]>
Commit | Line | Data |
---|---|---|
1a59d1b8 | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
1da177e4 LT |
2 | /* |
3 | * Fast Userspace Mutexes (which I call "Futexes!"). | |
4 | * (C) Rusty Russell, IBM 2002 | |
5 | * | |
6 | * Generalized futexes, futex requeueing, misc fixes by Ingo Molnar | |
7 | * (C) Copyright 2003 Red Hat Inc, All Rights Reserved | |
8 | * | |
9 | * Removed page pinning, fix privately mapped COW pages and other cleanups | |
10 | * (C) Copyright 2003, 2004 Jamie Lokier | |
11 | * | |
0771dfef IM |
12 | * Robust futex support started by Ingo Molnar |
13 | * (C) Copyright 2006 Red Hat Inc, All Rights Reserved | |
14 | * Thanks to Thomas Gleixner for suggestions, analysis and fixes. | |
15 | * | |
c87e2837 IM |
16 | * PI-futex support started by Ingo Molnar and Thomas Gleixner |
17 | * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> | |
18 | * Copyright (C) 2006 Timesys Corp., Thomas Gleixner <tglx@timesys.com> | |
19 | * | |
34f01cc1 ED |
20 | * PRIVATE futexes by Eric Dumazet |
21 | * Copyright (C) 2007 Eric Dumazet <dada1@cosmosbay.com> | |
22 | * | |
52400ba9 DH |
23 | * Requeue-PI support by Darren Hart <dvhltc@us.ibm.com> |
24 | * Copyright (C) IBM Corporation, 2009 | |
25 | * Thanks to Thomas Gleixner for conceptual design and careful reviews. | |
26 | * | |
1da177e4 LT |
27 | * Thanks to Ben LaHaise for yelling "hashed waitqueues" loudly |
28 | * enough at me, Linus for the original (flawed) idea, Matthew | |
29 | * Kirkwood for proof-of-concept implementation. | |
30 | * | |
31 | * "The futexes are also cursed." | |
32 | * "But they come in a choice of three flavours!" | |
1da177e4 | 33 | */ |
04e7712f | 34 | #include <linux/compat.h> |
1da177e4 | 35 | #include <linux/jhash.h> |
1da177e4 LT |
36 | #include <linux/pagemap.h> |
37 | #include <linux/syscalls.h> | |
88c8004f | 38 | #include <linux/freezer.h> |
57c8a661 | 39 | #include <linux/memblock.h> |
ab51fbab | 40 | #include <linux/fault-inject.h> |
c2f7d08c | 41 | #include <linux/time_namespace.h> |
b488893a | 42 | |
4732efbe | 43 | #include <asm/futex.h> |
1da177e4 | 44 | |
1696a8be | 45 | #include "locking/rtmutex_common.h" |
c87e2837 | 46 | |
99b60ce6 | 47 | /* |
d7e8af1a DB |
48 | * READ this before attempting to hack on futexes! |
49 | * | |
50 | * Basic futex operation and ordering guarantees | |
51 | * ============================================= | |
99b60ce6 TG |
52 | * |
53 | * The waiter reads the futex value in user space and calls | |
54 | * futex_wait(). This function computes the hash bucket and acquires | |
55 | * the hash bucket lock. After that it reads the futex user space value | |
b0c29f79 DB |
56 | * again and verifies that the data has not changed. If it has not changed |
57 | * it enqueues itself into the hash bucket, releases the hash bucket lock | |
58 | * and schedules. | |
99b60ce6 TG |
59 | * |
60 | * The waker side modifies the user space value of the futex and calls | |
b0c29f79 DB |
61 | * futex_wake(). This function computes the hash bucket and acquires the |
62 | * hash bucket lock. Then it looks for waiters on that futex in the hash | |
63 | * bucket and wakes them. | |
99b60ce6 | 64 | * |
b0c29f79 DB |
65 | * In futex wake up scenarios where no tasks are blocked on a futex, taking |
66 | * the hb spinlock can be avoided and simply return. In order for this | |
67 | * optimization to work, ordering guarantees must exist so that the waiter | |
68 | * being added to the list is acknowledged when the list is concurrently being | |
69 | * checked by the waker, avoiding scenarios like the following: | |
99b60ce6 TG |
70 | * |
71 | * CPU 0 CPU 1 | |
72 | * val = *futex; | |
73 | * sys_futex(WAIT, futex, val); | |
74 | * futex_wait(futex, val); | |
75 | * uval = *futex; | |
76 | * *futex = newval; | |
77 | * sys_futex(WAKE, futex); | |
78 | * futex_wake(futex); | |
79 | * if (queue_empty()) | |
80 | * return; | |
81 | * if (uval == val) | |
82 | * lock(hash_bucket(futex)); | |
83 | * queue(); | |
84 | * unlock(hash_bucket(futex)); | |
85 | * schedule(); | |
86 | * | |
87 | * This would cause the waiter on CPU 0 to wait forever because it | |
88 | * missed the transition of the user space value from val to newval | |
89 | * and the waker did not find the waiter in the hash bucket queue. | |
99b60ce6 | 90 | * |
b0c29f79 DB |
91 | * The correct serialization ensures that a waiter either observes |
92 | * the changed user space value before blocking or is woken by a | |
93 | * concurrent waker: | |
94 | * | |
95 | * CPU 0 CPU 1 | |
99b60ce6 TG |
96 | * val = *futex; |
97 | * sys_futex(WAIT, futex, val); | |
98 | * futex_wait(futex, val); | |
b0c29f79 | 99 | * |
d7e8af1a | 100 | * waiters++; (a) |
8ad7b378 DB |
101 | * smp_mb(); (A) <-- paired with -. |
102 | * | | |
103 | * lock(hash_bucket(futex)); | | |
104 | * | | |
105 | * uval = *futex; | | |
106 | * | *futex = newval; | |
107 | * | sys_futex(WAKE, futex); | |
108 | * | futex_wake(futex); | |
109 | * | | |
110 | * `--------> smp_mb(); (B) | |
99b60ce6 | 111 | * if (uval == val) |
b0c29f79 | 112 | * queue(); |
99b60ce6 | 113 | * unlock(hash_bucket(futex)); |
b0c29f79 DB |
114 | * schedule(); if (waiters) |
115 | * lock(hash_bucket(futex)); | |
d7e8af1a DB |
116 | * else wake_waiters(futex); |
117 | * waiters--; (b) unlock(hash_bucket(futex)); | |
b0c29f79 | 118 | * |
d7e8af1a DB |
119 | * Where (A) orders the waiters increment and the futex value read through |
120 | * atomic operations (see hb_waiters_inc) and where (B) orders the write | |
4b39f99c | 121 | * to futex and the waiters read (see hb_waiters_pending()). |
b0c29f79 DB |
122 | * |
123 | * This yields the following case (where X:=waiters, Y:=futex): | |
124 | * | |
125 | * X = Y = 0 | |
126 | * | |
127 | * w[X]=1 w[Y]=1 | |
128 | * MB MB | |
129 | * r[Y]=y r[X]=x | |
130 | * | |
131 | * Which guarantees that x==0 && y==0 is impossible; which translates back into | |
132 | * the guarantee that we cannot both miss the futex variable change and the | |
133 | * enqueue. | |
d7e8af1a DB |
134 | * |
135 | * Note that a new waiter is accounted for in (a) even when it is possible that | |
136 | * the wait call can return error, in which case we backtrack from it in (b). | |
137 | * Refer to the comment in queue_lock(). | |
138 | * | |
139 | * Similarly, in order to account for waiters being requeued on another | |
140 | * address we always increment the waiters for the destination bucket before | |
141 | * acquiring the lock. It then decrements them again after releasing it - | |
142 | * the code that actually moves the futex(es) between hash buckets (requeue_futex) | |
143 | * will do the additional required waiter count housekeeping. This is done for | |
144 | * double_lock_hb() and double_unlock_hb(), respectively. | |
99b60ce6 TG |
145 | */ |
146 | ||
04e7712f AB |
147 | #ifdef CONFIG_HAVE_FUTEX_CMPXCHG |
148 | #define futex_cmpxchg_enabled 1 | |
149 | #else | |
150 | static int __read_mostly futex_cmpxchg_enabled; | |
03b8c7b6 | 151 | #endif |
a0c1e907 | 152 | |
b41277dc DH |
153 | /* |
154 | * Futex flags used to encode options to functions and preserve them across | |
155 | * restarts. | |
156 | */ | |
784bdf3b TG |
157 | #ifdef CONFIG_MMU |
158 | # define FLAGS_SHARED 0x01 | |
159 | #else | |
160 | /* | |
161 | * NOMMU does not have per process address space. Let the compiler optimize | |
162 | * code away. | |
163 | */ | |
164 | # define FLAGS_SHARED 0x00 | |
165 | #endif | |
b41277dc DH |
166 | #define FLAGS_CLOCKRT 0x02 |
167 | #define FLAGS_HAS_TIMEOUT 0x04 | |
168 | ||
c87e2837 IM |
169 | /* |
170 | * Priority Inheritance state: | |
171 | */ | |
172 | struct futex_pi_state { | |
173 | /* | |
174 | * list of 'owned' pi_state instances - these have to be | |
175 | * cleaned up in do_exit() if the task exits prematurely: | |
176 | */ | |
177 | struct list_head list; | |
178 | ||
179 | /* | |
180 | * The PI object: | |
181 | */ | |
830e6acc | 182 | struct rt_mutex_base pi_mutex; |
c87e2837 IM |
183 | |
184 | struct task_struct *owner; | |
49262de2 | 185 | refcount_t refcount; |
c87e2837 IM |
186 | |
187 | union futex_key key; | |
3859a271 | 188 | } __randomize_layout; |
c87e2837 | 189 | |
d8d88fbb DH |
190 | /** |
191 | * struct futex_q - The hashed futex queue entry, one per waiting task | |
fb62db2b | 192 | * @list: priority-sorted list of tasks waiting on this futex |
d8d88fbb DH |
193 | * @task: the task waiting on the futex |
194 | * @lock_ptr: the hash bucket lock | |
195 | * @key: the key the futex is hashed on | |
196 | * @pi_state: optional priority inheritance state | |
197 | * @rt_waiter: rt_waiter storage for use with requeue_pi | |
198 | * @requeue_pi_key: the requeue_pi target futex key | |
199 | * @bitset: bitset for the optional bitmasked wakeup | |
200 | * | |
ac6424b9 | 201 | * We use this hashed waitqueue, instead of a normal wait_queue_entry_t, so |
1da177e4 LT |
202 | * we can wake only the relevant ones (hashed queues may be shared). |
203 | * | |
204 | * A futex_q has a woken state, just like tasks have TASK_RUNNING. | |
ec92d082 | 205 | * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. |
fb62db2b | 206 | * The order of wakeup is always to make the first condition true, then |
d8d88fbb DH |
207 | * the second. |
208 | * | |
209 | * PI futexes are typically woken before they are removed from the hash list via | |
210 | * the rt_mutex code. See unqueue_me_pi(). | |
1da177e4 LT |
211 | */ |
212 | struct futex_q { | |
ec92d082 | 213 | struct plist_node list; |
1da177e4 | 214 | |
d8d88fbb | 215 | struct task_struct *task; |
1da177e4 | 216 | spinlock_t *lock_ptr; |
1da177e4 | 217 | union futex_key key; |
c87e2837 | 218 | struct futex_pi_state *pi_state; |
52400ba9 | 219 | struct rt_mutex_waiter *rt_waiter; |
84bc4af5 | 220 | union futex_key *requeue_pi_key; |
cd689985 | 221 | u32 bitset; |
3859a271 | 222 | } __randomize_layout; |
1da177e4 | 223 | |
5bdb05f9 DH |
224 | static const struct futex_q futex_q_init = { |
225 | /* list gets initialized in queue_me()*/ | |
226 | .key = FUTEX_KEY_INIT, | |
227 | .bitset = FUTEX_BITSET_MATCH_ANY | |
228 | }; | |
229 | ||
1da177e4 | 230 | /* |
b2d0994b DH |
231 | * Hash buckets are shared by all the futex_keys that hash to the same |
232 | * location. Each key may have multiple futex_q structures, one for each task | |
233 | * waiting on a futex. | |
1da177e4 LT |
234 | */ |
235 | struct futex_hash_bucket { | |
11d4616b | 236 | atomic_t waiters; |
ec92d082 PP |
237 | spinlock_t lock; |
238 | struct plist_head chain; | |
a52b89eb | 239 | } ____cacheline_aligned_in_smp; |
1da177e4 | 240 | |
ac742d37 RV |
241 | /* |
242 | * The base of the bucket array and its size are always used together | |
243 | * (after initialization only in hash_futex()), so ensure that they | |
244 | * reside in the same cacheline. | |
245 | */ | |
246 | static struct { | |
247 | struct futex_hash_bucket *queues; | |
248 | unsigned long hashsize; | |
249 | } __futex_data __read_mostly __aligned(2*sizeof(long)); | |
250 | #define futex_queues (__futex_data.queues) | |
251 | #define futex_hashsize (__futex_data.hashsize) | |
a52b89eb | 252 | |
1da177e4 | 253 | |
ab51fbab DB |
254 | /* |
255 | * Fault injections for futexes. | |
256 | */ | |
257 | #ifdef CONFIG_FAIL_FUTEX | |
258 | ||
259 | static struct { | |
260 | struct fault_attr attr; | |
261 | ||
621a5f7a | 262 | bool ignore_private; |
ab51fbab DB |
263 | } fail_futex = { |
264 | .attr = FAULT_ATTR_INITIALIZER, | |
621a5f7a | 265 | .ignore_private = false, |
ab51fbab DB |
266 | }; |
267 | ||
268 | static int __init setup_fail_futex(char *str) | |
269 | { | |
270 | return setup_fault_attr(&fail_futex.attr, str); | |
271 | } | |
272 | __setup("fail_futex=", setup_fail_futex); | |
273 | ||
5d285a7f | 274 | static bool should_fail_futex(bool fshared) |
ab51fbab DB |
275 | { |
276 | if (fail_futex.ignore_private && !fshared) | |
277 | return false; | |
278 | ||
279 | return should_fail(&fail_futex.attr, 1); | |
280 | } | |
281 | ||
282 | #ifdef CONFIG_FAULT_INJECTION_DEBUG_FS | |
283 | ||
284 | static int __init fail_futex_debugfs(void) | |
285 | { | |
286 | umode_t mode = S_IFREG | S_IRUSR | S_IWUSR; | |
287 | struct dentry *dir; | |
288 | ||
289 | dir = fault_create_debugfs_attr("fail_futex", NULL, | |
290 | &fail_futex.attr); | |
291 | if (IS_ERR(dir)) | |
292 | return PTR_ERR(dir); | |
293 | ||
0365aeba GKH |
294 | debugfs_create_bool("ignore-private", mode, dir, |
295 | &fail_futex.ignore_private); | |
ab51fbab DB |
296 | return 0; |
297 | } | |
298 | ||
299 | late_initcall(fail_futex_debugfs); | |
300 | ||
301 | #endif /* CONFIG_FAULT_INJECTION_DEBUG_FS */ | |
302 | ||
303 | #else | |
304 | static inline bool should_fail_futex(bool fshared) | |
305 | { | |
306 | return false; | |
307 | } | |
308 | #endif /* CONFIG_FAIL_FUTEX */ | |
309 | ||
ba31c1a4 TG |
310 | #ifdef CONFIG_COMPAT |
311 | static void compat_exit_robust_list(struct task_struct *curr); | |
ba31c1a4 TG |
312 | #endif |
313 | ||
11d4616b LT |
314 | /* |
315 | * Reflects a new waiter being added to the waitqueue. | |
316 | */ | |
317 | static inline void hb_waiters_inc(struct futex_hash_bucket *hb) | |
b0c29f79 DB |
318 | { |
319 | #ifdef CONFIG_SMP | |
11d4616b | 320 | atomic_inc(&hb->waiters); |
b0c29f79 | 321 | /* |
11d4616b | 322 | * Full barrier (A), see the ordering comment above. |
b0c29f79 | 323 | */ |
4e857c58 | 324 | smp_mb__after_atomic(); |
11d4616b LT |
325 | #endif |
326 | } | |
327 | ||
328 | /* | |
329 | * Reflects a waiter being removed from the waitqueue by wakeup | |
330 | * paths. | |
331 | */ | |
332 | static inline void hb_waiters_dec(struct futex_hash_bucket *hb) | |
333 | { | |
334 | #ifdef CONFIG_SMP | |
335 | atomic_dec(&hb->waiters); | |
336 | #endif | |
337 | } | |
b0c29f79 | 338 | |
11d4616b LT |
339 | static inline int hb_waiters_pending(struct futex_hash_bucket *hb) |
340 | { | |
341 | #ifdef CONFIG_SMP | |
4b39f99c PZ |
342 | /* |
343 | * Full barrier (B), see the ordering comment above. | |
344 | */ | |
345 | smp_mb(); | |
11d4616b | 346 | return atomic_read(&hb->waiters); |
b0c29f79 | 347 | #else |
11d4616b | 348 | return 1; |
b0c29f79 DB |
349 | #endif |
350 | } | |
351 | ||
e8b61b3f TG |
352 | /** |
353 | * hash_futex - Return the hash bucket in the global hash | |
354 | * @key: Pointer to the futex key for which the hash is calculated | |
355 | * | |
356 | * We hash on the keys returned from get_futex_key (see below) and return the | |
357 | * corresponding hash bucket in the global hash. | |
1da177e4 LT |
358 | */ |
359 | static struct futex_hash_bucket *hash_futex(union futex_key *key) | |
360 | { | |
8d677436 | 361 | u32 hash = jhash2((u32 *)key, offsetof(typeof(*key), both.offset) / 4, |
1da177e4 | 362 | key->both.offset); |
8d677436 | 363 | |
a52b89eb | 364 | return &futex_queues[hash & (futex_hashsize - 1)]; |
1da177e4 LT |
365 | } |
366 | ||
e8b61b3f TG |
367 | |
368 | /** | |
369 | * match_futex - Check whether two futex keys are equal | |
370 | * @key1: Pointer to key1 | |
371 | * @key2: Pointer to key2 | |
372 | * | |
1da177e4 LT |
373 | * Return 1 if two futex_keys are equal, 0 otherwise. |
374 | */ | |
375 | static inline int match_futex(union futex_key *key1, union futex_key *key2) | |
376 | { | |
2bc87203 DH |
377 | return (key1 && key2 |
378 | && key1->both.word == key2->both.word | |
1da177e4 LT |
379 | && key1->both.ptr == key2->both.ptr |
380 | && key1->both.offset == key2->both.offset); | |
381 | } | |
382 | ||
96d4f267 LT |
383 | enum futex_access { |
384 | FUTEX_READ, | |
385 | FUTEX_WRITE | |
386 | }; | |
387 | ||
5ca584d9 WL |
388 | /** |
389 | * futex_setup_timer - set up the sleeping hrtimer. | |
390 | * @time: ptr to the given timeout value | |
391 | * @timeout: the hrtimer_sleeper structure to be set up | |
392 | * @flags: futex flags | |
393 | * @range_ns: optional range in ns | |
394 | * | |
395 | * Return: Initialized hrtimer_sleeper structure or NULL if no timeout | |
396 | * value given | |
397 | */ | |
398 | static inline struct hrtimer_sleeper * | |
399 | futex_setup_timer(ktime_t *time, struct hrtimer_sleeper *timeout, | |
400 | int flags, u64 range_ns) | |
401 | { | |
402 | if (!time) | |
403 | return NULL; | |
404 | ||
dbc1625f SAS |
405 | hrtimer_init_sleeper_on_stack(timeout, (flags & FLAGS_CLOCKRT) ? |
406 | CLOCK_REALTIME : CLOCK_MONOTONIC, | |
407 | HRTIMER_MODE_ABS); | |
5ca584d9 WL |
408 | /* |
409 | * If range_ns is 0, calling hrtimer_set_expires_range_ns() is | |
410 | * effectively the same as calling hrtimer_set_expires(). | |
411 | */ | |
412 | hrtimer_set_expires_range_ns(&timeout->timer, *time, range_ns); | |
413 | ||
414 | return timeout; | |
415 | } | |
416 | ||
8019ad13 PZ |
417 | /* |
418 | * Generate a machine wide unique identifier for this inode. | |
419 | * | |
420 | * This relies on u64 not wrapping in the life-time of the machine; which with | |
421 | * 1ns resolution means almost 585 years. | |
422 | * | |
423 | * This further relies on the fact that a well formed program will not unmap | |
424 | * the file while it has a (shared) futex waiting on it. This mapping will have | |
425 | * a file reference which pins the mount and inode. | |
426 | * | |
427 | * If for some reason an inode gets evicted and read back in again, it will get | |
428 | * a new sequence number and will _NOT_ match, even though it is the exact same | |
429 | * file. | |
430 | * | |
431 | * It is important that match_futex() will never have a false-positive, esp. | |
432 | * for PI futexes that can mess up the state. The above argues that false-negatives | |
433 | * are only possible for malformed programs. | |
434 | */ | |
435 | static u64 get_inode_sequence_number(struct inode *inode) | |
436 | { | |
437 | static atomic64_t i_seq; | |
438 | u64 old; | |
439 | ||
440 | /* Does the inode already have a sequence number? */ | |
441 | old = atomic64_read(&inode->i_sequence); | |
442 | if (likely(old)) | |
443 | return old; | |
444 | ||
445 | for (;;) { | |
446 | u64 new = atomic64_add_return(1, &i_seq); | |
447 | if (WARN_ON_ONCE(!new)) | |
448 | continue; | |
449 | ||
450 | old = atomic64_cmpxchg_relaxed(&inode->i_sequence, 0, new); | |
451 | if (old) | |
452 | return old; | |
453 | return new; | |
454 | } | |
455 | } | |
456 | ||
34f01cc1 | 457 | /** |
d96ee56c DH |
458 | * get_futex_key() - Get parameters which are the keys for a futex |
459 | * @uaddr: virtual address of the futex | |
92613085 | 460 | * @fshared: false for a PROCESS_PRIVATE futex, true for PROCESS_SHARED |
d96ee56c | 461 | * @key: address where result is stored. |
96d4f267 LT |
462 | * @rw: mapping needs to be read/write (values: FUTEX_READ, |
463 | * FUTEX_WRITE) | |
34f01cc1 | 464 | * |
6c23cbbd RD |
465 | * Return: a negative error code or 0 |
466 | * | |
7b4ff1ad | 467 | * The key words are stored in @key on success. |
1da177e4 | 468 | * |
8019ad13 | 469 | * For shared mappings (when @fshared), the key is: |
03c109d6 | 470 | * |
8019ad13 | 471 | * ( inode->i_sequence, page->index, offset_within_page ) |
03c109d6 | 472 | * |
8019ad13 PZ |
473 | * [ also see get_inode_sequence_number() ] |
474 | * | |
475 | * For private mappings (or when !@fshared), the key is: | |
03c109d6 | 476 | * |
8019ad13 PZ |
477 | * ( current->mm, address, 0 ) |
478 | * | |
479 | * This allows (cross process, where applicable) identification of the futex | |
480 | * without keeping the page pinned for the duration of the FUTEX_WAIT. | |
1da177e4 | 481 | * |
b2d0994b | 482 | * lock_page() might sleep, the caller should not hold a spinlock. |
1da177e4 | 483 | */ |
92613085 AA |
484 | static int get_futex_key(u32 __user *uaddr, bool fshared, union futex_key *key, |
485 | enum futex_access rw) | |
1da177e4 | 486 | { |
e2970f2f | 487 | unsigned long address = (unsigned long)uaddr; |
1da177e4 | 488 | struct mm_struct *mm = current->mm; |
077fa7ae | 489 | struct page *page, *tail; |
14d27abd | 490 | struct address_space *mapping; |
9ea71503 | 491 | int err, ro = 0; |
1da177e4 LT |
492 | |
493 | /* | |
494 | * The futex address must be "naturally" aligned. | |
495 | */ | |
e2970f2f | 496 | key->both.offset = address % PAGE_SIZE; |
34f01cc1 | 497 | if (unlikely((address % sizeof(u32)) != 0)) |
1da177e4 | 498 | return -EINVAL; |
e2970f2f | 499 | address -= key->both.offset; |
1da177e4 | 500 | |
96d4f267 | 501 | if (unlikely(!access_ok(uaddr, sizeof(u32)))) |
5cdec2d8 LT |
502 | return -EFAULT; |
503 | ||
ab51fbab DB |
504 | if (unlikely(should_fail_futex(fshared))) |
505 | return -EFAULT; | |
506 | ||
34f01cc1 ED |
507 | /* |
508 | * PROCESS_PRIVATE futexes are fast. | |
509 | * As the mm cannot disappear under us and the 'key' only needs | |
510 | * virtual address, we dont even have to find the underlying vma. | |
511 | * Note : We do have to check 'uaddr' is a valid user address, | |
512 | * but access_ok() should be faster than find_vma() | |
513 | */ | |
514 | if (!fshared) { | |
34f01cc1 ED |
515 | key->private.mm = mm; |
516 | key->private.address = address; | |
517 | return 0; | |
518 | } | |
1da177e4 | 519 | |
38d47c1b | 520 | again: |
ab51fbab | 521 | /* Ignore any VERIFY_READ mapping (futex common case) */ |
92613085 | 522 | if (unlikely(should_fail_futex(true))) |
ab51fbab DB |
523 | return -EFAULT; |
524 | ||
73b0140b | 525 | err = get_user_pages_fast(address, 1, FOLL_WRITE, &page); |
9ea71503 SB |
526 | /* |
527 | * If write access is not required (eg. FUTEX_WAIT), try | |
528 | * and get read-only access. | |
529 | */ | |
96d4f267 | 530 | if (err == -EFAULT && rw == FUTEX_READ) { |
9ea71503 SB |
531 | err = get_user_pages_fast(address, 1, 0, &page); |
532 | ro = 1; | |
533 | } | |
38d47c1b PZ |
534 | if (err < 0) |
535 | return err; | |
9ea71503 SB |
536 | else |
537 | err = 0; | |
38d47c1b | 538 | |
65d8fc77 MG |
539 | /* |
540 | * The treatment of mapping from this point on is critical. The page | |
541 | * lock protects many things but in this context the page lock | |
542 | * stabilizes mapping, prevents inode freeing in the shared | |
543 | * file-backed region case and guards against movement to swap cache. | |
544 | * | |
545 | * Strictly speaking the page lock is not needed in all cases being | |
546 | * considered here and page lock forces unnecessarily serialization | |
547 | * From this point on, mapping will be re-verified if necessary and | |
548 | * page lock will be acquired only if it is unavoidable | |
077fa7ae MG |
549 | * |
550 | * Mapping checks require the head page for any compound page so the | |
551 | * head page and mapping is looked up now. For anonymous pages, it | |
552 | * does not matter if the page splits in the future as the key is | |
553 | * based on the address. For filesystem-backed pages, the tail is | |
554 | * required as the index of the page determines the key. For | |
555 | * base pages, there is no tail page and tail == page. | |
65d8fc77 | 556 | */ |
077fa7ae | 557 | tail = page; |
65d8fc77 MG |
558 | page = compound_head(page); |
559 | mapping = READ_ONCE(page->mapping); | |
560 | ||
e6780f72 | 561 | /* |
14d27abd | 562 | * If page->mapping is NULL, then it cannot be a PageAnon |
e6780f72 HD |
563 | * page; but it might be the ZERO_PAGE or in the gate area or |
564 | * in a special mapping (all cases which we are happy to fail); | |
565 | * or it may have been a good file page when get_user_pages_fast | |
566 | * found it, but truncated or holepunched or subjected to | |
567 | * invalidate_complete_page2 before we got the page lock (also | |
568 | * cases which we are happy to fail). And we hold a reference, | |
569 | * so refcount care in invalidate_complete_page's remove_mapping | |
570 | * prevents drop_caches from setting mapping to NULL beneath us. | |
571 | * | |
572 | * The case we do have to guard against is when memory pressure made | |
573 | * shmem_writepage move it from filecache to swapcache beneath us: | |
14d27abd | 574 | * an unlikely race, but we do need to retry for page->mapping. |
e6780f72 | 575 | */ |
65d8fc77 MG |
576 | if (unlikely(!mapping)) { |
577 | int shmem_swizzled; | |
578 | ||
579 | /* | |
580 | * Page lock is required to identify which special case above | |
581 | * applies. If this is really a shmem page then the page lock | |
582 | * will prevent unexpected transitions. | |
583 | */ | |
584 | lock_page(page); | |
585 | shmem_swizzled = PageSwapCache(page) || page->mapping; | |
14d27abd KS |
586 | unlock_page(page); |
587 | put_page(page); | |
65d8fc77 | 588 | |
e6780f72 HD |
589 | if (shmem_swizzled) |
590 | goto again; | |
65d8fc77 | 591 | |
e6780f72 | 592 | return -EFAULT; |
38d47c1b | 593 | } |
1da177e4 LT |
594 | |
595 | /* | |
596 | * Private mappings are handled in a simple way. | |
597 | * | |
65d8fc77 MG |
598 | * If the futex key is stored on an anonymous page, then the associated |
599 | * object is the mm which is implicitly pinned by the calling process. | |
600 | * | |
1da177e4 LT |
601 | * NOTE: When userspace waits on a MAP_SHARED mapping, even if |
602 | * it's a read-only handle, it's expected that futexes attach to | |
38d47c1b | 603 | * the object not the particular process. |
1da177e4 | 604 | */ |
14d27abd | 605 | if (PageAnon(page)) { |
9ea71503 SB |
606 | /* |
607 | * A RO anonymous page will never change and thus doesn't make | |
608 | * sense for futex operations. | |
609 | */ | |
92613085 | 610 | if (unlikely(should_fail_futex(true)) || ro) { |
9ea71503 SB |
611 | err = -EFAULT; |
612 | goto out; | |
613 | } | |
614 | ||
38d47c1b | 615 | key->both.offset |= FUT_OFF_MMSHARED; /* ref taken on mm */ |
1da177e4 | 616 | key->private.mm = mm; |
e2970f2f | 617 | key->private.address = address; |
65d8fc77 | 618 | |
38d47c1b | 619 | } else { |
65d8fc77 MG |
620 | struct inode *inode; |
621 | ||
622 | /* | |
623 | * The associated futex object in this case is the inode and | |
624 | * the page->mapping must be traversed. Ordinarily this should | |
625 | * be stabilised under page lock but it's not strictly | |
626 | * necessary in this case as we just want to pin the inode, not | |
627 | * update the radix tree or anything like that. | |
628 | * | |
629 | * The RCU read lock is taken as the inode is finally freed | |
630 | * under RCU. If the mapping still matches expectations then the | |
631 | * mapping->host can be safely accessed as being a valid inode. | |
632 | */ | |
633 | rcu_read_lock(); | |
634 | ||
635 | if (READ_ONCE(page->mapping) != mapping) { | |
636 | rcu_read_unlock(); | |
637 | put_page(page); | |
638 | ||
639 | goto again; | |
640 | } | |
641 | ||
642 | inode = READ_ONCE(mapping->host); | |
643 | if (!inode) { | |
644 | rcu_read_unlock(); | |
645 | put_page(page); | |
646 | ||
647 | goto again; | |
648 | } | |
649 | ||
38d47c1b | 650 | key->both.offset |= FUT_OFF_INODE; /* inode-based key */ |
8019ad13 | 651 | key->shared.i_seq = get_inode_sequence_number(inode); |
fe19bd3d | 652 | key->shared.pgoff = page_to_pgoff(tail); |
65d8fc77 | 653 | rcu_read_unlock(); |
1da177e4 LT |
654 | } |
655 | ||
9ea71503 | 656 | out: |
14d27abd | 657 | put_page(page); |
9ea71503 | 658 | return err; |
1da177e4 LT |
659 | } |
660 | ||
d96ee56c DH |
661 | /** |
662 | * fault_in_user_writeable() - Fault in user address and verify RW access | |
d0725992 TG |
663 | * @uaddr: pointer to faulting user space address |
664 | * | |
665 | * Slow path to fixup the fault we just took in the atomic write | |
666 | * access to @uaddr. | |
667 | * | |
fb62db2b | 668 | * We have no generic implementation of a non-destructive write to the |
d0725992 TG |
669 | * user address. We know that we faulted in the atomic pagefault |
670 | * disabled section so we can as well avoid the #PF overhead by | |
671 | * calling get_user_pages() right away. | |
672 | */ | |
673 | static int fault_in_user_writeable(u32 __user *uaddr) | |
674 | { | |
722d0172 AK |
675 | struct mm_struct *mm = current->mm; |
676 | int ret; | |
677 | ||
d8ed45c5 | 678 | mmap_read_lock(mm); |
64019a2e | 679 | ret = fixup_user_fault(mm, (unsigned long)uaddr, |
4a9e1cda | 680 | FAULT_FLAG_WRITE, NULL); |
d8ed45c5 | 681 | mmap_read_unlock(mm); |
722d0172 | 682 | |
d0725992 TG |
683 | return ret < 0 ? ret : 0; |
684 | } | |
685 | ||
4b1c486b DH |
686 | /** |
687 | * futex_top_waiter() - Return the highest priority waiter on a futex | |
d96ee56c DH |
688 | * @hb: the hash bucket the futex_q's reside in |
689 | * @key: the futex key (to distinguish it from other futex futex_q's) | |
4b1c486b DH |
690 | * |
691 | * Must be called with the hb lock held. | |
692 | */ | |
693 | static struct futex_q *futex_top_waiter(struct futex_hash_bucket *hb, | |
694 | union futex_key *key) | |
695 | { | |
696 | struct futex_q *this; | |
697 | ||
698 | plist_for_each_entry(this, &hb->chain, list) { | |
699 | if (match_futex(&this->key, key)) | |
700 | return this; | |
701 | } | |
702 | return NULL; | |
703 | } | |
704 | ||
37a9d912 ML |
705 | static int cmpxchg_futex_value_locked(u32 *curval, u32 __user *uaddr, |
706 | u32 uval, u32 newval) | |
36cf3b5c | 707 | { |
37a9d912 | 708 | int ret; |
36cf3b5c TG |
709 | |
710 | pagefault_disable(); | |
37a9d912 | 711 | ret = futex_atomic_cmpxchg_inatomic(curval, uaddr, uval, newval); |
36cf3b5c TG |
712 | pagefault_enable(); |
713 | ||
37a9d912 | 714 | return ret; |
36cf3b5c TG |
715 | } |
716 | ||
717 | static int get_futex_value_locked(u32 *dest, u32 __user *from) | |
1da177e4 LT |
718 | { |
719 | int ret; | |
720 | ||
a866374a | 721 | pagefault_disable(); |
bd28b145 | 722 | ret = __get_user(*dest, from); |
a866374a | 723 | pagefault_enable(); |
1da177e4 LT |
724 | |
725 | return ret ? -EFAULT : 0; | |
726 | } | |
727 | ||
c87e2837 IM |
728 | |
729 | /* | |
730 | * PI code: | |
731 | */ | |
732 | static int refill_pi_state_cache(void) | |
733 | { | |
734 | struct futex_pi_state *pi_state; | |
735 | ||
736 | if (likely(current->pi_state_cache)) | |
737 | return 0; | |
738 | ||
4668edc3 | 739 | pi_state = kzalloc(sizeof(*pi_state), GFP_KERNEL); |
c87e2837 IM |
740 | |
741 | if (!pi_state) | |
742 | return -ENOMEM; | |
743 | ||
c87e2837 IM |
744 | INIT_LIST_HEAD(&pi_state->list); |
745 | /* pi_mutex gets initialized later */ | |
746 | pi_state->owner = NULL; | |
49262de2 | 747 | refcount_set(&pi_state->refcount, 1); |
38d47c1b | 748 | pi_state->key = FUTEX_KEY_INIT; |
c87e2837 IM |
749 | |
750 | current->pi_state_cache = pi_state; | |
751 | ||
752 | return 0; | |
753 | } | |
754 | ||
bf92cf3a | 755 | static struct futex_pi_state *alloc_pi_state(void) |
c87e2837 IM |
756 | { |
757 | struct futex_pi_state *pi_state = current->pi_state_cache; | |
758 | ||
759 | WARN_ON(!pi_state); | |
760 | current->pi_state_cache = NULL; | |
761 | ||
762 | return pi_state; | |
763 | } | |
764 | ||
c5cade20 TG |
765 | static void pi_state_update_owner(struct futex_pi_state *pi_state, |
766 | struct task_struct *new_owner) | |
767 | { | |
768 | struct task_struct *old_owner = pi_state->owner; | |
769 | ||
770 | lockdep_assert_held(&pi_state->pi_mutex.wait_lock); | |
771 | ||
772 | if (old_owner) { | |
773 | raw_spin_lock(&old_owner->pi_lock); | |
774 | WARN_ON(list_empty(&pi_state->list)); | |
775 | list_del_init(&pi_state->list); | |
776 | raw_spin_unlock(&old_owner->pi_lock); | |
777 | } | |
778 | ||
779 | if (new_owner) { | |
780 | raw_spin_lock(&new_owner->pi_lock); | |
781 | WARN_ON(!list_empty(&pi_state->list)); | |
782 | list_add(&pi_state->list, &new_owner->pi_state_list); | |
783 | pi_state->owner = new_owner; | |
784 | raw_spin_unlock(&new_owner->pi_lock); | |
785 | } | |
786 | } | |
787 | ||
bf92cf3a PZ |
788 | static void get_pi_state(struct futex_pi_state *pi_state) |
789 | { | |
49262de2 | 790 | WARN_ON_ONCE(!refcount_inc_not_zero(&pi_state->refcount)); |
bf92cf3a PZ |
791 | } |
792 | ||
30a6b803 | 793 | /* |
29e9ee5d TG |
794 | * Drops a reference to the pi_state object and frees or caches it |
795 | * when the last reference is gone. | |
30a6b803 | 796 | */ |
29e9ee5d | 797 | static void put_pi_state(struct futex_pi_state *pi_state) |
c87e2837 | 798 | { |
30a6b803 BS |
799 | if (!pi_state) |
800 | return; | |
801 | ||
49262de2 | 802 | if (!refcount_dec_and_test(&pi_state->refcount)) |
c87e2837 IM |
803 | return; |
804 | ||
805 | /* | |
806 | * If pi_state->owner is NULL, the owner is most probably dying | |
807 | * and has cleaned up the pi_state already | |
808 | */ | |
809 | if (pi_state->owner) { | |
1e106aa3 | 810 | unsigned long flags; |
c87e2837 | 811 | |
1e106aa3 | 812 | raw_spin_lock_irqsave(&pi_state->pi_mutex.wait_lock, flags); |
6ccc84f9 | 813 | pi_state_update_owner(pi_state, NULL); |
2156ac19 | 814 | rt_mutex_proxy_unlock(&pi_state->pi_mutex); |
1e106aa3 | 815 | raw_spin_unlock_irqrestore(&pi_state->pi_mutex.wait_lock, flags); |
c87e2837 IM |
816 | } |
817 | ||
c74aef2d | 818 | if (current->pi_state_cache) { |
c87e2837 | 819 | kfree(pi_state); |
c74aef2d | 820 | } else { |
c87e2837 IM |
821 | /* |
822 | * pi_state->list is already empty. | |
823 | * clear pi_state->owner. | |
824 | * refcount is at 0 - put it back to 1. | |
825 | */ | |
826 | pi_state->owner = NULL; | |
49262de2 | 827 | refcount_set(&pi_state->refcount, 1); |
c87e2837 IM |
828 | current->pi_state_cache = pi_state; |
829 | } | |
830 | } | |
831 | ||
bc2eecd7 NP |
832 | #ifdef CONFIG_FUTEX_PI |
833 | ||
c87e2837 IM |
834 | /* |
835 | * This task is holding PI mutexes at exit time => bad. | |
836 | * Kernel cleans up PI-state, but userspace is likely hosed. | |
837 | * (Robust-futex cleanup is separate and might save the day for userspace.) | |
838 | */ | |
ba31c1a4 | 839 | static void exit_pi_state_list(struct task_struct *curr) |
c87e2837 | 840 | { |
c87e2837 IM |
841 | struct list_head *next, *head = &curr->pi_state_list; |
842 | struct futex_pi_state *pi_state; | |
627371d7 | 843 | struct futex_hash_bucket *hb; |
38d47c1b | 844 | union futex_key key = FUTEX_KEY_INIT; |
c87e2837 | 845 | |
a0c1e907 TG |
846 | if (!futex_cmpxchg_enabled) |
847 | return; | |
c87e2837 IM |
848 | /* |
849 | * We are a ZOMBIE and nobody can enqueue itself on | |
850 | * pi_state_list anymore, but we have to be careful | |
627371d7 | 851 | * versus waiters unqueueing themselves: |
c87e2837 | 852 | */ |
1d615482 | 853 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 854 | while (!list_empty(head)) { |
c87e2837 IM |
855 | next = head->next; |
856 | pi_state = list_entry(next, struct futex_pi_state, list); | |
857 | key = pi_state->key; | |
627371d7 | 858 | hb = hash_futex(&key); |
153fbd12 PZ |
859 | |
860 | /* | |
861 | * We can race against put_pi_state() removing itself from the | |
862 | * list (a waiter going away). put_pi_state() will first | |
863 | * decrement the reference count and then modify the list, so | |
864 | * its possible to see the list entry but fail this reference | |
865 | * acquire. | |
866 | * | |
867 | * In that case; drop the locks to let put_pi_state() make | |
868 | * progress and retry the loop. | |
869 | */ | |
49262de2 | 870 | if (!refcount_inc_not_zero(&pi_state->refcount)) { |
153fbd12 PZ |
871 | raw_spin_unlock_irq(&curr->pi_lock); |
872 | cpu_relax(); | |
873 | raw_spin_lock_irq(&curr->pi_lock); | |
874 | continue; | |
875 | } | |
1d615482 | 876 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 877 | |
c87e2837 | 878 | spin_lock(&hb->lock); |
c74aef2d PZ |
879 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
880 | raw_spin_lock(&curr->pi_lock); | |
627371d7 IM |
881 | /* |
882 | * We dropped the pi-lock, so re-check whether this | |
883 | * task still owns the PI-state: | |
884 | */ | |
c87e2837 | 885 | if (head->next != next) { |
153fbd12 | 886 | /* retain curr->pi_lock for the loop invariant */ |
c74aef2d | 887 | raw_spin_unlock(&pi_state->pi_mutex.wait_lock); |
c87e2837 | 888 | spin_unlock(&hb->lock); |
153fbd12 | 889 | put_pi_state(pi_state); |
c87e2837 IM |
890 | continue; |
891 | } | |
892 | ||
c87e2837 | 893 | WARN_ON(pi_state->owner != curr); |
627371d7 IM |
894 | WARN_ON(list_empty(&pi_state->list)); |
895 | list_del_init(&pi_state->list); | |
c87e2837 | 896 | pi_state->owner = NULL; |
c87e2837 | 897 | |
153fbd12 | 898 | raw_spin_unlock(&curr->pi_lock); |
c74aef2d | 899 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
c87e2837 IM |
900 | spin_unlock(&hb->lock); |
901 | ||
16ffa12d PZ |
902 | rt_mutex_futex_unlock(&pi_state->pi_mutex); |
903 | put_pi_state(pi_state); | |
904 | ||
1d615482 | 905 | raw_spin_lock_irq(&curr->pi_lock); |
c87e2837 | 906 | } |
1d615482 | 907 | raw_spin_unlock_irq(&curr->pi_lock); |
c87e2837 | 908 | } |
ba31c1a4 TG |
909 | #else |
910 | static inline void exit_pi_state_list(struct task_struct *curr) { } | |
bc2eecd7 NP |
911 | #endif |
912 | ||
54a21788 TG |
913 | /* |
914 | * We need to check the following states: | |
915 | * | |
916 | * Waiter | pi_state | pi->owner | uTID | uODIED | ? | |
917 | * | |
918 | * [1] NULL | --- | --- | 0 | 0/1 | Valid | |
919 | * [2] NULL | --- | --- | >0 | 0/1 | Valid | |
920 | * | |
921 | * [3] Found | NULL | -- | Any | 0/1 | Invalid | |
922 | * | |
923 | * [4] Found | Found | NULL | 0 | 1 | Valid | |
924 | * [5] Found | Found | NULL | >0 | 1 | Invalid | |
925 | * | |
926 | * [6] Found | Found | task | 0 | 1 | Valid | |
927 | * | |
928 | * [7] Found | Found | NULL | Any | 0 | Invalid | |
929 | * | |
930 | * [8] Found | Found | task | ==taskTID | 0/1 | Valid | |
931 | * [9] Found | Found | task | 0 | 0 | Invalid | |
932 | * [10] Found | Found | task | !=taskTID | 0/1 | Invalid | |
933 | * | |
934 | * [1] Indicates that the kernel can acquire the futex atomically. We | |
7b7b8a2c | 935 | * came here due to a stale FUTEX_WAITERS/FUTEX_OWNER_DIED bit. |
54a21788 TG |
936 | * |
937 | * [2] Valid, if TID does not belong to a kernel thread. If no matching | |
938 | * thread is found then it indicates that the owner TID has died. | |
939 | * | |
940 | * [3] Invalid. The waiter is queued on a non PI futex | |
941 | * | |
942 | * [4] Valid state after exit_robust_list(), which sets the user space | |
943 | * value to FUTEX_WAITERS | FUTEX_OWNER_DIED. | |
944 | * | |
945 | * [5] The user space value got manipulated between exit_robust_list() | |
946 | * and exit_pi_state_list() | |
947 | * | |
948 | * [6] Valid state after exit_pi_state_list() which sets the new owner in | |
949 | * the pi_state but cannot access the user space value. | |
950 | * | |
951 | * [7] pi_state->owner can only be NULL when the OWNER_DIED bit is set. | |
952 | * | |
953 | * [8] Owner and user space value match | |
954 | * | |
955 | * [9] There is no transient state which sets the user space TID to 0 | |
956 | * except exit_robust_list(), but this is indicated by the | |
957 | * FUTEX_OWNER_DIED bit. See [4] | |
958 | * | |
959 | * [10] There is no transient state which leaves owner and user space | |
34b1a1ce TG |
960 | * TID out of sync. Except one error case where the kernel is denied |
961 | * write access to the user address, see fixup_pi_state_owner(). | |
734009e9 PZ |
962 | * |
963 | * | |
964 | * Serialization and lifetime rules: | |
965 | * | |
966 | * hb->lock: | |
967 | * | |
968 | * hb -> futex_q, relation | |
969 | * futex_q -> pi_state, relation | |
970 | * | |
971 | * (cannot be raw because hb can contain arbitrary amount | |
972 | * of futex_q's) | |
973 | * | |
974 | * pi_mutex->wait_lock: | |
975 | * | |
976 | * {uval, pi_state} | |
977 | * | |
978 | * (and pi_mutex 'obviously') | |
979 | * | |
980 | * p->pi_lock: | |
981 | * | |
982 | * p->pi_state_list -> pi_state->list, relation | |
c2e4bfe0 | 983 | * pi_mutex->owner -> pi_state->owner, relation |
734009e9 PZ |
984 | * |
985 | * pi_state->refcount: | |
986 | * | |
987 | * pi_state lifetime | |
988 | * | |
989 | * | |
990 | * Lock order: | |
991 | * | |
992 | * hb->lock | |
993 | * pi_mutex->wait_lock | |
994 | * p->pi_lock | |
995 | * | |
54a21788 | 996 | */ |
e60cbc5c TG |
997 | |
998 | /* | |
999 | * Validate that the existing waiter has a pi_state and sanity check | |
1000 | * the pi_state against the user space value. If correct, attach to | |
1001 | * it. | |
1002 | */ | |
734009e9 PZ |
1003 | static int attach_to_pi_state(u32 __user *uaddr, u32 uval, |
1004 | struct futex_pi_state *pi_state, | |
e60cbc5c | 1005 | struct futex_pi_state **ps) |
c87e2837 | 1006 | { |
778e9a9c | 1007 | pid_t pid = uval & FUTEX_TID_MASK; |
94ffac5d PZ |
1008 | u32 uval2; |
1009 | int ret; | |
c87e2837 | 1010 | |
e60cbc5c TG |
1011 | /* |
1012 | * Userspace might have messed up non-PI and PI futexes [3] | |
1013 | */ | |
1014 | if (unlikely(!pi_state)) | |
1015 | return -EINVAL; | |
06a9ec29 | 1016 | |
734009e9 PZ |
1017 | /* |
1018 | * We get here with hb->lock held, and having found a | |
1019 | * futex_top_waiter(). This means that futex_lock_pi() of said futex_q | |
1020 | * has dropped the hb->lock in between queue_me() and unqueue_me_pi(), | |
1021 | * which in turn means that futex_lock_pi() still has a reference on | |
1022 | * our pi_state. | |
16ffa12d PZ |
1023 | * |
1024 | * The waiter holding a reference on @pi_state also protects against | |
1025 | * the unlocked put_pi_state() in futex_unlock_pi(), futex_lock_pi() | |
1026 | * and futex_wait_requeue_pi() as it cannot go to 0 and consequently | |
1027 | * free pi_state before we can take a reference ourselves. | |
734009e9 | 1028 | */ |
49262de2 | 1029 | WARN_ON(!refcount_read(&pi_state->refcount)); |
59647b6a | 1030 | |
734009e9 PZ |
1031 | /* |
1032 | * Now that we have a pi_state, we can acquire wait_lock | |
1033 | * and do the state validation. | |
1034 | */ | |
1035 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); | |
1036 | ||
1037 | /* | |
1038 | * Since {uval, pi_state} is serialized by wait_lock, and our current | |
1039 | * uval was read without holding it, it can have changed. Verify it | |
1040 | * still is what we expect it to be, otherwise retry the entire | |
1041 | * operation. | |
1042 | */ | |
1043 | if (get_futex_value_locked(&uval2, uaddr)) | |
1044 | goto out_efault; | |
1045 | ||
1046 | if (uval != uval2) | |
1047 | goto out_eagain; | |
1048 | ||
e60cbc5c TG |
1049 | /* |
1050 | * Handle the owner died case: | |
1051 | */ | |
1052 | if (uval & FUTEX_OWNER_DIED) { | |
bd1dbcc6 | 1053 | /* |
e60cbc5c TG |
1054 | * exit_pi_state_list sets owner to NULL and wakes the |
1055 | * topmost waiter. The task which acquires the | |
1056 | * pi_state->rt_mutex will fixup owner. | |
bd1dbcc6 | 1057 | */ |
e60cbc5c | 1058 | if (!pi_state->owner) { |
59647b6a | 1059 | /* |
e60cbc5c TG |
1060 | * No pi state owner, but the user space TID |
1061 | * is not 0. Inconsistent state. [5] | |
59647b6a | 1062 | */ |
e60cbc5c | 1063 | if (pid) |
734009e9 | 1064 | goto out_einval; |
bd1dbcc6 | 1065 | /* |
e60cbc5c | 1066 | * Take a ref on the state and return success. [4] |
866293ee | 1067 | */ |
734009e9 | 1068 | goto out_attach; |
c87e2837 | 1069 | } |
bd1dbcc6 TG |
1070 | |
1071 | /* | |
e60cbc5c TG |
1072 | * If TID is 0, then either the dying owner has not |
1073 | * yet executed exit_pi_state_list() or some waiter | |
1074 | * acquired the rtmutex in the pi state, but did not | |
1075 | * yet fixup the TID in user space. | |
1076 | * | |
1077 | * Take a ref on the state and return success. [6] | |
1078 | */ | |
1079 | if (!pid) | |
734009e9 | 1080 | goto out_attach; |
e60cbc5c TG |
1081 | } else { |
1082 | /* | |
1083 | * If the owner died bit is not set, then the pi_state | |
1084 | * must have an owner. [7] | |
bd1dbcc6 | 1085 | */ |
e60cbc5c | 1086 | if (!pi_state->owner) |
734009e9 | 1087 | goto out_einval; |
c87e2837 IM |
1088 | } |
1089 | ||
e60cbc5c TG |
1090 | /* |
1091 | * Bail out if user space manipulated the futex value. If pi | |
1092 | * state exists then the owner TID must be the same as the | |
1093 | * user space TID. [9/10] | |
1094 | */ | |
1095 | if (pid != task_pid_vnr(pi_state->owner)) | |
734009e9 PZ |
1096 | goto out_einval; |
1097 | ||
1098 | out_attach: | |
bf92cf3a | 1099 | get_pi_state(pi_state); |
734009e9 | 1100 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
e60cbc5c TG |
1101 | *ps = pi_state; |
1102 | return 0; | |
734009e9 PZ |
1103 | |
1104 | out_einval: | |
1105 | ret = -EINVAL; | |
1106 | goto out_error; | |
1107 | ||
1108 | out_eagain: | |
1109 | ret = -EAGAIN; | |
1110 | goto out_error; | |
1111 | ||
1112 | out_efault: | |
1113 | ret = -EFAULT; | |
1114 | goto out_error; | |
1115 | ||
1116 | out_error: | |
1117 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); | |
1118 | return ret; | |
e60cbc5c TG |
1119 | } |
1120 | ||
3ef240ea TG |
1121 | /** |
1122 | * wait_for_owner_exiting - Block until the owner has exited | |
51bfb1d1 | 1123 | * @ret: owner's current futex lock status |
3ef240ea TG |
1124 | * @exiting: Pointer to the exiting task |
1125 | * | |
1126 | * Caller must hold a refcount on @exiting. | |
1127 | */ | |
1128 | static void wait_for_owner_exiting(int ret, struct task_struct *exiting) | |
1129 | { | |
1130 | if (ret != -EBUSY) { | |
1131 | WARN_ON_ONCE(exiting); | |
1132 | return; | |
1133 | } | |
1134 | ||
1135 | if (WARN_ON_ONCE(ret == -EBUSY && !exiting)) | |
1136 | return; | |
1137 | ||
1138 | mutex_lock(&exiting->futex_exit_mutex); | |
1139 | /* | |
1140 | * No point in doing state checking here. If the waiter got here | |
1141 | * while the task was in exec()->exec_futex_release() then it can | |
1142 | * have any FUTEX_STATE_* value when the waiter has acquired the | |
1143 | * mutex. OK, if running, EXITING or DEAD if it reached exit() | |
1144 | * already. Highly unlikely and not a problem. Just one more round | |
1145 | * through the futex maze. | |
1146 | */ | |
1147 | mutex_unlock(&exiting->futex_exit_mutex); | |
1148 | ||
1149 | put_task_struct(exiting); | |
1150 | } | |
1151 | ||
da791a66 TG |
1152 | static int handle_exit_race(u32 __user *uaddr, u32 uval, |
1153 | struct task_struct *tsk) | |
1154 | { | |
1155 | u32 uval2; | |
1156 | ||
1157 | /* | |
ac31c7ff TG |
1158 | * If the futex exit state is not yet FUTEX_STATE_DEAD, tell the |
1159 | * caller that the alleged owner is busy. | |
da791a66 | 1160 | */ |
3d4775df | 1161 | if (tsk && tsk->futex_state != FUTEX_STATE_DEAD) |
ac31c7ff | 1162 | return -EBUSY; |
da791a66 TG |
1163 | |
1164 | /* | |
1165 | * Reread the user space value to handle the following situation: | |
1166 | * | |
1167 | * CPU0 CPU1 | |
1168 | * | |
1169 | * sys_exit() sys_futex() | |
1170 | * do_exit() futex_lock_pi() | |
1171 | * futex_lock_pi_atomic() | |
1172 | * exit_signals(tsk) No waiters: | |
1173 | * tsk->flags |= PF_EXITING; *uaddr == 0x00000PID | |
1174 | * mm_release(tsk) Set waiter bit | |
1175 | * exit_robust_list(tsk) { *uaddr = 0x80000PID; | |
1176 | * Set owner died attach_to_pi_owner() { | |
1177 | * *uaddr = 0xC0000000; tsk = get_task(PID); | |
1178 | * } if (!tsk->flags & PF_EXITING) { | |
1179 | * ... attach(); | |
3d4775df TG |
1180 | * tsk->futex_state = } else { |
1181 | * FUTEX_STATE_DEAD; if (tsk->futex_state != | |
1182 | * FUTEX_STATE_DEAD) | |
da791a66 TG |
1183 | * return -EAGAIN; |
1184 | * return -ESRCH; <--- FAIL | |
1185 | * } | |
1186 | * | |
1187 | * Returning ESRCH unconditionally is wrong here because the | |
1188 | * user space value has been changed by the exiting task. | |
1189 | * | |
1190 | * The same logic applies to the case where the exiting task is | |
1191 | * already gone. | |
1192 | */ | |
1193 | if (get_futex_value_locked(&uval2, uaddr)) | |
1194 | return -EFAULT; | |
1195 | ||
1196 | /* If the user space value has changed, try again. */ | |
1197 | if (uval2 != uval) | |
1198 | return -EAGAIN; | |
1199 | ||
1200 | /* | |
1201 | * The exiting task did not have a robust list, the robust list was | |
1202 | * corrupted or the user space value in *uaddr is simply bogus. | |
1203 | * Give up and tell user space. | |
1204 | */ | |
1205 | return -ESRCH; | |
1206 | } | |
1207 | ||
04e1b2e5 TG |
1208 | /* |
1209 | * Lookup the task for the TID provided from user space and attach to | |
1210 | * it after doing proper sanity checks. | |
1211 | */ | |
da791a66 | 1212 | static int attach_to_pi_owner(u32 __user *uaddr, u32 uval, union futex_key *key, |
3ef240ea TG |
1213 | struct futex_pi_state **ps, |
1214 | struct task_struct **exiting) | |
e60cbc5c | 1215 | { |
e60cbc5c | 1216 | pid_t pid = uval & FUTEX_TID_MASK; |
04e1b2e5 TG |
1217 | struct futex_pi_state *pi_state; |
1218 | struct task_struct *p; | |
e60cbc5c | 1219 | |
c87e2837 | 1220 | /* |
e3f2ddea | 1221 | * We are the first waiter - try to look up the real owner and attach |
54a21788 | 1222 | * the new pi_state to it, but bail out when TID = 0 [1] |
da791a66 TG |
1223 | * |
1224 | * The !pid check is paranoid. None of the call sites should end up | |
1225 | * with pid == 0, but better safe than sorry. Let the caller retry | |
c87e2837 | 1226 | */ |
778e9a9c | 1227 | if (!pid) |
da791a66 | 1228 | return -EAGAIN; |
2ee08260 | 1229 | p = find_get_task_by_vpid(pid); |
7a0ea09a | 1230 | if (!p) |
da791a66 | 1231 | return handle_exit_race(uaddr, uval, NULL); |
778e9a9c | 1232 | |
a2129464 | 1233 | if (unlikely(p->flags & PF_KTHREAD)) { |
f0d71b3d TG |
1234 | put_task_struct(p); |
1235 | return -EPERM; | |
1236 | } | |
1237 | ||
778e9a9c | 1238 | /* |
3d4775df TG |
1239 | * We need to look at the task state to figure out, whether the |
1240 | * task is exiting. To protect against the change of the task state | |
1241 | * in futex_exit_release(), we do this protected by p->pi_lock: | |
778e9a9c | 1242 | */ |
1d615482 | 1243 | raw_spin_lock_irq(&p->pi_lock); |
3d4775df | 1244 | if (unlikely(p->futex_state != FUTEX_STATE_OK)) { |
778e9a9c | 1245 | /* |
3d4775df TG |
1246 | * The task is on the way out. When the futex state is |
1247 | * FUTEX_STATE_DEAD, we know that the task has finished | |
1248 | * the cleanup: | |
778e9a9c | 1249 | */ |
da791a66 | 1250 | int ret = handle_exit_race(uaddr, uval, p); |
778e9a9c | 1251 | |
1d615482 | 1252 | raw_spin_unlock_irq(&p->pi_lock); |
3ef240ea TG |
1253 | /* |
1254 | * If the owner task is between FUTEX_STATE_EXITING and | |
1255 | * FUTEX_STATE_DEAD then store the task pointer and keep | |
1256 | * the reference on the task struct. The calling code will | |
1257 | * drop all locks, wait for the task to reach | |
1258 | * FUTEX_STATE_DEAD and then drop the refcount. This is | |
1259 | * required to prevent a live lock when the current task | |
1260 | * preempted the exiting task between the two states. | |
1261 | */ | |
1262 | if (ret == -EBUSY) | |
1263 | *exiting = p; | |
1264 | else | |
1265 | put_task_struct(p); | |
778e9a9c AK |
1266 | return ret; |
1267 | } | |
c87e2837 | 1268 | |
54a21788 TG |
1269 | /* |
1270 | * No existing pi state. First waiter. [2] | |
734009e9 PZ |
1271 | * |
1272 | * This creates pi_state, we have hb->lock held, this means nothing can | |
1273 | * observe this state, wait_lock is irrelevant. | |
54a21788 | 1274 | */ |
c87e2837 IM |
1275 | pi_state = alloc_pi_state(); |
1276 | ||
1277 | /* | |
04e1b2e5 | 1278 | * Initialize the pi_mutex in locked state and make @p |
c87e2837 IM |
1279 | * the owner of it: |
1280 | */ | |
1281 | rt_mutex_init_proxy_locked(&pi_state->pi_mutex, p); | |
1282 | ||
1283 | /* Store the key for possible exit cleanups: */ | |
d0aa7a70 | 1284 | pi_state->key = *key; |
c87e2837 | 1285 | |
627371d7 | 1286 | WARN_ON(!list_empty(&pi_state->list)); |
c87e2837 | 1287 | list_add(&pi_state->list, &p->pi_state_list); |
c74aef2d PZ |
1288 | /* |
1289 | * Assignment without holding pi_state->pi_mutex.wait_lock is safe | |
1290 | * because there is no concurrency as the object is not published yet. | |
1291 | */ | |
c87e2837 | 1292 | pi_state->owner = p; |
1d615482 | 1293 | raw_spin_unlock_irq(&p->pi_lock); |
c87e2837 IM |
1294 | |
1295 | put_task_struct(p); | |
1296 | ||
d0aa7a70 | 1297 | *ps = pi_state; |
c87e2837 IM |
1298 | |
1299 | return 0; | |
1300 | } | |
1301 | ||
af54d6a1 TG |
1302 | static int lock_pi_update_atomic(u32 __user *uaddr, u32 uval, u32 newval) |
1303 | { | |
6b4f4bc9 | 1304 | int err; |
3f649ab7 | 1305 | u32 curval; |
af54d6a1 | 1306 | |
ab51fbab DB |
1307 | if (unlikely(should_fail_futex(true))) |
1308 | return -EFAULT; | |
1309 | ||
6b4f4bc9 WD |
1310 | err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
1311 | if (unlikely(err)) | |
1312 | return err; | |
af54d6a1 | 1313 | |
734009e9 | 1314 | /* If user space value changed, let the caller retry */ |
af54d6a1 TG |
1315 | return curval != uval ? -EAGAIN : 0; |
1316 | } | |
1317 | ||
1a52084d | 1318 | /** |
d96ee56c | 1319 | * futex_lock_pi_atomic() - Atomic work required to acquire a pi aware futex |
bab5bc9e DH |
1320 | * @uaddr: the pi futex user address |
1321 | * @hb: the pi futex hash bucket | |
1322 | * @key: the futex key associated with uaddr and hb | |
1323 | * @ps: the pi_state pointer where we store the result of the | |
1324 | * lookup | |
1325 | * @task: the task to perform the atomic lock work for. This will | |
1326 | * be "current" except in the case of requeue pi. | |
3ef240ea TG |
1327 | * @exiting: Pointer to store the task pointer of the owner task |
1328 | * which is in the middle of exiting | |
bab5bc9e | 1329 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) |
1a52084d | 1330 | * |
6c23cbbd | 1331 | * Return: |
7b4ff1ad MCC |
1332 | * - 0 - ready to wait; |
1333 | * - 1 - acquired the lock; | |
1334 | * - <0 - error | |
1a52084d | 1335 | * |
c363b7ed | 1336 | * The hb->lock must be held by the caller. |
3ef240ea TG |
1337 | * |
1338 | * @exiting is only set when the return value is -EBUSY. If so, this holds | |
1339 | * a refcount on the exiting task on return and the caller needs to drop it | |
1340 | * after waiting for the exit to complete. | |
1a52084d DH |
1341 | */ |
1342 | static int futex_lock_pi_atomic(u32 __user *uaddr, struct futex_hash_bucket *hb, | |
1343 | union futex_key *key, | |
1344 | struct futex_pi_state **ps, | |
3ef240ea TG |
1345 | struct task_struct *task, |
1346 | struct task_struct **exiting, | |
1347 | int set_waiters) | |
1a52084d | 1348 | { |
af54d6a1 | 1349 | u32 uval, newval, vpid = task_pid_vnr(task); |
499f5aca | 1350 | struct futex_q *top_waiter; |
af54d6a1 | 1351 | int ret; |
1a52084d DH |
1352 | |
1353 | /* | |
af54d6a1 TG |
1354 | * Read the user space value first so we can validate a few |
1355 | * things before proceeding further. | |
1a52084d | 1356 | */ |
af54d6a1 | 1357 | if (get_futex_value_locked(&uval, uaddr)) |
1a52084d DH |
1358 | return -EFAULT; |
1359 | ||
ab51fbab DB |
1360 | if (unlikely(should_fail_futex(true))) |
1361 | return -EFAULT; | |
1362 | ||
1a52084d DH |
1363 | /* |
1364 | * Detect deadlocks. | |
1365 | */ | |
af54d6a1 | 1366 | if ((unlikely((uval & FUTEX_TID_MASK) == vpid))) |
1a52084d DH |
1367 | return -EDEADLK; |
1368 | ||
ab51fbab DB |
1369 | if ((unlikely(should_fail_futex(true)))) |
1370 | return -EDEADLK; | |
1371 | ||
1a52084d | 1372 | /* |
af54d6a1 TG |
1373 | * Lookup existing state first. If it exists, try to attach to |
1374 | * its pi_state. | |
1a52084d | 1375 | */ |
499f5aca PZ |
1376 | top_waiter = futex_top_waiter(hb, key); |
1377 | if (top_waiter) | |
734009e9 | 1378 | return attach_to_pi_state(uaddr, uval, top_waiter->pi_state, ps); |
1a52084d DH |
1379 | |
1380 | /* | |
af54d6a1 TG |
1381 | * No waiter and user TID is 0. We are here because the |
1382 | * waiters or the owner died bit is set or called from | |
1383 | * requeue_cmp_pi or for whatever reason something took the | |
1384 | * syscall. | |
1a52084d | 1385 | */ |
af54d6a1 | 1386 | if (!(uval & FUTEX_TID_MASK)) { |
59fa6245 | 1387 | /* |
af54d6a1 TG |
1388 | * We take over the futex. No other waiters and the user space |
1389 | * TID is 0. We preserve the owner died bit. | |
59fa6245 | 1390 | */ |
af54d6a1 TG |
1391 | newval = uval & FUTEX_OWNER_DIED; |
1392 | newval |= vpid; | |
1a52084d | 1393 | |
af54d6a1 TG |
1394 | /* The futex requeue_pi code can enforce the waiters bit */ |
1395 | if (set_waiters) | |
1396 | newval |= FUTEX_WAITERS; | |
1397 | ||
1398 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1399 | /* If the take over worked, return 1 */ | |
1400 | return ret < 0 ? ret : 1; | |
1401 | } | |
1a52084d DH |
1402 | |
1403 | /* | |
af54d6a1 TG |
1404 | * First waiter. Set the waiters bit before attaching ourself to |
1405 | * the owner. If owner tries to unlock, it will be forced into | |
1406 | * the kernel and blocked on hb->lock. | |
1a52084d | 1407 | */ |
af54d6a1 TG |
1408 | newval = uval | FUTEX_WAITERS; |
1409 | ret = lock_pi_update_atomic(uaddr, uval, newval); | |
1410 | if (ret) | |
1411 | return ret; | |
1a52084d | 1412 | /* |
af54d6a1 TG |
1413 | * If the update of the user space value succeeded, we try to |
1414 | * attach to the owner. If that fails, no harm done, we only | |
1415 | * set the FUTEX_WAITERS bit in the user space variable. | |
1a52084d | 1416 | */ |
3ef240ea | 1417 | return attach_to_pi_owner(uaddr, newval, key, ps, exiting); |
1a52084d DH |
1418 | } |
1419 | ||
2e12978a LJ |
1420 | /** |
1421 | * __unqueue_futex() - Remove the futex_q from its futex_hash_bucket | |
1422 | * @q: The futex_q to unqueue | |
1423 | * | |
1424 | * The q->lock_ptr must not be NULL and must be held by the caller. | |
1425 | */ | |
1426 | static void __unqueue_futex(struct futex_q *q) | |
1427 | { | |
1428 | struct futex_hash_bucket *hb; | |
1429 | ||
4de1a293 | 1430 | if (WARN_ON_SMP(!q->lock_ptr) || WARN_ON(plist_node_empty(&q->list))) |
2e12978a | 1431 | return; |
4de1a293 | 1432 | lockdep_assert_held(q->lock_ptr); |
2e12978a LJ |
1433 | |
1434 | hb = container_of(q->lock_ptr, struct futex_hash_bucket, lock); | |
1435 | plist_del(&q->list, &hb->chain); | |
11d4616b | 1436 | hb_waiters_dec(hb); |
2e12978a LJ |
1437 | } |
1438 | ||
1da177e4 LT |
1439 | /* |
1440 | * The hash bucket lock must be held when this is called. | |
1d0dcb3a DB |
1441 | * Afterwards, the futex_q must not be accessed. Callers |
1442 | * must ensure to later call wake_up_q() for the actual | |
1443 | * wakeups to occur. | |
1da177e4 | 1444 | */ |
1d0dcb3a | 1445 | static void mark_wake_futex(struct wake_q_head *wake_q, struct futex_q *q) |
1da177e4 | 1446 | { |
f1a11e05 TG |
1447 | struct task_struct *p = q->task; |
1448 | ||
aa10990e DH |
1449 | if (WARN(q->pi_state || q->rt_waiter, "refusing to wake PI futex\n")) |
1450 | return; | |
1451 | ||
b061c38b | 1452 | get_task_struct(p); |
2e12978a | 1453 | __unqueue_futex(q); |
1da177e4 | 1454 | /* |
38fcd06e DHV |
1455 | * The waiting task can free the futex_q as soon as q->lock_ptr = NULL |
1456 | * is written, without taking any locks. This is possible in the event | |
1457 | * of a spurious wakeup, for example. A memory barrier is required here | |
1458 | * to prevent the following store to lock_ptr from getting ahead of the | |
1459 | * plist_del in __unqueue_futex(). | |
1da177e4 | 1460 | */ |
1b367ece | 1461 | smp_store_release(&q->lock_ptr, NULL); |
b061c38b PZ |
1462 | |
1463 | /* | |
1464 | * Queue the task for later wakeup for after we've released | |
75145904 | 1465 | * the hb->lock. |
b061c38b | 1466 | */ |
07879c6a | 1467 | wake_q_add_safe(wake_q, p); |
1da177e4 LT |
1468 | } |
1469 | ||
16ffa12d PZ |
1470 | /* |
1471 | * Caller must hold a reference on @pi_state. | |
1472 | */ | |
1473 | static int wake_futex_pi(u32 __user *uaddr, u32 uval, struct futex_pi_state *pi_state) | |
c87e2837 | 1474 | { |
9a4b99fc | 1475 | struct rt_mutex_waiter *top_waiter; |
16ffa12d | 1476 | struct task_struct *new_owner; |
aa2bfe55 | 1477 | bool postunlock = false; |
7980aa39 TG |
1478 | DEFINE_RT_WAKE_Q(wqh); |
1479 | u32 curval, newval; | |
13fbca4c | 1480 | int ret = 0; |
c87e2837 | 1481 | |
9a4b99fc DB |
1482 | top_waiter = rt_mutex_top_waiter(&pi_state->pi_mutex); |
1483 | if (WARN_ON_ONCE(!top_waiter)) { | |
16ffa12d | 1484 | /* |
bebe5b51 | 1485 | * As per the comment in futex_unlock_pi() this should not happen. |
16ffa12d PZ |
1486 | * |
1487 | * When this happens, give up our locks and try again, giving | |
1488 | * the futex_lock_pi() instance time to complete, either by | |
1489 | * waiting on the rtmutex or removing itself from the futex | |
1490 | * queue. | |
1491 | */ | |
1492 | ret = -EAGAIN; | |
1493 | goto out_unlock; | |
73d786bd | 1494 | } |
c87e2837 | 1495 | |
9a4b99fc DB |
1496 | new_owner = top_waiter->task; |
1497 | ||
c87e2837 | 1498 | /* |
16ffa12d PZ |
1499 | * We pass it to the next owner. The WAITERS bit is always kept |
1500 | * enabled while there is PI state around. We cleanup the owner | |
1501 | * died bit, because we are the owner. | |
c87e2837 | 1502 | */ |
13fbca4c | 1503 | newval = FUTEX_WAITERS | task_pid_vnr(new_owner); |
e3f2ddea | 1504 | |
921c7ebd | 1505 | if (unlikely(should_fail_futex(true))) { |
ab51fbab | 1506 | ret = -EFAULT; |
921c7ebd MN |
1507 | goto out_unlock; |
1508 | } | |
ab51fbab | 1509 | |
6b4f4bc9 WD |
1510 | ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
1511 | if (!ret && (curval != uval)) { | |
89e9e66b SAS |
1512 | /* |
1513 | * If a unconditional UNLOCK_PI operation (user space did not | |
1514 | * try the TID->0 transition) raced with a waiter setting the | |
1515 | * FUTEX_WAITERS flag between get_user() and locking the hash | |
1516 | * bucket lock, retry the operation. | |
1517 | */ | |
1518 | if ((FUTEX_TID_MASK & curval) == uval) | |
1519 | ret = -EAGAIN; | |
1520 | else | |
1521 | ret = -EINVAL; | |
1522 | } | |
734009e9 | 1523 | |
c5cade20 TG |
1524 | if (!ret) { |
1525 | /* | |
1526 | * This is a point of no return; once we modified the uval | |
1527 | * there is no going back and subsequent operations must | |
1528 | * not fail. | |
1529 | */ | |
1530 | pi_state_update_owner(pi_state, new_owner); | |
7980aa39 | 1531 | postunlock = __rt_mutex_futex_unlock(&pi_state->pi_mutex, &wqh); |
c5cade20 | 1532 | } |
5293c2ef | 1533 | |
16ffa12d | 1534 | out_unlock: |
5293c2ef | 1535 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
5293c2ef | 1536 | |
aa2bfe55 | 1537 | if (postunlock) |
7980aa39 | 1538 | rt_mutex_postunlock(&wqh); |
c87e2837 | 1539 | |
16ffa12d | 1540 | return ret; |
c87e2837 IM |
1541 | } |
1542 | ||
8b8f319f IM |
1543 | /* |
1544 | * Express the locking dependencies for lockdep: | |
1545 | */ | |
1546 | static inline void | |
1547 | double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1548 | { | |
1549 | if (hb1 <= hb2) { | |
1550 | spin_lock(&hb1->lock); | |
1551 | if (hb1 < hb2) | |
1552 | spin_lock_nested(&hb2->lock, SINGLE_DEPTH_NESTING); | |
1553 | } else { /* hb1 > hb2 */ | |
1554 | spin_lock(&hb2->lock); | |
1555 | spin_lock_nested(&hb1->lock, SINGLE_DEPTH_NESTING); | |
1556 | } | |
1557 | } | |
1558 | ||
5eb3dc62 DH |
1559 | static inline void |
1560 | double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2) | |
1561 | { | |
f061d351 | 1562 | spin_unlock(&hb1->lock); |
88f502fe IM |
1563 | if (hb1 != hb2) |
1564 | spin_unlock(&hb2->lock); | |
5eb3dc62 DH |
1565 | } |
1566 | ||
1da177e4 | 1567 | /* |
b2d0994b | 1568 | * Wake up waiters matching bitset queued on this futex (uaddr). |
1da177e4 | 1569 | */ |
b41277dc DH |
1570 | static int |
1571 | futex_wake(u32 __user *uaddr, unsigned int flags, int nr_wake, u32 bitset) | |
1da177e4 | 1572 | { |
e2970f2f | 1573 | struct futex_hash_bucket *hb; |
1da177e4 | 1574 | struct futex_q *this, *next; |
38d47c1b | 1575 | union futex_key key = FUTEX_KEY_INIT; |
1da177e4 | 1576 | int ret; |
194a6b5b | 1577 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1578 | |
cd689985 TG |
1579 | if (!bitset) |
1580 | return -EINVAL; | |
1581 | ||
96d4f267 | 1582 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, FUTEX_READ); |
1da177e4 | 1583 | if (unlikely(ret != 0)) |
d7c5ed73 | 1584 | return ret; |
1da177e4 | 1585 | |
e2970f2f | 1586 | hb = hash_futex(&key); |
b0c29f79 DB |
1587 | |
1588 | /* Make sure we really have tasks to wakeup */ | |
1589 | if (!hb_waiters_pending(hb)) | |
d7c5ed73 | 1590 | return ret; |
b0c29f79 | 1591 | |
e2970f2f | 1592 | spin_lock(&hb->lock); |
1da177e4 | 1593 | |
0d00c7b2 | 1594 | plist_for_each_entry_safe(this, next, &hb->chain, list) { |
1da177e4 | 1595 | if (match_futex (&this->key, &key)) { |
52400ba9 | 1596 | if (this->pi_state || this->rt_waiter) { |
ed6f7b10 IM |
1597 | ret = -EINVAL; |
1598 | break; | |
1599 | } | |
cd689985 TG |
1600 | |
1601 | /* Check if one of the bits is set in both bitsets */ | |
1602 | if (!(this->bitset & bitset)) | |
1603 | continue; | |
1604 | ||
1d0dcb3a | 1605 | mark_wake_futex(&wake_q, this); |
1da177e4 LT |
1606 | if (++ret >= nr_wake) |
1607 | break; | |
1608 | } | |
1609 | } | |
1610 | ||
e2970f2f | 1611 | spin_unlock(&hb->lock); |
1d0dcb3a | 1612 | wake_up_q(&wake_q); |
1da177e4 LT |
1613 | return ret; |
1614 | } | |
1615 | ||
30d6e0a4 JS |
1616 | static int futex_atomic_op_inuser(unsigned int encoded_op, u32 __user *uaddr) |
1617 | { | |
1618 | unsigned int op = (encoded_op & 0x70000000) >> 28; | |
1619 | unsigned int cmp = (encoded_op & 0x0f000000) >> 24; | |
d70ef228 JS |
1620 | int oparg = sign_extend32((encoded_op & 0x00fff000) >> 12, 11); |
1621 | int cmparg = sign_extend32(encoded_op & 0x00000fff, 11); | |
30d6e0a4 JS |
1622 | int oldval, ret; |
1623 | ||
1624 | if (encoded_op & (FUTEX_OP_OPARG_SHIFT << 28)) { | |
e78c38f6 JS |
1625 | if (oparg < 0 || oparg > 31) { |
1626 | char comm[sizeof(current->comm)]; | |
1627 | /* | |
1628 | * kill this print and return -EINVAL when userspace | |
1629 | * is sane again | |
1630 | */ | |
1631 | pr_info_ratelimited("futex_wake_op: %s tries to shift op by %d; fix this program\n", | |
1632 | get_task_comm(comm, current), oparg); | |
1633 | oparg &= 31; | |
1634 | } | |
30d6e0a4 JS |
1635 | oparg = 1 << oparg; |
1636 | } | |
1637 | ||
a08971e9 | 1638 | pagefault_disable(); |
30d6e0a4 | 1639 | ret = arch_futex_atomic_op_inuser(op, oparg, &oldval, uaddr); |
a08971e9 | 1640 | pagefault_enable(); |
30d6e0a4 JS |
1641 | if (ret) |
1642 | return ret; | |
1643 | ||
1644 | switch (cmp) { | |
1645 | case FUTEX_OP_CMP_EQ: | |
1646 | return oldval == cmparg; | |
1647 | case FUTEX_OP_CMP_NE: | |
1648 | return oldval != cmparg; | |
1649 | case FUTEX_OP_CMP_LT: | |
1650 | return oldval < cmparg; | |
1651 | case FUTEX_OP_CMP_GE: | |
1652 | return oldval >= cmparg; | |
1653 | case FUTEX_OP_CMP_LE: | |
1654 | return oldval <= cmparg; | |
1655 | case FUTEX_OP_CMP_GT: | |
1656 | return oldval > cmparg; | |
1657 | default: | |
1658 | return -ENOSYS; | |
1659 | } | |
1660 | } | |
1661 | ||
4732efbe JJ |
1662 | /* |
1663 | * Wake up all waiters hashed on the physical page that is mapped | |
1664 | * to this virtual address: | |
1665 | */ | |
e2970f2f | 1666 | static int |
b41277dc | 1667 | futex_wake_op(u32 __user *uaddr1, unsigned int flags, u32 __user *uaddr2, |
e2970f2f | 1668 | int nr_wake, int nr_wake2, int op) |
4732efbe | 1669 | { |
38d47c1b | 1670 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
e2970f2f | 1671 | struct futex_hash_bucket *hb1, *hb2; |
4732efbe | 1672 | struct futex_q *this, *next; |
e4dc5b7a | 1673 | int ret, op_ret; |
194a6b5b | 1674 | DEFINE_WAKE_Q(wake_q); |
4732efbe | 1675 | |
e4dc5b7a | 1676 | retry: |
96d4f267 | 1677 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, FUTEX_READ); |
4732efbe | 1678 | if (unlikely(ret != 0)) |
d7c5ed73 | 1679 | return ret; |
96d4f267 | 1680 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, FUTEX_WRITE); |
4732efbe | 1681 | if (unlikely(ret != 0)) |
d7c5ed73 | 1682 | return ret; |
4732efbe | 1683 | |
e2970f2f IM |
1684 | hb1 = hash_futex(&key1); |
1685 | hb2 = hash_futex(&key2); | |
4732efbe | 1686 | |
e4dc5b7a | 1687 | retry_private: |
eaaea803 | 1688 | double_lock_hb(hb1, hb2); |
e2970f2f | 1689 | op_ret = futex_atomic_op_inuser(op, uaddr2); |
4732efbe | 1690 | if (unlikely(op_ret < 0)) { |
5eb3dc62 | 1691 | double_unlock_hb(hb1, hb2); |
4732efbe | 1692 | |
6b4f4bc9 WD |
1693 | if (!IS_ENABLED(CONFIG_MMU) || |
1694 | unlikely(op_ret != -EFAULT && op_ret != -EAGAIN)) { | |
1695 | /* | |
1696 | * we don't get EFAULT from MMU faults if we don't have | |
1697 | * an MMU, but we might get them from range checking | |
1698 | */ | |
796f8d9b | 1699 | ret = op_ret; |
d7c5ed73 | 1700 | return ret; |
796f8d9b DG |
1701 | } |
1702 | ||
6b4f4bc9 WD |
1703 | if (op_ret == -EFAULT) { |
1704 | ret = fault_in_user_writeable(uaddr2); | |
1705 | if (ret) | |
d7c5ed73 | 1706 | return ret; |
6b4f4bc9 | 1707 | } |
4732efbe | 1708 | |
6b4f4bc9 | 1709 | cond_resched(); |
a82adc76 PB |
1710 | if (!(flags & FLAGS_SHARED)) |
1711 | goto retry_private; | |
e4dc5b7a | 1712 | goto retry; |
4732efbe JJ |
1713 | } |
1714 | ||
0d00c7b2 | 1715 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
4732efbe | 1716 | if (match_futex (&this->key, &key1)) { |
aa10990e DH |
1717 | if (this->pi_state || this->rt_waiter) { |
1718 | ret = -EINVAL; | |
1719 | goto out_unlock; | |
1720 | } | |
1d0dcb3a | 1721 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1722 | if (++ret >= nr_wake) |
1723 | break; | |
1724 | } | |
1725 | } | |
1726 | ||
1727 | if (op_ret > 0) { | |
4732efbe | 1728 | op_ret = 0; |
0d00c7b2 | 1729 | plist_for_each_entry_safe(this, next, &hb2->chain, list) { |
4732efbe | 1730 | if (match_futex (&this->key, &key2)) { |
aa10990e DH |
1731 | if (this->pi_state || this->rt_waiter) { |
1732 | ret = -EINVAL; | |
1733 | goto out_unlock; | |
1734 | } | |
1d0dcb3a | 1735 | mark_wake_futex(&wake_q, this); |
4732efbe JJ |
1736 | if (++op_ret >= nr_wake2) |
1737 | break; | |
1738 | } | |
1739 | } | |
1740 | ret += op_ret; | |
1741 | } | |
1742 | ||
aa10990e | 1743 | out_unlock: |
5eb3dc62 | 1744 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 1745 | wake_up_q(&wake_q); |
4732efbe JJ |
1746 | return ret; |
1747 | } | |
1748 | ||
9121e478 DH |
1749 | /** |
1750 | * requeue_futex() - Requeue a futex_q from one hb to another | |
1751 | * @q: the futex_q to requeue | |
1752 | * @hb1: the source hash_bucket | |
1753 | * @hb2: the target hash_bucket | |
1754 | * @key2: the new key for the requeued futex_q | |
1755 | */ | |
1756 | static inline | |
1757 | void requeue_futex(struct futex_q *q, struct futex_hash_bucket *hb1, | |
1758 | struct futex_hash_bucket *hb2, union futex_key *key2) | |
1759 | { | |
1760 | ||
1761 | /* | |
1762 | * If key1 and key2 hash to the same bucket, no need to | |
1763 | * requeue. | |
1764 | */ | |
1765 | if (likely(&hb1->chain != &hb2->chain)) { | |
1766 | plist_del(&q->list, &hb1->chain); | |
11d4616b | 1767 | hb_waiters_dec(hb1); |
11d4616b | 1768 | hb_waiters_inc(hb2); |
fe1bce9e | 1769 | plist_add(&q->list, &hb2->chain); |
9121e478 | 1770 | q->lock_ptr = &hb2->lock; |
9121e478 | 1771 | } |
9121e478 DH |
1772 | q->key = *key2; |
1773 | } | |
1774 | ||
52400ba9 DH |
1775 | /** |
1776 | * requeue_pi_wake_futex() - Wake a task that acquired the lock during requeue | |
d96ee56c DH |
1777 | * @q: the futex_q |
1778 | * @key: the key of the requeue target futex | |
1779 | * @hb: the hash_bucket of the requeue target futex | |
52400ba9 DH |
1780 | * |
1781 | * During futex_requeue, with requeue_pi=1, it is possible to acquire the | |
1782 | * target futex if it is uncontended or via a lock steal. Set the futex_q key | |
1783 | * to the requeue target futex so the waiter can detect the wakeup on the right | |
1784 | * futex, but remove it from the hb and NULL the rt_waiter so it can detect | |
beda2c7e DH |
1785 | * atomic lock acquisition. Set the q->lock_ptr to the requeue target hb->lock |
1786 | * to protect access to the pi_state to fixup the owner later. Must be called | |
1787 | * with both q->lock_ptr and hb->lock held. | |
52400ba9 DH |
1788 | */ |
1789 | static inline | |
beda2c7e DH |
1790 | void requeue_pi_wake_futex(struct futex_q *q, union futex_key *key, |
1791 | struct futex_hash_bucket *hb) | |
52400ba9 | 1792 | { |
52400ba9 DH |
1793 | q->key = *key; |
1794 | ||
2e12978a | 1795 | __unqueue_futex(q); |
52400ba9 DH |
1796 | |
1797 | WARN_ON(!q->rt_waiter); | |
1798 | q->rt_waiter = NULL; | |
1799 | ||
beda2c7e | 1800 | q->lock_ptr = &hb->lock; |
beda2c7e | 1801 | |
f1a11e05 | 1802 | wake_up_state(q->task, TASK_NORMAL); |
52400ba9 DH |
1803 | } |
1804 | ||
1805 | /** | |
1806 | * futex_proxy_trylock_atomic() - Attempt an atomic lock for the top waiter | |
bab5bc9e DH |
1807 | * @pifutex: the user address of the to futex |
1808 | * @hb1: the from futex hash bucket, must be locked by the caller | |
1809 | * @hb2: the to futex hash bucket, must be locked by the caller | |
1810 | * @key1: the from futex key | |
1811 | * @key2: the to futex key | |
1812 | * @ps: address to store the pi_state pointer | |
3ef240ea TG |
1813 | * @exiting: Pointer to store the task pointer of the owner task |
1814 | * which is in the middle of exiting | |
bab5bc9e | 1815 | * @set_waiters: force setting the FUTEX_WAITERS bit (1) or not (0) |
52400ba9 DH |
1816 | * |
1817 | * Try and get the lock on behalf of the top waiter if we can do it atomically. | |
bab5bc9e DH |
1818 | * Wake the top waiter if we succeed. If the caller specified set_waiters, |
1819 | * then direct futex_lock_pi_atomic() to force setting the FUTEX_WAITERS bit. | |
1820 | * hb1 and hb2 must be held by the caller. | |
52400ba9 | 1821 | * |
3ef240ea TG |
1822 | * @exiting is only set when the return value is -EBUSY. If so, this holds |
1823 | * a refcount on the exiting task on return and the caller needs to drop it | |
1824 | * after waiting for the exit to complete. | |
1825 | * | |
6c23cbbd | 1826 | * Return: |
7b4ff1ad MCC |
1827 | * - 0 - failed to acquire the lock atomically; |
1828 | * - >0 - acquired the lock, return value is vpid of the top_waiter | |
1829 | * - <0 - error | |
52400ba9 | 1830 | */ |
3ef240ea TG |
1831 | static int |
1832 | futex_proxy_trylock_atomic(u32 __user *pifutex, struct futex_hash_bucket *hb1, | |
1833 | struct futex_hash_bucket *hb2, union futex_key *key1, | |
1834 | union futex_key *key2, struct futex_pi_state **ps, | |
1835 | struct task_struct **exiting, int set_waiters) | |
52400ba9 | 1836 | { |
bab5bc9e | 1837 | struct futex_q *top_waiter = NULL; |
52400ba9 | 1838 | u32 curval; |
866293ee | 1839 | int ret, vpid; |
52400ba9 DH |
1840 | |
1841 | if (get_futex_value_locked(&curval, pifutex)) | |
1842 | return -EFAULT; | |
1843 | ||
ab51fbab DB |
1844 | if (unlikely(should_fail_futex(true))) |
1845 | return -EFAULT; | |
1846 | ||
bab5bc9e DH |
1847 | /* |
1848 | * Find the top_waiter and determine if there are additional waiters. | |
1849 | * If the caller intends to requeue more than 1 waiter to pifutex, | |
1850 | * force futex_lock_pi_atomic() to set the FUTEX_WAITERS bit now, | |
1851 | * as we have means to handle the possible fault. If not, don't set | |
93d0955e | 1852 | * the bit unnecessarily as it will force the subsequent unlock to enter |
bab5bc9e DH |
1853 | * the kernel. |
1854 | */ | |
52400ba9 DH |
1855 | top_waiter = futex_top_waiter(hb1, key1); |
1856 | ||
1857 | /* There are no waiters, nothing for us to do. */ | |
1858 | if (!top_waiter) | |
1859 | return 0; | |
1860 | ||
dc7109aa TG |
1861 | /* |
1862 | * Ensure that this is a waiter sitting in futex_wait_requeue_pi() | |
1863 | * and waiting on the 'waitqueue' futex which is always !PI. | |
1864 | */ | |
1865 | if (!top_waiter->rt_waiter || top_waiter->pi_state) | |
1866 | ret = -EINVAL; | |
1867 | ||
84bc4af5 DH |
1868 | /* Ensure we requeue to the expected futex. */ |
1869 | if (!match_futex(top_waiter->requeue_pi_key, key2)) | |
1870 | return -EINVAL; | |
1871 | ||
52400ba9 | 1872 | /* |
bab5bc9e DH |
1873 | * Try to take the lock for top_waiter. Set the FUTEX_WAITERS bit in |
1874 | * the contended case or if set_waiters is 1. The pi_state is returned | |
1875 | * in ps in contended cases. | |
52400ba9 | 1876 | */ |
866293ee | 1877 | vpid = task_pid_vnr(top_waiter->task); |
bab5bc9e | 1878 | ret = futex_lock_pi_atomic(pifutex, hb2, key2, ps, top_waiter->task, |
3ef240ea | 1879 | exiting, set_waiters); |
866293ee | 1880 | if (ret == 1) { |
beda2c7e | 1881 | requeue_pi_wake_futex(top_waiter, key2, hb2); |
866293ee TG |
1882 | return vpid; |
1883 | } | |
52400ba9 DH |
1884 | return ret; |
1885 | } | |
1886 | ||
1887 | /** | |
1888 | * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 | |
fb62db2b | 1889 | * @uaddr1: source futex user address |
b41277dc | 1890 | * @flags: futex flags (FLAGS_SHARED, etc.) |
fb62db2b RD |
1891 | * @uaddr2: target futex user address |
1892 | * @nr_wake: number of waiters to wake (must be 1 for requeue_pi) | |
1893 | * @nr_requeue: number of waiters to requeue (0-INT_MAX) | |
1894 | * @cmpval: @uaddr1 expected value (or %NULL) | |
1895 | * @requeue_pi: if we are attempting to requeue from a non-pi futex to a | |
b41277dc | 1896 | * pi futex (pi to pi requeue is not supported) |
52400ba9 DH |
1897 | * |
1898 | * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire | |
1899 | * uaddr2 atomically on behalf of the top waiter. | |
1900 | * | |
6c23cbbd | 1901 | * Return: |
7b4ff1ad MCC |
1902 | * - >=0 - on success, the number of tasks requeued or woken; |
1903 | * - <0 - on error | |
1da177e4 | 1904 | */ |
b41277dc DH |
1905 | static int futex_requeue(u32 __user *uaddr1, unsigned int flags, |
1906 | u32 __user *uaddr2, int nr_wake, int nr_requeue, | |
1907 | u32 *cmpval, int requeue_pi) | |
1da177e4 | 1908 | { |
38d47c1b | 1909 | union futex_key key1 = FUTEX_KEY_INIT, key2 = FUTEX_KEY_INIT; |
4b39f99c | 1910 | int task_count = 0, ret; |
52400ba9 | 1911 | struct futex_pi_state *pi_state = NULL; |
e2970f2f | 1912 | struct futex_hash_bucket *hb1, *hb2; |
1da177e4 | 1913 | struct futex_q *this, *next; |
194a6b5b | 1914 | DEFINE_WAKE_Q(wake_q); |
52400ba9 | 1915 | |
fbe0e839 LJ |
1916 | if (nr_wake < 0 || nr_requeue < 0) |
1917 | return -EINVAL; | |
1918 | ||
bc2eecd7 NP |
1919 | /* |
1920 | * When PI not supported: return -ENOSYS if requeue_pi is true, | |
1921 | * consequently the compiler knows requeue_pi is always false past | |
1922 | * this point which will optimize away all the conditional code | |
1923 | * further down. | |
1924 | */ | |
1925 | if (!IS_ENABLED(CONFIG_FUTEX_PI) && requeue_pi) | |
1926 | return -ENOSYS; | |
1927 | ||
52400ba9 | 1928 | if (requeue_pi) { |
e9c243a5 TG |
1929 | /* |
1930 | * Requeue PI only works on two distinct uaddrs. This | |
1931 | * check is only valid for private futexes. See below. | |
1932 | */ | |
1933 | if (uaddr1 == uaddr2) | |
1934 | return -EINVAL; | |
1935 | ||
52400ba9 DH |
1936 | /* |
1937 | * requeue_pi requires a pi_state, try to allocate it now | |
1938 | * without any locks in case it fails. | |
1939 | */ | |
1940 | if (refill_pi_state_cache()) | |
1941 | return -ENOMEM; | |
1942 | /* | |
1943 | * requeue_pi must wake as many tasks as it can, up to nr_wake | |
1944 | * + nr_requeue, since it acquires the rt_mutex prior to | |
1945 | * returning to userspace, so as to not leave the rt_mutex with | |
1946 | * waiters and no owner. However, second and third wake-ups | |
1947 | * cannot be predicted as they involve race conditions with the | |
1948 | * first wake and a fault while looking up the pi_state. Both | |
1949 | * pthread_cond_signal() and pthread_cond_broadcast() should | |
1950 | * use nr_wake=1. | |
1951 | */ | |
1952 | if (nr_wake != 1) | |
1953 | return -EINVAL; | |
1954 | } | |
1da177e4 | 1955 | |
42d35d48 | 1956 | retry: |
96d4f267 | 1957 | ret = get_futex_key(uaddr1, flags & FLAGS_SHARED, &key1, FUTEX_READ); |
1da177e4 | 1958 | if (unlikely(ret != 0)) |
d7c5ed73 | 1959 | return ret; |
9ea71503 | 1960 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, |
96d4f267 | 1961 | requeue_pi ? FUTEX_WRITE : FUTEX_READ); |
1da177e4 | 1962 | if (unlikely(ret != 0)) |
d7c5ed73 | 1963 | return ret; |
1da177e4 | 1964 | |
e9c243a5 TG |
1965 | /* |
1966 | * The check above which compares uaddrs is not sufficient for | |
1967 | * shared futexes. We need to compare the keys: | |
1968 | */ | |
d7c5ed73 AA |
1969 | if (requeue_pi && match_futex(&key1, &key2)) |
1970 | return -EINVAL; | |
e9c243a5 | 1971 | |
e2970f2f IM |
1972 | hb1 = hash_futex(&key1); |
1973 | hb2 = hash_futex(&key2); | |
1da177e4 | 1974 | |
e4dc5b7a | 1975 | retry_private: |
69cd9eba | 1976 | hb_waiters_inc(hb2); |
8b8f319f | 1977 | double_lock_hb(hb1, hb2); |
1da177e4 | 1978 | |
e2970f2f IM |
1979 | if (likely(cmpval != NULL)) { |
1980 | u32 curval; | |
1da177e4 | 1981 | |
e2970f2f | 1982 | ret = get_futex_value_locked(&curval, uaddr1); |
1da177e4 LT |
1983 | |
1984 | if (unlikely(ret)) { | |
5eb3dc62 | 1985 | double_unlock_hb(hb1, hb2); |
69cd9eba | 1986 | hb_waiters_dec(hb2); |
1da177e4 | 1987 | |
e2970f2f | 1988 | ret = get_user(curval, uaddr1); |
e4dc5b7a | 1989 | if (ret) |
d7c5ed73 | 1990 | return ret; |
1da177e4 | 1991 | |
b41277dc | 1992 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a | 1993 | goto retry_private; |
1da177e4 | 1994 | |
e4dc5b7a | 1995 | goto retry; |
1da177e4 | 1996 | } |
e2970f2f | 1997 | if (curval != *cmpval) { |
1da177e4 LT |
1998 | ret = -EAGAIN; |
1999 | goto out_unlock; | |
2000 | } | |
2001 | } | |
2002 | ||
8e74633d | 2003 | if (requeue_pi) { |
3ef240ea TG |
2004 | struct task_struct *exiting = NULL; |
2005 | ||
bab5bc9e DH |
2006 | /* |
2007 | * Attempt to acquire uaddr2 and wake the top waiter. If we | |
2008 | * intend to requeue waiters, force setting the FUTEX_WAITERS | |
2009 | * bit. We force this here where we are able to easily handle | |
2010 | * faults rather in the requeue loop below. | |
2011 | */ | |
52400ba9 | 2012 | ret = futex_proxy_trylock_atomic(uaddr2, hb1, hb2, &key1, |
3ef240ea TG |
2013 | &key2, &pi_state, |
2014 | &exiting, nr_requeue); | |
52400ba9 DH |
2015 | |
2016 | /* | |
2017 | * At this point the top_waiter has either taken uaddr2 or is | |
2018 | * waiting on it. If the former, then the pi_state will not | |
2019 | * exist yet, look it up one more time to ensure we have a | |
f6f4ec00 TG |
2020 | * reference to it. If the lock was taken, @ret contains the |
2021 | * VPID of the top waiter task. | |
ecb38b78 TG |
2022 | * If the lock was not taken, we have pi_state and an initial |
2023 | * refcount on it. In case of an error we have nothing. | |
52400ba9 | 2024 | */ |
866293ee | 2025 | if (ret > 0) { |
52400ba9 DH |
2026 | WARN_ON(pi_state); |
2027 | task_count++; | |
866293ee | 2028 | /* |
f6f4ec00 TG |
2029 | * If futex_proxy_trylock_atomic() acquired the |
2030 | * user space futex, then the user space value | |
2031 | * @uaddr2 has been set to the @hb1's top waiter | |
2032 | * task VPID. This task is guaranteed to be alive | |
2033 | * and cannot be exiting because it is either | |
2034 | * sleeping or blocked on @hb2 lock. | |
2035 | * | |
2036 | * The @uaddr2 futex cannot have waiters either as | |
2037 | * otherwise futex_proxy_trylock_atomic() would not | |
2038 | * have succeeded. | |
ecb38b78 | 2039 | * |
f6f4ec00 TG |
2040 | * In order to requeue waiters to @hb2, pi state is |
2041 | * required. Hand in the VPID value (@ret) and | |
2042 | * allocate PI state with an initial refcount on | |
2043 | * it. | |
866293ee | 2044 | */ |
f6f4ec00 TG |
2045 | ret = attach_to_pi_owner(uaddr2, ret, &key2, &pi_state, |
2046 | &exiting); | |
2047 | WARN_ON(ret); | |
52400ba9 DH |
2048 | } |
2049 | ||
2050 | switch (ret) { | |
2051 | case 0: | |
ecb38b78 | 2052 | /* We hold a reference on the pi state. */ |
52400ba9 | 2053 | break; |
4959f2de TG |
2054 | |
2055 | /* If the above failed, then pi_state is NULL */ | |
52400ba9 DH |
2056 | case -EFAULT: |
2057 | double_unlock_hb(hb1, hb2); | |
69cd9eba | 2058 | hb_waiters_dec(hb2); |
d0725992 | 2059 | ret = fault_in_user_writeable(uaddr2); |
52400ba9 DH |
2060 | if (!ret) |
2061 | goto retry; | |
d7c5ed73 | 2062 | return ret; |
ac31c7ff | 2063 | case -EBUSY: |
52400ba9 | 2064 | case -EAGAIN: |
af54d6a1 TG |
2065 | /* |
2066 | * Two reasons for this: | |
ac31c7ff | 2067 | * - EBUSY: Owner is exiting and we just wait for the |
af54d6a1 | 2068 | * exit to complete. |
ac31c7ff | 2069 | * - EAGAIN: The user space value changed. |
af54d6a1 | 2070 | */ |
52400ba9 | 2071 | double_unlock_hb(hb1, hb2); |
69cd9eba | 2072 | hb_waiters_dec(hb2); |
3ef240ea TG |
2073 | /* |
2074 | * Handle the case where the owner is in the middle of | |
2075 | * exiting. Wait for the exit to complete otherwise | |
2076 | * this task might loop forever, aka. live lock. | |
2077 | */ | |
2078 | wait_for_owner_exiting(ret, exiting); | |
52400ba9 DH |
2079 | cond_resched(); |
2080 | goto retry; | |
2081 | default: | |
2082 | goto out_unlock; | |
2083 | } | |
2084 | } | |
2085 | ||
0d00c7b2 | 2086 | plist_for_each_entry_safe(this, next, &hb1->chain, list) { |
52400ba9 DH |
2087 | if (task_count - nr_wake >= nr_requeue) |
2088 | break; | |
2089 | ||
2090 | if (!match_futex(&this->key, &key1)) | |
1da177e4 | 2091 | continue; |
52400ba9 | 2092 | |
392741e0 | 2093 | /* |
93d0955e | 2094 | * FUTEX_WAIT_REQUEUE_PI and FUTEX_CMP_REQUEUE_PI should always |
392741e0 | 2095 | * be paired with each other and no other futex ops. |
aa10990e DH |
2096 | * |
2097 | * We should never be requeueing a futex_q with a pi_state, | |
2098 | * which is awaiting a futex_unlock_pi(). | |
392741e0 DH |
2099 | */ |
2100 | if ((requeue_pi && !this->rt_waiter) || | |
aa10990e DH |
2101 | (!requeue_pi && this->rt_waiter) || |
2102 | this->pi_state) { | |
392741e0 DH |
2103 | ret = -EINVAL; |
2104 | break; | |
2105 | } | |
52400ba9 | 2106 | |
64b7b715 TG |
2107 | /* Plain futexes just wake or requeue and are done */ |
2108 | if (!requeue_pi) { | |
2109 | if (++task_count <= nr_wake) | |
2110 | mark_wake_futex(&wake_q, this); | |
2111 | else | |
2112 | requeue_futex(this, hb1, hb2, &key2); | |
52400ba9 DH |
2113 | continue; |
2114 | } | |
1da177e4 | 2115 | |
84bc4af5 | 2116 | /* Ensure we requeue to the expected futex for requeue_pi. */ |
64b7b715 | 2117 | if (!match_futex(this->requeue_pi_key, &key2)) { |
84bc4af5 DH |
2118 | ret = -EINVAL; |
2119 | break; | |
2120 | } | |
2121 | ||
52400ba9 DH |
2122 | /* |
2123 | * Requeue nr_requeue waiters and possibly one more in the case | |
2124 | * of requeue_pi if we couldn't acquire the lock atomically. | |
64b7b715 TG |
2125 | * |
2126 | * Prepare the waiter to take the rt_mutex. Take a refcount | |
2127 | * on the pi_state and store the pointer in the futex_q | |
2128 | * object of the waiter. | |
52400ba9 | 2129 | */ |
64b7b715 TG |
2130 | get_pi_state(pi_state); |
2131 | this->pi_state = pi_state; | |
2132 | ret = rt_mutex_start_proxy_lock(&pi_state->pi_mutex, | |
2133 | this->rt_waiter, this->task); | |
2134 | if (ret == 1) { | |
ecb38b78 | 2135 | /* |
64b7b715 TG |
2136 | * We got the lock. We do neither drop the refcount |
2137 | * on pi_state nor clear this->pi_state because the | |
2138 | * waiter needs the pi_state for cleaning up the | |
2139 | * user space value. It will drop the refcount | |
2140 | * after doing so. | |
ecb38b78 | 2141 | */ |
64b7b715 TG |
2142 | requeue_pi_wake_futex(this, &key2, hb2); |
2143 | task_count++; | |
2144 | continue; | |
2145 | } else if (ret) { | |
2146 | /* | |
2147 | * rt_mutex_start_proxy_lock() detected a potential | |
2148 | * deadlock when we tried to queue that waiter. | |
2149 | * Drop the pi_state reference which we took above | |
2150 | * and remove the pointer to the state from the | |
2151 | * waiters futex_q object. | |
2152 | */ | |
2153 | this->pi_state = NULL; | |
2154 | put_pi_state(pi_state); | |
2155 | /* | |
2156 | * We stop queueing more waiters and let user space | |
2157 | * deal with the mess. | |
2158 | */ | |
2159 | break; | |
1da177e4 | 2160 | } |
64b7b715 | 2161 | /* Waiter is queued, move it to hb2 */ |
52400ba9 | 2162 | requeue_futex(this, hb1, hb2, &key2); |
64b7b715 | 2163 | task_count++; |
1da177e4 LT |
2164 | } |
2165 | ||
ecb38b78 | 2166 | /* |
f6f4ec00 TG |
2167 | * We took an extra initial reference to the pi_state either in |
2168 | * futex_proxy_trylock_atomic() or in attach_to_pi_owner(). We need | |
2169 | * to drop it here again. | |
ecb38b78 | 2170 | */ |
29e9ee5d | 2171 | put_pi_state(pi_state); |
885c2cb7 TG |
2172 | |
2173 | out_unlock: | |
5eb3dc62 | 2174 | double_unlock_hb(hb1, hb2); |
1d0dcb3a | 2175 | wake_up_q(&wake_q); |
69cd9eba | 2176 | hb_waiters_dec(hb2); |
52400ba9 | 2177 | return ret ? ret : task_count; |
1da177e4 LT |
2178 | } |
2179 | ||
2180 | /* The key must be already stored in q->key. */ | |
82af7aca | 2181 | static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) |
15e408cd | 2182 | __acquires(&hb->lock) |
1da177e4 | 2183 | { |
e2970f2f | 2184 | struct futex_hash_bucket *hb; |
1da177e4 | 2185 | |
e2970f2f | 2186 | hb = hash_futex(&q->key); |
11d4616b LT |
2187 | |
2188 | /* | |
2189 | * Increment the counter before taking the lock so that | |
2190 | * a potential waker won't miss a to-be-slept task that is | |
2191 | * waiting for the spinlock. This is safe as all queue_lock() | |
2192 | * users end up calling queue_me(). Similarly, for housekeeping, | |
2193 | * decrement the counter at queue_unlock() when some error has | |
2194 | * occurred and we don't end up adding the task to the list. | |
2195 | */ | |
6f568ebe | 2196 | hb_waiters_inc(hb); /* implies smp_mb(); (A) */ |
11d4616b | 2197 | |
e2970f2f | 2198 | q->lock_ptr = &hb->lock; |
1da177e4 | 2199 | |
6f568ebe | 2200 | spin_lock(&hb->lock); |
e2970f2f | 2201 | return hb; |
1da177e4 LT |
2202 | } |
2203 | ||
d40d65c8 | 2204 | static inline void |
0d00c7b2 | 2205 | queue_unlock(struct futex_hash_bucket *hb) |
15e408cd | 2206 | __releases(&hb->lock) |
d40d65c8 DH |
2207 | { |
2208 | spin_unlock(&hb->lock); | |
11d4616b | 2209 | hb_waiters_dec(hb); |
d40d65c8 DH |
2210 | } |
2211 | ||
cfafcd11 | 2212 | static inline void __queue_me(struct futex_q *q, struct futex_hash_bucket *hb) |
1da177e4 | 2213 | { |
ec92d082 PP |
2214 | int prio; |
2215 | ||
2216 | /* | |
2217 | * The priority used to register this element is | |
2218 | * - either the real thread-priority for the real-time threads | |
2219 | * (i.e. threads with a priority lower than MAX_RT_PRIO) | |
2220 | * - or MAX_RT_PRIO for non-RT threads. | |
2221 | * Thus, all RT-threads are woken first in priority order, and | |
2222 | * the others are woken last, in FIFO order. | |
2223 | */ | |
2224 | prio = min(current->normal_prio, MAX_RT_PRIO); | |
2225 | ||
2226 | plist_node_init(&q->list, prio); | |
ec92d082 | 2227 | plist_add(&q->list, &hb->chain); |
c87e2837 | 2228 | q->task = current; |
cfafcd11 PZ |
2229 | } |
2230 | ||
2231 | /** | |
2232 | * queue_me() - Enqueue the futex_q on the futex_hash_bucket | |
2233 | * @q: The futex_q to enqueue | |
2234 | * @hb: The destination hash bucket | |
2235 | * | |
2236 | * The hb->lock must be held by the caller, and is released here. A call to | |
2237 | * queue_me() is typically paired with exactly one call to unqueue_me(). The | |
2238 | * exceptions involve the PI related operations, which may use unqueue_me_pi() | |
2239 | * or nothing if the unqueue is done as part of the wake process and the unqueue | |
2240 | * state is implicit in the state of woken task (see futex_wait_requeue_pi() for | |
2241 | * an example). | |
2242 | */ | |
2243 | static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) | |
2244 | __releases(&hb->lock) | |
2245 | { | |
2246 | __queue_me(q, hb); | |
e2970f2f | 2247 | spin_unlock(&hb->lock); |
1da177e4 LT |
2248 | } |
2249 | ||
d40d65c8 DH |
2250 | /** |
2251 | * unqueue_me() - Remove the futex_q from its futex_hash_bucket | |
2252 | * @q: The futex_q to unqueue | |
2253 | * | |
2254 | * The q->lock_ptr must not be held by the caller. A call to unqueue_me() must | |
2255 | * be paired with exactly one earlier call to queue_me(). | |
2256 | * | |
6c23cbbd | 2257 | * Return: |
7b4ff1ad MCC |
2258 | * - 1 - if the futex_q was still queued (and we removed unqueued it); |
2259 | * - 0 - if the futex_q was already removed by the waking thread | |
1da177e4 | 2260 | */ |
1da177e4 LT |
2261 | static int unqueue_me(struct futex_q *q) |
2262 | { | |
1da177e4 | 2263 | spinlock_t *lock_ptr; |
e2970f2f | 2264 | int ret = 0; |
1da177e4 LT |
2265 | |
2266 | /* In the common case we don't take the spinlock, which is nice. */ | |
42d35d48 | 2267 | retry: |
29b75eb2 JZ |
2268 | /* |
2269 | * q->lock_ptr can change between this read and the following spin_lock. | |
2270 | * Use READ_ONCE to forbid the compiler from reloading q->lock_ptr and | |
2271 | * optimizing lock_ptr out of the logic below. | |
2272 | */ | |
2273 | lock_ptr = READ_ONCE(q->lock_ptr); | |
c80544dc | 2274 | if (lock_ptr != NULL) { |
1da177e4 LT |
2275 | spin_lock(lock_ptr); |
2276 | /* | |
2277 | * q->lock_ptr can change between reading it and | |
2278 | * spin_lock(), causing us to take the wrong lock. This | |
2279 | * corrects the race condition. | |
2280 | * | |
2281 | * Reasoning goes like this: if we have the wrong lock, | |
2282 | * q->lock_ptr must have changed (maybe several times) | |
2283 | * between reading it and the spin_lock(). It can | |
2284 | * change again after the spin_lock() but only if it was | |
2285 | * already changed before the spin_lock(). It cannot, | |
2286 | * however, change back to the original value. Therefore | |
2287 | * we can detect whether we acquired the correct lock. | |
2288 | */ | |
2289 | if (unlikely(lock_ptr != q->lock_ptr)) { | |
2290 | spin_unlock(lock_ptr); | |
2291 | goto retry; | |
2292 | } | |
2e12978a | 2293 | __unqueue_futex(q); |
c87e2837 IM |
2294 | |
2295 | BUG_ON(q->pi_state); | |
2296 | ||
1da177e4 LT |
2297 | spin_unlock(lock_ptr); |
2298 | ret = 1; | |
2299 | } | |
2300 | ||
1da177e4 LT |
2301 | return ret; |
2302 | } | |
2303 | ||
c87e2837 | 2304 | /* |
93d0955e | 2305 | * PI futexes can not be requeued and must remove themselves from the |
a3f2428d | 2306 | * hash bucket. The hash bucket lock (i.e. lock_ptr) is held. |
c87e2837 | 2307 | */ |
d0aa7a70 | 2308 | static void unqueue_me_pi(struct futex_q *q) |
c87e2837 | 2309 | { |
2e12978a | 2310 | __unqueue_futex(q); |
c87e2837 IM |
2311 | |
2312 | BUG_ON(!q->pi_state); | |
29e9ee5d | 2313 | put_pi_state(q->pi_state); |
c87e2837 | 2314 | q->pi_state = NULL; |
c87e2837 IM |
2315 | } |
2316 | ||
f2dac39d TG |
2317 | static int __fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
2318 | struct task_struct *argowner) | |
d0aa7a70 | 2319 | { |
d0aa7a70 | 2320 | struct futex_pi_state *pi_state = q->pi_state; |
c1e2f0ea | 2321 | struct task_struct *oldowner, *newowner; |
f2dac39d TG |
2322 | u32 uval, curval, newval, newtid; |
2323 | int err = 0; | |
734009e9 PZ |
2324 | |
2325 | oldowner = pi_state->owner; | |
1b7558e4 TG |
2326 | |
2327 | /* | |
c1e2f0ea | 2328 | * We are here because either: |
16ffa12d | 2329 | * |
c1e2f0ea PZ |
2330 | * - we stole the lock and pi_state->owner needs updating to reflect |
2331 | * that (@argowner == current), | |
2332 | * | |
2333 | * or: | |
2334 | * | |
2335 | * - someone stole our lock and we need to fix things to point to the | |
2336 | * new owner (@argowner == NULL). | |
2337 | * | |
2338 | * Either way, we have to replace the TID in the user space variable. | |
8161239a | 2339 | * This must be atomic as we have to preserve the owner died bit here. |
1b7558e4 | 2340 | * |
b2d0994b DH |
2341 | * Note: We write the user space value _before_ changing the pi_state |
2342 | * because we can fault here. Imagine swapped out pages or a fork | |
2343 | * that marked all the anonymous memory readonly for cow. | |
1b7558e4 | 2344 | * |
734009e9 PZ |
2345 | * Modifying pi_state _before_ the user space value would leave the |
2346 | * pi_state in an inconsistent state when we fault here, because we | |
2347 | * need to drop the locks to handle the fault. This might be observed | |
f6f4ec00 | 2348 | * in the PID checks when attaching to PI state . |
1b7558e4 TG |
2349 | */ |
2350 | retry: | |
c1e2f0ea PZ |
2351 | if (!argowner) { |
2352 | if (oldowner != current) { | |
2353 | /* | |
2354 | * We raced against a concurrent self; things are | |
2355 | * already fixed up. Nothing to do. | |
2356 | */ | |
f2dac39d | 2357 | return 0; |
c1e2f0ea PZ |
2358 | } |
2359 | ||
2360 | if (__rt_mutex_futex_trylock(&pi_state->pi_mutex)) { | |
12bb3f7f | 2361 | /* We got the lock. pi_state is correct. Tell caller. */ |
f2dac39d | 2362 | return 1; |
c1e2f0ea PZ |
2363 | } |
2364 | ||
2365 | /* | |
9f5d1c33 MG |
2366 | * The trylock just failed, so either there is an owner or |
2367 | * there is a higher priority waiter than this one. | |
c1e2f0ea PZ |
2368 | */ |
2369 | newowner = rt_mutex_owner(&pi_state->pi_mutex); | |
9f5d1c33 MG |
2370 | /* |
2371 | * If the higher priority waiter has not yet taken over the | |
2372 | * rtmutex then newowner is NULL. We can't return here with | |
2373 | * that state because it's inconsistent vs. the user space | |
2374 | * state. So drop the locks and try again. It's a valid | |
2375 | * situation and not any different from the other retry | |
2376 | * conditions. | |
2377 | */ | |
2378 | if (unlikely(!newowner)) { | |
2379 | err = -EAGAIN; | |
2380 | goto handle_err; | |
2381 | } | |
c1e2f0ea PZ |
2382 | } else { |
2383 | WARN_ON_ONCE(argowner != current); | |
2384 | if (oldowner == current) { | |
2385 | /* | |
2386 | * We raced against a concurrent self; things are | |
2387 | * already fixed up. Nothing to do. | |
2388 | */ | |
f2dac39d | 2389 | return 1; |
c1e2f0ea PZ |
2390 | } |
2391 | newowner = argowner; | |
2392 | } | |
2393 | ||
2394 | newtid = task_pid_vnr(newowner) | FUTEX_WAITERS; | |
a97cb0e7 PZ |
2395 | /* Owner died? */ |
2396 | if (!pi_state->owner) | |
2397 | newtid |= FUTEX_OWNER_DIED; | |
c1e2f0ea | 2398 | |
6b4f4bc9 WD |
2399 | err = get_futex_value_locked(&uval, uaddr); |
2400 | if (err) | |
2401 | goto handle_err; | |
1b7558e4 | 2402 | |
16ffa12d | 2403 | for (;;) { |
1b7558e4 TG |
2404 | newval = (uval & FUTEX_OWNER_DIED) | newtid; |
2405 | ||
6b4f4bc9 WD |
2406 | err = cmpxchg_futex_value_locked(&curval, uaddr, uval, newval); |
2407 | if (err) | |
2408 | goto handle_err; | |
2409 | ||
1b7558e4 TG |
2410 | if (curval == uval) |
2411 | break; | |
2412 | uval = curval; | |
2413 | } | |
2414 | ||
2415 | /* | |
2416 | * We fixed up user space. Now we need to fix the pi_state | |
2417 | * itself. | |
2418 | */ | |
c5cade20 | 2419 | pi_state_update_owner(pi_state, newowner); |
d0aa7a70 | 2420 | |
12bb3f7f | 2421 | return argowner == current; |
d0aa7a70 | 2422 | |
d0aa7a70 | 2423 | /* |
6b4f4bc9 WD |
2424 | * In order to reschedule or handle a page fault, we need to drop the |
2425 | * locks here. In the case of a fault, this gives the other task | |
2426 | * (either the highest priority waiter itself or the task which stole | |
2427 | * the rtmutex) the chance to try the fixup of the pi_state. So once we | |
2428 | * are back from handling the fault we need to check the pi_state after | |
2429 | * reacquiring the locks and before trying to do another fixup. When | |
2430 | * the fixup has been done already we simply return. | |
734009e9 PZ |
2431 | * |
2432 | * Note: we hold both hb->lock and pi_mutex->wait_lock. We can safely | |
2433 | * drop hb->lock since the caller owns the hb -> futex_q relation. | |
2434 | * Dropping the pi_mutex->wait_lock requires the state revalidate. | |
d0aa7a70 | 2435 | */ |
6b4f4bc9 | 2436 | handle_err: |
734009e9 | 2437 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
1b7558e4 | 2438 | spin_unlock(q->lock_ptr); |
778e9a9c | 2439 | |
6b4f4bc9 WD |
2440 | switch (err) { |
2441 | case -EFAULT: | |
f2dac39d | 2442 | err = fault_in_user_writeable(uaddr); |
6b4f4bc9 WD |
2443 | break; |
2444 | ||
2445 | case -EAGAIN: | |
2446 | cond_resched(); | |
f2dac39d | 2447 | err = 0; |
6b4f4bc9 WD |
2448 | break; |
2449 | ||
2450 | default: | |
2451 | WARN_ON_ONCE(1); | |
6b4f4bc9 WD |
2452 | break; |
2453 | } | |
778e9a9c | 2454 | |
1b7558e4 | 2455 | spin_lock(q->lock_ptr); |
734009e9 | 2456 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
778e9a9c | 2457 | |
1b7558e4 TG |
2458 | /* |
2459 | * Check if someone else fixed it for us: | |
2460 | */ | |
f2dac39d TG |
2461 | if (pi_state->owner != oldowner) |
2462 | return argowner == current; | |
1b7558e4 | 2463 | |
f2dac39d TG |
2464 | /* Retry if err was -EAGAIN or the fault in succeeded */ |
2465 | if (!err) | |
2466 | goto retry; | |
1b7558e4 | 2467 | |
34b1a1ce TG |
2468 | /* |
2469 | * fault_in_user_writeable() failed so user state is immutable. At | |
2470 | * best we can make the kernel state consistent but user state will | |
2471 | * be most likely hosed and any subsequent unlock operation will be | |
2472 | * rejected due to PI futex rule [10]. | |
2473 | * | |
2474 | * Ensure that the rtmutex owner is also the pi_state owner despite | |
2475 | * the user space value claiming something different. There is no | |
2476 | * point in unlocking the rtmutex if current is the owner as it | |
2477 | * would need to wait until the next waiter has taken the rtmutex | |
2478 | * to guarantee consistent state. Keep it simple. Userspace asked | |
2479 | * for this wreckaged state. | |
2480 | * | |
2481 | * The rtmutex has an owner - either current or some other | |
2482 | * task. See the EAGAIN loop above. | |
2483 | */ | |
2484 | pi_state_update_owner(pi_state, rt_mutex_owner(&pi_state->pi_mutex)); | |
734009e9 | 2485 | |
f2dac39d TG |
2486 | return err; |
2487 | } | |
734009e9 | 2488 | |
f2dac39d TG |
2489 | static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q, |
2490 | struct task_struct *argowner) | |
2491 | { | |
2492 | struct futex_pi_state *pi_state = q->pi_state; | |
2493 | int ret; | |
2494 | ||
2495 | lockdep_assert_held(q->lock_ptr); | |
2496 | ||
2497 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); | |
2498 | ret = __fixup_pi_state_owner(uaddr, q, argowner); | |
734009e9 PZ |
2499 | raw_spin_unlock_irq(&pi_state->pi_mutex.wait_lock); |
2500 | return ret; | |
d0aa7a70 PP |
2501 | } |
2502 | ||
72c1bbf3 | 2503 | static long futex_wait_restart(struct restart_block *restart); |
36cf3b5c | 2504 | |
dd973998 DH |
2505 | /** |
2506 | * fixup_owner() - Post lock pi_state and corner case management | |
2507 | * @uaddr: user address of the futex | |
dd973998 DH |
2508 | * @q: futex_q (contains pi_state and access to the rt_mutex) |
2509 | * @locked: if the attempt to take the rt_mutex succeeded (1) or not (0) | |
2510 | * | |
2511 | * After attempting to lock an rt_mutex, this function is called to cleanup | |
2512 | * the pi_state owner as well as handle race conditions that may allow us to | |
2513 | * acquire the lock. Must be called with the hb lock held. | |
2514 | * | |
6c23cbbd | 2515 | * Return: |
7b4ff1ad MCC |
2516 | * - 1 - success, lock taken; |
2517 | * - 0 - success, lock not taken; | |
2518 | * - <0 - on error (-EFAULT) | |
dd973998 | 2519 | */ |
ae791a2d | 2520 | static int fixup_owner(u32 __user *uaddr, struct futex_q *q, int locked) |
dd973998 | 2521 | { |
dd973998 DH |
2522 | if (locked) { |
2523 | /* | |
2524 | * Got the lock. We might not be the anticipated owner if we | |
2525 | * did a lock-steal - fix up the PI-state in that case: | |
16ffa12d | 2526 | * |
c1e2f0ea PZ |
2527 | * Speculative pi_state->owner read (we don't hold wait_lock); |
2528 | * since we own the lock pi_state->owner == current is the | |
2529 | * stable state, anything else needs more attention. | |
dd973998 DH |
2530 | */ |
2531 | if (q->pi_state->owner != current) | |
12bb3f7f TG |
2532 | return fixup_pi_state_owner(uaddr, q, current); |
2533 | return 1; | |
dd973998 DH |
2534 | } |
2535 | ||
c1e2f0ea PZ |
2536 | /* |
2537 | * If we didn't get the lock; check if anybody stole it from us. In | |
2538 | * that case, we need to fix up the uval to point to them instead of | |
2539 | * us, otherwise bad things happen. [10] | |
2540 | * | |
2541 | * Another speculative read; pi_state->owner == current is unstable | |
2542 | * but needs our attention. | |
2543 | */ | |
12bb3f7f TG |
2544 | if (q->pi_state->owner == current) |
2545 | return fixup_pi_state_owner(uaddr, q, NULL); | |
c1e2f0ea | 2546 | |
dd973998 DH |
2547 | /* |
2548 | * Paranoia check. If we did not take the lock, then we should not be | |
04b79c55 | 2549 | * the owner of the rt_mutex. Warn and establish consistent state. |
dd973998 | 2550 | */ |
04b79c55 TG |
2551 | if (WARN_ON_ONCE(rt_mutex_owner(&q->pi_state->pi_mutex) == current)) |
2552 | return fixup_pi_state_owner(uaddr, q, current); | |
dd973998 | 2553 | |
12bb3f7f | 2554 | return 0; |
dd973998 DH |
2555 | } |
2556 | ||
ca5f9524 DH |
2557 | /** |
2558 | * futex_wait_queue_me() - queue_me() and wait for wakeup, timeout, or signal | |
2559 | * @hb: the futex hash bucket, must be locked by the caller | |
2560 | * @q: the futex_q to queue up on | |
2561 | * @timeout: the prepared hrtimer_sleeper, or null for no timeout | |
ca5f9524 DH |
2562 | */ |
2563 | static void futex_wait_queue_me(struct futex_hash_bucket *hb, struct futex_q *q, | |
f1a11e05 | 2564 | struct hrtimer_sleeper *timeout) |
ca5f9524 | 2565 | { |
9beba3c5 DH |
2566 | /* |
2567 | * The task state is guaranteed to be set before another task can | |
b92b8b35 | 2568 | * wake it. set_current_state() is implemented using smp_store_mb() and |
9beba3c5 DH |
2569 | * queue_me() calls spin_unlock() upon completion, both serializing |
2570 | * access to the hash list and forcing another memory barrier. | |
2571 | */ | |
f1a11e05 | 2572 | set_current_state(TASK_INTERRUPTIBLE); |
0729e196 | 2573 | queue_me(q, hb); |
ca5f9524 DH |
2574 | |
2575 | /* Arm the timer */ | |
2e4b0d3f | 2576 | if (timeout) |
9dd8813e | 2577 | hrtimer_sleeper_start_expires(timeout, HRTIMER_MODE_ABS); |
ca5f9524 DH |
2578 | |
2579 | /* | |
0729e196 DH |
2580 | * If we have been removed from the hash list, then another task |
2581 | * has tried to wake us, and we can skip the call to schedule(). | |
ca5f9524 DH |
2582 | */ |
2583 | if (likely(!plist_node_empty(&q->list))) { | |
2584 | /* | |
2585 | * If the timer has already expired, current will already be | |
2586 | * flagged for rescheduling. Only call schedule if there | |
2587 | * is no timeout, or if it has yet to expire. | |
2588 | */ | |
2589 | if (!timeout || timeout->task) | |
88c8004f | 2590 | freezable_schedule(); |
ca5f9524 DH |
2591 | } |
2592 | __set_current_state(TASK_RUNNING); | |
2593 | } | |
2594 | ||
f801073f DH |
2595 | /** |
2596 | * futex_wait_setup() - Prepare to wait on a futex | |
2597 | * @uaddr: the futex userspace address | |
2598 | * @val: the expected value | |
b41277dc | 2599 | * @flags: futex flags (FLAGS_SHARED, etc.) |
f801073f DH |
2600 | * @q: the associated futex_q |
2601 | * @hb: storage for hash_bucket pointer to be returned to caller | |
2602 | * | |
2603 | * Setup the futex_q and locate the hash_bucket. Get the futex value and | |
2604 | * compare it with the expected value. Handle atomic faults internally. | |
c363b7ed | 2605 | * Return with the hb lock held on success, and unlocked on failure. |
f801073f | 2606 | * |
6c23cbbd | 2607 | * Return: |
7b4ff1ad MCC |
2608 | * - 0 - uaddr contains val and hb has been locked; |
2609 | * - <1 - -EFAULT or -EWOULDBLOCK (uaddr does not contain val) and hb is unlocked | |
f801073f | 2610 | */ |
b41277dc | 2611 | static int futex_wait_setup(u32 __user *uaddr, u32 val, unsigned int flags, |
f801073f | 2612 | struct futex_q *q, struct futex_hash_bucket **hb) |
1da177e4 | 2613 | { |
e2970f2f IM |
2614 | u32 uval; |
2615 | int ret; | |
1da177e4 | 2616 | |
1da177e4 | 2617 | /* |
b2d0994b | 2618 | * Access the page AFTER the hash-bucket is locked. |
1da177e4 LT |
2619 | * Order is important: |
2620 | * | |
2621 | * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); | |
2622 | * Userspace waker: if (cond(var)) { var = new; futex_wake(&var); } | |
2623 | * | |
2624 | * The basic logical guarantee of a futex is that it blocks ONLY | |
2625 | * if cond(var) is known to be true at the time of blocking, for | |
8fe8f545 ML |
2626 | * any cond. If we locked the hash-bucket after testing *uaddr, that |
2627 | * would open a race condition where we could block indefinitely with | |
1da177e4 LT |
2628 | * cond(var) false, which would violate the guarantee. |
2629 | * | |
8fe8f545 ML |
2630 | * On the other hand, we insert q and release the hash-bucket only |
2631 | * after testing *uaddr. This guarantees that futex_wait() will NOT | |
2632 | * absorb a wakeup if *uaddr does not match the desired values | |
2633 | * while the syscall executes. | |
1da177e4 | 2634 | */ |
f801073f | 2635 | retry: |
96d4f267 | 2636 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q->key, FUTEX_READ); |
f801073f | 2637 | if (unlikely(ret != 0)) |
a5a2a0c7 | 2638 | return ret; |
f801073f DH |
2639 | |
2640 | retry_private: | |
2641 | *hb = queue_lock(q); | |
2642 | ||
e2970f2f | 2643 | ret = get_futex_value_locked(&uval, uaddr); |
1da177e4 | 2644 | |
f801073f | 2645 | if (ret) { |
0d00c7b2 | 2646 | queue_unlock(*hb); |
1da177e4 | 2647 | |
e2970f2f | 2648 | ret = get_user(uval, uaddr); |
e4dc5b7a | 2649 | if (ret) |
d7c5ed73 | 2650 | return ret; |
1da177e4 | 2651 | |
b41277dc | 2652 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2653 | goto retry_private; |
2654 | ||
e4dc5b7a | 2655 | goto retry; |
1da177e4 | 2656 | } |
ca5f9524 | 2657 | |
f801073f | 2658 | if (uval != val) { |
0d00c7b2 | 2659 | queue_unlock(*hb); |
f801073f | 2660 | ret = -EWOULDBLOCK; |
2fff78c7 | 2661 | } |
1da177e4 | 2662 | |
f801073f DH |
2663 | return ret; |
2664 | } | |
2665 | ||
b41277dc DH |
2666 | static int futex_wait(u32 __user *uaddr, unsigned int flags, u32 val, |
2667 | ktime_t *abs_time, u32 bitset) | |
f801073f | 2668 | { |
5ca584d9 | 2669 | struct hrtimer_sleeper timeout, *to; |
f801073f DH |
2670 | struct restart_block *restart; |
2671 | struct futex_hash_bucket *hb; | |
5bdb05f9 | 2672 | struct futex_q q = futex_q_init; |
f801073f DH |
2673 | int ret; |
2674 | ||
2675 | if (!bitset) | |
2676 | return -EINVAL; | |
f801073f DH |
2677 | q.bitset = bitset; |
2678 | ||
5ca584d9 WL |
2679 | to = futex_setup_timer(abs_time, &timeout, flags, |
2680 | current->timer_slack_ns); | |
d58e6576 | 2681 | retry: |
7ada876a | 2682 | /* |
c363b7ed TG |
2683 | * Prepare to wait on uaddr. On success, it holds hb->lock and q |
2684 | * is initialized. | |
7ada876a | 2685 | */ |
b41277dc | 2686 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
f801073f DH |
2687 | if (ret) |
2688 | goto out; | |
2689 | ||
ca5f9524 | 2690 | /* queue_me and wait for wakeup, timeout, or a signal. */ |
f1a11e05 | 2691 | futex_wait_queue_me(hb, &q, to); |
1da177e4 LT |
2692 | |
2693 | /* If we were woken (and unqueued), we succeeded, whatever. */ | |
2fff78c7 | 2694 | ret = 0; |
1da177e4 | 2695 | if (!unqueue_me(&q)) |
7ada876a | 2696 | goto out; |
2fff78c7 | 2697 | ret = -ETIMEDOUT; |
ca5f9524 | 2698 | if (to && !to->task) |
7ada876a | 2699 | goto out; |
72c1bbf3 | 2700 | |
e2970f2f | 2701 | /* |
d58e6576 TG |
2702 | * We expect signal_pending(current), but we might be the |
2703 | * victim of a spurious wakeup as well. | |
e2970f2f | 2704 | */ |
7ada876a | 2705 | if (!signal_pending(current)) |
d58e6576 | 2706 | goto retry; |
d58e6576 | 2707 | |
2fff78c7 | 2708 | ret = -ERESTARTSYS; |
c19384b5 | 2709 | if (!abs_time) |
7ada876a | 2710 | goto out; |
1da177e4 | 2711 | |
f56141e3 | 2712 | restart = ¤t->restart_block; |
a3c74c52 | 2713 | restart->futex.uaddr = uaddr; |
2fff78c7 | 2714 | restart->futex.val = val; |
2456e855 | 2715 | restart->futex.time = *abs_time; |
2fff78c7 | 2716 | restart->futex.bitset = bitset; |
0cd9c649 | 2717 | restart->futex.flags = flags | FLAGS_HAS_TIMEOUT; |
42d35d48 | 2718 | |
5abbe51a | 2719 | ret = set_restart_fn(restart, futex_wait_restart); |
2fff78c7 | 2720 | |
42d35d48 | 2721 | out: |
ca5f9524 DH |
2722 | if (to) { |
2723 | hrtimer_cancel(&to->timer); | |
2724 | destroy_hrtimer_on_stack(&to->timer); | |
2725 | } | |
c87e2837 IM |
2726 | return ret; |
2727 | } | |
2728 | ||
72c1bbf3 NP |
2729 | |
2730 | static long futex_wait_restart(struct restart_block *restart) | |
2731 | { | |
a3c74c52 | 2732 | u32 __user *uaddr = restart->futex.uaddr; |
a72188d8 | 2733 | ktime_t t, *tp = NULL; |
72c1bbf3 | 2734 | |
a72188d8 | 2735 | if (restart->futex.flags & FLAGS_HAS_TIMEOUT) { |
2456e855 | 2736 | t = restart->futex.time; |
a72188d8 DH |
2737 | tp = &t; |
2738 | } | |
72c1bbf3 | 2739 | restart->fn = do_no_restart_syscall; |
b41277dc DH |
2740 | |
2741 | return (long)futex_wait(uaddr, restart->futex.flags, | |
2742 | restart->futex.val, tp, restart->futex.bitset); | |
72c1bbf3 NP |
2743 | } |
2744 | ||
2745 | ||
c87e2837 IM |
2746 | /* |
2747 | * Userspace tried a 0 -> TID atomic transition of the futex value | |
2748 | * and failed. The kernel side here does the whole locking operation: | |
767f509c DB |
2749 | * if there are waiters then it will block as a consequence of relying |
2750 | * on rt-mutexes, it does PI, etc. (Due to races the kernel might see | |
2751 | * a 0 value of the futex too.). | |
2752 | * | |
2753 | * Also serves as futex trylock_pi()'ing, and due semantics. | |
c87e2837 | 2754 | */ |
996636dd | 2755 | static int futex_lock_pi(u32 __user *uaddr, unsigned int flags, |
b41277dc | 2756 | ktime_t *time, int trylock) |
c87e2837 | 2757 | { |
5ca584d9 | 2758 | struct hrtimer_sleeper timeout, *to; |
3ef240ea | 2759 | struct task_struct *exiting = NULL; |
cfafcd11 | 2760 | struct rt_mutex_waiter rt_waiter; |
c87e2837 | 2761 | struct futex_hash_bucket *hb; |
5bdb05f9 | 2762 | struct futex_q q = futex_q_init; |
dd973998 | 2763 | int res, ret; |
c87e2837 | 2764 | |
bc2eecd7 NP |
2765 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
2766 | return -ENOSYS; | |
2767 | ||
c87e2837 IM |
2768 | if (refill_pi_state_cache()) |
2769 | return -ENOMEM; | |
2770 | ||
e112c413 | 2771 | to = futex_setup_timer(time, &timeout, flags, 0); |
c5780e97 | 2772 | |
42d35d48 | 2773 | retry: |
96d4f267 | 2774 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &q.key, FUTEX_WRITE); |
c87e2837 | 2775 | if (unlikely(ret != 0)) |
42d35d48 | 2776 | goto out; |
c87e2837 | 2777 | |
e4dc5b7a | 2778 | retry_private: |
82af7aca | 2779 | hb = queue_lock(&q); |
c87e2837 | 2780 | |
3ef240ea TG |
2781 | ret = futex_lock_pi_atomic(uaddr, hb, &q.key, &q.pi_state, current, |
2782 | &exiting, 0); | |
c87e2837 | 2783 | if (unlikely(ret)) { |
767f509c DB |
2784 | /* |
2785 | * Atomic work succeeded and we got the lock, | |
2786 | * or failed. Either way, we do _not_ block. | |
2787 | */ | |
778e9a9c | 2788 | switch (ret) { |
1a52084d DH |
2789 | case 1: |
2790 | /* We got the lock. */ | |
2791 | ret = 0; | |
2792 | goto out_unlock_put_key; | |
2793 | case -EFAULT: | |
2794 | goto uaddr_faulted; | |
ac31c7ff | 2795 | case -EBUSY: |
778e9a9c AK |
2796 | case -EAGAIN: |
2797 | /* | |
af54d6a1 | 2798 | * Two reasons for this: |
ac31c7ff | 2799 | * - EBUSY: Task is exiting and we just wait for the |
af54d6a1 | 2800 | * exit to complete. |
ac31c7ff | 2801 | * - EAGAIN: The user space value changed. |
778e9a9c | 2802 | */ |
0d00c7b2 | 2803 | queue_unlock(hb); |
3ef240ea TG |
2804 | /* |
2805 | * Handle the case where the owner is in the middle of | |
2806 | * exiting. Wait for the exit to complete otherwise | |
2807 | * this task might loop forever, aka. live lock. | |
2808 | */ | |
2809 | wait_for_owner_exiting(ret, exiting); | |
778e9a9c AK |
2810 | cond_resched(); |
2811 | goto retry; | |
778e9a9c | 2812 | default: |
42d35d48 | 2813 | goto out_unlock_put_key; |
c87e2837 | 2814 | } |
c87e2837 IM |
2815 | } |
2816 | ||
cfafcd11 PZ |
2817 | WARN_ON(!q.pi_state); |
2818 | ||
c87e2837 IM |
2819 | /* |
2820 | * Only actually queue now that the atomic ops are done: | |
2821 | */ | |
cfafcd11 | 2822 | __queue_me(&q, hb); |
c87e2837 | 2823 | |
cfafcd11 | 2824 | if (trylock) { |
5293c2ef | 2825 | ret = rt_mutex_futex_trylock(&q.pi_state->pi_mutex); |
c87e2837 IM |
2826 | /* Fixup the trylock return value: */ |
2827 | ret = ret ? 0 : -EWOULDBLOCK; | |
cfafcd11 | 2828 | goto no_block; |
c87e2837 IM |
2829 | } |
2830 | ||
56222b21 PZ |
2831 | rt_mutex_init_waiter(&rt_waiter); |
2832 | ||
cfafcd11 | 2833 | /* |
56222b21 PZ |
2834 | * On PREEMPT_RT_FULL, when hb->lock becomes an rt_mutex, we must not |
2835 | * hold it while doing rt_mutex_start_proxy(), because then it will | |
2836 | * include hb->lock in the blocking chain, even through we'll not in | |
2837 | * fact hold it while blocking. This will lead it to report -EDEADLK | |
2838 | * and BUG when futex_unlock_pi() interleaves with this. | |
2839 | * | |
2840 | * Therefore acquire wait_lock while holding hb->lock, but drop the | |
1a1fb985 TG |
2841 | * latter before calling __rt_mutex_start_proxy_lock(). This |
2842 | * interleaves with futex_unlock_pi() -- which does a similar lock | |
2843 | * handoff -- such that the latter can observe the futex_q::pi_state | |
2844 | * before __rt_mutex_start_proxy_lock() is done. | |
cfafcd11 | 2845 | */ |
56222b21 PZ |
2846 | raw_spin_lock_irq(&q.pi_state->pi_mutex.wait_lock); |
2847 | spin_unlock(q.lock_ptr); | |
1a1fb985 TG |
2848 | /* |
2849 | * __rt_mutex_start_proxy_lock() unconditionally enqueues the @rt_waiter | |
2850 | * such that futex_unlock_pi() is guaranteed to observe the waiter when | |
2851 | * it sees the futex_q::pi_state. | |
2852 | */ | |
56222b21 PZ |
2853 | ret = __rt_mutex_start_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter, current); |
2854 | raw_spin_unlock_irq(&q.pi_state->pi_mutex.wait_lock); | |
2855 | ||
cfafcd11 PZ |
2856 | if (ret) { |
2857 | if (ret == 1) | |
2858 | ret = 0; | |
1a1fb985 | 2859 | goto cleanup; |
cfafcd11 PZ |
2860 | } |
2861 | ||
cfafcd11 | 2862 | if (unlikely(to)) |
9dd8813e | 2863 | hrtimer_sleeper_start_expires(to, HRTIMER_MODE_ABS); |
cfafcd11 PZ |
2864 | |
2865 | ret = rt_mutex_wait_proxy_lock(&q.pi_state->pi_mutex, to, &rt_waiter); | |
2866 | ||
1a1fb985 | 2867 | cleanup: |
a99e4e41 | 2868 | spin_lock(q.lock_ptr); |
cfafcd11 | 2869 | /* |
1a1fb985 | 2870 | * If we failed to acquire the lock (deadlock/signal/timeout), we must |
cfafcd11 | 2871 | * first acquire the hb->lock before removing the lock from the |
1a1fb985 TG |
2872 | * rt_mutex waitqueue, such that we can keep the hb and rt_mutex wait |
2873 | * lists consistent. | |
56222b21 PZ |
2874 | * |
2875 | * In particular; it is important that futex_unlock_pi() can not | |
2876 | * observe this inconsistency. | |
cfafcd11 PZ |
2877 | */ |
2878 | if (ret && !rt_mutex_cleanup_proxy_lock(&q.pi_state->pi_mutex, &rt_waiter)) | |
2879 | ret = 0; | |
2880 | ||
2881 | no_block: | |
dd973998 DH |
2882 | /* |
2883 | * Fixup the pi_state owner and possibly acquire the lock if we | |
2884 | * haven't already. | |
2885 | */ | |
ae791a2d | 2886 | res = fixup_owner(uaddr, &q, !ret); |
dd973998 | 2887 | /* |
93d0955e | 2888 | * If fixup_owner() returned an error, propagate that. If it acquired |
dd973998 DH |
2889 | * the lock, clear our -ETIMEDOUT or -EINTR. |
2890 | */ | |
2891 | if (res) | |
2892 | ret = (res < 0) ? res : 0; | |
c87e2837 | 2893 | |
778e9a9c | 2894 | unqueue_me_pi(&q); |
a3f2428d | 2895 | spin_unlock(q.lock_ptr); |
9180bd46 | 2896 | goto out; |
c87e2837 | 2897 | |
42d35d48 | 2898 | out_unlock_put_key: |
0d00c7b2 | 2899 | queue_unlock(hb); |
c87e2837 | 2900 | |
42d35d48 | 2901 | out: |
97181f9b TG |
2902 | if (to) { |
2903 | hrtimer_cancel(&to->timer); | |
237fc6e7 | 2904 | destroy_hrtimer_on_stack(&to->timer); |
97181f9b | 2905 | } |
dd973998 | 2906 | return ret != -EINTR ? ret : -ERESTARTNOINTR; |
c87e2837 | 2907 | |
42d35d48 | 2908 | uaddr_faulted: |
0d00c7b2 | 2909 | queue_unlock(hb); |
778e9a9c | 2910 | |
d0725992 | 2911 | ret = fault_in_user_writeable(uaddr); |
e4dc5b7a | 2912 | if (ret) |
9180bd46 | 2913 | goto out; |
c87e2837 | 2914 | |
b41277dc | 2915 | if (!(flags & FLAGS_SHARED)) |
e4dc5b7a DH |
2916 | goto retry_private; |
2917 | ||
e4dc5b7a | 2918 | goto retry; |
c87e2837 IM |
2919 | } |
2920 | ||
c87e2837 IM |
2921 | /* |
2922 | * Userspace attempted a TID -> 0 atomic transition, and failed. | |
2923 | * This is the in-kernel slowpath: we look up the PI state (if any), | |
2924 | * and do the rt-mutex unlock. | |
2925 | */ | |
b41277dc | 2926 | static int futex_unlock_pi(u32 __user *uaddr, unsigned int flags) |
c87e2837 | 2927 | { |
3f649ab7 | 2928 | u32 curval, uval, vpid = task_pid_vnr(current); |
38d47c1b | 2929 | union futex_key key = FUTEX_KEY_INIT; |
ccf9e6a8 | 2930 | struct futex_hash_bucket *hb; |
499f5aca | 2931 | struct futex_q *top_waiter; |
e4dc5b7a | 2932 | int ret; |
c87e2837 | 2933 | |
bc2eecd7 NP |
2934 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
2935 | return -ENOSYS; | |
2936 | ||
c87e2837 IM |
2937 | retry: |
2938 | if (get_user(uval, uaddr)) | |
2939 | return -EFAULT; | |
2940 | /* | |
2941 | * We release only a lock we actually own: | |
2942 | */ | |
c0c9ed15 | 2943 | if ((uval & FUTEX_TID_MASK) != vpid) |
c87e2837 | 2944 | return -EPERM; |
c87e2837 | 2945 | |
96d4f267 | 2946 | ret = get_futex_key(uaddr, flags & FLAGS_SHARED, &key, FUTEX_WRITE); |
ccf9e6a8 TG |
2947 | if (ret) |
2948 | return ret; | |
c87e2837 IM |
2949 | |
2950 | hb = hash_futex(&key); | |
2951 | spin_lock(&hb->lock); | |
2952 | ||
c87e2837 | 2953 | /* |
ccf9e6a8 TG |
2954 | * Check waiters first. We do not trust user space values at |
2955 | * all and we at least want to know if user space fiddled | |
2956 | * with the futex value instead of blindly unlocking. | |
c87e2837 | 2957 | */ |
499f5aca PZ |
2958 | top_waiter = futex_top_waiter(hb, &key); |
2959 | if (top_waiter) { | |
16ffa12d PZ |
2960 | struct futex_pi_state *pi_state = top_waiter->pi_state; |
2961 | ||
2962 | ret = -EINVAL; | |
2963 | if (!pi_state) | |
2964 | goto out_unlock; | |
2965 | ||
2966 | /* | |
2967 | * If current does not own the pi_state then the futex is | |
2968 | * inconsistent and user space fiddled with the futex value. | |
2969 | */ | |
2970 | if (pi_state->owner != current) | |
2971 | goto out_unlock; | |
2972 | ||
bebe5b51 | 2973 | get_pi_state(pi_state); |
802ab58d | 2974 | /* |
bebe5b51 PZ |
2975 | * By taking wait_lock while still holding hb->lock, we ensure |
2976 | * there is no point where we hold neither; and therefore | |
2977 | * wake_futex_pi() must observe a state consistent with what we | |
2978 | * observed. | |
1a1fb985 TG |
2979 | * |
2980 | * In particular; this forces __rt_mutex_start_proxy() to | |
2981 | * complete such that we're guaranteed to observe the | |
2982 | * rt_waiter. Also see the WARN in wake_futex_pi(). | |
16ffa12d | 2983 | */ |
bebe5b51 | 2984 | raw_spin_lock_irq(&pi_state->pi_mutex.wait_lock); |
16ffa12d PZ |
2985 | spin_unlock(&hb->lock); |
2986 | ||
c74aef2d | 2987 | /* drops pi_state->pi_mutex.wait_lock */ |
16ffa12d PZ |
2988 | ret = wake_futex_pi(uaddr, uval, pi_state); |
2989 | ||
2990 | put_pi_state(pi_state); | |
2991 | ||
2992 | /* | |
2993 | * Success, we're done! No tricky corner cases. | |
802ab58d SAS |
2994 | */ |
2995 | if (!ret) | |
0f943850 | 2996 | return ret; |
c87e2837 | 2997 | /* |
ccf9e6a8 TG |
2998 | * The atomic access to the futex value generated a |
2999 | * pagefault, so retry the user-access and the wakeup: | |
c87e2837 IM |
3000 | */ |
3001 | if (ret == -EFAULT) | |
3002 | goto pi_faulted; | |
89e9e66b SAS |
3003 | /* |
3004 | * A unconditional UNLOCK_PI op raced against a waiter | |
3005 | * setting the FUTEX_WAITERS bit. Try again. | |
3006 | */ | |
6b4f4bc9 WD |
3007 | if (ret == -EAGAIN) |
3008 | goto pi_retry; | |
802ab58d SAS |
3009 | /* |
3010 | * wake_futex_pi has detected invalid state. Tell user | |
3011 | * space. | |
3012 | */ | |
0f943850 | 3013 | return ret; |
c87e2837 | 3014 | } |
ccf9e6a8 | 3015 | |
c87e2837 | 3016 | /* |
ccf9e6a8 TG |
3017 | * We have no kernel internal state, i.e. no waiters in the |
3018 | * kernel. Waiters which are about to queue themselves are stuck | |
3019 | * on hb->lock. So we can safely ignore them. We do neither | |
3020 | * preserve the WAITERS bit not the OWNER_DIED one. We are the | |
3021 | * owner. | |
c87e2837 | 3022 | */ |
6b4f4bc9 | 3023 | if ((ret = cmpxchg_futex_value_locked(&curval, uaddr, uval, 0))) { |
16ffa12d | 3024 | spin_unlock(&hb->lock); |
6b4f4bc9 WD |
3025 | switch (ret) { |
3026 | case -EFAULT: | |
3027 | goto pi_faulted; | |
3028 | ||
3029 | case -EAGAIN: | |
3030 | goto pi_retry; | |
3031 | ||
3032 | default: | |
3033 | WARN_ON_ONCE(1); | |
0f943850 | 3034 | return ret; |
6b4f4bc9 | 3035 | } |
16ffa12d | 3036 | } |
c87e2837 | 3037 | |
ccf9e6a8 TG |
3038 | /* |
3039 | * If uval has changed, let user space handle it. | |
3040 | */ | |
3041 | ret = (curval == uval) ? 0 : -EAGAIN; | |
3042 | ||
c87e2837 IM |
3043 | out_unlock: |
3044 | spin_unlock(&hb->lock); | |
c87e2837 IM |
3045 | return ret; |
3046 | ||
6b4f4bc9 | 3047 | pi_retry: |
6b4f4bc9 WD |
3048 | cond_resched(); |
3049 | goto retry; | |
3050 | ||
c87e2837 | 3051 | pi_faulted: |
c87e2837 | 3052 | |
d0725992 | 3053 | ret = fault_in_user_writeable(uaddr); |
b5686363 | 3054 | if (!ret) |
c87e2837 IM |
3055 | goto retry; |
3056 | ||
1da177e4 LT |
3057 | return ret; |
3058 | } | |
3059 | ||
52400ba9 DH |
3060 | /** |
3061 | * handle_early_requeue_pi_wakeup() - Detect early wakeup on the initial futex | |
3062 | * @hb: the hash_bucket futex_q was original enqueued on | |
3063 | * @q: the futex_q woken while waiting to be requeued | |
3064 | * @key2: the futex_key of the requeue target futex | |
3065 | * @timeout: the timeout associated with the wait (NULL if none) | |
3066 | * | |
3067 | * Detect if the task was woken on the initial futex as opposed to the requeue | |
3068 | * target futex. If so, determine if it was a timeout or a signal that caused | |
3069 | * the wakeup and return the appropriate error code to the caller. Must be | |
3070 | * called with the hb lock held. | |
3071 | * | |
6c23cbbd | 3072 | * Return: |
7b4ff1ad MCC |
3073 | * - 0 = no early wakeup detected; |
3074 | * - <0 = -ETIMEDOUT or -ERESTARTNOINTR | |
52400ba9 DH |
3075 | */ |
3076 | static inline | |
3077 | int handle_early_requeue_pi_wakeup(struct futex_hash_bucket *hb, | |
3078 | struct futex_q *q, union futex_key *key2, | |
3079 | struct hrtimer_sleeper *timeout) | |
3080 | { | |
3081 | int ret = 0; | |
3082 | ||
3083 | /* | |
3084 | * With the hb lock held, we avoid races while we process the wakeup. | |
3085 | * We only need to hold hb (and not hb2) to ensure atomicity as the | |
3086 | * wakeup code can't change q.key from uaddr to uaddr2 if we hold hb. | |
3087 | * It can't be requeued from uaddr2 to something else since we don't | |
3088 | * support a PI aware source futex for requeue. | |
3089 | */ | |
3090 | if (!match_futex(&q->key, key2)) { | |
3091 | WARN_ON(q->lock_ptr && (&hb->lock != q->lock_ptr)); | |
3092 | /* | |
3093 | * We were woken prior to requeue by a timeout or a signal. | |
3094 | * Unqueue the futex_q and determine which it was. | |
3095 | */ | |
2e12978a | 3096 | plist_del(&q->list, &hb->chain); |
11d4616b | 3097 | hb_waiters_dec(hb); |
52400ba9 | 3098 | |
d58e6576 | 3099 | /* Handle spurious wakeups gracefully */ |
11df6ddd | 3100 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3101 | if (timeout && !timeout->task) |
3102 | ret = -ETIMEDOUT; | |
d58e6576 | 3103 | else if (signal_pending(current)) |
1c840c14 | 3104 | ret = -ERESTARTNOINTR; |
52400ba9 DH |
3105 | } |
3106 | return ret; | |
3107 | } | |
3108 | ||
3109 | /** | |
3110 | * futex_wait_requeue_pi() - Wait on uaddr and take uaddr2 | |
56ec1607 | 3111 | * @uaddr: the futex we initially wait on (non-pi) |
b41277dc | 3112 | * @flags: futex flags (FLAGS_SHARED, FLAGS_CLOCKRT, etc.), they must be |
ab51fbab | 3113 | * the same type, no requeueing from private to shared, etc. |
52400ba9 DH |
3114 | * @val: the expected value of uaddr |
3115 | * @abs_time: absolute timeout | |
56ec1607 | 3116 | * @bitset: 32 bit wakeup bitset set by userspace, defaults to all |
52400ba9 DH |
3117 | * @uaddr2: the pi futex we will take prior to returning to user-space |
3118 | * | |
3119 | * The caller will wait on uaddr and will be requeued by futex_requeue() to | |
6f7b0a2a DH |
3120 | * uaddr2 which must be PI aware and unique from uaddr. Normal wakeup will wake |
3121 | * on uaddr2 and complete the acquisition of the rt_mutex prior to returning to | |
3122 | * userspace. This ensures the rt_mutex maintains an owner when it has waiters; | |
3123 | * without one, the pi logic would not know which task to boost/deboost, if | |
3124 | * there was a need to. | |
52400ba9 DH |
3125 | * |
3126 | * We call schedule in futex_wait_queue_me() when we enqueue and return there | |
6c23cbbd | 3127 | * via the following-- |
52400ba9 | 3128 | * 1) wakeup on uaddr2 after an atomic lock acquisition by futex_requeue() |
cc6db4e6 DH |
3129 | * 2) wakeup on uaddr2 after a requeue |
3130 | * 3) signal | |
3131 | * 4) timeout | |
52400ba9 | 3132 | * |
cc6db4e6 | 3133 | * If 3, cleanup and return -ERESTARTNOINTR. |
52400ba9 DH |
3134 | * |
3135 | * If 2, we may then block on trying to take the rt_mutex and return via: | |
3136 | * 5) successful lock | |
3137 | * 6) signal | |
3138 | * 7) timeout | |
3139 | * 8) other lock acquisition failure | |
3140 | * | |
cc6db4e6 | 3141 | * If 6, return -EWOULDBLOCK (restarting the syscall would do the same). |
52400ba9 DH |
3142 | * |
3143 | * If 4 or 7, we cleanup and return with -ETIMEDOUT. | |
3144 | * | |
6c23cbbd | 3145 | * Return: |
7b4ff1ad MCC |
3146 | * - 0 - On success; |
3147 | * - <0 - On error | |
52400ba9 | 3148 | */ |
b41277dc | 3149 | static int futex_wait_requeue_pi(u32 __user *uaddr, unsigned int flags, |
52400ba9 | 3150 | u32 val, ktime_t *abs_time, u32 bitset, |
b41277dc | 3151 | u32 __user *uaddr2) |
52400ba9 | 3152 | { |
5ca584d9 | 3153 | struct hrtimer_sleeper timeout, *to; |
52400ba9 | 3154 | struct rt_mutex_waiter rt_waiter; |
52400ba9 | 3155 | struct futex_hash_bucket *hb; |
5bdb05f9 DH |
3156 | union futex_key key2 = FUTEX_KEY_INIT; |
3157 | struct futex_q q = futex_q_init; | |
52400ba9 | 3158 | int res, ret; |
52400ba9 | 3159 | |
bc2eecd7 NP |
3160 | if (!IS_ENABLED(CONFIG_FUTEX_PI)) |
3161 | return -ENOSYS; | |
3162 | ||
6f7b0a2a DH |
3163 | if (uaddr == uaddr2) |
3164 | return -EINVAL; | |
3165 | ||
52400ba9 DH |
3166 | if (!bitset) |
3167 | return -EINVAL; | |
3168 | ||
5ca584d9 WL |
3169 | to = futex_setup_timer(abs_time, &timeout, flags, |
3170 | current->timer_slack_ns); | |
52400ba9 DH |
3171 | |
3172 | /* | |
3173 | * The waiter is allocated on our stack, manipulated by the requeue | |
3174 | * code while we sleep on uaddr. | |
3175 | */ | |
50809358 | 3176 | rt_mutex_init_waiter(&rt_waiter); |
52400ba9 | 3177 | |
96d4f267 | 3178 | ret = get_futex_key(uaddr2, flags & FLAGS_SHARED, &key2, FUTEX_WRITE); |
52400ba9 DH |
3179 | if (unlikely(ret != 0)) |
3180 | goto out; | |
3181 | ||
84bc4af5 DH |
3182 | q.bitset = bitset; |
3183 | q.rt_waiter = &rt_waiter; | |
3184 | q.requeue_pi_key = &key2; | |
3185 | ||
7ada876a | 3186 | /* |
c363b7ed TG |
3187 | * Prepare to wait on uaddr. On success, it holds hb->lock and q |
3188 | * is initialized. | |
7ada876a | 3189 | */ |
b41277dc | 3190 | ret = futex_wait_setup(uaddr, val, flags, &q, &hb); |
c8b15a70 | 3191 | if (ret) |
9180bd46 | 3192 | goto out; |
52400ba9 | 3193 | |
e9c243a5 TG |
3194 | /* |
3195 | * The check above which compares uaddrs is not sufficient for | |
3196 | * shared futexes. We need to compare the keys: | |
3197 | */ | |
3198 | if (match_futex(&q.key, &key2)) { | |
13c42c2f | 3199 | queue_unlock(hb); |
e9c243a5 | 3200 | ret = -EINVAL; |
9180bd46 | 3201 | goto out; |
e9c243a5 TG |
3202 | } |
3203 | ||
52400ba9 | 3204 | /* Queue the futex_q, drop the hb lock, wait for wakeup. */ |
f1a11e05 | 3205 | futex_wait_queue_me(hb, &q, to); |
52400ba9 DH |
3206 | |
3207 | spin_lock(&hb->lock); | |
3208 | ret = handle_early_requeue_pi_wakeup(hb, &q, &key2, to); | |
3209 | spin_unlock(&hb->lock); | |
3210 | if (ret) | |
9180bd46 | 3211 | goto out; |
52400ba9 DH |
3212 | |
3213 | /* | |
3214 | * In order for us to be here, we know our q.key == key2, and since | |
3215 | * we took the hb->lock above, we also know that futex_requeue() has | |
3216 | * completed and we no longer have to concern ourselves with a wakeup | |
c363b7ed | 3217 | * race with the atomic proxy lock acquisition by the requeue code. |
52400ba9 DH |
3218 | */ |
3219 | ||
a1565aa4 DB |
3220 | /* |
3221 | * Check if the requeue code acquired the second futex for us and do | |
3222 | * any pertinent fixup. | |
3223 | */ | |
52400ba9 | 3224 | if (!q.rt_waiter) { |
52400ba9 DH |
3225 | if (q.pi_state && (q.pi_state->owner != current)) { |
3226 | spin_lock(q.lock_ptr); | |
a1565aa4 | 3227 | ret = fixup_owner(uaddr2, &q, true); |
fb75a428 TG |
3228 | /* |
3229 | * Drop the reference to the pi state which | |
3230 | * the requeue_pi() code acquired for us. | |
3231 | */ | |
29e9ee5d | 3232 | put_pi_state(q.pi_state); |
52400ba9 | 3233 | spin_unlock(q.lock_ptr); |
12bb3f7f TG |
3234 | /* |
3235 | * Adjust the return value. It's either -EFAULT or | |
3236 | * success (1) but the caller expects 0 for success. | |
3237 | */ | |
3238 | ret = ret < 0 ? ret : 0; | |
52400ba9 DH |
3239 | } |
3240 | } else { | |
830e6acc | 3241 | struct rt_mutex_base *pi_mutex; |
c236c8e9 | 3242 | |
52400ba9 DH |
3243 | /* |
3244 | * We have been woken up by futex_unlock_pi(), a timeout, or a | |
3245 | * signal. futex_unlock_pi() will not destroy the lock_ptr nor | |
3246 | * the pi_state. | |
3247 | */ | |
f27071cb | 3248 | WARN_ON(!q.pi_state); |
52400ba9 | 3249 | pi_mutex = &q.pi_state->pi_mutex; |
38d589f2 | 3250 | ret = rt_mutex_wait_proxy_lock(pi_mutex, to, &rt_waiter); |
52400ba9 DH |
3251 | |
3252 | spin_lock(q.lock_ptr); | |
38d589f2 PZ |
3253 | if (ret && !rt_mutex_cleanup_proxy_lock(pi_mutex, &rt_waiter)) |
3254 | ret = 0; | |
3255 | ||
3256 | debug_rt_mutex_free_waiter(&rt_waiter); | |
52400ba9 DH |
3257 | /* |
3258 | * Fixup the pi_state owner and possibly acquire the lock if we | |
3259 | * haven't already. | |
3260 | */ | |
ae791a2d | 3261 | res = fixup_owner(uaddr2, &q, !ret); |
52400ba9 | 3262 | /* |
93d0955e | 3263 | * If fixup_owner() returned an error, propagate that. If it |
56ec1607 | 3264 | * acquired the lock, clear -ETIMEDOUT or -EINTR. |
52400ba9 DH |
3265 | */ |
3266 | if (res) | |
3267 | ret = (res < 0) ? res : 0; | |
3268 | ||
52400ba9 | 3269 | unqueue_me_pi(&q); |
a3f2428d | 3270 | spin_unlock(q.lock_ptr); |
52400ba9 DH |
3271 | } |
3272 | ||
c236c8e9 | 3273 | if (ret == -EINTR) { |
52400ba9 | 3274 | /* |
cc6db4e6 DH |
3275 | * We've already been requeued, but cannot restart by calling |
3276 | * futex_lock_pi() directly. We could restart this syscall, but | |
3277 | * it would detect that the user space "val" changed and return | |
3278 | * -EWOULDBLOCK. Save the overhead of the restart and return | |
3279 | * -EWOULDBLOCK directly. | |
52400ba9 | 3280 | */ |
2070887f | 3281 | ret = -EWOULDBLOCK; |
52400ba9 DH |
3282 | } |
3283 | ||
52400ba9 DH |
3284 | out: |
3285 | if (to) { | |
3286 | hrtimer_cancel(&to->timer); | |
3287 | destroy_hrtimer_on_stack(&to->timer); | |
3288 | } | |
3289 | return ret; | |
3290 | } | |
3291 | ||
0771dfef IM |
3292 | /* |
3293 | * Support for robust futexes: the kernel cleans up held futexes at | |
3294 | * thread exit time. | |
3295 | * | |
3296 | * Implementation: user-space maintains a per-thread list of locks it | |
3297 | * is holding. Upon do_exit(), the kernel carefully walks this list, | |
3298 | * and marks all locks that are owned by this thread with the | |
c87e2837 | 3299 | * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is |
0771dfef IM |
3300 | * always manipulated with the lock held, so the list is private and |
3301 | * per-thread. Userspace also maintains a per-thread 'list_op_pending' | |
3302 | * field, to allow the kernel to clean up if the thread dies after | |
3303 | * acquiring the lock, but just before it could have added itself to | |
3304 | * the list. There can only be one such pending lock. | |
3305 | */ | |
3306 | ||
3307 | /** | |
d96ee56c DH |
3308 | * sys_set_robust_list() - Set the robust-futex list head of a task |
3309 | * @head: pointer to the list-head | |
3310 | * @len: length of the list-head, as userspace expects | |
0771dfef | 3311 | */ |
836f92ad HC |
3312 | SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, |
3313 | size_t, len) | |
0771dfef | 3314 | { |
a0c1e907 TG |
3315 | if (!futex_cmpxchg_enabled) |
3316 | return -ENOSYS; | |
0771dfef IM |
3317 | /* |
3318 | * The kernel knows only one size for now: | |
3319 | */ | |
3320 | if (unlikely(len != sizeof(*head))) | |
3321 | return -EINVAL; | |
3322 | ||
3323 | current->robust_list = head; | |
3324 | ||
3325 | return 0; | |
3326 | } | |
3327 | ||
3328 | /** | |
d96ee56c DH |
3329 | * sys_get_robust_list() - Get the robust-futex list head of a task |
3330 | * @pid: pid of the process [zero for current task] | |
3331 | * @head_ptr: pointer to a list-head pointer, the kernel fills it in | |
3332 | * @len_ptr: pointer to a length field, the kernel fills in the header size | |
0771dfef | 3333 | */ |
836f92ad HC |
3334 | SYSCALL_DEFINE3(get_robust_list, int, pid, |
3335 | struct robust_list_head __user * __user *, head_ptr, | |
3336 | size_t __user *, len_ptr) | |
0771dfef | 3337 | { |
ba46df98 | 3338 | struct robust_list_head __user *head; |
0771dfef | 3339 | unsigned long ret; |
bdbb776f | 3340 | struct task_struct *p; |
0771dfef | 3341 | |
a0c1e907 TG |
3342 | if (!futex_cmpxchg_enabled) |
3343 | return -ENOSYS; | |
3344 | ||
bdbb776f KC |
3345 | rcu_read_lock(); |
3346 | ||
3347 | ret = -ESRCH; | |
0771dfef | 3348 | if (!pid) |
bdbb776f | 3349 | p = current; |
0771dfef | 3350 | else { |
228ebcbe | 3351 | p = find_task_by_vpid(pid); |
0771dfef IM |
3352 | if (!p) |
3353 | goto err_unlock; | |
0771dfef IM |
3354 | } |
3355 | ||
bdbb776f | 3356 | ret = -EPERM; |
caaee623 | 3357 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) |
bdbb776f KC |
3358 | goto err_unlock; |
3359 | ||
3360 | head = p->robust_list; | |
3361 | rcu_read_unlock(); | |
3362 | ||
0771dfef IM |
3363 | if (put_user(sizeof(*head), len_ptr)) |
3364 | return -EFAULT; | |
3365 | return put_user(head, head_ptr); | |
3366 | ||
3367 | err_unlock: | |
aaa2a97e | 3368 | rcu_read_unlock(); |
0771dfef IM |
3369 | |
3370 | return ret; | |
3371 | } | |
3372 | ||
ca16d5be YT |
3373 | /* Constants for the pending_op argument of handle_futex_death */ |
3374 | #define HANDLE_DEATH_PENDING true | |
3375 | #define HANDLE_DEATH_LIST false | |
3376 | ||
0771dfef IM |
3377 | /* |
3378 | * Process a futex-list entry, check whether it's owned by the | |
3379 | * dying task, and do notification if so: | |
3380 | */ | |
ca16d5be YT |
3381 | static int handle_futex_death(u32 __user *uaddr, struct task_struct *curr, |
3382 | bool pi, bool pending_op) | |
0771dfef | 3383 | { |
3f649ab7 | 3384 | u32 uval, nval, mval; |
6b4f4bc9 | 3385 | int err; |
0771dfef | 3386 | |
5a07168d CJ |
3387 | /* Futex address must be 32bit aligned */ |
3388 | if ((((unsigned long)uaddr) % sizeof(*uaddr)) != 0) | |
3389 | return -1; | |
3390 | ||
8f17d3a5 IM |
3391 | retry: |
3392 | if (get_user(uval, uaddr)) | |
0771dfef IM |
3393 | return -1; |
3394 | ||
ca16d5be YT |
3395 | /* |
3396 | * Special case for regular (non PI) futexes. The unlock path in | |
3397 | * user space has two race scenarios: | |
3398 | * | |
3399 | * 1. The unlock path releases the user space futex value and | |
3400 | * before it can execute the futex() syscall to wake up | |
3401 | * waiters it is killed. | |
3402 | * | |
3403 | * 2. A woken up waiter is killed before it can acquire the | |
3404 | * futex in user space. | |
3405 | * | |
3406 | * In both cases the TID validation below prevents a wakeup of | |
3407 | * potential waiters which can cause these waiters to block | |
3408 | * forever. | |
3409 | * | |
3410 | * In both cases the following conditions are met: | |
3411 | * | |
3412 | * 1) task->robust_list->list_op_pending != NULL | |
3413 | * @pending_op == true | |
3414 | * 2) User space futex value == 0 | |
3415 | * 3) Regular futex: @pi == false | |
3416 | * | |
3417 | * If these conditions are met, it is safe to attempt waking up a | |
3418 | * potential waiter without touching the user space futex value and | |
3419 | * trying to set the OWNER_DIED bit. The user space futex value is | |
3420 | * uncontended and the rest of the user space mutex state is | |
3421 | * consistent, so a woken waiter will just take over the | |
3422 | * uncontended futex. Setting the OWNER_DIED bit would create | |
3423 | * inconsistent state and malfunction of the user space owner died | |
3424 | * handling. | |
3425 | */ | |
3426 | if (pending_op && !pi && !uval) { | |
3427 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); | |
3428 | return 0; | |
3429 | } | |
3430 | ||
6b4f4bc9 WD |
3431 | if ((uval & FUTEX_TID_MASK) != task_pid_vnr(curr)) |
3432 | return 0; | |
3433 | ||
3434 | /* | |
3435 | * Ok, this dying thread is truly holding a futex | |
3436 | * of interest. Set the OWNER_DIED bit atomically | |
3437 | * via cmpxchg, and if the value had FUTEX_WAITERS | |
3438 | * set, wake up a waiter (if any). (We have to do a | |
3439 | * futex_wake() even if OWNER_DIED is already set - | |
3440 | * to handle the rare but possible case of recursive | |
3441 | * thread-death.) The rest of the cleanup is done in | |
3442 | * userspace. | |
3443 | */ | |
3444 | mval = (uval & FUTEX_WAITERS) | FUTEX_OWNER_DIED; | |
3445 | ||
3446 | /* | |
3447 | * We are not holding a lock here, but we want to have | |
3448 | * the pagefault_disable/enable() protection because | |
3449 | * we want to handle the fault gracefully. If the | |
3450 | * access fails we try to fault in the futex with R/W | |
3451 | * verification via get_user_pages. get_user() above | |
3452 | * does not guarantee R/W access. If that fails we | |
3453 | * give up and leave the futex locked. | |
3454 | */ | |
3455 | if ((err = cmpxchg_futex_value_locked(&nval, uaddr, uval, mval))) { | |
3456 | switch (err) { | |
3457 | case -EFAULT: | |
6e0aa9f8 TG |
3458 | if (fault_in_user_writeable(uaddr)) |
3459 | return -1; | |
3460 | goto retry; | |
6b4f4bc9 WD |
3461 | |
3462 | case -EAGAIN: | |
3463 | cond_resched(); | |
8f17d3a5 | 3464 | goto retry; |
0771dfef | 3465 | |
6b4f4bc9 WD |
3466 | default: |
3467 | WARN_ON_ONCE(1); | |
3468 | return err; | |
3469 | } | |
0771dfef | 3470 | } |
6b4f4bc9 WD |
3471 | |
3472 | if (nval != uval) | |
3473 | goto retry; | |
3474 | ||
3475 | /* | |
3476 | * Wake robust non-PI futexes here. The wakeup of | |
3477 | * PI futexes happens in exit_pi_state(): | |
3478 | */ | |
3479 | if (!pi && (uval & FUTEX_WAITERS)) | |
3480 | futex_wake(uaddr, 1, 1, FUTEX_BITSET_MATCH_ANY); | |
3481 | ||
0771dfef IM |
3482 | return 0; |
3483 | } | |
3484 | ||
e3f2ddea IM |
3485 | /* |
3486 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3487 | */ | |
3488 | static inline int fetch_robust_entry(struct robust_list __user **entry, | |
ba46df98 | 3489 | struct robust_list __user * __user *head, |
1dcc41bb | 3490 | unsigned int *pi) |
e3f2ddea IM |
3491 | { |
3492 | unsigned long uentry; | |
3493 | ||
ba46df98 | 3494 | if (get_user(uentry, (unsigned long __user *)head)) |
e3f2ddea IM |
3495 | return -EFAULT; |
3496 | ||
ba46df98 | 3497 | *entry = (void __user *)(uentry & ~1UL); |
e3f2ddea IM |
3498 | *pi = uentry & 1; |
3499 | ||
3500 | return 0; | |
3501 | } | |
3502 | ||
0771dfef IM |
3503 | /* |
3504 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3505 | * and mark any locks found there dead, and notify any waiters. | |
3506 | * | |
3507 | * We silently return on any sign of list-walking problem. | |
3508 | */ | |
ba31c1a4 | 3509 | static void exit_robust_list(struct task_struct *curr) |
0771dfef IM |
3510 | { |
3511 | struct robust_list_head __user *head = curr->robust_list; | |
9f96cb1e | 3512 | struct robust_list __user *entry, *next_entry, *pending; |
4c115e95 | 3513 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; |
3f649ab7 | 3514 | unsigned int next_pi; |
0771dfef | 3515 | unsigned long futex_offset; |
9f96cb1e | 3516 | int rc; |
0771dfef | 3517 | |
a0c1e907 TG |
3518 | if (!futex_cmpxchg_enabled) |
3519 | return; | |
3520 | ||
0771dfef IM |
3521 | /* |
3522 | * Fetch the list head (which was registered earlier, via | |
3523 | * sys_set_robust_list()): | |
3524 | */ | |
e3f2ddea | 3525 | if (fetch_robust_entry(&entry, &head->list.next, &pi)) |
0771dfef IM |
3526 | return; |
3527 | /* | |
3528 | * Fetch the relative futex offset: | |
3529 | */ | |
3530 | if (get_user(futex_offset, &head->futex_offset)) | |
3531 | return; | |
3532 | /* | |
3533 | * Fetch any possibly pending lock-add first, and handle it | |
3534 | * if it exists: | |
3535 | */ | |
e3f2ddea | 3536 | if (fetch_robust_entry(&pending, &head->list_op_pending, &pip)) |
0771dfef | 3537 | return; |
e3f2ddea | 3538 | |
9f96cb1e | 3539 | next_entry = NULL; /* avoid warning with gcc */ |
0771dfef | 3540 | while (entry != &head->list) { |
9f96cb1e MS |
3541 | /* |
3542 | * Fetch the next entry in the list before calling | |
3543 | * handle_futex_death: | |
3544 | */ | |
3545 | rc = fetch_robust_entry(&next_entry, &entry->next, &next_pi); | |
0771dfef IM |
3546 | /* |
3547 | * A pending lock might already be on the list, so | |
c87e2837 | 3548 | * don't process it twice: |
0771dfef | 3549 | */ |
ca16d5be | 3550 | if (entry != pending) { |
ba46df98 | 3551 | if (handle_futex_death((void __user *)entry + futex_offset, |
ca16d5be | 3552 | curr, pi, HANDLE_DEATH_LIST)) |
0771dfef | 3553 | return; |
ca16d5be | 3554 | } |
9f96cb1e | 3555 | if (rc) |
0771dfef | 3556 | return; |
9f96cb1e MS |
3557 | entry = next_entry; |
3558 | pi = next_pi; | |
0771dfef IM |
3559 | /* |
3560 | * Avoid excessively long or circular lists: | |
3561 | */ | |
3562 | if (!--limit) | |
3563 | break; | |
3564 | ||
3565 | cond_resched(); | |
3566 | } | |
9f96cb1e | 3567 | |
ca16d5be | 3568 | if (pending) { |
9f96cb1e | 3569 | handle_futex_death((void __user *)pending + futex_offset, |
ca16d5be YT |
3570 | curr, pip, HANDLE_DEATH_PENDING); |
3571 | } | |
0771dfef IM |
3572 | } |
3573 | ||
af8cbda2 | 3574 | static void futex_cleanup(struct task_struct *tsk) |
ba31c1a4 TG |
3575 | { |
3576 | if (unlikely(tsk->robust_list)) { | |
3577 | exit_robust_list(tsk); | |
3578 | tsk->robust_list = NULL; | |
3579 | } | |
3580 | ||
3581 | #ifdef CONFIG_COMPAT | |
3582 | if (unlikely(tsk->compat_robust_list)) { | |
3583 | compat_exit_robust_list(tsk); | |
3584 | tsk->compat_robust_list = NULL; | |
3585 | } | |
3586 | #endif | |
3587 | ||
3588 | if (unlikely(!list_empty(&tsk->pi_state_list))) | |
3589 | exit_pi_state_list(tsk); | |
3590 | } | |
3591 | ||
18f69438 TG |
3592 | /** |
3593 | * futex_exit_recursive - Set the tasks futex state to FUTEX_STATE_DEAD | |
3594 | * @tsk: task to set the state on | |
3595 | * | |
3596 | * Set the futex exit state of the task lockless. The futex waiter code | |
3597 | * observes that state when a task is exiting and loops until the task has | |
3598 | * actually finished the futex cleanup. The worst case for this is that the | |
3599 | * waiter runs through the wait loop until the state becomes visible. | |
3600 | * | |
3601 | * This is called from the recursive fault handling path in do_exit(). | |
3602 | * | |
3603 | * This is best effort. Either the futex exit code has run already or | |
3604 | * not. If the OWNER_DIED bit has been set on the futex then the waiter can | |
3605 | * take it over. If not, the problem is pushed back to user space. If the | |
3606 | * futex exit code did not run yet, then an already queued waiter might | |
3607 | * block forever, but there is nothing which can be done about that. | |
3608 | */ | |
3609 | void futex_exit_recursive(struct task_struct *tsk) | |
3610 | { | |
3f186d97 TG |
3611 | /* If the state is FUTEX_STATE_EXITING then futex_exit_mutex is held */ |
3612 | if (tsk->futex_state == FUTEX_STATE_EXITING) | |
3613 | mutex_unlock(&tsk->futex_exit_mutex); | |
18f69438 TG |
3614 | tsk->futex_state = FUTEX_STATE_DEAD; |
3615 | } | |
3616 | ||
af8cbda2 | 3617 | static void futex_cleanup_begin(struct task_struct *tsk) |
150d7158 | 3618 | { |
3f186d97 TG |
3619 | /* |
3620 | * Prevent various race issues against a concurrent incoming waiter | |
3621 | * including live locks by forcing the waiter to block on | |
3622 | * tsk->futex_exit_mutex when it observes FUTEX_STATE_EXITING in | |
3623 | * attach_to_pi_owner(). | |
3624 | */ | |
3625 | mutex_lock(&tsk->futex_exit_mutex); | |
3626 | ||
18f69438 | 3627 | /* |
4a8e991b TG |
3628 | * Switch the state to FUTEX_STATE_EXITING under tsk->pi_lock. |
3629 | * | |
3630 | * This ensures that all subsequent checks of tsk->futex_state in | |
3631 | * attach_to_pi_owner() must observe FUTEX_STATE_EXITING with | |
3632 | * tsk->pi_lock held. | |
3633 | * | |
3634 | * It guarantees also that a pi_state which was queued right before | |
3635 | * the state change under tsk->pi_lock by a concurrent waiter must | |
3636 | * be observed in exit_pi_state_list(). | |
18f69438 TG |
3637 | */ |
3638 | raw_spin_lock_irq(&tsk->pi_lock); | |
4a8e991b | 3639 | tsk->futex_state = FUTEX_STATE_EXITING; |
18f69438 | 3640 | raw_spin_unlock_irq(&tsk->pi_lock); |
af8cbda2 | 3641 | } |
18f69438 | 3642 | |
af8cbda2 TG |
3643 | static void futex_cleanup_end(struct task_struct *tsk, int state) |
3644 | { | |
3645 | /* | |
3646 | * Lockless store. The only side effect is that an observer might | |
3647 | * take another loop until it becomes visible. | |
3648 | */ | |
3649 | tsk->futex_state = state; | |
3f186d97 TG |
3650 | /* |
3651 | * Drop the exit protection. This unblocks waiters which observed | |
3652 | * FUTEX_STATE_EXITING to reevaluate the state. | |
3653 | */ | |
3654 | mutex_unlock(&tsk->futex_exit_mutex); | |
af8cbda2 | 3655 | } |
18f69438 | 3656 | |
af8cbda2 TG |
3657 | void futex_exec_release(struct task_struct *tsk) |
3658 | { | |
3659 | /* | |
3660 | * The state handling is done for consistency, but in the case of | |
93d0955e | 3661 | * exec() there is no way to prevent further damage as the PID stays |
af8cbda2 TG |
3662 | * the same. But for the unlikely and arguably buggy case that a |
3663 | * futex is held on exec(), this provides at least as much state | |
3664 | * consistency protection which is possible. | |
3665 | */ | |
3666 | futex_cleanup_begin(tsk); | |
3667 | futex_cleanup(tsk); | |
3668 | /* | |
3669 | * Reset the state to FUTEX_STATE_OK. The task is alive and about | |
3670 | * exec a new binary. | |
3671 | */ | |
3672 | futex_cleanup_end(tsk, FUTEX_STATE_OK); | |
3673 | } | |
3674 | ||
3675 | void futex_exit_release(struct task_struct *tsk) | |
3676 | { | |
3677 | futex_cleanup_begin(tsk); | |
3678 | futex_cleanup(tsk); | |
3679 | futex_cleanup_end(tsk, FUTEX_STATE_DEAD); | |
150d7158 TG |
3680 | } |
3681 | ||
c19384b5 | 3682 | long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout, |
e2970f2f | 3683 | u32 __user *uaddr2, u32 val2, u32 val3) |
1da177e4 | 3684 | { |
81b40539 | 3685 | int cmd = op & FUTEX_CMD_MASK; |
b41277dc | 3686 | unsigned int flags = 0; |
34f01cc1 ED |
3687 | |
3688 | if (!(op & FUTEX_PRIVATE_FLAG)) | |
b41277dc | 3689 | flags |= FLAGS_SHARED; |
1da177e4 | 3690 | |
b41277dc DH |
3691 | if (op & FUTEX_CLOCK_REALTIME) { |
3692 | flags |= FLAGS_CLOCKRT; | |
bf22a697 TG |
3693 | if (cmd != FUTEX_WAIT_BITSET && cmd != FUTEX_WAIT_REQUEUE_PI && |
3694 | cmd != FUTEX_LOCK_PI2) | |
b41277dc DH |
3695 | return -ENOSYS; |
3696 | } | |
1da177e4 | 3697 | |
59263b51 TG |
3698 | switch (cmd) { |
3699 | case FUTEX_LOCK_PI: | |
bf22a697 | 3700 | case FUTEX_LOCK_PI2: |
59263b51 TG |
3701 | case FUTEX_UNLOCK_PI: |
3702 | case FUTEX_TRYLOCK_PI: | |
3703 | case FUTEX_WAIT_REQUEUE_PI: | |
3704 | case FUTEX_CMP_REQUEUE_PI: | |
3705 | if (!futex_cmpxchg_enabled) | |
3706 | return -ENOSYS; | |
3707 | } | |
3708 | ||
34f01cc1 | 3709 | switch (cmd) { |
1da177e4 | 3710 | case FUTEX_WAIT: |
cd689985 | 3711 | val3 = FUTEX_BITSET_MATCH_ANY; |
405fa8ac | 3712 | fallthrough; |
cd689985 | 3713 | case FUTEX_WAIT_BITSET: |
81b40539 | 3714 | return futex_wait(uaddr, flags, val, timeout, val3); |
1da177e4 | 3715 | case FUTEX_WAKE: |
cd689985 | 3716 | val3 = FUTEX_BITSET_MATCH_ANY; |
405fa8ac | 3717 | fallthrough; |
cd689985 | 3718 | case FUTEX_WAKE_BITSET: |
81b40539 | 3719 | return futex_wake(uaddr, flags, val, val3); |
1da177e4 | 3720 | case FUTEX_REQUEUE: |
81b40539 | 3721 | return futex_requeue(uaddr, flags, uaddr2, val, val2, NULL, 0); |
1da177e4 | 3722 | case FUTEX_CMP_REQUEUE: |
81b40539 | 3723 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 0); |
4732efbe | 3724 | case FUTEX_WAKE_OP: |
81b40539 | 3725 | return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3); |
c87e2837 | 3726 | case FUTEX_LOCK_PI: |
e112c413 | 3727 | flags |= FLAGS_CLOCKRT; |
bf22a697 TG |
3728 | fallthrough; |
3729 | case FUTEX_LOCK_PI2: | |
996636dd | 3730 | return futex_lock_pi(uaddr, flags, timeout, 0); |
c87e2837 | 3731 | case FUTEX_UNLOCK_PI: |
81b40539 | 3732 | return futex_unlock_pi(uaddr, flags); |
c87e2837 | 3733 | case FUTEX_TRYLOCK_PI: |
996636dd | 3734 | return futex_lock_pi(uaddr, flags, NULL, 1); |
52400ba9 DH |
3735 | case FUTEX_WAIT_REQUEUE_PI: |
3736 | val3 = FUTEX_BITSET_MATCH_ANY; | |
81b40539 TG |
3737 | return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3, |
3738 | uaddr2); | |
52400ba9 | 3739 | case FUTEX_CMP_REQUEUE_PI: |
81b40539 | 3740 | return futex_requeue(uaddr, flags, uaddr2, val, val2, &val3, 1); |
1da177e4 | 3741 | } |
81b40539 | 3742 | return -ENOSYS; |
1da177e4 LT |
3743 | } |
3744 | ||
51cf94d1 TG |
3745 | static __always_inline bool futex_cmd_has_timeout(u32 cmd) |
3746 | { | |
3747 | switch (cmd) { | |
3748 | case FUTEX_WAIT: | |
3749 | case FUTEX_LOCK_PI: | |
bf22a697 | 3750 | case FUTEX_LOCK_PI2: |
51cf94d1 TG |
3751 | case FUTEX_WAIT_BITSET: |
3752 | case FUTEX_WAIT_REQUEUE_PI: | |
3753 | return true; | |
3754 | } | |
3755 | return false; | |
3756 | } | |
3757 | ||
3758 | static __always_inline int | |
3759 | futex_init_timeout(u32 cmd, u32 op, struct timespec64 *ts, ktime_t *t) | |
3760 | { | |
3761 | if (!timespec64_valid(ts)) | |
3762 | return -EINVAL; | |
3763 | ||
3764 | *t = timespec64_to_ktime(*ts); | |
3765 | if (cmd == FUTEX_WAIT) | |
3766 | *t = ktime_add_safe(ktime_get(), *t); | |
3767 | else if (cmd != FUTEX_LOCK_PI && !(op & FUTEX_CLOCK_REALTIME)) | |
3768 | *t = timens_ktime_to_host(CLOCK_MONOTONIC, *t); | |
3769 | return 0; | |
3770 | } | |
1da177e4 | 3771 | |
17da2bd9 | 3772 | SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val, |
1ce53e2c AC |
3773 | const struct __kernel_timespec __user *, utime, |
3774 | u32 __user *, uaddr2, u32, val3) | |
1da177e4 | 3775 | { |
51cf94d1 | 3776 | int ret, cmd = op & FUTEX_CMD_MASK; |
c19384b5 | 3777 | ktime_t t, *tp = NULL; |
51cf94d1 | 3778 | struct timespec64 ts; |
1da177e4 | 3779 | |
51cf94d1 | 3780 | if (utime && futex_cmd_has_timeout(cmd)) { |
ab51fbab DB |
3781 | if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG)))) |
3782 | return -EFAULT; | |
bec2f7cb | 3783 | if (get_timespec64(&ts, utime)) |
1da177e4 | 3784 | return -EFAULT; |
51cf94d1 TG |
3785 | ret = futex_init_timeout(cmd, op, &ts, &t); |
3786 | if (ret) | |
3787 | return ret; | |
c19384b5 | 3788 | tp = &t; |
1da177e4 | 3789 | } |
1da177e4 | 3790 | |
b097d5ed | 3791 | return do_futex(uaddr, op, val, tp, uaddr2, (unsigned long)utime, val3); |
1da177e4 LT |
3792 | } |
3793 | ||
04e7712f AB |
3794 | #ifdef CONFIG_COMPAT |
3795 | /* | |
3796 | * Fetch a robust-list pointer. Bit 0 signals PI futexes: | |
3797 | */ | |
3798 | static inline int | |
3799 | compat_fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry, | |
3800 | compat_uptr_t __user *head, unsigned int *pi) | |
3801 | { | |
3802 | if (get_user(*uentry, head)) | |
3803 | return -EFAULT; | |
3804 | ||
3805 | *entry = compat_ptr((*uentry) & ~1); | |
3806 | *pi = (unsigned int)(*uentry) & 1; | |
3807 | ||
3808 | return 0; | |
3809 | } | |
3810 | ||
3811 | static void __user *futex_uaddr(struct robust_list __user *entry, | |
3812 | compat_long_t futex_offset) | |
3813 | { | |
3814 | compat_uptr_t base = ptr_to_compat(entry); | |
3815 | void __user *uaddr = compat_ptr(base + futex_offset); | |
3816 | ||
3817 | return uaddr; | |
3818 | } | |
3819 | ||
3820 | /* | |
3821 | * Walk curr->robust_list (very carefully, it's a userspace list!) | |
3822 | * and mark any locks found there dead, and notify any waiters. | |
3823 | * | |
3824 | * We silently return on any sign of list-walking problem. | |
3825 | */ | |
ba31c1a4 | 3826 | static void compat_exit_robust_list(struct task_struct *curr) |
04e7712f AB |
3827 | { |
3828 | struct compat_robust_list_head __user *head = curr->compat_robust_list; | |
3829 | struct robust_list __user *entry, *next_entry, *pending; | |
3830 | unsigned int limit = ROBUST_LIST_LIMIT, pi, pip; | |
3f649ab7 | 3831 | unsigned int next_pi; |
04e7712f AB |
3832 | compat_uptr_t uentry, next_uentry, upending; |
3833 | compat_long_t futex_offset; | |
3834 | int rc; | |
3835 | ||
3836 | if (!futex_cmpxchg_enabled) | |
3837 | return; | |
3838 | ||
3839 | /* | |
3840 | * Fetch the list head (which was registered earlier, via | |
3841 | * sys_set_robust_list()): | |
3842 | */ | |
3843 | if (compat_fetch_robust_entry(&uentry, &entry, &head->list.next, &pi)) | |
3844 | return; | |
3845 | /* | |
3846 | * Fetch the relative futex offset: | |
3847 | */ | |
3848 | if (get_user(futex_offset, &head->futex_offset)) | |
3849 | return; | |
3850 | /* | |
3851 | * Fetch any possibly pending lock-add first, and handle it | |
3852 | * if it exists: | |
3853 | */ | |
3854 | if (compat_fetch_robust_entry(&upending, &pending, | |
3855 | &head->list_op_pending, &pip)) | |
3856 | return; | |
3857 | ||
3858 | next_entry = NULL; /* avoid warning with gcc */ | |
3859 | while (entry != (struct robust_list __user *) &head->list) { | |
3860 | /* | |
3861 | * Fetch the next entry in the list before calling | |
3862 | * handle_futex_death: | |
3863 | */ | |
3864 | rc = compat_fetch_robust_entry(&next_uentry, &next_entry, | |
3865 | (compat_uptr_t __user *)&entry->next, &next_pi); | |
3866 | /* | |
3867 | * A pending lock might already be on the list, so | |
3868 | * dont process it twice: | |
3869 | */ | |
3870 | if (entry != pending) { | |
3871 | void __user *uaddr = futex_uaddr(entry, futex_offset); | |
3872 | ||
ca16d5be YT |
3873 | if (handle_futex_death(uaddr, curr, pi, |
3874 | HANDLE_DEATH_LIST)) | |
04e7712f AB |
3875 | return; |
3876 | } | |
3877 | if (rc) | |
3878 | return; | |
3879 | uentry = next_uentry; | |
3880 | entry = next_entry; | |
3881 | pi = next_pi; | |
3882 | /* | |
3883 | * Avoid excessively long or circular lists: | |
3884 | */ | |
3885 | if (!--limit) | |
3886 | break; | |
3887 | ||
3888 | cond_resched(); | |
3889 | } | |
3890 | if (pending) { | |
3891 | void __user *uaddr = futex_uaddr(pending, futex_offset); | |
3892 | ||
ca16d5be | 3893 | handle_futex_death(uaddr, curr, pip, HANDLE_DEATH_PENDING); |
04e7712f AB |
3894 | } |
3895 | } | |
3896 | ||
3897 | COMPAT_SYSCALL_DEFINE2(set_robust_list, | |
3898 | struct compat_robust_list_head __user *, head, | |
3899 | compat_size_t, len) | |
3900 | { | |
3901 | if (!futex_cmpxchg_enabled) | |
3902 | return -ENOSYS; | |
3903 | ||
3904 | if (unlikely(len != sizeof(*head))) | |
3905 | return -EINVAL; | |
3906 | ||
3907 | current->compat_robust_list = head; | |
3908 | ||
3909 | return 0; | |
3910 | } | |
3911 | ||
3912 | COMPAT_SYSCALL_DEFINE3(get_robust_list, int, pid, | |
3913 | compat_uptr_t __user *, head_ptr, | |
3914 | compat_size_t __user *, len_ptr) | |
3915 | { | |
3916 | struct compat_robust_list_head __user *head; | |
3917 | unsigned long ret; | |
3918 | struct task_struct *p; | |
3919 | ||
3920 | if (!futex_cmpxchg_enabled) | |
3921 | return -ENOSYS; | |
3922 | ||
3923 | rcu_read_lock(); | |
3924 | ||
3925 | ret = -ESRCH; | |
3926 | if (!pid) | |
3927 | p = current; | |
3928 | else { | |
3929 | p = find_task_by_vpid(pid); | |
3930 | if (!p) | |
3931 | goto err_unlock; | |
3932 | } | |
3933 | ||
3934 | ret = -EPERM; | |
3935 | if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS)) | |
3936 | goto err_unlock; | |
3937 | ||
3938 | head = p->compat_robust_list; | |
3939 | rcu_read_unlock(); | |
3940 | ||
3941 | if (put_user(sizeof(*head), len_ptr)) | |
3942 | return -EFAULT; | |
3943 | return put_user(ptr_to_compat(head), head_ptr); | |
3944 | ||
3945 | err_unlock: | |
3946 | rcu_read_unlock(); | |
3947 | ||
3948 | return ret; | |
3949 | } | |
bec2f7cb | 3950 | #endif /* CONFIG_COMPAT */ |
04e7712f | 3951 | |
bec2f7cb | 3952 | #ifdef CONFIG_COMPAT_32BIT_TIME |
8dabe724 | 3953 | SYSCALL_DEFINE6(futex_time32, u32 __user *, uaddr, int, op, u32, val, |
1ce53e2c | 3954 | const struct old_timespec32 __user *, utime, u32 __user *, uaddr2, |
04e7712f AB |
3955 | u32, val3) |
3956 | { | |
51cf94d1 | 3957 | int ret, cmd = op & FUTEX_CMD_MASK; |
04e7712f | 3958 | ktime_t t, *tp = NULL; |
51cf94d1 | 3959 | struct timespec64 ts; |
04e7712f | 3960 | |
51cf94d1 | 3961 | if (utime && futex_cmd_has_timeout(cmd)) { |
bec2f7cb | 3962 | if (get_old_timespec32(&ts, utime)) |
04e7712f | 3963 | return -EFAULT; |
51cf94d1 TG |
3964 | ret = futex_init_timeout(cmd, op, &ts, &t); |
3965 | if (ret) | |
3966 | return ret; | |
04e7712f AB |
3967 | tp = &t; |
3968 | } | |
04e7712f | 3969 | |
b097d5ed | 3970 | return do_futex(uaddr, op, val, tp, uaddr2, (unsigned long)utime, val3); |
04e7712f | 3971 | } |
bec2f7cb | 3972 | #endif /* CONFIG_COMPAT_32BIT_TIME */ |
04e7712f | 3973 | |
03b8c7b6 | 3974 | static void __init futex_detect_cmpxchg(void) |
1da177e4 | 3975 | { |
03b8c7b6 | 3976 | #ifndef CONFIG_HAVE_FUTEX_CMPXCHG |
a0c1e907 | 3977 | u32 curval; |
03b8c7b6 HC |
3978 | |
3979 | /* | |
3980 | * This will fail and we want it. Some arch implementations do | |
3981 | * runtime detection of the futex_atomic_cmpxchg_inatomic() | |
3982 | * functionality. We want to know that before we call in any | |
3983 | * of the complex code paths. Also we want to prevent | |
3984 | * registration of robust lists in that case. NULL is | |
3985 | * guaranteed to fault and we get -EFAULT on functional | |
3986 | * implementation, the non-functional ones will return | |
3987 | * -ENOSYS. | |
3988 | */ | |
3989 | if (cmpxchg_futex_value_locked(&curval, NULL, 0, 0) == -EFAULT) | |
3990 | futex_cmpxchg_enabled = 1; | |
3991 | #endif | |
3992 | } | |
3993 | ||
3994 | static int __init futex_init(void) | |
3995 | { | |
63b1a816 | 3996 | unsigned int futex_shift; |
a52b89eb DB |
3997 | unsigned long i; |
3998 | ||
3999 | #if CONFIG_BASE_SMALL | |
4000 | futex_hashsize = 16; | |
4001 | #else | |
4002 | futex_hashsize = roundup_pow_of_two(256 * num_possible_cpus()); | |
4003 | #endif | |
4004 | ||
4005 | futex_queues = alloc_large_system_hash("futex", sizeof(*futex_queues), | |
4006 | futex_hashsize, 0, | |
4007 | futex_hashsize < 256 ? HASH_SMALL : 0, | |
63b1a816 HC |
4008 | &futex_shift, NULL, |
4009 | futex_hashsize, futex_hashsize); | |
4010 | futex_hashsize = 1UL << futex_shift; | |
03b8c7b6 HC |
4011 | |
4012 | futex_detect_cmpxchg(); | |
a0c1e907 | 4013 | |
a52b89eb | 4014 | for (i = 0; i < futex_hashsize; i++) { |
11d4616b | 4015 | atomic_set(&futex_queues[i].waiters, 0); |
732375c6 | 4016 | plist_head_init(&futex_queues[i].chain); |
3e4ab747 TG |
4017 | spin_lock_init(&futex_queues[i].lock); |
4018 | } | |
4019 | ||
1da177e4 LT |
4020 | return 0; |
4021 | } | |
25f71d1c | 4022 | core_initcall(futex_init); |