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 {
  94         struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
  95         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  96         struct f2fs_dir_entry *de;
  97         struct qstr name;
  98         struct page *page;
  99         struct inode *dir, *einode;
 100         int err = 0;
 101
 102         dir = f2fs_iget(inode->i_sb, pino);
 103         if (IS_ERR(dir)) {
 104                 err = PTR_ERR(dir);
 105                 goto out;
 106         }
 107
 108         if (file_enc_name(inode)) {
 109                 iput(dir);
 110                 return 0;
 111         }
 112
 113         name.len = le32_to_cpu(raw_inode->i_namelen);
 114         name.name = raw_inode->i_name;
 115
 116         if (unlikely(name.len > F2FS_NAME_LEN)) {
 117                 WARN_ON(1);
 118                 err = -ENAMETOOLONG;
 119                 goto out_err;
 120         }
 121 retry:
 122         de = f2fs_find_entry(dir, &name, &page);
 123         if (de && inode->i_ino == le32_to_cpu(de->ino))
 124                 goto out_unmap_put;
 125
 126         if (de) {
 127                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
 128                 if (IS_ERR(einode)) {
 129                         WARN_ON(1);
 130                         err = PTR_ERR(einode);
 131                         if (err == -ENOENT)
 132                                 err = -EEXIST;
 133                         goto out_unmap_put;
 134                 }
 135                 err = acquire_orphan_inode(F2FS_I_SB(inode));
 136                 if (err) {
 137                         iput(einode);
 138                         goto out_unmap_put;
 139                 }
 140                 f2fs_delete_entry(de, page, dir, einode);
 141                 iput(einode);
 142                 goto retry;
 143         }
 144         err = __f2fs_add_link(dir, &name, inode, inode->i_ino, inode->i_mode);
 145         if (err)
 146                 goto out_err;
 147
 148         if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
 149                 iput(dir);
 150         } else {
 151                 add_dirty_dir_inode(dir);
 152                 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
 153         }
 154
 155         goto out;
 156
 157 out_unmap_put:
 158         f2fs_dentry_kunmap(dir, page);
 159         f2fs_put_page(page, 0);
 160 out_err:
 161         iput(dir);
 162 out:
 163         f2fs_msg(inode->i_sb, KERN_NOTICE,
 164                         "%s: ino = %x, name = %s, dir = %lx, err = %d",
 165                         __func__, ino_of_node(ipage), raw_inode->i_name,
 166                         IS_ERR(dir) ? 0 : dir->i_ino, err);
 167         return err;
 168 }
 169
 170 static void recover_inode(struct inode *inode, struct page *page)
 171 {
 172         struct f2fs_inode *raw = F2FS_INODE(page);
 173         char *name;
 174
 175         inode->i_mode = le16_to_cpu(raw->i_mode);
 176         i_size_write(inode, le64_to_cpu(raw->i_size));
 177         inode->i_atime.tv_sec = le64_to_cpu(raw->i_mtime);
 178         inode->i_ctime.tv_sec = le64_to_cpu(raw->i_ctime);
 179         inode->i_mtime.tv_sec = le64_to_cpu(raw->i_mtime);
 180         inode->i_atime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 181         inode->i_ctime.tv_nsec = le32_to_cpu(raw->i_ctime_nsec);
 182         inode->i_mtime.tv_nsec = le32_to_cpu(raw->i_mtime_nsec);
 183
 184         if (file_enc_name(inode))
 185                 name = "<encrypted>";
 186         else
 187                 name = F2FS_INODE(page)->i_name;
 188
 189         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 190                         ino_of_node(page), name);
 191 }
 192
 193 static bool is_same_inode(struct inode *inode, struct page *ipage)
 194 {
 195         struct f2fs_inode *ri = F2FS_INODE(ipage);
 196         struct timespec disk;
 197
 198         if (!IS_INODE(ipage))
 199                 return true;
 200
 201         disk.tv_sec = le64_to_cpu(ri->i_ctime);
 202         disk.tv_nsec = le32_to_cpu(ri->i_ctime_nsec);
 203         if (timespec_compare(&inode->i_ctime, &disk) > 0)
 204                 return false;
 205
 206         disk.tv_sec = le64_to_cpu(ri->i_atime);
 207         disk.tv_nsec = le32_to_cpu(ri->i_atime_nsec);
 208         if (timespec_compare(&inode->i_atime, &disk) > 0)
 209                 return false;
 210
 211         disk.tv_sec = le64_to_cpu(ri->i_mtime);
 212         disk.tv_nsec = le32_to_cpu(ri->i_mtime_nsec);
 213         if (timespec_compare(&inode->i_mtime, &disk) > 0)
 214                 return false;
 215
 216         return true;
 217 }
 218
 219 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 220 {
 221         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 222         struct curseg_info *curseg;
 223         struct inode *inode;
 224         struct page *page = NULL;
 225         block_t blkaddr;
 226         int err = 0;
 227
 228         /* get node pages in the current segment */
 229         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 230         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 231
 232         while (1) {
 233                 struct fsync_inode_entry *entry;
 234
 235                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 236                         return 0;
 237
 238                 page = get_tmp_page(sbi, blkaddr);
 239
 240                 if (cp_ver != cpver_of_node(page))
 241                         break;
 242
 243                 if (!is_fsync_dnode(page))
 244                         goto next;
 245
 246                 entry = get_fsync_inode(head, ino_of_node(page));
 247                 if (entry) {
 248                         if (!is_same_inode(entry->inode, page))
 249                                 goto next;
 250                 } else {
 251                         if (IS_INODE(page) && is_dent_dnode(page)) {
 252                                 err = recover_inode_page(sbi, page);
 253                                 if (err)
 254                                         break;
 255                         }
 256
 257                         /*
 258                          * CP | dnode(F) | inode(DF)
 259                          * For this case, we should not give up now.
 260                          */
 261                         inode = f2fs_iget(sbi->sb, ino_of_node(page));
 262                         if (IS_ERR(inode)) {
 263                                 err = PTR_ERR(inode);
 264                                 if (err == -ENOENT) {
 265                                         err = 0;
 266                                         goto next;
 267                                 }
 268                                 break;
 269                         }
 270
 271                         /* add this fsync inode to the list */
 272                         entry = add_fsync_inode(head, inode);
 273                         if (!entry) {
 274                                 err = -ENOMEM;
 275                                 iput(inode);
 276                                 break;
 277                         }
 278                 }
 279                 entry->blkaddr = blkaddr;
 280
 281                 if (IS_INODE(page) && is_dent_dnode(page))
 282                         entry->last_dentry = blkaddr;
 283 next:
 284                 /* check next segment */
 285                 blkaddr = next_blkaddr_of_node(page);
 286                 f2fs_put_page(page, 1);
 287
 288                 ra_meta_pages_cond(sbi, blkaddr);
 289         }
 290         f2fs_put_page(page, 1);
 291         return err;
 292 }
 293
 294 static void destroy_fsync_dnodes(struct list_head *head)
 295 {
 296         struct fsync_inode_entry *entry, *tmp;
 297
 298         list_for_each_entry_safe(entry, tmp, head, list)
 299                 del_fsync_inode(entry);
 300 }
 301
 302 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 303                         block_t blkaddr, struct dnode_of_data *dn)
 304 {
 305         struct seg_entry *sentry;
 306         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 307         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 308         struct f2fs_summary_block *sum_node;
 309         struct f2fs_summary sum;
 310         struct page *sum_page, *node_page;
 311         struct dnode_of_data tdn = *dn;
 312         nid_t ino, nid;
 313         struct inode *inode;
 314         unsigned int offset;
 315         block_t bidx;
 316         int i;
 317
 318         sentry = get_seg_entry(sbi, segno);
 319         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 320                 return 0;
 321
 322         /* Get the previous summary */
 323         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 324                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 325                 if (curseg->segno == segno) {
 326                         sum = curseg->sum_blk->entries[blkoff];
 327                         goto got_it;
 328                 }
 329         }
 330
 331         sum_page = get_sum_page(sbi, segno);
 332         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 333         sum = sum_node->entries[blkoff];
 334         f2fs_put_page(sum_page, 1);
 335 got_it:
 336         /* Use the locked dnode page and inode */
 337         nid = le32_to_cpu(sum.nid);
 338         if (dn->inode->i_ino == nid) {
 339                 tdn.nid = nid;
 340                 if (!dn->inode_page_locked)
 341                         lock_page(dn->inode_page);
 342                 tdn.node_page = dn->inode_page;
 343                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 344                 goto truncate_out;
 345         } else if (dn->nid == nid) {
 346                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 347                 goto truncate_out;
 348         }
 349
 350         /* Get the node page */
 351         node_page = get_node_page(sbi, nid);
 352         if (IS_ERR(node_page))
 353                 return PTR_ERR(node_page);
 354
 355         offset = ofs_of_node(node_page);
 356         ino = ino_of_node(node_page);
 357         f2fs_put_page(node_page, 1);
 358
 359         if (ino != dn->inode->i_ino) {
 360                 /* Deallocate previous index in the node page */
 361                 inode = f2fs_iget(sbi->sb, ino);
 362                 if (IS_ERR(inode))
 363                         return PTR_ERR(inode);
 364         } else {
 365                 inode = dn->inode;
 366         }
 367
 368         bidx = start_bidx_of_node(offset, inode) + le16_to_cpu(sum.ofs_in_node);
 369
 370         /*
 371          * if inode page is locked, unlock temporarily, but its reference
 372          * count keeps alive.
 373          */
 374         if (ino == dn->inode->i_ino && dn->inode_page_locked)
 375                 unlock_page(dn->inode_page);
 376
 377         set_new_dnode(&tdn, inode, NULL, NULL, 0);
 378         if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
 379                 goto out;
 380
 381         if (tdn.data_blkaddr == blkaddr)
 382                 truncate_data_blocks_range(&tdn, 1);
 383
 384         f2fs_put_dnode(&tdn);
 385 out:
 386         if (ino != dn->inode->i_ino)
 387                 iput(inode);
 388         else if (dn->inode_page_locked)
 389                 lock_page(dn->inode_page);
 390         return 0;
 391
 392 truncate_out:
 393         if (datablock_addr(tdn.node_page, tdn.ofs_in_node) == blkaddr)
 394                 truncate_data_blocks_range(&tdn, 1);
 395         if (dn->inode->i_ino == nid && !dn->inode_page_locked)
 396                 unlock_page(dn->inode_page);
 397         return 0;
 398 }
 399
 400 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 401                                         struct page *page, block_t blkaddr)
 402 {
 403         struct dnode_of_data dn;
 404         struct node_info ni;
 405         unsigned int start, end;
 406         int err = 0, recovered = 0;
 407
 408         /* step 1: recover xattr */
 409         if (IS_INODE(page)) {
 410                 recover_inline_xattr(inode, page);
 411         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 412                 /*
 413                  * Deprecated; xattr blocks should be found from cold log.
 414                  * But, we should remain this for backward compatibility.
 415                  */
 416                 recover_xattr_data(inode, page, blkaddr);
 417                 goto out;
 418         }
 419
 420         /* step 2: recover inline data */
 421         if (recover_inline_data(inode, page))
 422                 goto out;
 423
 424         /* step 3: recover data indices */
 425         start = start_bidx_of_node(ofs_of_node(page), inode);
 426         end = start + ADDRS_PER_PAGE(page, inode);
 427
 428         set_new_dnode(&dn, inode, NULL, NULL, 0);
 429
 430         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 431         if (err)
 432                 goto out;
 433
 434         f2fs_wait_on_page_writeback(dn.node_page, NODE, true);
 435
 436         get_node_info(sbi, dn.nid, &ni);
 437         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 438         f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
 439
 440         for (; start < end; start++, dn.ofs_in_node++) {
 441                 block_t src, dest;
 442
 443                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 444                 dest = datablock_addr(page, dn.ofs_in_node);
 445
 446                 /* skip recovering if dest is the same as src */
 447                 if (src == dest)
 448                         continue;
 449
 450                 /* dest is invalid, just invalidate src block */
 451                 if (dest == NULL_ADDR) {
 452                         truncate_data_blocks_range(&dn, 1);
 453                         continue;
 454                 }
 455
 456                 /*
 457                  * dest is reserved block, invalidate src block
 458                  * and then reserve one new block in dnode page.
 459                  */
 460                 if (dest == NEW_ADDR) {
 461                         truncate_data_blocks_range(&dn, 1);
 462                         err = reserve_new_block(&dn);
 463                         f2fs_bug_on(sbi, err);
 464                         continue;
 465                 }
 466
 467                 /* dest is valid block, try to recover from src to dest */
 468                 if (is_valid_blkaddr(sbi, dest, META_POR)) {
 469
 470                         if (src == NULL_ADDR) {
 471                                 err = reserve_new_block(&dn);
 472                                 /* We should not get -ENOSPC */
 473                                 f2fs_bug_on(sbi, err);
 474                         }
 475
 476                         /* Check the previous node page having this index */
 477                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 478                         if (err)
 479                                 goto err;
 480
 481                         /* write dummy data page */
 482                         f2fs_replace_block(sbi, &dn, src, dest,
 483                                                 ni.version, false, false);
 484                         recovered++;
 485                 }
 486         }
 487
 488         if (IS_INODE(dn.node_page))
 489                 sync_inode_page(&dn);
 490
 491         copy_node_footer(dn.node_page, page);
 492         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 493                                         ofs_of_node(page), false);
 494         set_page_dirty(dn.node_page);
 495 err:
 496         f2fs_put_dnode(&dn);
 497 out:
 498         f2fs_msg(sbi->sb, KERN_NOTICE,
 499                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 500                 inode->i_ino, recovered, err);
 501         return err;
 502 }
 503
 504 static int recover_data(struct f2fs_sb_info *sbi, struct list_head *head)
 505 {
 506         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 507         struct curseg_info *curseg;
 508         struct page *page = NULL;
 509         int err = 0;
 510         block_t blkaddr;
 511
 512         /* get node pages in the current segment */
 513         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 514         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 515
 516         while (1) {
 517                 struct fsync_inode_entry *entry;
 518
 519                 if (!is_valid_blkaddr(sbi, blkaddr, META_POR))
 520                         break;
 521
 522                 ra_meta_pages_cond(sbi, blkaddr);
 523
 524                 page = get_tmp_page(sbi, blkaddr);
 525
 526                 if (cp_ver != cpver_of_node(page)) {
 527                         f2fs_put_page(page, 1);
 528                         break;
 529                 }
 530
 531                 entry = get_fsync_inode(head, ino_of_node(page));
 532                 if (!entry)
 533                         goto next;
 534                 /*
 535                  * inode(x) | CP | inode(x) | dnode(F)
 536                  * In this case, we can lose the latest inode(x).
 537                  * So, call recover_inode for the inode update.
 538                  */
 539                 if (IS_INODE(page))
 540                         recover_inode(entry->inode, page);
 541                 if (entry->last_dentry == blkaddr) {
 542                         err = recover_dentry(entry->inode, page);
 543                         if (err) {
 544                                 f2fs_put_page(page, 1);
 545                                 break;
 546                         }
 547                 }
 548                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 549                 if (err) {
 550                         f2fs_put_page(page, 1);
 551                         break;
 552                 }
 553
 554                 if (entry->blkaddr == blkaddr)
 555                         del_fsync_inode(entry);
 556 next:
 557                 /* check next segment */
 558                 blkaddr = next_blkaddr_of_node(page);
 559                 f2fs_put_page(page, 1);
 560         }
 561         if (!err)
 562                 allocate_new_segments(sbi);
 563         return err;
 564 }
 565
 566 int recover_fsync_data(struct f2fs_sb_info *sbi, bool check_only)
 567 {
 568         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 569         struct list_head inode_list;
 570         block_t blkaddr;
 571         int err;
 572         int ret = 0;
 573         bool need_writecp = false;
 574
 575         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 576                         sizeof(struct fsync_inode_entry));
 577         if (!fsync_entry_slab)
 578                 return -ENOMEM;
 579
 580         INIT_LIST_HEAD(&inode_list);
 581
 582         /* prevent checkpoint */
 583         mutex_lock(&sbi->cp_mutex);
 584
 585         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 586
 587         /* step #1: find fsynced inode numbers */
 588         err = find_fsync_dnodes(sbi, &inode_list);
 589         if (err || list_empty(&inode_list))
 590                 goto out;
 591
 592         if (check_only) {
 593                 ret = 1;
 594                 goto out;
 595         }
 596
 597         need_writecp = true;
 598
 599         /* step #2: recover data */
 600         err = recover_data(sbi, &inode_list);
 601         if (!err)
 602                 f2fs_bug_on(sbi, !list_empty(&inode_list));
 603 out:
 604         destroy_fsync_dnodes(&inode_list);
 605         kmem_cache_destroy(fsync_entry_slab);
 606
 607         /* truncate meta pages to be used by the recovery */
 608         truncate_inode_pages_range(META_MAPPING(sbi),
 609                         (loff_t)MAIN_BLKADDR(sbi) << PAGE_SHIFT, -1);
 610
 611         if (err) {
 612                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 613                 truncate_inode_pages_final(META_MAPPING(sbi));
 614         }
 615
 616         clear_sbi_flag(sbi, SBI_POR_DOING);
 617         if (err) {
 618                 bool invalidate = false;
 619
 620                 if (discard_next_dnode(sbi, blkaddr))
 621                         invalidate = true;
 622
 623                 /* Flush all the NAT/SIT pages */
 624                 while (get_pages(sbi, F2FS_DIRTY_META))
 625                         sync_meta_pages(sbi, META, LONG_MAX);
 626
 627                 /* invalidate temporary meta page */
 628                 if (invalidate)
 629                         invalidate_mapping_pages(META_MAPPING(sbi),
 630                                                         blkaddr, blkaddr);
 631
 632                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 633                 mutex_unlock(&sbi->cp_mutex);
 634         } else if (need_writecp) {
 635                 struct cp_control cpc = {
 636                         .reason = CP_RECOVERY,
 637                 };
 638                 mutex_unlock(&sbi->cp_mutex);
 639                 err = write_checkpoint(sbi, &cpc);
 640         } else {
 641                 mutex_unlock(&sbi->cp_mutex);
 642         }
 643         return ret ? ret: err;
 644 }