Loading...
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#ifndef BTRFS_INODE_H
7#define BTRFS_INODE_H
8
9#include <linux/hash.h>
10#include "extent_map.h"
11#include "extent_io.h"
12#include "ordered-data.h"
13#include "delayed-inode.h"
14
15/*
16 * ordered_data_close is set by truncate when a file that used
17 * to have good data has been truncated to zero. When it is set
18 * the btrfs file release call will add this inode to the
19 * ordered operations list so that we make sure to flush out any
20 * new data the application may have written before commit.
21 */
22#define BTRFS_INODE_ORDERED_DATA_CLOSE 0
23#define BTRFS_INODE_ORPHAN_META_RESERVED 1
24#define BTRFS_INODE_DUMMY 2
25#define BTRFS_INODE_IN_DEFRAG 3
26#define BTRFS_INODE_HAS_ORPHAN_ITEM 4
27#define BTRFS_INODE_HAS_ASYNC_EXTENT 5
28#define BTRFS_INODE_NEEDS_FULL_SYNC 6
29#define BTRFS_INODE_COPY_EVERYTHING 7
30#define BTRFS_INODE_IN_DELALLOC_LIST 8
31#define BTRFS_INODE_READDIO_NEED_LOCK 9
32#define BTRFS_INODE_HAS_PROPS 10
33
34/* in memory btrfs inode */
35struct btrfs_inode {
36 /* which subvolume this inode belongs to */
37 struct btrfs_root *root;
38
39 /* key used to find this inode on disk. This is used by the code
40 * to read in roots of subvolumes
41 */
42 struct btrfs_key location;
43
44 /*
45 * Lock for counters and all fields used to determine if the inode is in
46 * the log or not (last_trans, last_sub_trans, last_log_commit,
47 * logged_trans).
48 */
49 spinlock_t lock;
50
51 /* the extent_tree has caches of all the extent mappings to disk */
52 struct extent_map_tree extent_tree;
53
54 /* the io_tree does range state (DIRTY, LOCKED etc) */
55 struct extent_io_tree io_tree;
56
57 /* special utility tree used to record which mirrors have already been
58 * tried when checksums fail for a given block
59 */
60 struct extent_io_tree io_failure_tree;
61
62 /* held while logging the inode in tree-log.c */
63 struct mutex log_mutex;
64
65 /* held while doing delalloc reservations */
66 struct mutex delalloc_mutex;
67
68 /* used to order data wrt metadata */
69 struct btrfs_ordered_inode_tree ordered_tree;
70
71 /* list of all the delalloc inodes in the FS. There are times we need
72 * to write all the delalloc pages to disk, and this list is used
73 * to walk them all.
74 */
75 struct list_head delalloc_inodes;
76
77 /* node for the red-black tree that links inodes in subvolume root */
78 struct rb_node rb_node;
79
80 unsigned long runtime_flags;
81
82 /* Keep track of who's O_SYNC/fsyncing currently */
83 atomic_t sync_writers;
84
85 /* full 64 bit generation number, struct vfs_inode doesn't have a big
86 * enough field for this.
87 */
88 u64 generation;
89
90 /*
91 * transid of the trans_handle that last modified this inode
92 */
93 u64 last_trans;
94
95 /*
96 * transid that last logged this inode
97 */
98 u64 logged_trans;
99
100 /*
101 * log transid when this inode was last modified
102 */
103 int last_sub_trans;
104
105 /* a local copy of root's last_log_commit */
106 int last_log_commit;
107
108 /* total number of bytes pending delalloc, used by stat to calc the
109 * real block usage of the file
110 */
111 u64 delalloc_bytes;
112
113 /*
114 * Total number of bytes pending delalloc that fall within a file
115 * range that is either a hole or beyond EOF (and no prealloc extent
116 * exists in the range). This is always <= delalloc_bytes.
117 */
118 u64 new_delalloc_bytes;
119
120 /*
121 * total number of bytes pending defrag, used by stat to check whether
122 * it needs COW.
123 */
124 u64 defrag_bytes;
125
126 /*
127 * the size of the file stored in the metadata on disk. data=ordered
128 * means the in-memory i_size might be larger than the size on disk
129 * because not all the blocks are written yet.
130 */
131 u64 disk_i_size;
132
133 /*
134 * if this is a directory then index_cnt is the counter for the index
135 * number for new files that are created
136 */
137 u64 index_cnt;
138
139 /* Cache the directory index number to speed the dir/file remove */
140 u64 dir_index;
141
142 /* the fsync log has some corner cases that mean we have to check
143 * directories to see if any unlinks have been done before
144 * the directory was logged. See tree-log.c for all the
145 * details
146 */
147 u64 last_unlink_trans;
148
149 /*
150 * Number of bytes outstanding that are going to need csums. This is
151 * used in ENOSPC accounting.
152 */
153 u64 csum_bytes;
154
155 /* flags field from the on disk inode */
156 u32 flags;
157
158 /*
159 * Counters to keep track of the number of extent item's we may use due
160 * to delalloc and such. outstanding_extents is the number of extent
161 * items we think we'll end up using, and reserved_extents is the number
162 * of extent items we've reserved metadata for.
163 */
164 unsigned outstanding_extents;
165
166 struct btrfs_block_rsv block_rsv;
167
168 /*
169 * Cached values of inode properties
170 */
171 unsigned prop_compress; /* per-file compression algorithm */
172 /*
173 * Force compression on the file using the defrag ioctl, could be
174 * different from prop_compress and takes precedence if set
175 */
176 unsigned defrag_compress;
177
178 struct btrfs_delayed_node *delayed_node;
179
180 /* File creation time. */
181 struct timespec i_otime;
182
183 /* Hook into fs_info->delayed_iputs */
184 struct list_head delayed_iput;
185
186 /*
187 * To avoid races between lockless (i_mutex not held) direct IO writes
188 * and concurrent fsync requests. Direct IO writes must acquire read
189 * access on this semaphore for creating an extent map and its
190 * corresponding ordered extent. The fast fsync path must acquire write
191 * access on this semaphore before it collects ordered extents and
192 * extent maps.
193 */
194 struct rw_semaphore dio_sem;
195
196 struct inode vfs_inode;
197};
198
199extern unsigned char btrfs_filetype_table[];
200
201static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
202{
203 return container_of(inode, struct btrfs_inode, vfs_inode);
204}
205
206static inline unsigned long btrfs_inode_hash(u64 objectid,
207 const struct btrfs_root *root)
208{
209 u64 h = objectid ^ (root->objectid * GOLDEN_RATIO_PRIME);
210
211#if BITS_PER_LONG == 32
212 h = (h >> 32) ^ (h & 0xffffffff);
213#endif
214
215 return (unsigned long)h;
216}
217
218static inline void btrfs_insert_inode_hash(struct inode *inode)
219{
220 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
221
222 __insert_inode_hash(inode, h);
223}
224
225static inline u64 btrfs_ino(const struct btrfs_inode *inode)
226{
227 u64 ino = inode->location.objectid;
228
229 /*
230 * !ino: btree_inode
231 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
232 */
233 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
234 ino = inode->vfs_inode.i_ino;
235 return ino;
236}
237
238static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
239{
240 i_size_write(&inode->vfs_inode, size);
241 inode->disk_i_size = size;
242}
243
244static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
245{
246 struct btrfs_root *root = inode->root;
247
248 if (root == root->fs_info->tree_root &&
249 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
250 return true;
251 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
252 return true;
253 return false;
254}
255
256static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
257 int mod)
258{
259 lockdep_assert_held(&inode->lock);
260 inode->outstanding_extents += mod;
261 if (btrfs_is_free_space_inode(inode))
262 return;
263 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
264 mod);
265}
266
267static inline int btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
268{
269 int ret = 0;
270
271 spin_lock(&inode->lock);
272 if (inode->logged_trans == generation &&
273 inode->last_sub_trans <= inode->last_log_commit &&
274 inode->last_sub_trans <= inode->root->last_log_commit) {
275 /*
276 * After a ranged fsync we might have left some extent maps
277 * (that fall outside the fsync's range). So return false
278 * here if the list isn't empty, to make sure btrfs_log_inode()
279 * will be called and process those extent maps.
280 */
281 smp_mb();
282 if (list_empty(&inode->extent_tree.modified_extents))
283 ret = 1;
284 }
285 spin_unlock(&inode->lock);
286 return ret;
287}
288
289#define BTRFS_DIO_ORIG_BIO_SUBMITTED 0x1
290
291struct btrfs_dio_private {
292 struct inode *inode;
293 unsigned long flags;
294 u64 logical_offset;
295 u64 disk_bytenr;
296 u64 bytes;
297 void *private;
298
299 /* number of bios pending for this dio */
300 atomic_t pending_bios;
301
302 /* IO errors */
303 int errors;
304
305 /* orig_bio is our btrfs_io_bio */
306 struct bio *orig_bio;
307
308 /* dio_bio came from fs/direct-io.c */
309 struct bio *dio_bio;
310
311 /*
312 * The original bio may be split to several sub-bios, this is
313 * done during endio of sub-bios
314 */
315 blk_status_t (*subio_endio)(struct inode *, struct btrfs_io_bio *,
316 blk_status_t);
317};
318
319/*
320 * Disable DIO read nolock optimization, so new dio readers will be forced
321 * to grab i_mutex. It is used to avoid the endless truncate due to
322 * nonlocked dio read.
323 */
324static inline void btrfs_inode_block_unlocked_dio(struct btrfs_inode *inode)
325{
326 set_bit(BTRFS_INODE_READDIO_NEED_LOCK, &inode->runtime_flags);
327 smp_mb();
328}
329
330static inline void btrfs_inode_resume_unlocked_dio(struct btrfs_inode *inode)
331{
332 smp_mb__before_atomic();
333 clear_bit(BTRFS_INODE_READDIO_NEED_LOCK, &inode->runtime_flags);
334}
335
336static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
337 u64 logical_start, u32 csum, u32 csum_expected, int mirror_num)
338{
339 struct btrfs_root *root = inode->root;
340
341 /* Output minus objectid, which is more meaningful */
342 if (root->objectid >= BTRFS_LAST_FREE_OBJECTID)
343 btrfs_warn_rl(root->fs_info,
344 "csum failed root %lld ino %lld off %llu csum 0x%08x expected csum 0x%08x mirror %d",
345 root->objectid, btrfs_ino(inode),
346 logical_start, csum, csum_expected, mirror_num);
347 else
348 btrfs_warn_rl(root->fs_info,
349 "csum failed root %llu ino %llu off %llu csum 0x%08x expected csum 0x%08x mirror %d",
350 root->objectid, btrfs_ino(inode),
351 logical_start, csum, csum_expected, mirror_num);
352}
353
354#endif
1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * Copyright (C) 2007 Oracle. All rights reserved.
4 */
5
6#ifndef BTRFS_INODE_H
7#define BTRFS_INODE_H
8
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"
15
16/*
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.
22 */
23enum {
24 BTRFS_INODE_FLUSH_ON_CLOSE,
25 BTRFS_INODE_DUMMY,
26 BTRFS_INODE_IN_DEFRAG,
27 BTRFS_INODE_HAS_ASYNC_EXTENT,
28 /*
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).
32 */
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,
38 /*
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
42 * to an inode.
43 */
44 BTRFS_INODE_NO_XATTRS,
45 /*
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).
52 */
53 BTRFS_INODE_NO_DELALLOC_FLUSH,
54};
55
56/* in memory btrfs inode */
57struct btrfs_inode {
58 /* which subvolume this inode belongs to */
59 struct btrfs_root *root;
60
61 /* key used to find this inode on disk. This is used by the code
62 * to read in roots of subvolumes
63 */
64 struct btrfs_key location;
65
66 /*
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,
69 * logged_trans), to access/update new_delalloc_bytes and to update the
70 * VFS' inode number of bytes used.
71 */
72 spinlock_t lock;
73
74 /* the extent_tree has caches of all the extent mappings to disk */
75 struct extent_map_tree extent_tree;
76
77 /* the io_tree does range state (DIRTY, LOCKED etc) */
78 struct extent_io_tree io_tree;
79
80 /* special utility tree used to record which mirrors have already been
81 * tried when checksums fail for a given block
82 */
83 struct extent_io_tree io_failure_tree;
84
85 /*
86 * Keep track of where the inode has extent items mapped in order to
87 * make sure the i_size adjustments are accurate
88 */
89 struct extent_io_tree file_extent_tree;
90
91 /* held while logging the inode in tree-log.c */
92 struct mutex log_mutex;
93
94 /* used to order data wrt metadata */
95 struct btrfs_ordered_inode_tree ordered_tree;
96
97 /* list of all the delalloc inodes in the FS. There are times we need
98 * to write all the delalloc pages to disk, and this list is used
99 * to walk them all.
100 */
101 struct list_head delalloc_inodes;
102
103 /* node for the red-black tree that links inodes in subvolume root */
104 struct rb_node rb_node;
105
106 unsigned long runtime_flags;
107
108 /* Keep track of who's O_SYNC/fsyncing currently */
109 atomic_t sync_writers;
110
111 /* full 64 bit generation number, struct vfs_inode doesn't have a big
112 * enough field for this.
113 */
114 u64 generation;
115
116 /*
117 * transid of the trans_handle that last modified this inode
118 */
119 u64 last_trans;
120
121 /*
122 * transid that last logged this inode
123 */
124 u64 logged_trans;
125
126 /*
127 * log transid when this inode was last modified
128 */
129 int last_sub_trans;
130
131 /* a local copy of root's last_log_commit */
132 int last_log_commit;
133
134 /* total number of bytes pending delalloc, used by stat to calc the
135 * real block usage of the file
136 */
137 u64 delalloc_bytes;
138
139 /*
140 * Total number of bytes pending delalloc that fall within a file
141 * range that is either a hole or beyond EOF (and no prealloc extent
142 * exists in the range). This is always <= delalloc_bytes.
143 */
144 u64 new_delalloc_bytes;
145
146 /*
147 * total number of bytes pending defrag, used by stat to check whether
148 * it needs COW.
149 */
150 u64 defrag_bytes;
151
152 /*
153 * the size of the file stored in the metadata on disk. data=ordered
154 * means the in-memory i_size might be larger than the size on disk
155 * because not all the blocks are written yet.
156 */
157 u64 disk_i_size;
158
159 /*
160 * if this is a directory then index_cnt is the counter for the index
161 * number for new files that are created
162 */
163 u64 index_cnt;
164
165 /* Cache the directory index number to speed the dir/file remove */
166 u64 dir_index;
167
168 /* the fsync log has some corner cases that mean we have to check
169 * directories to see if any unlinks have been done before
170 * the directory was logged. See tree-log.c for all the
171 * details
172 */
173 u64 last_unlink_trans;
174
175 /*
176 * The id/generation of the last transaction where this inode was
177 * either the source or the destination of a clone/dedupe operation.
178 * Used when logging an inode to know if there are shared extents that
179 * need special care when logging checksum items, to avoid duplicate
180 * checksum items in a log (which can lead to a corruption where we end
181 * up with missing checksum ranges after log replay).
182 * Protected by the vfs inode lock.
183 */
184 u64 last_reflink_trans;
185
186 /*
187 * Number of bytes outstanding that are going to need csums. This is
188 * used in ENOSPC accounting.
189 */
190 u64 csum_bytes;
191
192 /* flags field from the on disk inode */
193 u32 flags;
194
195 /*
196 * Counters to keep track of the number of extent item's we may use due
197 * to delalloc and such. outstanding_extents is the number of extent
198 * items we think we'll end up using, and reserved_extents is the number
199 * of extent items we've reserved metadata for.
200 */
201 unsigned outstanding_extents;
202
203 struct btrfs_block_rsv block_rsv;
204
205 /*
206 * Cached values of inode properties
207 */
208 unsigned prop_compress; /* per-file compression algorithm */
209 /*
210 * Force compression on the file using the defrag ioctl, could be
211 * different from prop_compress and takes precedence if set
212 */
213 unsigned defrag_compress;
214
215 struct btrfs_delayed_node *delayed_node;
216
217 /* File creation time. */
218 struct timespec64 i_otime;
219
220 /* Hook into fs_info->delayed_iputs */
221 struct list_head delayed_iput;
222
223 struct rw_semaphore i_mmap_lock;
224 struct inode vfs_inode;
225};
226
227static inline u32 btrfs_inode_sectorsize(const struct btrfs_inode *inode)
228{
229 return inode->root->fs_info->sectorsize;
230}
231
232static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
233{
234 return container_of(inode, struct btrfs_inode, vfs_inode);
235}
236
237static inline unsigned long btrfs_inode_hash(u64 objectid,
238 const struct btrfs_root *root)
239{
240 u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
241
242#if BITS_PER_LONG == 32
243 h = (h >> 32) ^ (h & 0xffffffff);
244#endif
245
246 return (unsigned long)h;
247}
248
249static inline void btrfs_insert_inode_hash(struct inode *inode)
250{
251 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
252
253 __insert_inode_hash(inode, h);
254}
255
256static inline u64 btrfs_ino(const struct btrfs_inode *inode)
257{
258 u64 ino = inode->location.objectid;
259
260 /*
261 * !ino: btree_inode
262 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
263 */
264 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
265 ino = inode->vfs_inode.i_ino;
266 return ino;
267}
268
269static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
270{
271 i_size_write(&inode->vfs_inode, size);
272 inode->disk_i_size = size;
273}
274
275static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
276{
277 struct btrfs_root *root = inode->root;
278
279 if (root == root->fs_info->tree_root &&
280 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
281 return true;
282 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
283 return true;
284 return false;
285}
286
287static inline bool is_data_inode(struct inode *inode)
288{
289 return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
290}
291
292static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
293 int mod)
294{
295 lockdep_assert_held(&inode->lock);
296 inode->outstanding_extents += mod;
297 if (btrfs_is_free_space_inode(inode))
298 return;
299 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
300 mod);
301}
302
303/*
304 * Called every time after doing a buffered, direct IO or memory mapped write.
305 *
306 * This is to ensure that if we write to a file that was previously fsynced in
307 * the current transaction, then try to fsync it again in the same transaction,
308 * we will know that there were changes in the file and that it needs to be
309 * logged.
310 */
311static inline void btrfs_set_inode_last_sub_trans(struct btrfs_inode *inode)
312{
313 spin_lock(&inode->lock);
314 inode->last_sub_trans = inode->root->log_transid;
315 spin_unlock(&inode->lock);
316}
317
318static inline bool btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
319{
320 bool ret = false;
321
322 spin_lock(&inode->lock);
323 if (inode->logged_trans == generation &&
324 inode->last_sub_trans <= inode->last_log_commit &&
325 inode->last_sub_trans <= inode->root->last_log_commit)
326 ret = true;
327 spin_unlock(&inode->lock);
328 return ret;
329}
330
331struct btrfs_dio_private {
332 struct inode *inode;
333 u64 logical_offset;
334 u64 disk_bytenr;
335 /* Used for bio::bi_size */
336 u32 bytes;
337
338 /*
339 * References to this structure. There is one reference per in-flight
340 * bio plus one while we're still setting up.
341 */
342 refcount_t refs;
343
344 /* dio_bio came from fs/direct-io.c */
345 struct bio *dio_bio;
346
347 /* Array of checksums */
348 u8 csums[];
349};
350
351/* Array of bytes with variable length, hexadecimal format 0x1234 */
352#define CSUM_FMT "0x%*phN"
353#define CSUM_FMT_VALUE(size, bytes) size, bytes
354
355static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
356 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
357{
358 struct btrfs_root *root = inode->root;
359 const u32 csum_size = root->fs_info->csum_size;
360
361 /* Output minus objectid, which is more meaningful */
362 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
363 btrfs_warn_rl(root->fs_info,
364"csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
365 root->root_key.objectid, btrfs_ino(inode),
366 logical_start,
367 CSUM_FMT_VALUE(csum_size, csum),
368 CSUM_FMT_VALUE(csum_size, csum_expected),
369 mirror_num);
370 else
371 btrfs_warn_rl(root->fs_info,
372"csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
373 root->root_key.objectid, btrfs_ino(inode),
374 logical_start,
375 CSUM_FMT_VALUE(csum_size, csum),
376 CSUM_FMT_VALUE(csum_size, csum_expected),
377 mirror_num);
378}
379
380#endif