Loading...
1/*
2 * Copyright (C) 2007 Oracle. All rights reserved.
3 *
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
17 */
18
19#ifndef __BTRFS_I__
20#define __BTRFS_I__
21
22#include <linux/hash.h>
23#include "extent_map.h"
24#include "extent_io.h"
25#include "ordered-data.h"
26#include "delayed-inode.h"
27
28/*
29 * ordered_data_close is set by truncate when a file that used
30 * to have good data has been truncated to zero. When it is set
31 * the btrfs file release call will add this inode to the
32 * ordered operations list so that we make sure to flush out any
33 * new data the application may have written before commit.
34 */
35#define BTRFS_INODE_ORDERED_DATA_CLOSE 0
36#define BTRFS_INODE_ORPHAN_META_RESERVED 1
37#define BTRFS_INODE_DUMMY 2
38#define BTRFS_INODE_IN_DEFRAG 3
39#define BTRFS_INODE_DELALLOC_META_RESERVED 4
40#define BTRFS_INODE_HAS_ORPHAN_ITEM 5
41#define BTRFS_INODE_HAS_ASYNC_EXTENT 6
42#define BTRFS_INODE_NEEDS_FULL_SYNC 7
43#define BTRFS_INODE_COPY_EVERYTHING 8
44#define BTRFS_INODE_IN_DELALLOC_LIST 9
45#define BTRFS_INODE_READDIO_NEED_LOCK 10
46#define BTRFS_INODE_HAS_PROPS 11
47
48/* in memory btrfs inode */
49struct btrfs_inode {
50 /* which subvolume this inode belongs to */
51 struct btrfs_root *root;
52
53 /* key used to find this inode on disk. This is used by the code
54 * to read in roots of subvolumes
55 */
56 struct btrfs_key location;
57
58 /*
59 * Lock for counters and all fields used to determine if the inode is in
60 * the log or not (last_trans, last_sub_trans, last_log_commit,
61 * logged_trans).
62 */
63 spinlock_t lock;
64
65 /* the extent_tree has caches of all the extent mappings to disk */
66 struct extent_map_tree extent_tree;
67
68 /* the io_tree does range state (DIRTY, LOCKED etc) */
69 struct extent_io_tree io_tree;
70
71 /* special utility tree used to record which mirrors have already been
72 * tried when checksums fail for a given block
73 */
74 struct extent_io_tree io_failure_tree;
75
76 /* held while logging the inode in tree-log.c */
77 struct mutex log_mutex;
78
79 /* held while doing delalloc reservations */
80 struct mutex delalloc_mutex;
81
82 /* used to order data wrt metadata */
83 struct btrfs_ordered_inode_tree ordered_tree;
84
85 /* list of all the delalloc inodes in the FS. There are times we need
86 * to write all the delalloc pages to disk, and this list is used
87 * to walk them all.
88 */
89 struct list_head delalloc_inodes;
90
91 /* node for the red-black tree that links inodes in subvolume root */
92 struct rb_node rb_node;
93
94 unsigned long runtime_flags;
95
96 /* Keep track of who's O_SYNC/fsyncing currently */
97 atomic_t sync_writers;
98
99 /* full 64 bit generation number, struct vfs_inode doesn't have a big
100 * enough field for this.
101 */
102 u64 generation;
103
104 /*
105 * transid of the trans_handle that last modified this inode
106 */
107 u64 last_trans;
108
109 /*
110 * transid that last logged this inode
111 */
112 u64 logged_trans;
113
114 /*
115 * log transid when this inode was last modified
116 */
117 int last_sub_trans;
118
119 /* a local copy of root's last_log_commit */
120 int last_log_commit;
121
122 /* total number of bytes pending delalloc, used by stat to calc the
123 * real block usage of the file
124 */
125 u64 delalloc_bytes;
126
127 /*
128 * total number of bytes pending defrag, used by stat to check whether
129 * it needs COW.
130 */
131 u64 defrag_bytes;
132
133 /*
134 * the size of the file stored in the metadata on disk. data=ordered
135 * means the in-memory i_size might be larger than the size on disk
136 * because not all the blocks are written yet.
137 */
138 u64 disk_i_size;
139
140 /*
141 * if this is a directory then index_cnt is the counter for the index
142 * number for new files that are created
143 */
144 u64 index_cnt;
145
146 /* Cache the directory index number to speed the dir/file remove */
147 u64 dir_index;
148
149 /* the fsync log has some corner cases that mean we have to check
150 * directories to see if any unlinks have been done before
151 * the directory was logged. See tree-log.c for all the
152 * details
153 */
154 u64 last_unlink_trans;
155
156 /*
157 * Number of bytes outstanding that are going to need csums. This is
158 * used in ENOSPC accounting.
159 */
160 u64 csum_bytes;
161
162 /* flags field from the on disk inode */
163 u32 flags;
164
165 /*
166 * Counters to keep track of the number of extent item's we may use due
167 * to delalloc and such. outstanding_extents is the number of extent
168 * items we think we'll end up using, and reserved_extents is the number
169 * of extent items we've reserved metadata for.
170 */
171 unsigned outstanding_extents;
172 unsigned reserved_extents;
173
174 /*
175 * always compress this one file
176 */
177 unsigned force_compress;
178
179 struct btrfs_delayed_node *delayed_node;
180
181 /* File creation time. */
182 struct timespec i_otime;
183
184 /* Hook into fs_info->delayed_iputs */
185 struct list_head delayed_iput;
186 long delayed_iput_count;
187
188 /*
189 * To avoid races between lockless (i_mutex not held) direct IO writes
190 * and concurrent fsync requests. Direct IO writes must acquire read
191 * access on this semaphore for creating an extent map and its
192 * corresponding ordered extent. The fast fsync path must acquire write
193 * access on this semaphore before it collects ordered extents and
194 * extent maps.
195 */
196 struct rw_semaphore dio_sem;
197
198 struct inode vfs_inode;
199};
200
201extern unsigned char btrfs_filetype_table[];
202
203static inline struct btrfs_inode *BTRFS_I(struct inode *inode)
204{
205 return container_of(inode, struct btrfs_inode, vfs_inode);
206}
207
208static inline unsigned long btrfs_inode_hash(u64 objectid,
209 const struct btrfs_root *root)
210{
211 u64 h = objectid ^ (root->objectid * GOLDEN_RATIO_PRIME);
212
213#if BITS_PER_LONG == 32
214 h = (h >> 32) ^ (h & 0xffffffff);
215#endif
216
217 return (unsigned long)h;
218}
219
220static inline void btrfs_insert_inode_hash(struct inode *inode)
221{
222 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
223
224 __insert_inode_hash(inode, h);
225}
226
227static inline u64 btrfs_ino(struct inode *inode)
228{
229 u64 ino = BTRFS_I(inode)->location.objectid;
230
231 /*
232 * !ino: btree_inode
233 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
234 */
235 if (!ino || BTRFS_I(inode)->location.type == BTRFS_ROOT_ITEM_KEY)
236 ino = inode->i_ino;
237 return ino;
238}
239
240static inline void btrfs_i_size_write(struct inode *inode, u64 size)
241{
242 i_size_write(inode, size);
243 BTRFS_I(inode)->disk_i_size = size;
244}
245
246static inline bool btrfs_is_free_space_inode(struct inode *inode)
247{
248 struct btrfs_root *root = BTRFS_I(inode)->root;
249
250 if (root == root->fs_info->tree_root &&
251 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
252 return true;
253 if (BTRFS_I(inode)->location.objectid == BTRFS_FREE_INO_OBJECTID)
254 return true;
255 return false;
256}
257
258static inline int btrfs_inode_in_log(struct inode *inode, u64 generation)
259{
260 int ret = 0;
261
262 spin_lock(&BTRFS_I(inode)->lock);
263 if (BTRFS_I(inode)->logged_trans == generation &&
264 BTRFS_I(inode)->last_sub_trans <=
265 BTRFS_I(inode)->last_log_commit &&
266 BTRFS_I(inode)->last_sub_trans <=
267 BTRFS_I(inode)->root->last_log_commit) {
268 /*
269 * After a ranged fsync we might have left some extent maps
270 * (that fall outside the fsync's range). So return false
271 * here if the list isn't empty, to make sure btrfs_log_inode()
272 * will be called and process those extent maps.
273 */
274 smp_mb();
275 if (list_empty(&BTRFS_I(inode)->extent_tree.modified_extents))
276 ret = 1;
277 }
278 spin_unlock(&BTRFS_I(inode)->lock);
279 return ret;
280}
281
282#define BTRFS_DIO_ORIG_BIO_SUBMITTED 0x1
283
284struct btrfs_dio_private {
285 struct inode *inode;
286 unsigned long flags;
287 u64 logical_offset;
288 u64 disk_bytenr;
289 u64 bytes;
290 void *private;
291
292 /* number of bios pending for this dio */
293 atomic_t pending_bios;
294
295 /* IO errors */
296 int errors;
297
298 /* orig_bio is our btrfs_io_bio */
299 struct bio *orig_bio;
300
301 /* dio_bio came from fs/direct-io.c */
302 struct bio *dio_bio;
303
304 /*
305 * The original bio may be split to several sub-bios, this is
306 * done during endio of sub-bios
307 */
308 int (*subio_endio)(struct inode *, struct btrfs_io_bio *, int);
309};
310
311/*
312 * Disable DIO read nolock optimization, so new dio readers will be forced
313 * to grab i_mutex. It is used to avoid the endless truncate due to
314 * nonlocked dio read.
315 */
316static inline void btrfs_inode_block_unlocked_dio(struct inode *inode)
317{
318 set_bit(BTRFS_INODE_READDIO_NEED_LOCK, &BTRFS_I(inode)->runtime_flags);
319 smp_mb();
320}
321
322static inline void btrfs_inode_resume_unlocked_dio(struct inode *inode)
323{
324 smp_mb__before_atomic();
325 clear_bit(BTRFS_INODE_READDIO_NEED_LOCK,
326 &BTRFS_I(inode)->runtime_flags);
327}
328
329bool btrfs_page_exists_in_range(struct inode *inode, loff_t start, loff_t end);
330
331#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_ORDERED_DATA_CLOSE,
25 BTRFS_INODE_DUMMY,
26 BTRFS_INODE_IN_DEFRAG,
27 BTRFS_INODE_HAS_ASYNC_EXTENT,
28 BTRFS_INODE_NEEDS_FULL_SYNC,
29 BTRFS_INODE_COPY_EVERYTHING,
30 BTRFS_INODE_IN_DELALLOC_LIST,
31 BTRFS_INODE_READDIO_NEED_LOCK,
32 BTRFS_INODE_HAS_PROPS,
33 BTRFS_INODE_SNAPSHOT_FLUSH,
34};
35
36/* in memory btrfs inode */
37struct btrfs_inode {
38 /* which subvolume this inode belongs to */
39 struct btrfs_root *root;
40
41 /* key used to find this inode on disk. This is used by the code
42 * to read in roots of subvolumes
43 */
44 struct btrfs_key location;
45
46 /*
47 * Lock for counters and all fields used to determine if the inode is in
48 * the log or not (last_trans, last_sub_trans, last_log_commit,
49 * logged_trans).
50 */
51 spinlock_t lock;
52
53 /* the extent_tree has caches of all the extent mappings to disk */
54 struct extent_map_tree extent_tree;
55
56 /* the io_tree does range state (DIRTY, LOCKED etc) */
57 struct extent_io_tree io_tree;
58
59 /* special utility tree used to record which mirrors have already been
60 * tried when checksums fail for a given block
61 */
62 struct extent_io_tree io_failure_tree;
63
64 /*
65 * Keep track of where the inode has extent items mapped in order to
66 * make sure the i_size adjustments are accurate
67 */
68 struct extent_io_tree file_extent_tree;
69
70 /* held while logging the inode in tree-log.c */
71 struct mutex log_mutex;
72
73 /* used to order data wrt metadata */
74 struct btrfs_ordered_inode_tree ordered_tree;
75
76 /* list of all the delalloc inodes in the FS. There are times we need
77 * to write all the delalloc pages to disk, and this list is used
78 * to walk them all.
79 */
80 struct list_head delalloc_inodes;
81
82 /* node for the red-black tree that links inodes in subvolume root */
83 struct rb_node rb_node;
84
85 unsigned long runtime_flags;
86
87 /* Keep track of who's O_SYNC/fsyncing currently */
88 atomic_t sync_writers;
89
90 /* full 64 bit generation number, struct vfs_inode doesn't have a big
91 * enough field for this.
92 */
93 u64 generation;
94
95 /*
96 * transid of the trans_handle that last modified this inode
97 */
98 u64 last_trans;
99
100 /*
101 * transid that last logged this inode
102 */
103 u64 logged_trans;
104
105 /*
106 * log transid when this inode was last modified
107 */
108 int last_sub_trans;
109
110 /* a local copy of root's last_log_commit */
111 int last_log_commit;
112
113 /* total number of bytes pending delalloc, used by stat to calc the
114 * real block usage of the file
115 */
116 u64 delalloc_bytes;
117
118 /*
119 * Total number of bytes pending delalloc that fall within a file
120 * range that is either a hole or beyond EOF (and no prealloc extent
121 * exists in the range). This is always <= delalloc_bytes.
122 */
123 u64 new_delalloc_bytes;
124
125 /*
126 * total number of bytes pending defrag, used by stat to check whether
127 * it needs COW.
128 */
129 u64 defrag_bytes;
130
131 /*
132 * the size of the file stored in the metadata on disk. data=ordered
133 * means the in-memory i_size might be larger than the size on disk
134 * because not all the blocks are written yet.
135 */
136 u64 disk_i_size;
137
138 /*
139 * if this is a directory then index_cnt is the counter for the index
140 * number for new files that are created
141 */
142 u64 index_cnt;
143
144 /* Cache the directory index number to speed the dir/file remove */
145 u64 dir_index;
146
147 /* the fsync log has some corner cases that mean we have to check
148 * directories to see if any unlinks have been done before
149 * the directory was logged. See tree-log.c for all the
150 * details
151 */
152 u64 last_unlink_trans;
153
154 /*
155 * The id/generation of the last transaction where this inode was
156 * either the source or the destination of a clone/dedupe operation.
157 * Used when logging an inode to know if there are shared extents that
158 * need special care when logging checksum items, to avoid duplicate
159 * checksum items in a log (which can lead to a corruption where we end
160 * up with missing checksum ranges after log replay).
161 * Protected by the vfs inode lock.
162 */
163 u64 last_reflink_trans;
164
165 /*
166 * Number of bytes outstanding that are going to need csums. This is
167 * used in ENOSPC accounting.
168 */
169 u64 csum_bytes;
170
171 /* flags field from the on disk inode */
172 u32 flags;
173
174 /*
175 * Counters to keep track of the number of extent item's we may use due
176 * to delalloc and such. outstanding_extents is the number of extent
177 * items we think we'll end up using, and reserved_extents is the number
178 * of extent items we've reserved metadata for.
179 */
180 unsigned outstanding_extents;
181
182 struct btrfs_block_rsv block_rsv;
183
184 /*
185 * Cached values of inode properties
186 */
187 unsigned prop_compress; /* per-file compression algorithm */
188 /*
189 * Force compression on the file using the defrag ioctl, could be
190 * different from prop_compress and takes precedence if set
191 */
192 unsigned defrag_compress;
193
194 struct btrfs_delayed_node *delayed_node;
195
196 /* File creation time. */
197 struct timespec64 i_otime;
198
199 /* Hook into fs_info->delayed_iputs */
200 struct list_head delayed_iput;
201
202 /*
203 * To avoid races between lockless (i_mutex not held) direct IO writes
204 * and concurrent fsync requests. Direct IO writes must acquire read
205 * access on this semaphore for creating an extent map and its
206 * corresponding ordered extent. The fast fsync path must acquire write
207 * access on this semaphore before it collects ordered extents and
208 * extent maps.
209 */
210 struct rw_semaphore dio_sem;
211
212 struct inode vfs_inode;
213};
214
215static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
216{
217 return container_of(inode, struct btrfs_inode, vfs_inode);
218}
219
220static inline unsigned long btrfs_inode_hash(u64 objectid,
221 const struct btrfs_root *root)
222{
223 u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
224
225#if BITS_PER_LONG == 32
226 h = (h >> 32) ^ (h & 0xffffffff);
227#endif
228
229 return (unsigned long)h;
230}
231
232static inline void btrfs_insert_inode_hash(struct inode *inode)
233{
234 unsigned long h = btrfs_inode_hash(inode->i_ino, BTRFS_I(inode)->root);
235
236 __insert_inode_hash(inode, h);
237}
238
239static inline u64 btrfs_ino(const struct btrfs_inode *inode)
240{
241 u64 ino = inode->location.objectid;
242
243 /*
244 * !ino: btree_inode
245 * type == BTRFS_ROOT_ITEM_KEY: subvol dir
246 */
247 if (!ino || inode->location.type == BTRFS_ROOT_ITEM_KEY)
248 ino = inode->vfs_inode.i_ino;
249 return ino;
250}
251
252static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
253{
254 i_size_write(&inode->vfs_inode, size);
255 inode->disk_i_size = size;
256}
257
258static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
259{
260 struct btrfs_root *root = inode->root;
261
262 if (root == root->fs_info->tree_root &&
263 btrfs_ino(inode) != BTRFS_BTREE_INODE_OBJECTID)
264 return true;
265 if (inode->location.objectid == BTRFS_FREE_INO_OBJECTID)
266 return true;
267 return false;
268}
269
270static inline bool is_data_inode(struct inode *inode)
271{
272 return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
273}
274
275static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
276 int mod)
277{
278 lockdep_assert_held(&inode->lock);
279 inode->outstanding_extents += mod;
280 if (btrfs_is_free_space_inode(inode))
281 return;
282 trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
283 mod);
284}
285
286static inline int btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
287{
288 int ret = 0;
289
290 spin_lock(&inode->lock);
291 if (inode->logged_trans == generation &&
292 inode->last_sub_trans <= inode->last_log_commit &&
293 inode->last_sub_trans <= inode->root->last_log_commit) {
294 /*
295 * After a ranged fsync we might have left some extent maps
296 * (that fall outside the fsync's range). So return false
297 * here if the list isn't empty, to make sure btrfs_log_inode()
298 * will be called and process those extent maps.
299 */
300 smp_mb();
301 if (list_empty(&inode->extent_tree.modified_extents))
302 ret = 1;
303 }
304 spin_unlock(&inode->lock);
305 return ret;
306}
307
308struct btrfs_dio_private {
309 struct inode *inode;
310 u64 logical_offset;
311 u64 disk_bytenr;
312 u64 bytes;
313
314 /*
315 * References to this structure. There is one reference per in-flight
316 * bio plus one while we're still setting up.
317 */
318 refcount_t refs;
319
320 /* dio_bio came from fs/direct-io.c */
321 struct bio *dio_bio;
322
323 /* Array of checksums */
324 u8 csums[];
325};
326
327/*
328 * Disable DIO read nolock optimization, so new dio readers will be forced
329 * to grab i_mutex. It is used to avoid the endless truncate due to
330 * nonlocked dio read.
331 */
332static inline void btrfs_inode_block_unlocked_dio(struct btrfs_inode *inode)
333{
334 set_bit(BTRFS_INODE_READDIO_NEED_LOCK, &inode->runtime_flags);
335 smp_mb();
336}
337
338static inline void btrfs_inode_resume_unlocked_dio(struct btrfs_inode *inode)
339{
340 smp_mb__before_atomic();
341 clear_bit(BTRFS_INODE_READDIO_NEED_LOCK, &inode->runtime_flags);
342}
343
344/* Array of bytes with variable length, hexadecimal format 0x1234 */
345#define CSUM_FMT "0x%*phN"
346#define CSUM_FMT_VALUE(size, bytes) size, bytes
347
348static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
349 u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
350{
351 struct btrfs_root *root = inode->root;
352 struct btrfs_super_block *sb = root->fs_info->super_copy;
353 const u16 csum_size = btrfs_super_csum_size(sb);
354
355 /* Output minus objectid, which is more meaningful */
356 if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
357 btrfs_warn_rl(root->fs_info,
358"csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
359 root->root_key.objectid, btrfs_ino(inode),
360 logical_start,
361 CSUM_FMT_VALUE(csum_size, csum),
362 CSUM_FMT_VALUE(csum_size, csum_expected),
363 mirror_num);
364 else
365 btrfs_warn_rl(root->fs_info,
366"csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
367 root->root_key.objectid, btrfs_ino(inode),
368 logical_start,
369 CSUM_FMT_VALUE(csum_size, csum),
370 CSUM_FMT_VALUE(csum_size, csum_expected),
371 mirror_num);
372}
373
374#endif