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         struct dentry dent, parent;
  46         struct f2fs_dir_entry *de;
  47         struct page *page;
  48         struct inode *dir;
  49         int err = 0;
  50
  51         if (!is_dent_dnode(ipage))
  52                 goto out;
  53
  54         dir = f2fs_iget(inode->i_sb, le32_to_cpu(raw_inode->i_pino));
  55         if (IS_ERR(dir)) {
  56                 err = -EINVAL;
  57                 goto out;
  58         }
  59
  60         parent.d_inode = dir;
  61         dent.d_parent = &parent;
  62         dent.d_name.len = le32_to_cpu(raw_inode->i_namelen);
  63         dent.d_name.name = raw_inode->i_name;
  64
  65         de = f2fs_find_entry(dir, &dent.d_name, &page);
  66         if (de) {
  67                 kunmap(page);
  68                 f2fs_put_page(page, 0);
  69         } else {
  70                 f2fs_add_link(&dent, inode);
  71         }
  72         iput(dir);
  73 out:
  74         kunmap(ipage);
  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         inode->i_mode = le16_to_cpu(raw_inode->i_mode);
  85         i_size_write(inode, le64_to_cpu(raw_inode->i_size));
  86         inode->i_atime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
  87         inode->i_ctime.tv_sec = le64_to_cpu(raw_inode->i_ctime);
  88         inode->i_mtime.tv_sec = le64_to_cpu(raw_inode->i_mtime);
  89         inode->i_atime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
  90         inode->i_ctime.tv_nsec = le32_to_cpu(raw_inode->i_ctime_nsec);
  91         inode->i_mtime.tv_nsec = le32_to_cpu(raw_inode->i_mtime_nsec);
  92
  93         return recover_dentry(node_page, inode);
  94 }
  95
  96 static int find_fsync_dnodes(struct f2fs_sb_info *sbi, struct list_head *head)
  97 {
  98         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
  99         struct curseg_info *curseg;
 100         struct page *page;
 101         block_t blkaddr;
 102         int err = 0;
 103
 104         /* get node pages in the current segment */
 105         curseg = CURSEG_I(sbi, CURSEG_WARM_NODE);
 106         blkaddr = START_BLOCK(sbi, curseg->segno) + curseg->next_blkoff;
 107
 108         /* read node page */
 109         page = alloc_page(GFP_F2FS_ZERO);
 110         if (IS_ERR(page))
 111                 return PTR_ERR(page);
 112         lock_page(page);
 113
 114         while (1) {
 115                 struct fsync_inode_entry *entry;
 116
 117                 if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
 118                         goto out;
 119
 120                 if (cp_ver != cpver_of_node(page))
 121                         goto out;
 122
 123                 if (!is_fsync_dnode(page))
 124                         goto next;
 125
 126                 entry = get_fsync_inode(head, ino_of_node(page));
 127                 if (entry) {
 128                         entry->blkaddr = blkaddr;
 129                         if (IS_INODE(page) && is_dent_dnode(page))
 130                                 set_inode_flag(F2FS_I(entry->inode),
 131                                                         FI_INC_LINK);
 132                 } else {
 133                         if (IS_INODE(page) && is_dent_dnode(page)) {
 134                                 if (recover_inode_page(sbi, page)) {
 135                                         err = -ENOMEM;
 136                                         goto out;
 137                                 }
 138                         }
 139
 140                         /* add this fsync inode to the list */
 141                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 142                         if (!entry) {
 143                                 err = -ENOMEM;
 144                                 goto out;
 145                         }
 146
 147                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 148                         if (IS_ERR(entry->inode)) {
 149                                 err = PTR_ERR(entry->inode);
 150                                 kmem_cache_free(fsync_entry_slab, entry);
 151                                 goto out;
 152                         }
 153
 154                         INIT_LIST_HEAD(&entry->list);
 155                         list_add_tail(&entry->list, head);
 156                         entry->blkaddr = blkaddr;
 157                 }
 158                 if (IS_INODE(page)) {
 159                         err = recover_inode(entry->inode, page);
 160                         if (err)
 161                                 goto out;
 162                 }
 163 next:
 164                 /* check next segment */
 165                 blkaddr = next_blkaddr_of_node(page);
 166                 ClearPageUptodate(page);
 167         }
 168 out:
 169         unlock_page(page);
 170         __free_pages(page, 0);
 171         return err;
 172 }
 173
 174 static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi,
 175                                         struct list_head *head)
 176 {
 177         struct list_head *this;
 178         struct fsync_inode_entry *entry;
 179         list_for_each(this, head) {
 180                 entry = list_entry(this, struct fsync_inode_entry, list);
 181                 iput(entry->inode);
 182                 list_del(&entry->list);
 183                 kmem_cache_free(fsync_entry_slab, entry);
 184         }
 185 }
 186
 187 static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 188                                                 block_t blkaddr)
 189 {
 190         struct seg_entry *sentry;
 191         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 192         unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
 193                                         (sbi->blocks_per_seg - 1);
 194         struct f2fs_summary sum;
 195         nid_t ino;
 196         void *kaddr;
 197         struct inode *inode;
 198         struct page *node_page;
 199         block_t bidx;
 200         int i;
 201
 202         sentry = get_seg_entry(sbi, segno);
 203         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 204                 return;
 205
 206         /* Get the previous summary */
 207         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 208                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 209                 if (curseg->segno == segno) {
 210                         sum = curseg->sum_blk->entries[blkoff];
 211                         break;
 212                 }
 213         }
 214         if (i > CURSEG_COLD_DATA) {
 215                 struct page *sum_page = get_sum_page(sbi, segno);
 216                 struct f2fs_summary_block *sum_node;
 217                 kaddr = page_address(sum_page);
 218                 sum_node = (struct f2fs_summary_block *)kaddr;
 219                 sum = sum_node->entries[blkoff];
 220                 f2fs_put_page(sum_page, 1);
 221         }
 222
 223         /* Get the node page */
 224         node_page = get_node_page(sbi, le32_to_cpu(sum.nid));
 225         bidx = start_bidx_of_node(ofs_of_node(node_page)) +
 226                                 le16_to_cpu(sum.ofs_in_node);
 227         ino = ino_of_node(node_page);
 228         f2fs_put_page(node_page, 1);
 229
 230         /* Deallocate previous index in the node page */
 231         inode = f2fs_iget_nowait(sbi->sb, ino);
 232         if (IS_ERR(inode))
 233                 return;
 234
 235         truncate_hole(inode, bidx, bidx + 1);
 236         iput(inode);
 237 }
 238
 239 static void do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 240                                         struct page *page, block_t blkaddr)
 241 {
 242         unsigned int start, end;
 243         struct dnode_of_data dn;
 244         struct f2fs_summary sum;
 245         struct node_info ni;
 246
 247         start = start_bidx_of_node(ofs_of_node(page));
 248         if (IS_INODE(page))
 249                 end = start + ADDRS_PER_INODE;
 250         else
 251                 end = start + ADDRS_PER_BLOCK;
 252
 253         set_new_dnode(&dn, inode, NULL, NULL, 0);
 254         if (get_dnode_of_data(&dn, start, 0))
 255                 return;
 256
 257         wait_on_page_writeback(dn.node_page);
 258
 259         get_node_info(sbi, dn.nid, &ni);
 260         BUG_ON(ni.ino != ino_of_node(page));
 261         BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page));
 262
 263         for (; start < end; start++) {
 264                 block_t src, dest;
 265
 266                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 267                 dest = datablock_addr(page, dn.ofs_in_node);
 268
 269                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 270                         if (src == NULL_ADDR) {
 271                                 int err = reserve_new_block(&dn);
 272                                 /* We should not get -ENOSPC */
 273                                 BUG_ON(err);
 274                         }
 275
 276                         /* Check the previous node page having this index */
 277                         check_index_in_prev_nodes(sbi, dest);
 278
 279                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 280
 281                         /* write dummy data page */
 282                         recover_data_page(sbi, NULL, &sum, src, dest);
 283                         update_extent_cache(dest, &dn);
 284                 }
 285                 dn.ofs_in_node++;
 286         }
 287
 288         /* write node page in place */
 289         set_summary(&sum, dn.nid, 0, 0);
 290         if (IS_INODE(dn.node_page))
 291                 sync_inode_page(&dn);
 292
 293         copy_node_footer(dn.node_page, page);
 294         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 295                                         ofs_of_node(page), false);
 296         set_page_dirty(dn.node_page);
 297
 298         recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
 299         f2fs_put_dnode(&dn);
 300 }
 301
 302 static void recover_data(struct f2fs_sb_info *sbi,
 303                                 struct list_head *head, int type)
 304 {
 305         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
 306         struct curseg_info *curseg;
 307         struct page *page;
 308         block_t blkaddr;
 309
 310         /* get node pages in the current segment */
 311         curseg = CURSEG_I(sbi, type);
 312         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 313
 314         /* read node page */
 315         page = alloc_page(GFP_NOFS | __GFP_ZERO);
 316         if (IS_ERR(page))
 317                 return;
 318         lock_page(page);
 319
 320         while (1) {
 321                 struct fsync_inode_entry *entry;
 322
 323                 if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
 324                         goto out;
 325
 326                 if (cp_ver != cpver_of_node(page))
 327                         goto out;
 328
 329                 entry = get_fsync_inode(head, ino_of_node(page));
 330                 if (!entry)
 331                         goto next;
 332
 333                 do_recover_data(sbi, entry->inode, page, blkaddr);
 334
 335                 if (entry->blkaddr == blkaddr) {
 336                         iput(entry->inode);
 337                         list_del(&entry->list);
 338                         kmem_cache_free(fsync_entry_slab, entry);
 339                 }
 340 next:
 341                 /* check next segment */
 342                 blkaddr = next_blkaddr_of_node(page);
 343                 ClearPageUptodate(page);
 344         }
 345 out:
 346         unlock_page(page);
 347         __free_pages(page, 0);
 348
 349         allocate_new_segments(sbi);
 350 }
 351
 352 void recover_fsync_data(struct f2fs_sb_info *sbi)
 353 {
 354         struct list_head inode_list;
 355
 356         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 357                         sizeof(struct fsync_inode_entry), NULL);
 358         if (unlikely(!fsync_entry_slab))
 359                 return;
 360
 361         INIT_LIST_HEAD(&inode_list);
 362
 363         /* step #1: find fsynced inode numbers */
 364         if (find_fsync_dnodes(sbi, &inode_list))
 365                 goto out;
 366
 367         if (list_empty(&inode_list))
 368                 goto out;
 369
 370         /* step #2: recover data */
 371         sbi->por_doing = 1;
 372         recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 373         sbi->por_doing = 0;
 374         BUG_ON(!list_empty(&inode_list));
 375 out:
 376         destroy_fsync_dnodes(sbi, &inode_list);
 377         kmem_cache_destroy(fsync_entry_slab);
 378         write_checkpoint(sbi, false, false);
 379 }