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