1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Copyright (C) 2007 Oracle. All rights reserved.
9 #include <linux/hash.h>
10 #include <linux/refcount.h>
11 #include "extent_map.h"
12 #include "extent_io.h"
13 #include "ordered-data.h"
14 #include "delayed-inode.h"
17 * Since we search a directory based on f_pos (struct dir_context::pos) we have
18 * to start at 2 since '.' and '..' have f_pos of 0 and 1 respectively, so
19 * everybody else has to start at 2 (see btrfs_real_readdir() and dir_emit_dots()).
21 #define BTRFS_DIR_START_INDEX 2
24 * ordered_data_close is set by truncate when a file that used
25 * to have good data has been truncated to zero. When it is set
26 * the btrfs file release call will add this inode to the
27 * ordered operations list so that we make sure to flush out any
28 * new data the application may have written before commit.
31 BTRFS_INODE_FLUSH_ON_CLOSE,
33 BTRFS_INODE_IN_DEFRAG,
34 BTRFS_INODE_HAS_ASYNC_EXTENT,
36 * Always set under the VFS' inode lock, otherwise it can cause races
37 * during fsync (we start as a fast fsync and then end up in a full
38 * fsync racing with ordered extent completion).
40 BTRFS_INODE_NEEDS_FULL_SYNC,
41 BTRFS_INODE_COPY_EVERYTHING,
42 BTRFS_INODE_IN_DELALLOC_LIST,
43 BTRFS_INODE_HAS_PROPS,
44 BTRFS_INODE_SNAPSHOT_FLUSH,
46 * Set and used when logging an inode and it serves to signal that an
47 * inode does not have xattrs, so subsequent fsyncs can avoid searching
48 * for xattrs to log. This bit must be cleared whenever a xattr is added
51 BTRFS_INODE_NO_XATTRS,
53 * Set when we are in a context where we need to start a transaction and
54 * have dirty pages with the respective file range locked. This is to
55 * ensure that when reserving space for the transaction, if we are low
56 * on available space and need to flush delalloc, we will not flush
57 * delalloc for this inode, because that could result in a deadlock (on
58 * the file range, inode's io_tree).
60 BTRFS_INODE_NO_DELALLOC_FLUSH,
62 * Set when we are working on enabling verity for a file. Computing and
63 * writing the whole Merkle tree can take a while so we want to prevent
64 * races where two separate tasks attempt to simultaneously start verity
67 BTRFS_INODE_VERITY_IN_PROGRESS,
70 /* in memory btrfs inode */
72 /* which subvolume this inode belongs to */
73 struct btrfs_root *root;
75 /* key used to find this inode on disk. This is used by the code
76 * to read in roots of subvolumes
78 struct btrfs_key location;
81 * Lock for counters and all fields used to determine if the inode is in
82 * the log or not (last_trans, last_sub_trans, last_log_commit,
83 * logged_trans), to access/update new_delalloc_bytes and to update the
84 * VFS' inode number of bytes used.
88 /* the extent_tree has caches of all the extent mappings to disk */
89 struct extent_map_tree extent_tree;
91 /* the io_tree does range state (DIRTY, LOCKED etc) */
92 struct extent_io_tree io_tree;
94 /* special utility tree used to record which mirrors have already been
95 * tried when checksums fail for a given block
97 struct extent_io_tree io_failure_tree;
100 * Keep track of where the inode has extent items mapped in order to
101 * make sure the i_size adjustments are accurate
103 struct extent_io_tree file_extent_tree;
105 /* held while logging the inode in tree-log.c */
106 struct mutex log_mutex;
108 /* used to order data wrt metadata */
109 struct btrfs_ordered_inode_tree ordered_tree;
111 /* list of all the delalloc inodes in the FS. There are times we need
112 * to write all the delalloc pages to disk, and this list is used
115 struct list_head delalloc_inodes;
117 /* node for the red-black tree that links inodes in subvolume root */
118 struct rb_node rb_node;
120 unsigned long runtime_flags;
122 /* Keep track of who's O_SYNC/fsyncing currently */
123 atomic_t sync_writers;
125 /* full 64 bit generation number, struct vfs_inode doesn't have a big
126 * enough field for this.
131 * transid of the trans_handle that last modified this inode
136 * transid that last logged this inode
141 * log transid when this inode was last modified
145 /* a local copy of root's last_log_commit */
149 * Total number of bytes pending delalloc, used by stat to calculate the
150 * real block usage of the file. This is used only for files.
156 * Total number of bytes pending delalloc that fall within a file
157 * range that is either a hole or beyond EOF (and no prealloc extent
158 * exists in the range). This is always <= delalloc_bytes and this
159 * is used only for files.
161 u64 new_delalloc_bytes;
163 * The offset of the last dir index key that was logged.
164 * This is used only for directories.
166 u64 last_dir_index_offset;
170 * total number of bytes pending defrag, used by stat to check whether
176 * the size of the file stored in the metadata on disk. data=ordered
177 * means the in-memory i_size might be larger than the size on disk
178 * because not all the blocks are written yet.
183 * If this is a directory then index_cnt is the counter for the index
184 * number for new files that are created. For an empty directory, this
185 * must be initialized to BTRFS_DIR_START_INDEX.
189 /* Cache the directory index number to speed the dir/file remove */
192 /* the fsync log has some corner cases that mean we have to check
193 * directories to see if any unlinks have been done before
194 * the directory was logged. See tree-log.c for all the
197 u64 last_unlink_trans;
200 * The id/generation of the last transaction where this inode was
201 * either the source or the destination of a clone/dedupe operation.
202 * Used when logging an inode to know if there are shared extents that
203 * need special care when logging checksum items, to avoid duplicate
204 * checksum items in a log (which can lead to a corruption where we end
205 * up with missing checksum ranges after log replay).
206 * Protected by the vfs inode lock.
208 u64 last_reflink_trans;
211 * Number of bytes outstanding that are going to need csums. This is
212 * used in ENOSPC accounting.
216 /* Backwards incompatible flags, lower half of inode_item::flags */
218 /* Read-only compatibility flags, upper half of inode_item::flags */
222 * Counters to keep track of the number of extent item's we may use due
223 * to delalloc and such. outstanding_extents is the number of extent
224 * items we think we'll end up using, and reserved_extents is the number
225 * of extent items we've reserved metadata for.
227 unsigned outstanding_extents;
229 struct btrfs_block_rsv block_rsv;
232 * Cached values of inode properties
234 unsigned prop_compress; /* per-file compression algorithm */
236 * Force compression on the file using the defrag ioctl, could be
237 * different from prop_compress and takes precedence if set
239 unsigned defrag_compress;
241 struct btrfs_delayed_node *delayed_node;
243 /* File creation time. */
244 struct timespec64 i_otime;
246 /* Hook into fs_info->delayed_iputs */
247 struct list_head delayed_iput;
249 struct rw_semaphore i_mmap_lock;
250 struct inode vfs_inode;
253 static inline u32 btrfs_inode_sectorsize(const struct btrfs_inode *inode)
255 return inode->root->fs_info->sectorsize;
258 static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
260 return container_of(inode, struct btrfs_inode, vfs_inode);
263 static inline unsigned long btrfs_inode_hash(u64 objectid,
264 const struct btrfs_root *root)
266 u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
268 #if BITS_PER_LONG == 32
269 h = (h >> 32) ^ (h & 0xffffffff);
272 return (unsigned long)h;
275 static inline void btrfs_insert_inode_hash(struct inode *inode)
277 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
279 __insert_inode_hash(inode, h);
282 static inline u64 btrfs_ino(const struct btrfs_inode *inode)
284 u64 ino = inode->location.objectid;
288 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
290 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
291 ino = inode->vfs_inode.i_ino;
295 static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
297 i_size_write(&inode->vfs_inode, size);
298 inode->disk_i_size = size;
301 static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
303 struct btrfs_root *root = inode->root;
305 if (root == root->fs_info->tree_root &&
306 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
308 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
313 static inline bool is_data_inode(struct inode *inode)
315 return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
318 static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
321 lockdep_assert_held(&inode->lock);
322 inode->outstanding_extents += mod;
323 if (btrfs_is_free_space_inode(inode))
325 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
330 * Called every time after doing a buffered, direct IO or memory mapped write.
332 * This is to ensure that if we write to a file that was previously fsynced in
333 * the current transaction, then try to fsync it again in the same transaction,
334 * we will know that there were changes in the file and that it needs to be
337 static inline void btrfs_set_inode_last_sub_trans(struct btrfs_inode *inode)
339 spin_lock(&inode->lock);
340 inode->last_sub_trans = inode->root->log_transid;
341 spin_unlock(&inode->lock);
345 * Should be called while holding the inode's VFS lock in exclusive mode or in a
346 * context where no one else can access the inode concurrently (during inode
347 * creation or when loading an inode from disk).
349 static inline void btrfs_set_inode_full_sync(struct btrfs_inode *inode)
351 set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
353 * The inode may have been part of a reflink operation in the last
354 * transaction that modified it, and then a fsync has reset the
355 * last_reflink_trans to avoid subsequent fsyncs in the same
356 * transaction to do unnecessary work. So update last_reflink_trans
357 * to the last_trans value (we have to be pessimistic and assume a
360 * The ->last_trans is protected by the inode's spinlock and we can
361 * have a concurrent ordered extent completion update it. Also set
362 * last_reflink_trans to ->last_trans only if the former is less than
363 * the later, because we can be called in a context where
364 * last_reflink_trans was set to the current transaction generation
365 * while ->last_trans was not yet updated in the current transaction,
366 * and therefore has a lower value.
368 spin_lock(&inode->lock);
369 if (inode->last_reflink_trans < inode->last_trans)
370 inode->last_reflink_trans = inode->last_trans;
371 spin_unlock(&inode->lock);
374 static inline bool btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
378 spin_lock(&inode->lock);
379 if (inode->logged_trans == generation &&
380 inode->last_sub_trans <= inode->last_log_commit &&
381 inode->last_sub_trans <= inode->root->last_log_commit)
383 spin_unlock(&inode->lock);
388 * Check if the inode has flags compatible with compression
390 static inline bool btrfs_inode_can_compress(const struct btrfs_inode *inode)
392 if (inode->flags & BTRFS_INODE_NODATACOW ||
393 inode->flags & BTRFS_INODE_NODATASUM)
399 * btrfs_inode_item stores flags in a u64, btrfs_inode stores them in two
400 * separate u32s. These two functions convert between the two representations.
402 static inline u64 btrfs_inode_combine_flags(u32 flags, u32 ro_flags)
404 return (flags | ((u64)ro_flags << 32));
407 static inline void btrfs_inode_split_flags(u64 inode_item_flags,
408 u32 *flags, u32 *ro_flags)
410 *flags = (u32)inode_item_flags;
411 *ro_flags = (u32)(inode_item_flags >> 32);
414 /* Array of bytes with variable length, hexadecimal format 0x1234 */
415 #define CSUM_FMT "0x%*phN"
416 #define CSUM_FMT_VALUE(size, bytes) size, bytes
418 static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
419 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
421 struct btrfs_root *root = inode->root;
422 const u32 csum_size = root->fs_info->csum_size;
424 /* Output minus objectid, which is more meaningful */
425 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
426 btrfs_warn_rl(root->fs_info,
427 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
428 root->root_key.objectid, btrfs_ino(inode),
430 CSUM_FMT_VALUE(csum_size, csum),
431 CSUM_FMT_VALUE(csum_size, csum_expected),
434 btrfs_warn_rl(root->fs_info,
435 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
436 root->root_key.objectid, btrfs_ino(inode),
438 CSUM_FMT_VALUE(csum_size, csum),
439 CSUM_FMT_VALUE(csum_size, csum_expected),