projects/linux·blob master · refs

526 lines · 13.5 KiB

1// SPDX-License-Identifier: GPL-2.0-or-later
2
3#include <linux/syscalls.h>
4#include <linux/time_namespace.h>
5
6#include "futex.h"
7
8/*
9 * Support for robust futexes: the kernel cleans up held futexes at
10 * thread exit time.
11 *
12 * Implementation: user-space maintains a per-thread list of locks it
13 * is holding. Upon do_exit(), the kernel carefully walks this list,
14 * and marks all locks that are owned by this thread with the
15 * FUTEX_OWNER_DIED bit, and wakes up a waiter (if any). The list is
16 * always manipulated with the lock held, so the list is private and
17 * per-thread. Userspace also maintains a per-thread 'list_op_pending'
18 * field, to allow the kernel to clean up if the thread dies after
19 * acquiring the lock, but just before it could have added itself to
20 * the list. There can only be one such pending lock.
21 */
22
23/**
24 * sys_set_robust_list() - Set the robust-futex list head of a task
25 * @head: pointer to the list-head
26 * @len: length of the list-head, as userspace expects
27 */
28SYSCALL_DEFINE2(set_robust_list, struct robust_list_head __user *, head, size_t, len)
29{
30 /* The kernel knows only one size for now. */
31 if (unlikely(len != sizeof(*head)))
32 return -EINVAL;
33
34 current->futex.robust_list = head;
35 return 0;
36}
37
38static inline void __user *futex_task_robust_list(struct task_struct *p, bool compat)
39{
40#ifdef CONFIG_COMPAT
41 if (compat)
42 return p->futex.compat_robust_list;
43#endif
44 return p->futex.robust_list;
45}
46
47static void __user *futex_get_robust_list_common(int pid, bool compat)
48{
49 struct task_struct *p = current;
50 void __user *head;
51 int ret;
52
53 scoped_guard(rcu) {
54 if (pid) {
55 p = find_task_by_vpid(pid);
56 if (!p)
57 return (void __user *)ERR_PTR(-ESRCH);
58 }
59 get_task_struct(p);
60 }
61
62 /*
63 * Hold exec_update_lock to serialize with concurrent exec()
64 * so ptrace_may_access() is checked against stable credentials
65 */
66 ret = down_read_killable(&p->signal->exec_update_lock);
67 if (ret)
68 goto err_put;
69
70 ret = -EPERM;
71 if (!ptrace_may_access(p, PTRACE_MODE_READ_REALCREDS))
72 goto err_unlock;
73
74 head = futex_task_robust_list(p, compat);
75
76 up_read(&p->signal->exec_update_lock);
77 put_task_struct(p);
78
79 return head;
80
81err_unlock:
82 up_read(&p->signal->exec_update_lock);
83err_put:
84 put_task_struct(p);
85 return (void __user *)ERR_PTR(ret);
86}
87
88/**
89 * sys_get_robust_list() - Get the robust-futex list head of a task
90 * @pid: pid of the process [zero for current task]
91 * @head_ptr: pointer to a list-head pointer, the kernel fills it in
92 * @len_ptr: pointer to a length field, the kernel fills in the header size
93 */
94SYSCALL_DEFINE3(get_robust_list, int, pid,
95 struct robust_list_head __user * __user *, head_ptr,
96 size_t __user *, len_ptr)
97{
98 struct robust_list_head __user *head = futex_get_robust_list_common(pid, false);
99
100 if (IS_ERR(head))
101 return PTR_ERR(head);
102
103 if (put_user(sizeof(*head), len_ptr))
104 return -EFAULT;
105 return put_user(head, head_ptr);
106}
107
108long do_futex(u32 __user *uaddr, int op, u32 val, ktime_t *timeout,
109 u32 __user *uaddr2, u32 val2, u32 val3)
110{
111 unsigned int flags = futex_to_flags(op);
112 int cmd = op & FUTEX_CMD_MASK;
113
114 if (flags & FLAGS_CLOCKRT) {
115 if (cmd != FUTEX_WAIT_BITSET &&
116 cmd != FUTEX_WAIT_REQUEUE_PI &&
117 cmd != FUTEX_LOCK_PI2)
118 return -ENOSYS;
119 }
120
121 if (flags & FLAGS_ROBUST_UNLOCK) {
122 if (cmd != FUTEX_WAKE &&
123 cmd != FUTEX_WAKE_BITSET &&
124 cmd != FUTEX_UNLOCK_PI)
125 return -ENOSYS;
126 }
127
128 switch (cmd) {
129 case FUTEX_WAIT:
130 val3 = FUTEX_BITSET_MATCH_ANY;
131 fallthrough;
132 case FUTEX_WAIT_BITSET:
133 return futex_wait(uaddr, flags, val, timeout, val3);
134 case FUTEX_WAKE:
135 val3 = FUTEX_BITSET_MATCH_ANY;
136 fallthrough;
137 case FUTEX_WAKE_BITSET:
138 return futex_wake(uaddr, flags, uaddr2, val, val3);
139 case FUTEX_REQUEUE:
140 return futex_requeue(uaddr, flags, uaddr2, flags, val, val2, NULL, 0);
141 case FUTEX_CMP_REQUEUE:
142 return futex_requeue(uaddr, flags, uaddr2, flags, val, val2, &val3, 0);
143 case FUTEX_WAKE_OP:
144 return futex_wake_op(uaddr, flags, uaddr2, val, val2, val3);
145 case FUTEX_LOCK_PI:
146 flags |= FLAGS_CLOCKRT;
147 fallthrough;
148 case FUTEX_LOCK_PI2:
149 return futex_lock_pi(uaddr, flags, timeout, 0);
150 case FUTEX_UNLOCK_PI:
151 return futex_unlock_pi(uaddr, flags, uaddr2);
152 case FUTEX_TRYLOCK_PI:
153 return futex_lock_pi(uaddr, flags, NULL, 1);
154 case FUTEX_WAIT_REQUEUE_PI:
155 val3 = FUTEX_BITSET_MATCH_ANY;
156 return futex_wait_requeue_pi(uaddr, flags, val, timeout, val3,
157 uaddr2);
158 case FUTEX_CMP_REQUEUE_PI:
159 return futex_requeue(uaddr, flags, uaddr2, flags, val, val2, &val3, 1);
160 }
161 return -ENOSYS;
162}
163
164static __always_inline bool futex_cmd_has_timeout(u32 cmd)
165{
166 switch (cmd) {
167 case FUTEX_WAIT:
168 case FUTEX_LOCK_PI:
169 case FUTEX_LOCK_PI2:
170 case FUTEX_WAIT_BITSET:
171 case FUTEX_WAIT_REQUEUE_PI:
172 return true;
173 }
174 return false;
175}
176
177static __always_inline int
178futex_init_timeout(u32 cmd, u32 op, struct timespec64 *ts, ktime_t *t)
179{
180 if (!timespec64_valid(ts))
181 return -EINVAL;
182
183 *t = timespec64_to_ktime(*ts);
184 if (cmd == FUTEX_WAIT)
185 *t = ktime_add_safe(ktime_get(), *t);
186 else if (cmd != FUTEX_LOCK_PI && !(op & FUTEX_CLOCK_REALTIME))
187 *t = timens_ktime_to_host(CLOCK_MONOTONIC, *t);
188 return 0;
189}
190
191SYSCALL_DEFINE6(futex, u32 __user *, uaddr, int, op, u32, val,
192 const struct __kernel_timespec __user *, utime,
193 u32 __user *, uaddr2, u32, val3)
194{
195 int ret, cmd = op & FUTEX_CMD_MASK;
196 ktime_t t, *tp = NULL;
197 struct timespec64 ts;
198
199 if (utime && futex_cmd_has_timeout(cmd)) {
200 if (unlikely(should_fail_futex(!(op & FUTEX_PRIVATE_FLAG))))
201 return -EFAULT;
202 if (get_timespec64(&ts, utime))
203 return -EFAULT;
204 ret = futex_init_timeout(cmd, op, &ts, &t);
205 if (ret)
206 return ret;
207 tp = &t;
208 }
209
210 return do_futex(uaddr, op, val, tp, uaddr2, (unsigned long)utime, val3);
211}
212
213/**
214 * futex_parse_waitv - Parse a waitv array from userspace
215 * @futexv: Kernel side list of waiters to be filled
216 * @uwaitv: Userspace list to be parsed
217 * @nr_futexes: Length of futexv
218 * @wake: Wake to call when futex is woken
219 * @wake_data: Data for the wake handler
220 *
221 * Return: Error code on failure, 0 on success
222 */
223int futex_parse_waitv(struct futex_vector *futexv,
224 struct futex_waitv __user *uwaitv,
225 unsigned int nr_futexes, futex_wake_fn *wake,
226 void *wake_data)
227{
228 struct futex_waitv aux;
229 unsigned int i;
230
231 for (i = 0; i < nr_futexes; i++) {
232 unsigned int flags;
233
234 if (copy_from_user(&aux, &uwaitv[i], sizeof(aux)))
235 return -EFAULT;
236
237 if ((aux.flags & ~FUTEX2_VALID_MASK) || aux.__reserved)
238 return -EINVAL;
239
240 flags = futex2_to_flags(aux.flags);
241 if (!futex_flags_valid(flags))
242 return -EINVAL;
243
244 if (!futex_validate_input(flags, aux.val))
245 return -EINVAL;
246
247 futexv[i].w.flags = flags;
248 futexv[i].w.val = aux.val;
249 futexv[i].w.uaddr = aux.uaddr;
250 futexv[i].q = futex_q_init;
251 futexv[i].q.wake = wake;
252 futexv[i].q.wake_data = wake_data;
253 }
254
255 return 0;
256}
257
258static int futex2_setup_timeout(struct __kernel_timespec __user *timeout,
259 clockid_t clockid, struct hrtimer_sleeper *to)
260{
261 int flag_clkid = 0, flag_init = 0;
262 struct timespec64 ts;
263 ktime_t time;
264 int ret;
265
266 if (!timeout)
267 return 0;
268
269 if (clockid == CLOCK_REALTIME) {
270 flag_clkid = FLAGS_CLOCKRT;
271 flag_init = FUTEX_CLOCK_REALTIME;
272 }
273
274 if (clockid != CLOCK_REALTIME && clockid != CLOCK_MONOTONIC)
275 return -EINVAL;
276
277 if (get_timespec64(&ts, timeout))
278 return -EFAULT;
279
280 /*
281 * Since there's no opcode for futex_waitv, use
282 * FUTEX_WAIT_BITSET that uses absolute timeout as well
283 */
284 ret = futex_init_timeout(FUTEX_WAIT_BITSET, flag_init, &ts, &time);
285 if (ret)
286 return ret;
287
288 futex_setup_timer(&time, to, flag_clkid, 0);
289 return 0;
290}
291
292static inline void futex2_destroy_timeout(struct hrtimer_sleeper *to)
293{
294 hrtimer_cancel(&to->timer);
295 destroy_hrtimer_on_stack(&to->timer);
296}
297
298/**
299 * sys_futex_waitv - Wait on a list of futexes
300 * @waiters: List of futexes to wait on
301 * @nr_futexes: Length of futexv
302 * @flags: Flag for timeout (monotonic/realtime)
303 * @timeout: Optional absolute timeout.
304 * @clockid: Clock to be used for the timeout, realtime or monotonic.
305 *
306 * Given an array of `struct futex_waitv`, wait on each uaddr. The thread wakes
307 * if a futex_wake() is performed at any uaddr. The syscall returns immediately
308 * if any waiter has *uaddr != val. *timeout is an optional timeout value for
309 * the operation. Each waiter has individual flags. The `flags` argument for
310 * the syscall should be used solely for specifying the timeout as realtime, if
311 * needed. Flags for private futexes, sizes, etc. should be used on the
312 * individual flags of each waiter.
313 *
314 * Returns the array index of one of the woken futexes. No further information
315 * is provided: any number of other futexes may also have been woken by the
316 * same event, and if more than one futex was woken, the retrned index may
317 * refer to any one of them. (It is not necessaryily the futex with the
318 * smallest index, nor the one most recently woken, nor...)
319 */
320
321SYSCALL_DEFINE5(futex_waitv, struct futex_waitv __user *, waiters,
322 unsigned int, nr_futexes, unsigned int, flags,
323 struct __kernel_timespec __user *, timeout, clockid_t, clockid)
324{
325 struct hrtimer_sleeper to;
326 struct futex_vector *futexv;
327 int ret;
328
329 /* This syscall supports no flags for now */
330 if (flags)
331 return -EINVAL;
332
333 if (!nr_futexes || nr_futexes > FUTEX_WAITV_MAX || !waiters)
334 return -EINVAL;
335
336 if (timeout && (ret = futex2_setup_timeout(timeout, clockid, &to)))
337 return ret;
338
339 futexv = kzalloc_objs(*futexv, nr_futexes);
340 if (!futexv) {
341 ret = -ENOMEM;
342 goto destroy_timer;
343 }
344
345 ret = futex_parse_waitv(futexv, waiters, nr_futexes, futex_wake_mark,
346 NULL);
347 if (!ret)
348 ret = futex_wait_multiple(futexv, nr_futexes, timeout ? &to : NULL);
349
350 kfree(futexv);
351
352destroy_timer:
353 if (timeout)
354 futex2_destroy_timeout(&to);
355 return ret;
356}
357
358/*
359 * sys_futex_wake - Wake a number of futexes
360 * @uaddr: Address of the futex(es) to wake
361 * @mask: bitmask
362 * @nr: Number of the futexes to wake
363 * @flags: FUTEX2 flags
364 *
365 * Identical to the traditional FUTEX_WAKE_BITSET op, except it is part of the
366 * futex2 family of calls.
367 */
368
369SYSCALL_DEFINE4(futex_wake,
370 void __user *, uaddr,
371 unsigned long, mask,
372 int, nr,
373 unsigned int, flags)
374{
375 if (flags & ~FUTEX2_VALID_MASK)
376 return -EINVAL;
377
378 flags = futex2_to_flags(flags);
379 if (!futex_flags_valid(flags))
380 return -EINVAL;
381
382 if (!futex_validate_input(flags, mask))
383 return -EINVAL;
384
385 return futex_wake(uaddr, FLAGS_STRICT | flags, NULL, nr, mask);
386}
387
388/*
389 * sys_futex_wait - Wait on a futex
390 * @uaddr: Address of the futex to wait on
391 * @val: Value of @uaddr
392 * @mask: bitmask
393 * @flags: FUTEX2 flags
394 * @timeout: Optional absolute timeout
395 * @clockid: Clock to be used for the timeout, realtime or monotonic
396 *
397 * Identical to the traditional FUTEX_WAIT_BITSET op, except it is part of the
398 * futex2 familiy of calls.
399 */
400
401SYSCALL_DEFINE6(futex_wait,
402 void __user *, uaddr,
403 unsigned long, val,
404 unsigned long, mask,
405 unsigned int, flags,
406 struct __kernel_timespec __user *, timeout,
407 clockid_t, clockid)
408{
409 struct hrtimer_sleeper to;
410 int ret;
411
412 if (flags & ~FUTEX2_VALID_MASK)
413 return -EINVAL;
414
415 flags = futex2_to_flags(flags);
416 if (!futex_flags_valid(flags))
417 return -EINVAL;
418
419 if (!futex_validate_input(flags, val) ||
420 !futex_validate_input(flags, mask))
421 return -EINVAL;
422
423 if (timeout && (ret = futex2_setup_timeout(timeout, clockid, &to)))
424 return ret;
425
426 ret = __futex_wait(uaddr, flags, val, timeout ? &to : NULL, mask);
427
428 if (timeout)
429 futex2_destroy_timeout(&to);
430
431 return ret;
432}
433
434/*
435 * sys_futex_requeue - Requeue a waiter from one futex to another
436 * @waiters: array describing the source and destination futex
437 * @flags: unused
438 * @nr_wake: number of futexes to wake
439 * @nr_requeue: number of futexes to requeue
440 *
441 * Identical to the traditional FUTEX_CMP_REQUEUE op, except it is part of the
442 * futex2 family of calls.
443 */
444
445SYSCALL_DEFINE4(futex_requeue,
446 struct futex_waitv __user *, waiters,
447 unsigned int, flags,
448 int, nr_wake,
449 int, nr_requeue)
450{
451 struct futex_vector futexes[2];
452 u32 cmpval;
453 int ret;
454
455 if (flags)
456 return -EINVAL;
457
458 if (!waiters)
459 return -EINVAL;
460
461 ret = futex_parse_waitv(futexes, waiters, 2, futex_wake_mark, NULL);
462 if (ret)
463 return ret;
464
465 /*
466 * For now mandate both flags are identical, like the sys_futex()
467 * interface has. If/when we merge the variable sized futex support,
468 * that patch can modify this test to allow a difference in size.
469 */
470 if (futexes[0].w.flags != futexes[1].w.flags)
471 return -EINVAL;
472
473 cmpval = futexes[0].w.val;
474
475 return futex_requeue(u64_to_user_ptr(futexes[0].w.uaddr), futexes[0].w.flags,
476 u64_to_user_ptr(futexes[1].w.uaddr), futexes[1].w.flags,
477 nr_wake, nr_requeue, &cmpval, 0);
478}
479
480#ifdef CONFIG_COMPAT
481COMPAT_SYSCALL_DEFINE2(set_robust_list, struct compat_robust_list_head __user *, head,
482 compat_size_t, len)
483{
484 if (unlikely(len != sizeof(*head)))
485 return -EINVAL;
486
487 current->futex.compat_robust_list = head;
488 return 0;
489}
490
491COMPAT_SYSCALL_DEFINE3(get_robust_list, int, pid,
492 compat_uptr_t __user *, head_ptr,
493 compat_size_t __user *, len_ptr)
494{
495 struct compat_robust_list_head __user *head = futex_get_robust_list_common(pid, true);
496
497 if (IS_ERR(head))
498 return PTR_ERR(head);
499
500 if (put_user(sizeof(*head), len_ptr))
501 return -EFAULT;
502 return put_user(ptr_to_compat(head), head_ptr);
503}
504#endif /* CONFIG_COMPAT */
505
506#ifdef CONFIG_COMPAT_32BIT_TIME
507SYSCALL_DEFINE6(futex_time32, u32 __user *, uaddr, int, op, u32, val,
508 const struct old_timespec32 __user *, utime, u32 __user *, uaddr2,
509 u32, val3)
510{
511 int ret, cmd = op & FUTEX_CMD_MASK;
512 ktime_t t, *tp = NULL;
513 struct timespec64 ts;
514
515 if (utime && futex_cmd_has_timeout(cmd)) {
516 if (get_old_timespec32(&ts, utime))
517 return -EFAULT;
518 ret = futex_init_timeout(cmd, op, &ts, &t);
519 if (ret)
520 return ret;
521 tp = &t;
522 }
523
524 return do_futex(uaddr, op, val, tp, uaddr2, (unsigned long)utime, val3);
525}
526#endif /* CONFIG_COMPAT_32BIT_TIME */