikely(err)) { |
| 769 | printk(KERN_ERR "NILFS: Oops! recovery failed. " |
| 770 | "(err=%d)\n", err); |
| 771 | goto failed; |
| 772 | } |
| 773 | |
| 774 | nilfs_finish_roll_forward(nilfs, sbi, ri); |
| 775 | } |
| 776 | |
| 777 | nilfs_detach_checkpoint(sbi); |
| 778 | return 0; |
| 779 | |
| 780 | failed: |
| 781 | nilfs_detach_checkpoint(sbi); |
| 782 | nilfs_mdt_clear(nilfs->ns_cpfile); |
| 783 | nilfs_mdt_clear(nilfs->ns_sufile); |
| 784 | nilfs_mdt_clear(nilfs->ns_dat); |
| 785 | return err; |
| 786 | } |
| 787 | |
| 788 | /** |
| 789 | * nilfs_search_super_root - search the latest valid super root |
| 790 | * @nilfs: the_nilfs |
| 791 | * @sbi: nilfs_sb_info |
| 792 | * @ri: pointer to a nilfs_recovery_info struct to store search results. |
| 793 | * |
| 794 | * nilfs_search_super_root() looks for the latest super-root from a partial |
| 795 | * segment pointed by the superblock. It sets up struct the_nilfs through |
| 796 | * this search. It fills nilfs_recovery_info (ri) required for recovery. |
| 797 | * |
| 798 | * Return Value: On success, 0 is returned. On error, one of the following |
| 799 | * negative error code is returned. |
| 800 | * |
| 801 | * %-EINVAL - No valid segment found |
| 802 | * |
| 803 | * %-EIO - I/O error |
| 804 | */ |
| 805 | int nilfs_search_super_root(struct the_nilfs *nilfs, struct nilfs_sb_info *sbi, |
| 806 | struct nilfs_recovery_info *ri) |
| 807 | { |
| 808 | struct nilfs_segsum_info ssi; |
| 809 | sector_t pseg_start, pseg_end, sr_pseg_start = 0; |
| 810 | sector_t seg_start, seg_end; /* range of full segment (block number) */ |
| 811 | u64 seg_seq; |
| 812 | __u64 segnum, nextnum = 0; |
| 813 | __u64 cno; |
| 814 | struct nilfs_segment_entry *ent; |
| 815 | LIST_HEAD(segments); |
| 816 | int empty_seg = 0, scan_newer = 0; |
| 817 | int ret; |
| 818 | |
| 819 | pseg_start = nilfs->ns_last_pseg; |
| 820 | seg_seq = nilfs->ns_last_seq; |
| 821 | cno = nilfs->ns_last_cno; |
| 822 | segnum = nilfs_get_segnum_of_block(nilfs, pseg_start); |
| 823 | |
| 824 | /* Calculate range of segment */ |
| 825 | nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); |
| 826 | |
| 827 | for (;;) { |
| 828 | /* Load segment summary */ |
| 829 | ret = load_segment_summary(sbi, pseg_start, seg_seq, &ssi, 1); |
| 830 | if (ret) { |
| 831 | if (ret == NILFS_SEG_FAIL_IO) |
| 832 | goto failed; |
| 833 | goto strayed; |
| 834 | } |
| 835 | pseg_end = pseg_start + ssi.nblocks - 1; |
| 836 | if (unlikely(pseg_end > seg_end)) { |
| 837 | ret = NILFS_SEG_FAIL_CONSISTENCY; |
| 838 | goto strayed; |
| 839 | } |
| 840 | |
| 841 | /* A valid partial segment */ |
| 842 | ri->ri_pseg_start = pseg_start; |
| 843 | ri->ri_seq = seg_seq; |
| 844 | ri->ri_segnum = segnum; |
| 845 | nextnum = nilfs_get_segnum_of_block(nilfs, ssi.next); |
| 846 | ri->ri_nextnum = nextnum; |
| 847 | empty_seg = 0; |
| 848 | |
| 849 | if (!NILFS_SEG_HAS_SR(&ssi)) { |
| 850 | if (!scan_newer) { |
| 851 | /* This will never happen because a superblock |
| 852 | (last_segment) always points to a pseg |
| 853 | having a super root. */ |
| 854 | ret = NILFS_SEG_FAIL_CONSISTENCY; |
| 855 | goto failed; |
| 856 | } |
| 857 | if (!ri->ri_lsegs_start && NILFS_SEG_LOGBGN(&ssi)) { |
| 858 | ri->ri_lsegs_start = pseg_start; |
| 859 | ri->ri_lsegs_start_seq = seg_seq; |
| 860 | } |
| 861 | if (NILFS_SEG_LOGEND(&ssi)) |
| 862 | ri->ri_lsegs_end = pseg_start; |
| 863 | goto try_next_pseg; |
| 864 | } |
| 865 | |
| 866 | /* A valid super root was found. */ |
| 867 | ri->ri_cno = cno++; |
| 868 | ri->ri_super_root = pseg_end; |
| 869 | ri->ri_lsegs_start = ri->ri_lsegs_end = 0; |
| 870 | |
| 871 | nilfs_dispose_segment_list(&segments); |
| 872 | nilfs->ns_pseg_offset = (sr_pseg_start = pseg_start) |
| 873 | + ssi.nblocks - seg_start; |
| 874 | nilfs->ns_seg_seq = seg_seq; |
| 875 | nilfs->ns_segnum = segnum; |
| 876 | nilfs->ns_cno = cno; /* nilfs->ns_cno = ri->ri_cno + 1 */ |
| 877 | nilfs->ns_ctime = ssi.ctime; |
| 878 | nilfs->ns_nextnum = nextnum; |
| 879 | |
| 880 | if (scan_newer) |
| 881 | ri->ri_need_recovery = NILFS_RECOVERY_SR_UPDATED; |
Ryusuke Konishi | 2c2e52f | 2009-04-06 19:01:54 -0700 | [diff] [blame] | 882 | else { |
| 883 | nilfs->ns_prot_seq = ssi.seg_seq; |
| 884 | if (nilfs->ns_mount_state & NILFS_VALID_FS) |
| 885 | goto super_root_found; |
| 886 | scan_newer = 1; |
| 887 | } |
Ryusuke Konishi | 0f3e1c7 | 2009-04-06 19:01:38 -0700 | [diff] [blame] | 888 | |
| 889 | /* reset region for roll-forward */ |
| 890 | pseg_start += ssi.nblocks; |
| 891 | if (pseg_start < seg_end) |
| 892 | continue; |
| 893 | goto feed_segment; |
| 894 | |
| 895 | try_next_pseg: |
| 896 | /* Standing on a course, or met an inconsistent state */ |
| 897 | pseg_start += ssi.nblocks; |
| 898 | if (pseg_start < seg_end) |
| 899 | continue; |
| 900 | goto feed_segment; |
| 901 | |
| 902 | strayed: |
| 903 | /* Off the trail */ |
| 904 | if (!scan_newer) |
| 905 | /* |
| 906 | * This can happen if a checkpoint was written without |
| 907 | * barriers, or as a result of an I/O failure. |
| 908 | */ |
| 909 | goto failed; |
| 910 | |
| 911 | feed_segment: |
| 912 | /* Looking to the next full segment */ |
| 913 | if (empty_seg++) |
| 914 | goto super_root_found; /* found a valid super root */ |
| 915 | |
| 916 | ent = nilfs_alloc_segment_entry(segnum); |
| 917 | if (unlikely(!ent)) { |
| 918 | ret = -ENOMEM; |
| 919 | goto failed; |
| 920 | } |
| 921 | list_add_tail(&ent->list, &segments); |
| 922 | |
| 923 | seg_seq++; |
| 924 | segnum = nextnum; |
| 925 | nilfs_get_segment_range(nilfs, segnum, &seg_start, &seg_end); |
| 926 | pseg_start = seg_start; |
| 927 | } |
| 928 | |
| 929 | super_root_found: |
| 930 | /* Updating pointers relating to the latest checkpoint */ |
| 931 | list_splice(&segments, ri->ri_used_segments.prev); |
| 932 | nilfs->ns_last_pseg = sr_pseg_start; |
| 933 | nilfs->ns_last_seq = nilfs->ns_seg_seq; |
| 934 | nilfs->ns_last_cno = ri->ri_cno; |
| 935 | return 0; |
| 936 | |
| 937 | failed: |
| 938 | nilfs_dispose_segment_list(&segments); |
| 939 | return (ret < 0) ? ret : nilfs_warn_segment_error(ret); |
| 940 | } |