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 list_head *this;
  31         struct fsync_inode_entry *entry;
  32
  33         list_for_each(this, head) {
  34                 entry = list_entry(this, struct fsync_inode_entry, list);
  35                 if (entry->inode->i_ino == ino)
  36                         return entry;
  37         }
  38         return NULL;
  39 }
  40
  41 static int recover_dentry(struct page *ipage, struct inode *inode)
  42 {
  43         void *kaddr = page_address(ipage);
  44         struct f2fs_node *raw_node = (struct f2fs_node *)kaddr;
  45         struct f2fs_inode *raw_inode = &(raw_node->i);
  46         nid_t pino = le32_to_cpu(raw_inode->i_pino);
  47         struct f2fs_dir_entry *de;
  48         struct qstr name;
  49         struct page *page;
  50         struct inode *dir, *einode;
  51         int err = 0;
  52
  53         dir = check_dirty_dir_inode(F2FS_SB(inode->i_sb), pino);
  54         if (!dir) {
  55                 dir = f2fs_iget(inode->i_sb, pino);
  56                 if (IS_ERR(dir)) {
  57                         err = PTR_ERR(dir);
  58                         goto out;
  59                 }
  60                 set_inode_flag(F2FS_I(dir), FI_DELAY_IPUT);
  61                 add_dirty_dir_inode(dir);
  62         }
  63
  64         name.len = le32_to_cpu(raw_inode->i_namelen);
  65         name.name = raw_inode->i_name;
  66 retry:
  67         de = f2fs_find_entry(dir, &name, &page);
  68         if (de && inode->i_ino == le32_to_cpu(de->ino)) {
  69                 kunmap(page);
  70                 f2fs_put_page(page, 0);
  71                 goto out;
  72         }
  73         if (de) {
  74                 einode = f2fs_iget(inode->i_sb, le32_to_cpu(de->ino));
  75                 if (IS_ERR(einode)) {
  76                         WARN_ON(1);
  77                         if (PTR_ERR(einode) == -ENOENT)
  78                                 err = -EEXIST;
  79                         goto out;
  80                 }
  81                 f2fs_delete_entry(de, page, einode);
  82                 iput(einode);
  83                 goto retry;
  84         }
  85         err = __f2fs_add_link(dir, &name, inode);
  86 out:
  87         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode and its dentry: "
  88                         "ino = %x, name = %s, dir = %lx, err = %d",
  89                         ino_of_node(ipage), raw_inode->i_name,
  90                         IS_ERR(dir) ? 0 : dir->i_ino, err);
  91         return err;
  92 }
  93
  94 static int recover_inode(struct inode *inode, struct page *node_page)
  95 {
  96         void *kaddr = page_address(node_page);
  97         struct f2fs_node *raw_node = (struct f2fs_node *)kaddr;
  98         struct f2fs_inode *raw_inode = &(raw_node->i);
  99
 100         if (!IS_INODE(node_page))
 101                 return 0;
 102
 103         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
 104         i_size_write(inode, le64_to_cpu(raw_inode->i_size));
 105         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 106         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
 107         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
 108         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 109         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
 110         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
 111
 112         if (is_dent_dnode(node_page))
 113                 return recover_dentry(node_page, inode);
 114
 115         f2fs_msg(inode->i_sb, KERN_NOTICE, "recover_inode: ino = %x, name = %s",
 116                         ino_of_node(node_page), raw_inode->i_name);
 117         return 0;
 118 }
 119
 120 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
 121 {
 122         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
 123         struct curseg_info *curseg;
 124         struct page *page;
 125         block_t blkaddr;
 126         int err = 0;
 127
 128         /* get node pages in the current segment */
 129         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 130         blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff;
 131
 132         /* read node page */
 133         page = alloc_page(GFP_F2FS_ZERO);
 134         if (IS_ERR(page))
 135                 return PTR_ERR(page);
 136         lock_page(page);
 137
 138         while (1) {
 139                 struct fsync_inode_entry *entry;
 140
 141                 err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
 142                 if (err)
 143                         goto out;
 144
 145                 lock_page(page);
 146
 147                 if (cp_ver != cpver_of_node(page))
 148                         break;
 149
 150                 if (!is_fsync_dnode(page))
 151                         goto next;
 152
 153                 entry = get_fsync_inode(head, ino_of_node(page));
 154                 if (entry) {
 155                         if (IS_INODE(page) && is_dent_dnode(page))
 156                                 set_inode_flag(F2FS_I(entry->inode),
 157                                                         FI_INC_LINK);
 158                 } else {
 159                         if (IS_INODE(page) && is_dent_dnode(page)) {
 160                                 err = recover_inode_page(sbi, page);
 161                                 if (err)
 162                                         break;
 163                         }
 164
 165                         /* add this fsync inode to the list */
 166                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 167                         if (!entry) {
 168                                 err = -ENOMEM;
 169                                 break;
 170                         }
 171
 172                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 173                         if (IS_ERR(entry->inode)) {
 174                                 err = PTR_ERR(entry->inode);
 175                                 kmem_cache_free(fsync_entry_slab, entry);
 176                                 break;
 177                         }
 178                         list_add_tail(&entry->list, head);
 179                 }
 180                 entry->blkaddr = blkaddr;
 181
 182                 err = recover_inode(entry->inode, page);
 183                 if (err && err != -ENOENT)
 184                         break;
 185 next:
 186                 /* check next segment */
 187                 blkaddr = next_blkaddr_of_node(page);
 188         }
 189         unlock_page(page);
 190 out:
 191         __free_pages(page, 0);
 192         return err;
 193 }
 194
 195 static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi,
 196                                         struct list_head *head)
 197 {
 198         struct fsync_inode_entry *entry, *tmp;
 199
 200         list_for_each_entry_safe(entry, tmp, head, list) {
 201                 iput(entry->inode);
 202                 list_del(&entry->list);
 203                 kmem_cache_free(fsync_entry_slab, entry);
 204         }
 205 }
 206
 207 static int check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 208                         block_t blkaddr, struct dnode_of_data *dn)
 209 {
 210         struct seg_entry *sentry;
 211         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 212         unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
 213                                         (sbi->blocks_per_seg - 1);
 214         struct f2fs_summary sum;
 215         nid_t ino, nid;
 216         void *kaddr;
 217         struct inode *inode;
 218         struct page *node_page;
 219         block_t bidx;
 220         int i;
 221
 222         sentry = get_seg_entry(sbi, segno);
 223         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 224                 return 0;
 225
 226         /* Get the previous summary */
 227         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 228                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 229                 if (curseg->segno == segno) {
 230                         sum = curseg->sum_blk->entries[blkoff];
 231                         break;
 232                 }
 233         }
 234         if (i > CURSEG_COLD_DATA) {
 235                 struct page *sum_page = get_sum_page(sbi, segno);
 236                 struct f2fs_summary_block *sum_node;
 237                 kaddr = page_address(sum_page);
 238                 sum_node = (struct f2fs_summary_block *)kaddr;
 239                 sum = sum_node->entries[blkoff];
 240                 f2fs_put_page(sum_page, 1);
 241         }
 242
 243         /* Use the locked dnode page and inode */
 244         nid = le32_to_cpu(sum.nid);
 245         if (dn->inode->i_ino == nid) {
 246                 struct dnode_of_data tdn = *dn;
 247                 tdn.nid = nid;
 248                 tdn.node_page = dn->inode_page;
 249                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 250                 truncate_data_blocks_range(&tdn, 1);
 251                 return 0;
 252         } else if (dn->nid == nid) {
 253                 struct dnode_of_data tdn = *dn;
 254                 tdn.ofs_in_node = le16_to_cpu(sum.ofs_in_node);
 255                 truncate_data_blocks_range(&tdn, 1);
 256                 return 0;
 257         }
 258
 259         /* Get the node page */
 260         node_page = get_node_page(sbi, nid);
 261         if (IS_ERR(node_page))
 262                 return PTR_ERR(node_page);
 263         bidx = start_bidx_of_node(ofs_of_node(node_page)) +
 264                                         le16_to_cpu(sum.ofs_in_node);
 265         ino = ino_of_node(node_page);
 266         f2fs_put_page(node_page, 1);
 267
 268         /* Deallocate previous index in the node page */
 269         inode = f2fs_iget(sbi->sb, ino);
 270         if (IS_ERR(inode))
 271                 return PTR_ERR(inode);
 272
 273         truncate_hole(inode, bidx, bidx + 1);
 274         iput(inode);
 275         return 0;
 276 }
 277
 278 static int do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 279                                         struct page *page, block_t blkaddr)
 280 {
 281         unsigned int start, end;
 282         struct dnode_of_data dn;
 283         struct f2fs_summary sum;
 284         struct node_info ni;
 285         int err = 0, recovered = 0;
 286         int ilock;
 287
 288         start = start_bidx_of_node(ofs_of_node(page));
 289         if (IS_INODE(page))
 290                 end = start + ADDRS_PER_INODE;
 291         else
 292                 end = start + ADDRS_PER_BLOCK;
 293
 294         ilock = mutex_lock_op(sbi);
 295         set_new_dnode(&dn, inode, NULL, NULL, 0);
 296
 297         err = get_dnode_of_data(&dn, start, ALLOC_NODE);
 298         if (err) {
 299                 mutex_unlock_op(sbi, ilock);
 300                 return err;
 301         }
 302
 303         wait_on_page_writeback(dn.node_page);
 304
 305         get_node_info(sbi, dn.nid, &ni);
 306         BUG_ON(ni.ino != ino_of_node(page));
 307         BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page));
 308
 309         for (; start < end; start++) {
 310                 block_t src, dest;
 311
 312                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 313                 dest = datablock_addr(page, dn.ofs_in_node);
 314
 315                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 316                         if (src == NULL_ADDR) {
 317                                 int err = reserve_new_block(&dn);
 318                                 /* We should not get -ENOSPC */
 319                                 BUG_ON(err);
 320                         }
 321
 322                         /* Check the previous node page having this index */
 323                         err = check_index_in_prev_nodes(sbi, dest, &dn);
 324                         if (err)
 325                                 goto err;
 326
 327                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 328
 329                         /* write dummy data page */
 330                         recover_data_page(sbi, NULL, &sum, src, dest);
 331                         update_extent_cache(dest, &dn);
 332                         recovered++;
 333                 }
 334                 dn.ofs_in_node++;
 335         }
 336
 337         /* write node page in place */
 338         set_summary(&sum, dn.nid, 0, 0);
 339         if (IS_INODE(dn.node_page))
 340                 sync_inode_page(&dn);
 341
 342         copy_node_footer(dn.node_page, page);
 343         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 344                                         ofs_of_node(page), false);
 345         set_page_dirty(dn.node_page);
 346
 347         recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
 348 err:
 349         f2fs_put_dnode(&dn);
 350         mutex_unlock_op(sbi, ilock);
 351
 352         f2fs_msg(sbi->sb, KERN_NOTICE, "recover_data: ino = %lx, "
 353                         "recovered_data = %d blocks, err = %d",
 354                         inode->i_ino, recovered, err);
 355         return err;
 356 }
 357
 358 static int recover_data(struct f2fs_sb_info *sbi,
 359                                 struct list_head *head, int type)
 360 {
 361         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
 362         struct curseg_info *curseg;
 363         struct page *page;
 364         int err = 0;
 365         block_t blkaddr;
 366
 367         /* get node pages in the current segment */
 368         curseg = CURSEG_I(sbi, type);
 369         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 370
 371         /* read node page */
 372         page = alloc_page(GFP_NOFS | __GFP_ZERO);
 373         if (IS_ERR(page))
 374                 return -ENOMEM;
 375
 376         lock_page(page);
 377
 378         while (1) {
 379                 struct fsync_inode_entry *entry;
 380
 381                 err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
 382                 if (err)
 383                         goto out;
 384
 385                 lock_page(page);
 386
 387                 if (cp_ver != cpver_of_node(page))
 388                         break;
 389
 390                 entry = get_fsync_inode(head, ino_of_node(page));
 391                 if (!entry)
 392                         goto next;
 393
 394                 err = do_recover_data(sbi, entry->inode, page, blkaddr);
 395                 if (err)
 396                         break;
 397
 398                 if (entry->blkaddr == blkaddr) {
 399                         iput(entry->inode);
 400                         list_del(&entry->list);
 401                         kmem_cache_free(fsync_entry_slab, entry);
 402                 }
 403 next:
 404                 /* check next segment */
 405                 blkaddr = next_blkaddr_of_node(page);
 406         }
 407         unlock_page(page);
 408 out:
 409         __free_pages(page, 0);
 410
 411         if (!err)
 412                 allocate_new_segments(sbi);
 413         return err;
 414 }
 415
 416 int recover_fsync_data(struct f2fs_sb_info *sbi)
 417 {
 418         struct list_head inode_list;
 419         int err;
 420
 421         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 422                         sizeof(struct fsync_inode_entry), NULL);
 423         if (unlikely(!fsync_entry_slab))
 424                 return -ENOMEM;
 425
 426         INIT_LIST_HEAD(&inode_list);
 427
 428         /* step #1: find fsynced inode numbers */
 429         sbi->por_doing = 1;
 430         err = find_fsync_dnodes(sbi, &inode_list);
 431         if (err)
 432                 goto out;
 433
 434         if (list_empty(&inode_list))
 435                 goto out;
 436
 437         /* step #2: recover data */
 438         err = recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 439         BUG_ON(!list_empty(&inode_list));
 440 out:
 441         destroy_fsync_dnodes(sbi, &inode_list);
 442         kmem_cache_destroy(fsync_entry_slab);
 443         sbi->por_doing = 0;
 444         if (!err)
 445                 write_checkpoint(sbi, false);
 446         return err;
 447 }