fs/f2fs/inode.c

   1 /*
   2  * fs/f2fs/inode.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 <linux/buffer_head.h>
  14 #include <linux/backing-dev.h>
  15 #include <linux/writeback.h>
  16
  17 #include "f2fs.h"
  18 #include "node.h"
  19
  20 #include <trace/events/f2fs.h>
  21
  22 void f2fs_mark_inode_dirty_sync(struct inode *inode)
  23 {
  24         if (f2fs_inode_dirtied(inode))
  25                 return;
  26         mark_inode_dirty_sync(inode);
  27 }
  28
  29 void f2fs_set_inode_flags(struct inode *inode)
  30 {
  31         unsigned int flags = F2FS_I(inode)->i_flags;
  32         unsigned int new_fl = 0;
  33
  34         if (flags & FS_SYNC_FL)
  35                 new_fl |= S_SYNC;
  36         if (flags & FS_APPEND_FL)
  37                 new_fl |= S_APPEND;
  38         if (flags & FS_IMMUTABLE_FL)
  39                 new_fl |= S_IMMUTABLE;
  40         if (flags & FS_NOATIME_FL)
  41                 new_fl |= S_NOATIME;
  42         if (flags & FS_DIRSYNC_FL)
  43                 new_fl |= S_DIRSYNC;
  44         inode_set_flags(inode, new_fl,
  45                         S_SYNC|S_APPEND|S_IMMUTABLE|S_NOATIME|S_DIRSYNC);
  46         f2fs_mark_inode_dirty_sync(inode);
  47 }
  48
  49 static void __get_inode_rdev(struct inode *inode, struct f2fs_inode *ri)
  50 {
  51         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
  52                         S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
  53                 if (ri->i_addr[0])
  54                         inode->i_rdev =
  55                                 old_decode_dev(le32_to_cpu(ri->i_addr[0]));
  56                 else
  57                         inode->i_rdev =
  58                                 new_decode_dev(le32_to_cpu(ri->i_addr[1]));
  59         }
  60 }
  61
  62 static int __written_first_block(struct f2fs_sb_info *sbi,
  63                                         struct f2fs_inode *ri)
  64 {
  65         block_t addr = le32_to_cpu(ri->i_addr[0]);
  66
  67         if (!__is_valid_data_blkaddr(addr))
  68                 return 1;
  69         if (!f2fs_is_valid_blkaddr(sbi, addr, DATA_GENERIC))
  70                 return -EFAULT;
  71         return 0;
  72 }
  73
  74 static void __set_inode_rdev(struct inode *inode, struct f2fs_inode *ri)
  75 {
  76         if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode)) {
  77                 if (old_valid_dev(inode->i_rdev)) {
  78                         ri->i_addr[0] =
  79                                 cpu_to_le32(old_encode_dev(inode->i_rdev));
  80                         ri->i_addr[1] = 0;
  81                 } else {
  82                         ri->i_addr[0] = 0;
  83                         ri->i_addr[1] =
  84                                 cpu_to_le32(new_encode_dev(inode->i_rdev));
  85                         ri->i_addr[2] = 0;
  86                 }
  87         }
  88 }
  89
  90 static void __recover_inline_status(struct inode *inode, struct page *ipage)
  91 {
  92         void *inline_data = inline_data_addr(ipage);
  93         __le32 *start = inline_data;
  94         __le32 *end = start + MAX_INLINE_DATA / sizeof(__le32);
  95
  96         while (start < end) {
  97                 if (*start++) {
  98                         f2fs_wait_on_page_writeback(ipage, NODE, true);
  99
 100                         set_inode_flag(inode, FI_DATA_EXIST);
 101                         set_raw_inline(inode, F2FS_INODE(ipage));
 102                         set_page_dirty(ipage);
 103                         return;
 104                 }
 105         }
 106         return;
 107 }
 108
 109 static bool sanity_check_inode(struct inode *inode, struct page *node_page)
 110 {
 111         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 112         unsigned long long iblocks;
 113
 114         iblocks = le64_to_cpu(F2FS_INODE(node_page)->i_blocks);
 115         if (!iblocks) {
 116                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 117                 f2fs_msg(sbi->sb, KERN_WARNING,
 118                         "%s: corrupted inode i_blocks i_ino=%lx iblocks=%llu, "
 119                         "run fsck to fix.",
 120                         __func__, inode->i_ino, iblocks);
 121                 return false;
 122         }
 123
 124         if (ino_of_node(node_page) != nid_of_node(node_page)) {
 125                 set_sbi_flag(sbi, SBI_NEED_FSCK);
 126                 f2fs_msg(sbi->sb, KERN_WARNING,
 127                         "%s: corrupted inode footer i_ino=%lx, ino,nid: "
 128                         "[%u, %u] run fsck to fix.",
 129                         __func__, inode->i_ino,
 130                         ino_of_node(node_page), nid_of_node(node_page));
 131                 return false;
 132         }
 133
 134         if (F2FS_I(inode)->extent_tree) {
 135                 struct extent_info *ei = &F2FS_I(inode)->extent_tree->largest;
 136
 137                 if (ei->len &&
 138                         (!f2fs_is_valid_blkaddr(sbi, ei->blk, DATA_GENERIC) ||
 139                         !f2fs_is_valid_blkaddr(sbi, ei->blk + ei->len - 1,
 140                                                         DATA_GENERIC))) {
 141                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 142                         f2fs_msg(sbi->sb, KERN_WARNING,
 143                                 "%s: inode (ino=%lx) extent info [%u, %u, %u] "
 144                                 "is incorrect, run fsck to fix",
 145                                 __func__, inode->i_ino,
 146                                 ei->blk, ei->fofs, ei->len);
 147                         return false;
 148                 }
 149         }
 150         return true;
 151 }
 152
 153 static int do_read_inode(struct inode *inode)
 154 {
 155         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 156         struct f2fs_inode_info *fi = F2FS_I(inode);
 157         struct page *node_page;
 158         struct f2fs_inode *ri;
 159         int err;
 160
 161         /* Check if ino is within scope */
 162         if (check_nid_range(sbi, inode->i_ino)) {
 163                 f2fs_msg(inode->i_sb, KERN_ERR, "bad inode number: %lu",
 164                          (unsigned long) inode->i_ino);
 165                 WARN_ON(1);
 166                 return -EINVAL;
 167         }
 168
 169         node_page = get_node_page(sbi, inode->i_ino);
 170         if (IS_ERR(node_page))
 171                 return PTR_ERR(node_page);
 172
 173         ri = F2FS_INODE(node_page);
 174
 175         inode->i_mode = le16_to_cpu(ri->i_mode);
 176         i_uid_write(inode, le32_to_cpu(ri->i_uid));
 177         i_gid_write(inode, le32_to_cpu(ri->i_gid));
 178         set_nlink(inode, le32_to_cpu(ri->i_links));
 179         inode->i_size = le64_to_cpu(ri->i_size);
 180         inode->i_blocks = le64_to_cpu(ri->i_blocks);
 181
 182         inode->i_atime.tv_sec = le64_to_cpu(ri->i_atime);
 183         inode->i_ctime.tv_sec = le64_to_cpu(ri->i_ctime);
 184         inode->i_mtime.tv_sec = le64_to_cpu(ri->i_mtime);
 185         inode->i_atime.tv_nsec = le32_to_cpu(ri->i_atime_nsec);
 186         inode->i_ctime.tv_nsec = le32_to_cpu(ri->i_ctime_nsec);
 187         inode->i_mtime.tv_nsec = le32_to_cpu(ri->i_mtime_nsec);
 188         inode->i_generation = le32_to_cpu(ri->i_generation);
 189
 190         fi->i_current_depth = le32_to_cpu(ri->i_current_depth);
 191         fi->i_xattr_nid = le32_to_cpu(ri->i_xattr_nid);
 192         fi->i_flags = le32_to_cpu(ri->i_flags);
 193         fi->flags = 0;
 194         fi->i_advise = ri->i_advise;
 195         fi->i_pino = le32_to_cpu(ri->i_pino);
 196         fi->i_dir_level = ri->i_dir_level;
 197
 198         if (f2fs_init_extent_tree(inode, &ri->i_ext))
 199                 set_page_dirty(node_page);
 200
 201         get_inline_info(inode, ri);
 202
 203         if (!sanity_check_inode(inode, node_page)) {
 204                 f2fs_put_page(node_page, 1);
 205                 return -EINVAL;
 206         }
 207
 208         /* check data exist */
 209         if (f2fs_has_inline_data(inode) && !f2fs_exist_data(inode))
 210                 __recover_inline_status(inode, node_page);
 211
 212         /* get rdev by using inline_info */
 213         __get_inode_rdev(inode, ri);
 214
 215         err = __written_first_block(sbi, ri);
 216         if (err < 0) {
 217                 f2fs_put_page(node_page, 1);
 218                 return err;
 219         }
 220         if (!err)
 221                 set_inode_flag(inode, FI_FIRST_BLOCK_WRITTEN);
 222
 223         if (!need_inode_block_update(sbi, inode->i_ino))
 224                 fi->last_disk_size = inode->i_size;
 225
 226         f2fs_put_page(node_page, 1);
 227
 228         stat_inc_inline_xattr(inode);
 229         stat_inc_inline_inode(inode);
 230         stat_inc_inline_dir(inode);
 231
 232         return 0;
 233 }
 234
 235 struct inode *f2fs_iget(struct super_block *sb, unsigned long ino)
 236 {
 237         struct f2fs_sb_info *sbi = F2FS_SB(sb);
 238         struct inode *inode;
 239         int ret = 0;
 240
 241         inode = iget_locked(sb, ino);
 242         if (!inode)
 243                 return ERR_PTR(-ENOMEM);
 244
 245         if (!(inode->i_state & I_NEW)) {
 246                 trace_f2fs_iget(inode);
 247                 return inode;
 248         }
 249         if (ino == F2FS_NODE_INO(sbi) || ino == F2FS_META_INO(sbi))
 250                 goto make_now;
 251
 252         ret = do_read_inode(inode);
 253         if (ret)
 254                 goto bad_inode;
 255 make_now:
 256         if (ino == F2FS_NODE_INO(sbi)) {
 257                 inode->i_mapping->a_ops = &f2fs_node_aops;
 258                 mapping_set_gfp_mask(inode->i_mapping, GFP_F2FS_ZERO);
 259         } else if (ino == F2FS_META_INO(sbi)) {
 260                 inode->i_mapping->a_ops = &f2fs_meta_aops;
 261                 mapping_set_gfp_mask(inode->i_mapping, GFP_F2FS_ZERO);
 262         } else if (S_ISREG(inode->i_mode)) {
 263                 inode->i_op = &f2fs_file_inode_operations;
 264                 inode->i_fop = &f2fs_file_operations;
 265                 inode->i_mapping->a_ops = &f2fs_dblock_aops;
 266         } else if (S_ISDIR(inode->i_mode)) {
 267                 inode->i_op = &f2fs_dir_inode_operations;
 268                 inode->i_fop = &f2fs_dir_operations;
 269                 inode->i_mapping->a_ops = &f2fs_dblock_aops;
 270                 mapping_set_gfp_mask(inode->i_mapping, GFP_F2FS_HIGH_ZERO);
 271         } else if (S_ISLNK(inode->i_mode)) {
 272                 if (f2fs_encrypted_inode(inode))
 273                         inode->i_op = &f2fs_encrypted_symlink_inode_operations;
 274                 else
 275                         inode->i_op = &f2fs_symlink_inode_operations;
 276                 inode_nohighmem(inode);
 277                 inode->i_mapping->a_ops = &f2fs_dblock_aops;
 278         } else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 279                         S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 280                 inode->i_op = &f2fs_special_inode_operations;
 281                 init_special_inode(inode, inode->i_mode, inode->i_rdev);
 282         } else {
 283                 ret = -EIO;
 284                 goto bad_inode;
 285         }
 286         unlock_new_inode(inode);
 287         trace_f2fs_iget(inode);
 288         return inode;
 289
 290 bad_inode:
 291         f2fs_inode_synced(inode);
 292         iget_failed(inode);
 293         trace_f2fs_iget_exit(inode, ret);
 294         return ERR_PTR(ret);
 295 }
 296
 297 struct inode *f2fs_iget_retry(struct super_block *sb, unsigned long ino)
 298 {
 299         struct inode *inode;
 300 retry:
 301         inode = f2fs_iget(sb, ino);
 302         if (IS_ERR(inode)) {
 303                 if (PTR_ERR(inode) == -ENOMEM) {
 304                         congestion_wait(BLK_RW_ASYNC, HZ/50);
 305                         goto retry;
 306                 }
 307         }
 308         return inode;
 309 }
 310
 311 int update_inode(struct inode *inode, struct page *node_page)
 312 {
 313         struct f2fs_inode *ri;
 314
 315         f2fs_inode_synced(inode);
 316
 317         f2fs_wait_on_page_writeback(node_page, NODE, true);
 318
 319         ri = F2FS_INODE(node_page);
 320
 321         ri->i_mode = cpu_to_le16(inode->i_mode);
 322         ri->i_advise = F2FS_I(inode)->i_advise;
 323         ri->i_uid = cpu_to_le32(i_uid_read(inode));
 324         ri->i_gid = cpu_to_le32(i_gid_read(inode));
 325         ri->i_links = cpu_to_le32(inode->i_nlink);
 326         ri->i_size = cpu_to_le64(i_size_read(inode));
 327         ri->i_blocks = cpu_to_le64(inode->i_blocks);
 328
 329         if (F2FS_I(inode)->extent_tree)
 330                 set_raw_extent(&F2FS_I(inode)->extent_tree->largest,
 331                                                         &ri->i_ext);
 332         else
 333                 memset(&ri->i_ext, 0, sizeof(ri->i_ext));
 334         set_raw_inline(inode, ri);
 335
 336         ri->i_atime = cpu_to_le64(inode->i_atime.tv_sec);
 337         ri->i_ctime = cpu_to_le64(inode->i_ctime.tv_sec);
 338         ri->i_mtime = cpu_to_le64(inode->i_mtime.tv_sec);
 339         ri->i_atime_nsec = cpu_to_le32(inode->i_atime.tv_nsec);
 340         ri->i_ctime_nsec = cpu_to_le32(inode->i_ctime.tv_nsec);
 341         ri->i_mtime_nsec = cpu_to_le32(inode->i_mtime.tv_nsec);
 342         ri->i_current_depth = cpu_to_le32(F2FS_I(inode)->i_current_depth);
 343         ri->i_xattr_nid = cpu_to_le32(F2FS_I(inode)->i_xattr_nid);
 344         ri->i_flags = cpu_to_le32(F2FS_I(inode)->i_flags);
 345         ri->i_pino = cpu_to_le32(F2FS_I(inode)->i_pino);
 346         ri->i_generation = cpu_to_le32(inode->i_generation);
 347         ri->i_dir_level = F2FS_I(inode)->i_dir_level;
 348
 349         __set_inode_rdev(inode, ri);
 350         set_cold_node(inode, node_page);
 351
 352         /* deleted inode */
 353         if (inode->i_nlink == 0)
 354                 clear_inline_node(node_page);
 355
 356         return set_page_dirty(node_page);
 357 }
 358
 359 int update_inode_page(struct inode *inode)
 360 {
 361         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 362         struct page *node_page;
 363         int ret = 0;
 364 retry:
 365         node_page = get_node_page(sbi, inode->i_ino);
 366         if (IS_ERR(node_page)) {
 367                 int err = PTR_ERR(node_page);
 368                 if (err == -ENOMEM) {
 369                         cond_resched();
 370                         goto retry;
 371                 } else if (err != -ENOENT) {
 372                         f2fs_stop_checkpoint(sbi, false);
 373                 }
 374                 f2fs_inode_synced(inode);
 375                 return 0;
 376         }
 377         ret = update_inode(inode, node_page);
 378         f2fs_put_page(node_page, 1);
 379         return ret;
 380 }
 381
 382 int f2fs_write_inode(struct inode *inode, struct writeback_control *wbc)
 383 {
 384         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 385
 386         if (inode->i_ino == F2FS_NODE_INO(sbi) ||
 387                         inode->i_ino == F2FS_META_INO(sbi))
 388                 return 0;
 389
 390         if (!is_inode_flag_set(inode, FI_DIRTY_INODE))
 391                 return 0;
 392
 393         /*
 394          * We need to balance fs here to prevent from producing dirty node pages
 395          * during the urgent cleaning time when runing out of free sections.
 396          */
 397         if (update_inode_page(inode))
 398                 f2fs_balance_fs(sbi, true);
 399         return 0;
 400 }
 401
 402 /*
 403  * Called at the last iput() if i_nlink is zero
 404  */
 405 void f2fs_evict_inode(struct inode *inode)
 406 {
 407         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 408         nid_t xnid = F2FS_I(inode)->i_xattr_nid;
 409         int err = 0;
 410
 411         /* some remained atomic pages should discarded */
 412         if (f2fs_is_atomic_file(inode))
 413                 drop_inmem_pages(inode);
 414
 415         trace_f2fs_evict_inode(inode);
 416         truncate_inode_pages_final(&inode->i_data);
 417
 418         if (inode->i_ino == F2FS_NODE_INO(sbi) ||
 419                         inode->i_ino == F2FS_META_INO(sbi))
 420                 goto out_clear;
 421
 422         f2fs_bug_on(sbi, get_dirty_pages(inode));
 423         remove_dirty_inode(inode);
 424
 425         f2fs_destroy_extent_tree(inode);
 426
 427         if (inode->i_nlink || is_bad_inode(inode))
 428                 goto no_delete;
 429
 430 #ifdef CONFIG_F2FS_FAULT_INJECTION
 431         if (time_to_inject(sbi, FAULT_EVICT_INODE))
 432                 goto no_delete;
 433 #endif
 434
 435         sb_start_intwrite(inode->i_sb);
 436         set_inode_flag(inode, FI_NO_ALLOC);
 437         i_size_write(inode, 0);
 438 retry:
 439         if (F2FS_HAS_BLOCKS(inode))
 440                 err = f2fs_truncate(inode);
 441
 442         if (!err) {
 443                 f2fs_lock_op(sbi);
 444                 err = remove_inode_page(inode);
 445                 f2fs_unlock_op(sbi);
 446         }
 447
 448         /* give more chances, if ENOMEM case */
 449         if (err == -ENOMEM) {
 450                 err = 0;
 451                 goto retry;
 452         }
 453
 454         if (err)
 455                 update_inode_page(inode);
 456         sb_end_intwrite(inode->i_sb);
 457 no_delete:
 458         stat_dec_inline_xattr(inode);
 459         stat_dec_inline_dir(inode);
 460         stat_dec_inline_inode(inode);
 461
 462         invalidate_mapping_pages(NODE_MAPPING(sbi), inode->i_ino, inode->i_ino);
 463         if (xnid)
 464                 invalidate_mapping_pages(NODE_MAPPING(sbi), xnid, xnid);
 465         if (is_inode_flag_set(inode, FI_APPEND_WRITE))
 466                 add_ino_entry(sbi, inode->i_ino, APPEND_INO);
 467         if (is_inode_flag_set(inode, FI_UPDATE_WRITE))
 468                 add_ino_entry(sbi, inode->i_ino, UPDATE_INO);
 469         if (is_inode_flag_set(inode, FI_FREE_NID)) {
 470                 alloc_nid_failed(sbi, inode->i_ino);
 471                 clear_inode_flag(inode, FI_FREE_NID);
 472         }
 473         f2fs_bug_on(sbi, err &&
 474                 !exist_written_data(sbi, inode->i_ino, ORPHAN_INO));
 475 out_clear:
 476         fscrypt_put_encryption_info(inode, NULL);
 477         clear_inode(inode);
 478 }
 479
 480 /* caller should call f2fs_lock_op() */
 481 void handle_failed_inode(struct inode *inode)
 482 {
 483         struct f2fs_sb_info *sbi = F2FS_I_SB(inode);
 484         struct node_info ni;
 485
 486         /* don't make bad inode, since it becomes a regular file. */
 487         unlock_new_inode(inode);
 488
 489         /*
 490          * Note: we should add inode to orphan list before f2fs_unlock_op()
 491          * so we can prevent losing this orphan when encoutering checkpoint
 492          * and following suddenly power-off.
 493          */
 494         get_node_info(sbi, inode->i_ino, &ni);
 495
 496         if (ni.blk_addr != NULL_ADDR) {
 497                 int err = acquire_orphan_inode(sbi);
 498                 if (err) {
 499                         set_sbi_flag(sbi, SBI_NEED_FSCK);
 500                         f2fs_msg(sbi->sb, KERN_WARNING,
 501                                 "Too many orphan inodes, run fsck to fix.");
 502                 } else {
 503                         add_orphan_inode(inode);
 504                 }
 505                 alloc_nid_done(sbi, inode->i_ino);
 506         } else {
 507                 set_inode_flag(inode, FI_FREE_NID);
 508         }
 509
 510         f2fs_unlock_op(sbi);
 511
 512         /* iput will drop the inode object */
 513         iput(inode);
 514 }