3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
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
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
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
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.
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
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.
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>
90 #include <asm/uaccess.h>
93 /* One semaphore structure for each semaphore in the system. */
95 int semval
; /* current value */
96 int sempid
; /* pid of last operation */
97 spinlock_t lock
; /* spinlock for fine-grained semtimedop */
98 struct list_head pending_alter
; /* pending single-sop operations */
99 /* that alter the semaphore */
100 struct list_head pending_const
; /* pending single-sop operations */
101 /* that do not alter the semaphore*/
102 time_t sem_otime
; /* candidate for sem_otime */
103 } ____cacheline_aligned_in_smp
;
105 /* One queue for each sleeping process in the system. */
107 struct list_head list
; /* queue of pending operations */
108 struct task_struct
*sleeper
; /* this process */
109 struct sem_undo
*undo
; /* undo structure */
110 int pid
; /* process id of requesting process */
111 int status
; /* completion status of operation */
112 struct sembuf
*sops
; /* array of pending operations */
113 int nsops
; /* number of operations */
114 int alter
; /* does *sops alter the array? */
117 /* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
121 struct list_head list_proc
; /* per-process list: *
122 * all undos from one process
124 struct rcu_head rcu
; /* rcu struct for sem_undo */
125 struct sem_undo_list
*ulp
; /* back ptr to sem_undo_list */
126 struct list_head list_id
; /* per semaphore array list:
127 * all undos for one array */
128 int semid
; /* semaphore set identifier */
129 short *semadj
; /* array of adjustments */
130 /* one per semaphore */
133 /* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
136 struct sem_undo_list
{
139 struct list_head list_proc
;
143 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
145 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
147 static int newary(struct ipc_namespace
*, struct ipc_params
*);
148 static void freeary(struct ipc_namespace
*, struct kern_ipc_perm
*);
149 #ifdef CONFIG_PROC_FS
150 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
);
153 #define SEMMSL_FAST 256 /* 512 bytes on stack */
154 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
157 * linked list protection:
159 * sem_array.pending{_alter,_cont},
160 * sem_array.sem_undo: sem_lock() for read/write
161 * sem_undo.proc_next: only "current" is allowed to read/write that field.
165 #define sc_semmsl sem_ctls[0]
166 #define sc_semmns sem_ctls[1]
167 #define sc_semopm sem_ctls[2]
168 #define sc_semmni sem_ctls[3]
170 void sem_init_ns(struct ipc_namespace
*ns
)
172 ns
->sc_semmsl
= SEMMSL
;
173 ns
->sc_semmns
= SEMMNS
;
174 ns
->sc_semopm
= SEMOPM
;
175 ns
->sc_semmni
= SEMMNI
;
177 ipc_init_ids(&ns
->ids
[IPC_SEM_IDS
]);
181 void sem_exit_ns(struct ipc_namespace
*ns
)
183 free_ipcs(ns
, &sem_ids(ns
), freeary
);
184 idr_destroy(&ns
->ids
[IPC_SEM_IDS
].ipcs_idr
);
188 void __init
sem_init (void)
190 sem_init_ns(&init_ipc_ns
);
191 ipc_init_proc_interface("sysvipc/sem",
192 " key semid perms nsems uid gid cuid cgid otime ctime\n",
193 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
197 * unmerge_queues - unmerge queues, if possible.
198 * @sma: semaphore array
200 * The function unmerges the wait queues if complex_count is 0.
201 * It must be called prior to dropping the global semaphore array lock.
203 static void unmerge_queues(struct sem_array
*sma
)
205 struct sem_queue
*q
, *tq
;
207 /* complex operations still around? */
208 if (sma
->complex_count
)
211 * We will switch back to simple mode.
212 * Move all pending operation back into the per-semaphore
215 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
217 curr
= &sma
->sem_base
[q
->sops
[0].sem_num
];
219 list_add_tail(&q
->list
, &curr
->pending_alter
);
221 INIT_LIST_HEAD(&sma
->pending_alter
);
225 * merge_queues - Merge single semop queues into global queue
226 * @sma: semaphore array
228 * This function merges all per-semaphore queues into the global queue.
229 * It is necessary to achieve FIFO ordering for the pending single-sop
230 * operations when a multi-semop operation must sleep.
231 * Only the alter operations must be moved, the const operations can stay.
233 static void merge_queues(struct sem_array
*sma
)
236 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
237 struct sem
*sem
= sma
->sem_base
+ i
;
239 list_splice_init(&sem
->pending_alter
, &sma
->pending_alter
);
244 * If the request contains only one semaphore operation, and there are
245 * no complex transactions pending, lock only the semaphore involved.
246 * Otherwise, lock the entire semaphore array, since we either have
247 * multiple semaphores in our own semops, or we need to look at
248 * semaphores from other pending complex operations.
250 * Carefully guard against sma->complex_count changing between zero
251 * and non-zero while we are spinning for the lock. The value of
252 * sma->complex_count cannot change while we are holding the lock,
253 * so sem_unlock should be fine.
255 * The global lock path checks that all the local locks have been released,
256 * checking each local lock once. This means that the local lock paths
257 * cannot start their critical sections while the global lock is held.
259 static inline int sem_lock(struct sem_array
*sma
, struct sembuf
*sops
,
264 if (nsops
== 1 && !sma
->complex_count
) {
265 struct sem
*sem
= sma
->sem_base
+ sops
->sem_num
;
267 /* Lock just the semaphore we are interested in. */
268 spin_lock(&sem
->lock
);
271 * If sma->complex_count was set while we were spinning,
272 * we may need to look at things we did not lock here.
274 if (unlikely(sma
->complex_count
)) {
275 spin_unlock(&sem
->lock
);
280 * Another process is holding the global lock on the
281 * sem_array; we cannot enter our critical section,
282 * but have to wait for the global lock to be released.
284 if (unlikely(spin_is_locked(&sma
->sem_perm
.lock
))) {
285 spin_unlock(&sem
->lock
);
286 spin_unlock_wait(&sma
->sem_perm
.lock
);
290 locknum
= sops
->sem_num
;
294 * Lock the semaphore array, and wait for all of the
295 * individual semaphore locks to go away. The code
296 * above ensures no new single-lock holders will enter
297 * their critical section while the array lock is held.
300 ipc_lock_object(&sma
->sem_perm
);
301 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
302 struct sem
*sem
= sma
->sem_base
+ i
;
303 spin_unlock_wait(&sem
->lock
);
310 static inline void sem_unlock(struct sem_array
*sma
, int locknum
)
314 ipc_unlock_object(&sma
->sem_perm
);
316 struct sem
*sem
= sma
->sem_base
+ locknum
;
317 spin_unlock(&sem
->lock
);
322 * sem_lock_(check_) routines are called in the paths where the rw_mutex
325 * The caller holds the RCU read lock.
327 static inline struct sem_array
*sem_obtain_lock(struct ipc_namespace
*ns
,
328 int id
, struct sembuf
*sops
, int nsops
, int *locknum
)
330 struct kern_ipc_perm
*ipcp
;
331 struct sem_array
*sma
;
333 ipcp
= ipc_obtain_object(&sem_ids(ns
), id
);
335 return ERR_CAST(ipcp
);
337 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
338 *locknum
= sem_lock(sma
, sops
, nsops
);
340 /* ipc_rmid() may have already freed the ID while sem_lock
341 * was spinning: verify that the structure is still valid
344 return container_of(ipcp
, struct sem_array
, sem_perm
);
346 sem_unlock(sma
, *locknum
);
347 return ERR_PTR(-EINVAL
);
350 static inline struct sem_array
*sem_obtain_object(struct ipc_namespace
*ns
, int id
)
352 struct kern_ipc_perm
*ipcp
= ipc_obtain_object(&sem_ids(ns
), id
);
355 return ERR_CAST(ipcp
);
357 return container_of(ipcp
, struct sem_array
, sem_perm
);
360 static inline struct sem_array
*sem_obtain_object_check(struct ipc_namespace
*ns
,
363 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&sem_ids(ns
), id
);
366 return ERR_CAST(ipcp
);
368 return container_of(ipcp
, struct sem_array
, sem_perm
);
371 static inline void sem_lock_and_putref(struct sem_array
*sma
)
373 sem_lock(sma
, NULL
, -1);
377 static inline void sem_putref(struct sem_array
*sma
)
382 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
384 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
388 * Lockless wakeup algorithm:
389 * Without the check/retry algorithm a lockless wakeup is possible:
390 * - queue.status is initialized to -EINTR before blocking.
391 * - wakeup is performed by
392 * * unlinking the queue entry from the pending list
393 * * setting queue.status to IN_WAKEUP
394 * This is the notification for the blocked thread that a
395 * result value is imminent.
396 * * call wake_up_process
397 * * set queue.status to the final value.
398 * - the previously blocked thread checks queue.status:
399 * * if it's IN_WAKEUP, then it must wait until the value changes
400 * * if it's not -EINTR, then the operation was completed by
401 * update_queue. semtimedop can return queue.status without
402 * performing any operation on the sem array.
403 * * otherwise it must acquire the spinlock and check what's up.
405 * The two-stage algorithm is necessary to protect against the following
407 * - if queue.status is set after wake_up_process, then the woken up idle
408 * thread could race forward and try (and fail) to acquire sma->lock
409 * before update_queue had a chance to set queue.status
410 * - if queue.status is written before wake_up_process and if the
411 * blocked process is woken up by a signal between writing
412 * queue.status and the wake_up_process, then the woken up
413 * process could return from semtimedop and die by calling
414 * sys_exit before wake_up_process is called. Then wake_up_process
415 * will oops, because the task structure is already invalid.
416 * (yes, this happened on s390 with sysv msg).
422 * newary - Create a new semaphore set
424 * @params: ptr to the structure that contains key, semflg and nsems
426 * Called with sem_ids.rw_mutex held (as a writer)
429 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
433 struct sem_array
*sma
;
435 key_t key
= params
->key
;
436 int nsems
= params
->u
.nsems
;
437 int semflg
= params
->flg
;
442 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
445 size
= sizeof (*sma
) + nsems
* sizeof (struct sem
);
446 sma
= ipc_rcu_alloc(size
);
450 memset (sma
, 0, size
);
452 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
453 sma
->sem_perm
.key
= key
;
455 sma
->sem_perm
.security
= NULL
;
456 retval
= security_sem_alloc(sma
);
462 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
464 security_sem_free(sma
);
468 ns
->used_sems
+= nsems
;
470 sma
->sem_base
= (struct sem
*) &sma
[1];
472 for (i
= 0; i
< nsems
; i
++) {
473 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_alter
);
474 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_const
);
475 spin_lock_init(&sma
->sem_base
[i
].lock
);
478 sma
->complex_count
= 0;
479 INIT_LIST_HEAD(&sma
->pending_alter
);
480 INIT_LIST_HEAD(&sma
->pending_const
);
481 INIT_LIST_HEAD(&sma
->list_id
);
482 sma
->sem_nsems
= nsems
;
483 sma
->sem_ctime
= get_seconds();
487 return sma
->sem_perm
.id
;
492 * Called with sem_ids.rw_mutex and ipcp locked.
494 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
496 struct sem_array
*sma
;
498 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
499 return security_sem_associate(sma
, semflg
);
503 * Called with sem_ids.rw_mutex and ipcp locked.
505 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
506 struct ipc_params
*params
)
508 struct sem_array
*sma
;
510 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
511 if (params
->u
.nsems
> sma
->sem_nsems
)
517 SYSCALL_DEFINE3(semget
, key_t
, key
, int, nsems
, int, semflg
)
519 struct ipc_namespace
*ns
;
520 struct ipc_ops sem_ops
;
521 struct ipc_params sem_params
;
523 ns
= current
->nsproxy
->ipc_ns
;
525 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
528 sem_ops
.getnew
= newary
;
529 sem_ops
.associate
= sem_security
;
530 sem_ops
.more_checks
= sem_more_checks
;
532 sem_params
.key
= key
;
533 sem_params
.flg
= semflg
;
534 sem_params
.u
.nsems
= nsems
;
536 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
540 * Determine whether a sequence of semaphore operations would succeed
541 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
544 static int try_atomic_semop (struct sem_array
* sma
, struct sembuf
* sops
,
545 int nsops
, struct sem_undo
*un
, int pid
)
551 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
552 curr
= sma
->sem_base
+ sop
->sem_num
;
553 sem_op
= sop
->sem_op
;
554 result
= curr
->semval
;
556 if (!sem_op
&& result
)
564 if (sop
->sem_flg
& SEM_UNDO
) {
565 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
567 * Exceeding the undo range is an error.
569 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
572 curr
->semval
= result
;
576 while (sop
>= sops
) {
577 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
578 if (sop
->sem_flg
& SEM_UNDO
)
579 un
->semadj
[sop
->sem_num
] -= sop
->sem_op
;
590 if (sop
->sem_flg
& IPC_NOWAIT
)
597 while (sop
>= sops
) {
598 sma
->sem_base
[sop
->sem_num
].semval
-= sop
->sem_op
;
605 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
606 * @q: queue entry that must be signaled
607 * @error: Error value for the signal
609 * Prepare the wake-up of the queue entry q.
611 static void wake_up_sem_queue_prepare(struct list_head
*pt
,
612 struct sem_queue
*q
, int error
)
614 if (list_empty(pt
)) {
616 * Hold preempt off so that we don't get preempted and have the
617 * wakee busy-wait until we're scheduled back on.
621 q
->status
= IN_WAKEUP
;
624 list_add_tail(&q
->list
, pt
);
628 * wake_up_sem_queue_do(pt) - do the actual wake-up
629 * @pt: list of tasks to be woken up
631 * Do the actual wake-up.
632 * The function is called without any locks held, thus the semaphore array
633 * could be destroyed already and the tasks can disappear as soon as the
634 * status is set to the actual return code.
636 static void wake_up_sem_queue_do(struct list_head
*pt
)
638 struct sem_queue
*q
, *t
;
641 did_something
= !list_empty(pt
);
642 list_for_each_entry_safe(q
, t
, pt
, list
) {
643 wake_up_process(q
->sleeper
);
644 /* q can disappear immediately after writing q->status. */
652 static void unlink_queue(struct sem_array
*sma
, struct sem_queue
*q
)
656 sma
->complex_count
--;
659 /** check_restart(sma, q)
660 * @sma: semaphore array
661 * @q: the operation that just completed
663 * update_queue is O(N^2) when it restarts scanning the whole queue of
664 * waiting operations. Therefore this function checks if the restart is
665 * really necessary. It is called after a previously waiting operation
666 * modified the array.
667 * Note that wait-for-zero operations are handled without restart.
669 static int check_restart(struct sem_array
*sma
, struct sem_queue
*q
)
671 /* pending complex alter operations are too difficult to analyse */
672 if (!list_empty(&sma
->pending_alter
))
675 /* we were a sleeping complex operation. Too difficult */
679 /* It is impossible that someone waits for the new value:
680 * - complex operations always restart.
681 * - wait-for-zero are handled seperately.
682 * - q is a previously sleeping simple operation that
683 * altered the array. It must be a decrement, because
684 * simple increments never sleep.
685 * - If there are older (higher priority) decrements
686 * in the queue, then they have observed the original
687 * semval value and couldn't proceed. The operation
688 * decremented to value - thus they won't proceed either.
694 * wake_const_ops(sma, semnum, pt) - Wake up non-alter tasks
695 * @sma: semaphore array.
696 * @semnum: semaphore that was modified.
697 * @pt: list head for the tasks that must be woken up.
699 * wake_const_ops must be called after a semaphore in a semaphore array
700 * was set to 0. If complex const operations are pending, wake_const_ops must
701 * be called with semnum = -1, as well as with the number of each modified
703 * The tasks that must be woken up are added to @pt. The return code
704 * is stored in q->pid.
705 * The function returns 1 if at least one operation was completed successfully.
707 static int wake_const_ops(struct sem_array
*sma
, int semnum
,
708 struct list_head
*pt
)
711 struct list_head
*walk
;
712 struct list_head
*pending_list
;
713 int semop_completed
= 0;
716 pending_list
= &sma
->pending_const
;
718 pending_list
= &sma
->sem_base
[semnum
].pending_const
;
720 walk
= pending_list
->next
;
721 while (walk
!= pending_list
) {
724 q
= container_of(walk
, struct sem_queue
, list
);
727 error
= try_atomic_semop(sma
, q
->sops
, q
->nsops
,
731 /* operation completed, remove from queue & wakeup */
733 unlink_queue(sma
, q
);
735 wake_up_sem_queue_prepare(pt
, q
, error
);
740 return semop_completed
;
744 * do_smart_wakeup_zero(sma, sops, nsops, pt) - wakeup all wait for zero tasks
745 * @sma: semaphore array
746 * @sops: operations that were performed
747 * @nsops: number of operations
748 * @pt: list head of the tasks that must be woken up.
750 * do_smart_wakeup_zero() checks all required queue for wait-for-zero
751 * operations, based on the actual changes that were performed on the
753 * The function returns 1 if at least one operation was completed successfully.
755 static int do_smart_wakeup_zero(struct sem_array
*sma
, struct sembuf
*sops
,
756 int nsops
, struct list_head
*pt
)
759 int semop_completed
= 0;
762 /* first: the per-semaphore queues, if known */
764 for (i
= 0; i
< nsops
; i
++) {
765 int num
= sops
[i
].sem_num
;
767 if (sma
->sem_base
[num
].semval
== 0) {
769 semop_completed
|= wake_const_ops(sma
, num
, pt
);
774 * No sops means modified semaphores not known.
775 * Assume all were changed.
777 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
778 if (sma
->sem_base
[i
].semval
== 0) {
780 semop_completed
|= wake_const_ops(sma
, i
, pt
);
785 * If one of the modified semaphores got 0,
786 * then check the global queue, too.
789 semop_completed
|= wake_const_ops(sma
, -1, pt
);
791 return semop_completed
;
796 * update_queue(sma, semnum): Look for tasks that can be completed.
797 * @sma: semaphore array.
798 * @semnum: semaphore that was modified.
799 * @pt: list head for the tasks that must be woken up.
801 * update_queue must be called after a semaphore in a semaphore array
802 * was modified. If multiple semaphores were modified, update_queue must
803 * be called with semnum = -1, as well as with the number of each modified
805 * The tasks that must be woken up are added to @pt. The return code
806 * is stored in q->pid.
807 * The function internally checks if const operations can now succeed.
809 * The function return 1 if at least one semop was completed successfully.
811 static int update_queue(struct sem_array
*sma
, int semnum
, struct list_head
*pt
)
814 struct list_head
*walk
;
815 struct list_head
*pending_list
;
816 int semop_completed
= 0;
819 pending_list
= &sma
->pending_alter
;
821 pending_list
= &sma
->sem_base
[semnum
].pending_alter
;
824 walk
= pending_list
->next
;
825 while (walk
!= pending_list
) {
828 q
= container_of(walk
, struct sem_queue
, list
);
831 /* If we are scanning the single sop, per-semaphore list of
832 * one semaphore and that semaphore is 0, then it is not
833 * necessary to scan further: simple increments
834 * that affect only one entry succeed immediately and cannot
835 * be in the per semaphore pending queue, and decrements
836 * cannot be successful if the value is already 0.
838 if (semnum
!= -1 && sma
->sem_base
[semnum
].semval
== 0)
841 error
= try_atomic_semop(sma
, q
->sops
, q
->nsops
,
844 /* Does q->sleeper still need to sleep? */
848 unlink_queue(sma
, q
);
854 do_smart_wakeup_zero(sma
, q
->sops
, q
->nsops
, pt
);
855 restart
= check_restart(sma
, q
);
858 wake_up_sem_queue_prepare(pt
, q
, error
);
862 return semop_completed
;
866 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
867 * @sma: semaphore array
868 * @sops: operations that were performed
869 * @nsops: number of operations
870 * @otime: force setting otime
871 * @pt: list head of the tasks that must be woken up.
873 * do_smart_update() does the required calls to update_queue and wakeup_zero,
874 * based on the actual changes that were performed on the semaphore array.
875 * Note that the function does not do the actual wake-up: the caller is
876 * responsible for calling wake_up_sem_queue_do(@pt).
877 * It is safe to perform this call after dropping all locks.
879 static void do_smart_update(struct sem_array
*sma
, struct sembuf
*sops
, int nsops
,
880 int otime
, struct list_head
*pt
)
884 otime
|= do_smart_wakeup_zero(sma
, sops
, nsops
, pt
);
886 if (!list_empty(&sma
->pending_alter
)) {
887 /* semaphore array uses the global queue - just process it. */
888 otime
|= update_queue(sma
, -1, pt
);
892 * No sops, thus the modified semaphores are not
895 for (i
= 0; i
< sma
->sem_nsems
; i
++)
896 otime
|= update_queue(sma
, i
, pt
);
899 * Check the semaphores that were increased:
900 * - No complex ops, thus all sleeping ops are
902 * - if we decreased the value, then any sleeping
903 * semaphore ops wont be able to run: If the
904 * previous value was too small, then the new
905 * value will be too small, too.
907 for (i
= 0; i
< nsops
; i
++) {
908 if (sops
[i
].sem_op
> 0) {
909 otime
|= update_queue(sma
,
910 sops
[i
].sem_num
, pt
);
917 sma
->sem_base
[0].sem_otime
= get_seconds();
919 sma
->sem_base
[sops
[0].sem_num
].sem_otime
=
926 /* The following counts are associated to each semaphore:
927 * semncnt number of tasks waiting on semval being nonzero
928 * semzcnt number of tasks waiting on semval being zero
929 * This model assumes that a task waits on exactly one semaphore.
930 * Since semaphore operations are to be performed atomically, tasks actually
931 * wait on a whole sequence of semaphores simultaneously.
932 * The counts we return here are a rough approximation, but still
933 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
935 static int count_semncnt (struct sem_array
* sma
, ushort semnum
)
938 struct sem_queue
* q
;
941 list_for_each_entry(q
, &sma
->sem_base
[semnum
].pending_alter
, list
) {
942 struct sembuf
* sops
= q
->sops
;
943 BUG_ON(sops
->sem_num
!= semnum
);
944 if ((sops
->sem_op
< 0) && !(sops
->sem_flg
& IPC_NOWAIT
))
948 list_for_each_entry(q
, &sma
->pending_alter
, list
) {
949 struct sembuf
* sops
= q
->sops
;
950 int nsops
= q
->nsops
;
952 for (i
= 0; i
< nsops
; i
++)
953 if (sops
[i
].sem_num
== semnum
954 && (sops
[i
].sem_op
< 0)
955 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
961 static int count_semzcnt (struct sem_array
* sma
, ushort semnum
)
964 struct sem_queue
* q
;
967 list_for_each_entry(q
, &sma
->sem_base
[semnum
].pending_const
, list
) {
968 struct sembuf
* sops
= q
->sops
;
969 BUG_ON(sops
->sem_num
!= semnum
);
970 if ((sops
->sem_op
== 0) && !(sops
->sem_flg
& IPC_NOWAIT
))
974 list_for_each_entry(q
, &sma
->pending_const
, list
) {
975 struct sembuf
* sops
= q
->sops
;
976 int nsops
= q
->nsops
;
978 for (i
= 0; i
< nsops
; i
++)
979 if (sops
[i
].sem_num
== semnum
980 && (sops
[i
].sem_op
== 0)
981 && !(sops
[i
].sem_flg
& IPC_NOWAIT
))
987 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
988 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
989 * remains locked on exit.
991 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
993 struct sem_undo
*un
, *tu
;
994 struct sem_queue
*q
, *tq
;
995 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
996 struct list_head tasks
;
999 /* Free the existing undo structures for this semaphore set. */
1000 ipc_assert_locked_object(&sma
->sem_perm
);
1001 list_for_each_entry_safe(un
, tu
, &sma
->list_id
, list_id
) {
1002 list_del(&un
->list_id
);
1003 spin_lock(&un
->ulp
->lock
);
1005 list_del_rcu(&un
->list_proc
);
1006 spin_unlock(&un
->ulp
->lock
);
1010 /* Wake up all pending processes and let them fail with EIDRM. */
1011 INIT_LIST_HEAD(&tasks
);
1012 list_for_each_entry_safe(q
, tq
, &sma
->pending_const
, list
) {
1013 unlink_queue(sma
, q
);
1014 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1017 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
1018 unlink_queue(sma
, q
);
1019 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1021 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
1022 struct sem
*sem
= sma
->sem_base
+ i
;
1023 list_for_each_entry_safe(q
, tq
, &sem
->pending_const
, list
) {
1024 unlink_queue(sma
, q
);
1025 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1027 list_for_each_entry_safe(q
, tq
, &sem
->pending_alter
, list
) {
1028 unlink_queue(sma
, q
);
1029 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1033 /* Remove the semaphore set from the IDR */
1035 sem_unlock(sma
, -1);
1038 wake_up_sem_queue_do(&tasks
);
1039 ns
->used_sems
-= sma
->sem_nsems
;
1040 security_sem_free(sma
);
1041 ipc_rcu_putref(sma
);
1044 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
1048 return copy_to_user(buf
, in
, sizeof(*in
));
1051 struct semid_ds out
;
1053 memset(&out
, 0, sizeof(out
));
1055 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
1057 out
.sem_otime
= in
->sem_otime
;
1058 out
.sem_ctime
= in
->sem_ctime
;
1059 out
.sem_nsems
= in
->sem_nsems
;
1061 return copy_to_user(buf
, &out
, sizeof(out
));
1068 static time_t get_semotime(struct sem_array
*sma
)
1073 res
= sma
->sem_base
[0].sem_otime
;
1074 for (i
= 1; i
< sma
->sem_nsems
; i
++) {
1075 time_t to
= sma
->sem_base
[i
].sem_otime
;
1083 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
,
1084 int cmd
, int version
, void __user
*p
)
1087 struct sem_array
*sma
;
1093 struct seminfo seminfo
;
1096 err
= security_sem_semctl(NULL
, cmd
);
1100 memset(&seminfo
,0,sizeof(seminfo
));
1101 seminfo
.semmni
= ns
->sc_semmni
;
1102 seminfo
.semmns
= ns
->sc_semmns
;
1103 seminfo
.semmsl
= ns
->sc_semmsl
;
1104 seminfo
.semopm
= ns
->sc_semopm
;
1105 seminfo
.semvmx
= SEMVMX
;
1106 seminfo
.semmnu
= SEMMNU
;
1107 seminfo
.semmap
= SEMMAP
;
1108 seminfo
.semume
= SEMUME
;
1109 down_read(&sem_ids(ns
).rw_mutex
);
1110 if (cmd
== SEM_INFO
) {
1111 seminfo
.semusz
= sem_ids(ns
).in_use
;
1112 seminfo
.semaem
= ns
->used_sems
;
1114 seminfo
.semusz
= SEMUSZ
;
1115 seminfo
.semaem
= SEMAEM
;
1117 max_id
= ipc_get_maxid(&sem_ids(ns
));
1118 up_read(&sem_ids(ns
).rw_mutex
);
1119 if (copy_to_user(p
, &seminfo
, sizeof(struct seminfo
)))
1121 return (max_id
< 0) ? 0: max_id
;
1126 struct semid64_ds tbuf
;
1129 memset(&tbuf
, 0, sizeof(tbuf
));
1132 if (cmd
== SEM_STAT
) {
1133 sma
= sem_obtain_object(ns
, semid
);
1138 id
= sma
->sem_perm
.id
;
1140 sma
= sem_obtain_object_check(ns
, semid
);
1148 if (ipcperms(ns
, &sma
->sem_perm
, S_IRUGO
))
1151 err
= security_sem_semctl(sma
, cmd
);
1155 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
1156 tbuf
.sem_otime
= get_semotime(sma
);
1157 tbuf
.sem_ctime
= sma
->sem_ctime
;
1158 tbuf
.sem_nsems
= sma
->sem_nsems
;
1160 if (copy_semid_to_user(p
, &tbuf
, version
))
1172 static int semctl_setval(struct ipc_namespace
*ns
, int semid
, int semnum
,
1175 struct sem_undo
*un
;
1176 struct sem_array
*sma
;
1179 struct list_head tasks
;
1181 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1182 /* big-endian 64bit */
1185 /* 32bit or little-endian 64bit */
1189 if (val
> SEMVMX
|| val
< 0)
1192 INIT_LIST_HEAD(&tasks
);
1195 sma
= sem_obtain_object_check(ns
, semid
);
1198 return PTR_ERR(sma
);
1201 if (semnum
< 0 || semnum
>= sma
->sem_nsems
) {
1207 if (ipcperms(ns
, &sma
->sem_perm
, S_IWUGO
)) {
1212 err
= security_sem_semctl(sma
, SETVAL
);
1218 sem_lock(sma
, NULL
, -1);
1220 curr
= &sma
->sem_base
[semnum
];
1222 ipc_assert_locked_object(&sma
->sem_perm
);
1223 list_for_each_entry(un
, &sma
->list_id
, list_id
)
1224 un
->semadj
[semnum
] = 0;
1227 curr
->sempid
= task_tgid_vnr(current
);
1228 sma
->sem_ctime
= get_seconds();
1229 /* maybe some queued-up processes were waiting for this */
1230 do_smart_update(sma
, NULL
, 0, 0, &tasks
);
1231 sem_unlock(sma
, -1);
1233 wake_up_sem_queue_do(&tasks
);
1237 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
1238 int cmd
, void __user
*p
)
1240 struct sem_array
*sma
;
1243 ushort fast_sem_io
[SEMMSL_FAST
];
1244 ushort
* sem_io
= fast_sem_io
;
1245 struct list_head tasks
;
1247 INIT_LIST_HEAD(&tasks
);
1250 sma
= sem_obtain_object_check(ns
, semid
);
1253 return PTR_ERR(sma
);
1256 nsems
= sma
->sem_nsems
;
1259 if (ipcperms(ns
, &sma
->sem_perm
, cmd
== SETALL
? S_IWUGO
: S_IRUGO
))
1260 goto out_rcu_wakeup
;
1262 err
= security_sem_semctl(sma
, cmd
);
1264 goto out_rcu_wakeup
;
1270 ushort __user
*array
= p
;
1273 sem_lock(sma
, NULL
, -1);
1274 if(nsems
> SEMMSL_FAST
) {
1275 if (!ipc_rcu_getref(sma
)) {
1276 sem_unlock(sma
, -1);
1281 sem_unlock(sma
, -1);
1283 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1284 if(sem_io
== NULL
) {
1290 sem_lock_and_putref(sma
);
1291 if (sma
->sem_perm
.deleted
) {
1292 sem_unlock(sma
, -1);
1298 for (i
= 0; i
< sma
->sem_nsems
; i
++)
1299 sem_io
[i
] = sma
->sem_base
[i
].semval
;
1300 sem_unlock(sma
, -1);
1303 if(copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
1310 struct sem_undo
*un
;
1312 if (!ipc_rcu_getref(sma
)) {
1318 if(nsems
> SEMMSL_FAST
) {
1319 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1320 if(sem_io
== NULL
) {
1326 if (copy_from_user (sem_io
, p
, nsems
*sizeof(ushort
))) {
1332 for (i
= 0; i
< nsems
; i
++) {
1333 if (sem_io
[i
] > SEMVMX
) {
1340 sem_lock_and_putref(sma
);
1341 if (sma
->sem_perm
.deleted
) {
1342 sem_unlock(sma
, -1);
1348 for (i
= 0; i
< nsems
; i
++)
1349 sma
->sem_base
[i
].semval
= sem_io
[i
];
1351 ipc_assert_locked_object(&sma
->sem_perm
);
1352 list_for_each_entry(un
, &sma
->list_id
, list_id
) {
1353 for (i
= 0; i
< nsems
; i
++)
1356 sma
->sem_ctime
= get_seconds();
1357 /* maybe some queued-up processes were waiting for this */
1358 do_smart_update(sma
, NULL
, 0, 0, &tasks
);
1362 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1365 if (semnum
< 0 || semnum
>= nsems
)
1366 goto out_rcu_wakeup
;
1368 sem_lock(sma
, NULL
, -1);
1369 curr
= &sma
->sem_base
[semnum
];
1379 err
= count_semncnt(sma
,semnum
);
1382 err
= count_semzcnt(sma
,semnum
);
1387 sem_unlock(sma
, -1);
1390 wake_up_sem_queue_do(&tasks
);
1392 if(sem_io
!= fast_sem_io
)
1393 ipc_free(sem_io
, sizeof(ushort
)*nsems
);
1397 static inline unsigned long
1398 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
1402 if (copy_from_user(out
, buf
, sizeof(*out
)))
1407 struct semid_ds tbuf_old
;
1409 if(copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
1412 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
1413 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
1414 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
1424 * This function handles some semctl commands which require the rw_mutex
1425 * to be held in write mode.
1426 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1428 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
1429 int cmd
, int version
, void __user
*p
)
1431 struct sem_array
*sma
;
1433 struct semid64_ds semid64
;
1434 struct kern_ipc_perm
*ipcp
;
1436 if(cmd
== IPC_SET
) {
1437 if (copy_semid_from_user(&semid64
, p
, version
))
1441 down_write(&sem_ids(ns
).rw_mutex
);
1444 ipcp
= ipcctl_pre_down_nolock(ns
, &sem_ids(ns
), semid
, cmd
,
1445 &semid64
.sem_perm
, 0);
1447 err
= PTR_ERR(ipcp
);
1451 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1453 err
= security_sem_semctl(sma
, cmd
);
1459 sem_lock(sma
, NULL
, -1);
1460 /* freeary unlocks the ipc object and rcu */
1464 sem_lock(sma
, NULL
, -1);
1465 err
= ipc_update_perm(&semid64
.sem_perm
, ipcp
);
1468 sma
->sem_ctime
= get_seconds();
1476 sem_unlock(sma
, -1);
1480 up_write(&sem_ids(ns
).rw_mutex
);
1484 SYSCALL_DEFINE4(semctl
, int, semid
, int, semnum
, int, cmd
, unsigned long, arg
)
1487 struct ipc_namespace
*ns
;
1488 void __user
*p
= (void __user
*)arg
;
1493 version
= ipc_parse_version(&cmd
);
1494 ns
= current
->nsproxy
->ipc_ns
;
1501 return semctl_nolock(ns
, semid
, cmd
, version
, p
);
1508 return semctl_main(ns
, semid
, semnum
, cmd
, p
);
1510 return semctl_setval(ns
, semid
, semnum
, arg
);
1513 return semctl_down(ns
, semid
, cmd
, version
, p
);
1519 /* If the task doesn't already have a undo_list, then allocate one
1520 * here. We guarantee there is only one thread using this undo list,
1521 * and current is THE ONE
1523 * If this allocation and assignment succeeds, but later
1524 * portions of this code fail, there is no need to free the sem_undo_list.
1525 * Just let it stay associated with the task, and it'll be freed later
1528 * This can block, so callers must hold no locks.
1530 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
1532 struct sem_undo_list
*undo_list
;
1534 undo_list
= current
->sysvsem
.undo_list
;
1536 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
1537 if (undo_list
== NULL
)
1539 spin_lock_init(&undo_list
->lock
);
1540 atomic_set(&undo_list
->refcnt
, 1);
1541 INIT_LIST_HEAD(&undo_list
->list_proc
);
1543 current
->sysvsem
.undo_list
= undo_list
;
1545 *undo_listp
= undo_list
;
1549 static struct sem_undo
*__lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1551 struct sem_undo
*un
;
1553 list_for_each_entry_rcu(un
, &ulp
->list_proc
, list_proc
) {
1554 if (un
->semid
== semid
)
1560 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1562 struct sem_undo
*un
;
1564 assert_spin_locked(&ulp
->lock
);
1566 un
= __lookup_undo(ulp
, semid
);
1568 list_del_rcu(&un
->list_proc
);
1569 list_add_rcu(&un
->list_proc
, &ulp
->list_proc
);
1575 * find_alloc_undo - Lookup (and if not present create) undo array
1577 * @semid: semaphore array id
1579 * The function looks up (and if not present creates) the undo structure.
1580 * The size of the undo structure depends on the size of the semaphore
1581 * array, thus the alloc path is not that straightforward.
1582 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1583 * performs a rcu_read_lock().
1585 static struct sem_undo
*find_alloc_undo(struct ipc_namespace
*ns
, int semid
)
1587 struct sem_array
*sma
;
1588 struct sem_undo_list
*ulp
;
1589 struct sem_undo
*un
, *new;
1592 error
= get_undo_list(&ulp
);
1594 return ERR_PTR(error
);
1597 spin_lock(&ulp
->lock
);
1598 un
= lookup_undo(ulp
, semid
);
1599 spin_unlock(&ulp
->lock
);
1600 if (likely(un
!=NULL
))
1603 /* no undo structure around - allocate one. */
1604 /* step 1: figure out the size of the semaphore array */
1605 sma
= sem_obtain_object_check(ns
, semid
);
1608 return ERR_CAST(sma
);
1611 nsems
= sma
->sem_nsems
;
1612 if (!ipc_rcu_getref(sma
)) {
1614 un
= ERR_PTR(-EIDRM
);
1619 /* step 2: allocate new undo structure */
1620 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1623 return ERR_PTR(-ENOMEM
);
1626 /* step 3: Acquire the lock on semaphore array */
1628 sem_lock_and_putref(sma
);
1629 if (sma
->sem_perm
.deleted
) {
1630 sem_unlock(sma
, -1);
1633 un
= ERR_PTR(-EIDRM
);
1636 spin_lock(&ulp
->lock
);
1639 * step 4: check for races: did someone else allocate the undo struct?
1641 un
= lookup_undo(ulp
, semid
);
1646 /* step 5: initialize & link new undo structure */
1647 new->semadj
= (short *) &new[1];
1650 assert_spin_locked(&ulp
->lock
);
1651 list_add_rcu(&new->list_proc
, &ulp
->list_proc
);
1652 ipc_assert_locked_object(&sma
->sem_perm
);
1653 list_add(&new->list_id
, &sma
->list_id
);
1657 spin_unlock(&ulp
->lock
);
1658 sem_unlock(sma
, -1);
1665 * get_queue_result - Retrieve the result code from sem_queue
1666 * @q: Pointer to queue structure
1668 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1669 * q->status, then we must loop until the value is replaced with the final
1670 * value: This may happen if a task is woken up by an unrelated event (e.g.
1671 * signal) and in parallel the task is woken up by another task because it got
1672 * the requested semaphores.
1674 * The function can be called with or without holding the semaphore spinlock.
1676 static int get_queue_result(struct sem_queue
*q
)
1681 while (unlikely(error
== IN_WAKEUP
)) {
1690 SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsops
,
1691 unsigned, nsops
, const struct timespec __user
*, timeout
)
1693 int error
= -EINVAL
;
1694 struct sem_array
*sma
;
1695 struct sembuf fast_sops
[SEMOPM_FAST
];
1696 struct sembuf
* sops
= fast_sops
, *sop
;
1697 struct sem_undo
*un
;
1698 int undos
= 0, alter
= 0, max
, locknum
;
1699 struct sem_queue queue
;
1700 unsigned long jiffies_left
= 0;
1701 struct ipc_namespace
*ns
;
1702 struct list_head tasks
;
1704 ns
= current
->nsproxy
->ipc_ns
;
1706 if (nsops
< 1 || semid
< 0)
1708 if (nsops
> ns
->sc_semopm
)
1710 if(nsops
> SEMOPM_FAST
) {
1711 sops
= kmalloc(sizeof(*sops
)*nsops
,GFP_KERNEL
);
1715 if (copy_from_user (sops
, tsops
, nsops
* sizeof(*tsops
))) {
1720 struct timespec _timeout
;
1721 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1725 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1726 _timeout
.tv_nsec
>= 1000000000L) {
1730 jiffies_left
= timespec_to_jiffies(&_timeout
);
1733 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1734 if (sop
->sem_num
>= max
)
1736 if (sop
->sem_flg
& SEM_UNDO
)
1738 if (sop
->sem_op
!= 0)
1742 INIT_LIST_HEAD(&tasks
);
1745 /* On success, find_alloc_undo takes the rcu_read_lock */
1746 un
= find_alloc_undo(ns
, semid
);
1748 error
= PTR_ERR(un
);
1756 sma
= sem_obtain_object_check(ns
, semid
);
1759 error
= PTR_ERR(sma
);
1764 if (max
>= sma
->sem_nsems
)
1765 goto out_rcu_wakeup
;
1768 if (ipcperms(ns
, &sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
))
1769 goto out_rcu_wakeup
;
1771 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1773 goto out_rcu_wakeup
;
1776 * semid identifiers are not unique - find_alloc_undo may have
1777 * allocated an undo structure, it was invalidated by an RMID
1778 * and now a new array with received the same id. Check and fail.
1779 * This case can be detected checking un->semid. The existence of
1780 * "un" itself is guaranteed by rcu.
1783 locknum
= sem_lock(sma
, sops
, nsops
);
1784 if (un
&& un
->semid
== -1)
1785 goto out_unlock_free
;
1787 error
= try_atomic_semop (sma
, sops
, nsops
, un
, task_tgid_vnr(current
));
1789 if (alter
&& error
== 0)
1790 do_smart_update(sma
, sops
, nsops
, 1, &tasks
);
1792 goto out_unlock_free
;
1795 /* We need to sleep on this operation, so we put the current
1796 * task into the pending queue and go to sleep.
1800 queue
.nsops
= nsops
;
1802 queue
.pid
= task_tgid_vnr(current
);
1803 queue
.alter
= alter
;
1807 curr
= &sma
->sem_base
[sops
->sem_num
];
1810 if (sma
->complex_count
) {
1811 list_add_tail(&queue
.list
,
1812 &sma
->pending_alter
);
1815 list_add_tail(&queue
.list
,
1816 &curr
->pending_alter
);
1819 list_add_tail(&queue
.list
, &curr
->pending_const
);
1822 if (!sma
->complex_count
)
1826 list_add_tail(&queue
.list
, &sma
->pending_alter
);
1828 list_add_tail(&queue
.list
, &sma
->pending_const
);
1830 sma
->complex_count
++;
1833 queue
.status
= -EINTR
;
1834 queue
.sleeper
= current
;
1837 current
->state
= TASK_INTERRUPTIBLE
;
1838 sem_unlock(sma
, locknum
);
1842 jiffies_left
= schedule_timeout(jiffies_left
);
1846 error
= get_queue_result(&queue
);
1848 if (error
!= -EINTR
) {
1849 /* fast path: update_queue already obtained all requested
1851 * Perform a smp_mb(): User space could assume that semop()
1852 * is a memory barrier: Without the mb(), the cpu could
1853 * speculatively read in user space stale data that was
1854 * overwritten by the previous owner of the semaphore.
1862 sma
= sem_obtain_lock(ns
, semid
, sops
, nsops
, &locknum
);
1865 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1867 error
= get_queue_result(&queue
);
1870 * Array removed? If yes, leave without sem_unlock().
1879 * If queue.status != -EINTR we are woken up by another process.
1880 * Leave without unlink_queue(), but with sem_unlock().
1883 if (error
!= -EINTR
) {
1884 goto out_unlock_free
;
1888 * If an interrupt occurred we have to clean up the queue
1890 if (timeout
&& jiffies_left
== 0)
1894 * If the wakeup was spurious, just retry
1896 if (error
== -EINTR
&& !signal_pending(current
))
1899 unlink_queue(sma
, &queue
);
1902 sem_unlock(sma
, locknum
);
1905 wake_up_sem_queue_do(&tasks
);
1907 if(sops
!= fast_sops
)
1912 SYSCALL_DEFINE3(semop
, int, semid
, struct sembuf __user
*, tsops
,
1915 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
1918 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1919 * parent and child tasks.
1922 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
1924 struct sem_undo_list
*undo_list
;
1927 if (clone_flags
& CLONE_SYSVSEM
) {
1928 error
= get_undo_list(&undo_list
);
1931 atomic_inc(&undo_list
->refcnt
);
1932 tsk
->sysvsem
.undo_list
= undo_list
;
1934 tsk
->sysvsem
.undo_list
= NULL
;
1940 * add semadj values to semaphores, free undo structures.
1941 * undo structures are not freed when semaphore arrays are destroyed
1942 * so some of them may be out of date.
1943 * IMPLEMENTATION NOTE: There is some confusion over whether the
1944 * set of adjustments that needs to be done should be done in an atomic
1945 * manner or not. That is, if we are attempting to decrement the semval
1946 * should we queue up and wait until we can do so legally?
1947 * The original implementation attempted to do this (queue and wait).
1948 * The current implementation does not do so. The POSIX standard
1949 * and SVID should be consulted to determine what behavior is mandated.
1951 void exit_sem(struct task_struct
*tsk
)
1953 struct sem_undo_list
*ulp
;
1955 ulp
= tsk
->sysvsem
.undo_list
;
1958 tsk
->sysvsem
.undo_list
= NULL
;
1960 if (!atomic_dec_and_test(&ulp
->refcnt
))
1964 struct sem_array
*sma
;
1965 struct sem_undo
*un
;
1966 struct list_head tasks
;
1970 un
= list_entry_rcu(ulp
->list_proc
.next
,
1971 struct sem_undo
, list_proc
);
1972 if (&un
->list_proc
== &ulp
->list_proc
)
1982 sma
= sem_obtain_object_check(tsk
->nsproxy
->ipc_ns
, un
->semid
);
1983 /* exit_sem raced with IPC_RMID, nothing to do */
1989 sem_lock(sma
, NULL
, -1);
1990 un
= __lookup_undo(ulp
, semid
);
1992 /* exit_sem raced with IPC_RMID+semget() that created
1993 * exactly the same semid. Nothing to do.
1995 sem_unlock(sma
, -1);
2000 /* remove un from the linked lists */
2001 ipc_assert_locked_object(&sma
->sem_perm
);
2002 list_del(&un
->list_id
);
2004 spin_lock(&ulp
->lock
);
2005 list_del_rcu(&un
->list_proc
);
2006 spin_unlock(&ulp
->lock
);
2008 /* perform adjustments registered in un */
2009 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
2010 struct sem
* semaphore
= &sma
->sem_base
[i
];
2011 if (un
->semadj
[i
]) {
2012 semaphore
->semval
+= un
->semadj
[i
];
2014 * Range checks of the new semaphore value,
2015 * not defined by sus:
2016 * - Some unices ignore the undo entirely
2017 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2018 * - some cap the value (e.g. FreeBSD caps
2019 * at 0, but doesn't enforce SEMVMX)
2021 * Linux caps the semaphore value, both at 0
2024 * Manfred <manfred@colorfullife.com>
2026 if (semaphore
->semval
< 0)
2027 semaphore
->semval
= 0;
2028 if (semaphore
->semval
> SEMVMX
)
2029 semaphore
->semval
= SEMVMX
;
2030 semaphore
->sempid
= task_tgid_vnr(current
);
2033 /* maybe some queued-up processes were waiting for this */
2034 INIT_LIST_HEAD(&tasks
);
2035 do_smart_update(sma
, NULL
, 0, 1, &tasks
);
2036 sem_unlock(sma
, -1);
2038 wake_up_sem_queue_do(&tasks
);
2045 #ifdef CONFIG_PROC_FS
2046 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
2048 struct user_namespace
*user_ns
= seq_user_ns(s
);
2049 struct sem_array
*sma
= it
;
2052 sem_otime
= get_semotime(sma
);
2054 return seq_printf(s
,
2055 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2060 from_kuid_munged(user_ns
, sma
->sem_perm
.uid
),
2061 from_kgid_munged(user_ns
, sma
->sem_perm
.gid
),
2062 from_kuid_munged(user_ns
, sma
->sem_perm
.cuid
),
2063 from_kgid_munged(user_ns
, sma
->sem_perm
.cgid
),