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                 err = f2fs_readpage(sbi, page, blkaddr, READ_SYNC);
 116                 if (err)
 117                         goto out;
 118
 119                 lock_page(page);
 120
 121                 if (cp_ver != cpver_of_node(page)) {
 122                         err = -EINVAL;
 123                         goto unlock_out;
 124                 }
 125
 126                 if (!is_fsync_dnode(page))
 127                         goto next;
 128
 129                 entry = get_fsync_inode(head, ino_of_node(page));
 130                 if (entry) {
 131                         entry->blkaddr = blkaddr;
 132                         if (IS_INODE(page) && is_dent_dnode(page))
 133                                 set_inode_flag(F2FS_I(entry->inode),
 134                                                         FI_INC_LINK);
 135                 } else {
 136                         if (IS_INODE(page) && is_dent_dnode(page)) {
 137                                 if (recover_inode_page(sbi, page)) {
 138                                         err = -ENOMEM;
 139                                         goto unlock_out;
 140                                 }
 141                         }
 142
 143                         /* add this fsync inode to the list */
 144                         entry = kmem_cache_alloc(fsync_entry_slab, GFP_NOFS);
 145                         if (!entry) {
 146                                 err = -ENOMEM;
 147                                 goto unlock_out;
 148                         }
 149
 150                         entry->inode = f2fs_iget(sbi->sb, ino_of_node(page));
 151                         if (IS_ERR(entry->inode)) {
 152                                 err = PTR_ERR(entry->inode);
 153                                 kmem_cache_free(fsync_entry_slab, entry);
 154                                 goto unlock_out;
 155                         }
 156
 157                         list_add_tail(&entry->list, head);
 158                         entry->blkaddr = blkaddr;
 159                 }
 160                 if (IS_INODE(page)) {
 161                         err = recover_inode(entry->inode, page);
 162                         if (err)
 163                                 goto unlock_out;
 164                 }
 165 next:
 166                 /* check next segment */
 167                 blkaddr = next_blkaddr_of_node(page);
 168         }
 169 unlock_out:
 170         unlock_page(page);
 171 out:
 172         __free_pages(page, 0);
 173         return err;
 174 }
 175
 176 static void destroy_fsync_dnodes(struct f2fs_sb_info *sbi,
 177                                         struct list_head *head)
 178 {
 179         struct fsync_inode_entry *entry, *tmp;
 180
 181         list_for_each_entry_safe(entry, tmp, head, list) {
 182                 iput(entry->inode);
 183                 list_del(&entry->list);
 184                 kmem_cache_free(fsync_entry_slab, entry);
 185         }
 186 }
 187
 188 static void check_index_in_prev_nodes(struct f2fs_sb_info *sbi,
 189                                                 block_t blkaddr)
 190 {
 191         struct seg_entry *sentry;
 192         unsigned int segno = GET_SEGNO(sbi, blkaddr);
 193         unsigned short blkoff = GET_SEGOFF_FROM_SEG0(sbi, blkaddr) &
 194                                         (sbi->blocks_per_seg - 1);
 195         struct f2fs_summary sum;
 196         nid_t ino;
 197         void *kaddr;
 198         struct inode *inode;
 199         struct page *node_page;
 200         block_t bidx;
 201         int i;
 202
 203         sentry = get_seg_entry(sbi, segno);
 204         if (!f2fs_test_bit(blkoff, sentry->cur_valid_map))
 205                 return;
 206
 207         /* Get the previous summary */
 208         for (i = CURSEG_WARM_DATA; i <= CURSEG_COLD_DATA; i++) {
 209                 struct curseg_info *curseg = CURSEG_I(sbi, i);
 210                 if (curseg->segno == segno) {
 211                         sum = curseg->sum_blk->entries[blkoff];
 212                         break;
 213                 }
 214         }
 215         if (i > CURSEG_COLD_DATA) {
 216                 struct page *sum_page = get_sum_page(sbi, segno);
 217                 struct f2fs_summary_block *sum_node;
 218                 kaddr = page_address(sum_page);
 219                 sum_node = (struct f2fs_summary_block *)kaddr;
 220                 sum = sum_node->entries[blkoff];
 221                 f2fs_put_page(sum_page, 1);
 222         }
 223
 224         /* Get the node page */
 225         node_page = get_node_page(sbi, le32_to_cpu(sum.nid));
 226         bidx = start_bidx_of_node(ofs_of_node(node_page)) +
 227                                 le16_to_cpu(sum.ofs_in_node);
 228         ino = ino_of_node(node_page);
 229         f2fs_put_page(node_page, 1);
 230
 231         /* Deallocate previous index in the node page */
 232         inode = f2fs_iget(sbi->sb, ino);
 233         if (IS_ERR(inode))
 234                 return;
 235
 236         truncate_hole(inode, bidx, bidx + 1);
 237         iput(inode);
 238 }
 239
 240 static void do_recover_data(struct f2fs_sb_info *sbi, struct inode *inode,
 241                                         struct page *page, block_t blkaddr)
 242 {
 243         unsigned int start, end;
 244         struct dnode_of_data dn;
 245         struct f2fs_summary sum;
 246         struct node_info ni;
 247
 248         start = start_bidx_of_node(ofs_of_node(page));
 249         if (IS_INODE(page))
 250                 end = start + ADDRS_PER_INODE;
 251         else
 252                 end = start + ADDRS_PER_BLOCK;
 253
 254         set_new_dnode(&dn, inode, NULL, NULL, 0);
 255         if (get_dnode_of_data(&dn, start, ALLOC_NODE))
 256                 return;
 257
 258         wait_on_page_writeback(dn.node_page);
 259
 260         get_node_info(sbi, dn.nid, &ni);
 261         BUG_ON(ni.ino != ino_of_node(page));
 262         BUG_ON(ofs_of_node(dn.node_page) != ofs_of_node(page));
 263
 264         for (; start < end; start++) {
 265                 block_t src, dest;
 266
 267                 src = datablock_addr(dn.node_page, dn.ofs_in_node);
 268                 dest = datablock_addr(page, dn.ofs_in_node);
 269
 270                 if (src != dest && dest != NEW_ADDR && dest != NULL_ADDR) {
 271                         if (src == NULL_ADDR) {
 272                                 int err = reserve_new_block(&dn);
 273                                 /* We should not get -ENOSPC */
 274                                 BUG_ON(err);
 275                         }
 276
 277                         /* Check the previous node page having this index */
 278                         check_index_in_prev_nodes(sbi, dest);
 279
 280                         set_summary(&sum, dn.nid, dn.ofs_in_node, ni.version);
 281
 282                         /* write dummy data page */
 283                         recover_data_page(sbi, NULL, &sum, src, dest);
 284                         update_extent_cache(dest, &dn);
 285                 }
 286                 dn.ofs_in_node++;
 287         }
 288
 289         /* write node page in place */
 290         set_summary(&sum, dn.nid, 0, 0);
 291         if (IS_INODE(dn.node_page))
 292                 sync_inode_page(&dn);
 293
 294         copy_node_footer(dn.node_page, page);
 295         fill_node_footer(dn.node_page, dn.nid, ni.ino,
 296                                         ofs_of_node(page), false);
 297         set_page_dirty(dn.node_page);
 298
 299         recover_node_page(sbi, dn.node_page, &sum, &ni, blkaddr);
 300         f2fs_put_dnode(&dn);
 301 }
 302
 303 static void recover_data(struct f2fs_sb_info *sbi,
 304                                 struct list_head *head, int type)
 305 {
 306         unsigned long long cp_ver = le64_to_cpu(sbi->ckpt->checkpoint_ver);
 307         struct curseg_info *curseg;
 308         struct page *page;
 309         block_t blkaddr;
 310
 311         /* get node pages in the current segment */
 312         curseg = CURSEG_I(sbi, type);
 313         blkaddr = NEXT_FREE_BLKADDR(sbi, curseg);
 314
 315         /* read node page */
 316         page = alloc_page(GFP_NOFS | __GFP_ZERO);
 317         if (IS_ERR(page))
 318                 return;
 319         lock_page(page);
 320
 321         while (1) {
 322                 struct fsync_inode_entry *entry;
 323
 324                 if (f2fs_readpage(sbi, page, blkaddr, READ_SYNC))
 325                         goto out;
 326
 327                 lock_page(page);
 328
 329                 if (cp_ver != cpver_of_node(page))
 330                         goto unlock_out;
 331
 332                 entry = get_fsync_inode(head, ino_of_node(page));
 333                 if (!entry)
 334                         goto next;
 335
 336                 do_recover_data(sbi, entry->inode, page, blkaddr);
 337
 338                 if (entry->blkaddr == blkaddr) {
 339                         iput(entry->inode);
 340                         list_del(&entry->list);
 341                         kmem_cache_free(fsync_entry_slab, entry);
 342                 }
 343 next:
 344                 /* check next segment */
 345                 blkaddr = next_blkaddr_of_node(page);
 346         }
 347 unlock_out:
 348         unlock_page(page);
 349 out:
 350         __free_pages(page, 0);
 351
 352         allocate_new_segments(sbi);
 353 }
 354
 355 void recover_fsync_data(struct f2fs_sb_info *sbi)
 356 {
 357         struct list_head inode_list;
 358
 359         fsync_entry_slab = f2fs_kmem_cache_create("f2fs_fsync_inode_entry",
 360                         sizeof(struct fsync_inode_entry), NULL);
 361         if (unlikely(!fsync_entry_slab))
 362                 return;
 363
 364         INIT_LIST_HEAD(&inode_list);
 365
 366         /* step #1: find fsynced inode numbers */
 367         if (find_fsync_dnodes(sbi, &inode_list))
 368                 goto out;
 369
 370         if (list_empty(&inode_list))
 371                 goto out;
 372
 373         /* step #2: recover data */
 374         sbi->por_doing = 1;
 375         recover_data(sbi, &inode_list, CURSEG_WARM_NODE);
 376         sbi->por_doing = 0;
 377         BUG_ON(!list_empty(&inode_list));
 378 out:
 379         destroy_fsync_dnodes(sbi, &inode_list);
 380         kmem_cache_destroy(fsync_entry_slab);
 381         write_checkpoint(sbi, false);
 382 }