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         if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
  53                         > sbi->user_block_count)
  54                 return false;
  55         return true;
  56 }
  57
  58 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
  59                                                                 nid_t ino)
  60 {
  61         struct fsync_inode_entry *entry;
  62
  63         list_for_each_entry(entry, head, list)
  64                 if (entry->inode->i_ino == ino)
  65                         return entry;
  66
  67         return NULL;
  68 }
  69
  70 static struct fsync_inode_entry *add_fsync_inode(struct list_head *head,
  71                                                         struct inode *inode)
  72 {
  73         struct fsync_inode_entry *entry;
  74
  75         entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
  76         if (!entry)
  77                 return NULL;
  78
  79         entry->inode = inode;
  80         list_add_tail(&entry->list, head);
  81
  82         return entry;
  83 }
  84
  85 static void del_fsync_inode(struct fsync_inode_entry *entry)
  86 {
  87         iput(entry->inode);
  88         list_del(&entry->list);
  89         kmem_cache_free(fsync_entry_slab, entry);
  90 }
  91
  92 static int recover_dentry(struct inode *inode, struct page *ipage,
  93                                                 struct list_head *dir_list)
  94 {
  95         struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
  96         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  97         struct f2fs_dir_entry *de;
  98         struct qstr name;
  99         struct page *page;
 100         struct inode *dir, *einode;
 101         struct fsync_inode_entry *entry;
 102         int err = 0;
 103
 104         entry = get_fsync_inode(dir_list, pino);
 105         if (!entry) {
 106                 dir = f2fs_iget(inode->i_sb, pino);
 107                 if (IS_ERR(dir)) {
 108                         err = PTR_ERR(dir);
 109                         goto out;
 110                 }
 111
 112                 entry = add_fsync_inode(dir_list, dir);
 113                 if (!entry) {
 114                         err = -ENOMEM;
 115                         iput(dir);
 116                         goto out;
 117                 }
 118         }
 119
 120         dir = entry->inode;
 121
 122         if (file_enc_name(inode))
 123                 return 0;
 124
 125         name.len = le32_to_cpu(raw_inode->i_namelen);
 126         name.name = raw_inode->i_name;
 127
 128         if (unlikely(name.len > F2FS_NAME_LEN)) {
 129                 WARN_ON(1);
 130                 err = -ENAMETOOLONG;
 131                 goto out;
 132         }
 133 retry:
 134         de = f2fs_find_entry(dir, &name, &page);
 135         if (de && inode->i_ino == le32_to_cpu(de->ino))
 136                 goto out_unmap_put;
 137
 138         if (de) {
 139                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
 140                 if (IS_ERR(einode)) {
 141                         WARN_ON(1);
 142                         err = PTR_ERR(einode);
 143                         if (err == -ENOENT)
 144                                 err = -EEXIST;
 145                         goto out_unmap_put;
 146                 }
 147                 err = acquire_orphan_inode(F2FS_I_SB(inode));
 148                 if (err) {
 149                         iput(einode);
 150                         goto out_unmap_put;
 151                 }
 152                 f2fs_delete_entry(de, page, dir, einode);
 153                 iput(einode);
 154                 goto retry;
 155         }
 156         err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
 157
 158         goto out;
 159
 160 out_unmap_put:
 161         f2fs_dentry_kunmap(dir, page);
 162         f2fs_put_page(page, 0);
 163 out:
 164         f2fs_msg(inode->i_sb, KERN_NOTICE,
 165                         "%s: ino = %x, name = %s, dir = %lx, err = %d",
 166                         __func__, ino_of_node(ipage), raw_inode->i_name,
 167                         IS_ERR(dir) ? 0 : dir->i_ino, err);
 168         return err;
 169 }
 170
 171 static void recover_inode(struct inode *inode, struct page *page)
 172 {
 173         struct f2fs_inode *raw = F2FS_INODE(page);
 174         char *name;
 175
 176         inode->i_mode = le16_to_cpu(raw->i_mode);
 177         i_size_write(inode, le64_to_cpu(raw->i_size));
 178         inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
 179         inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
 180         inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
 181         inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 182         inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
 183         inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 184
 185         if (file_enc_name(inode))
 186                 name = "<encrypted>";
 187         else
 188                 name = F2FS_INODE(page)->i_name;
 189
 190         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 191                         ino_of_node(page), name);
 192 }
 193
 194 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 195 {
 196         struct curseg_info *curseg;
 197         struct inode *inode;
 198         struct page *page = NULL;
 199         block_t blkaddr;
 200         int err = 0;
 201
 202         /* get node pages in the current segment */
 203         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 204         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 205
 206         ra_meta_pages(sbi, blkaddr, 1, META_POR, true);
 207
 208         while (1) {
 209                 struct fsync_inode_entry *entry;
 210
 211                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
 212                         return 0;
 213
 214                 page = get_tmp_page(sbi, blkaddr);
 215
 216                 if (!is_recoverable_dnode(page))
 217                         break;
 218
 219                 if (!is_fsync_dnode(page))
 220                         goto next;
 221
 222                 entry = get_fsync_inode(head, ino_of_node(page));
 223                 if (!entry) {
 224                         if (IS_INODE(page) && is_dent_dnode(page)) {
 225                                 err = recover_inode_page(sbi, page);
 226                                 if (err)
 227                                         break;
 228                         }
 229
 230                         /*
 231                          * CP | dnode(F) | inode(DF)
 232                          * For this case, we should not give up now.
 233                          */
 234                         inode = f2fs_iget(sbi->sb, ino_of_node(page));
 235                         if (IS_ERR(inode)) {
 236                                 err = PTR_ERR(inode);
 237                                 if (err == -ENOENT) {
 238                                         err = 0;
 239                                         goto next;
 240                                 }
 241                                 break;
 242                         }
 243
 244                         /* add this fsync inode to the list */
 245                         entry = add_fsync_inode(head, inode);
 246                         if (!entry) {
 247                                 err = -ENOMEM;
 248                                 iput(inode);
 249                                 break;
 250                         }
 251                 }
 252                 entry->blkaddr = blkaddr;
 253
 254                 if (IS_INODE(page)) {
 255                         entry->last_inode = blkaddr;
 256                         if (is_dent_dnode(page))
 257                                 entry->last_dentry = blkaddr;
 258                 }
 259 next:
 260                 /* check next segment */
 261                 blkaddr = next_blkaddr_of_node(page);
 262                 f2fs_put_page(page, 1);
 263
 264                 ra_meta_pages_cond(sbi, blkaddr);
 265         }
 266         f2fs_put_page(page, 1);
 267         return err;
 268 }
 269
 270 static void destroy_fsync_dnodes(struct list_head *head)
 271 {
 272         struct fsync_inode_entry *entry, *tmp;
 273
 274         list_for_each_entry_safe(entry, tmp, head, list)
 275                 del_fsync_inode(entry);
 276 }
 277
 278 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 279                         block_t blkaddr, struct dnode_of_data *dn)
 280 {
 281         struct seg_entry *sentry;
 282         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 283         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 284         struct f2fs_summary_block *sum_node;
 285         struct f2fs_summary sum;
 286         struct page *sum_page, *node_page;
 287         struct dnode_of_data tdn = *dn;
 288         nid_t ino, nid;
 289         struct inode *inode;
 290         unsigned int offset;
 291         block_t bidx;
 292         int i;
 293
 294         sentry = get_seg_entry(sbi, segno);
 295         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 296                 return 0;
 297
 298         /* Get the previous summary */
 299         for (i = CURSEG_HOT_DATA; i <= CURSEG_COLD_DATA; i++) {
 300                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 301                 if (curseg->segno == segno) {
 302                         sum = curseg->sum_blk->entries[blkoff];
 303                         goto got_it;
 304                 }
 305         }
 306
 307         sum_page = get_sum_page(sbi, segno);
 308         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 309         sum = sum_node->entries[blkoff];
 310         f2fs_put_page(sum_page, 1);
 311 got_it:
 312         /* Use the locked dnode page and inode */
 313         nid = le32_to_cpu(sum.nid);
 314         if (dn->inode->i_ino == nid) {
 315                 tdn.nid = nid;
 316                 if (!dn->inode_page_locked)
 317                         lock_page(dn->inode_page);
 318                 tdn.node_page = dn->inode_page;
 319                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 320                 goto truncate_out;
 321         } else if (dn->nid == nid) {
 322                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 323                 goto truncate_out;
 324         }
 325
 326         /* Get the node page */
 327         node_page = get_node_page(sbi, nid);
 328         if (IS_ERR(node_page))
 329                 return PTR_ERR(node_page);
 330
 331         offset = ofs_of_node(node_page);
 332         ino = ino_of_node(node_page);
 333         f2fs_put_page(node_page, 1);
 334
 335         if (ino != dn->inode->i_ino) {
 336                 /* Deallocate previous index in the node page */
 337                 inode = f2fs_iget(sbi->sb, ino);
 338                 if (IS_ERR(inode))
 339                         return PTR_ERR(inode);
 340         } else {
 341                 inode = dn->inode;
 342         }
 343
 344         bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
 345                         le16_to_cpu(sum.ofs_in_node);
 346
 347         /*
 348          * if inode page is locked, unlock temporarily, but its reference
 349          * count keeps alive.
 350          */
 351         if (ino == dn->inode->i_ino && dn->inode_page_locked)
 352                 unlock_page(dn->inode_page);
 353
 354         set_new_dnode(&tdn, inode, NULL, NULL, 0);
 355         if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
 356                 goto out;
 357
 358         if (tdn.data_blkaddr == blkaddr)
 359                 truncate_data_blocks_range(&tdn, 1);
 360
 361         f2fs_put_dnode(&tdn);
 362 out:
 363         if (ino != dn->inode->i_ino)
 364                 iput(inode);
 365         else if (dn->inode_page_locked)
 366                 lock_page(dn->inode_page);
 367         return 0;
 368
 369 truncate_out:
 370         if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
 371                 truncate_data_blocks_range(&tdn, 1);
 372         if (dn->inode->i_ino == nid && !dn->inode_page_locked)
 373                 unlock_page(dn->inode_page);
 374         return 0;
 375 }
 376
 377 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 378                                         struct page *page, block_t blkaddr)
 379 {
 380         struct f2fs_inode_info *fi = F2FS_I(inode);
 381         unsigned int start, end;
 382         struct dnode_of_data dn;
 383         struct node_info ni;
 384         int err = 0, recovered = 0;
 385
 386         /* step 1: recover xattr */
 387         if (IS_INODE(page)) {
 388                 recover_inline_xattr(inode, page);
 389         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 390                 /*
 391                  * Deprecated; xattr blocks should be found from cold log.
 392                  * But, we should remain this for backward compatibility.
 393                  */
 394                 recover_xattr_data(inode, page, blkaddr);
 395                 goto out;
 396         }
 397
 398         /* step 2: recover inline data */
 399         if (recover_inline_data(inode, page))
 400                 goto out;
 401
 402         /* step 3: recover data indices */
 403         start = start_bidx_of_node(ofs_of_node(page), fi);
 404         end = start + ADDRS_PER_PAGE(page, fi);
 405
 406         set_new_dnode(&dn, inode, NULL, NULL, 0);
 407
 408         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 409         if (err)
 410                 goto out;
 411
 412         f2fs_wait_on_page_writeback(dn.node_page, NODE);
 413
 414         get_node_info(sbi, dn.nid, &ni);
 415         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 416
 417         if (ofs_of_node(dn.node_page) != ofs_of_node(page)) {
 418                 f2fs_msg(sbi->sb, KERN_WARNING,
 419                         "Inconsistent ofs_of_node, ino:%lu, ofs:%u, %u",
 420                         inode->i_ino, ofs_of_node(dn.node_page),
 421                         ofs_of_node(page));
 422                 err = -EFAULT;
 423                 goto err;
 424         }
 425
 426         for (; start < end; start++, dn.ofs_in_node++) {
 427                 block_t src, dest;
 428
 429                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 430                 dest = datablock_addr(page, dn.ofs_in_node);
 431
 432                 /* skip recovering if dest is the same as src */
 433                 if (src == dest)
 434                         continue;
 435
 436                 /* dest is invalid, just invalidate src block */
 437                 if (dest == NULL_ADDR) {
 438                         truncate_data_blocks_range(&dn, 1);
 439                         continue;
 440                 }
 441
 442                 /*
 443                  * dest is reserved block, invalidate src block
 444                  * and then reserve one new block in dnode page.
 445                  */
 446                 if (dest == NEW_ADDR) {
 447                         truncate_data_blocks_range(&dn, 1);
 448                         err = reserve_new_block(&dn);
 449                         f2fs_bug_on(sbi, err);
 450                         continue;
 451                 }
 452
 453                 /* dest is valid block, try to recover from src to dest */
 454                 if (f2fs_is_valid_blkaddr(sbi, dest, META_POR)) {
 455
 456                         if (src == NULL_ADDR) {
 457                                 err = reserve_new_block(&dn);
 458                                 /* We should not get -ENOSPC */
 459                                 f2fs_bug_on(sbi, err);
 460                         }
 461
 462                         /* Check the previous node page having this index */
 463                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 464                         if (err)
 465                                 goto err;
 466
 467                         /* write dummy data page */
 468                         f2fs_replace_block(sbi, &dn, src, dest,
 469                                                         ni.version, false);
 470                         recovered++;
 471                 }
 472         }
 473
 474         if (IS_INODE(dn.node_page))
 475                 sync_inode_page(&dn);
 476
 477         copy_node_footer(dn.node_page, page);
 478         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 479                                         ofs_of_node(page), false);
 480         set_page_dirty(dn.node_page);
 481 err:
 482         f2fs_put_dnode(&dn);
 483 out:
 484         f2fs_msg(sbi->sb, KERN_NOTICE,
 485                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 486                 inode->i_ino, recovered, err);
 487         return err;
 488 }
 489
 490 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *inode_list,
 491                                                 struct list_head *dir_list)
 492 {
 493         struct curseg_info *curseg;
 494         struct page *page = NULL;
 495         int err = 0;
 496         block_t blkaddr;
 497
 498         /* get node pages in the current segment */
 499         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 500         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 501
 502         while (1) {
 503                 struct fsync_inode_entry *entry;
 504
 505                 if (!f2fs_is_valid_blkaddr(sbi, blkaddr, META_POR))
 506                         break;
 507
 508                 ra_meta_pages_cond(sbi, blkaddr);
 509
 510                 page = get_tmp_page(sbi, blkaddr);
 511
 512                 if (!is_recoverable_dnode(page)) {
 513                         f2fs_put_page(page, 1);
 514                         break;
 515                 }
 516
 517                 entry = get_fsync_inode(inode_list, ino_of_node(page));
 518                 if (!entry)
 519                         goto next;
 520                 /*
 521                  * inode(x) | CP | inode(x) | dnode(F)
 522                  * In this case, we can lose the latest inode(x).
 523                  * So, call recover_inode for the inode update.
 524                  */
 525                 if (entry->last_inode == blkaddr)
 526                         recover_inode(entry->inode, page);
 527                 if (entry->last_dentry == blkaddr) {
 528                         err = recover_dentry(entry->inode, page, dir_list);
 529                         if (err) {
 530                                 f2fs_put_page(page, 1);
 531                                 break;
 532                         }
 533                 }
 534                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 535                 if (err) {
 536                         f2fs_put_page(page, 1);
 537                         break;
 538                 }
 539
 540                 if (entry->blkaddr == blkaddr)
 541                         del_fsync_inode(entry);
 542 next:
 543                 /* check next segment */
 544                 blkaddr = next_blkaddr_of_node(page);
 545                 f2fs_put_page(page, 1);
 546         }
 547         if (!err)
 548                 allocate_new_segments(sbi);
 549         return err;
 550 }
 551
 552 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 553 {
 554         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 555         struct list_head inode_list;
 556         struct list_head dir_list;
 557         block_t blkaddr;
 558         int err;
 559         int ret = 0;
 560         bool need_writecp = false;
 561
 562         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 563                         sizeof(struct fsync_inode_entry));
 564         if (!fsync_entry_slab)
 565                 return -ENOMEM;
 566
 567         INIT_LIST_HEAD(&inode_list);
 568         INIT_LIST_HEAD(&dir_list);
 569
 570         /* prevent checkpoint */
 571         mutex_lock(&sbi->cp_mutex);
 572
 573         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 574
 575         /* step #1: find fsynced inode numbers */
 576         err = find_fsync_dnodes(sbi, &inode_list);
 577         if (err || list_empty(&inode_list))
 578                 goto out;
 579
 580         if (check_only) {
 581                 ret = 1;
 582                 goto out;
 583         }
 584
 585         need_writecp = true;
 586
 587         /* step #2: recover data */
 588         err = recover_data(sbi, &inode_list, &dir_list);
 589         if (!err)
 590                 f2fs_bug_on(sbi, !list_empty(&inode_list));
 591 out:
 592         destroy_fsync_dnodes(&inode_list);
 593
 594         /* truncate meta pages to be used by the recovery */
 595         truncate_inode_pages_range(META_MAPPING(sbi),
 596                         (loff_t)MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
 597
 598         if (err) {
 599                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 600                 truncate_inode_pages_final(META_MAPPING(sbi));
 601         }
 602
 603         clear_sbi_flag(sbi, SBI_POR_DOING);
 604         if (err)
 605                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 606         mutex_unlock(&sbi->cp_mutex);
 607
 608         /* let's drop all the directory inodes for clean checkpoint */
 609         destroy_fsync_dnodes(&dir_list);
 610
 611         if (!err && need_writecp) {
 612                 struct cp_control cpc = {
 613                         .reason = CP_RECOVERY,
 614                 };
 615                 write_checkpoint(sbi, &cpc);
 616         }
 617
 618         kmem_cache_destroy(fsync_entry_slab);
 619         return ret ? ret: err;
 620 }