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