fs/f2fs/recovery.c

   1 /*
   2  * fs/f2fs/recovery.c
   3  *
   4  * Copyright (c) 2012 Samsung Electronics Co., Ltd.
   5  *             http://www.samsung.com/
   6  *
   7  * This program is free software; you can redistribute it and/or modify
   8  * it under the terms of the GNU General Public License version 2 as
   9  * published by the Free Software Foundation.
  10  */
  11 #include <linux/fs.h>
  12 #include <linux/f2fs_fs.h>
  13 #include "f2fs.h"
  14 #include "node.h"
  15 #include "segment.h"
  16
  17 /*
  18  * Roll forward recovery scenarios.
  19  *
  20  * [Term] F: fsync_mark, D: dentry_mark
  21  *
  22  * 1. inode(x) | CP | inode(x) | dnode(F)
  23  * -> Update the latest inode(x).
  24  *
  25  * 2. inode(x) | CP | inode(F) | dnode(F)
  26  * -> No problem.
  27  *
  28  * 3. inode(x) | CP | dnode(F) | inode(x)
  29  * -> Recover to the latest dnode(F), and drop the last inode(x)
  30  *
  31  * 4. inode(x) | CP | dnode(F) | inode(F)
  32  * -> No problem.
  33  *
  34  * 5. CP | inode(x) | dnode(F)
  35  * -> The inode(DF) was missing. Should drop this dnode(F).
  36  *
  37  * 6. CP | inode(DF) | dnode(F)
  38  * -> No problem.
  39  *
  40  * 7. CP | dnode(F) | inode(DF)
  41  * -> If f2fs_iget fails, then goto next to find inode(DF).
  42  *
  43  * 8. CP | dnode(F) | inode(x)
  44  * -> If f2fs_iget fails, then goto next to find inode(DF).
  45  *    But it will fail due to no inode(DF).
  46  */
  47
  48 static struct kmem_cache *fsync_entry_slab;
  49
  50 bool space_for_roll_forward(struct f2fs_sb_info *sbi)
  51 {
  52         s64 nalloc = percpu_counter_sum_positive(&sbi->alloc_valid_block_count);
  53
  54         if (sbi->last_valid_block_count + nalloc > sbi->user_block_count)
  55                 return false;
  56         return true;
  57 }
  58
  59 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
  60                                                                 nid_t ino)
  61 {
  62         struct fsync_inode_entry *entry;
  63
  64         list_for_each_entry(entry, head, list)
  65                 if (entry->inode->i_ino == ino)
  66                         return entry;
  67
  68         return NULL;
  69 }
  70
  71 static struct fsync_inode_entry *add_fsync_inode(struct list_head *head,
  72                                                         struct inode *inode)
  73 {
  74         struct fsync_inode_entry *entry;
  75
  76         entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
  77         if (!entry)
  78                 return NULL;
  79
  80         entry->inode = inode;
  81         list_add_tail(&entry->list, head);
  82
  83         return entry;
  84 }
  85
  86 static void del_fsync_inode(struct fsync_inode_entry *entry)
  87 {
  88         iput(entry->inode);
  89         list_del(&entry->list);
  90         kmem_cache_free(fsync_entry_slab, entry);
  91 }
  92
  93 static int recover_dentry(struct inode *inode, struct page *ipage,
  94                                                 struct list_head *dir_list)
  95 {
  96         struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
  97         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  98         struct f2fs_dir_entry *de;
  99         struct qstr name;
 100         struct page *page;
 101         struct inode *dir, *einode;
 102         struct fsync_inode_entry *entry;
 103         int err = 0;
 104
 105         entry = get_fsync_inode(dir_list, pino);
 106         if (!entry) {
 107                 dir = f2fs_iget(inode->i_sb, pino);
 108                 if (IS_ERR(dir)) {
 109                         err = PTR_ERR(dir);
 110                         goto out;
 111                 }
 112
 113                 entry = add_fsync_inode(dir_list, dir);
 114                 if (!entry) {
 115                         err = -ENOMEM;
 116                         iput(dir);
 117                         goto out;
 118                 }
 119         }
 120
 121         dir = entry->inode;
 122
 123         if (file_enc_name(inode))
 124                 return 0;
 125
 126         name.len = le32_to_cpu(raw_inode->i_namelen);
 127         name.name = raw_inode->i_name;
 128
 129         if (unlikely(name.len > F2FS_NAME_LEN)) {
 130                 WARN_ON(1);
 131                 err = -ENAMETOOLONG;
 132                 goto out;
 133         }
 134 retry:
 135         de = f2fs_find_entry(dir, &name, &page);
 136         if (de && inode->i_ino == le32_to_cpu(de->ino))
 137                 goto out_unmap_put;
 138
 139         if (de) {
 140                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
 141                 if (IS_ERR(einode)) {
 142                         WARN_ON(1);
 143                         err = PTR_ERR(einode);
 144                         if (err == -ENOENT)
 145                                 err = -EEXIST;
 146                         goto out_unmap_put;
 147                 }
 148                 err = acquire_orphan_inode(F2FS_I_SB(inode));
 149                 if (err) {
 150                         iput(einode);
 151                         goto out_unmap_put;
 152                 }
 153                 f2fs_delete_entry(de, page, dir, einode);
 154                 iput(einode);
 155                 goto retry;
 156         }
 157         err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
 158
 159         goto out;
 160
 161 out_unmap_put:
 162         f2fs_dentry_kunmap(dir, page);
 163         f2fs_put_page(page, 0);
 164 out:
 165         f2fs_msg(inode->i_sb, KERN_NOTICE,
 166                         "%s: ino = %x, name = %s, dir = %lx, err = %d",
 167                         __func__, ino_of_node(ipage), raw_inode->i_name,
 168                         IS_ERR(dir) ? 0 : dir->i_ino, err);
 169         return err;
 170 }
 171
 172 static void recover_inode(struct inode *inode, struct page *page)
 173 {
 174         struct f2fs_inode *raw = F2FS_INODE(page);
 175         char *name;
 176
 177         inode->i_mode = le16_to_cpu(raw->i_mode);
 178         i_size_write(inode, le64_to_cpu(raw->i_size));
 179         inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
 180         inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
 181         inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
 182         inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 183         inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
 184         inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 185
 186         if (file_enc_name(inode))
 187                 name = "<encrypted>";
 188         else
 189                 name = F2FS_INODE(page)->i_name;
 190
 191         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 192                         ino_of_node(page), name);
 193 }
 194
 195 static bool is_same_inode(struct inode *inode, struct page *ipage)
 196 {
 197         struct f2fs_inode *ri = F2FS_INODE(ipage);
 198         struct timespec disk;
 199
 200         if (!IS_INODE(ipage))
 201                 return true;
 202
 203         disk.tv_sec = le64_to_cpu(ri->i_ctime);
 204         disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec);
 205         if (timespec_compare(&inode->i_ctime, &disk) > 0)
 206                 return false;
 207
 208         disk.tv_sec = le64_to_cpu(ri->i_atime);
 209         disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec);
 210         if (timespec_compare(&inode->i_atime, &disk) > 0)
 211                 return false;
 212
 213         disk.tv_sec = le64_to_cpu(ri->i_mtime);
 214         disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec);
 215         if (timespec_compare(&inode->i_mtime, &disk) > 0)
 216                 return false;
 217
 218         return true;
 219 }
 220
 221 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 222 {
 223         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 224         struct curseg_info *curseg;
 225         struct inode *inode;
 226         struct page *page = NULL;
 227         block_t blkaddr;
 228         int err = 0;
 229
 230         /* get node pages in the current segment */
 231         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 232         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 233
 234         while (1) {
 235                 struct fsync_inode_entry *entry;
 236
 237                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 238                         return 0;
 239
 240                 page = get_tmp_page(sbi, blkaddr);
 241
 242                 if (cp_ver != cpver_of_node(page))
 243                         break;
 244
 245                 if (!is_fsync_dnode(page))
 246                         goto next;
 247
 248                 entry = get_fsync_inode(head, ino_of_node(page));
 249                 if (entry) {
 250                         if (!is_same_inode(entry->inode, page))
 251                                 goto next;
 252                 } else {
 253                         if (IS_INODE(page) && is_dent_dnode(page)) {
 254                                 err = recover_inode_page(sbi, page);
 255                                 if (err)
 256                                         break;
 257                         }
 258
 259                         /*
 260                          * CP | dnode(F) | inode(DF)
 261                          * For this case, we should not give up now.
 262                          */
 263                         inode = f2fs_iget(sbi->sb, ino_of_node(page));
 264                         if (IS_ERR(inode)) {
 265                                 err = PTR_ERR(inode);
 266                                 if (err == -ENOENT) {
 267                                         err = 0;
 268                                         goto next;
 269                                 }
 270                                 break;
 271                         }
 272
 273                         /* add this fsync inode to the list */
 274                         entry = add_fsync_inode(head, inode);
 275                         if (!entry) {
 276                                 err = -ENOMEM;
 277                                 iput(inode);
 278                                 break;
 279                         }
 280                 }
 281                 entry->blkaddr = blkaddr;
 282
 283                 if (IS_INODE(page) && is_dent_dnode(page))
 284                         entry->last_dentry = blkaddr;
 285 next:
 286                 /* check next segment */
 287                 blkaddr = next_blkaddr_of_node(page);
 288                 f2fs_put_page(page, 1);
 289
 290                 ra_meta_pages_cond(sbi, blkaddr);
 291         }
 292         f2fs_put_page(page, 1);
 293         return err;
 294 }
 295
 296 static void destroy_fsync_dnodes(struct list_head *head)
 297 {
 298         struct fsync_inode_entry *entry, *tmp;
 299
 300         list_for_each_entry_safe(entry, tmp, head, list)
 301                 del_fsync_inode(entry);
 302 }
 303
 304 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 305                         block_t blkaddr, struct dnode_of_data *dn)
 306 {
 307         struct seg_entry *sentry;
 308         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 309         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 310         struct f2fs_summary_block *sum_node;
 311         struct f2fs_summary sum;
 312         struct page *sum_page, *node_page;
 313         struct dnode_of_data tdn = *dn;
 314         nid_t ino, nid;
 315         struct inode *inode;
 316         unsigned int offset;
 317         block_t bidx;
 318         int i;
 319
 320         sentry = get_seg_entry(sbi, segno);
 321         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 322                 return 0;
 323
 324         /* Get the previous summary */
 325         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 326                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 327                 if (curseg->segno == segno) {
 328                         sum = curseg->sum_blk->entries[blkoff];
 329                         goto got_it;
 330                 }
 331         }
 332
 333         sum_page = get_sum_page(sbi, segno);
 334         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 335         sum = sum_node->entries[blkoff];
 336         f2fs_put_page(sum_page, 1);
 337 got_it:
 338         /* Use the locked dnode page and inode */
 339         nid = le32_to_cpu(sum.nid);
 340         if (dn->inode->i_ino == nid) {
 341                 tdn.nid = nid;
 342                 if (!dn->inode_page_locked)
 343                         lock_page(dn->inode_page);
 344                 tdn.node_page = dn->inode_page;
 345                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 346                 goto truncate_out;
 347         } else if (dn->nid == nid) {
 348                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 349                 goto truncate_out;
 350         }
 351
 352         /* Get the node page */
 353         node_page = get_node_page(sbi, nid);
 354         if (IS_ERR(node_page))
 355                 return PTR_ERR(node_page);
 356
 357         offset = ofs_of_node(node_page);
 358         ino = ino_of_node(node_page);
 359         f2fs_put_page(node_page, 1);
 360
 361         if (ino != dn->inode->i_ino) {
 362                 /* Deallocate previous index in the node page */
 363                 inode = f2fs_iget(sbi->sb, ino);
 364                 if (IS_ERR(inode))
 365                         return PTR_ERR(inode);
 366         } else {
 367                 inode = dn->inode;
 368         }
 369
 370         bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node);
 371
 372         /*
 373          * if inode page is locked, unlock temporarily, but its reference
 374          * count keeps alive.
 375          */
 376         if (ino == dn->inode->i_ino && dn->inode_page_locked)
 377                 unlock_page(dn->inode_page);
 378
 379         set_new_dnode(&tdn, inode, NULL, NULL, 0);
 380         if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
 381                 goto out;
 382
 383         if (tdn.data_blkaddr == blkaddr)
 384                 truncate_data_blocks_range(&tdn, 1);
 385
 386         f2fs_put_dnode(&tdn);
 387 out:
 388         if (ino != dn->inode->i_ino)
 389                 iput(inode);
 390         else if (dn->inode_page_locked)
 391                 lock_page(dn->inode_page);
 392         return 0;
 393
 394 truncate_out:
 395         if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
 396                 truncate_data_blocks_range(&tdn, 1);
 397         if (dn->inode->i_ino == nid && !dn->inode_page_locked)
 398                 unlock_page(dn->inode_page);
 399         return 0;
 400 }
 401
 402 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 403                                         struct page *page, block_t blkaddr)
 404 {
 405         struct dnode_of_data dn;
 406         struct node_info ni;
 407         unsigned int start, end;
 408         int err = 0, recovered = 0;
 409
 410         /* step 1: recover xattr */
 411         if (IS_INODE(page)) {
 412                 recover_inline_xattr(inode, page);
 413         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 414                 /*
 415                  * Deprecated; xattr blocks should be found from cold log.
 416                  * But, we should remain this for backward compatibility.
 417                  */
 418                 recover_xattr_data(inode, page, blkaddr);
 419                 goto out;
 420         }
 421
 422         /* step 2: recover inline data */
 423         if (recover_inline_data(inode, page))
 424                 goto out;
 425
 426         /* step 3: recover data indices */
 427         start = start_bidx_of_node(ofs_of_node(page), inode);
 428         end = start + ADDRS_PER_PAGE(page, inode);
 429
 430         set_new_dnode(&dn, inode, NULL, NULL, 0);
 431
 432         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 433         if (err)
 434                 goto out;
 435
 436         f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 437
 438         get_node_info(sbi, dn.nid, &ni);
 439         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 440         f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
 441
 442         for (; start < end; start++, dn.ofs_in_node++) {
 443                 block_t src, dest;
 444
 445                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 446                 dest = datablock_addr(page, dn.ofs_in_node);
 447
 448                 /* skip recovering if dest is the same as src */
 449                 if (src == dest)
 450                         continue;
 451
 452                 /* dest is invalid, just invalidate src block */
 453                 if (dest == NULL_ADDR) {
 454                         truncate_data_blocks_range(&dn, 1);
 455                         continue;
 456                 }
 457
 458                 /*
 459                  * dest is reserved block, invalidate src block
 460                  * and then reserve one new block in dnode page.
 461                  */
 462                 if (dest == NEW_ADDR) {
 463                         truncate_data_blocks_range(&dn, 1);
 464                         reserve_new_block(&dn);
 465                         continue;
 466                 }
 467
 468                 /* dest is valid block, try to recover from src to dest */
 469                 if (is_valid_blkaddr(sbi, dest, META_POR)) {
 470
 471                         if (src == NULL_ADDR) {
 472                                 err = reserve_new_block(&dn);
 473 #ifdef CONFIG_F2FS_FAULT_INJECTION
 474                                 while (err)
 475                                         err = reserve_new_block(&dn);
 476 #endif
 477                                 /* We should not get -ENOSPC */
 478                                 f2fs_bug_on(sbi, err);
 479                         }
 480
 481                         /* Check the previous node page having this index */
 482                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 483                         if (err)
 484                                 goto err;
 485
 486                         /* write dummy data page */
 487                         f2fs_replace_block(sbi, &dn, src, dest,
 488                                                 ni.version, false, false);
 489                         recovered++;
 490                 }
 491         }
 492
 493         if (IS_INODE(dn.node_page))
 494                 sync_inode_page(&dn);
 495
 496         copy_node_footer(dn.node_page, page);
 497         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 498                                         ofs_of_node(page), false);
 499         set_page_dirty(dn.node_page);
 500 err:
 501         f2fs_put_dnode(&dn);
 502 out:
 503         f2fs_msg(sbi->sb, KERN_NOTICE,
 504                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 505                 inode->i_ino, recovered, err);
 506         return err;
 507 }
 508
 509 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
 510                                                 struct list_head *dir_list)
 511 {
 512         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 513         struct curseg_info *curseg;
 514         struct page *page = NULL;
 515         int err = 0;
 516         block_t blkaddr;
 517
 518         /* get node pages in the current segment */
 519         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 520         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 521
 522         while (1) {
 523                 struct fsync_inode_entry *entry;
 524
 525                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 526                         break;
 527
 528                 ra_meta_pages_cond(sbi, blkaddr);
 529
 530                 page = get_tmp_page(sbi, blkaddr);
 531
 532                 if (cp_ver != cpver_of_node(page)) {
 533                         f2fs_put_page(page, 1);
 534                         break;
 535                 }
 536
 537                 entry = get_fsync_inode(inode_list, ino_of_node(page));
 538                 if (!entry)
 539                         goto next;
 540                 /*
 541                  * inode(x) | CP | inode(x) | dnode(F)
 542                  * In this case, we can lose the latest inode(x).
 543                  * So, call recover_inode for the inode update.
 544                  */
 545                 if (IS_INODE(page))
 546                         recover_inode(entry->inode, page);
 547                 if (entry->last_dentry == blkaddr) {
 548                         err = recover_dentry(entry->inode, page, dir_list);
 549                         if (err) {
 550                                 f2fs_put_page(page, 1);
 551                                 break;
 552                         }
 553                 }
 554                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 555                 if (err) {
 556                         f2fs_put_page(page, 1);
 557                         break;
 558                 }
 559
 560                 if (entry->blkaddr == blkaddr)
 561                         del_fsync_inode(entry);
 562 next:
 563                 /* check next segment */
 564                 blkaddr = next_blkaddr_of_node(page);
 565                 f2fs_put_page(page, 1);
 566         }
 567         if (!err)
 568                 allocate_new_segments(sbi);
 569         return err;
 570 }
 571
 572 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 573 {
 574         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 575         struct list_head inode_list;
 576         struct list_head dir_list;
 577         block_t blkaddr;
 578         int err;
 579         int ret = 0;
 580         bool need_writecp = false;
 581
 582         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 583                         sizeof(struct fsync_inode_entry));
 584         if (!fsync_entry_slab)
 585                 return -ENOMEM;
 586
 587         INIT_LIST_HEAD(&inode_list);
 588         INIT_LIST_HEAD(&dir_list);
 589
 590         /* prevent checkpoint */
 591         mutex_lock(&sbi->cp_mutex);
 592
 593         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 594
 595         /* step #1: find fsynced inode numbers */
 596         err = find_fsync_dnodes(sbi, &inode_list);
 597         if (err || list_empty(&inode_list))
 598                 goto out;
 599
 600         if (check_only) {
 601                 ret = 1;
 602                 goto out;
 603         }
 604
 605         need_writecp = true;
 606
 607         /* step #2: recover data */
 608         err = recover_data(sbi, &inode_list, &dir_list);
 609         if (!err)
 610                 f2fs_bug_on(sbi, !list_empty(&inode_list));
 611 out:
 612         destroy_fsync_dnodes(&inode_list);
 613
 614         /* truncate meta pages to be used by the recovery */
 615         truncate_inode_pages_range(META_MAPPING(sbi),
 616                         (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
 617
 618         if (err) {
 619                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 620                 truncate_inode_pages_final(META_MAPPING(sbi));
 621         }
 622
 623         clear_sbi_flag(sbi, SBI_POR_DOING);
 624         if (err) {
 625                 bool invalidate = false;
 626
 627                 if (discard_next_dnode(sbi, blkaddr))
 628                         invalidate = true;
 629
 630                 /* Flush all the NAT/SIT pages */
 631                 while (get_pages(sbi, F2FS_DIRTY_META))
 632                         sync_meta_pages(sbi, META, LONG_MAX);
 633
 634                 /* invalidate temporary meta page */
 635                 if (invalidate)
 636                         invalidate_mapping_pages(META_MAPPING(sbi),
 637                                                         blkaddr, blkaddr);
 638
 639                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 640                 mutex_unlock(&sbi->cp_mutex);
 641         } else if (need_writecp) {
 642                 struct cp_control cpc = {
 643                         .reason = CP_RECOVERY,
 644                 };
 645                 mutex_unlock(&sbi->cp_mutex);
 646                 err = write_checkpoint(sbi, &cpc);
 647         } else {
 648                 mutex_unlock(&sbi->cp_mutex);
 649         }
 650
 651         destroy_fsync_dnodes(&dir_list);
 652         kmem_cache_destroy(fsync_entry_slab);
 653         return ret ? ret: err;
 654 }