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 static struct kmem_cache *fsync_entry_slab;
  18
  19 bool space_for_roll_forward(struct f2fs_sb_info *sbi)
  20 {
  21         if (sbi->last_valid_block_count + sbi->alloc_valid_block_count
  22                         > sbi->user_block_count)
  23                 return false;
  24         return true;
  25 }
  26
  27 static struct fsync_inode_entry *get_fsync_inode(struct list_head *head,
  28                                                                 nid_t ino)
  29 {
  30         struct fsync_inode_entry *entry;
  31
  32         list_for_each_entry(entry, head, list)
  33                 if (entry->inode->i_ino == ino)
  34                         return entry;
  35
  36         return NULL;
  37 }
  38
  39 static int recover_dentry(struct page *ipage, struct inode *inode)
  40 {
  41         struct f2fs_inode *raw_inode = F2FS_INODE(ipage);
  42         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  43         struct f2fs_dir_entry *de;
  44         struct qstr name;
  45         struct page *page;
  46         struct inode *dir, *einode;
  47         int err = 0;
  48
  49         dir = f2fs_iget(inode->i_sb, pino);
  50         if (IS_ERR(dir)) {
  51                 err = PTR_ERR(dir);
  52                 goto out;
  53         }
  54
  55         name.len = le32_to_cpu(raw_inode->i_namelen);
  56         name.name = raw_inode->i_name;
  57
  58         if (unlikely(name.len > F2FS_NAME_LEN)) {
  59                 WARN_ON(1);
  60                 err = -ENAMETOOLONG;
  61                 goto out_err;
  62         }
  63 retry:
  64         de = f2fs_find_entry(dir, &name, &page);
  65         if (de && inode->i_ino == le32_to_cpu(de->ino)) {
  66                 clear_inode_flag(F2FS_I(inode), FI_INC_LINK);
  67                 goto out_unmap_put;
  68         }
  69         if (de) {
  70                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
  71                 if (IS_ERR(einode)) {
  72                         WARN_ON(1);
  73                         err = PTR_ERR(einode);
  74                         if (err == -ENOENT)
  75                                 err = -EEXIST;
  76                         goto out_unmap_put;
  77                 }
  78                 err = acquire_orphan_inode(F2FS_I_SB(inode));
  79                 if (err) {
  80                         iput(einode);
  81                         goto out_unmap_put;
  82                 }
  83                 f2fs_delete_entry(de, page, einode);
  84                 iput(einode);
  85                 goto retry;
  86         }
  87         err = __f2fs_add_link(dir, &name, inode);
  88         if (err)
  89                 goto out_err;
  90
  91         if (is_inode_flag_set(F2FS_I(dir), FI_DELAY_IPUT)) {
  92                 iput(dir);
  93         } else {
  94                 add_dirty_dir_inode(dir);
  95                 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
  96         }
  97
  98         goto out;
  99
 100 out_unmap_put:
 101         kunmap(page);
 102         f2fs_put_page(page, 0);
 103 out_err:
 104         iput(dir);
 105 out:
 106         f2fs_msg(inode->i_sb, KERN_NOTICE,
 107                         "%s: ino = %x, name = %s, dir = %lx, err = %d",
 108                         __func__, ino_of_node(ipage), raw_inode->i_name,
 109                         IS_ERR(dir) ? 0 : dir->i_ino, err);
 110         return err;
 111 }
 112
 113 static int recover_inode(struct inode *inode, struct page *node_page)
 114 {
 115         struct f2fs_inode *raw_inode = F2FS_INODE(node_page);
 116
 117         if (!IS_INODE(node_page))
 118                 return 0;
 119
 120         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 121         i_size_write(inode, le64_to_cpu(raw_inode->i_size));
 122         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 123         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 124         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 125         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 126         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 127         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 128
 129         if (is_dent_dnode(node_page))
 130                 return recover_dentry(node_page, inode);
 131
 132         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 133                         ino_of_node(node_page), raw_inode->i_name);
 134         return 0;
 135 }
 136
 137 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 138 {
 139         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 140         struct curseg_info *curseg;
 141         struct page *page;
 142         block_t blkaddr;
 143         int err = 0;
 144
 145         /* get node pages in the current segment */
 146         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 147         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 148
 149         /* read node page */
 150         page = alloc_page(GFP_F2FS_ZERO);
 151         if (!page)
 152                 return -ENOMEM;
 153         lock_page(page);
 154
 155         while (1) {
 156                 struct fsync_inode_entry *entry;
 157
 158                 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
 159                 if (err)
 160                         return err;
 161
 162                 lock_page(page);
 163
 164                 if (cp_ver != cpver_of_node(page))
 165                         break;
 166
 167                 if (!is_fsync_dnode(page))
 168                         goto next;
 169
 170                 entry = get_fsync_inode(head, ino_of_node(page));
 171                 if (entry) {
 172                         if (IS_INODE(page) && is_dent_dnode(page))
 173                                 set_inode_flag(F2FS_I(entry->inode),
 174                                                         FI_INC_LINK);
 175                 } else {
 176                         if (IS_INODE(page) && is_dent_dnode(page)) {
 177                                 err = recover_inode_page(sbi, page);
 178                                 if (err)
 179                                         break;
 180                         }
 181
 182                         /* add this fsync inode to the list */
 183                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 184                         if (!entry) {
 185                                 err = -ENOMEM;
 186                                 break;
 187                         }
 188
 189                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 190                         if (IS_ERR(entry->inode)) {
 191                                 err = PTR_ERR(entry->inode);
 192                                 kmem_cache_free(fsync_entry_slab, entry);
 193                                 break;
 194                         }
 195                         list_add_tail(&entry->list, head);
 196                 }
 197                 entry->blkaddr = blkaddr;
 198
 199                 err = recover_inode(entry->inode, page);
 200                 if (err && err != -ENOENT)
 201                         break;
 202 next:
 203                 /* check next segment */
 204                 blkaddr = next_blkaddr_of_node(page);
 205         }
 206
 207         unlock_page(page);
 208         __free_pages(page, 0);
 209
 210         return err;
 211 }
 212
 213 static void destroy_fsync_dnodes(struct list_head *head)
 214 {
 215         struct fsync_inode_entry *entry, *tmp;
 216
 217         list_for_each_entry_safe(entry, tmp, head, list) {
 218                 iput(entry->inode);
 219                 list_del(&entry->list);
 220                 kmem_cache_free(fsync_entry_slab, entry);
 221         }
 222 }
 223
 224 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 225                         block_t blkaddr, struct dnode_of_data *dn)
 226 {
 227         struct seg_entry *sentry;
 228         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 229         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 230         struct f2fs_summary_block *sum_node;
 231         struct f2fs_summary sum;
 232         struct page *sum_page, *node_page;
 233         nid_t ino, nid;
 234         struct inode *inode;
 235         unsigned int offset;
 236         block_t bidx;
 237         int i;
 238
 239         sentry = get_seg_entry(sbi, segno);
 240         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 241                 return 0;
 242
 243         /* Get the previous summary */
 244         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 245                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 246                 if (curseg->segno == segno) {
 247                         sum = curseg->sum_blk->entries[blkoff];
 248                         goto got_it;
 249                 }
 250         }
 251
 252         sum_page = get_sum_page(sbi, segno);
 253         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 254         sum = sum_node->entries[blkoff];
 255         f2fs_put_page(sum_page, 1);
 256 got_it:
 257         /* Use the locked dnode page and inode */
 258         nid = le32_to_cpu(sum.nid);
 259         if (dn->inode->i_ino == nid) {
 260                 struct dnode_of_data tdn = *dn;
 261                 tdn.nid = nid;
 262                 tdn.node_page = dn->inode_page;
 263                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 264                 truncate_data_blocks_range(&tdn, 1);
 265                 return 0;
 266         } else if (dn->nid == nid) {
 267                 struct dnode_of_data tdn = *dn;
 268                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 269                 truncate_data_blocks_range(&tdn, 1);
 270                 return 0;
 271         }
 272
 273         /* Get the node page */
 274         node_page = get_node_page(sbi, nid);
 275         if (IS_ERR(node_page))
 276                 return PTR_ERR(node_page);
 277
 278         offset = ofs_of_node(node_page);
 279         ino = ino_of_node(node_page);
 280         f2fs_put_page(node_page, 1);
 281
 282         /* Deallocate previous index in the node page */
 283         inode = f2fs_iget(sbi->sb, ino);
 284         if (IS_ERR(inode))
 285                 return PTR_ERR(inode);
 286
 287         bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
 288                                         le16_to_cpu(sum.ofs_in_node);
 289
 290         truncate_hole(inode, bidx, bidx + 1);
 291         iput(inode);
 292         return 0;
 293 }
 294
 295 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 296                                         struct page *page, block_t blkaddr)
 297 {
 298         struct f2fs_inode_info *fi = F2FS_I(inode);
 299         unsigned int start, end;
 300         struct dnode_of_data dn;
 301         struct f2fs_summary sum;
 302         struct node_info ni;
 303         int err = 0, recovered = 0;
 304
 305         /* step 1: recover xattr */
 306         if (IS_INODE(page)) {
 307                 recover_inline_xattr(inode, page);
 308         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 309                 recover_xattr_data(inode, page, blkaddr);
 310                 goto out;
 311         }
 312
 313         /* step 2: recover inline data */
 314         if (recover_inline_data(inode, page))
 315                 goto out;
 316
 317         /* step 3: recover data indices */
 318         start = start_bidx_of_node(ofs_of_node(page), fi);
 319         end = start + ADDRS_PER_PAGE(page, fi);
 320
 321         f2fs_lock_op(sbi);
 322
 323         set_new_dnode(&dn, inode, NULL, NULL, 0);
 324
 325         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 326         if (err) {
 327                 f2fs_unlock_op(sbi);
 328                 goto out;
 329         }
 330
 331         f2fs_wait_on_page_writeback(dn.node_page, NODE);
 332
 333         get_node_info(sbi, dn.nid, &ni);
 334         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 335         f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
 336
 337         for (; start < end; start++) {
 338                 block_t src, dest;
 339
 340                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 341                 dest = datablock_addr(page, dn.ofs_in_node);
 342
 343                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 344                         if (src == NULL_ADDR) {
 345                                 err = reserve_new_block(&dn);
 346                                 /* We should not get -ENOSPC */
 347                                 f2fs_bug_on(sbi, err);
 348                         }
 349
 350                         /* Check the previous node page having this index */
 351                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 352                         if (err)
 353                                 goto err;
 354
 355                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 356
 357                         /* write dummy data page */
 358                         recover_data_page(sbi, NULL, &sum, src, dest);
 359                         update_extent_cache(dest, &dn);
 360                         recovered++;
 361                 }
 362                 dn.ofs_in_node++;
 363         }
 364
 365         /* write node page in place */
 366         set_summary(&sum, dn.nid, 0, 0);
 367         if (IS_INODE(dn.node_page))
 368                 sync_inode_page(&dn);
 369
 370         copy_node_footer(dn.node_page, page);
 371         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 372                                         ofs_of_node(page), false);
 373         set_page_dirty(dn.node_page);
 374 err:
 375         f2fs_put_dnode(&dn);
 376         f2fs_unlock_op(sbi);
 377 out:
 378         f2fs_msg(sbi->sb, KERN_NOTICE,
 379                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 380                 inode->i_ino, recovered, err);
 381         return err;
 382 }
 383
 384 static int recover_data(struct f2fs_sb_info *sbi,
 385                                 struct list_head *head, int type)
 386 {
 387         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 388         struct curseg_info *curseg;
 389         struct page *page;
 390         int err = 0;
 391         block_t blkaddr;
 392
 393         /* get node pages in the current segment */
 394         curseg = CURSEG_I(sbi, type);
 395         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 396
 397         /* read node page */
 398         page = alloc_page(GFP_F2FS_ZERO);
 399         if (!page)
 400                 return -ENOMEM;
 401
 402         lock_page(page);
 403
 404         while (1) {
 405                 struct fsync_inode_entry *entry;
 406
 407                 err = f2fs_submit_page_bio(sbi, page, blkaddr, READ_SYNC);
 408                 if (err)
 409                         return err;
 410
 411                 lock_page(page);
 412
 413                 if (cp_ver != cpver_of_node(page))
 414                         break;
 415
 416                 entry = get_fsync_inode(head, ino_of_node(page));
 417                 if (!entry)
 418                         goto next;
 419
 420                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 421                 if (err)
 422                         break;
 423
 424                 if (entry->blkaddr == blkaddr) {
 425                         iput(entry->inode);
 426                         list_del(&entry->list);
 427                         kmem_cache_free(fsync_entry_slab, entry);
 428                 }
 429 next:
 430                 /* check next segment */
 431                 blkaddr = next_blkaddr_of_node(page);
 432         }
 433
 434         unlock_page(page);
 435         __free_pages(page, 0);
 436
 437         if (!err)
 438                 allocate_new_segments(sbi);
 439         return err;
 440 }
 441
 442 int recover_fsync_data(struct f2fs_sb_info *sbi)
 443 {
 444         struct curseg_info *curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 445         struct list_head inode_list;
 446         block_t blkaddr;
 447         int err;
 448         bool need_writecp = false;
 449
 450         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 451                         sizeof(struct fsync_inode_entry));
 452         if (!fsync_entry_slab)
 453                 return -ENOMEM;
 454
 455         INIT_LIST_HEAD(&inode_list);
 456
 457         /* step #1: find fsynced inode numbers */
 458         sbi->por_doing = true;
 459
 460         /* prevent checkpoint */
 461         mutex_lock(&sbi->cp_mutex);
 462
 463         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 464
 465         err = find_fsync_dnodes(sbi, &inode_list);
 466         if (err)
 467                 goto out;
 468
 469         if (list_empty(&inode_list))
 470                 goto out;
 471
 472         need_writecp = true;
 473
 474         /* step #2: recover data */
 475         err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 476         if (!err)
 477                 f2fs_bug_on(sbi, !list_empty(&inode_list));
 478 out:
 479         destroy_fsync_dnodes(&inode_list);
 480         kmem_cache_destroy(fsync_entry_slab);
 481
 482         if (err) {
 483                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 484                 truncate_inode_pages_final(META_MAPPING(sbi));
 485         }
 486
 487         sbi->por_doing = false;
 488         if (err) {
 489                 discard_next_dnode(sbi, blkaddr);
 490
 491                 /* Flush all the NAT/SIT pages */
 492                 while (get_pages(sbi, F2FS_DIRTY_META))
 493                         sync_meta_pages(sbi, META, LONG_MAX);
 494                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 495                 mutex_unlock(&sbi->cp_mutex);
 496         } else if (need_writecp) {
 497                 mutex_unlock(&sbi->cp_mutex);
 498                 write_checkpoint(sbi, false);
 499         } else {
 500                 mutex_unlock(&sbi->cp_mutex);
 501         }
 502         return err;
 503 }