xref: /linux/fs/bcachefs/six.h (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 
3 #ifndef _LINUX_SIX_H
4 #define _LINUX_SIX_H
5 
6 /**
7  * DOC: SIX locks overview
8  *
9  * Shared/intent/exclusive locks: sleepable read/write locks, like rw semaphores
10  * but with an additional state: read/shared, intent, exclusive/write
11  *
12  * The purpose of the intent state is to allow for greater concurrency on tree
13  * structures without deadlocking. In general, a read can't be upgraded to a
14  * write lock without deadlocking, so an operation that updates multiple nodes
15  * will have to take write locks for the full duration of the operation.
16  *
17  * But by adding an intent state, which is exclusive with other intent locks but
18  * not with readers, we can take intent locks at the start of the operation,
19  * and then take write locks only for the actual update to each individual
20  * nodes, without deadlocking.
21  *
22  * Example usage:
23  *   six_lock_read(&foo->lock);
24  *   six_unlock_read(&foo->lock);
25  *
26  * An intent lock must be held before taking a write lock:
27  *   six_lock_intent(&foo->lock);
28  *   six_lock_write(&foo->lock);
29  *   six_unlock_write(&foo->lock);
30  *   six_unlock_intent(&foo->lock);
31  *
32  * Other operations:
33  *   six_trylock_read()
34  *   six_trylock_intent()
35  *   six_trylock_write()
36  *
37  *   six_lock_downgrade()	convert from intent to read
38  *   six_lock_tryupgrade()	attempt to convert from read to intent, may fail
39  *
40  * There are also interfaces that take the lock type as an enum:
41  *
42  *   six_lock_type(&foo->lock, SIX_LOCK_read);
43  *   six_trylock_convert(&foo->lock, SIX_LOCK_read, SIX_LOCK_intent)
44  *   six_lock_type(&foo->lock, SIX_LOCK_write);
45  *   six_unlock_type(&foo->lock, SIX_LOCK_write);
46  *   six_unlock_type(&foo->lock, SIX_LOCK_intent);
47  *
48  * Lock sequence numbers - unlock(), relock():
49  *
50  *   Locks embed sequences numbers, which are incremented on write lock/unlock.
51  *   This allows locks to be dropped and the retaken iff the state they protect
52  *   hasn't changed; this makes it much easier to avoid holding locks while e.g.
53  *   doing IO or allocating memory.
54  *
55  *   Example usage:
56  *     six_lock_read(&foo->lock);
57  *     u32 seq = six_lock_seq(&foo->lock);
58  *     six_unlock_read(&foo->lock);
59  *
60  *     some_operation_that_may_block();
61  *
62  *     if (six_relock_read(&foo->lock, seq)) { ... }
63  *
64  *   If the relock operation succeeds, it is as if the lock was never unlocked.
65  *
66  * Reentrancy:
67  *
68  *   Six locks are not by themselves reentrant, but have counters for both the
69  *   read and intent states that can be used to provide reentrancy by an upper
70  *   layer that tracks held locks. If a lock is known to already be held in the
71  *   read or intent state, six_lock_increment() can be used to bump the "lock
72  *   held in this state" counter, increasing the number of unlock calls that
73  *   will be required to fully unlock it.
74  *
75  *   Example usage:
76  *     six_lock_read(&foo->lock);
77  *     six_lock_increment(&foo->lock, SIX_LOCK_read);
78  *     six_unlock_read(&foo->lock);
79  *     six_unlock_read(&foo->lock);
80  *   foo->lock is now fully unlocked.
81  *
82  *   Since the intent state supercedes read, it's legal to increment the read
83  *   counter when holding an intent lock, but not the reverse.
84  *
85  *   A lock may only be held once for write: six_lock_increment(.., SIX_LOCK_write)
86  *   is not legal.
87  *
88  * should_sleep_fn:
89  *
90  *   There is a six_lock() variant that takes a function pointer that is called
91  *   immediately prior to schedule() when blocking, and may return an error to
92  *   abort.
93  *
94  *   One possible use for this feature is when objects being locked are part of
95  *   a cache and may reused, and lock ordering is based on a property of the
96  *   object that will change when the object is reused - i.e. logical key order.
97  *
98  *   If looking up an object in the cache may race with object reuse, and lock
99  *   ordering is required to prevent deadlock, object reuse may change the
100  *   correct lock order for that object and cause a deadlock. should_sleep_fn
101  *   can be used to check if the object is still the object we want and avoid
102  *   this deadlock.
103  *
104  * Wait list entry interface:
105  *
106  *   There is a six_lock() variant, six_lock_waiter(), that takes a pointer to a
107  *   wait list entry. By embedding six_lock_waiter into another object, and by
108  *   traversing lock waitlists, it is then possible for an upper layer to
109  *   implement full cycle detection for deadlock avoidance.
110  *
111  *   should_sleep_fn should be used for invoking the cycle detector, walking the
112  *   graph of held locks to check for a deadlock. The upper layer must track
113  *   held locks for each thread, and each thread's held locks must be reachable
114  *   from its six_lock_waiter object.
115  *
116  *   six_lock_waiter() will add the wait object to the waitlist re-trying taking
117  *   the lock, and before calling should_sleep_fn, and the wait object will not
118  *   be removed from the waitlist until either the lock has been successfully
119  *   acquired, or we aborted because should_sleep_fn returned an error.
120  *
121  *   Also, six_lock_waiter contains a timestamp, and waiters on a waitlist will
122  *   have timestamps in strictly ascending order - this is so the timestamp can
123  *   be used as a cursor for lock graph traverse.
124  */
125 
126 #include <linux/lockdep.h>
127 #include <linux/sched.h>
128 #include <linux/types.h>
129 
130 enum six_lock_type {
131 	SIX_LOCK_read,
132 	SIX_LOCK_intent,
133 	SIX_LOCK_write,
134 };
135 
136 struct six_lock {
137 	atomic_t		state;
138 	u32			seq;
139 	unsigned		intent_lock_recurse;
140 	struct task_struct	*owner;
141 	unsigned __percpu	*readers;
142 	raw_spinlock_t		wait_lock;
143 	struct list_head	wait_list;
144 #ifdef CONFIG_DEBUG_LOCK_ALLOC
145 	struct lockdep_map	dep_map;
146 #endif
147 };
148 
149 struct six_lock_waiter {
150 	struct list_head	list;
151 	struct task_struct	*task;
152 	enum six_lock_type	lock_want;
153 	bool			lock_acquired;
154 	u64			start_time;
155 };
156 
157 typedef int (*six_lock_should_sleep_fn)(struct six_lock *lock, void *);
158 
159 void six_lock_exit(struct six_lock *lock);
160 
161 enum six_lock_init_flags {
162 	SIX_LOCK_INIT_PCPU	= 1U << 0,
163 };
164 
165 void __six_lock_init(struct six_lock *lock, const char *name,
166 		     struct lock_class_key *key, enum six_lock_init_flags flags);
167 
168 /**
169  * six_lock_init - initialize a six lock
170  * @lock:	lock to initialize
171  * @flags:	optional flags, i.e. SIX_LOCK_INIT_PCPU
172  */
173 #define six_lock_init(lock, flags)					\
174 do {									\
175 	static struct lock_class_key __key;				\
176 									\
177 	__six_lock_init((lock), #lock, &__key, flags);			\
178 } while (0)
179 
180 /**
181  * six_lock_seq - obtain current lock sequence number
182  * @lock:	six_lock to obtain sequence number for
183  *
184  * @lock should be held for read or intent, and not write
185  *
186  * By saving the lock sequence number, we can unlock @lock and then (typically
187  * after some blocking operation) attempt to relock it: the relock will succeed
188  * if the sequence number hasn't changed, meaning no write locks have been taken
189  * and state corresponding to what @lock protects is still valid.
190  */
191 static inline u32 six_lock_seq(const struct six_lock *lock)
192 {
193 	return lock->seq;
194 }
195 
196 bool six_trylock_ip(struct six_lock *lock, enum six_lock_type type, unsigned long ip);
197 
198 /**
199  * six_trylock_type - attempt to take a six lock without blocking
200  * @lock:	lock to take
201  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
202  *
203  * Return: true on success, false on failure.
204  */
205 static inline bool six_trylock_type(struct six_lock *lock, enum six_lock_type type)
206 {
207 	return six_trylock_ip(lock, type, _THIS_IP_);
208 }
209 
210 int six_lock_ip_waiter(struct six_lock *lock, enum six_lock_type type,
211 		       struct six_lock_waiter *wait,
212 		       six_lock_should_sleep_fn should_sleep_fn, void *p,
213 		       unsigned long ip);
214 
215 /**
216  * six_lock_waiter - take a lock, with full waitlist interface
217  * @lock:	lock to take
218  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
219  * @wait:	pointer to wait object, which will be added to lock's waitlist
220  * @should_sleep_fn: callback run after adding to waitlist, immediately prior
221  *		to scheduling
222  * @p:		passed through to @should_sleep_fn
223  *
224  * This is a convenience wrapper around six_lock_ip_waiter(), see that function
225  * for full documentation.
226  *
227  * Return: 0 on success, or the return code from @should_sleep_fn on failure.
228  */
229 static inline int six_lock_waiter(struct six_lock *lock, enum six_lock_type type,
230 				  struct six_lock_waiter *wait,
231 				  six_lock_should_sleep_fn should_sleep_fn, void *p)
232 {
233 	return six_lock_ip_waiter(lock, type, wait, should_sleep_fn, p, _THIS_IP_);
234 }
235 
236 /**
237  * six_lock_ip - take a six lock lock
238  * @lock:	lock to take
239  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
240  * @should_sleep_fn: callback run after adding to waitlist, immediately prior
241  *		to scheduling
242  * @p:		passed through to @should_sleep_fn
243  * @ip:		ip parameter for lockdep/lockstat, i.e. _THIS_IP_
244  *
245  * Return: 0 on success, or the return code from @should_sleep_fn on failure.
246  */
247 static inline int six_lock_ip(struct six_lock *lock, enum six_lock_type type,
248 			      six_lock_should_sleep_fn should_sleep_fn, void *p,
249 			      unsigned long ip)
250 {
251 	struct six_lock_waiter wait;
252 
253 	return six_lock_ip_waiter(lock, type, &wait, should_sleep_fn, p, ip);
254 }
255 
256 /**
257  * six_lock_type - take a six lock lock
258  * @lock:	lock to take
259  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
260  * @should_sleep_fn: callback run after adding to waitlist, immediately prior
261  *		to scheduling
262  * @p:		passed through to @should_sleep_fn
263  *
264  * Return: 0 on success, or the return code from @should_sleep_fn on failure.
265  */
266 static inline int six_lock_type(struct six_lock *lock, enum six_lock_type type,
267 				six_lock_should_sleep_fn should_sleep_fn, void *p)
268 {
269 	struct six_lock_waiter wait;
270 
271 	return six_lock_ip_waiter(lock, type, &wait, should_sleep_fn, p, _THIS_IP_);
272 }
273 
274 bool six_relock_ip(struct six_lock *lock, enum six_lock_type type,
275 		   unsigned seq, unsigned long ip);
276 
277 /**
278  * six_relock_type - attempt to re-take a lock that was held previously
279  * @lock:	lock to take
280  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
281  * @seq:	lock sequence number obtained from six_lock_seq() while lock was
282  *		held previously
283  *
284  * Return: true on success, false on failure.
285  */
286 static inline bool six_relock_type(struct six_lock *lock, enum six_lock_type type,
287 				   unsigned seq)
288 {
289 	return six_relock_ip(lock, type, seq, _THIS_IP_);
290 }
291 
292 void six_unlock_ip(struct six_lock *lock, enum six_lock_type type, unsigned long ip);
293 
294 /**
295  * six_unlock_type - drop a six lock
296  * @lock:	lock to unlock
297  * @type:	SIX_LOCK_read, SIX_LOCK_intent, or SIX_LOCK_write
298  *
299  * When a lock is held multiple times (because six_lock_incement()) was used),
300  * this decrements the 'lock held' counter by one.
301  *
302  * For example:
303  * six_lock_read(&foo->lock);				read count 1
304  * six_lock_increment(&foo->lock, SIX_LOCK_read);	read count 2
305  * six_lock_unlock(&foo->lock, SIX_LOCK_read);		read count 1
306  * six_lock_unlock(&foo->lock, SIX_LOCK_read);		read count 0
307  */
308 static inline void six_unlock_type(struct six_lock *lock, enum six_lock_type type)
309 {
310 	six_unlock_ip(lock, type, _THIS_IP_);
311 }
312 
313 #define __SIX_LOCK(type)						\
314 static inline bool six_trylock_ip_##type(struct six_lock *lock, unsigned long ip)\
315 {									\
316 	return six_trylock_ip(lock, SIX_LOCK_##type, ip);		\
317 }									\
318 									\
319 static inline bool six_trylock_##type(struct six_lock *lock)		\
320 {									\
321 	return six_trylock_ip(lock, SIX_LOCK_##type, _THIS_IP_);	\
322 }									\
323 									\
324 static inline int six_lock_ip_waiter_##type(struct six_lock *lock,	\
325 			   struct six_lock_waiter *wait,		\
326 			   six_lock_should_sleep_fn should_sleep_fn, void *p,\
327 			   unsigned long ip)				\
328 {									\
329 	return six_lock_ip_waiter(lock, SIX_LOCK_##type, wait, should_sleep_fn, p, ip);\
330 }									\
331 									\
332 static inline int six_lock_ip_##type(struct six_lock *lock,		\
333 		    six_lock_should_sleep_fn should_sleep_fn, void *p,	\
334 		    unsigned long ip)					\
335 {									\
336 	return six_lock_ip(lock, SIX_LOCK_##type, should_sleep_fn, p, ip);\
337 }									\
338 									\
339 static inline bool six_relock_ip_##type(struct six_lock *lock, u32 seq, unsigned long ip)\
340 {									\
341 	return six_relock_ip(lock, SIX_LOCK_##type, seq, ip);		\
342 }									\
343 									\
344 static inline bool six_relock_##type(struct six_lock *lock, u32 seq)	\
345 {									\
346 	return six_relock_ip(lock, SIX_LOCK_##type, seq, _THIS_IP_);	\
347 }									\
348 									\
349 static inline int six_lock_##type(struct six_lock *lock,		\
350 				  six_lock_should_sleep_fn fn, void *p)\
351 {									\
352 	return six_lock_ip_##type(lock, fn, p, _THIS_IP_);		\
353 }									\
354 									\
355 static inline void six_unlock_ip_##type(struct six_lock *lock, unsigned long ip)	\
356 {									\
357 	six_unlock_ip(lock, SIX_LOCK_##type, ip);			\
358 }									\
359 									\
360 static inline void six_unlock_##type(struct six_lock *lock)		\
361 {									\
362 	six_unlock_ip(lock, SIX_LOCK_##type, _THIS_IP_);		\
363 }
364 
365 __SIX_LOCK(read)
366 __SIX_LOCK(intent)
367 __SIX_LOCK(write)
368 #undef __SIX_LOCK
369 
370 void six_lock_downgrade(struct six_lock *);
371 bool six_lock_tryupgrade(struct six_lock *);
372 bool six_trylock_convert(struct six_lock *, enum six_lock_type,
373 			 enum six_lock_type);
374 
375 void six_lock_increment(struct six_lock *, enum six_lock_type);
376 
377 void six_lock_wakeup_all(struct six_lock *);
378 
379 struct six_lock_count {
380 	unsigned n[3];
381 };
382 
383 struct six_lock_count six_lock_counts(struct six_lock *);
384 void six_lock_readers_add(struct six_lock *, int);
385 
386 #endif /* _LINUX_SIX_H */
387