Loading...
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
23 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
74 */
75
76#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
81#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
84#include <linux/capability.h>
85#include <linux/seq_file.h>
86#include <linux/rwsem.h>
87#include <linux/nsproxy.h>
88#include <linux/ipc_namespace.h>
89
90#include <asm/uaccess.h>
91#include "util.h"
92
93#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
94
95#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
96#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
97
98static int newary(struct ipc_namespace *, struct ipc_params *);
99static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
100#ifdef CONFIG_PROC_FS
101static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
102#endif
103
104#define SEMMSL_FAST 256 /* 512 bytes on stack */
105#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
106
107/*
108 * linked list protection:
109 * sem_undo.id_next,
110 * sem_array.sem_pending{,last},
111 * sem_array.sem_undo: sem_lock() for read/write
112 * sem_undo.proc_next: only "current" is allowed to read/write that field.
113 *
114 */
115
116#define sc_semmsl sem_ctls[0]
117#define sc_semmns sem_ctls[1]
118#define sc_semopm sem_ctls[2]
119#define sc_semmni sem_ctls[3]
120
121void sem_init_ns(struct ipc_namespace *ns)
122{
123 ns->sc_semmsl = SEMMSL;
124 ns->sc_semmns = SEMMNS;
125 ns->sc_semopm = SEMOPM;
126 ns->sc_semmni = SEMMNI;
127 ns->used_sems = 0;
128 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
129}
130
131#ifdef CONFIG_IPC_NS
132void sem_exit_ns(struct ipc_namespace *ns)
133{
134 free_ipcs(ns, &sem_ids(ns), freeary);
135 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
136}
137#endif
138
139void __init sem_init (void)
140{
141 sem_init_ns(&init_ipc_ns);
142 ipc_init_proc_interface("sysvipc/sem",
143 " key semid perms nsems uid gid cuid cgid otime ctime\n",
144 IPC_SEM_IDS, sysvipc_sem_proc_show);
145}
146
147/*
148 * sem_lock_(check_) routines are called in the paths where the rw_mutex
149 * is not held.
150 */
151static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
152{
153 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
154
155 if (IS_ERR(ipcp))
156 return (struct sem_array *)ipcp;
157
158 return container_of(ipcp, struct sem_array, sem_perm);
159}
160
161static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
162 int id)
163{
164 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
165
166 if (IS_ERR(ipcp))
167 return (struct sem_array *)ipcp;
168
169 return container_of(ipcp, struct sem_array, sem_perm);
170}
171
172static inline void sem_lock_and_putref(struct sem_array *sma)
173{
174 ipc_lock_by_ptr(&sma->sem_perm);
175 ipc_rcu_putref(sma);
176}
177
178static inline void sem_getref_and_unlock(struct sem_array *sma)
179{
180 ipc_rcu_getref(sma);
181 ipc_unlock(&(sma)->sem_perm);
182}
183
184static inline void sem_putref(struct sem_array *sma)
185{
186 ipc_lock_by_ptr(&sma->sem_perm);
187 ipc_rcu_putref(sma);
188 ipc_unlock(&(sma)->sem_perm);
189}
190
191static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
192{
193 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194}
195
196/*
197 * Lockless wakeup algorithm:
198 * Without the check/retry algorithm a lockless wakeup is possible:
199 * - queue.status is initialized to -EINTR before blocking.
200 * - wakeup is performed by
201 * * unlinking the queue entry from sma->sem_pending
202 * * setting queue.status to IN_WAKEUP
203 * This is the notification for the blocked thread that a
204 * result value is imminent.
205 * * call wake_up_process
206 * * set queue.status to the final value.
207 * - the previously blocked thread checks queue.status:
208 * * if it's IN_WAKEUP, then it must wait until the value changes
209 * * if it's not -EINTR, then the operation was completed by
210 * update_queue. semtimedop can return queue.status without
211 * performing any operation on the sem array.
212 * * otherwise it must acquire the spinlock and check what's up.
213 *
214 * The two-stage algorithm is necessary to protect against the following
215 * races:
216 * - if queue.status is set after wake_up_process, then the woken up idle
217 * thread could race forward and try (and fail) to acquire sma->lock
218 * before update_queue had a chance to set queue.status
219 * - if queue.status is written before wake_up_process and if the
220 * blocked process is woken up by a signal between writing
221 * queue.status and the wake_up_process, then the woken up
222 * process could return from semtimedop and die by calling
223 * sys_exit before wake_up_process is called. Then wake_up_process
224 * will oops, because the task structure is already invalid.
225 * (yes, this happened on s390 with sysv msg).
226 *
227 */
228#define IN_WAKEUP 1
229
230/**
231 * newary - Create a new semaphore set
232 * @ns: namespace
233 * @params: ptr to the structure that contains key, semflg and nsems
234 *
235 * Called with sem_ids.rw_mutex held (as a writer)
236 */
237
238static int newary(struct ipc_namespace *ns, struct ipc_params *params)
239{
240 int id;
241 int retval;
242 struct sem_array *sma;
243 int size;
244 key_t key = params->key;
245 int nsems = params->u.nsems;
246 int semflg = params->flg;
247 int i;
248
249 if (!nsems)
250 return -EINVAL;
251 if (ns->used_sems + nsems > ns->sc_semmns)
252 return -ENOSPC;
253
254 size = sizeof (*sma) + nsems * sizeof (struct sem);
255 sma = ipc_rcu_alloc(size);
256 if (!sma) {
257 return -ENOMEM;
258 }
259 memset (sma, 0, size);
260
261 sma->sem_perm.mode = (semflg & S_IRWXUGO);
262 sma->sem_perm.key = key;
263
264 sma->sem_perm.security = NULL;
265 retval = security_sem_alloc(sma);
266 if (retval) {
267 ipc_rcu_putref(sma);
268 return retval;
269 }
270
271 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
272 if (id < 0) {
273 security_sem_free(sma);
274 ipc_rcu_putref(sma);
275 return id;
276 }
277 ns->used_sems += nsems;
278
279 sma->sem_base = (struct sem *) &sma[1];
280
281 for (i = 0; i < nsems; i++)
282 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
283
284 sma->complex_count = 0;
285 INIT_LIST_HEAD(&sma->sem_pending);
286 INIT_LIST_HEAD(&sma->list_id);
287 sma->sem_nsems = nsems;
288 sma->sem_ctime = get_seconds();
289 sem_unlock(sma);
290
291 return sma->sem_perm.id;
292}
293
294
295/*
296 * Called with sem_ids.rw_mutex and ipcp locked.
297 */
298static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
299{
300 struct sem_array *sma;
301
302 sma = container_of(ipcp, struct sem_array, sem_perm);
303 return security_sem_associate(sma, semflg);
304}
305
306/*
307 * Called with sem_ids.rw_mutex and ipcp locked.
308 */
309static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
310 struct ipc_params *params)
311{
312 struct sem_array *sma;
313
314 sma = container_of(ipcp, struct sem_array, sem_perm);
315 if (params->u.nsems > sma->sem_nsems)
316 return -EINVAL;
317
318 return 0;
319}
320
321SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
322{
323 struct ipc_namespace *ns;
324 struct ipc_ops sem_ops;
325 struct ipc_params sem_params;
326
327 ns = current->nsproxy->ipc_ns;
328
329 if (nsems < 0 || nsems > ns->sc_semmsl)
330 return -EINVAL;
331
332 sem_ops.getnew = newary;
333 sem_ops.associate = sem_security;
334 sem_ops.more_checks = sem_more_checks;
335
336 sem_params.key = key;
337 sem_params.flg = semflg;
338 sem_params.u.nsems = nsems;
339
340 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
341}
342
343/*
344 * Determine whether a sequence of semaphore operations would succeed
345 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
346 */
347
348static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
349 int nsops, struct sem_undo *un, int pid)
350{
351 int result, sem_op;
352 struct sembuf *sop;
353 struct sem * curr;
354
355 for (sop = sops; sop < sops + nsops; sop++) {
356 curr = sma->sem_base + sop->sem_num;
357 sem_op = sop->sem_op;
358 result = curr->semval;
359
360 if (!sem_op && result)
361 goto would_block;
362
363 result += sem_op;
364 if (result < 0)
365 goto would_block;
366 if (result > SEMVMX)
367 goto out_of_range;
368 if (sop->sem_flg & SEM_UNDO) {
369 int undo = un->semadj[sop->sem_num] - sem_op;
370 /*
371 * Exceeding the undo range is an error.
372 */
373 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
374 goto out_of_range;
375 }
376 curr->semval = result;
377 }
378
379 sop--;
380 while (sop >= sops) {
381 sma->sem_base[sop->sem_num].sempid = pid;
382 if (sop->sem_flg & SEM_UNDO)
383 un->semadj[sop->sem_num] -= sop->sem_op;
384 sop--;
385 }
386
387 return 0;
388
389out_of_range:
390 result = -ERANGE;
391 goto undo;
392
393would_block:
394 if (sop->sem_flg & IPC_NOWAIT)
395 result = -EAGAIN;
396 else
397 result = 1;
398
399undo:
400 sop--;
401 while (sop >= sops) {
402 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
403 sop--;
404 }
405
406 return result;
407}
408
409/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
410 * @q: queue entry that must be signaled
411 * @error: Error value for the signal
412 *
413 * Prepare the wake-up of the queue entry q.
414 */
415static void wake_up_sem_queue_prepare(struct list_head *pt,
416 struct sem_queue *q, int error)
417{
418 if (list_empty(pt)) {
419 /*
420 * Hold preempt off so that we don't get preempted and have the
421 * wakee busy-wait until we're scheduled back on.
422 */
423 preempt_disable();
424 }
425 q->status = IN_WAKEUP;
426 q->pid = error;
427
428 list_add_tail(&q->simple_list, pt);
429}
430
431/**
432 * wake_up_sem_queue_do(pt) - do the actual wake-up
433 * @pt: list of tasks to be woken up
434 *
435 * Do the actual wake-up.
436 * The function is called without any locks held, thus the semaphore array
437 * could be destroyed already and the tasks can disappear as soon as the
438 * status is set to the actual return code.
439 */
440static void wake_up_sem_queue_do(struct list_head *pt)
441{
442 struct sem_queue *q, *t;
443 int did_something;
444
445 did_something = !list_empty(pt);
446 list_for_each_entry_safe(q, t, pt, simple_list) {
447 wake_up_process(q->sleeper);
448 /* q can disappear immediately after writing q->status. */
449 smp_wmb();
450 q->status = q->pid;
451 }
452 if (did_something)
453 preempt_enable();
454}
455
456static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
457{
458 list_del(&q->list);
459 if (q->nsops == 1)
460 list_del(&q->simple_list);
461 else
462 sma->complex_count--;
463}
464
465/** check_restart(sma, q)
466 * @sma: semaphore array
467 * @q: the operation that just completed
468 *
469 * update_queue is O(N^2) when it restarts scanning the whole queue of
470 * waiting operations. Therefore this function checks if the restart is
471 * really necessary. It is called after a previously waiting operation
472 * was completed.
473 */
474static int check_restart(struct sem_array *sma, struct sem_queue *q)
475{
476 struct sem *curr;
477 struct sem_queue *h;
478
479 /* if the operation didn't modify the array, then no restart */
480 if (q->alter == 0)
481 return 0;
482
483 /* pending complex operations are too difficult to analyse */
484 if (sma->complex_count)
485 return 1;
486
487 /* we were a sleeping complex operation. Too difficult */
488 if (q->nsops > 1)
489 return 1;
490
491 curr = sma->sem_base + q->sops[0].sem_num;
492
493 /* No-one waits on this queue */
494 if (list_empty(&curr->sem_pending))
495 return 0;
496
497 /* the new semaphore value */
498 if (curr->semval) {
499 /* It is impossible that someone waits for the new value:
500 * - q is a previously sleeping simple operation that
501 * altered the array. It must be a decrement, because
502 * simple increments never sleep.
503 * - The value is not 0, thus wait-for-zero won't proceed.
504 * - If there are older (higher priority) decrements
505 * in the queue, then they have observed the original
506 * semval value and couldn't proceed. The operation
507 * decremented to value - thus they won't proceed either.
508 */
509 BUG_ON(q->sops[0].sem_op >= 0);
510 return 0;
511 }
512 /*
513 * semval is 0. Check if there are wait-for-zero semops.
514 * They must be the first entries in the per-semaphore simple queue
515 */
516 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
517 BUG_ON(h->nsops != 1);
518 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
519
520 /* Yes, there is a wait-for-zero semop. Restart */
521 if (h->sops[0].sem_op == 0)
522 return 1;
523
524 /* Again - no-one is waiting for the new value. */
525 return 0;
526}
527
528
529/**
530 * update_queue(sma, semnum): Look for tasks that can be completed.
531 * @sma: semaphore array.
532 * @semnum: semaphore that was modified.
533 * @pt: list head for the tasks that must be woken up.
534 *
535 * update_queue must be called after a semaphore in a semaphore array
536 * was modified. If multiple semaphore were modified, then @semnum
537 * must be set to -1.
538 * The tasks that must be woken up are added to @pt. The return code
539 * is stored in q->pid.
540 * The function return 1 if at least one semop was completed successfully.
541 */
542static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
543{
544 struct sem_queue *q;
545 struct list_head *walk;
546 struct list_head *pending_list;
547 int offset;
548 int semop_completed = 0;
549
550 /* if there are complex operations around, then knowing the semaphore
551 * that was modified doesn't help us. Assume that multiple semaphores
552 * were modified.
553 */
554 if (sma->complex_count)
555 semnum = -1;
556
557 if (semnum == -1) {
558 pending_list = &sma->sem_pending;
559 offset = offsetof(struct sem_queue, list);
560 } else {
561 pending_list = &sma->sem_base[semnum].sem_pending;
562 offset = offsetof(struct sem_queue, simple_list);
563 }
564
565again:
566 walk = pending_list->next;
567 while (walk != pending_list) {
568 int error, restart;
569
570 q = (struct sem_queue *)((char *)walk - offset);
571 walk = walk->next;
572
573 /* If we are scanning the single sop, per-semaphore list of
574 * one semaphore and that semaphore is 0, then it is not
575 * necessary to scan the "alter" entries: simple increments
576 * that affect only one entry succeed immediately and cannot
577 * be in the per semaphore pending queue, and decrements
578 * cannot be successful if the value is already 0.
579 */
580 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
581 q->alter)
582 break;
583
584 error = try_atomic_semop(sma, q->sops, q->nsops,
585 q->undo, q->pid);
586
587 /* Does q->sleeper still need to sleep? */
588 if (error > 0)
589 continue;
590
591 unlink_queue(sma, q);
592
593 if (error) {
594 restart = 0;
595 } else {
596 semop_completed = 1;
597 restart = check_restart(sma, q);
598 }
599
600 wake_up_sem_queue_prepare(pt, q, error);
601 if (restart)
602 goto again;
603 }
604 return semop_completed;
605}
606
607/**
608 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
609 * @sma: semaphore array
610 * @sops: operations that were performed
611 * @nsops: number of operations
612 * @otime: force setting otime
613 * @pt: list head of the tasks that must be woken up.
614 *
615 * do_smart_update() does the required called to update_queue, based on the
616 * actual changes that were performed on the semaphore array.
617 * Note that the function does not do the actual wake-up: the caller is
618 * responsible for calling wake_up_sem_queue_do(@pt).
619 * It is safe to perform this call after dropping all locks.
620 */
621static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
622 int otime, struct list_head *pt)
623{
624 int i;
625
626 if (sma->complex_count || sops == NULL) {
627 if (update_queue(sma, -1, pt))
628 otime = 1;
629 goto done;
630 }
631
632 for (i = 0; i < nsops; i++) {
633 if (sops[i].sem_op > 0 ||
634 (sops[i].sem_op < 0 &&
635 sma->sem_base[sops[i].sem_num].semval == 0))
636 if (update_queue(sma, sops[i].sem_num, pt))
637 otime = 1;
638 }
639done:
640 if (otime)
641 sma->sem_otime = get_seconds();
642}
643
644
645/* The following counts are associated to each semaphore:
646 * semncnt number of tasks waiting on semval being nonzero
647 * semzcnt number of tasks waiting on semval being zero
648 * This model assumes that a task waits on exactly one semaphore.
649 * Since semaphore operations are to be performed atomically, tasks actually
650 * wait on a whole sequence of semaphores simultaneously.
651 * The counts we return here are a rough approximation, but still
652 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
653 */
654static int count_semncnt (struct sem_array * sma, ushort semnum)
655{
656 int semncnt;
657 struct sem_queue * q;
658
659 semncnt = 0;
660 list_for_each_entry(q, &sma->sem_pending, list) {
661 struct sembuf * sops = q->sops;
662 int nsops = q->nsops;
663 int i;
664 for (i = 0; i < nsops; i++)
665 if (sops[i].sem_num == semnum
666 && (sops[i].sem_op < 0)
667 && !(sops[i].sem_flg & IPC_NOWAIT))
668 semncnt++;
669 }
670 return semncnt;
671}
672
673static int count_semzcnt (struct sem_array * sma, ushort semnum)
674{
675 int semzcnt;
676 struct sem_queue * q;
677
678 semzcnt = 0;
679 list_for_each_entry(q, &sma->sem_pending, list) {
680 struct sembuf * sops = q->sops;
681 int nsops = q->nsops;
682 int i;
683 for (i = 0; i < nsops; i++)
684 if (sops[i].sem_num == semnum
685 && (sops[i].sem_op == 0)
686 && !(sops[i].sem_flg & IPC_NOWAIT))
687 semzcnt++;
688 }
689 return semzcnt;
690}
691
692/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
693 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
694 * remains locked on exit.
695 */
696static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
697{
698 struct sem_undo *un, *tu;
699 struct sem_queue *q, *tq;
700 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
701 struct list_head tasks;
702
703 /* Free the existing undo structures for this semaphore set. */
704 assert_spin_locked(&sma->sem_perm.lock);
705 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
706 list_del(&un->list_id);
707 spin_lock(&un->ulp->lock);
708 un->semid = -1;
709 list_del_rcu(&un->list_proc);
710 spin_unlock(&un->ulp->lock);
711 kfree_rcu(un, rcu);
712 }
713
714 /* Wake up all pending processes and let them fail with EIDRM. */
715 INIT_LIST_HEAD(&tasks);
716 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
717 unlink_queue(sma, q);
718 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
719 }
720
721 /* Remove the semaphore set from the IDR */
722 sem_rmid(ns, sma);
723 sem_unlock(sma);
724
725 wake_up_sem_queue_do(&tasks);
726 ns->used_sems -= sma->sem_nsems;
727 security_sem_free(sma);
728 ipc_rcu_putref(sma);
729}
730
731static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
732{
733 switch(version) {
734 case IPC_64:
735 return copy_to_user(buf, in, sizeof(*in));
736 case IPC_OLD:
737 {
738 struct semid_ds out;
739
740 memset(&out, 0, sizeof(out));
741
742 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
743
744 out.sem_otime = in->sem_otime;
745 out.sem_ctime = in->sem_ctime;
746 out.sem_nsems = in->sem_nsems;
747
748 return copy_to_user(buf, &out, sizeof(out));
749 }
750 default:
751 return -EINVAL;
752 }
753}
754
755static int semctl_nolock(struct ipc_namespace *ns, int semid,
756 int cmd, int version, union semun arg)
757{
758 int err;
759 struct sem_array *sma;
760
761 switch(cmd) {
762 case IPC_INFO:
763 case SEM_INFO:
764 {
765 struct seminfo seminfo;
766 int max_id;
767
768 err = security_sem_semctl(NULL, cmd);
769 if (err)
770 return err;
771
772 memset(&seminfo,0,sizeof(seminfo));
773 seminfo.semmni = ns->sc_semmni;
774 seminfo.semmns = ns->sc_semmns;
775 seminfo.semmsl = ns->sc_semmsl;
776 seminfo.semopm = ns->sc_semopm;
777 seminfo.semvmx = SEMVMX;
778 seminfo.semmnu = SEMMNU;
779 seminfo.semmap = SEMMAP;
780 seminfo.semume = SEMUME;
781 down_read(&sem_ids(ns).rw_mutex);
782 if (cmd == SEM_INFO) {
783 seminfo.semusz = sem_ids(ns).in_use;
784 seminfo.semaem = ns->used_sems;
785 } else {
786 seminfo.semusz = SEMUSZ;
787 seminfo.semaem = SEMAEM;
788 }
789 max_id = ipc_get_maxid(&sem_ids(ns));
790 up_read(&sem_ids(ns).rw_mutex);
791 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
792 return -EFAULT;
793 return (max_id < 0) ? 0: max_id;
794 }
795 case IPC_STAT:
796 case SEM_STAT:
797 {
798 struct semid64_ds tbuf;
799 int id;
800
801 if (cmd == SEM_STAT) {
802 sma = sem_lock(ns, semid);
803 if (IS_ERR(sma))
804 return PTR_ERR(sma);
805 id = sma->sem_perm.id;
806 } else {
807 sma = sem_lock_check(ns, semid);
808 if (IS_ERR(sma))
809 return PTR_ERR(sma);
810 id = 0;
811 }
812
813 err = -EACCES;
814 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
815 goto out_unlock;
816
817 err = security_sem_semctl(sma, cmd);
818 if (err)
819 goto out_unlock;
820
821 memset(&tbuf, 0, sizeof(tbuf));
822
823 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
824 tbuf.sem_otime = sma->sem_otime;
825 tbuf.sem_ctime = sma->sem_ctime;
826 tbuf.sem_nsems = sma->sem_nsems;
827 sem_unlock(sma);
828 if (copy_semid_to_user (arg.buf, &tbuf, version))
829 return -EFAULT;
830 return id;
831 }
832 default:
833 return -EINVAL;
834 }
835out_unlock:
836 sem_unlock(sma);
837 return err;
838}
839
840static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
841 int cmd, int version, union semun arg)
842{
843 struct sem_array *sma;
844 struct sem* curr;
845 int err;
846 ushort fast_sem_io[SEMMSL_FAST];
847 ushort* sem_io = fast_sem_io;
848 int nsems;
849 struct list_head tasks;
850
851 sma = sem_lock_check(ns, semid);
852 if (IS_ERR(sma))
853 return PTR_ERR(sma);
854
855 INIT_LIST_HEAD(&tasks);
856 nsems = sma->sem_nsems;
857
858 err = -EACCES;
859 if (ipcperms(ns, &sma->sem_perm,
860 (cmd == SETVAL || cmd == SETALL) ? S_IWUGO : S_IRUGO))
861 goto out_unlock;
862
863 err = security_sem_semctl(sma, cmd);
864 if (err)
865 goto out_unlock;
866
867 err = -EACCES;
868 switch (cmd) {
869 case GETALL:
870 {
871 ushort __user *array = arg.array;
872 int i;
873
874 if(nsems > SEMMSL_FAST) {
875 sem_getref_and_unlock(sma);
876
877 sem_io = ipc_alloc(sizeof(ushort)*nsems);
878 if(sem_io == NULL) {
879 sem_putref(sma);
880 return -ENOMEM;
881 }
882
883 sem_lock_and_putref(sma);
884 if (sma->sem_perm.deleted) {
885 sem_unlock(sma);
886 err = -EIDRM;
887 goto out_free;
888 }
889 }
890
891 for (i = 0; i < sma->sem_nsems; i++)
892 sem_io[i] = sma->sem_base[i].semval;
893 sem_unlock(sma);
894 err = 0;
895 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
896 err = -EFAULT;
897 goto out_free;
898 }
899 case SETALL:
900 {
901 int i;
902 struct sem_undo *un;
903
904 sem_getref_and_unlock(sma);
905
906 if(nsems > SEMMSL_FAST) {
907 sem_io = ipc_alloc(sizeof(ushort)*nsems);
908 if(sem_io == NULL) {
909 sem_putref(sma);
910 return -ENOMEM;
911 }
912 }
913
914 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
915 sem_putref(sma);
916 err = -EFAULT;
917 goto out_free;
918 }
919
920 for (i = 0; i < nsems; i++) {
921 if (sem_io[i] > SEMVMX) {
922 sem_putref(sma);
923 err = -ERANGE;
924 goto out_free;
925 }
926 }
927 sem_lock_and_putref(sma);
928 if (sma->sem_perm.deleted) {
929 sem_unlock(sma);
930 err = -EIDRM;
931 goto out_free;
932 }
933
934 for (i = 0; i < nsems; i++)
935 sma->sem_base[i].semval = sem_io[i];
936
937 assert_spin_locked(&sma->sem_perm.lock);
938 list_for_each_entry(un, &sma->list_id, list_id) {
939 for (i = 0; i < nsems; i++)
940 un->semadj[i] = 0;
941 }
942 sma->sem_ctime = get_seconds();
943 /* maybe some queued-up processes were waiting for this */
944 do_smart_update(sma, NULL, 0, 0, &tasks);
945 err = 0;
946 goto out_unlock;
947 }
948 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
949 }
950 err = -EINVAL;
951 if(semnum < 0 || semnum >= nsems)
952 goto out_unlock;
953
954 curr = &sma->sem_base[semnum];
955
956 switch (cmd) {
957 case GETVAL:
958 err = curr->semval;
959 goto out_unlock;
960 case GETPID:
961 err = curr->sempid;
962 goto out_unlock;
963 case GETNCNT:
964 err = count_semncnt(sma,semnum);
965 goto out_unlock;
966 case GETZCNT:
967 err = count_semzcnt(sma,semnum);
968 goto out_unlock;
969 case SETVAL:
970 {
971 int val = arg.val;
972 struct sem_undo *un;
973
974 err = -ERANGE;
975 if (val > SEMVMX || val < 0)
976 goto out_unlock;
977
978 assert_spin_locked(&sma->sem_perm.lock);
979 list_for_each_entry(un, &sma->list_id, list_id)
980 un->semadj[semnum] = 0;
981
982 curr->semval = val;
983 curr->sempid = task_tgid_vnr(current);
984 sma->sem_ctime = get_seconds();
985 /* maybe some queued-up processes were waiting for this */
986 do_smart_update(sma, NULL, 0, 0, &tasks);
987 err = 0;
988 goto out_unlock;
989 }
990 }
991out_unlock:
992 sem_unlock(sma);
993 wake_up_sem_queue_do(&tasks);
994
995out_free:
996 if(sem_io != fast_sem_io)
997 ipc_free(sem_io, sizeof(ushort)*nsems);
998 return err;
999}
1000
1001static inline unsigned long
1002copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1003{
1004 switch(version) {
1005 case IPC_64:
1006 if (copy_from_user(out, buf, sizeof(*out)))
1007 return -EFAULT;
1008 return 0;
1009 case IPC_OLD:
1010 {
1011 struct semid_ds tbuf_old;
1012
1013 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1014 return -EFAULT;
1015
1016 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1017 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1018 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1019
1020 return 0;
1021 }
1022 default:
1023 return -EINVAL;
1024 }
1025}
1026
1027/*
1028 * This function handles some semctl commands which require the rw_mutex
1029 * to be held in write mode.
1030 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1031 */
1032static int semctl_down(struct ipc_namespace *ns, int semid,
1033 int cmd, int version, union semun arg)
1034{
1035 struct sem_array *sma;
1036 int err;
1037 struct semid64_ds semid64;
1038 struct kern_ipc_perm *ipcp;
1039
1040 if(cmd == IPC_SET) {
1041 if (copy_semid_from_user(&semid64, arg.buf, version))
1042 return -EFAULT;
1043 }
1044
1045 ipcp = ipcctl_pre_down(ns, &sem_ids(ns), semid, cmd,
1046 &semid64.sem_perm, 0);
1047 if (IS_ERR(ipcp))
1048 return PTR_ERR(ipcp);
1049
1050 sma = container_of(ipcp, struct sem_array, sem_perm);
1051
1052 err = security_sem_semctl(sma, cmd);
1053 if (err)
1054 goto out_unlock;
1055
1056 switch(cmd){
1057 case IPC_RMID:
1058 freeary(ns, ipcp);
1059 goto out_up;
1060 case IPC_SET:
1061 ipc_update_perm(&semid64.sem_perm, ipcp);
1062 sma->sem_ctime = get_seconds();
1063 break;
1064 default:
1065 err = -EINVAL;
1066 }
1067
1068out_unlock:
1069 sem_unlock(sma);
1070out_up:
1071 up_write(&sem_ids(ns).rw_mutex);
1072 return err;
1073}
1074
1075SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1076{
1077 int err = -EINVAL;
1078 int version;
1079 struct ipc_namespace *ns;
1080
1081 if (semid < 0)
1082 return -EINVAL;
1083
1084 version = ipc_parse_version(&cmd);
1085 ns = current->nsproxy->ipc_ns;
1086
1087 switch(cmd) {
1088 case IPC_INFO:
1089 case SEM_INFO:
1090 case IPC_STAT:
1091 case SEM_STAT:
1092 err = semctl_nolock(ns, semid, cmd, version, arg);
1093 return err;
1094 case GETALL:
1095 case GETVAL:
1096 case GETPID:
1097 case GETNCNT:
1098 case GETZCNT:
1099 case SETVAL:
1100 case SETALL:
1101 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1102 return err;
1103 case IPC_RMID:
1104 case IPC_SET:
1105 err = semctl_down(ns, semid, cmd, version, arg);
1106 return err;
1107 default:
1108 return -EINVAL;
1109 }
1110}
1111#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1112asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1113{
1114 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1115}
1116SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1117#endif
1118
1119/* If the task doesn't already have a undo_list, then allocate one
1120 * here. We guarantee there is only one thread using this undo list,
1121 * and current is THE ONE
1122 *
1123 * If this allocation and assignment succeeds, but later
1124 * portions of this code fail, there is no need to free the sem_undo_list.
1125 * Just let it stay associated with the task, and it'll be freed later
1126 * at exit time.
1127 *
1128 * This can block, so callers must hold no locks.
1129 */
1130static inline int get_undo_list(struct sem_undo_list **undo_listp)
1131{
1132 struct sem_undo_list *undo_list;
1133
1134 undo_list = current->sysvsem.undo_list;
1135 if (!undo_list) {
1136 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1137 if (undo_list == NULL)
1138 return -ENOMEM;
1139 spin_lock_init(&undo_list->lock);
1140 atomic_set(&undo_list->refcnt, 1);
1141 INIT_LIST_HEAD(&undo_list->list_proc);
1142
1143 current->sysvsem.undo_list = undo_list;
1144 }
1145 *undo_listp = undo_list;
1146 return 0;
1147}
1148
1149static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1150{
1151 struct sem_undo *un;
1152
1153 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1154 if (un->semid == semid)
1155 return un;
1156 }
1157 return NULL;
1158}
1159
1160static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1161{
1162 struct sem_undo *un;
1163
1164 assert_spin_locked(&ulp->lock);
1165
1166 un = __lookup_undo(ulp, semid);
1167 if (un) {
1168 list_del_rcu(&un->list_proc);
1169 list_add_rcu(&un->list_proc, &ulp->list_proc);
1170 }
1171 return un;
1172}
1173
1174/**
1175 * find_alloc_undo - Lookup (and if not present create) undo array
1176 * @ns: namespace
1177 * @semid: semaphore array id
1178 *
1179 * The function looks up (and if not present creates) the undo structure.
1180 * The size of the undo structure depends on the size of the semaphore
1181 * array, thus the alloc path is not that straightforward.
1182 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1183 * performs a rcu_read_lock().
1184 */
1185static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1186{
1187 struct sem_array *sma;
1188 struct sem_undo_list *ulp;
1189 struct sem_undo *un, *new;
1190 int nsems;
1191 int error;
1192
1193 error = get_undo_list(&ulp);
1194 if (error)
1195 return ERR_PTR(error);
1196
1197 rcu_read_lock();
1198 spin_lock(&ulp->lock);
1199 un = lookup_undo(ulp, semid);
1200 spin_unlock(&ulp->lock);
1201 if (likely(un!=NULL))
1202 goto out;
1203 rcu_read_unlock();
1204
1205 /* no undo structure around - allocate one. */
1206 /* step 1: figure out the size of the semaphore array */
1207 sma = sem_lock_check(ns, semid);
1208 if (IS_ERR(sma))
1209 return ERR_CAST(sma);
1210
1211 nsems = sma->sem_nsems;
1212 sem_getref_and_unlock(sma);
1213
1214 /* step 2: allocate new undo structure */
1215 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1216 if (!new) {
1217 sem_putref(sma);
1218 return ERR_PTR(-ENOMEM);
1219 }
1220
1221 /* step 3: Acquire the lock on semaphore array */
1222 sem_lock_and_putref(sma);
1223 if (sma->sem_perm.deleted) {
1224 sem_unlock(sma);
1225 kfree(new);
1226 un = ERR_PTR(-EIDRM);
1227 goto out;
1228 }
1229 spin_lock(&ulp->lock);
1230
1231 /*
1232 * step 4: check for races: did someone else allocate the undo struct?
1233 */
1234 un = lookup_undo(ulp, semid);
1235 if (un) {
1236 kfree(new);
1237 goto success;
1238 }
1239 /* step 5: initialize & link new undo structure */
1240 new->semadj = (short *) &new[1];
1241 new->ulp = ulp;
1242 new->semid = semid;
1243 assert_spin_locked(&ulp->lock);
1244 list_add_rcu(&new->list_proc, &ulp->list_proc);
1245 assert_spin_locked(&sma->sem_perm.lock);
1246 list_add(&new->list_id, &sma->list_id);
1247 un = new;
1248
1249success:
1250 spin_unlock(&ulp->lock);
1251 rcu_read_lock();
1252 sem_unlock(sma);
1253out:
1254 return un;
1255}
1256
1257
1258/**
1259 * get_queue_result - Retrieve the result code from sem_queue
1260 * @q: Pointer to queue structure
1261 *
1262 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1263 * q->status, then we must loop until the value is replaced with the final
1264 * value: This may happen if a task is woken up by an unrelated event (e.g.
1265 * signal) and in parallel the task is woken up by another task because it got
1266 * the requested semaphores.
1267 *
1268 * The function can be called with or without holding the semaphore spinlock.
1269 */
1270static int get_queue_result(struct sem_queue *q)
1271{
1272 int error;
1273
1274 error = q->status;
1275 while (unlikely(error == IN_WAKEUP)) {
1276 cpu_relax();
1277 error = q->status;
1278 }
1279
1280 return error;
1281}
1282
1283
1284SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1285 unsigned, nsops, const struct timespec __user *, timeout)
1286{
1287 int error = -EINVAL;
1288 struct sem_array *sma;
1289 struct sembuf fast_sops[SEMOPM_FAST];
1290 struct sembuf* sops = fast_sops, *sop;
1291 struct sem_undo *un;
1292 int undos = 0, alter = 0, max;
1293 struct sem_queue queue;
1294 unsigned long jiffies_left = 0;
1295 struct ipc_namespace *ns;
1296 struct list_head tasks;
1297
1298 ns = current->nsproxy->ipc_ns;
1299
1300 if (nsops < 1 || semid < 0)
1301 return -EINVAL;
1302 if (nsops > ns->sc_semopm)
1303 return -E2BIG;
1304 if(nsops > SEMOPM_FAST) {
1305 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1306 if(sops==NULL)
1307 return -ENOMEM;
1308 }
1309 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1310 error=-EFAULT;
1311 goto out_free;
1312 }
1313 if (timeout) {
1314 struct timespec _timeout;
1315 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1316 error = -EFAULT;
1317 goto out_free;
1318 }
1319 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1320 _timeout.tv_nsec >= 1000000000L) {
1321 error = -EINVAL;
1322 goto out_free;
1323 }
1324 jiffies_left = timespec_to_jiffies(&_timeout);
1325 }
1326 max = 0;
1327 for (sop = sops; sop < sops + nsops; sop++) {
1328 if (sop->sem_num >= max)
1329 max = sop->sem_num;
1330 if (sop->sem_flg & SEM_UNDO)
1331 undos = 1;
1332 if (sop->sem_op != 0)
1333 alter = 1;
1334 }
1335
1336 if (undos) {
1337 un = find_alloc_undo(ns, semid);
1338 if (IS_ERR(un)) {
1339 error = PTR_ERR(un);
1340 goto out_free;
1341 }
1342 } else
1343 un = NULL;
1344
1345 INIT_LIST_HEAD(&tasks);
1346
1347 sma = sem_lock_check(ns, semid);
1348 if (IS_ERR(sma)) {
1349 if (un)
1350 rcu_read_unlock();
1351 error = PTR_ERR(sma);
1352 goto out_free;
1353 }
1354
1355 /*
1356 * semid identifiers are not unique - find_alloc_undo may have
1357 * allocated an undo structure, it was invalidated by an RMID
1358 * and now a new array with received the same id. Check and fail.
1359 * This case can be detected checking un->semid. The existence of
1360 * "un" itself is guaranteed by rcu.
1361 */
1362 error = -EIDRM;
1363 if (un) {
1364 if (un->semid == -1) {
1365 rcu_read_unlock();
1366 goto out_unlock_free;
1367 } else {
1368 /*
1369 * rcu lock can be released, "un" cannot disappear:
1370 * - sem_lock is acquired, thus IPC_RMID is
1371 * impossible.
1372 * - exit_sem is impossible, it always operates on
1373 * current (or a dead task).
1374 */
1375
1376 rcu_read_unlock();
1377 }
1378 }
1379
1380 error = -EFBIG;
1381 if (max >= sma->sem_nsems)
1382 goto out_unlock_free;
1383
1384 error = -EACCES;
1385 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1386 goto out_unlock_free;
1387
1388 error = security_sem_semop(sma, sops, nsops, alter);
1389 if (error)
1390 goto out_unlock_free;
1391
1392 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1393 if (error <= 0) {
1394 if (alter && error == 0)
1395 do_smart_update(sma, sops, nsops, 1, &tasks);
1396
1397 goto out_unlock_free;
1398 }
1399
1400 /* We need to sleep on this operation, so we put the current
1401 * task into the pending queue and go to sleep.
1402 */
1403
1404 queue.sops = sops;
1405 queue.nsops = nsops;
1406 queue.undo = un;
1407 queue.pid = task_tgid_vnr(current);
1408 queue.alter = alter;
1409 if (alter)
1410 list_add_tail(&queue.list, &sma->sem_pending);
1411 else
1412 list_add(&queue.list, &sma->sem_pending);
1413
1414 if (nsops == 1) {
1415 struct sem *curr;
1416 curr = &sma->sem_base[sops->sem_num];
1417
1418 if (alter)
1419 list_add_tail(&queue.simple_list, &curr->sem_pending);
1420 else
1421 list_add(&queue.simple_list, &curr->sem_pending);
1422 } else {
1423 INIT_LIST_HEAD(&queue.simple_list);
1424 sma->complex_count++;
1425 }
1426
1427 queue.status = -EINTR;
1428 queue.sleeper = current;
1429 current->state = TASK_INTERRUPTIBLE;
1430 sem_unlock(sma);
1431
1432 if (timeout)
1433 jiffies_left = schedule_timeout(jiffies_left);
1434 else
1435 schedule();
1436
1437 error = get_queue_result(&queue);
1438
1439 if (error != -EINTR) {
1440 /* fast path: update_queue already obtained all requested
1441 * resources.
1442 * Perform a smp_mb(): User space could assume that semop()
1443 * is a memory barrier: Without the mb(), the cpu could
1444 * speculatively read in user space stale data that was
1445 * overwritten by the previous owner of the semaphore.
1446 */
1447 smp_mb();
1448
1449 goto out_free;
1450 }
1451
1452 sma = sem_lock(ns, semid);
1453
1454 /*
1455 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1456 */
1457 error = get_queue_result(&queue);
1458
1459 /*
1460 * Array removed? If yes, leave without sem_unlock().
1461 */
1462 if (IS_ERR(sma)) {
1463 error = -EIDRM;
1464 goto out_free;
1465 }
1466
1467
1468 /*
1469 * If queue.status != -EINTR we are woken up by another process.
1470 * Leave without unlink_queue(), but with sem_unlock().
1471 */
1472
1473 if (error != -EINTR) {
1474 goto out_unlock_free;
1475 }
1476
1477 /*
1478 * If an interrupt occurred we have to clean up the queue
1479 */
1480 if (timeout && jiffies_left == 0)
1481 error = -EAGAIN;
1482 unlink_queue(sma, &queue);
1483
1484out_unlock_free:
1485 sem_unlock(sma);
1486
1487 wake_up_sem_queue_do(&tasks);
1488out_free:
1489 if(sops != fast_sops)
1490 kfree(sops);
1491 return error;
1492}
1493
1494SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1495 unsigned, nsops)
1496{
1497 return sys_semtimedop(semid, tsops, nsops, NULL);
1498}
1499
1500/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1501 * parent and child tasks.
1502 */
1503
1504int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1505{
1506 struct sem_undo_list *undo_list;
1507 int error;
1508
1509 if (clone_flags & CLONE_SYSVSEM) {
1510 error = get_undo_list(&undo_list);
1511 if (error)
1512 return error;
1513 atomic_inc(&undo_list->refcnt);
1514 tsk->sysvsem.undo_list = undo_list;
1515 } else
1516 tsk->sysvsem.undo_list = NULL;
1517
1518 return 0;
1519}
1520
1521/*
1522 * add semadj values to semaphores, free undo structures.
1523 * undo structures are not freed when semaphore arrays are destroyed
1524 * so some of them may be out of date.
1525 * IMPLEMENTATION NOTE: There is some confusion over whether the
1526 * set of adjustments that needs to be done should be done in an atomic
1527 * manner or not. That is, if we are attempting to decrement the semval
1528 * should we queue up and wait until we can do so legally?
1529 * The original implementation attempted to do this (queue and wait).
1530 * The current implementation does not do so. The POSIX standard
1531 * and SVID should be consulted to determine what behavior is mandated.
1532 */
1533void exit_sem(struct task_struct *tsk)
1534{
1535 struct sem_undo_list *ulp;
1536
1537 ulp = tsk->sysvsem.undo_list;
1538 if (!ulp)
1539 return;
1540 tsk->sysvsem.undo_list = NULL;
1541
1542 if (!atomic_dec_and_test(&ulp->refcnt))
1543 return;
1544
1545 for (;;) {
1546 struct sem_array *sma;
1547 struct sem_undo *un;
1548 struct list_head tasks;
1549 int semid;
1550 int i;
1551
1552 rcu_read_lock();
1553 un = list_entry_rcu(ulp->list_proc.next,
1554 struct sem_undo, list_proc);
1555 if (&un->list_proc == &ulp->list_proc)
1556 semid = -1;
1557 else
1558 semid = un->semid;
1559 rcu_read_unlock();
1560
1561 if (semid == -1)
1562 break;
1563
1564 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1565
1566 /* exit_sem raced with IPC_RMID, nothing to do */
1567 if (IS_ERR(sma))
1568 continue;
1569
1570 un = __lookup_undo(ulp, semid);
1571 if (un == NULL) {
1572 /* exit_sem raced with IPC_RMID+semget() that created
1573 * exactly the same semid. Nothing to do.
1574 */
1575 sem_unlock(sma);
1576 continue;
1577 }
1578
1579 /* remove un from the linked lists */
1580 assert_spin_locked(&sma->sem_perm.lock);
1581 list_del(&un->list_id);
1582
1583 spin_lock(&ulp->lock);
1584 list_del_rcu(&un->list_proc);
1585 spin_unlock(&ulp->lock);
1586
1587 /* perform adjustments registered in un */
1588 for (i = 0; i < sma->sem_nsems; i++) {
1589 struct sem * semaphore = &sma->sem_base[i];
1590 if (un->semadj[i]) {
1591 semaphore->semval += un->semadj[i];
1592 /*
1593 * Range checks of the new semaphore value,
1594 * not defined by sus:
1595 * - Some unices ignore the undo entirely
1596 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1597 * - some cap the value (e.g. FreeBSD caps
1598 * at 0, but doesn't enforce SEMVMX)
1599 *
1600 * Linux caps the semaphore value, both at 0
1601 * and at SEMVMX.
1602 *
1603 * Manfred <manfred@colorfullife.com>
1604 */
1605 if (semaphore->semval < 0)
1606 semaphore->semval = 0;
1607 if (semaphore->semval > SEMVMX)
1608 semaphore->semval = SEMVMX;
1609 semaphore->sempid = task_tgid_vnr(current);
1610 }
1611 }
1612 /* maybe some queued-up processes were waiting for this */
1613 INIT_LIST_HEAD(&tasks);
1614 do_smart_update(sma, NULL, 0, 1, &tasks);
1615 sem_unlock(sma);
1616 wake_up_sem_queue_do(&tasks);
1617
1618 kfree_rcu(un, rcu);
1619 }
1620 kfree(ulp);
1621}
1622
1623#ifdef CONFIG_PROC_FS
1624static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1625{
1626 struct sem_array *sma = it;
1627
1628 return seq_printf(s,
1629 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1630 sma->sem_perm.key,
1631 sma->sem_perm.id,
1632 sma->sem_perm.mode,
1633 sma->sem_nsems,
1634 sma->sem_perm.uid,
1635 sma->sem_perm.gid,
1636 sma->sem_perm.cuid,
1637 sma->sem_perm.cgid,
1638 sma->sem_otime,
1639 sma->sem_ctime);
1640}
1641#endif
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * linux/ipc/sem.c
4 * Copyright (C) 1992 Krishna Balasubramanian
5 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 *
7 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 *
9 * SMP-threaded, sysctl's added
10 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
11 * Enforced range limit on SEM_UNDO
12 * (c) 2001 Red Hat Inc
13 * Lockless wakeup
14 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
15 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
16 * Further wakeup optimizations, documentation
17 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
18 *
19 * support for audit of ipc object properties and permission changes
20 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
21 *
22 * namespaces support
23 * OpenVZ, SWsoft Inc.
24 * Pavel Emelianov <xemul@openvz.org>
25 *
26 * Implementation notes: (May 2010)
27 * This file implements System V semaphores.
28 *
29 * User space visible behavior:
30 * - FIFO ordering for semop() operations (just FIFO, not starvation
31 * protection)
32 * - multiple semaphore operations that alter the same semaphore in
33 * one semop() are handled.
34 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
35 * SETALL calls.
36 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
37 * - undo adjustments at process exit are limited to 0..SEMVMX.
38 * - namespace are supported.
39 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
40 * to /proc/sys/kernel/sem.
41 * - statistics about the usage are reported in /proc/sysvipc/sem.
42 *
43 * Internals:
44 * - scalability:
45 * - all global variables are read-mostly.
46 * - semop() calls and semctl(RMID) are synchronized by RCU.
47 * - most operations do write operations (actually: spin_lock calls) to
48 * the per-semaphore array structure.
49 * Thus: Perfect SMP scaling between independent semaphore arrays.
50 * If multiple semaphores in one array are used, then cache line
51 * trashing on the semaphore array spinlock will limit the scaling.
52 * - semncnt and semzcnt are calculated on demand in count_semcnt()
53 * - the task that performs a successful semop() scans the list of all
54 * sleeping tasks and completes any pending operations that can be fulfilled.
55 * Semaphores are actively given to waiting tasks (necessary for FIFO).
56 * (see update_queue())
57 * - To improve the scalability, the actual wake-up calls are performed after
58 * dropping all locks. (see wake_up_sem_queue_prepare())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - UNDO values are stored in an array (one per process and per
64 * semaphore array, lazily allocated). For backwards compatibility, multiple
65 * modes for the UNDO variables are supported (per process, per thread)
66 * (see copy_semundo, CLONE_SYSVSEM)
67 * - There are two lists of the pending operations: a per-array list
68 * and per-semaphore list (stored in the array). This allows to achieve FIFO
69 * ordering without always scanning all pending operations.
70 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
71 */
72
73#include <linux/compat.h>
74#include <linux/slab.h>
75#include <linux/spinlock.h>
76#include <linux/init.h>
77#include <linux/proc_fs.h>
78#include <linux/time.h>
79#include <linux/security.h>
80#include <linux/syscalls.h>
81#include <linux/audit.h>
82#include <linux/capability.h>
83#include <linux/seq_file.h>
84#include <linux/rwsem.h>
85#include <linux/nsproxy.h>
86#include <linux/ipc_namespace.h>
87#include <linux/sched/wake_q.h>
88#include <linux/nospec.h>
89#include <linux/rhashtable.h>
90
91#include <linux/uaccess.h>
92#include "util.h"
93
94/* One semaphore structure for each semaphore in the system. */
95struct sem {
96 int semval; /* current value */
97 /*
98 * PID of the process that last modified the semaphore. For
99 * Linux, specifically these are:
100 * - semop
101 * - semctl, via SETVAL and SETALL.
102 * - at task exit when performing undo adjustments (see exit_sem).
103 */
104 struct pid *sempid;
105 spinlock_t lock; /* spinlock for fine-grained semtimedop */
106 struct list_head pending_alter; /* pending single-sop operations */
107 /* that alter the semaphore */
108 struct list_head pending_const; /* pending single-sop operations */
109 /* that do not alter the semaphore*/
110 time64_t sem_otime; /* candidate for sem_otime */
111} ____cacheline_aligned_in_smp;
112
113/* One sem_array data structure for each set of semaphores in the system. */
114struct sem_array {
115 struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */
116 time64_t sem_ctime; /* create/last semctl() time */
117 struct list_head pending_alter; /* pending operations */
118 /* that alter the array */
119 struct list_head pending_const; /* pending complex operations */
120 /* that do not alter semvals */
121 struct list_head list_id; /* undo requests on this array */
122 int sem_nsems; /* no. of semaphores in array */
123 int complex_count; /* pending complex operations */
124 unsigned int use_global_lock;/* >0: global lock required */
125
126 struct sem sems[];
127} __randomize_layout;
128
129/* One queue for each sleeping process in the system. */
130struct sem_queue {
131 struct list_head list; /* queue of pending operations */
132 struct task_struct *sleeper; /* this process */
133 struct sem_undo *undo; /* undo structure */
134 struct pid *pid; /* process id of requesting process */
135 int status; /* completion status of operation */
136 struct sembuf *sops; /* array of pending operations */
137 struct sembuf *blocking; /* the operation that blocked */
138 int nsops; /* number of operations */
139 bool alter; /* does *sops alter the array? */
140 bool dupsop; /* sops on more than one sem_num */
141};
142
143/* Each task has a list of undo requests. They are executed automatically
144 * when the process exits.
145 */
146struct sem_undo {
147 struct list_head list_proc; /* per-process list: *
148 * all undos from one process
149 * rcu protected */
150 struct rcu_head rcu; /* rcu struct for sem_undo */
151 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
152 struct list_head list_id; /* per semaphore array list:
153 * all undos for one array */
154 int semid; /* semaphore set identifier */
155 short *semadj; /* array of adjustments */
156 /* one per semaphore */
157};
158
159/* sem_undo_list controls shared access to the list of sem_undo structures
160 * that may be shared among all a CLONE_SYSVSEM task group.
161 */
162struct sem_undo_list {
163 refcount_t refcnt;
164 spinlock_t lock;
165 struct list_head list_proc;
166};
167
168
169#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
170
171static int newary(struct ipc_namespace *, struct ipc_params *);
172static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
173#ifdef CONFIG_PROC_FS
174static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
175#endif
176
177#define SEMMSL_FAST 256 /* 512 bytes on stack */
178#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
179
180/*
181 * Switching from the mode suitable for simple ops
182 * to the mode for complex ops is costly. Therefore:
183 * use some hysteresis
184 */
185#define USE_GLOBAL_LOCK_HYSTERESIS 10
186
187/*
188 * Locking:
189 * a) global sem_lock() for read/write
190 * sem_undo.id_next,
191 * sem_array.complex_count,
192 * sem_array.pending{_alter,_const},
193 * sem_array.sem_undo
194 *
195 * b) global or semaphore sem_lock() for read/write:
196 * sem_array.sems[i].pending_{const,alter}:
197 *
198 * c) special:
199 * sem_undo_list.list_proc:
200 * * undo_list->lock for write
201 * * rcu for read
202 * use_global_lock:
203 * * global sem_lock() for write
204 * * either local or global sem_lock() for read.
205 *
206 * Memory ordering:
207 * Most ordering is enforced by using spin_lock() and spin_unlock().
208 * The special case is use_global_lock:
209 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
210 * using smp_store_release().
211 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
212 * smp_load_acquire().
213 * Setting it from 0 to non-zero must be ordered with regards to
214 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
215 * is inside a spin_lock() and after a write from 0 to non-zero a
216 * spin_lock()+spin_unlock() is done.
217 */
218
219#define sc_semmsl sem_ctls[0]
220#define sc_semmns sem_ctls[1]
221#define sc_semopm sem_ctls[2]
222#define sc_semmni sem_ctls[3]
223
224void sem_init_ns(struct ipc_namespace *ns)
225{
226 ns->sc_semmsl = SEMMSL;
227 ns->sc_semmns = SEMMNS;
228 ns->sc_semopm = SEMOPM;
229 ns->sc_semmni = SEMMNI;
230 ns->used_sems = 0;
231 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
232}
233
234#ifdef CONFIG_IPC_NS
235void sem_exit_ns(struct ipc_namespace *ns)
236{
237 free_ipcs(ns, &sem_ids(ns), freeary);
238 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
239 rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht);
240}
241#endif
242
243void __init sem_init(void)
244{
245 sem_init_ns(&init_ipc_ns);
246 ipc_init_proc_interface("sysvipc/sem",
247 " key semid perms nsems uid gid cuid cgid otime ctime\n",
248 IPC_SEM_IDS, sysvipc_sem_proc_show);
249}
250
251/**
252 * unmerge_queues - unmerge queues, if possible.
253 * @sma: semaphore array
254 *
255 * The function unmerges the wait queues if complex_count is 0.
256 * It must be called prior to dropping the global semaphore array lock.
257 */
258static void unmerge_queues(struct sem_array *sma)
259{
260 struct sem_queue *q, *tq;
261
262 /* complex operations still around? */
263 if (sma->complex_count)
264 return;
265 /*
266 * We will switch back to simple mode.
267 * Move all pending operation back into the per-semaphore
268 * queues.
269 */
270 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
271 struct sem *curr;
272 curr = &sma->sems[q->sops[0].sem_num];
273
274 list_add_tail(&q->list, &curr->pending_alter);
275 }
276 INIT_LIST_HEAD(&sma->pending_alter);
277}
278
279/**
280 * merge_queues - merge single semop queues into global queue
281 * @sma: semaphore array
282 *
283 * This function merges all per-semaphore queues into the global queue.
284 * It is necessary to achieve FIFO ordering for the pending single-sop
285 * operations when a multi-semop operation must sleep.
286 * Only the alter operations must be moved, the const operations can stay.
287 */
288static void merge_queues(struct sem_array *sma)
289{
290 int i;
291 for (i = 0; i < sma->sem_nsems; i++) {
292 struct sem *sem = &sma->sems[i];
293
294 list_splice_init(&sem->pending_alter, &sma->pending_alter);
295 }
296}
297
298static void sem_rcu_free(struct rcu_head *head)
299{
300 struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu);
301 struct sem_array *sma = container_of(p, struct sem_array, sem_perm);
302
303 security_sem_free(&sma->sem_perm);
304 kvfree(sma);
305}
306
307/*
308 * Enter the mode suitable for non-simple operations:
309 * Caller must own sem_perm.lock.
310 */
311static void complexmode_enter(struct sem_array *sma)
312{
313 int i;
314 struct sem *sem;
315
316 if (sma->use_global_lock > 0) {
317 /*
318 * We are already in global lock mode.
319 * Nothing to do, just reset the
320 * counter until we return to simple mode.
321 */
322 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
323 return;
324 }
325 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
326
327 for (i = 0; i < sma->sem_nsems; i++) {
328 sem = &sma->sems[i];
329 spin_lock(&sem->lock);
330 spin_unlock(&sem->lock);
331 }
332}
333
334/*
335 * Try to leave the mode that disallows simple operations:
336 * Caller must own sem_perm.lock.
337 */
338static void complexmode_tryleave(struct sem_array *sma)
339{
340 if (sma->complex_count) {
341 /* Complex ops are sleeping.
342 * We must stay in complex mode
343 */
344 return;
345 }
346 if (sma->use_global_lock == 1) {
347 /*
348 * Immediately after setting use_global_lock to 0,
349 * a simple op can start. Thus: all memory writes
350 * performed by the current operation must be visible
351 * before we set use_global_lock to 0.
352 */
353 smp_store_release(&sma->use_global_lock, 0);
354 } else {
355 sma->use_global_lock--;
356 }
357}
358
359#define SEM_GLOBAL_LOCK (-1)
360/*
361 * If the request contains only one semaphore operation, and there are
362 * no complex transactions pending, lock only the semaphore involved.
363 * Otherwise, lock the entire semaphore array, since we either have
364 * multiple semaphores in our own semops, or we need to look at
365 * semaphores from other pending complex operations.
366 */
367static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
368 int nsops)
369{
370 struct sem *sem;
371 int idx;
372
373 if (nsops != 1) {
374 /* Complex operation - acquire a full lock */
375 ipc_lock_object(&sma->sem_perm);
376
377 /* Prevent parallel simple ops */
378 complexmode_enter(sma);
379 return SEM_GLOBAL_LOCK;
380 }
381
382 /*
383 * Only one semaphore affected - try to optimize locking.
384 * Optimized locking is possible if no complex operation
385 * is either enqueued or processed right now.
386 *
387 * Both facts are tracked by use_global_mode.
388 */
389 idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
390 sem = &sma->sems[idx];
391
392 /*
393 * Initial check for use_global_lock. Just an optimization,
394 * no locking, no memory barrier.
395 */
396 if (!sma->use_global_lock) {
397 /*
398 * It appears that no complex operation is around.
399 * Acquire the per-semaphore lock.
400 */
401 spin_lock(&sem->lock);
402
403 /* pairs with smp_store_release() */
404 if (!smp_load_acquire(&sma->use_global_lock)) {
405 /* fast path successful! */
406 return sops->sem_num;
407 }
408 spin_unlock(&sem->lock);
409 }
410
411 /* slow path: acquire the full lock */
412 ipc_lock_object(&sma->sem_perm);
413
414 if (sma->use_global_lock == 0) {
415 /*
416 * The use_global_lock mode ended while we waited for
417 * sma->sem_perm.lock. Thus we must switch to locking
418 * with sem->lock.
419 * Unlike in the fast path, there is no need to recheck
420 * sma->use_global_lock after we have acquired sem->lock:
421 * We own sma->sem_perm.lock, thus use_global_lock cannot
422 * change.
423 */
424 spin_lock(&sem->lock);
425
426 ipc_unlock_object(&sma->sem_perm);
427 return sops->sem_num;
428 } else {
429 /*
430 * Not a false alarm, thus continue to use the global lock
431 * mode. No need for complexmode_enter(), this was done by
432 * the caller that has set use_global_mode to non-zero.
433 */
434 return SEM_GLOBAL_LOCK;
435 }
436}
437
438static inline void sem_unlock(struct sem_array *sma, int locknum)
439{
440 if (locknum == SEM_GLOBAL_LOCK) {
441 unmerge_queues(sma);
442 complexmode_tryleave(sma);
443 ipc_unlock_object(&sma->sem_perm);
444 } else {
445 struct sem *sem = &sma->sems[locknum];
446 spin_unlock(&sem->lock);
447 }
448}
449
450/*
451 * sem_lock_(check_) routines are called in the paths where the rwsem
452 * is not held.
453 *
454 * The caller holds the RCU read lock.
455 */
456static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
457{
458 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
459
460 if (IS_ERR(ipcp))
461 return ERR_CAST(ipcp);
462
463 return container_of(ipcp, struct sem_array, sem_perm);
464}
465
466static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
467 int id)
468{
469 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
470
471 if (IS_ERR(ipcp))
472 return ERR_CAST(ipcp);
473
474 return container_of(ipcp, struct sem_array, sem_perm);
475}
476
477static inline void sem_lock_and_putref(struct sem_array *sma)
478{
479 sem_lock(sma, NULL, -1);
480 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
481}
482
483static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
484{
485 ipc_rmid(&sem_ids(ns), &s->sem_perm);
486}
487
488static struct sem_array *sem_alloc(size_t nsems)
489{
490 struct sem_array *sma;
491
492 if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0]))
493 return NULL;
494
495 sma = kvzalloc(struct_size(sma, sems, nsems), GFP_KERNEL);
496 if (unlikely(!sma))
497 return NULL;
498
499 return sma;
500}
501
502/**
503 * newary - Create a new semaphore set
504 * @ns: namespace
505 * @params: ptr to the structure that contains key, semflg and nsems
506 *
507 * Called with sem_ids.rwsem held (as a writer)
508 */
509static int newary(struct ipc_namespace *ns, struct ipc_params *params)
510{
511 int retval;
512 struct sem_array *sma;
513 key_t key = params->key;
514 int nsems = params->u.nsems;
515 int semflg = params->flg;
516 int i;
517
518 if (!nsems)
519 return -EINVAL;
520 if (ns->used_sems + nsems > ns->sc_semmns)
521 return -ENOSPC;
522
523 sma = sem_alloc(nsems);
524 if (!sma)
525 return -ENOMEM;
526
527 sma->sem_perm.mode = (semflg & S_IRWXUGO);
528 sma->sem_perm.key = key;
529
530 sma->sem_perm.security = NULL;
531 retval = security_sem_alloc(&sma->sem_perm);
532 if (retval) {
533 kvfree(sma);
534 return retval;
535 }
536
537 for (i = 0; i < nsems; i++) {
538 INIT_LIST_HEAD(&sma->sems[i].pending_alter);
539 INIT_LIST_HEAD(&sma->sems[i].pending_const);
540 spin_lock_init(&sma->sems[i].lock);
541 }
542
543 sma->complex_count = 0;
544 sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS;
545 INIT_LIST_HEAD(&sma->pending_alter);
546 INIT_LIST_HEAD(&sma->pending_const);
547 INIT_LIST_HEAD(&sma->list_id);
548 sma->sem_nsems = nsems;
549 sma->sem_ctime = ktime_get_real_seconds();
550
551 /* ipc_addid() locks sma upon success. */
552 retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
553 if (retval < 0) {
554 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
555 return retval;
556 }
557 ns->used_sems += nsems;
558
559 sem_unlock(sma, -1);
560 rcu_read_unlock();
561
562 return sma->sem_perm.id;
563}
564
565
566/*
567 * Called with sem_ids.rwsem and ipcp locked.
568 */
569static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
570 struct ipc_params *params)
571{
572 struct sem_array *sma;
573
574 sma = container_of(ipcp, struct sem_array, sem_perm);
575 if (params->u.nsems > sma->sem_nsems)
576 return -EINVAL;
577
578 return 0;
579}
580
581long ksys_semget(key_t key, int nsems, int semflg)
582{
583 struct ipc_namespace *ns;
584 static const struct ipc_ops sem_ops = {
585 .getnew = newary,
586 .associate = security_sem_associate,
587 .more_checks = sem_more_checks,
588 };
589 struct ipc_params sem_params;
590
591 ns = current->nsproxy->ipc_ns;
592
593 if (nsems < 0 || nsems > ns->sc_semmsl)
594 return -EINVAL;
595
596 sem_params.key = key;
597 sem_params.flg = semflg;
598 sem_params.u.nsems = nsems;
599
600 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
601}
602
603SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
604{
605 return ksys_semget(key, nsems, semflg);
606}
607
608/**
609 * perform_atomic_semop[_slow] - Attempt to perform semaphore
610 * operations on a given array.
611 * @sma: semaphore array
612 * @q: struct sem_queue that describes the operation
613 *
614 * Caller blocking are as follows, based the value
615 * indicated by the semaphore operation (sem_op):
616 *
617 * (1) >0 never blocks.
618 * (2) 0 (wait-for-zero operation): semval is non-zero.
619 * (3) <0 attempting to decrement semval to a value smaller than zero.
620 *
621 * Returns 0 if the operation was possible.
622 * Returns 1 if the operation is impossible, the caller must sleep.
623 * Returns <0 for error codes.
624 */
625static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
626{
627 int result, sem_op, nsops;
628 struct pid *pid;
629 struct sembuf *sop;
630 struct sem *curr;
631 struct sembuf *sops;
632 struct sem_undo *un;
633
634 sops = q->sops;
635 nsops = q->nsops;
636 un = q->undo;
637
638 for (sop = sops; sop < sops + nsops; sop++) {
639 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
640 curr = &sma->sems[idx];
641 sem_op = sop->sem_op;
642 result = curr->semval;
643
644 if (!sem_op && result)
645 goto would_block;
646
647 result += sem_op;
648 if (result < 0)
649 goto would_block;
650 if (result > SEMVMX)
651 goto out_of_range;
652
653 if (sop->sem_flg & SEM_UNDO) {
654 int undo = un->semadj[sop->sem_num] - sem_op;
655 /* Exceeding the undo range is an error. */
656 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
657 goto out_of_range;
658 un->semadj[sop->sem_num] = undo;
659 }
660
661 curr->semval = result;
662 }
663
664 sop--;
665 pid = q->pid;
666 while (sop >= sops) {
667 ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid);
668 sop--;
669 }
670
671 return 0;
672
673out_of_range:
674 result = -ERANGE;
675 goto undo;
676
677would_block:
678 q->blocking = sop;
679
680 if (sop->sem_flg & IPC_NOWAIT)
681 result = -EAGAIN;
682 else
683 result = 1;
684
685undo:
686 sop--;
687 while (sop >= sops) {
688 sem_op = sop->sem_op;
689 sma->sems[sop->sem_num].semval -= sem_op;
690 if (sop->sem_flg & SEM_UNDO)
691 un->semadj[sop->sem_num] += sem_op;
692 sop--;
693 }
694
695 return result;
696}
697
698static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
699{
700 int result, sem_op, nsops;
701 struct sembuf *sop;
702 struct sem *curr;
703 struct sembuf *sops;
704 struct sem_undo *un;
705
706 sops = q->sops;
707 nsops = q->nsops;
708 un = q->undo;
709
710 if (unlikely(q->dupsop))
711 return perform_atomic_semop_slow(sma, q);
712
713 /*
714 * We scan the semaphore set twice, first to ensure that the entire
715 * operation can succeed, therefore avoiding any pointless writes
716 * to shared memory and having to undo such changes in order to block
717 * until the operations can go through.
718 */
719 for (sop = sops; sop < sops + nsops; sop++) {
720 int idx = array_index_nospec(sop->sem_num, sma->sem_nsems);
721
722 curr = &sma->sems[idx];
723 sem_op = sop->sem_op;
724 result = curr->semval;
725
726 if (!sem_op && result)
727 goto would_block; /* wait-for-zero */
728
729 result += sem_op;
730 if (result < 0)
731 goto would_block;
732
733 if (result > SEMVMX)
734 return -ERANGE;
735
736 if (sop->sem_flg & SEM_UNDO) {
737 int undo = un->semadj[sop->sem_num] - sem_op;
738
739 /* Exceeding the undo range is an error. */
740 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
741 return -ERANGE;
742 }
743 }
744
745 for (sop = sops; sop < sops + nsops; sop++) {
746 curr = &sma->sems[sop->sem_num];
747 sem_op = sop->sem_op;
748 result = curr->semval;
749
750 if (sop->sem_flg & SEM_UNDO) {
751 int undo = un->semadj[sop->sem_num] - sem_op;
752
753 un->semadj[sop->sem_num] = undo;
754 }
755 curr->semval += sem_op;
756 ipc_update_pid(&curr->sempid, q->pid);
757 }
758
759 return 0;
760
761would_block:
762 q->blocking = sop;
763 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
764}
765
766static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
767 struct wake_q_head *wake_q)
768{
769 wake_q_add(wake_q, q->sleeper);
770 /*
771 * Rely on the above implicit barrier, such that we can
772 * ensure that we hold reference to the task before setting
773 * q->status. Otherwise we could race with do_exit if the
774 * task is awoken by an external event before calling
775 * wake_up_process().
776 */
777 WRITE_ONCE(q->status, error);
778}
779
780static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
781{
782 list_del(&q->list);
783 if (q->nsops > 1)
784 sma->complex_count--;
785}
786
787/** check_restart(sma, q)
788 * @sma: semaphore array
789 * @q: the operation that just completed
790 *
791 * update_queue is O(N^2) when it restarts scanning the whole queue of
792 * waiting operations. Therefore this function checks if the restart is
793 * really necessary. It is called after a previously waiting operation
794 * modified the array.
795 * Note that wait-for-zero operations are handled without restart.
796 */
797static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
798{
799 /* pending complex alter operations are too difficult to analyse */
800 if (!list_empty(&sma->pending_alter))
801 return 1;
802
803 /* we were a sleeping complex operation. Too difficult */
804 if (q->nsops > 1)
805 return 1;
806
807 /* It is impossible that someone waits for the new value:
808 * - complex operations always restart.
809 * - wait-for-zero are handled seperately.
810 * - q is a previously sleeping simple operation that
811 * altered the array. It must be a decrement, because
812 * simple increments never sleep.
813 * - If there are older (higher priority) decrements
814 * in the queue, then they have observed the original
815 * semval value and couldn't proceed. The operation
816 * decremented to value - thus they won't proceed either.
817 */
818 return 0;
819}
820
821/**
822 * wake_const_ops - wake up non-alter tasks
823 * @sma: semaphore array.
824 * @semnum: semaphore that was modified.
825 * @wake_q: lockless wake-queue head.
826 *
827 * wake_const_ops must be called after a semaphore in a semaphore array
828 * was set to 0. If complex const operations are pending, wake_const_ops must
829 * be called with semnum = -1, as well as with the number of each modified
830 * semaphore.
831 * The tasks that must be woken up are added to @wake_q. The return code
832 * is stored in q->pid.
833 * The function returns 1 if at least one operation was completed successfully.
834 */
835static int wake_const_ops(struct sem_array *sma, int semnum,
836 struct wake_q_head *wake_q)
837{
838 struct sem_queue *q, *tmp;
839 struct list_head *pending_list;
840 int semop_completed = 0;
841
842 if (semnum == -1)
843 pending_list = &sma->pending_const;
844 else
845 pending_list = &sma->sems[semnum].pending_const;
846
847 list_for_each_entry_safe(q, tmp, pending_list, list) {
848 int error = perform_atomic_semop(sma, q);
849
850 if (error > 0)
851 continue;
852 /* operation completed, remove from queue & wakeup */
853 unlink_queue(sma, q);
854
855 wake_up_sem_queue_prepare(q, error, wake_q);
856 if (error == 0)
857 semop_completed = 1;
858 }
859
860 return semop_completed;
861}
862
863/**
864 * do_smart_wakeup_zero - wakeup all wait for zero tasks
865 * @sma: semaphore array
866 * @sops: operations that were performed
867 * @nsops: number of operations
868 * @wake_q: lockless wake-queue head
869 *
870 * Checks all required queue for wait-for-zero operations, based
871 * on the actual changes that were performed on the semaphore array.
872 * The function returns 1 if at least one operation was completed successfully.
873 */
874static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
875 int nsops, struct wake_q_head *wake_q)
876{
877 int i;
878 int semop_completed = 0;
879 int got_zero = 0;
880
881 /* first: the per-semaphore queues, if known */
882 if (sops) {
883 for (i = 0; i < nsops; i++) {
884 int num = sops[i].sem_num;
885
886 if (sma->sems[num].semval == 0) {
887 got_zero = 1;
888 semop_completed |= wake_const_ops(sma, num, wake_q);
889 }
890 }
891 } else {
892 /*
893 * No sops means modified semaphores not known.
894 * Assume all were changed.
895 */
896 for (i = 0; i < sma->sem_nsems; i++) {
897 if (sma->sems[i].semval == 0) {
898 got_zero = 1;
899 semop_completed |= wake_const_ops(sma, i, wake_q);
900 }
901 }
902 }
903 /*
904 * If one of the modified semaphores got 0,
905 * then check the global queue, too.
906 */
907 if (got_zero)
908 semop_completed |= wake_const_ops(sma, -1, wake_q);
909
910 return semop_completed;
911}
912
913
914/**
915 * update_queue - look for tasks that can be completed.
916 * @sma: semaphore array.
917 * @semnum: semaphore that was modified.
918 * @wake_q: lockless wake-queue head.
919 *
920 * update_queue must be called after a semaphore in a semaphore array
921 * was modified. If multiple semaphores were modified, update_queue must
922 * be called with semnum = -1, as well as with the number of each modified
923 * semaphore.
924 * The tasks that must be woken up are added to @wake_q. The return code
925 * is stored in q->pid.
926 * The function internally checks if const operations can now succeed.
927 *
928 * The function return 1 if at least one semop was completed successfully.
929 */
930static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
931{
932 struct sem_queue *q, *tmp;
933 struct list_head *pending_list;
934 int semop_completed = 0;
935
936 if (semnum == -1)
937 pending_list = &sma->pending_alter;
938 else
939 pending_list = &sma->sems[semnum].pending_alter;
940
941again:
942 list_for_each_entry_safe(q, tmp, pending_list, list) {
943 int error, restart;
944
945 /* If we are scanning the single sop, per-semaphore list of
946 * one semaphore and that semaphore is 0, then it is not
947 * necessary to scan further: simple increments
948 * that affect only one entry succeed immediately and cannot
949 * be in the per semaphore pending queue, and decrements
950 * cannot be successful if the value is already 0.
951 */
952 if (semnum != -1 && sma->sems[semnum].semval == 0)
953 break;
954
955 error = perform_atomic_semop(sma, q);
956
957 /* Does q->sleeper still need to sleep? */
958 if (error > 0)
959 continue;
960
961 unlink_queue(sma, q);
962
963 if (error) {
964 restart = 0;
965 } else {
966 semop_completed = 1;
967 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
968 restart = check_restart(sma, q);
969 }
970
971 wake_up_sem_queue_prepare(q, error, wake_q);
972 if (restart)
973 goto again;
974 }
975 return semop_completed;
976}
977
978/**
979 * set_semotime - set sem_otime
980 * @sma: semaphore array
981 * @sops: operations that modified the array, may be NULL
982 *
983 * sem_otime is replicated to avoid cache line trashing.
984 * This function sets one instance to the current time.
985 */
986static void set_semotime(struct sem_array *sma, struct sembuf *sops)
987{
988 if (sops == NULL) {
989 sma->sems[0].sem_otime = ktime_get_real_seconds();
990 } else {
991 sma->sems[sops[0].sem_num].sem_otime =
992 ktime_get_real_seconds();
993 }
994}
995
996/**
997 * do_smart_update - optimized update_queue
998 * @sma: semaphore array
999 * @sops: operations that were performed
1000 * @nsops: number of operations
1001 * @otime: force setting otime
1002 * @wake_q: lockless wake-queue head
1003 *
1004 * do_smart_update() does the required calls to update_queue and wakeup_zero,
1005 * based on the actual changes that were performed on the semaphore array.
1006 * Note that the function does not do the actual wake-up: the caller is
1007 * responsible for calling wake_up_q().
1008 * It is safe to perform this call after dropping all locks.
1009 */
1010static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1011 int otime, struct wake_q_head *wake_q)
1012{
1013 int i;
1014
1015 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
1016
1017 if (!list_empty(&sma->pending_alter)) {
1018 /* semaphore array uses the global queue - just process it. */
1019 otime |= update_queue(sma, -1, wake_q);
1020 } else {
1021 if (!sops) {
1022 /*
1023 * No sops, thus the modified semaphores are not
1024 * known. Check all.
1025 */
1026 for (i = 0; i < sma->sem_nsems; i++)
1027 otime |= update_queue(sma, i, wake_q);
1028 } else {
1029 /*
1030 * Check the semaphores that were increased:
1031 * - No complex ops, thus all sleeping ops are
1032 * decrease.
1033 * - if we decreased the value, then any sleeping
1034 * semaphore ops wont be able to run: If the
1035 * previous value was too small, then the new
1036 * value will be too small, too.
1037 */
1038 for (i = 0; i < nsops; i++) {
1039 if (sops[i].sem_op > 0) {
1040 otime |= update_queue(sma,
1041 sops[i].sem_num, wake_q);
1042 }
1043 }
1044 }
1045 }
1046 if (otime)
1047 set_semotime(sma, sops);
1048}
1049
1050/*
1051 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1052 */
1053static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1054 bool count_zero)
1055{
1056 struct sembuf *sop = q->blocking;
1057
1058 /*
1059 * Linux always (since 0.99.10) reported a task as sleeping on all
1060 * semaphores. This violates SUS, therefore it was changed to the
1061 * standard compliant behavior.
1062 * Give the administrators a chance to notice that an application
1063 * might misbehave because it relies on the Linux behavior.
1064 */
1065 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1066 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1067 current->comm, task_pid_nr(current));
1068
1069 if (sop->sem_num != semnum)
1070 return 0;
1071
1072 if (count_zero && sop->sem_op == 0)
1073 return 1;
1074 if (!count_zero && sop->sem_op < 0)
1075 return 1;
1076
1077 return 0;
1078}
1079
1080/* The following counts are associated to each semaphore:
1081 * semncnt number of tasks waiting on semval being nonzero
1082 * semzcnt number of tasks waiting on semval being zero
1083 *
1084 * Per definition, a task waits only on the semaphore of the first semop
1085 * that cannot proceed, even if additional operation would block, too.
1086 */
1087static int count_semcnt(struct sem_array *sma, ushort semnum,
1088 bool count_zero)
1089{
1090 struct list_head *l;
1091 struct sem_queue *q;
1092 int semcnt;
1093
1094 semcnt = 0;
1095 /* First: check the simple operations. They are easy to evaluate */
1096 if (count_zero)
1097 l = &sma->sems[semnum].pending_const;
1098 else
1099 l = &sma->sems[semnum].pending_alter;
1100
1101 list_for_each_entry(q, l, list) {
1102 /* all task on a per-semaphore list sleep on exactly
1103 * that semaphore
1104 */
1105 semcnt++;
1106 }
1107
1108 /* Then: check the complex operations. */
1109 list_for_each_entry(q, &sma->pending_alter, list) {
1110 semcnt += check_qop(sma, semnum, q, count_zero);
1111 }
1112 if (count_zero) {
1113 list_for_each_entry(q, &sma->pending_const, list) {
1114 semcnt += check_qop(sma, semnum, q, count_zero);
1115 }
1116 }
1117 return semcnt;
1118}
1119
1120/* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1121 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1122 * remains locked on exit.
1123 */
1124static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1125{
1126 struct sem_undo *un, *tu;
1127 struct sem_queue *q, *tq;
1128 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1129 int i;
1130 DEFINE_WAKE_Q(wake_q);
1131
1132 /* Free the existing undo structures for this semaphore set. */
1133 ipc_assert_locked_object(&sma->sem_perm);
1134 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1135 list_del(&un->list_id);
1136 spin_lock(&un->ulp->lock);
1137 un->semid = -1;
1138 list_del_rcu(&un->list_proc);
1139 spin_unlock(&un->ulp->lock);
1140 kfree_rcu(un, rcu);
1141 }
1142
1143 /* Wake up all pending processes and let them fail with EIDRM. */
1144 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1145 unlink_queue(sma, q);
1146 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1147 }
1148
1149 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1150 unlink_queue(sma, q);
1151 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1152 }
1153 for (i = 0; i < sma->sem_nsems; i++) {
1154 struct sem *sem = &sma->sems[i];
1155 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1156 unlink_queue(sma, q);
1157 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1158 }
1159 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1160 unlink_queue(sma, q);
1161 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1162 }
1163 ipc_update_pid(&sem->sempid, NULL);
1164 }
1165
1166 /* Remove the semaphore set from the IDR */
1167 sem_rmid(ns, sma);
1168 sem_unlock(sma, -1);
1169 rcu_read_unlock();
1170
1171 wake_up_q(&wake_q);
1172 ns->used_sems -= sma->sem_nsems;
1173 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1174}
1175
1176static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1177{
1178 switch (version) {
1179 case IPC_64:
1180 return copy_to_user(buf, in, sizeof(*in));
1181 case IPC_OLD:
1182 {
1183 struct semid_ds out;
1184
1185 memset(&out, 0, sizeof(out));
1186
1187 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1188
1189 out.sem_otime = in->sem_otime;
1190 out.sem_ctime = in->sem_ctime;
1191 out.sem_nsems = in->sem_nsems;
1192
1193 return copy_to_user(buf, &out, sizeof(out));
1194 }
1195 default:
1196 return -EINVAL;
1197 }
1198}
1199
1200static time64_t get_semotime(struct sem_array *sma)
1201{
1202 int i;
1203 time64_t res;
1204
1205 res = sma->sems[0].sem_otime;
1206 for (i = 1; i < sma->sem_nsems; i++) {
1207 time64_t to = sma->sems[i].sem_otime;
1208
1209 if (to > res)
1210 res = to;
1211 }
1212 return res;
1213}
1214
1215static int semctl_stat(struct ipc_namespace *ns, int semid,
1216 int cmd, struct semid64_ds *semid64)
1217{
1218 struct sem_array *sma;
1219 time64_t semotime;
1220 int err;
1221
1222 memset(semid64, 0, sizeof(*semid64));
1223
1224 rcu_read_lock();
1225 if (cmd == SEM_STAT || cmd == SEM_STAT_ANY) {
1226 sma = sem_obtain_object(ns, semid);
1227 if (IS_ERR(sma)) {
1228 err = PTR_ERR(sma);
1229 goto out_unlock;
1230 }
1231 } else { /* IPC_STAT */
1232 sma = sem_obtain_object_check(ns, semid);
1233 if (IS_ERR(sma)) {
1234 err = PTR_ERR(sma);
1235 goto out_unlock;
1236 }
1237 }
1238
1239 /* see comment for SHM_STAT_ANY */
1240 if (cmd == SEM_STAT_ANY)
1241 audit_ipc_obj(&sma->sem_perm);
1242 else {
1243 err = -EACCES;
1244 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1245 goto out_unlock;
1246 }
1247
1248 err = security_sem_semctl(&sma->sem_perm, cmd);
1249 if (err)
1250 goto out_unlock;
1251
1252 ipc_lock_object(&sma->sem_perm);
1253
1254 if (!ipc_valid_object(&sma->sem_perm)) {
1255 ipc_unlock_object(&sma->sem_perm);
1256 err = -EIDRM;
1257 goto out_unlock;
1258 }
1259
1260 kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm);
1261 semotime = get_semotime(sma);
1262 semid64->sem_otime = semotime;
1263 semid64->sem_ctime = sma->sem_ctime;
1264#ifndef CONFIG_64BIT
1265 semid64->sem_otime_high = semotime >> 32;
1266 semid64->sem_ctime_high = sma->sem_ctime >> 32;
1267#endif
1268 semid64->sem_nsems = sma->sem_nsems;
1269
1270 if (cmd == IPC_STAT) {
1271 /*
1272 * As defined in SUS:
1273 * Return 0 on success
1274 */
1275 err = 0;
1276 } else {
1277 /*
1278 * SEM_STAT and SEM_STAT_ANY (both Linux specific)
1279 * Return the full id, including the sequence number
1280 */
1281 err = sma->sem_perm.id;
1282 }
1283 ipc_unlock_object(&sma->sem_perm);
1284out_unlock:
1285 rcu_read_unlock();
1286 return err;
1287}
1288
1289static int semctl_info(struct ipc_namespace *ns, int semid,
1290 int cmd, void __user *p)
1291{
1292 struct seminfo seminfo;
1293 int max_idx;
1294 int err;
1295
1296 err = security_sem_semctl(NULL, cmd);
1297 if (err)
1298 return err;
1299
1300 memset(&seminfo, 0, sizeof(seminfo));
1301 seminfo.semmni = ns->sc_semmni;
1302 seminfo.semmns = ns->sc_semmns;
1303 seminfo.semmsl = ns->sc_semmsl;
1304 seminfo.semopm = ns->sc_semopm;
1305 seminfo.semvmx = SEMVMX;
1306 seminfo.semmnu = SEMMNU;
1307 seminfo.semmap = SEMMAP;
1308 seminfo.semume = SEMUME;
1309 down_read(&sem_ids(ns).rwsem);
1310 if (cmd == SEM_INFO) {
1311 seminfo.semusz = sem_ids(ns).in_use;
1312 seminfo.semaem = ns->used_sems;
1313 } else {
1314 seminfo.semusz = SEMUSZ;
1315 seminfo.semaem = SEMAEM;
1316 }
1317 max_idx = ipc_get_maxidx(&sem_ids(ns));
1318 up_read(&sem_ids(ns).rwsem);
1319 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1320 return -EFAULT;
1321 return (max_idx < 0) ? 0 : max_idx;
1322}
1323
1324static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1325 int val)
1326{
1327 struct sem_undo *un;
1328 struct sem_array *sma;
1329 struct sem *curr;
1330 int err;
1331 DEFINE_WAKE_Q(wake_q);
1332
1333 if (val > SEMVMX || val < 0)
1334 return -ERANGE;
1335
1336 rcu_read_lock();
1337 sma = sem_obtain_object_check(ns, semid);
1338 if (IS_ERR(sma)) {
1339 rcu_read_unlock();
1340 return PTR_ERR(sma);
1341 }
1342
1343 if (semnum < 0 || semnum >= sma->sem_nsems) {
1344 rcu_read_unlock();
1345 return -EINVAL;
1346 }
1347
1348
1349 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1350 rcu_read_unlock();
1351 return -EACCES;
1352 }
1353
1354 err = security_sem_semctl(&sma->sem_perm, SETVAL);
1355 if (err) {
1356 rcu_read_unlock();
1357 return -EACCES;
1358 }
1359
1360 sem_lock(sma, NULL, -1);
1361
1362 if (!ipc_valid_object(&sma->sem_perm)) {
1363 sem_unlock(sma, -1);
1364 rcu_read_unlock();
1365 return -EIDRM;
1366 }
1367
1368 semnum = array_index_nospec(semnum, sma->sem_nsems);
1369 curr = &sma->sems[semnum];
1370
1371 ipc_assert_locked_object(&sma->sem_perm);
1372 list_for_each_entry(un, &sma->list_id, list_id)
1373 un->semadj[semnum] = 0;
1374
1375 curr->semval = val;
1376 ipc_update_pid(&curr->sempid, task_tgid(current));
1377 sma->sem_ctime = ktime_get_real_seconds();
1378 /* maybe some queued-up processes were waiting for this */
1379 do_smart_update(sma, NULL, 0, 0, &wake_q);
1380 sem_unlock(sma, -1);
1381 rcu_read_unlock();
1382 wake_up_q(&wake_q);
1383 return 0;
1384}
1385
1386static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1387 int cmd, void __user *p)
1388{
1389 struct sem_array *sma;
1390 struct sem *curr;
1391 int err, nsems;
1392 ushort fast_sem_io[SEMMSL_FAST];
1393 ushort *sem_io = fast_sem_io;
1394 DEFINE_WAKE_Q(wake_q);
1395
1396 rcu_read_lock();
1397 sma = sem_obtain_object_check(ns, semid);
1398 if (IS_ERR(sma)) {
1399 rcu_read_unlock();
1400 return PTR_ERR(sma);
1401 }
1402
1403 nsems = sma->sem_nsems;
1404
1405 err = -EACCES;
1406 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1407 goto out_rcu_wakeup;
1408
1409 err = security_sem_semctl(&sma->sem_perm, cmd);
1410 if (err)
1411 goto out_rcu_wakeup;
1412
1413 err = -EACCES;
1414 switch (cmd) {
1415 case GETALL:
1416 {
1417 ushort __user *array = p;
1418 int i;
1419
1420 sem_lock(sma, NULL, -1);
1421 if (!ipc_valid_object(&sma->sem_perm)) {
1422 err = -EIDRM;
1423 goto out_unlock;
1424 }
1425 if (nsems > SEMMSL_FAST) {
1426 if (!ipc_rcu_getref(&sma->sem_perm)) {
1427 err = -EIDRM;
1428 goto out_unlock;
1429 }
1430 sem_unlock(sma, -1);
1431 rcu_read_unlock();
1432 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1433 GFP_KERNEL);
1434 if (sem_io == NULL) {
1435 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1436 return -ENOMEM;
1437 }
1438
1439 rcu_read_lock();
1440 sem_lock_and_putref(sma);
1441 if (!ipc_valid_object(&sma->sem_perm)) {
1442 err = -EIDRM;
1443 goto out_unlock;
1444 }
1445 }
1446 for (i = 0; i < sma->sem_nsems; i++)
1447 sem_io[i] = sma->sems[i].semval;
1448 sem_unlock(sma, -1);
1449 rcu_read_unlock();
1450 err = 0;
1451 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1452 err = -EFAULT;
1453 goto out_free;
1454 }
1455 case SETALL:
1456 {
1457 int i;
1458 struct sem_undo *un;
1459
1460 if (!ipc_rcu_getref(&sma->sem_perm)) {
1461 err = -EIDRM;
1462 goto out_rcu_wakeup;
1463 }
1464 rcu_read_unlock();
1465
1466 if (nsems > SEMMSL_FAST) {
1467 sem_io = kvmalloc_array(nsems, sizeof(ushort),
1468 GFP_KERNEL);
1469 if (sem_io == NULL) {
1470 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1471 return -ENOMEM;
1472 }
1473 }
1474
1475 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1476 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1477 err = -EFAULT;
1478 goto out_free;
1479 }
1480
1481 for (i = 0; i < nsems; i++) {
1482 if (sem_io[i] > SEMVMX) {
1483 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1484 err = -ERANGE;
1485 goto out_free;
1486 }
1487 }
1488 rcu_read_lock();
1489 sem_lock_and_putref(sma);
1490 if (!ipc_valid_object(&sma->sem_perm)) {
1491 err = -EIDRM;
1492 goto out_unlock;
1493 }
1494
1495 for (i = 0; i < nsems; i++) {
1496 sma->sems[i].semval = sem_io[i];
1497 ipc_update_pid(&sma->sems[i].sempid, task_tgid(current));
1498 }
1499
1500 ipc_assert_locked_object(&sma->sem_perm);
1501 list_for_each_entry(un, &sma->list_id, list_id) {
1502 for (i = 0; i < nsems; i++)
1503 un->semadj[i] = 0;
1504 }
1505 sma->sem_ctime = ktime_get_real_seconds();
1506 /* maybe some queued-up processes were waiting for this */
1507 do_smart_update(sma, NULL, 0, 0, &wake_q);
1508 err = 0;
1509 goto out_unlock;
1510 }
1511 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1512 }
1513 err = -EINVAL;
1514 if (semnum < 0 || semnum >= nsems)
1515 goto out_rcu_wakeup;
1516
1517 sem_lock(sma, NULL, -1);
1518 if (!ipc_valid_object(&sma->sem_perm)) {
1519 err = -EIDRM;
1520 goto out_unlock;
1521 }
1522
1523 semnum = array_index_nospec(semnum, nsems);
1524 curr = &sma->sems[semnum];
1525
1526 switch (cmd) {
1527 case GETVAL:
1528 err = curr->semval;
1529 goto out_unlock;
1530 case GETPID:
1531 err = pid_vnr(curr->sempid);
1532 goto out_unlock;
1533 case GETNCNT:
1534 err = count_semcnt(sma, semnum, 0);
1535 goto out_unlock;
1536 case GETZCNT:
1537 err = count_semcnt(sma, semnum, 1);
1538 goto out_unlock;
1539 }
1540
1541out_unlock:
1542 sem_unlock(sma, -1);
1543out_rcu_wakeup:
1544 rcu_read_unlock();
1545 wake_up_q(&wake_q);
1546out_free:
1547 if (sem_io != fast_sem_io)
1548 kvfree(sem_io);
1549 return err;
1550}
1551
1552static inline unsigned long
1553copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1554{
1555 switch (version) {
1556 case IPC_64:
1557 if (copy_from_user(out, buf, sizeof(*out)))
1558 return -EFAULT;
1559 return 0;
1560 case IPC_OLD:
1561 {
1562 struct semid_ds tbuf_old;
1563
1564 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1565 return -EFAULT;
1566
1567 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1568 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1569 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1570
1571 return 0;
1572 }
1573 default:
1574 return -EINVAL;
1575 }
1576}
1577
1578/*
1579 * This function handles some semctl commands which require the rwsem
1580 * to be held in write mode.
1581 * NOTE: no locks must be held, the rwsem is taken inside this function.
1582 */
1583static int semctl_down(struct ipc_namespace *ns, int semid,
1584 int cmd, struct semid64_ds *semid64)
1585{
1586 struct sem_array *sma;
1587 int err;
1588 struct kern_ipc_perm *ipcp;
1589
1590 down_write(&sem_ids(ns).rwsem);
1591 rcu_read_lock();
1592
1593 ipcp = ipcctl_obtain_check(ns, &sem_ids(ns), semid, cmd,
1594 &semid64->sem_perm, 0);
1595 if (IS_ERR(ipcp)) {
1596 err = PTR_ERR(ipcp);
1597 goto out_unlock1;
1598 }
1599
1600 sma = container_of(ipcp, struct sem_array, sem_perm);
1601
1602 err = security_sem_semctl(&sma->sem_perm, cmd);
1603 if (err)
1604 goto out_unlock1;
1605
1606 switch (cmd) {
1607 case IPC_RMID:
1608 sem_lock(sma, NULL, -1);
1609 /* freeary unlocks the ipc object and rcu */
1610 freeary(ns, ipcp);
1611 goto out_up;
1612 case IPC_SET:
1613 sem_lock(sma, NULL, -1);
1614 err = ipc_update_perm(&semid64->sem_perm, ipcp);
1615 if (err)
1616 goto out_unlock0;
1617 sma->sem_ctime = ktime_get_real_seconds();
1618 break;
1619 default:
1620 err = -EINVAL;
1621 goto out_unlock1;
1622 }
1623
1624out_unlock0:
1625 sem_unlock(sma, -1);
1626out_unlock1:
1627 rcu_read_unlock();
1628out_up:
1629 up_write(&sem_ids(ns).rwsem);
1630 return err;
1631}
1632
1633static long ksys_semctl(int semid, int semnum, int cmd, unsigned long arg, int version)
1634{
1635 struct ipc_namespace *ns;
1636 void __user *p = (void __user *)arg;
1637 struct semid64_ds semid64;
1638 int err;
1639
1640 if (semid < 0)
1641 return -EINVAL;
1642
1643 ns = current->nsproxy->ipc_ns;
1644
1645 switch (cmd) {
1646 case IPC_INFO:
1647 case SEM_INFO:
1648 return semctl_info(ns, semid, cmd, p);
1649 case IPC_STAT:
1650 case SEM_STAT:
1651 case SEM_STAT_ANY:
1652 err = semctl_stat(ns, semid, cmd, &semid64);
1653 if (err < 0)
1654 return err;
1655 if (copy_semid_to_user(p, &semid64, version))
1656 err = -EFAULT;
1657 return err;
1658 case GETALL:
1659 case GETVAL:
1660 case GETPID:
1661 case GETNCNT:
1662 case GETZCNT:
1663 case SETALL:
1664 return semctl_main(ns, semid, semnum, cmd, p);
1665 case SETVAL: {
1666 int val;
1667#if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1668 /* big-endian 64bit */
1669 val = arg >> 32;
1670#else
1671 /* 32bit or little-endian 64bit */
1672 val = arg;
1673#endif
1674 return semctl_setval(ns, semid, semnum, val);
1675 }
1676 case IPC_SET:
1677 if (copy_semid_from_user(&semid64, p, version))
1678 return -EFAULT;
1679 /* fall through */
1680 case IPC_RMID:
1681 return semctl_down(ns, semid, cmd, &semid64);
1682 default:
1683 return -EINVAL;
1684 }
1685}
1686
1687SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1688{
1689 return ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1690}
1691
1692#ifdef CONFIG_ARCH_WANT_IPC_PARSE_VERSION
1693long ksys_old_semctl(int semid, int semnum, int cmd, unsigned long arg)
1694{
1695 int version = ipc_parse_version(&cmd);
1696
1697 return ksys_semctl(semid, semnum, cmd, arg, version);
1698}
1699
1700SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1701{
1702 return ksys_old_semctl(semid, semnum, cmd, arg);
1703}
1704#endif
1705
1706#ifdef CONFIG_COMPAT
1707
1708struct compat_semid_ds {
1709 struct compat_ipc_perm sem_perm;
1710 old_time32_t sem_otime;
1711 old_time32_t sem_ctime;
1712 compat_uptr_t sem_base;
1713 compat_uptr_t sem_pending;
1714 compat_uptr_t sem_pending_last;
1715 compat_uptr_t undo;
1716 unsigned short sem_nsems;
1717};
1718
1719static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf,
1720 int version)
1721{
1722 memset(out, 0, sizeof(*out));
1723 if (version == IPC_64) {
1724 struct compat_semid64_ds __user *p = buf;
1725 return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm);
1726 } else {
1727 struct compat_semid_ds __user *p = buf;
1728 return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm);
1729 }
1730}
1731
1732static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in,
1733 int version)
1734{
1735 if (version == IPC_64) {
1736 struct compat_semid64_ds v;
1737 memset(&v, 0, sizeof(v));
1738 to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm);
1739 v.sem_otime = lower_32_bits(in->sem_otime);
1740 v.sem_otime_high = upper_32_bits(in->sem_otime);
1741 v.sem_ctime = lower_32_bits(in->sem_ctime);
1742 v.sem_ctime_high = upper_32_bits(in->sem_ctime);
1743 v.sem_nsems = in->sem_nsems;
1744 return copy_to_user(buf, &v, sizeof(v));
1745 } else {
1746 struct compat_semid_ds v;
1747 memset(&v, 0, sizeof(v));
1748 to_compat_ipc_perm(&v.sem_perm, &in->sem_perm);
1749 v.sem_otime = in->sem_otime;
1750 v.sem_ctime = in->sem_ctime;
1751 v.sem_nsems = in->sem_nsems;
1752 return copy_to_user(buf, &v, sizeof(v));
1753 }
1754}
1755
1756static long compat_ksys_semctl(int semid, int semnum, int cmd, int arg, int version)
1757{
1758 void __user *p = compat_ptr(arg);
1759 struct ipc_namespace *ns;
1760 struct semid64_ds semid64;
1761 int err;
1762
1763 ns = current->nsproxy->ipc_ns;
1764
1765 if (semid < 0)
1766 return -EINVAL;
1767
1768 switch (cmd & (~IPC_64)) {
1769 case IPC_INFO:
1770 case SEM_INFO:
1771 return semctl_info(ns, semid, cmd, p);
1772 case IPC_STAT:
1773 case SEM_STAT:
1774 case SEM_STAT_ANY:
1775 err = semctl_stat(ns, semid, cmd, &semid64);
1776 if (err < 0)
1777 return err;
1778 if (copy_compat_semid_to_user(p, &semid64, version))
1779 err = -EFAULT;
1780 return err;
1781 case GETVAL:
1782 case GETPID:
1783 case GETNCNT:
1784 case GETZCNT:
1785 case GETALL:
1786 case SETALL:
1787 return semctl_main(ns, semid, semnum, cmd, p);
1788 case SETVAL:
1789 return semctl_setval(ns, semid, semnum, arg);
1790 case IPC_SET:
1791 if (copy_compat_semid_from_user(&semid64, p, version))
1792 return -EFAULT;
1793 /* fallthru */
1794 case IPC_RMID:
1795 return semctl_down(ns, semid, cmd, &semid64);
1796 default:
1797 return -EINVAL;
1798 }
1799}
1800
1801COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg)
1802{
1803 return compat_ksys_semctl(semid, semnum, cmd, arg, IPC_64);
1804}
1805
1806#ifdef CONFIG_ARCH_WANT_COMPAT_IPC_PARSE_VERSION
1807long compat_ksys_old_semctl(int semid, int semnum, int cmd, int arg)
1808{
1809 int version = compat_ipc_parse_version(&cmd);
1810
1811 return compat_ksys_semctl(semid, semnum, cmd, arg, version);
1812}
1813
1814COMPAT_SYSCALL_DEFINE4(old_semctl, int, semid, int, semnum, int, cmd, int, arg)
1815{
1816 return compat_ksys_old_semctl(semid, semnum, cmd, arg);
1817}
1818#endif
1819#endif
1820
1821/* If the task doesn't already have a undo_list, then allocate one
1822 * here. We guarantee there is only one thread using this undo list,
1823 * and current is THE ONE
1824 *
1825 * If this allocation and assignment succeeds, but later
1826 * portions of this code fail, there is no need to free the sem_undo_list.
1827 * Just let it stay associated with the task, and it'll be freed later
1828 * at exit time.
1829 *
1830 * This can block, so callers must hold no locks.
1831 */
1832static inline int get_undo_list(struct sem_undo_list **undo_listp)
1833{
1834 struct sem_undo_list *undo_list;
1835
1836 undo_list = current->sysvsem.undo_list;
1837 if (!undo_list) {
1838 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1839 if (undo_list == NULL)
1840 return -ENOMEM;
1841 spin_lock_init(&undo_list->lock);
1842 refcount_set(&undo_list->refcnt, 1);
1843 INIT_LIST_HEAD(&undo_list->list_proc);
1844
1845 current->sysvsem.undo_list = undo_list;
1846 }
1847 *undo_listp = undo_list;
1848 return 0;
1849}
1850
1851static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1852{
1853 struct sem_undo *un;
1854
1855 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc,
1856 spin_is_locked(&ulp->lock)) {
1857 if (un->semid == semid)
1858 return un;
1859 }
1860 return NULL;
1861}
1862
1863static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1864{
1865 struct sem_undo *un;
1866
1867 assert_spin_locked(&ulp->lock);
1868
1869 un = __lookup_undo(ulp, semid);
1870 if (un) {
1871 list_del_rcu(&un->list_proc);
1872 list_add_rcu(&un->list_proc, &ulp->list_proc);
1873 }
1874 return un;
1875}
1876
1877/**
1878 * find_alloc_undo - lookup (and if not present create) undo array
1879 * @ns: namespace
1880 * @semid: semaphore array id
1881 *
1882 * The function looks up (and if not present creates) the undo structure.
1883 * The size of the undo structure depends on the size of the semaphore
1884 * array, thus the alloc path is not that straightforward.
1885 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1886 * performs a rcu_read_lock().
1887 */
1888static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1889{
1890 struct sem_array *sma;
1891 struct sem_undo_list *ulp;
1892 struct sem_undo *un, *new;
1893 int nsems, error;
1894
1895 error = get_undo_list(&ulp);
1896 if (error)
1897 return ERR_PTR(error);
1898
1899 rcu_read_lock();
1900 spin_lock(&ulp->lock);
1901 un = lookup_undo(ulp, semid);
1902 spin_unlock(&ulp->lock);
1903 if (likely(un != NULL))
1904 goto out;
1905
1906 /* no undo structure around - allocate one. */
1907 /* step 1: figure out the size of the semaphore array */
1908 sma = sem_obtain_object_check(ns, semid);
1909 if (IS_ERR(sma)) {
1910 rcu_read_unlock();
1911 return ERR_CAST(sma);
1912 }
1913
1914 nsems = sma->sem_nsems;
1915 if (!ipc_rcu_getref(&sma->sem_perm)) {
1916 rcu_read_unlock();
1917 un = ERR_PTR(-EIDRM);
1918 goto out;
1919 }
1920 rcu_read_unlock();
1921
1922 /* step 2: allocate new undo structure */
1923 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1924 if (!new) {
1925 ipc_rcu_putref(&sma->sem_perm, sem_rcu_free);
1926 return ERR_PTR(-ENOMEM);
1927 }
1928
1929 /* step 3: Acquire the lock on semaphore array */
1930 rcu_read_lock();
1931 sem_lock_and_putref(sma);
1932 if (!ipc_valid_object(&sma->sem_perm)) {
1933 sem_unlock(sma, -1);
1934 rcu_read_unlock();
1935 kfree(new);
1936 un = ERR_PTR(-EIDRM);
1937 goto out;
1938 }
1939 spin_lock(&ulp->lock);
1940
1941 /*
1942 * step 4: check for races: did someone else allocate the undo struct?
1943 */
1944 un = lookup_undo(ulp, semid);
1945 if (un) {
1946 kfree(new);
1947 goto success;
1948 }
1949 /* step 5: initialize & link new undo structure */
1950 new->semadj = (short *) &new[1];
1951 new->ulp = ulp;
1952 new->semid = semid;
1953 assert_spin_locked(&ulp->lock);
1954 list_add_rcu(&new->list_proc, &ulp->list_proc);
1955 ipc_assert_locked_object(&sma->sem_perm);
1956 list_add(&new->list_id, &sma->list_id);
1957 un = new;
1958
1959success:
1960 spin_unlock(&ulp->lock);
1961 sem_unlock(sma, -1);
1962out:
1963 return un;
1964}
1965
1966static long do_semtimedop(int semid, struct sembuf __user *tsops,
1967 unsigned nsops, const struct timespec64 *timeout)
1968{
1969 int error = -EINVAL;
1970 struct sem_array *sma;
1971 struct sembuf fast_sops[SEMOPM_FAST];
1972 struct sembuf *sops = fast_sops, *sop;
1973 struct sem_undo *un;
1974 int max, locknum;
1975 bool undos = false, alter = false, dupsop = false;
1976 struct sem_queue queue;
1977 unsigned long dup = 0, jiffies_left = 0;
1978 struct ipc_namespace *ns;
1979
1980 ns = current->nsproxy->ipc_ns;
1981
1982 if (nsops < 1 || semid < 0)
1983 return -EINVAL;
1984 if (nsops > ns->sc_semopm)
1985 return -E2BIG;
1986 if (nsops > SEMOPM_FAST) {
1987 sops = kvmalloc_array(nsops, sizeof(*sops), GFP_KERNEL);
1988 if (sops == NULL)
1989 return -ENOMEM;
1990 }
1991
1992 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1993 error = -EFAULT;
1994 goto out_free;
1995 }
1996
1997 if (timeout) {
1998 if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 ||
1999 timeout->tv_nsec >= 1000000000L) {
2000 error = -EINVAL;
2001 goto out_free;
2002 }
2003 jiffies_left = timespec64_to_jiffies(timeout);
2004 }
2005
2006 max = 0;
2007 for (sop = sops; sop < sops + nsops; sop++) {
2008 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
2009
2010 if (sop->sem_num >= max)
2011 max = sop->sem_num;
2012 if (sop->sem_flg & SEM_UNDO)
2013 undos = true;
2014 if (dup & mask) {
2015 /*
2016 * There was a previous alter access that appears
2017 * to have accessed the same semaphore, thus use
2018 * the dupsop logic. "appears", because the detection
2019 * can only check % BITS_PER_LONG.
2020 */
2021 dupsop = true;
2022 }
2023 if (sop->sem_op != 0) {
2024 alter = true;
2025 dup |= mask;
2026 }
2027 }
2028
2029 if (undos) {
2030 /* On success, find_alloc_undo takes the rcu_read_lock */
2031 un = find_alloc_undo(ns, semid);
2032 if (IS_ERR(un)) {
2033 error = PTR_ERR(un);
2034 goto out_free;
2035 }
2036 } else {
2037 un = NULL;
2038 rcu_read_lock();
2039 }
2040
2041 sma = sem_obtain_object_check(ns, semid);
2042 if (IS_ERR(sma)) {
2043 rcu_read_unlock();
2044 error = PTR_ERR(sma);
2045 goto out_free;
2046 }
2047
2048 error = -EFBIG;
2049 if (max >= sma->sem_nsems) {
2050 rcu_read_unlock();
2051 goto out_free;
2052 }
2053
2054 error = -EACCES;
2055 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
2056 rcu_read_unlock();
2057 goto out_free;
2058 }
2059
2060 error = security_sem_semop(&sma->sem_perm, sops, nsops, alter);
2061 if (error) {
2062 rcu_read_unlock();
2063 goto out_free;
2064 }
2065
2066 error = -EIDRM;
2067 locknum = sem_lock(sma, sops, nsops);
2068 /*
2069 * We eventually might perform the following check in a lockless
2070 * fashion, considering ipc_valid_object() locking constraints.
2071 * If nsops == 1 and there is no contention for sem_perm.lock, then
2072 * only a per-semaphore lock is held and it's OK to proceed with the
2073 * check below. More details on the fine grained locking scheme
2074 * entangled here and why it's RMID race safe on comments at sem_lock()
2075 */
2076 if (!ipc_valid_object(&sma->sem_perm))
2077 goto out_unlock_free;
2078 /*
2079 * semid identifiers are not unique - find_alloc_undo may have
2080 * allocated an undo structure, it was invalidated by an RMID
2081 * and now a new array with received the same id. Check and fail.
2082 * This case can be detected checking un->semid. The existence of
2083 * "un" itself is guaranteed by rcu.
2084 */
2085 if (un && un->semid == -1)
2086 goto out_unlock_free;
2087
2088 queue.sops = sops;
2089 queue.nsops = nsops;
2090 queue.undo = un;
2091 queue.pid = task_tgid(current);
2092 queue.alter = alter;
2093 queue.dupsop = dupsop;
2094
2095 error = perform_atomic_semop(sma, &queue);
2096 if (error == 0) { /* non-blocking succesfull path */
2097 DEFINE_WAKE_Q(wake_q);
2098
2099 /*
2100 * If the operation was successful, then do
2101 * the required updates.
2102 */
2103 if (alter)
2104 do_smart_update(sma, sops, nsops, 1, &wake_q);
2105 else
2106 set_semotime(sma, sops);
2107
2108 sem_unlock(sma, locknum);
2109 rcu_read_unlock();
2110 wake_up_q(&wake_q);
2111
2112 goto out_free;
2113 }
2114 if (error < 0) /* non-blocking error path */
2115 goto out_unlock_free;
2116
2117 /*
2118 * We need to sleep on this operation, so we put the current
2119 * task into the pending queue and go to sleep.
2120 */
2121 if (nsops == 1) {
2122 struct sem *curr;
2123 int idx = array_index_nospec(sops->sem_num, sma->sem_nsems);
2124 curr = &sma->sems[idx];
2125
2126 if (alter) {
2127 if (sma->complex_count) {
2128 list_add_tail(&queue.list,
2129 &sma->pending_alter);
2130 } else {
2131
2132 list_add_tail(&queue.list,
2133 &curr->pending_alter);
2134 }
2135 } else {
2136 list_add_tail(&queue.list, &curr->pending_const);
2137 }
2138 } else {
2139 if (!sma->complex_count)
2140 merge_queues(sma);
2141
2142 if (alter)
2143 list_add_tail(&queue.list, &sma->pending_alter);
2144 else
2145 list_add_tail(&queue.list, &sma->pending_const);
2146
2147 sma->complex_count++;
2148 }
2149
2150 do {
2151 WRITE_ONCE(queue.status, -EINTR);
2152 queue.sleeper = current;
2153
2154 __set_current_state(TASK_INTERRUPTIBLE);
2155 sem_unlock(sma, locknum);
2156 rcu_read_unlock();
2157
2158 if (timeout)
2159 jiffies_left = schedule_timeout(jiffies_left);
2160 else
2161 schedule();
2162
2163 /*
2164 * fastpath: the semop has completed, either successfully or
2165 * not, from the syscall pov, is quite irrelevant to us at this
2166 * point; we're done.
2167 *
2168 * We _do_ care, nonetheless, about being awoken by a signal or
2169 * spuriously. The queue.status is checked again in the
2170 * slowpath (aka after taking sem_lock), such that we can detect
2171 * scenarios where we were awakened externally, during the
2172 * window between wake_q_add() and wake_up_q().
2173 */
2174 error = READ_ONCE(queue.status);
2175 if (error != -EINTR) {
2176 /*
2177 * User space could assume that semop() is a memory
2178 * barrier: Without the mb(), the cpu could
2179 * speculatively read in userspace stale data that was
2180 * overwritten by the previous owner of the semaphore.
2181 */
2182 smp_mb();
2183 goto out_free;
2184 }
2185
2186 rcu_read_lock();
2187 locknum = sem_lock(sma, sops, nsops);
2188
2189 if (!ipc_valid_object(&sma->sem_perm))
2190 goto out_unlock_free;
2191
2192 error = READ_ONCE(queue.status);
2193
2194 /*
2195 * If queue.status != -EINTR we are woken up by another process.
2196 * Leave without unlink_queue(), but with sem_unlock().
2197 */
2198 if (error != -EINTR)
2199 goto out_unlock_free;
2200
2201 /*
2202 * If an interrupt occurred we have to clean up the queue.
2203 */
2204 if (timeout && jiffies_left == 0)
2205 error = -EAGAIN;
2206 } while (error == -EINTR && !signal_pending(current)); /* spurious */
2207
2208 unlink_queue(sma, &queue);
2209
2210out_unlock_free:
2211 sem_unlock(sma, locknum);
2212 rcu_read_unlock();
2213out_free:
2214 if (sops != fast_sops)
2215 kvfree(sops);
2216 return error;
2217}
2218
2219long ksys_semtimedop(int semid, struct sembuf __user *tsops,
2220 unsigned int nsops, const struct __kernel_timespec __user *timeout)
2221{
2222 if (timeout) {
2223 struct timespec64 ts;
2224 if (get_timespec64(&ts, timeout))
2225 return -EFAULT;
2226 return do_semtimedop(semid, tsops, nsops, &ts);
2227 }
2228 return do_semtimedop(semid, tsops, nsops, NULL);
2229}
2230
2231SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
2232 unsigned int, nsops, const struct __kernel_timespec __user *, timeout)
2233{
2234 return ksys_semtimedop(semid, tsops, nsops, timeout);
2235}
2236
2237#ifdef CONFIG_COMPAT_32BIT_TIME
2238long compat_ksys_semtimedop(int semid, struct sembuf __user *tsems,
2239 unsigned int nsops,
2240 const struct old_timespec32 __user *timeout)
2241{
2242 if (timeout) {
2243 struct timespec64 ts;
2244 if (get_old_timespec32(&ts, timeout))
2245 return -EFAULT;
2246 return do_semtimedop(semid, tsems, nsops, &ts);
2247 }
2248 return do_semtimedop(semid, tsems, nsops, NULL);
2249}
2250
2251SYSCALL_DEFINE4(semtimedop_time32, int, semid, struct sembuf __user *, tsems,
2252 unsigned int, nsops,
2253 const struct old_timespec32 __user *, timeout)
2254{
2255 return compat_ksys_semtimedop(semid, tsems, nsops, timeout);
2256}
2257#endif
2258
2259SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2260 unsigned, nsops)
2261{
2262 return do_semtimedop(semid, tsops, nsops, NULL);
2263}
2264
2265/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2266 * parent and child tasks.
2267 */
2268
2269int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2270{
2271 struct sem_undo_list *undo_list;
2272 int error;
2273
2274 if (clone_flags & CLONE_SYSVSEM) {
2275 error = get_undo_list(&undo_list);
2276 if (error)
2277 return error;
2278 refcount_inc(&undo_list->refcnt);
2279 tsk->sysvsem.undo_list = undo_list;
2280 } else
2281 tsk->sysvsem.undo_list = NULL;
2282
2283 return 0;
2284}
2285
2286/*
2287 * add semadj values to semaphores, free undo structures.
2288 * undo structures are not freed when semaphore arrays are destroyed
2289 * so some of them may be out of date.
2290 * IMPLEMENTATION NOTE: There is some confusion over whether the
2291 * set of adjustments that needs to be done should be done in an atomic
2292 * manner or not. That is, if we are attempting to decrement the semval
2293 * should we queue up and wait until we can do so legally?
2294 * The original implementation attempted to do this (queue and wait).
2295 * The current implementation does not do so. The POSIX standard
2296 * and SVID should be consulted to determine what behavior is mandated.
2297 */
2298void exit_sem(struct task_struct *tsk)
2299{
2300 struct sem_undo_list *ulp;
2301
2302 ulp = tsk->sysvsem.undo_list;
2303 if (!ulp)
2304 return;
2305 tsk->sysvsem.undo_list = NULL;
2306
2307 if (!refcount_dec_and_test(&ulp->refcnt))
2308 return;
2309
2310 for (;;) {
2311 struct sem_array *sma;
2312 struct sem_undo *un;
2313 int semid, i;
2314 DEFINE_WAKE_Q(wake_q);
2315
2316 cond_resched();
2317
2318 rcu_read_lock();
2319 un = list_entry_rcu(ulp->list_proc.next,
2320 struct sem_undo, list_proc);
2321 if (&un->list_proc == &ulp->list_proc) {
2322 /*
2323 * We must wait for freeary() before freeing this ulp,
2324 * in case we raced with last sem_undo. There is a small
2325 * possibility where we exit while freeary() didn't
2326 * finish unlocking sem_undo_list.
2327 */
2328 spin_lock(&ulp->lock);
2329 spin_unlock(&ulp->lock);
2330 rcu_read_unlock();
2331 break;
2332 }
2333 spin_lock(&ulp->lock);
2334 semid = un->semid;
2335 spin_unlock(&ulp->lock);
2336
2337 /* exit_sem raced with IPC_RMID, nothing to do */
2338 if (semid == -1) {
2339 rcu_read_unlock();
2340 continue;
2341 }
2342
2343 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2344 /* exit_sem raced with IPC_RMID, nothing to do */
2345 if (IS_ERR(sma)) {
2346 rcu_read_unlock();
2347 continue;
2348 }
2349
2350 sem_lock(sma, NULL, -1);
2351 /* exit_sem raced with IPC_RMID, nothing to do */
2352 if (!ipc_valid_object(&sma->sem_perm)) {
2353 sem_unlock(sma, -1);
2354 rcu_read_unlock();
2355 continue;
2356 }
2357 un = __lookup_undo(ulp, semid);
2358 if (un == NULL) {
2359 /* exit_sem raced with IPC_RMID+semget() that created
2360 * exactly the same semid. Nothing to do.
2361 */
2362 sem_unlock(sma, -1);
2363 rcu_read_unlock();
2364 continue;
2365 }
2366
2367 /* remove un from the linked lists */
2368 ipc_assert_locked_object(&sma->sem_perm);
2369 list_del(&un->list_id);
2370
2371 /* we are the last process using this ulp, acquiring ulp->lock
2372 * isn't required. Besides that, we are also protected against
2373 * IPC_RMID as we hold sma->sem_perm lock now
2374 */
2375 list_del_rcu(&un->list_proc);
2376
2377 /* perform adjustments registered in un */
2378 for (i = 0; i < sma->sem_nsems; i++) {
2379 struct sem *semaphore = &sma->sems[i];
2380 if (un->semadj[i]) {
2381 semaphore->semval += un->semadj[i];
2382 /*
2383 * Range checks of the new semaphore value,
2384 * not defined by sus:
2385 * - Some unices ignore the undo entirely
2386 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2387 * - some cap the value (e.g. FreeBSD caps
2388 * at 0, but doesn't enforce SEMVMX)
2389 *
2390 * Linux caps the semaphore value, both at 0
2391 * and at SEMVMX.
2392 *
2393 * Manfred <manfred@colorfullife.com>
2394 */
2395 if (semaphore->semval < 0)
2396 semaphore->semval = 0;
2397 if (semaphore->semval > SEMVMX)
2398 semaphore->semval = SEMVMX;
2399 ipc_update_pid(&semaphore->sempid, task_tgid(current));
2400 }
2401 }
2402 /* maybe some queued-up processes were waiting for this */
2403 do_smart_update(sma, NULL, 0, 1, &wake_q);
2404 sem_unlock(sma, -1);
2405 rcu_read_unlock();
2406 wake_up_q(&wake_q);
2407
2408 kfree_rcu(un, rcu);
2409 }
2410 kfree(ulp);
2411}
2412
2413#ifdef CONFIG_PROC_FS
2414static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2415{
2416 struct user_namespace *user_ns = seq_user_ns(s);
2417 struct kern_ipc_perm *ipcp = it;
2418 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
2419 time64_t sem_otime;
2420
2421 /*
2422 * The proc interface isn't aware of sem_lock(), it calls
2423 * ipc_lock_object() directly (in sysvipc_find_ipc).
2424 * In order to stay compatible with sem_lock(), we must
2425 * enter / leave complex_mode.
2426 */
2427 complexmode_enter(sma);
2428
2429 sem_otime = get_semotime(sma);
2430
2431 seq_printf(s,
2432 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2433 sma->sem_perm.key,
2434 sma->sem_perm.id,
2435 sma->sem_perm.mode,
2436 sma->sem_nsems,
2437 from_kuid_munged(user_ns, sma->sem_perm.uid),
2438 from_kgid_munged(user_ns, sma->sem_perm.gid),
2439 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2440 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2441 sem_otime,
2442 sma->sem_ctime);
2443
2444 complexmode_tryleave(sma);
2445
2446 return 0;
2447}
2448#endif