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 * ordered_data_close is set by truncate when a file that used
18 * to have good data has been truncated to zero. When it is set
19 * the btrfs file release call will add this inode to the
20 * ordered operations list so that we make sure to flush out any
21 * new data the application may have written before commit.
24 BTRFS_INODE_FLUSH_ON_CLOSE,
26 BTRFS_INODE_IN_DEFRAG,
27 BTRFS_INODE_HAS_ASYNC_EXTENT,
29 * Always set under the VFS' inode lock, otherwise it can cause races
30 * during fsync (we start as a fast fsync and then end up in a full
31 * fsync racing with ordered extent completion).
33 BTRFS_INODE_NEEDS_FULL_SYNC,
34 BTRFS_INODE_COPY_EVERYTHING,
35 BTRFS_INODE_IN_DELALLOC_LIST,
36 BTRFS_INODE_HAS_PROPS,
37 BTRFS_INODE_SNAPSHOT_FLUSH,
40 /* in memory btrfs inode */
42 /* which subvolume this inode belongs to */
43 struct btrfs_root *root;
45 /* key used to find this inode on disk. This is used by the code
46 * to read in roots of subvolumes
48 struct btrfs_key location;
51 * Lock for counters and all fields used to determine if the inode is in
52 * the log or not (last_trans, last_sub_trans, last_log_commit,
57 /* the extent_tree has caches of all the extent mappings to disk */
58 struct extent_map_tree extent_tree;
60 /* the io_tree does range state (DIRTY, LOCKED etc) */
61 struct extent_io_tree io_tree;
63 /* special utility tree used to record which mirrors have already been
64 * tried when checksums fail for a given block
66 struct extent_io_tree io_failure_tree;
69 * Keep track of where the inode has extent items mapped in order to
70 * make sure the i_size adjustments are accurate
72 struct extent_io_tree file_extent_tree;
74 /* held while logging the inode in tree-log.c */
75 struct mutex log_mutex;
77 /* used to order data wrt metadata */
78 struct btrfs_ordered_inode_tree ordered_tree;
80 /* list of all the delalloc inodes in the FS. There are times we need
81 * to write all the delalloc pages to disk, and this list is used
84 struct list_head delalloc_inodes;
86 /* node for the red-black tree that links inodes in subvolume root */
87 struct rb_node rb_node;
89 unsigned long runtime_flags;
91 /* Keep track of who's O_SYNC/fsyncing currently */
92 atomic_t sync_writers;
94 /* full 64 bit generation number, struct vfs_inode doesn't have a big
95 * enough field for this.
100 * transid of the trans_handle that last modified this inode
105 * transid that last logged this inode
110 * log transid when this inode was last modified
114 /* a local copy of root's last_log_commit */
117 /* total number of bytes pending delalloc, used by stat to calc the
118 * real block usage of the file
123 * Total number of bytes pending delalloc that fall within a file
124 * range that is either a hole or beyond EOF (and no prealloc extent
125 * exists in the range). This is always <= delalloc_bytes.
127 u64 new_delalloc_bytes;
130 * total number of bytes pending defrag, used by stat to check whether
136 * the size of the file stored in the metadata on disk. data=ordered
137 * means the in-memory i_size might be larger than the size on disk
138 * because not all the blocks are written yet.
143 * if this is a directory then index_cnt is the counter for the index
144 * number for new files that are created
148 /* Cache the directory index number to speed the dir/file remove */
151 /* the fsync log has some corner cases that mean we have to check
152 * directories to see if any unlinks have been done before
153 * the directory was logged. See tree-log.c for all the
156 u64 last_unlink_trans;
159 * The id/generation of the last transaction where this inode was
160 * either the source or the destination of a clone/dedupe operation.
161 * Used when logging an inode to know if there are shared extents that
162 * need special care when logging checksum items, to avoid duplicate
163 * checksum items in a log (which can lead to a corruption where we end
164 * up with missing checksum ranges after log replay).
165 * Protected by the vfs inode lock.
167 u64 last_reflink_trans;
170 * Number of bytes outstanding that are going to need csums. This is
171 * used in ENOSPC accounting.
175 /* flags field from the on disk inode */
179 * Counters to keep track of the number of extent item's we may use due
180 * to delalloc and such. outstanding_extents is the number of extent
181 * items we think we'll end up using, and reserved_extents is the number
182 * of extent items we've reserved metadata for.
184 unsigned outstanding_extents;
186 struct btrfs_block_rsv block_rsv;
189 * Cached values of inode properties
191 unsigned prop_compress; /* per-file compression algorithm */
193 * Force compression on the file using the defrag ioctl, could be
194 * different from prop_compress and takes precedence if set
196 unsigned defrag_compress;
198 struct btrfs_delayed_node *delayed_node;
200 /* File creation time. */
201 struct timespec64 i_otime;
203 /* Hook into fs_info->delayed_iputs */
204 struct list_head delayed_iput;
207 * To avoid races between lockless (i_mutex not held) direct IO writes
208 * and concurrent fsync requests. Direct IO writes must acquire read
209 * access on this semaphore for creating an extent map and its
210 * corresponding ordered extent. The fast fsync path must acquire write
211 * access on this semaphore before it collects ordered extents and
214 struct rw_semaphore dio_sem;
216 struct inode vfs_inode;
219 static inline u32 btrfs_inode_sectorsize(const struct btrfs_inode *inode)
221 return inode->root->fs_info->sectorsize;
224 static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
226 return container_of(inode, struct btrfs_inode, vfs_inode);
229 static inline unsigned long btrfs_inode_hash(u64 objectid,
230 const struct btrfs_root *root)
232 u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
234 #if BITS_PER_LONG == 32
235 h = (h >> 32) ^ (h & 0xffffffff);
238 return (unsigned long)h;
241 static inline void btrfs_insert_inode_hash(struct inode *inode)
243 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
245 __insert_inode_hash(inode, h);
248 static inline u64 btrfs_ino(const struct btrfs_inode *inode)
250 u64 ino = inode->location.objectid;
254 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
256 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
257 ino = inode->vfs_inode.i_ino;
261 static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
263 i_size_write(&inode->vfs_inode, size);
264 inode->disk_i_size = size;
267 static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
269 struct btrfs_root *root = inode->root;
271 if (root == root->fs_info->tree_root &&
272 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
274 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
279 static inline bool is_data_inode(struct inode *inode)
281 return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
284 static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
287 lockdep_assert_held(&inode->lock);
288 inode->outstanding_extents += mod;
289 if (btrfs_is_free_space_inode(inode))
291 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
295 static inline int btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
299 spin_lock(&inode->lock);
300 if (inode->logged_trans == generation &&
301 inode->last_sub_trans <= inode->last_log_commit &&
302 inode->last_sub_trans <= inode->root->last_log_commit) {
304 * After a ranged fsync we might have left some extent maps
305 * (that fall outside the fsync's range). So return false
306 * here if the list isn't empty, to make sure btrfs_log_inode()
307 * will be called and process those extent maps.
310 if (list_empty(&inode->extent_tree.modified_extents))
313 spin_unlock(&inode->lock);
317 struct btrfs_dio_private {
324 * References to this structure. There is one reference per in-flight
325 * bio plus one while we're still setting up.
329 /* dio_bio came from fs/direct-io.c */
332 /* Array of checksums */
336 /* Array of bytes with variable length, hexadecimal format 0x1234 */
337 #define CSUM_FMT "0x%*phN"
338 #define CSUM_FMT_VALUE(size, bytes) size, bytes
340 static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
341 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
343 struct btrfs_root *root = inode->root;
344 struct btrfs_super_block *sb = root->fs_info->super_copy;
345 const u16 csum_size = btrfs_super_csum_size(sb);
347 /* Output minus objectid, which is more meaningful */
348 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
349 btrfs_warn_rl(root->fs_info,
350 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
351 root->root_key.objectid, btrfs_ino(inode),
353 CSUM_FMT_VALUE(csum_size, csum),
354 CSUM_FMT_VALUE(csum_size, csum_expected),
357 btrfs_warn_rl(root->fs_info,
358 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
359 root->root_key.objectid, btrfs_ino(inode),
361 CSUM_FMT_VALUE(csum_size, csum),
362 CSUM_FMT_VALUE(csum_size, csum_expected),