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 = NULL;
 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         while (1) {
 150                 struct fsync_inode_entry *entry;
 151
 152                 if (blkaddr < SM_I(sbi)->main_blkaddr ||
 153                         blkaddr >= (SM_I(sbi)->seg0_blkaddr + TOTAL_BLKS(sbi)))
 154                         return 0;
 155
 156                 page = get_meta_page_ra(sbi, blkaddr);
 157
 158                 if (cp_ver != cpver_of_node(page))
 159                         break;
 160
 161                 if (!is_fsync_dnode(page))
 162                         goto next;
 163
 164                 entry = get_fsync_inode(head, ino_of_node(page));
 165                 if (entry) {
 166                         if (IS_INODE(page) && is_dent_dnode(page))
 167                                 set_inode_flag(F2FS_I(entry->inode),
 168                                                         FI_INC_LINK);
 169                 } else {
 170                         if (IS_INODE(page) && is_dent_dnode(page)) {
 171                                 err = recover_inode_page(sbi, page);
 172                                 if (err)
 173                                         break;
 174                         }
 175
 176                         /* add this fsync inode to the list */
 177                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 178                         if (!entry) {
 179                                 err = -ENOMEM;
 180                                 break;
 181                         }
 182
 183                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 184                         if (IS_ERR(entry->inode)) {
 185                                 err = PTR_ERR(entry->inode);
 186                                 kmem_cache_free(fsync_entry_slab, entry);
 187                                 break;
 188                         }
 189                         list_add_tail(&entry->list, head);
 190                 }
 191                 entry->blkaddr = blkaddr;
 192
 193                 err = recover_inode(entry->inode, page);
 194                 if (err && err != -ENOENT)
 195                         break;
 196 next:
 197                 /* check next segment */
 198                 blkaddr = next_blkaddr_of_node(page);
 199                 f2fs_put_page(page, 1);
 200         }
 201         f2fs_put_page(page, 1);
 202         return err;
 203 }
 204
 205 static void destroy_fsync_dnodes(struct list_head *head)
 206 {
 207         struct fsync_inode_entry *entry, *tmp;
 208
 209         list_for_each_entry_safe(entry, tmp, head, list) {
 210                 iput(entry->inode);
 211                 list_del(&entry->list);
 212                 kmem_cache_free(fsync_entry_slab, entry);
 213         }
 214 }
 215
 216 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 217                         block_t blkaddr, struct dnode_of_data *dn)
 218 {
 219         struct seg_entry *sentry;
 220         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 221         unsigned short blkoff = GET_BLKOFF_FROM_SEG0(sbi, blkaddr);
 222         struct f2fs_summary_block *sum_node;
 223         struct f2fs_summary sum;
 224         struct page *sum_page, *node_page;
 225         nid_t ino, nid;
 226         struct inode *inode;
 227         unsigned int offset;
 228         block_t bidx;
 229         int i;
 230
 231         sentry = get_seg_entry(sbi, segno);
 232         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 233                 return 0;
 234
 235         /* Get the previous summary */
 236         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 237                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 238                 if (curseg->segno == segno) {
 239                         sum = curseg->sum_blk->entries[blkoff];
 240                         goto got_it;
 241                 }
 242         }
 243
 244         sum_page = get_sum_page(sbi, segno);
 245         sum_node = (struct f2fs_summary_block *)page_address(sum_page);
 246         sum = sum_node->entries[blkoff];
 247         f2fs_put_page(sum_page, 1);
 248 got_it:
 249         /* Use the locked dnode page and inode */
 250         nid = le32_to_cpu(sum.nid);
 251         if (dn->inode->i_ino == nid) {
 252                 struct dnode_of_data tdn = *dn;
 253                 tdn.nid = nid;
 254                 tdn.node_page = dn->inode_page;
 255                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 256                 truncate_data_blocks_range(&tdn, 1);
 257                 return 0;
 258         } else if (dn->nid == nid) {
 259                 struct dnode_of_data tdn = *dn;
 260                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 261                 truncate_data_blocks_range(&tdn, 1);
 262                 return 0;
 263         }
 264
 265         /* Get the node page */
 266         node_page = get_node_page(sbi, nid);
 267         if (IS_ERR(node_page))
 268                 return PTR_ERR(node_page);
 269
 270         offset = ofs_of_node(node_page);
 271         ino = ino_of_node(node_page);
 272         f2fs_put_page(node_page, 1);
 273
 274         if (ino != dn->inode->i_ino) {
 275                 /* Deallocate previous index in the node page */
 276                 inode = f2fs_iget(sbi->sb, ino);
 277                 if (IS_ERR(inode))
 278                         return PTR_ERR(inode);
 279         } else {
 280                 inode = dn->inode;
 281         }
 282
 283         bidx = start_bidx_of_node(offset, F2FS_I(inode)) +
 284                         le16_to_cpu(sum.ofs_in_node);
 285
 286         if (ino != dn->inode->i_ino) {
 287                 truncate_hole(inode, bidx, bidx + 1);
 288                 iput(inode);
 289         } else {
 290                 struct dnode_of_data tdn;
 291                 set_new_dnode(&tdn, inode, dn->inode_page, NULL, 0);
 292                 if (get_dnode_of_data(&tdn, bidx, LOOKUP_NODE))
 293                         return 0;
 294                 if (tdn.data_blkaddr != NULL_ADDR)
 295                         truncate_data_blocks_range(&tdn, 1);
 296                 f2fs_put_page(tdn.node_page, 1);
 297         }
 298         return 0;
 299 }
 300
 301 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 302                                         struct page *page, block_t blkaddr)
 303 {
 304         struct f2fs_inode_info *fi = F2FS_I(inode);
 305         unsigned int start, end;
 306         struct dnode_of_data dn;
 307         struct f2fs_summary sum;
 308         struct node_info ni;
 309         int err = 0, recovered = 0;
 310
 311         /* step 1: recover xattr */
 312         if (IS_INODE(page)) {
 313                 recover_inline_xattr(inode, page);
 314         } else if (f2fs_has_xattr_block(ofs_of_node(page))) {
 315                 recover_xattr_data(inode, page, blkaddr);
 316                 goto out;
 317         }
 318
 319         /* step 2: recover inline data */
 320         if (recover_inline_data(inode, page))
 321                 goto out;
 322
 323         /* step 3: recover data indices */
 324         start = start_bidx_of_node(ofs_of_node(page), fi);
 325         end = start + ADDRS_PER_PAGE(page, fi);
 326
 327         f2fs_lock_op(sbi);
 328
 329         set_new_dnode(&dn, inode, NULL, NULL, 0);
 330
 331         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 332         if (err) {
 333                 f2fs_unlock_op(sbi);
 334                 goto out;
 335         }
 336
 337         f2fs_wait_on_page_writeback(dn.node_page, NODE);
 338
 339         get_node_info(sbi, dn.nid, &ni);
 340         f2fs_bug_on(sbi, ni.ino != ino_of_node(page));
 341         f2fs_bug_on(sbi, ofs_of_node(dn.node_page) != ofs_of_node(page));
 342
 343         for (; start < end; start++) {
 344                 block_t src, dest;
 345
 346                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 347                 dest = datablock_addr(page, dn.ofs_in_node);
 348
 349                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 350                         if (src == NULL_ADDR) {
 351                                 err = reserve_new_block(&dn);
 352                                 /* We should not get -ENOSPC */
 353                                 f2fs_bug_on(sbi, err);
 354                         }
 355
 356                         /* Check the previous node page having this index */
 357                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 358                         if (err)
 359                                 goto err;
 360
 361                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 362
 363                         /* write dummy data page */
 364                         recover_data_page(sbi, NULL, &sum, src, dest);
 365                         update_extent_cache(dest, &dn);
 366                         recovered++;
 367                 }
 368                 dn.ofs_in_node++;
 369         }
 370
 371         /* write node page in place */
 372         set_summary(&sum, dn.nid, 0, 0);
 373         if (IS_INODE(dn.node_page))
 374                 sync_inode_page(&dn);
 375
 376         copy_node_footer(dn.node_page, page);
 377         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 378                                         ofs_of_node(page), false);
 379         set_page_dirty(dn.node_page);
 380 err:
 381         f2fs_put_dnode(&dn);
 382         f2fs_unlock_op(sbi);
 383 out:
 384         f2fs_msg(sbi->sb, KERN_NOTICE,
 385                 "recover_data: ino = %lx, recovered = %d blocks, err = %d",
 386                 inode->i_ino, recovered, err);
 387         return err;
 388 }
 389
 390 static int recover_data(struct f2fs_sb_info *sbi,
 391                                 struct list_head *head, int type)
 392 {
 393         unsigned long long cp_ver = cur_cp_version(F2FS_CKPT(sbi));
 394         struct curseg_info *curseg;
 395         struct page *page = NULL;
 396         int err = 0;
 397         block_t blkaddr;
 398
 399         /* get node pages in the current segment */
 400         curseg = CURSEG_I(sbi, type);
 401         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 402
 403         while (1) {
 404                 struct fsync_inode_entry *entry;
 405
 406                 if (blkaddr < SM_I(sbi)->main_blkaddr ||
 407                         blkaddr >= (SM_I(sbi)->seg0_blkaddr + TOTAL_BLKS(sbi)))
 408                         break;
 409
 410                 page = get_meta_page_ra(sbi, blkaddr);
 411
 412                 if (cp_ver != cpver_of_node(page)) {
 413                         f2fs_put_page(page, 1);
 414                         break;
 415                 }
 416
 417                 entry = get_fsync_inode(head, ino_of_node(page));
 418                 if (!entry)
 419                         goto next;
 420
 421                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 422                 if (err) {
 423                         f2fs_put_page(page, 1);
 424                         break;
 425                 }
 426
 427                 if (entry->blkaddr == blkaddr) {
 428                         iput(entry->inode);
 429                         list_del(&entry->list);
 430                         kmem_cache_free(fsync_entry_slab, entry);
 431                 }
 432 next:
 433                 /* check next segment */
 434                 blkaddr = next_blkaddr_of_node(page);
 435                 f2fs_put_page(page, 1);
 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         /* truncate meta pages to be used by the recovery */
 483         truncate_inode_pages_range(META_MAPPING(sbi),
 484                 SM_I(sbi)->main_blkaddr << PAGE_CACHE_SHIFT, -1);
 485
 486         if (err) {
 487                 truncate_inode_pages_final(NODE_MAPPING(sbi));
 488                 truncate_inode_pages_final(META_MAPPING(sbi));
 489         }
 490
 491         sbi->por_doing = false;
 492         if (err) {
 493                 discard_next_dnode(sbi, blkaddr);
 494
 495                 /* Flush all the NAT/SIT pages */
 496                 while (get_pages(sbi, F2FS_DIRTY_META))
 497                         sync_meta_pages(sbi, META, LONG_MAX);
 498                 set_ckpt_flags(sbi->ckpt, CP_ERROR_FLAG);
 499                 mutex_unlock(&sbi->cp_mutex);
 500         } else if (need_writecp) {
 501                 mutex_unlock(&sbi->cp_mutex);
 502                 write_checkpoint(sbi, false);
 503         } else {
 504                 mutex_unlock(&sbi->cp_mutex);
 505         }
 506         return err;
 507 }