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                 __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                         INIT_LIST_HEAD(&entry->list);
 153                         list_add_tail(&entry->list, head);
 154                         entry->blkaddr = blkaddr;
 155                 }
 156                 if (IS_INODE(page)) {
 157                         err = recover_inode(entry->inode, page);
 158                         if (err)
 159                                 goto out;
 160                 }
 161 next:
 162                 /* check next segment */
 163                 blkaddr = next_blkaddr_of_node(page);
 164                 ClearPageUptodate(page);
 165         }
 166 out:
 167         unlock_page(page);
 168         __free_pages(page, 0);
 169         return err;
 170 }
 171
 172 static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi,
 173                                         struct list_head *head)
 174 {
 175         struct list_head *this;
 176         struct fsync_inode_entry *entry;
 177         list_for_each(this, head) {
 178                 entry = list_entry(this, struct fsync_inode_entry, list);
 179                 iput(entry->inode);
 180                 list_del(&entry->list);
 181                 kmem_cache_free(fsync_entry_slab, entry);
 182         }
 183 }
 184
 185 static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 186                                                 block_t blkaddr)
 187 {
 188         struct seg_entry *sentry;
 189         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 190         unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
 191                                         (sbi->blocks_per_seg - 1);
 192         struct f2fs_summary sum;
 193         nid_t ino;
 194         void *kaddr;
 195         struct inode *inode;
 196         struct page *node_page;
 197         block_t bidx;
 198         int i;
 199
 200         sentry = get_seg_entry(sbi, segno);
 201         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 202                 return;
 203
 204         /* Get the previous summary */
 205         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 206                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 207                 if (curseg->segno == segno) {
 208                         sum = curseg->sum_blk->entries[blkoff];
 209                         break;
 210                 }
 211         }
 212         if (i > CURSEG_COLD_DATA) {
 213                 struct page *sum_page = get_sum_page(sbi, segno);
 214                 struct f2fs_summary_block *sum_node;
 215                 kaddr = page_address(sum_page);
 216                 sum_node = (struct f2fs_summary_block *)kaddr;
 217                 sum = sum_node->entries[blkoff];
 218                 f2fs_put_page(sum_page, 1);
 219         }
 220
 221         /* Get the node page */
 222         node_page = get_node_page(sbi, le32_to_cpu(sum.nid));
 223         bidx = start_bidx_of_node(ofs_of_node(node_page)) +
 224                                 le16_to_cpu(sum.ofs_in_node);
 225         ino = ino_of_node(node_page);
 226         f2fs_put_page(node_page, 1);
 227
 228         /* Deallocate previous index in the node page */
 229         inode = f2fs_iget_nowait(sbi->sb, ino);
 230         if (IS_ERR(inode))
 231                 return;
 232
 233         truncate_hole(inode, bidx, bidx + 1);
 234         iput(inode);
 235 }
 236
 237 static void do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 238                                         struct page *page, block_t blkaddr)
 239 {
 240         unsigned int start, end;
 241         struct dnode_of_data dn;
 242         struct f2fs_summary sum;
 243         struct node_info ni;
 244
 245         start = start_bidx_of_node(ofs_of_node(page));
 246         if (IS_INODE(page))
 247                 end = start + ADDRS_PER_INODE;
 248         else
 249                 end = start + ADDRS_PER_BLOCK;
 250
 251         set_new_dnode(&dn, inode, NULL, NULL, 0);
 252         if (get_dnode_of_data(&dn, start, 0))
 253                 return;
 254
 255         wait_on_page_writeback(dn.node_page);
 256
 257         get_node_info(sbi, dn.nid, &ni);
 258         BUG_ON(ni.ino != ino_of_node(page));
 259         BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page));
 260
 261         for (; start < end; start++) {
 262                 block_t src, dest;
 263
 264                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 265                 dest = datablock_addr(page, dn.ofs_in_node);
 266
 267                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 268                         if (src == NULL_ADDR) {
 269                                 int err = reserve_new_block(&dn);
 270                                 /* We should not get -ENOSPC */
 271                                 BUG_ON(err);
 272                         }
 273
 274                         /* Check the previous node page having this index */
 275                         check_index_in_prev_nodes(sbi, dest);
 276
 277                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 278
 279                         /* write dummy data page */
 280                         recover_data_page(sbi, NULL, &sum, src, dest);
 281                         update_extent_cache(dest, &dn);
 282                 }
 283                 dn.ofs_in_node++;
 284         }
 285
 286         /* write node page in place */
 287         set_summary(&sum, dn.nid, 0, 0);
 288         if (IS_INODE(dn.node_page))
 289                 sync_inode_page(&dn);
 290
 291         copy_node_footer(dn.node_page, page);
 292         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 293                                         ofs_of_node(page), false);
 294         set_page_dirty(dn.node_page);
 295
 296         recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
 297         f2fs_put_dnode(&dn);
 298 }
 299
 300 static void recover_data(struct f2fs_sb_info *sbi,
 301                                 struct list_head *head, int type)
 302 {
 303         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
 304         struct curseg_info *curseg;
 305         struct page *page;
 306         block_t blkaddr;
 307
 308         /* get node pages in the current segment */
 309         curseg = CURSEG_I(sbi, type);
 310         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 311
 312         /* read node page */
 313         page = alloc_page(GFP_NOFS | __GFP_ZERO);
 314         if (IS_ERR(page))
 315                 return;
 316         lock_page(page);
 317
 318         while (1) {
 319                 struct fsync_inode_entry *entry;
 320
 321                 if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
 322                         goto out;
 323
 324                 if (cp_ver != cpver_of_node(page))
 325                         goto out;
 326
 327                 entry = get_fsync_inode(head, ino_of_node(page));
 328                 if (!entry)
 329                         goto next;
 330
 331                 do_recover_data(sbi, entry->inode, page, blkaddr);
 332
 333                 if (entry->blkaddr == blkaddr) {
 334                         iput(entry->inode);
 335                         list_del(&entry->list);
 336                         kmem_cache_free(fsync_entry_slab, entry);
 337                 }
 338 next:
 339                 /* check next segment */
 340                 blkaddr = next_blkaddr_of_node(page);
 341                 ClearPageUptodate(page);
 342         }
 343 out:
 344         unlock_page(page);
 345         __free_pages(page, 0);
 346
 347         allocate_new_segments(sbi);
 348 }
 349
 350 void recover_fsync_data(struct f2fs_sb_info *sbi)
 351 {
 352         struct list_head inode_list;
 353
 354         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 355                         sizeof(struct fsync_inode_entry), NULL);
 356         if (unlikely(!fsync_entry_slab))
 357                 return;
 358
 359         INIT_LIST_HEAD(&inode_list);
 360
 361         /* step #1: find fsynced inode numbers */
 362         if (find_fsync_dnodes(sbi, &inode_list))
 363                 goto out;
 364
 365         if (list_empty(&inode_list))
 366                 goto out;
 367
 368         /* step #2: recover data */
 369         sbi->por_doing = 1;
 370         recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 371         sbi->por_doing = 0;
 372         BUG_ON(!list_empty(&inode_list));
 373 out:
 374         destroy_fsync_dnodes(sbi, &inode_list);
 375         kmem_cache_destroy(fsync_entry_slab);
 376         write_checkpoint(sbi, false, false);
 377 }