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 int recover_dentry(struct inode *inode, struct page *ipage)
  71 {
  72         struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
  73         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  74         struct f2fs_dir_entry *de;
  75         struct qstr name;
  76         struct page *page;
  77         struct inode *dir, *einode;
  78         int err = 0;
  79
  80         dir = f2fs_iget(inode->i_sb, pino);
  81         if (IS_ERR(dir)) {
  82                 err = PTR_ERR(dir);
  83                 goto out;
  84         }
  85
  86         if (file_enc_name(inode)) {
  87                 iput(dir);
  88                 return 0;
  89         }
  90
  91         name.len = le32_to_cpu(raw_inode->i_namelen);
  92         name.name = raw_inode->i_name;
  93
  94         if (unlikely(name.len > F2FS_NAME_LEN)) {
  95                 WARN_ON(1);
  96                 err = -ENAMETOOLONG;
  97                 goto out_err;
  98         }
  99 retry:
 100         de = f2fs_find_entry(dir, &name, &page);
 101         if (de && inode->i_ino == le32_to_cpu(de->ino))
 102                 goto out_unmap_put;
 103
 104         if (de) {
 105                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
 106                 if (IS_ERR(einode)) {
 107                         WARN_ON(1);
 108                         err = PTR_ERR(einode);
 109                         if (err == -ENOENT)
 110                                 err = -EEXIST;
 111                         goto out_unmap_put;
 112                 }
 113                 err = acquire_orphan_inode(F2FS_I_SB(inode));
 114                 if (err) {
 115                         iput(einode);
 116                         goto out_unmap_put;
 117                 }
 118                 f2fs_delete_entry(de, page, dir, einode);
 119                 iput(einode);
 120                 goto retry;
 121         }
 122         err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
 123         if (err)
 124                 goto out_err;
 125
 126         if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
 127                 iput(dir);
 128         } else {
 129                 add_dirty_dir_inode(dir);
 130                 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
 131         }
 132
 133         goto out;
 134
 135 out_unmap_put:
 136         f2fs_dentry_kunmap(dir, page);
 137         f2fs_put_page(page, 0);
 138 out_err:
 139         iput(dir);
 140 out:
 141         f2fs_msg(inode->i_sb, KERN_NOTICE,
 142                         "%s: ino = %x, name = %s, dir = %lx, err = %d",
 143                         __func__, ino_of_node(ipage), raw_inode->i_name,
 144                         IS_ERR(dir) ? 0 : dir->i_ino, err);
 145         return err;
 146 }
 147
 148 static void recover_inode(struct inode *inode, struct page *page)
 149 {
 150         struct f2fs_inode *raw = F2FS_INODE(page);
 151         char *name;
 152
 153         inode->i_mode = le16_to_cpu(raw->i_mode);
 154         i_size_write(inode, le64_to_cpu(raw->i_size));
 155         inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
 156         inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
 157         inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
 158         inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 159         inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
 160         inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 161
 162         if (file_enc_name(inode))
 163                 name = "<encrypted>";
 164         else
 165                 name = F2FS_INODE(page)->i_name;
 166
 167         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 168                         ino_of_node(page), name);
 169 }
 170
 171 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 172 {
 173         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 174         struct curseg_info *curseg;
 175         struct page *page = NULL;
 176         block_t blkaddr;
 177         int err = 0;
 178
 179         /* get node pages in the current segment */
 180         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 181         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 182
 183         ra_meta_pages(sbi, blkaddr, 1, META_POR, true);
 184
 185         while (1) {
 186                 struct fsync_inode_entry *entry;
 187
 188                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 189                         return 0;
 190
 191                 page = get_tmp_page(sbi, blkaddr);
 192
 193                 if (cp_ver != cpver_of_node(page))
 194                         break;
 195
 196                 if (!is_fsync_dnode(page))
 197                         goto next;
 198
 199                 entry = get_fsync_inode(head, ino_of_node(page));
 200                 if (!entry) {
 201                         if (IS_INODE(page) && is_dent_dnode(page)) {
 202                                 err = recover_inode_page(sbi, page);
 203                                 if (err)
 204                                         break;
 205                         }
 206
 207                         /* add this fsync inode to the list */
 208                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_F2FS_ZERO);
 209                         if (!entry) {
 210                                 err = -ENOMEM;
 211                                 break;
 212                         }
 213                         /*
 214                          * CP | dnode(F) | inode(DF)
 215                          * For this case, we should not give up now.
 216                          */
 217                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 218                         if (IS_ERR(entry->inode)) {
 219                                 err = PTR_ERR(entry->inode);
 220                                 kmem_cache_free(fsync_entry_slab, entry);
 221                                 if (err == -ENOENT) {
 222                                         err = 0;
 223                                         goto next;
 224                                 }
 225                                 break;
 226                         }
 227                         list_add_tail(&entry->list, head);
 228                 }
 229                 entry->blkaddr = blkaddr;
 230
 231                 if (IS_INODE(page)) {
 232                         entry->last_inode = blkaddr;
 233                         if (is_dent_dnode(page))
 234                                 entry->last_dentry = blkaddr;
 235                 }
 236 next:
 237                 /* check next segment */
 238                 blkaddr = next_blkaddr_of_node(page);
 239                 f2fs_put_page(page, 1);
 240
 241                 ra_meta_pages_cond(sbi, blkaddr);
 242         }
 243         f2fs_put_page(page, 1);
 244         return err;
 245 }
 246
 247 static void destroy_fsync_dnodes(struct list_head *head)
 248 {
 249         struct fsync_inode_entry *entry, *tmp;
 250
 251         list_for_each_entry_safe(entry, tmp, head, list) {
 252                 iput(entry->inode);
 253                 list_del(&entry->list);
 254                 kmem_cache_free(fsync_entry_slab, entry);
 255         }
 256 }
 257
 258 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 259                         block_t blkaddr, struct dnode_of_data *dn)
 260 {
 261         struct seg_entry *sentry;
 262         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 263         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 264         struct f2fs_summary_block *sum_node;
 265         struct f2fs_summary sum;
 266         struct page *sum_page, *node_page;
 267         struct dnode_of_data tdn = *dn;
 268         nid_t ino, nid;
 269         struct inode *inode;
 270         unsigned int offset;
 271         block_t bidx;
 272         int i;
 273
 274         sentry = get_seg_entry(sbi, segno);
 275         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 276                 return 0;
 277
 278         /* Get the previous summary */
 279         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 280                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 281                 if (curseg->segno == segno) {
 282                         sum = curseg->sum_blk->entries[blkoff];
 283                         goto got_it;
 284                 }
 285         }
 286
 287         sum_page = get_sum_page(sbi, segno);
 288         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 289         sum = sum_node->entries[blkoff];
 290         f2fs_put_page(sum_page, 1);
 291 got_it:
 292         /* Use the locked dnode page and inode */
 293         nid = le32_to_cpu(sum.nid);
 294         if (dn->inode->i_ino == nid) {
 295                 tdn.nid = nid;
 296                 if (!dn->inode_page_locked)
 297                         lock_page(dn->inode_page);
 298                 tdn.node_page = dn->inode_page;
 299                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 300                 goto truncate_out;
 301         } else if (dn->nid == nid) {
 302                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 303                 goto truncate_out;
 304         }
 305
 306         /* Get the node page */
 307         node_page = get_node_page(sbi, nid);
 308         if (IS_ERR(node_page))
 309                 return PTR_ERR(node_page);
 310
 311         offset = ofs_of_node(node_page);
 312         ino = ino_of_node(node_page);
 313         f2fs_put_page(node_page, 1);
 314
 315         if (ino != dn->inode->i_ino) {
 316                 /* Deallocate previous index in the node page */
 317                 inode = f2fs_iget(sbi->sb, ino);
 318                 if (IS_ERR(inode))
 319                         return PTR_ERR(inode);
 320         } else {
 321                 inode = dn->inode;
 322         }
 323
 324         bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
 325                         le16_to_cpu(sum.ofs_in_node);
 326
 327         /*
 328          * if inode page is locked, unlock temporarily, but its reference
 329          * count keeps alive.
 330          */
 331         if (ino == dn->inode->i_ino && dn->inode_page_locked)
 332                 unlock_page(dn->inode_page);
 333
 334         set_new_dnode(&tdn, inode, NULL, NULL, 0);
 335         if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
 336                 goto out;
 337
 338         if (tdn.data_blkaddr == blkaddr)
 339                 truncate_data_blocks_range(&tdn, 1);
 340
 341         f2fs_put_dnode(&tdn);
 342 out:
 343         if (ino != dn->inode->i_ino)
 344                 iput(inode);
 345         else if (dn->inode_page_locked)
 346                 lock_page(dn->inode_page);
 347         return 0;
 348
 349 truncate_out:
 350         if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
 351                 truncate_data_blocks_range(&tdn, 1);
 352         if (dn->inode->i_ino == nid && !dn->inode_page_locked)
 353                 unlock_page(dn->inode_page);
 354         return 0;
 355 }
 356
 357 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 358                                         struct page *page, block_t blkaddr)
 359 {
 360         struct f2fs_inode_info *fi = F2FS_I(inode);
 361         unsigned int start, end;
 362         struct dnode_of_data dn;
 363         struct node_info ni;
 364         int err = 0, recovered = 0;
 365
 366         /* step 1: recover xattr */
 367         if (IS_INODE(page)) {
 368                 recover_inline_xattr(inode, page);
 369         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 370                 /*
 371                  * Deprecated; xattr blocks should be found from cold log.
 372                  * But, we should remain this for backward compatibility.
 373                  */
 374                 recover_xattr_data(inode, page, blkaddr);
 375                 goto out;
 376         }
 377
 378         /* step 2: recover inline data */
 379         if (recover_inline_data(inode, page))
 380                 goto out;
 381
 382         /* step 3: recover data indices */
 383         start = start_bidx_of_node(ofs_of_node(page), fi);
 384         end = start + ADDRS_PER_PAGE(page, fi);
 385
 386         set_new_dnode(&dn, inode, NULL, NULL, 0);
 387
 388         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 389         if (err)
 390                 goto out;
 391
 392         f2fs_wait_on_page_writeback(dn.node_page, NODE);
 393
 394         get_node_info(sbi, dn.nid, &ni);
 395         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 396         f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
 397
 398         for (; start < end; start++, dn.ofs_in_node++) {
 399                 block_t src, dest;
 400
 401                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 402                 dest = datablock_addr(page, dn.ofs_in_node);
 403
 404                 /* skip recovering if dest is the same as src */
 405                 if (src == dest)
 406                         continue;
 407
 408                 /* dest is invalid, just invalidate src block */
 409                 if (dest == NULL_ADDR) {
 410                         truncate_data_blocks_range(&dn, 1);
 411                         continue;
 412                 }
 413
 414                 /*
 415                  * dest is reserved block, invalidate src block
 416                  * and then reserve one new block in dnode page.
 417                  */
 418                 if (dest == NEW_ADDR) {
 419                         truncate_data_blocks_range(&dn, 1);
 420                         err = reserve_new_block(&dn);
 421                         f2fs_bug_on(sbi, err);
 422                         continue;
 423                 }
 424
 425                 /* dest is valid block, try to recover from src to dest */
 426                 if (is_valid_blkaddr(sbi, dest, META_POR)) {
 427
 428                         if (src == NULL_ADDR) {
 429                                 err = reserve_new_block(&dn);
 430                                 /* We should not get -ENOSPC */
 431                                 f2fs_bug_on(sbi, err);
 432                         }
 433
 434                         /* Check the previous node page having this index */
 435                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 436                         if (err)
 437                                 goto err;
 438
 439                         /* write dummy data page */
 440                         f2fs_replace_block(sbi, &dn, src, dest,
 441                                                         ni.version, false);
 442                         recovered++;
 443                 }
 444         }
 445
 446         if (IS_INODE(dn.node_page))
 447                 sync_inode_page(&dn);
 448
 449         copy_node_footer(dn.node_page, page);
 450         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 451                                         ofs_of_node(page), false);
 452         set_page_dirty(dn.node_page);
 453 err:
 454         f2fs_put_dnode(&dn);
 455 out:
 456         f2fs_msg(sbi->sb, KERN_NOTICE,
 457                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 458                 inode->i_ino, recovered, err);
 459         return err;
 460 }
 461
 462 static int recover_data(struct f2fs_sb_info *sbi,
 463                                 struct list_head *head, int type)
 464 {
 465         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 466         struct curseg_info *curseg;
 467         struct page *page = NULL;
 468         int err = 0;
 469         block_t blkaddr;
 470
 471         /* get node pages in the current segment */
 472         curseg = CURSEG_I(sbi, type);
 473         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 474
 475         while (1) {
 476                 struct fsync_inode_entry *entry;
 477
 478                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 479                         break;
 480
 481                 ra_meta_pages_cond(sbi, blkaddr);
 482
 483                 page = get_tmp_page(sbi, blkaddr);
 484
 485                 if (cp_ver != cpver_of_node(page)) {
 486                         f2fs_put_page(page, 1);
 487                         break;
 488                 }
 489
 490                 entry = get_fsync_inode(head, ino_of_node(page));
 491                 if (!entry)
 492                         goto next;
 493                 /*
 494                  * inode(x) | CP | inode(x) | dnode(F)
 495                  * In this case, we can lose the latest inode(x).
 496                  * So, call recover_inode for the inode update.
 497                  */
 498                 if (entry->last_inode == blkaddr)
 499                         recover_inode(entry->inode, page);
 500                 if (entry->last_dentry == blkaddr) {
 501                         err = recover_dentry(entry->inode, page);
 502                         if (err) {
 503                                 f2fs_put_page(page, 1);
 504                                 break;
 505                         }
 506                 }
 507                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 508                 if (err) {
 509                         f2fs_put_page(page, 1);
 510                         break;
 511                 }
 512
 513                 if (entry->blkaddr == blkaddr) {
 514                         iput(entry->inode);
 515                         list_del(&entry->list);
 516                         kmem_cache_free(fsync_entry_slab, entry);
 517                 }
 518 next:
 519                 /* check next segment */
 520                 blkaddr = next_blkaddr_of_node(page);
 521                 f2fs_put_page(page, 1);
 522         }
 523         if (!err)
 524                 allocate_new_segments(sbi);
 525         return err;
 526 }
 527
 528 int recover_fsync_data(struct f2fs_sb_info *sbi)
 529 {
 530         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 531         struct list_head inode_list;
 532         block_t blkaddr;
 533         int err;
 534         bool need_writecp = false;
 535
 536         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 537                         sizeof(struct fsync_inode_entry));
 538         if (!fsync_entry_slab)
 539                 return -ENOMEM;
 540
 541         INIT_LIST_HEAD(&inode_list);
 542
 543         /* prevent checkpoint */
 544         mutex_lock(&sbi->cp_mutex);
 545
 546         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 547
 548         /* step #1: find fsynced inode numbers */
 549         err = find_fsync_dnodes(sbi, &inode_list);
 550         if (err)
 551                 goto out;
 552
 553         if (list_empty(&inode_list))
 554                 goto out;
 555
 556         need_writecp = true;
 557
 558         /* step #2: recover data */
 559         err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 560         if (!err)
 561                 f2fs_bug_on(sbi, !list_empty(&inode_list));
 562 out:
 563         destroy_fsync_dnodes(&inode_list);
 564         kmem_cache_destroy(fsync_entry_slab);
 565
 566         /* truncate meta pages to be used by the recovery */
 567         truncate_inode_pages_range(META_MAPPING(sbi),
 568                         (loff_t)MAIN_BLKADDR(sbi) << PAGE_CACHE_SHIFT, -1);
 569
 570         if (err) {
 571                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 572                 truncate_inode_pages_final(META_MAPPING(sbi));
 573         }
 574
 575         clear_sbi_flag(sbi, SBI_POR_DOING);
 576         if (err) {
 577                 bool invalidate = false;
 578
 579                 if (discard_next_dnode(sbi, blkaddr))
 580                         invalidate = true;
 581
 582                 /* Flush all the NAT/SIT pages */
 583                 while (get_pages(sbi, F2FS_DIRTY_META))
 584                         sync_meta_pages(sbi, META, LONG_MAX);
 585
 586                 /* invalidate temporary meta page */
 587                 if (invalidate)
 588                         invalidate_mapping_pages(META_MAPPING(sbi),
 589                                                         blkaddr, blkaddr);
 590
 591                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 592                 mutex_unlock(&sbi->cp_mutex);
 593         } else if (need_writecp) {
 594                 struct cp_control cpc = {
 595                         .reason = CP_RECOVERY,
 596                 };
 597                 mutex_unlock(&sbi->cp_mutex);
 598                 write_checkpoint(sbi, &cpc);
 599         } else {
 600                 mutex_unlock(&sbi->cp_mutex);
 601         }
 602         return err;
 603 }