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  * 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()).
20  */
21 #define BTRFS_DIR_START_INDEX 2
22 
23 /*
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.
29  */
30 enum {
31 	BTRFS_INODE_FLUSH_ON_CLOSE,
32 	BTRFS_INODE_DUMMY,
33 	BTRFS_INODE_IN_DEFRAG,
34 	BTRFS_INODE_HAS_ASYNC_EXTENT,
35 	 /*
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).
39 	  */
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,
45 	/*
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
49 	 * to an inode.
50 	 */
51 	BTRFS_INODE_NO_XATTRS,
52 	/*
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).
59 	 */
60 	BTRFS_INODE_NO_DELALLOC_FLUSH,
61 	/*
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
65 	 * on the same file.
66 	 */
67 	BTRFS_INODE_VERITY_IN_PROGRESS,
68 	/* Set when this inode is a free space inode. */
69 	BTRFS_INODE_FREE_SPACE_INODE,
70 };
71 
72 /* in memory btrfs inode */
73 struct btrfs_inode {
74 	/* which subvolume this inode belongs to */
75 	struct btrfs_root *root;
76 
77 	/* key used to find this inode on disk.  This is used by the code
78 	 * to read in roots of subvolumes
79 	 */
80 	struct btrfs_key location;
81 
82 	/*
83 	 * Lock for counters and all fields used to determine if the inode is in
84 	 * the log or not (last_trans, last_sub_trans, last_log_commit,
85 	 * logged_trans), to access/update new_delalloc_bytes and to update the
86 	 * VFS' inode number of bytes used.
87 	 */
88 	spinlock_t lock;
89 
90 	/* the extent_tree has caches of all the extent mappings to disk */
91 	struct extent_map_tree extent_tree;
92 
93 	/* the io_tree does range state (DIRTY, LOCKED etc) */
94 	struct extent_io_tree io_tree;
95 
96 	/* special utility tree used to record which mirrors have already been
97 	 * tried when checksums fail for a given block
98 	 */
99 	struct rb_root io_failure_tree;
100 	spinlock_t io_failure_lock;
101 
102 	/*
103 	 * Keep track of where the inode has extent items mapped in order to
104 	 * make sure the i_size adjustments are accurate
105 	 */
106 	struct extent_io_tree file_extent_tree;
107 
108 	/* held while logging the inode in tree-log.c */
109 	struct mutex log_mutex;
110 
111 	/* used to order data wrt metadata */
112 	struct btrfs_ordered_inode_tree ordered_tree;
113 
114 	/* list of all the delalloc inodes in the FS.  There are times we need
115 	 * to write all the delalloc pages to disk, and this list is used
116 	 * to walk them all.
117 	 */
118 	struct list_head delalloc_inodes;
119 
120 	/* node for the red-black tree that links inodes in subvolume root */
121 	struct rb_node rb_node;
122 
123 	unsigned long runtime_flags;
124 
125 	/* Keep track of who's O_SYNC/fsyncing currently */
126 	atomic_t sync_writers;
127 
128 	/* full 64 bit generation number, struct vfs_inode doesn't have a big
129 	 * enough field for this.
130 	 */
131 	u64 generation;
132 
133 	/*
134 	 * transid of the trans_handle that last modified this inode
135 	 */
136 	u64 last_trans;
137 
138 	/*
139 	 * transid that last logged this inode
140 	 */
141 	u64 logged_trans;
142 
143 	/*
144 	 * log transid when this inode was last modified
145 	 */
146 	int last_sub_trans;
147 
148 	/* a local copy of root's last_log_commit */
149 	int last_log_commit;
150 
151 	/*
152 	 * Total number of bytes pending delalloc, used by stat to calculate the
153 	 * real block usage of the file. This is used only for files.
154 	 */
155 	u64 delalloc_bytes;
156 
157 	union {
158 		/*
159 		 * Total number of bytes pending delalloc that fall within a file
160 		 * range that is either a hole or beyond EOF (and no prealloc extent
161 		 * exists in the range). This is always <= delalloc_bytes and this
162 		 * is used only for files.
163 		 */
164 		u64 new_delalloc_bytes;
165 		/*
166 		 * The offset of the last dir index key that was logged.
167 		 * This is used only for directories.
168 		 */
169 		u64 last_dir_index_offset;
170 	};
171 
172 	/*
173 	 * total number of bytes pending defrag, used by stat to check whether
174 	 * it needs COW.
175 	 */
176 	u64 defrag_bytes;
177 
178 	/*
179 	 * the size of the file stored in the metadata on disk.  data=ordered
180 	 * means the in-memory i_size might be larger than the size on disk
181 	 * because not all the blocks are written yet.
182 	 */
183 	u64 disk_i_size;
184 
185 	/*
186 	 * If this is a directory then index_cnt is the counter for the index
187 	 * number for new files that are created. For an empty directory, this
188 	 * must be initialized to BTRFS_DIR_START_INDEX.
189 	 */
190 	u64 index_cnt;
191 
192 	/* Cache the directory index number to speed the dir/file remove */
193 	u64 dir_index;
194 
195 	/* the fsync log has some corner cases that mean we have to check
196 	 * directories to see if any unlinks have been done before
197 	 * the directory was logged.  See tree-log.c for all the
198 	 * details
199 	 */
200 	u64 last_unlink_trans;
201 
202 	/*
203 	 * The id/generation of the last transaction where this inode was
204 	 * either the source or the destination of a clone/dedupe operation.
205 	 * Used when logging an inode to know if there are shared extents that
206 	 * need special care when logging checksum items, to avoid duplicate
207 	 * checksum items in a log (which can lead to a corruption where we end
208 	 * up with missing checksum ranges after log replay).
209 	 * Protected by the vfs inode lock.
210 	 */
211 	u64 last_reflink_trans;
212 
213 	/*
214 	 * Number of bytes outstanding that are going to need csums.  This is
215 	 * used in ENOSPC accounting.
216 	 */
217 	u64 csum_bytes;
218 
219 	/* Backwards incompatible flags, lower half of inode_item::flags  */
220 	u32 flags;
221 	/* Read-only compatibility flags, upper half of inode_item::flags */
222 	u32 ro_flags;
223 
224 	/*
225 	 * Counters to keep track of the number of extent item's we may use due
226 	 * to delalloc and such.  outstanding_extents is the number of extent
227 	 * items we think we'll end up using, and reserved_extents is the number
228 	 * of extent items we've reserved metadata for.
229 	 */
230 	unsigned outstanding_extents;
231 
232 	struct btrfs_block_rsv block_rsv;
233 
234 	/*
235 	 * Cached values of inode properties
236 	 */
237 	unsigned prop_compress;		/* per-file compression algorithm */
238 	/*
239 	 * Force compression on the file using the defrag ioctl, could be
240 	 * different from prop_compress and takes precedence if set
241 	 */
242 	unsigned defrag_compress;
243 
244 	struct btrfs_delayed_node *delayed_node;
245 
246 	/* File creation time. */
247 	struct timespec64 i_otime;
248 
249 	/* Hook into fs_info->delayed_iputs */
250 	struct list_head delayed_iput;
251 
252 	struct rw_semaphore i_mmap_lock;
253 	struct inode vfs_inode;
254 };
255 
BTRFS_I(const struct inode * inode)256 static inline struct btrfs_inode *BTRFS_I(const struct inode *inode)
257 {
258 	return container_of(inode, struct btrfs_inode, vfs_inode);
259 }
260 
btrfs_inode_hash(u64 objectid,const struct btrfs_root * root)261 static inline unsigned long btrfs_inode_hash(u64 objectid,
262 					     const struct btrfs_root *root)
263 {
264 	u64 h = objectid ^ (root->root_key.objectid * GOLDEN_RATIO_PRIME);
265 
266 #if BITS_PER_LONG == 32
267 	h = (h >> 32) ^ (h & 0xffffffff);
268 #endif
269 
270 	return (unsigned long)h;
271 }
272 
273 #if BITS_PER_LONG == 32
274 
275 /*
276  * On 32 bit systems the i_ino of struct inode is 32 bits (unsigned long), so
277  * we use the inode's location objectid which is a u64 to avoid truncation.
278  */
btrfs_ino(const struct btrfs_inode * inode)279 static inline u64 btrfs_ino(const struct btrfs_inode *inode)
280 {
281 	u64 ino = inode->location.objectid;
282 
283 	/* type == BTRFS_ROOT_ITEM_KEY: subvol dir */
284 	if (inode->location.type == BTRFS_ROOT_ITEM_KEY)
285 		ino = inode->vfs_inode.i_ino;
286 	return ino;
287 }
288 
289 #else
290 
btrfs_ino(const struct btrfs_inode * inode)291 static inline u64 btrfs_ino(const struct btrfs_inode *inode)
292 {
293 	return inode->vfs_inode.i_ino;
294 }
295 
296 #endif
297 
btrfs_i_size_write(struct btrfs_inode * inode,u64 size)298 static inline void btrfs_i_size_write(struct btrfs_inode *inode, u64 size)
299 {
300 	i_size_write(&inode->vfs_inode, size);
301 	inode->disk_i_size = size;
302 }
303 
btrfs_is_free_space_inode(struct btrfs_inode * inode)304 static inline bool btrfs_is_free_space_inode(struct btrfs_inode *inode)
305 {
306 	return test_bit(BTRFS_INODE_FREE_SPACE_INODE, &inode->runtime_flags);
307 }
308 
is_data_inode(struct inode * inode)309 static inline bool is_data_inode(struct inode *inode)
310 {
311 	return btrfs_ino(BTRFS_I(inode)) != BTRFS_BTREE_INODE_OBJECTID;
312 }
313 
btrfs_mod_outstanding_extents(struct btrfs_inode * inode,int mod)314 static inline void btrfs_mod_outstanding_extents(struct btrfs_inode *inode,
315 						 int mod)
316 {
317 	lockdep_assert_held(&inode->lock);
318 	inode->outstanding_extents += mod;
319 	if (btrfs_is_free_space_inode(inode))
320 		return;
321 	trace_btrfs_inode_mod_outstanding_extents(inode->root, btrfs_ino(inode),
322 						  mod);
323 }
324 
325 /*
326  * Called every time after doing a buffered, direct IO or memory mapped write.
327  *
328  * This is to ensure that if we write to a file that was previously fsynced in
329  * the current transaction, then try to fsync it again in the same transaction,
330  * we will know that there were changes in the file and that it needs to be
331  * logged.
332  */
btrfs_set_inode_last_sub_trans(struct btrfs_inode * inode)333 static inline void btrfs_set_inode_last_sub_trans(struct btrfs_inode *inode)
334 {
335 	spin_lock(&inode->lock);
336 	inode->last_sub_trans = inode->root->log_transid;
337 	spin_unlock(&inode->lock);
338 }
339 
340 /*
341  * Should be called while holding the inode's VFS lock in exclusive mode or in a
342  * context where no one else can access the inode concurrently (during inode
343  * creation or when loading an inode from disk).
344  */
btrfs_set_inode_full_sync(struct btrfs_inode * inode)345 static inline void btrfs_set_inode_full_sync(struct btrfs_inode *inode)
346 {
347 	set_bit(BTRFS_INODE_NEEDS_FULL_SYNC, &inode->runtime_flags);
348 	/*
349 	 * The inode may have been part of a reflink operation in the last
350 	 * transaction that modified it, and then a fsync has reset the
351 	 * last_reflink_trans to avoid subsequent fsyncs in the same
352 	 * transaction to do unnecessary work. So update last_reflink_trans
353 	 * to the last_trans value (we have to be pessimistic and assume a
354 	 * reflink happened).
355 	 *
356 	 * The ->last_trans is protected by the inode's spinlock and we can
357 	 * have a concurrent ordered extent completion update it. Also set
358 	 * last_reflink_trans to ->last_trans only if the former is less than
359 	 * the later, because we can be called in a context where
360 	 * last_reflink_trans was set to the current transaction generation
361 	 * while ->last_trans was not yet updated in the current transaction,
362 	 * and therefore has a lower value.
363 	 */
364 	spin_lock(&inode->lock);
365 	if (inode->last_reflink_trans < inode->last_trans)
366 		inode->last_reflink_trans = inode->last_trans;
367 	spin_unlock(&inode->lock);
368 }
369 
btrfs_inode_in_log(struct btrfs_inode * inode,u64 generation)370 static inline bool btrfs_inode_in_log(struct btrfs_inode *inode, u64 generation)
371 {
372 	bool ret = false;
373 
374 	spin_lock(&inode->lock);
375 	if (inode->logged_trans == generation &&
376 	    inode->last_sub_trans <= inode->last_log_commit &&
377 	    inode->last_sub_trans <= inode->root->last_log_commit)
378 		ret = true;
379 	spin_unlock(&inode->lock);
380 	return ret;
381 }
382 
383 /*
384  * Check if the inode has flags compatible with compression
385  */
btrfs_inode_can_compress(const struct btrfs_inode * inode)386 static inline bool btrfs_inode_can_compress(const struct btrfs_inode *inode)
387 {
388 	if (inode->flags & BTRFS_INODE_NODATACOW ||
389 	    inode->flags & BTRFS_INODE_NODATASUM)
390 		return false;
391 	return true;
392 }
393 
394 /*
395  * btrfs_inode_item stores flags in a u64, btrfs_inode stores them in two
396  * separate u32s. These two functions convert between the two representations.
397  */
btrfs_inode_combine_flags(u32 flags,u32 ro_flags)398 static inline u64 btrfs_inode_combine_flags(u32 flags, u32 ro_flags)
399 {
400 	return (flags | ((u64)ro_flags << 32));
401 }
402 
btrfs_inode_split_flags(u64 inode_item_flags,u32 * flags,u32 * ro_flags)403 static inline void btrfs_inode_split_flags(u64 inode_item_flags,
404 					   u32 *flags, u32 *ro_flags)
405 {
406 	*flags = (u32)inode_item_flags;
407 	*ro_flags = (u32)(inode_item_flags >> 32);
408 }
409 
410 /* Array of bytes with variable length, hexadecimal format 0x1234 */
411 #define CSUM_FMT				"0x%*phN"
412 #define CSUM_FMT_VALUE(size, bytes)		size, bytes
413 
btrfs_print_data_csum_error(struct btrfs_inode * inode,u64 logical_start,u8 * csum,u8 * csum_expected,int mirror_num)414 static inline void btrfs_print_data_csum_error(struct btrfs_inode *inode,
415 		u64 logical_start, u8 *csum, u8 *csum_expected, int mirror_num)
416 {
417 	struct btrfs_root *root = inode->root;
418 	const u32 csum_size = root->fs_info->csum_size;
419 
420 	/* Output minus objectid, which is more meaningful */
421 	if (root->root_key.objectid >= BTRFS_LAST_FREE_OBJECTID)
422 		btrfs_warn_rl(root->fs_info,
423 "csum failed root %lld ino %lld off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
424 			root->root_key.objectid, btrfs_ino(inode),
425 			logical_start,
426 			CSUM_FMT_VALUE(csum_size, csum),
427 			CSUM_FMT_VALUE(csum_size, csum_expected),
428 			mirror_num);
429 	else
430 		btrfs_warn_rl(root->fs_info,
431 "csum failed root %llu ino %llu off %llu csum " CSUM_FMT " expected csum " CSUM_FMT " mirror %d",
432 			root->root_key.objectid, btrfs_ino(inode),
433 			logical_start,
434 			CSUM_FMT_VALUE(csum_size, csum),
435 			CSUM_FMT_VALUE(csum_size, csum_expected),
436 			mirror_num);
437 }
438 
439 #endif
440