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