xref: /linux/fs/btrfs/locking.c (revision 69bfec7548f4c1595bac0e3ddfc0458a5af31f4c)
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 "misc.h"
12 #include "ctree.h"
13 #include "extent_io.h"
14 #include "locking.h"
15 #include "accessors.h"
16 
17 /*
18  * Lockdep class keys for extent_buffer->lock's in this root.  For a given
19  * eb, the lockdep key is determined by the btrfs_root it belongs to and
20  * the level the eb occupies in the tree.
21  *
22  * Different roots are used for different purposes and may nest inside each
23  * other and they require separate keysets.  As lockdep keys should be
24  * static, assign keysets according to the purpose of the root as indicated
25  * by btrfs_root->root_key.objectid.  This ensures that all special purpose
26  * roots have separate keysets.
27  *
28  * Lock-nesting across peer nodes is always done with the immediate parent
29  * node locked thus preventing deadlock.  As lockdep doesn't know this, use
30  * subclass to avoid triggering lockdep warning in such cases.
31  *
32  * The key is set by the readpage_end_io_hook after the buffer has passed
33  * csum validation but before the pages are unlocked.  It is also set by
34  * btrfs_init_new_buffer on freshly allocated blocks.
35  *
36  * We also add a check to make sure the highest level of the tree is the
37  * same as our lockdep setup here.  If BTRFS_MAX_LEVEL changes, this code
38  * needs update as well.
39  */
40 #ifdef CONFIG_DEBUG_LOCK_ALLOC
41 #if BTRFS_MAX_LEVEL != 8
42 #error
43 #endif
44 
45 #define DEFINE_LEVEL(stem, level)					\
46 	.names[level] = "btrfs-" stem "-0" #level,
47 
48 #define DEFINE_NAME(stem)						\
49 	DEFINE_LEVEL(stem, 0)						\
50 	DEFINE_LEVEL(stem, 1)						\
51 	DEFINE_LEVEL(stem, 2)						\
52 	DEFINE_LEVEL(stem, 3)						\
53 	DEFINE_LEVEL(stem, 4)						\
54 	DEFINE_LEVEL(stem, 5)						\
55 	DEFINE_LEVEL(stem, 6)						\
56 	DEFINE_LEVEL(stem, 7)
57 
58 static struct btrfs_lockdep_keyset {
59 	u64			id;		/* root objectid */
60 	/* Longest entry: btrfs-free-space-00 */
61 	char			names[BTRFS_MAX_LEVEL][20];
62 	struct lock_class_key	keys[BTRFS_MAX_LEVEL];
63 } btrfs_lockdep_keysets[] = {
64 	{ .id = BTRFS_ROOT_TREE_OBJECTID,	DEFINE_NAME("root")	},
65 	{ .id = BTRFS_EXTENT_TREE_OBJECTID,	DEFINE_NAME("extent")	},
66 	{ .id = BTRFS_CHUNK_TREE_OBJECTID,	DEFINE_NAME("chunk")	},
67 	{ .id = BTRFS_DEV_TREE_OBJECTID,	DEFINE_NAME("dev")	},
68 	{ .id = BTRFS_CSUM_TREE_OBJECTID,	DEFINE_NAME("csum")	},
69 	{ .id = BTRFS_QUOTA_TREE_OBJECTID,	DEFINE_NAME("quota")	},
70 	{ .id = BTRFS_TREE_LOG_OBJECTID,	DEFINE_NAME("log")	},
71 	{ .id = BTRFS_TREE_RELOC_OBJECTID,	DEFINE_NAME("treloc")	},
72 	{ .id = BTRFS_DATA_RELOC_TREE_OBJECTID,	DEFINE_NAME("dreloc")	},
73 	{ .id = BTRFS_UUID_TREE_OBJECTID,	DEFINE_NAME("uuid")	},
74 	{ .id = BTRFS_FREE_SPACE_TREE_OBJECTID,	DEFINE_NAME("free-space") },
75 	{ .id = 0,				DEFINE_NAME("tree")	},
76 };
77 
78 #undef DEFINE_LEVEL
79 #undef DEFINE_NAME
80 
81 void btrfs_set_buffer_lockdep_class(u64 objectid, struct extent_buffer *eb, int level)
82 {
83 	struct btrfs_lockdep_keyset *ks;
84 
85 	BUG_ON(level >= ARRAY_SIZE(ks->keys));
86 
87 	/* Find the matching keyset, id 0 is the default entry */
88 	for (ks = btrfs_lockdep_keysets; ks->id; ks++)
89 		if (ks->id == objectid)
90 			break;
91 
92 	lockdep_set_class_and_name(&eb->lock, &ks->keys[level], ks->names[level]);
93 }
94 
95 void btrfs_maybe_reset_lockdep_class(struct btrfs_root *root, struct extent_buffer *eb)
96 {
97 	if (test_bit(BTRFS_ROOT_RESET_LOCKDEP_CLASS, &root->state))
98 		btrfs_set_buffer_lockdep_class(root->root_key.objectid,
99 					       eb, btrfs_header_level(eb));
100 }
101 
102 #endif
103 
104 /*
105  * Extent buffer locking
106  * =====================
107  *
108  * We use a rw_semaphore for tree locking, and the semantics are exactly the
109  * same:
110  *
111  * - reader/writer exclusion
112  * - writer/writer exclusion
113  * - reader/reader sharing
114  * - try-lock semantics for readers and writers
115  *
116  * The rwsem implementation does opportunistic spinning which reduces number of
117  * times the locking task needs to sleep.
118  */
119 
120 /*
121  * __btrfs_tree_read_lock - lock extent buffer for read
122  * @eb:		the eb to be locked
123  * @nest:	the nesting level to be used for lockdep
124  *
125  * This takes the read lock on the extent buffer, using the specified nesting
126  * level for lockdep purposes.
127  */
128 void __btrfs_tree_read_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
129 {
130 	u64 start_ns = 0;
131 
132 	if (trace_btrfs_tree_read_lock_enabled())
133 		start_ns = ktime_get_ns();
134 
135 	down_read_nested(&eb->lock, nest);
136 	trace_btrfs_tree_read_lock(eb, start_ns);
137 }
138 
139 void btrfs_tree_read_lock(struct extent_buffer *eb)
140 {
141 	__btrfs_tree_read_lock(eb, BTRFS_NESTING_NORMAL);
142 }
143 
144 /*
145  * Try-lock for read.
146  *
147  * Return 1 if the rwlock has been taken, 0 otherwise
148  */
149 int btrfs_try_tree_read_lock(struct extent_buffer *eb)
150 {
151 	if (down_read_trylock(&eb->lock)) {
152 		trace_btrfs_try_tree_read_lock(eb);
153 		return 1;
154 	}
155 	return 0;
156 }
157 
158 /*
159  * Try-lock for write.
160  *
161  * Return 1 if the rwlock has been taken, 0 otherwise
162  */
163 int btrfs_try_tree_write_lock(struct extent_buffer *eb)
164 {
165 	if (down_write_trylock(&eb->lock)) {
166 		eb->lock_owner = current->pid;
167 		trace_btrfs_try_tree_write_lock(eb);
168 		return 1;
169 	}
170 	return 0;
171 }
172 
173 /*
174  * Release read lock.
175  */
176 void btrfs_tree_read_unlock(struct extent_buffer *eb)
177 {
178 	trace_btrfs_tree_read_unlock(eb);
179 	up_read(&eb->lock);
180 }
181 
182 /*
183  * __btrfs_tree_lock - lock eb for write
184  * @eb:		the eb to lock
185  * @nest:	the nesting to use for the lock
186  *
187  * Returns with the eb->lock write locked.
188  */
189 void __btrfs_tree_lock(struct extent_buffer *eb, enum btrfs_lock_nesting nest)
190 	__acquires(&eb->lock)
191 {
192 	u64 start_ns = 0;
193 
194 	if (trace_btrfs_tree_lock_enabled())
195 		start_ns = ktime_get_ns();
196 
197 	down_write_nested(&eb->lock, nest);
198 	eb->lock_owner = current->pid;
199 	trace_btrfs_tree_lock(eb, start_ns);
200 }
201 
202 void btrfs_tree_lock(struct extent_buffer *eb)
203 {
204 	__btrfs_tree_lock(eb, BTRFS_NESTING_NORMAL);
205 }
206 
207 /*
208  * Release the write lock.
209  */
210 void btrfs_tree_unlock(struct extent_buffer *eb)
211 {
212 	trace_btrfs_tree_unlock(eb);
213 	eb->lock_owner = 0;
214 	up_write(&eb->lock);
215 }
216 
217 /*
218  * This releases any locks held in the path starting at level and going all the
219  * way up to the root.
220  *
221  * btrfs_search_slot will keep the lock held on higher nodes in a few corner
222  * cases, such as COW of the block at slot zero in the node.  This ignores
223  * those rules, and it should only be called when there are no more updates to
224  * be done higher up in the tree.
225  */
226 void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
227 {
228 	int i;
229 
230 	if (path->keep_locks)
231 		return;
232 
233 	for (i = level; i < BTRFS_MAX_LEVEL; i++) {
234 		if (!path->nodes[i])
235 			continue;
236 		if (!path->locks[i])
237 			continue;
238 		btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
239 		path->locks[i] = 0;
240 	}
241 }
242 
243 /*
244  * Loop around taking references on and locking the root node of the tree until
245  * we end up with a lock on the root node.
246  *
247  * Return: root extent buffer with write lock held
248  */
249 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
250 {
251 	struct extent_buffer *eb;
252 
253 	while (1) {
254 		eb = btrfs_root_node(root);
255 
256 		btrfs_maybe_reset_lockdep_class(root, eb);
257 		btrfs_tree_lock(eb);
258 		if (eb == root->node)
259 			break;
260 		btrfs_tree_unlock(eb);
261 		free_extent_buffer(eb);
262 	}
263 	return eb;
264 }
265 
266 /*
267  * Loop around taking references on and locking the root node of the tree until
268  * we end up with a lock on the root node.
269  *
270  * Return: root extent buffer with read lock held
271  */
272 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
273 {
274 	struct extent_buffer *eb;
275 
276 	while (1) {
277 		eb = btrfs_root_node(root);
278 
279 		btrfs_maybe_reset_lockdep_class(root, eb);
280 		btrfs_tree_read_lock(eb);
281 		if (eb == root->node)
282 			break;
283 		btrfs_tree_read_unlock(eb);
284 		free_extent_buffer(eb);
285 	}
286 	return eb;
287 }
288 
289 /*
290  * Loop around taking references on and locking the root node of the tree in
291  * nowait mode until we end up with a lock on the root node or returning to
292  * avoid blocking.
293  *
294  * Return: root extent buffer with read lock held or -EAGAIN.
295  */
296 struct extent_buffer *btrfs_try_read_lock_root_node(struct btrfs_root *root)
297 {
298 	struct extent_buffer *eb;
299 
300 	while (1) {
301 		eb = btrfs_root_node(root);
302 		if (!btrfs_try_tree_read_lock(eb)) {
303 			free_extent_buffer(eb);
304 			return ERR_PTR(-EAGAIN);
305 		}
306 		if (eb == root->node)
307 			break;
308 		btrfs_tree_read_unlock(eb);
309 		free_extent_buffer(eb);
310 	}
311 	return eb;
312 }
313 
314 /*
315  * DREW locks
316  * ==========
317  *
318  * DREW stands for double-reader-writer-exclusion lock. It's used in situation
319  * where you want to provide A-B exclusion but not AA or BB.
320  *
321  * Currently implementation gives more priority to reader. If a reader and a
322  * writer both race to acquire their respective sides of the lock the writer
323  * would yield its lock as soon as it detects a concurrent reader. Additionally
324  * if there are pending readers no new writers would be allowed to come in and
325  * acquire the lock.
326  */
327 
328 int btrfs_drew_lock_init(struct btrfs_drew_lock *lock)
329 {
330 	int ret;
331 
332 	ret = percpu_counter_init(&lock->writers, 0, GFP_KERNEL);
333 	if (ret)
334 		return ret;
335 
336 	atomic_set(&lock->readers, 0);
337 	init_waitqueue_head(&lock->pending_readers);
338 	init_waitqueue_head(&lock->pending_writers);
339 
340 	return 0;
341 }
342 
343 void btrfs_drew_lock_destroy(struct btrfs_drew_lock *lock)
344 {
345 	percpu_counter_destroy(&lock->writers);
346 }
347 
348 /* Return true if acquisition is successful, false otherwise */
349 bool btrfs_drew_try_write_lock(struct btrfs_drew_lock *lock)
350 {
351 	if (atomic_read(&lock->readers))
352 		return false;
353 
354 	percpu_counter_inc(&lock->writers);
355 
356 	/* Ensure writers count is updated before we check for pending readers */
357 	smp_mb();
358 	if (atomic_read(&lock->readers)) {
359 		btrfs_drew_write_unlock(lock);
360 		return false;
361 	}
362 
363 	return true;
364 }
365 
366 void btrfs_drew_write_lock(struct btrfs_drew_lock *lock)
367 {
368 	while (true) {
369 		if (btrfs_drew_try_write_lock(lock))
370 			return;
371 		wait_event(lock->pending_writers, !atomic_read(&lock->readers));
372 	}
373 }
374 
375 void btrfs_drew_write_unlock(struct btrfs_drew_lock *lock)
376 {
377 	percpu_counter_dec(&lock->writers);
378 	cond_wake_up(&lock->pending_readers);
379 }
380 
381 void btrfs_drew_read_lock(struct btrfs_drew_lock *lock)
382 {
383 	atomic_inc(&lock->readers);
384 
385 	/*
386 	 * Ensure the pending reader count is perceieved BEFORE this reader
387 	 * goes to sleep in case of active writers. This guarantees new writers
388 	 * won't be allowed and that the current reader will be woken up when
389 	 * the last active writer finishes its jobs.
390 	 */
391 	smp_mb__after_atomic();
392 
393 	wait_event(lock->pending_readers,
394 		   percpu_counter_sum(&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