xref: /linux/fs/btrfs/locking.c (revision b1a54551dd9ed5ef1763b97b35a0999ca002b95c)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2008 Oracle.  All rights reserved.
4  */
5 
6 #include <linux/sched.h>
7 #include <linux/pagemap.h>
8 #include <linux/spinlock.h>
9 #include <linux/page-flags.h>
10 #include <asm/bug.h>
11 #include <trace/events/btrfs.h>
12 #include "misc.h"
13 #include "ctree.h"
14 #include "extent_io.h"
15 #include "locking.h"
16 #include "accessors.h"
17 
18 /*
19  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
20  * eb, the lockdep key is determined by the btrfs_root it belongs to and
21  * the level the eb occupies in the tree.
22  *
23  * Different roots are used for different purposes and may nest inside each
24  * other and they require separate keysets.  As lockdep keys should be
25  * static, assign keysets according to the purpose of the root as indicated
26  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
27  * roots have separate keysets.
28  *
29  * Lock-nesting across peer nodes is always done with the immediate parent
30  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
31  * subclass to avoid triggering lockdep warning in such cases.
32  *
33  * The key is set by the readpage_end_io_hook after the buffer has passed
34  * csum validation but before the pages are unlocked.  It is also set by
35  * btrfs_init_new_buffer on freshly allocated blocks.
36  *
37  * We also add a check to make sure the highest level of the tree is the
38  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
39  * needs update as well.
40  */
41 #ifdef CONFIG_DEBUG_LOCK_ALLOC
42 #if BTRFS_MAX_LEVEL != 8
43 #error
44 #endif
45 
46 #define DEFINE_LEVEL(stem, level)					\
47 	.names[level] = "btrfs-" stem "-0" #level,
48 
49 #define DEFINE_NAME(stem)						\
50 	DEFINE_LEVEL(stem, 0)						\
51 	DEFINE_LEVEL(stem, 1)						\
52 	DEFINE_LEVEL(stem, 2)						\
53 	DEFINE_LEVEL(stem, 3)						\
54 	DEFINE_LEVEL(stem, 4)						\
55 	DEFINE_LEVEL(stem, 5)						\
56 	DEFINE_LEVEL(stem, 6)						\
57 	DEFINE_LEVEL(stem, 7)
58 
59 static struct btrfs_lockdep_keyset {
60 	u64			id;		/* root objectid */
61 	/* Longest entry: btrfs-block-group-00 */
62 	char			names[BTRFS_MAX_LEVEL][24];
63 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
64 } btrfs_lockdep_keysets[] = {
65 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
66 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
67 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
68 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
69 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
70 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
71 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
72 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
73 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
74 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
75 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
76 	{ .id = BTRFS_BLOCK_GROUP_TREE_OBJECTID, DEFINE_NAME("block-group") },
77 	{ .id = BTRFS_RAID_STRIPE_TREE_OBJECTID, DEFINE_NAME("raid-stripe") },
78 	{ .id = 0,				DEFINE_NAME("tree")	},
79 };
80 
81 #undef DEFINE_LEVEL
82 #undef DEFINE_NAME
83 
84 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, int level)
85 {
86 	struct btrfs_lockdep_keyset *ks;
87 
88 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
89 
90 	/* Find the matching keyset, id 0 is the default entry */
91 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
92 		if (ks->id == objectid)
93 			break;
94 
95 	lockdep_set_class_and_name(&eb->lock, &ks->keys[level], ks->names[level]);
96 }
97 
98 void btrfs_maybe_reset_lockdep_class(struct btrfs_root *root, struct extent_buffer *eb)
99 {
100 	if (test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
101 		btrfs_set_buffer_lockdep_class(root->root_key.objectid,
102 					       eb, btrfs_header_level(eb));
103 }
104 
105 #endif
106 
107 #ifdef CONFIG_BTRFS_DEBUG
108 static void btrfs_set_eb_lock_owner(struct extent_buffer *eb, pid_t owner)
109 {
110 	eb->lock_owner = owner;
111 }
112 #else
113 static void btrfs_set_eb_lock_owner(struct extent_buffer *eb, pid_t owner) { }
114 #endif
115 
116 /*
117  * Extent buffer locking
118  * =====================
119  *
120  * We use a rw_semaphore for tree locking, and the semantics are exactly the
121  * same:
122  *
123  * - reader/writer exclusion
124  * - writer/writer exclusion
125  * - reader/reader sharing
126  * - try-lock semantics for readers and writers
127  *
128  * The rwsem implementation does opportunistic spinning which reduces number of
129  * times the locking task needs to sleep.
130  */
131 
132 /*
133  * __btrfs_tree_read_lock - lock extent buffer for read
134  * @eb:		the eb to be locked
135  * @nest:	the nesting level to be used for lockdep
136  *
137  * This takes the read lock on the extent buffer, using the specified nesting
138  * level for lockdep purposes.
139  */
140 void __btrfs_tree_read_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
141 {
142 	u64 start_ns = 0;
143 
144 	if (trace_btrfs_tree_read_lock_enabled())
145 		start_ns = ktime_get_ns();
146 
147 	down_read_nested(&eb->lock, nest);
148 	trace_btrfs_tree_read_lock(eb, start_ns);
149 }
150 
151 void btrfs_tree_read_lock(struct extent_buffer *eb)
152 {
153 	__btrfs_tree_read_lock(eb, BTRFS_NESTING_NORMAL);
154 }
155 
156 /*
157  * Try-lock for read.
158  *
159  * Return 1 if the rwlock has been taken, 0 otherwise
160  */
161 int btrfs_try_tree_read_lock(struct extent_buffer *eb)
162 {
163 	if (down_read_trylock(&eb->lock)) {
164 		trace_btrfs_try_tree_read_lock(eb);
165 		return 1;
166 	}
167 	return 0;
168 }
169 
170 /*
171  * Try-lock for write.
172  *
173  * Return 1 if the rwlock has been taken, 0 otherwise
174  */
175 int btrfs_try_tree_write_lock(struct extent_buffer *eb)
176 {
177 	if (down_write_trylock(&eb->lock)) {
178 		btrfs_set_eb_lock_owner(eb, current->pid);
179 		trace_btrfs_try_tree_write_lock(eb);
180 		return 1;
181 	}
182 	return 0;
183 }
184 
185 /*
186  * Release read lock.
187  */
188 void btrfs_tree_read_unlock(struct extent_buffer *eb)
189 {
190 	trace_btrfs_tree_read_unlock(eb);
191 	up_read(&eb->lock);
192 }
193 
194 /*
195  * Lock eb for write.
196  *
197  * @eb:		the eb to lock
198  * @nest:	the nesting to use for the lock
199  *
200  * Returns with the eb->lock write locked.
201  */
202 void __btrfs_tree_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
203 	__acquires(&eb->lock)
204 {
205 	u64 start_ns = 0;
206 
207 	if (trace_btrfs_tree_lock_enabled())
208 		start_ns = ktime_get_ns();
209 
210 	down_write_nested(&eb->lock, nest);
211 	btrfs_set_eb_lock_owner(eb, current->pid);
212 	trace_btrfs_tree_lock(eb, start_ns);
213 }
214 
215 void btrfs_tree_lock(struct extent_buffer *eb)
216 {
217 	__btrfs_tree_lock(eb, BTRFS_NESTING_NORMAL);
218 }
219 
220 /*
221  * Release the write lock.
222  */
223 void btrfs_tree_unlock(struct extent_buffer *eb)
224 {
225 	trace_btrfs_tree_unlock(eb);
226 	btrfs_set_eb_lock_owner(eb, 0);
227 	up_write(&eb->lock);
228 }
229 
230 /*
231  * This releases any locks held in the path starting at level and going all the
232  * way up to the root.
233  *
234  * btrfs_search_slot will keep the lock held on higher nodes in a few corner
235  * cases, such as COW of the block at slot zero in the node.  This ignores
236  * those rules, and it should only be called when there are no more updates to
237  * be done higher up in the tree.
238  */
239 void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
240 {
241 	int i;
242 
243 	if (path->keep_locks)
244 		return;
245 
246 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
247 		if (!path->nodes[i])
248 			continue;
249 		if (!path->locks[i])
250 			continue;
251 		btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
252 		path->locks[i] = 0;
253 	}
254 }
255 
256 /*
257  * Loop around taking references on and locking the root node of the tree until
258  * we end up with a lock on the root node.
259  *
260  * Return: root extent buffer with write lock held
261  */
262 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
263 {
264 	struct extent_buffer *eb;
265 
266 	while (1) {
267 		eb = btrfs_root_node(root);
268 
269 		btrfs_maybe_reset_lockdep_class(root, eb);
270 		btrfs_tree_lock(eb);
271 		if (eb == root->node)
272 			break;
273 		btrfs_tree_unlock(eb);
274 		free_extent_buffer(eb);
275 	}
276 	return eb;
277 }
278 
279 /*
280  * Loop around taking references on and locking the root node of the tree until
281  * we end up with a lock on the root node.
282  *
283  * Return: root extent buffer with read lock held
284  */
285 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
286 {
287 	struct extent_buffer *eb;
288 
289 	while (1) {
290 		eb = btrfs_root_node(root);
291 
292 		btrfs_maybe_reset_lockdep_class(root, eb);
293 		btrfs_tree_read_lock(eb);
294 		if (eb == root->node)
295 			break;
296 		btrfs_tree_read_unlock(eb);
297 		free_extent_buffer(eb);
298 	}
299 	return eb;
300 }
301 
302 /*
303  * Loop around taking references on and locking the root node of the tree in
304  * nowait mode until we end up with a lock on the root node or returning to
305  * avoid blocking.
306  *
307  * Return: root extent buffer with read lock held or -EAGAIN.
308  */
309 struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root)
310 {
311 	struct extent_buffer *eb;
312 
313 	while (1) {
314 		eb = btrfs_root_node(root);
315 		if (!btrfs_try_tree_read_lock(eb)) {
316 			free_extent_buffer(eb);
317 			return ERR_PTR(-EAGAIN);
318 		}
319 		if (eb == root->node)
320 			break;
321 		btrfs_tree_read_unlock(eb);
322 		free_extent_buffer(eb);
323 	}
324 	return eb;
325 }
326 
327 /*
328  * DREW locks
329  * ==========
330  *
331  * DREW stands for double-reader-writer-exclusion lock. It's used in situation
332  * where you want to provide A-B exclusion but not AA or BB.
333  *
334  * Currently implementation gives more priority to reader. If a reader and a
335  * writer both race to acquire their respective sides of the lock the writer
336  * would yield its lock as soon as it detects a concurrent reader. Additionally
337  * if there are pending readers no new writers would be allowed to come in and
338  * acquire the lock.
339  */
340 
341 void btrfs_drew_lock_init(struct btrfs_drew_lock *lock)
342 {
343 	atomic_set(&lock->readers, 0);
344 	atomic_set(&lock->writers, 0);
345 	init_waitqueue_head(&lock->pending_readers);
346 	init_waitqueue_head(&lock->pending_writers);
347 }
348 
349 /* Return true if acquisition is successful, false otherwise */
350 bool btrfs_drew_try_write_lock(struct btrfs_drew_lock *lock)
351 {
352 	if (atomic_read(&lock->readers))
353 		return false;
354 
355 	atomic_inc(&lock->writers);
356 
357 	/* Ensure writers count is updated before we check for pending readers */
358 	smp_mb__after_atomic();
359 	if (atomic_read(&lock->readers)) {
360 		btrfs_drew_write_unlock(lock);
361 		return false;
362 	}
363 
364 	return true;
365 }
366 
367 void btrfs_drew_write_lock(struct btrfs_drew_lock *lock)
368 {
369 	while (true) {
370 		if (btrfs_drew_try_write_lock(lock))
371 			return;
372 		wait_event(lock->pending_writers, !atomic_read(&lock->readers));
373 	}
374 }
375 
376 void btrfs_drew_write_unlock(struct btrfs_drew_lock *lock)
377 {
378 	atomic_dec(&lock->writers);
379 	cond_wake_up(&lock->pending_readers);
380 }
381 
382 void btrfs_drew_read_lock(struct btrfs_drew_lock *lock)
383 {
384 	atomic_inc(&lock->readers);
385 
386 	/*
387 	 * Ensure the pending reader count is perceieved BEFORE this reader
388 	 * goes to sleep in case of active writers. This guarantees new writers
389 	 * won't be allowed and that the current reader will be woken up when
390 	 * the last active writer finishes its jobs.
391 	 */
392 	smp_mb__after_atomic();
393 
394 	wait_event(lock->pending_readers, atomic_read(&lock->writers) == 0);
395 }
396 
397 void btrfs_drew_read_unlock(struct btrfs_drew_lock *lock)
398 {
399 	/*
400 	 * atomic_dec_and_test implies a full barrier, so woken up writers
401 	 * are guaranteed to see the decrement
402 	 */
403 	if (atomic_dec_and_test(&lock->readers))
404 		wake_up(&lock->pending_writers);
405 }
406