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,
39 * Set and used when logging an inode and it serves to signal that an
40 * inode does not have xattrs, so subsequent fsyncs can avoid searching
41 * for xattrs to log. This bit must be cleared whenever a xattr is added
44 BTRFS_INODE_NO_XATTRS,
46 * Set when we are in a context where we need to start a transaction and
47 * have dirty pages with the respective file range locked. This is to
48 * ensure that when reserving space for the transaction, if we are low
49 * on available space and need to flush delalloc, we will not flush
50 * delalloc for this inode, because that could result in a deadlock (on
51 * the file range, inode's io_tree).
53 BTRFS_INODE_NO_DELALLOC_FLUSH,
56 /* in memory btrfs inode */
58 /* which subvolume this inode belongs to */
59 struct btrfs_root *root;
61 /* key used to find this inode on disk. This is used by the code
62 * to read in roots of subvolumes
64 struct btrfs_key location;
67 * Lock for counters and all fields used to determine if the inode is in
68 * the log or not (last_trans, last_sub_trans, last_log_commit,
73 /* the extent_tree has caches of all the extent mappings to disk */
74 struct extent_map_tree extent_tree;
76 /* the io_tree does range state (DIRTY, LOCKED etc) */
77 struct extent_io_tree io_tree;
79 /* special utility tree used to record which mirrors have already been
80 * tried when checksums fail for a given block
82 struct extent_io_tree io_failure_tree;
85 * Keep track of where the inode has extent items mapped in order to
86 * make sure the i_size adjustments are accurate
88 struct extent_io_tree file_extent_tree;
90 /* held while logging the inode in tree-log.c */
91 struct mutex log_mutex;
93 /* used to order data wrt metadata */
94 struct btrfs_ordered_inode_tree ordered_tree;
96 /* list of all the delalloc inodes in the FS. There are times we need
97 * to write all the delalloc pages to disk, and this list is used
100 struct list_head delalloc_inodes;
102 /* node for the red-black tree that links inodes in subvolume root */
103 struct rb_node rb_node;
105 unsigned long runtime_flags;
107 /* Keep track of who's O_SYNC/fsyncing currently */
108 atomic_t sync_writers;
110 /* full 64 bit generation number, struct vfs_inode doesn't have a big
111 * enough field for this.
116 * transid of the trans_handle that last modified this inode
121 * transid that last logged this inode
126 * log transid when this inode was last modified
130 /* a local copy of root's last_log_commit */
133 /* total number of bytes pending delalloc, used by stat to calc the
134 * real block usage of the file
139 * Total number of bytes pending delalloc that fall within a file
140 * range that is either a hole or beyond EOF (and no prealloc extent
141 * exists in the range). This is always <= delalloc_bytes.
143 u64 new_delalloc_bytes;
146 * total number of bytes pending defrag, used by stat to check whether
152 * the size of the file stored in the metadata on disk. data=ordered
153 * means the in-memory i_size might be larger than the size on disk
154 * because not all the blocks are written yet.
159 * if this is a directory then index_cnt is the counter for the index
160 * number for new files that are created
164 /* Cache the directory index number to speed the dir/file remove */
167 /* the fsync log has some corner cases that mean we have to check
168 * directories to see if any unlinks have been done before
169 * the directory was logged. See tree-log.c for all the
172 u64 last_unlink_trans;
175 * The id/generation of the last transaction where this inode was
176 * either the source or the destination of a clone/dedupe operation.
177 * Used when logging an inode to know if there are shared extents that
178 * need special care when logging checksum items, to avoid duplicate
179 * checksum items in a log (which can lead to a corruption where we end
180 * up with missing checksum ranges after log replay).
181 * Protected by the vfs inode lock.
183 u64 last_reflink_trans;
186 * Number of bytes outstanding that are going to need csums. This is
187 * used in ENOSPC accounting.
191 /* flags field from the on disk inode */
195 * Counters to keep track of the number of extent item's we may use due
196 * to delalloc and such. outstanding_extents is the number of extent
197 * items we think we'll end up using, and reserved_extents is the number
198 * of extent items we've reserved metadata for.
200 unsigned outstanding_extents;
202 struct btrfs_block_rsv block_rsv;
205 * Cached values of inode properties
207 unsigned prop_compress; /* per-file compression algorithm */
209 * Force compression on the file using the defrag ioctl, could be
210 * different from prop_compress and takes precedence if set
212 unsigned defrag_compress;
214 struct btrfs_delayed_node *delayed_node;
216 /* File creation time. */
217 struct timespec64 i_otime;
219 /* Hook into fs_info->delayed_iputs */
220 struct list_head delayed_iput;
223 * To avoid races between lockless (i_mutex not held) direct IO writes
224 * and concurrent fsync requests. Direct IO writes must acquire read
225 * access on this semaphore for creating an extent map and its
226 * corresponding ordered extent. The fast fsync path must acquire write
227 * access on this semaphore before it collects ordered extents and
230 struct rw_semaphore dio_sem;
232 struct inode vfs_inode;
235 static inline u32 btrfs_inode_sectorsize(const struct btrfs_inode *inode)
237 return inode->root->fs_info->sectorsize;
240 static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
242 return container_of(inode, struct btrfs_inode, vfs_inode);
245 static inline unsigned long btrfs_inode_hash(u64 objectid,
246 const struct btrfs_root *root)
248 u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
250 #if BITS_PER_LONG == 32
251 h = (h >> 32) ^ (h & 0xffffffff);
254 return (unsigned long)h;
257 static inline void btrfs_insert_inode_hash(struct inode *inode)
259 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
261 __insert_inode_hash(inode, h);
264 static inline u64 btrfs_ino(const struct btrfs_inode *inode)
266 u64 ino = inode->location.objectid;
270 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
272 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
273 ino = inode->vfs_inode.i_ino;
277 static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
279 i_size_write(&inode->vfs_inode, size);
280 inode->disk_i_size = size;
283 static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
285 struct btrfs_root *root = inode->root;
287 if (root == root->fs_info->tree_root &&
288 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
290 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
295 static inline bool is_data_inode(struct inode *inode)
297 return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
300 static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
303 lockdep_assert_held(&inode->lock);
304 inode->outstanding_extents += mod;
305 if (btrfs_is_free_space_inode(inode))
307 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
312 * Called every time after doing a buffered, direct IO or memory mapped write.
314 * This is to ensure that if we write to a file that was previously fsynced in
315 * the current transaction, then try to fsync it again in the same transaction,
316 * we will know that there were changes in the file and that it needs to be
319 static inline void btrfs_set_inode_last_sub_trans(struct btrfs_inode *inode)
321 spin_lock(&inode->lock);
322 inode->last_sub_trans = inode->root->log_transid;
323 spin_unlock(&inode->lock);
326 static inline int btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
330 spin_lock(&inode->lock);
331 if (inode->logged_trans == generation &&
332 inode->last_sub_trans <= inode->last_log_commit &&
333 inode->last_sub_trans <= inode->root->last_log_commit) {
335 * After a ranged fsync we might have left some extent maps
336 * (that fall outside the fsync's range). So return false
337 * here if the list isn't empty, to make sure btrfs_log_inode()
338 * will be called and process those extent maps.
341 if (list_empty(&inode->extent_tree.modified_extents))
344 spin_unlock(&inode->lock);
348 struct btrfs_dio_private {
355 * References to this structure. There is one reference per in-flight
356 * bio plus one while we're still setting up.
360 /* dio_bio came from fs/direct-io.c */
363 /* Array of checksums */
367 /* Array of bytes with variable length, hexadecimal format 0x1234 */
368 #define CSUM_FMT "0x%*phN"
369 #define CSUM_FMT_VALUE(size, bytes) size, bytes
371 static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
372 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
374 struct btrfs_root *root = inode->root;
375 struct btrfs_super_block *sb = root->fs_info->super_copy;
376 const u16 csum_size = btrfs_super_csum_size(sb);
378 /* Output minus objectid, which is more meaningful */
379 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
380 btrfs_warn_rl(root->fs_info,
381 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
382 root->root_key.objectid, btrfs_ino(inode),
384 CSUM_FMT_VALUE(csum_size, csum),
385 CSUM_FMT_VALUE(csum_size, csum_expected),
388 btrfs_warn_rl(root->fs_info,
389 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
390 root->root_key.objectid, btrfs_ino(inode),
392 CSUM_FMT_VALUE(csum_size, csum),
393 CSUM_FMT_VALUE(csum_size, csum_expected),