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_semcnt()
51 * - the task that performs a successful semop() scans the list of all
52 * sleeping tasks and completes any pending operations that can be fulfilled.
53 * Semaphores are actively given to waiting tasks (necessary for FIFO).
54 * (see update_queue())
55 * - To improve the scalability, the actual wake-up calls are performed after
56 * dropping all locks. (see wake_up_sem_queue_prepare(),
57 * wake_up_sem_queue_do())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - The synchronizations between wake-ups due to a timeout/signal and a
63 * wake-up due to a completed semaphore operation is achieved by using an
64 * intermediate state (IN_WAKEUP).
65 * - UNDO values are stored in an array (one per process and per
66 * semaphore array, lazily allocated). For backwards compatibility, multiple
67 * modes for the UNDO variables are supported (per process, per thread)
68 * (see copy_semundo, CLONE_SYSVSEM)
69 * - There are two lists of the pending operations: a per-array list
70 * and per-semaphore list (stored in the array). This allows to achieve FIFO
71 * ordering without always scanning all pending operations.
72 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
75 #include <linux/slab.h>
76 #include <linux/spinlock.h>
77 #include <linux/init.h>
78 #include <linux/proc_fs.h>
79 #include <linux/time.h>
80 #include <linux/security.h>
81 #include <linux/syscalls.h>
82 #include <linux/audit.h>
83 #include <linux/capability.h>
84 #include <linux/seq_file.h>
85 #include <linux/rwsem.h>
86 #include <linux/nsproxy.h>
87 #include <linux/ipc_namespace.h>
89 #include <linux/uaccess.h>
92 /* One semaphore structure for each semaphore in the system. */
94 int semval
; /* current value */
95 int sempid
; /* pid of last operation */
96 spinlock_t lock
; /* spinlock for fine-grained semtimedop */
97 struct list_head pending_alter
; /* pending single-sop operations */
98 /* that alter the semaphore */
99 struct list_head pending_const
; /* pending single-sop operations */
100 /* that do not alter the semaphore*/
101 time_t sem_otime
; /* candidate for sem_otime */
102 } ____cacheline_aligned_in_smp
;
104 /* One queue for each sleeping process in the system. */
106 struct list_head list
; /* queue of pending operations */
107 struct task_struct
*sleeper
; /* this process */
108 struct sem_undo
*undo
; /* undo structure */
109 int pid
; /* process id of requesting process */
110 int status
; /* completion status of operation */
111 struct sembuf
*sops
; /* array of pending operations */
112 struct sembuf
*blocking
; /* the operation that blocked */
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 */
159 * sem_array.complex_count,
160 * sem_array.pending{_alter,_cont},
161 * sem_array.sem_undo: global sem_lock() for read/write
162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
168 #define sc_semmsl sem_ctls[0]
169 #define sc_semmns sem_ctls[1]
170 #define sc_semopm sem_ctls[2]
171 #define sc_semmni sem_ctls[3]
173 void sem_init_ns(struct ipc_namespace
*ns
)
175 ns
->sc_semmsl
= SEMMSL
;
176 ns
->sc_semmns
= SEMMNS
;
177 ns
->sc_semopm
= SEMOPM
;
178 ns
->sc_semmni
= SEMMNI
;
180 ipc_init_ids(&ns
->ids
[IPC_SEM_IDS
]);
184 void sem_exit_ns(struct ipc_namespace
*ns
)
186 free_ipcs(ns
, &sem_ids(ns
), freeary
);
187 idr_destroy(&ns
->ids
[IPC_SEM_IDS
].ipcs_idr
);
191 void __init
sem_init(void)
193 sem_init_ns(&init_ipc_ns
);
194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
196 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
206 static void unmerge_queues(struct sem_array
*sma
)
208 struct sem_queue
*q
, *tq
;
210 /* complex operations still around? */
211 if (sma
->complex_count
)
214 * We will switch back to simple mode.
215 * Move all pending operation back into the per-semaphore
218 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
220 curr
= &sma
->sem_base
[q
->sops
[0].sem_num
];
222 list_add_tail(&q
->list
, &curr
->pending_alter
);
224 INIT_LIST_HEAD(&sma
->pending_alter
);
228 * merge_queues - merge single semop queues into global queue
229 * @sma: semaphore array
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
236 static void merge_queues(struct sem_array
*sma
)
239 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
240 struct sem
*sem
= sma
->sem_base
+ i
;
242 list_splice_init(&sem
->pending_alter
, &sma
->pending_alter
);
246 static void sem_rcu_free(struct rcu_head
*head
)
248 struct ipc_rcu
*p
= container_of(head
, struct ipc_rcu
, rcu
);
249 struct sem_array
*sma
= ipc_rcu_to_struct(p
);
251 security_sem_free(sma
);
256 * Wait until all currently ongoing simple ops have completed.
257 * Caller must own sem_perm.lock.
258 * New simple ops cannot start, because simple ops first check
259 * that sem_perm.lock is free.
260 * that a) sem_perm.lock is free and b) complex_count is 0.
262 static void sem_wait_array(struct sem_array
*sma
)
267 if (sma
->complex_count
) {
268 /* The thread that increased sma->complex_count waited on
269 * all sem->lock locks. Thus we don't need to wait again.
274 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
275 sem
= sma
->sem_base
+ i
;
276 spin_unlock_wait(&sem
->lock
);
281 * If the request contains only one semaphore operation, and there are
282 * no complex transactions pending, lock only the semaphore involved.
283 * Otherwise, lock the entire semaphore array, since we either have
284 * multiple semaphores in our own semops, or we need to look at
285 * semaphores from other pending complex operations.
287 static inline int sem_lock(struct sem_array
*sma
, struct sembuf
*sops
,
293 /* Complex operation - acquire a full lock */
294 ipc_lock_object(&sma
->sem_perm
);
296 /* And wait until all simple ops that are processed
297 * right now have dropped their locks.
304 * Only one semaphore affected - try to optimize locking.
306 * - optimized locking is possible if no complex operation
307 * is either enqueued or processed right now.
308 * - The test for enqueued complex ops is simple:
309 * sma->complex_count != 0
310 * - Testing for complex ops that are processed right now is
311 * a bit more difficult. Complex ops acquire the full lock
312 * and first wait that the running simple ops have completed.
314 * Thus: If we own a simple lock and the global lock is free
315 * and complex_count is now 0, then it will stay 0 and
316 * thus just locking sem->lock is sufficient.
318 sem
= sma
->sem_base
+ sops
->sem_num
;
320 if (sma
->complex_count
== 0) {
322 * It appears that no complex operation is around.
323 * Acquire the per-semaphore lock.
325 spin_lock(&sem
->lock
);
327 /* Then check that the global lock is free */
328 if (!spin_is_locked(&sma
->sem_perm
.lock
)) {
329 /* spin_is_locked() is not a memory barrier */
332 /* Now repeat the test of complex_count:
333 * It can't change anymore until we drop sem->lock.
334 * Thus: if is now 0, then it will stay 0.
336 if (sma
->complex_count
== 0) {
337 /* fast path successful! */
338 return sops
->sem_num
;
341 spin_unlock(&sem
->lock
);
344 /* slow path: acquire the full lock */
345 ipc_lock_object(&sma
->sem_perm
);
347 if (sma
->complex_count
== 0) {
349 * There is no complex operation, thus we can switch
350 * back to the fast path.
352 spin_lock(&sem
->lock
);
353 ipc_unlock_object(&sma
->sem_perm
);
354 return sops
->sem_num
;
356 /* Not a false alarm, thus complete the sequence for a
364 static inline void sem_unlock(struct sem_array
*sma
, int locknum
)
368 ipc_unlock_object(&sma
->sem_perm
);
370 struct sem
*sem
= sma
->sem_base
+ locknum
;
371 spin_unlock(&sem
->lock
);
376 * sem_lock_(check_) routines are called in the paths where the rwsem
379 * The caller holds the RCU read lock.
381 static inline struct sem_array
*sem_obtain_lock(struct ipc_namespace
*ns
,
382 int id
, struct sembuf
*sops
, int nsops
, int *locknum
)
384 struct kern_ipc_perm
*ipcp
;
385 struct sem_array
*sma
;
387 ipcp
= ipc_obtain_object(&sem_ids(ns
), id
);
389 return ERR_CAST(ipcp
);
391 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
392 *locknum
= sem_lock(sma
, sops
, nsops
);
394 /* ipc_rmid() may have already freed the ID while sem_lock
395 * was spinning: verify that the structure is still valid
397 if (ipc_valid_object(ipcp
))
398 return container_of(ipcp
, struct sem_array
, sem_perm
);
400 sem_unlock(sma
, *locknum
);
401 return ERR_PTR(-EINVAL
);
404 static inline struct sem_array
*sem_obtain_object(struct ipc_namespace
*ns
, int id
)
406 struct kern_ipc_perm
*ipcp
= ipc_obtain_object(&sem_ids(ns
), id
);
409 return ERR_CAST(ipcp
);
411 return container_of(ipcp
, struct sem_array
, sem_perm
);
414 static inline struct sem_array
*sem_obtain_object_check(struct ipc_namespace
*ns
,
417 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&sem_ids(ns
), id
);
420 return ERR_CAST(ipcp
);
422 return container_of(ipcp
, struct sem_array
, sem_perm
);
425 static inline void sem_lock_and_putref(struct sem_array
*sma
)
427 sem_lock(sma
, NULL
, -1);
428 ipc_rcu_putref(sma
, ipc_rcu_free
);
431 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
433 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
437 * Lockless wakeup algorithm:
438 * Without the check/retry algorithm a lockless wakeup is possible:
439 * - queue.status is initialized to -EINTR before blocking.
440 * - wakeup is performed by
441 * * unlinking the queue entry from the pending list
442 * * setting queue.status to IN_WAKEUP
443 * This is the notification for the blocked thread that a
444 * result value is imminent.
445 * * call wake_up_process
446 * * set queue.status to the final value.
447 * - the previously blocked thread checks queue.status:
448 * * if it's IN_WAKEUP, then it must wait until the value changes
449 * * if it's not -EINTR, then the operation was completed by
450 * update_queue. semtimedop can return queue.status without
451 * performing any operation on the sem array.
452 * * otherwise it must acquire the spinlock and check what's up.
454 * The two-stage algorithm is necessary to protect against the following
456 * - if queue.status is set after wake_up_process, then the woken up idle
457 * thread could race forward and try (and fail) to acquire sma->lock
458 * before update_queue had a chance to set queue.status
459 * - if queue.status is written before wake_up_process and if the
460 * blocked process is woken up by a signal between writing
461 * queue.status and the wake_up_process, then the woken up
462 * process could return from semtimedop and die by calling
463 * sys_exit before wake_up_process is called. Then wake_up_process
464 * will oops, because the task structure is already invalid.
465 * (yes, this happened on s390 with sysv msg).
471 * newary - Create a new semaphore set
473 * @params: ptr to the structure that contains key, semflg and nsems
475 * Called with sem_ids.rwsem held (as a writer)
477 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
481 struct sem_array
*sma
;
483 key_t key
= params
->key
;
484 int nsems
= params
->u
.nsems
;
485 int semflg
= params
->flg
;
490 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
493 size
= sizeof(*sma
) + nsems
* sizeof(struct sem
);
494 sma
= ipc_rcu_alloc(size
);
498 memset(sma
, 0, size
);
500 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
501 sma
->sem_perm
.key
= key
;
503 sma
->sem_perm
.security
= NULL
;
504 retval
= security_sem_alloc(sma
);
506 ipc_rcu_putref(sma
, ipc_rcu_free
);
510 sma
->sem_base
= (struct sem
*) &sma
[1];
512 for (i
= 0; i
< nsems
; i
++) {
513 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_alter
);
514 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_const
);
515 spin_lock_init(&sma
->sem_base
[i
].lock
);
518 sma
->complex_count
= 0;
519 INIT_LIST_HEAD(&sma
->pending_alter
);
520 INIT_LIST_HEAD(&sma
->pending_const
);
521 INIT_LIST_HEAD(&sma
->list_id
);
522 sma
->sem_nsems
= nsems
;
523 sma
->sem_ctime
= get_seconds();
525 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
527 ipc_rcu_putref(sma
, sem_rcu_free
);
530 ns
->used_sems
+= nsems
;
535 return sma
->sem_perm
.id
;
540 * Called with sem_ids.rwsem and ipcp locked.
542 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
544 struct sem_array
*sma
;
546 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
547 return security_sem_associate(sma
, semflg
);
551 * Called with sem_ids.rwsem and ipcp locked.
553 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
554 struct ipc_params
*params
)
556 struct sem_array
*sma
;
558 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
559 if (params
->u
.nsems
> sma
->sem_nsems
)
565 SYSCALL_DEFINE3(semget
, key_t
, key
, int, nsems
, int, semflg
)
567 struct ipc_namespace
*ns
;
568 static const struct ipc_ops sem_ops
= {
570 .associate
= sem_security
,
571 .more_checks
= sem_more_checks
,
573 struct ipc_params sem_params
;
575 ns
= current
->nsproxy
->ipc_ns
;
577 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
580 sem_params
.key
= key
;
581 sem_params
.flg
= semflg
;
582 sem_params
.u
.nsems
= nsems
;
584 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
588 * perform_atomic_semop - Perform (if possible) a semaphore operation
589 * @sma: semaphore array
590 * @q: struct sem_queue that describes the operation
592 * Returns 0 if the operation was possible.
593 * Returns 1 if the operation is impossible, the caller must sleep.
594 * Negative values are error codes.
596 static int perform_atomic_semop(struct sem_array
*sma
, struct sem_queue
*q
)
598 int result
, sem_op
, nsops
, pid
;
608 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
609 curr
= sma
->sem_base
+ sop
->sem_num
;
610 sem_op
= sop
->sem_op
;
611 result
= curr
->semval
;
613 if (!sem_op
&& result
)
622 if (sop
->sem_flg
& SEM_UNDO
) {
623 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
624 /* Exceeding the undo range is an error. */
625 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
627 un
->semadj
[sop
->sem_num
] = undo
;
630 curr
->semval
= result
;
635 while (sop
>= sops
) {
636 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
649 if (sop
->sem_flg
& IPC_NOWAIT
)
656 while (sop
>= sops
) {
657 sem_op
= sop
->sem_op
;
658 sma
->sem_base
[sop
->sem_num
].semval
-= sem_op
;
659 if (sop
->sem_flg
& SEM_UNDO
)
660 un
->semadj
[sop
->sem_num
] += sem_op
;
667 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
668 * @q: queue entry that must be signaled
669 * @error: Error value for the signal
671 * Prepare the wake-up of the queue entry q.
673 static void wake_up_sem_queue_prepare(struct list_head
*pt
,
674 struct sem_queue
*q
, int error
)
676 if (list_empty(pt
)) {
678 * Hold preempt off so that we don't get preempted and have the
679 * wakee busy-wait until we're scheduled back on.
683 q
->status
= IN_WAKEUP
;
686 list_add_tail(&q
->list
, pt
);
690 * wake_up_sem_queue_do - do the actual wake-up
691 * @pt: list of tasks to be woken up
693 * Do the actual wake-up.
694 * The function is called without any locks held, thus the semaphore array
695 * could be destroyed already and the tasks can disappear as soon as the
696 * status is set to the actual return code.
698 static void wake_up_sem_queue_do(struct list_head
*pt
)
700 struct sem_queue
*q
, *t
;
703 did_something
= !list_empty(pt
);
704 list_for_each_entry_safe(q
, t
, pt
, list
) {
705 wake_up_process(q
->sleeper
);
706 /* q can disappear immediately after writing q->status. */
714 static void unlink_queue(struct sem_array
*sma
, struct sem_queue
*q
)
718 sma
->complex_count
--;
721 /** check_restart(sma, q)
722 * @sma: semaphore array
723 * @q: the operation that just completed
725 * update_queue is O(N^2) when it restarts scanning the whole queue of
726 * waiting operations. Therefore this function checks if the restart is
727 * really necessary. It is called after a previously waiting operation
728 * modified the array.
729 * Note that wait-for-zero operations are handled without restart.
731 static int check_restart(struct sem_array
*sma
, struct sem_queue
*q
)
733 /* pending complex alter operations are too difficult to analyse */
734 if (!list_empty(&sma
->pending_alter
))
737 /* we were a sleeping complex operation. Too difficult */
741 /* It is impossible that someone waits for the new value:
742 * - complex operations always restart.
743 * - wait-for-zero are handled seperately.
744 * - q is a previously sleeping simple operation that
745 * altered the array. It must be a decrement, because
746 * simple increments never sleep.
747 * - If there are older (higher priority) decrements
748 * in the queue, then they have observed the original
749 * semval value and couldn't proceed. The operation
750 * decremented to value - thus they won't proceed either.
756 * wake_const_ops - wake up non-alter tasks
757 * @sma: semaphore array.
758 * @semnum: semaphore that was modified.
759 * @pt: list head for the tasks that must be woken up.
761 * wake_const_ops must be called after a semaphore in a semaphore array
762 * was set to 0. If complex const operations are pending, wake_const_ops must
763 * be called with semnum = -1, as well as with the number of each modified
765 * The tasks that must be woken up are added to @pt. The return code
766 * is stored in q->pid.
767 * The function returns 1 if at least one operation was completed successfully.
769 static int wake_const_ops(struct sem_array
*sma
, int semnum
,
770 struct list_head
*pt
)
773 struct list_head
*walk
;
774 struct list_head
*pending_list
;
775 int semop_completed
= 0;
778 pending_list
= &sma
->pending_const
;
780 pending_list
= &sma
->sem_base
[semnum
].pending_const
;
782 walk
= pending_list
->next
;
783 while (walk
!= pending_list
) {
786 q
= container_of(walk
, struct sem_queue
, list
);
789 error
= perform_atomic_semop(sma
, q
);
792 /* operation completed, remove from queue & wakeup */
794 unlink_queue(sma
, q
);
796 wake_up_sem_queue_prepare(pt
, q
, error
);
801 return semop_completed
;
805 * do_smart_wakeup_zero - wakeup all wait for zero tasks
806 * @sma: semaphore array
807 * @sops: operations that were performed
808 * @nsops: number of operations
809 * @pt: list head of the tasks that must be woken up.
811 * Checks all required queue for wait-for-zero operations, based
812 * on the actual changes that were performed on the semaphore array.
813 * The function returns 1 if at least one operation was completed successfully.
815 static int do_smart_wakeup_zero(struct sem_array
*sma
, struct sembuf
*sops
,
816 int nsops
, struct list_head
*pt
)
819 int semop_completed
= 0;
822 /* first: the per-semaphore queues, if known */
824 for (i
= 0; i
< nsops
; i
++) {
825 int num
= sops
[i
].sem_num
;
827 if (sma
->sem_base
[num
].semval
== 0) {
829 semop_completed
|= wake_const_ops(sma
, num
, pt
);
834 * No sops means modified semaphores not known.
835 * Assume all were changed.
837 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
838 if (sma
->sem_base
[i
].semval
== 0) {
840 semop_completed
|= wake_const_ops(sma
, i
, pt
);
845 * If one of the modified semaphores got 0,
846 * then check the global queue, too.
849 semop_completed
|= wake_const_ops(sma
, -1, pt
);
851 return semop_completed
;
856 * update_queue - look for tasks that can be completed.
857 * @sma: semaphore array.
858 * @semnum: semaphore that was modified.
859 * @pt: list head for the tasks that must be woken up.
861 * update_queue must be called after a semaphore in a semaphore array
862 * was modified. If multiple semaphores were modified, update_queue must
863 * be called with semnum = -1, as well as with the number of each modified
865 * The tasks that must be woken up are added to @pt. The return code
866 * is stored in q->pid.
867 * The function internally checks if const operations can now succeed.
869 * The function return 1 if at least one semop was completed successfully.
871 static int update_queue(struct sem_array
*sma
, int semnum
, struct list_head
*pt
)
874 struct list_head
*walk
;
875 struct list_head
*pending_list
;
876 int semop_completed
= 0;
879 pending_list
= &sma
->pending_alter
;
881 pending_list
= &sma
->sem_base
[semnum
].pending_alter
;
884 walk
= pending_list
->next
;
885 while (walk
!= pending_list
) {
888 q
= container_of(walk
, struct sem_queue
, list
);
891 /* If we are scanning the single sop, per-semaphore list of
892 * one semaphore and that semaphore is 0, then it is not
893 * necessary to scan further: simple increments
894 * that affect only one entry succeed immediately and cannot
895 * be in the per semaphore pending queue, and decrements
896 * cannot be successful if the value is already 0.
898 if (semnum
!= -1 && sma
->sem_base
[semnum
].semval
== 0)
901 error
= perform_atomic_semop(sma
, q
);
903 /* Does q->sleeper still need to sleep? */
907 unlink_queue(sma
, q
);
913 do_smart_wakeup_zero(sma
, q
->sops
, q
->nsops
, pt
);
914 restart
= check_restart(sma
, q
);
917 wake_up_sem_queue_prepare(pt
, q
, error
);
921 return semop_completed
;
925 * set_semotime - set sem_otime
926 * @sma: semaphore array
927 * @sops: operations that modified the array, may be NULL
929 * sem_otime is replicated to avoid cache line trashing.
930 * This function sets one instance to the current time.
932 static void set_semotime(struct sem_array
*sma
, struct sembuf
*sops
)
935 sma
->sem_base
[0].sem_otime
= get_seconds();
937 sma
->sem_base
[sops
[0].sem_num
].sem_otime
=
943 * do_smart_update - optimized update_queue
944 * @sma: semaphore array
945 * @sops: operations that were performed
946 * @nsops: number of operations
947 * @otime: force setting otime
948 * @pt: list head of the tasks that must be woken up.
950 * do_smart_update() does the required calls to update_queue and wakeup_zero,
951 * based on the actual changes that were performed on the semaphore array.
952 * Note that the function does not do the actual wake-up: the caller is
953 * responsible for calling wake_up_sem_queue_do(@pt).
954 * It is safe to perform this call after dropping all locks.
956 static void do_smart_update(struct sem_array
*sma
, struct sembuf
*sops
, int nsops
,
957 int otime
, struct list_head
*pt
)
961 otime
|= do_smart_wakeup_zero(sma
, sops
, nsops
, pt
);
963 if (!list_empty(&sma
->pending_alter
)) {
964 /* semaphore array uses the global queue - just process it. */
965 otime
|= update_queue(sma
, -1, pt
);
969 * No sops, thus the modified semaphores are not
972 for (i
= 0; i
< sma
->sem_nsems
; i
++)
973 otime
|= update_queue(sma
, i
, pt
);
976 * Check the semaphores that were increased:
977 * - No complex ops, thus all sleeping ops are
979 * - if we decreased the value, then any sleeping
980 * semaphore ops wont be able to run: If the
981 * previous value was too small, then the new
982 * value will be too small, too.
984 for (i
= 0; i
< nsops
; i
++) {
985 if (sops
[i
].sem_op
> 0) {
986 otime
|= update_queue(sma
,
987 sops
[i
].sem_num
, pt
);
993 set_semotime(sma
, sops
);
997 * check_qop: Test if a queued operation sleeps on the semaphore semnum
999 static int check_qop(struct sem_array
*sma
, int semnum
, struct sem_queue
*q
,
1002 struct sembuf
*sop
= q
->blocking
;
1005 * Linux always (since 0.99.10) reported a task as sleeping on all
1006 * semaphores. This violates SUS, therefore it was changed to the
1007 * standard compliant behavior.
1008 * Give the administrators a chance to notice that an application
1009 * might misbehave because it relies on the Linux behavior.
1011 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1012 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1013 current
->comm
, task_pid_nr(current
));
1015 if (sop
->sem_num
!= semnum
)
1018 if (count_zero
&& sop
->sem_op
== 0)
1020 if (!count_zero
&& sop
->sem_op
< 0)
1026 /* The following counts are associated to each semaphore:
1027 * semncnt number of tasks waiting on semval being nonzero
1028 * semzcnt number of tasks waiting on semval being zero
1030 * Per definition, a task waits only on the semaphore of the first semop
1031 * that cannot proceed, even if additional operation would block, too.
1033 static int count_semcnt(struct sem_array
*sma
, ushort semnum
,
1036 struct list_head
*l
;
1037 struct sem_queue
*q
;
1041 /* First: check the simple operations. They are easy to evaluate */
1043 l
= &sma
->sem_base
[semnum
].pending_const
;
1045 l
= &sma
->sem_base
[semnum
].pending_alter
;
1047 list_for_each_entry(q
, l
, list
) {
1048 /* all task on a per-semaphore list sleep on exactly
1054 /* Then: check the complex operations. */
1055 list_for_each_entry(q
, &sma
->pending_alter
, list
) {
1056 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1059 list_for_each_entry(q
, &sma
->pending_const
, list
) {
1060 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1066 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1067 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1068 * remains locked on exit.
1070 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
1072 struct sem_undo
*un
, *tu
;
1073 struct sem_queue
*q
, *tq
;
1074 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1075 struct list_head tasks
;
1078 /* Free the existing undo structures for this semaphore set. */
1079 ipc_assert_locked_object(&sma
->sem_perm
);
1080 list_for_each_entry_safe(un
, tu
, &sma
->list_id
, list_id
) {
1081 list_del(&un
->list_id
);
1082 spin_lock(&un
->ulp
->lock
);
1084 list_del_rcu(&un
->list_proc
);
1085 spin_unlock(&un
->ulp
->lock
);
1089 /* Wake up all pending processes and let them fail with EIDRM. */
1090 INIT_LIST_HEAD(&tasks
);
1091 list_for_each_entry_safe(q
, tq
, &sma
->pending_const
, list
) {
1092 unlink_queue(sma
, q
);
1093 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1096 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
1097 unlink_queue(sma
, q
);
1098 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1100 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
1101 struct sem
*sem
= sma
->sem_base
+ i
;
1102 list_for_each_entry_safe(q
, tq
, &sem
->pending_const
, list
) {
1103 unlink_queue(sma
, q
);
1104 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1106 list_for_each_entry_safe(q
, tq
, &sem
->pending_alter
, list
) {
1107 unlink_queue(sma
, q
);
1108 wake_up_sem_queue_prepare(&tasks
, q
, -EIDRM
);
1112 /* Remove the semaphore set from the IDR */
1114 sem_unlock(sma
, -1);
1117 wake_up_sem_queue_do(&tasks
);
1118 ns
->used_sems
-= sma
->sem_nsems
;
1119 ipc_rcu_putref(sma
, sem_rcu_free
);
1122 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
1126 return copy_to_user(buf
, in
, sizeof(*in
));
1129 struct semid_ds out
;
1131 memset(&out
, 0, sizeof(out
));
1133 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
1135 out
.sem_otime
= in
->sem_otime
;
1136 out
.sem_ctime
= in
->sem_ctime
;
1137 out
.sem_nsems
= in
->sem_nsems
;
1139 return copy_to_user(buf
, &out
, sizeof(out
));
1146 static time_t get_semotime(struct sem_array
*sma
)
1151 res
= sma
->sem_base
[0].sem_otime
;
1152 for (i
= 1; i
< sma
->sem_nsems
; i
++) {
1153 time_t to
= sma
->sem_base
[i
].sem_otime
;
1161 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
,
1162 int cmd
, int version
, void __user
*p
)
1165 struct sem_array
*sma
;
1171 struct seminfo seminfo
;
1174 err
= security_sem_semctl(NULL
, cmd
);
1178 memset(&seminfo
, 0, sizeof(seminfo
));
1179 seminfo
.semmni
= ns
->sc_semmni
;
1180 seminfo
.semmns
= ns
->sc_semmns
;
1181 seminfo
.semmsl
= ns
->sc_semmsl
;
1182 seminfo
.semopm
= ns
->sc_semopm
;
1183 seminfo
.semvmx
= SEMVMX
;
1184 seminfo
.semmnu
= SEMMNU
;
1185 seminfo
.semmap
= SEMMAP
;
1186 seminfo
.semume
= SEMUME
;
1187 down_read(&sem_ids(ns
).rwsem
);
1188 if (cmd
== SEM_INFO
) {
1189 seminfo
.semusz
= sem_ids(ns
).in_use
;
1190 seminfo
.semaem
= ns
->used_sems
;
1192 seminfo
.semusz
= SEMUSZ
;
1193 seminfo
.semaem
= SEMAEM
;
1195 max_id
= ipc_get_maxid(&sem_ids(ns
));
1196 up_read(&sem_ids(ns
).rwsem
);
1197 if (copy_to_user(p
, &seminfo
, sizeof(struct seminfo
)))
1199 return (max_id
< 0) ? 0 : max_id
;
1204 struct semid64_ds tbuf
;
1207 memset(&tbuf
, 0, sizeof(tbuf
));
1210 if (cmd
== SEM_STAT
) {
1211 sma
= sem_obtain_object(ns
, semid
);
1216 id
= sma
->sem_perm
.id
;
1218 sma
= sem_obtain_object_check(ns
, semid
);
1226 if (ipcperms(ns
, &sma
->sem_perm
, S_IRUGO
))
1229 err
= security_sem_semctl(sma
, cmd
);
1233 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
1234 tbuf
.sem_otime
= get_semotime(sma
);
1235 tbuf
.sem_ctime
= sma
->sem_ctime
;
1236 tbuf
.sem_nsems
= sma
->sem_nsems
;
1238 if (copy_semid_to_user(p
, &tbuf
, version
))
1250 static int semctl_setval(struct ipc_namespace
*ns
, int semid
, int semnum
,
1253 struct sem_undo
*un
;
1254 struct sem_array
*sma
;
1257 struct list_head tasks
;
1259 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1260 /* big-endian 64bit */
1263 /* 32bit or little-endian 64bit */
1267 if (val
> SEMVMX
|| val
< 0)
1270 INIT_LIST_HEAD(&tasks
);
1273 sma
= sem_obtain_object_check(ns
, semid
);
1276 return PTR_ERR(sma
);
1279 if (semnum
< 0 || semnum
>= sma
->sem_nsems
) {
1285 if (ipcperms(ns
, &sma
->sem_perm
, S_IWUGO
)) {
1290 err
= security_sem_semctl(sma
, SETVAL
);
1296 sem_lock(sma
, NULL
, -1);
1298 if (!ipc_valid_object(&sma
->sem_perm
)) {
1299 sem_unlock(sma
, -1);
1304 curr
= &sma
->sem_base
[semnum
];
1306 ipc_assert_locked_object(&sma
->sem_perm
);
1307 list_for_each_entry(un
, &sma
->list_id
, list_id
)
1308 un
->semadj
[semnum
] = 0;
1311 curr
->sempid
= task_tgid_vnr(current
);
1312 sma
->sem_ctime
= get_seconds();
1313 /* maybe some queued-up processes were waiting for this */
1314 do_smart_update(sma
, NULL
, 0, 0, &tasks
);
1315 sem_unlock(sma
, -1);
1317 wake_up_sem_queue_do(&tasks
);
1321 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
1322 int cmd
, void __user
*p
)
1324 struct sem_array
*sma
;
1327 ushort fast_sem_io
[SEMMSL_FAST
];
1328 ushort
*sem_io
= fast_sem_io
;
1329 struct list_head tasks
;
1331 INIT_LIST_HEAD(&tasks
);
1334 sma
= sem_obtain_object_check(ns
, semid
);
1337 return PTR_ERR(sma
);
1340 nsems
= sma
->sem_nsems
;
1343 if (ipcperms(ns
, &sma
->sem_perm
, cmd
== SETALL
? S_IWUGO
: S_IRUGO
))
1344 goto out_rcu_wakeup
;
1346 err
= security_sem_semctl(sma
, cmd
);
1348 goto out_rcu_wakeup
;
1354 ushort __user
*array
= p
;
1357 sem_lock(sma
, NULL
, -1);
1358 if (!ipc_valid_object(&sma
->sem_perm
)) {
1362 if (nsems
> SEMMSL_FAST
) {
1363 if (!ipc_rcu_getref(sma
)) {
1367 sem_unlock(sma
, -1);
1369 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1370 if (sem_io
== NULL
) {
1371 ipc_rcu_putref(sma
, ipc_rcu_free
);
1376 sem_lock_and_putref(sma
);
1377 if (!ipc_valid_object(&sma
->sem_perm
)) {
1382 for (i
= 0; i
< sma
->sem_nsems
; i
++)
1383 sem_io
[i
] = sma
->sem_base
[i
].semval
;
1384 sem_unlock(sma
, -1);
1387 if (copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
1394 struct sem_undo
*un
;
1396 if (!ipc_rcu_getref(sma
)) {
1398 goto out_rcu_wakeup
;
1402 if (nsems
> SEMMSL_FAST
) {
1403 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1404 if (sem_io
== NULL
) {
1405 ipc_rcu_putref(sma
, ipc_rcu_free
);
1410 if (copy_from_user(sem_io
, p
, nsems
*sizeof(ushort
))) {
1411 ipc_rcu_putref(sma
, ipc_rcu_free
);
1416 for (i
= 0; i
< nsems
; i
++) {
1417 if (sem_io
[i
] > SEMVMX
) {
1418 ipc_rcu_putref(sma
, ipc_rcu_free
);
1424 sem_lock_and_putref(sma
);
1425 if (!ipc_valid_object(&sma
->sem_perm
)) {
1430 for (i
= 0; i
< nsems
; i
++)
1431 sma
->sem_base
[i
].semval
= sem_io
[i
];
1433 ipc_assert_locked_object(&sma
->sem_perm
);
1434 list_for_each_entry(un
, &sma
->list_id
, list_id
) {
1435 for (i
= 0; i
< nsems
; i
++)
1438 sma
->sem_ctime
= get_seconds();
1439 /* maybe some queued-up processes were waiting for this */
1440 do_smart_update(sma
, NULL
, 0, 0, &tasks
);
1444 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1447 if (semnum
< 0 || semnum
>= nsems
)
1448 goto out_rcu_wakeup
;
1450 sem_lock(sma
, NULL
, -1);
1451 if (!ipc_valid_object(&sma
->sem_perm
)) {
1455 curr
= &sma
->sem_base
[semnum
];
1465 err
= count_semcnt(sma
, semnum
, 0);
1468 err
= count_semcnt(sma
, semnum
, 1);
1473 sem_unlock(sma
, -1);
1476 wake_up_sem_queue_do(&tasks
);
1478 if (sem_io
!= fast_sem_io
)
1479 ipc_free(sem_io
, sizeof(ushort
)*nsems
);
1483 static inline unsigned long
1484 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
1488 if (copy_from_user(out
, buf
, sizeof(*out
)))
1493 struct semid_ds tbuf_old
;
1495 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
1498 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
1499 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
1500 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
1510 * This function handles some semctl commands which require the rwsem
1511 * to be held in write mode.
1512 * NOTE: no locks must be held, the rwsem is taken inside this function.
1514 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
1515 int cmd
, int version
, void __user
*p
)
1517 struct sem_array
*sma
;
1519 struct semid64_ds semid64
;
1520 struct kern_ipc_perm
*ipcp
;
1522 if (cmd
== IPC_SET
) {
1523 if (copy_semid_from_user(&semid64
, p
, version
))
1527 down_write(&sem_ids(ns
).rwsem
);
1530 ipcp
= ipcctl_pre_down_nolock(ns
, &sem_ids(ns
), semid
, cmd
,
1531 &semid64
.sem_perm
, 0);
1533 err
= PTR_ERR(ipcp
);
1537 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1539 err
= security_sem_semctl(sma
, cmd
);
1545 sem_lock(sma
, NULL
, -1);
1546 /* freeary unlocks the ipc object and rcu */
1550 sem_lock(sma
, NULL
, -1);
1551 err
= ipc_update_perm(&semid64
.sem_perm
, ipcp
);
1554 sma
->sem_ctime
= get_seconds();
1562 sem_unlock(sma
, -1);
1566 up_write(&sem_ids(ns
).rwsem
);
1570 SYSCALL_DEFINE4(semctl
, int, semid
, int, semnum
, int, cmd
, unsigned long, arg
)
1573 struct ipc_namespace
*ns
;
1574 void __user
*p
= (void __user
*)arg
;
1579 version
= ipc_parse_version(&cmd
);
1580 ns
= current
->nsproxy
->ipc_ns
;
1587 return semctl_nolock(ns
, semid
, cmd
, version
, p
);
1594 return semctl_main(ns
, semid
, semnum
, cmd
, p
);
1596 return semctl_setval(ns
, semid
, semnum
, arg
);
1599 return semctl_down(ns
, semid
, cmd
, version
, p
);
1605 /* If the task doesn't already have a undo_list, then allocate one
1606 * here. We guarantee there is only one thread using this undo list,
1607 * and current is THE ONE
1609 * If this allocation and assignment succeeds, but later
1610 * portions of this code fail, there is no need to free the sem_undo_list.
1611 * Just let it stay associated with the task, and it'll be freed later
1614 * This can block, so callers must hold no locks.
1616 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
1618 struct sem_undo_list
*undo_list
;
1620 undo_list
= current
->sysvsem
.undo_list
;
1622 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
1623 if (undo_list
== NULL
)
1625 spin_lock_init(&undo_list
->lock
);
1626 atomic_set(&undo_list
->refcnt
, 1);
1627 INIT_LIST_HEAD(&undo_list
->list_proc
);
1629 current
->sysvsem
.undo_list
= undo_list
;
1631 *undo_listp
= undo_list
;
1635 static struct sem_undo
*__lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1637 struct sem_undo
*un
;
1639 list_for_each_entry_rcu(un
, &ulp
->list_proc
, list_proc
) {
1640 if (un
->semid
== semid
)
1646 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1648 struct sem_undo
*un
;
1650 assert_spin_locked(&ulp
->lock
);
1652 un
= __lookup_undo(ulp
, semid
);
1654 list_del_rcu(&un
->list_proc
);
1655 list_add_rcu(&un
->list_proc
, &ulp
->list_proc
);
1661 * find_alloc_undo - lookup (and if not present create) undo array
1663 * @semid: semaphore array id
1665 * The function looks up (and if not present creates) the undo structure.
1666 * The size of the undo structure depends on the size of the semaphore
1667 * array, thus the alloc path is not that straightforward.
1668 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1669 * performs a rcu_read_lock().
1671 static struct sem_undo
*find_alloc_undo(struct ipc_namespace
*ns
, int semid
)
1673 struct sem_array
*sma
;
1674 struct sem_undo_list
*ulp
;
1675 struct sem_undo
*un
, *new;
1678 error
= get_undo_list(&ulp
);
1680 return ERR_PTR(error
);
1683 spin_lock(&ulp
->lock
);
1684 un
= lookup_undo(ulp
, semid
);
1685 spin_unlock(&ulp
->lock
);
1686 if (likely(un
!= NULL
))
1689 /* no undo structure around - allocate one. */
1690 /* step 1: figure out the size of the semaphore array */
1691 sma
= sem_obtain_object_check(ns
, semid
);
1694 return ERR_CAST(sma
);
1697 nsems
= sma
->sem_nsems
;
1698 if (!ipc_rcu_getref(sma
)) {
1700 un
= ERR_PTR(-EIDRM
);
1705 /* step 2: allocate new undo structure */
1706 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1708 ipc_rcu_putref(sma
, ipc_rcu_free
);
1709 return ERR_PTR(-ENOMEM
);
1712 /* step 3: Acquire the lock on semaphore array */
1714 sem_lock_and_putref(sma
);
1715 if (!ipc_valid_object(&sma
->sem_perm
)) {
1716 sem_unlock(sma
, -1);
1719 un
= ERR_PTR(-EIDRM
);
1722 spin_lock(&ulp
->lock
);
1725 * step 4: check for races: did someone else allocate the undo struct?
1727 un
= lookup_undo(ulp
, semid
);
1732 /* step 5: initialize & link new undo structure */
1733 new->semadj
= (short *) &new[1];
1736 assert_spin_locked(&ulp
->lock
);
1737 list_add_rcu(&new->list_proc
, &ulp
->list_proc
);
1738 ipc_assert_locked_object(&sma
->sem_perm
);
1739 list_add(&new->list_id
, &sma
->list_id
);
1743 spin_unlock(&ulp
->lock
);
1744 sem_unlock(sma
, -1);
1751 * get_queue_result - retrieve the result code from sem_queue
1752 * @q: Pointer to queue structure
1754 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1755 * q->status, then we must loop until the value is replaced with the final
1756 * value: This may happen if a task is woken up by an unrelated event (e.g.
1757 * signal) and in parallel the task is woken up by another task because it got
1758 * the requested semaphores.
1760 * The function can be called with or without holding the semaphore spinlock.
1762 static int get_queue_result(struct sem_queue
*q
)
1767 while (unlikely(error
== IN_WAKEUP
)) {
1775 SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsops
,
1776 unsigned, nsops
, const struct timespec __user
*, timeout
)
1778 int error
= -EINVAL
;
1779 struct sem_array
*sma
;
1780 struct sembuf fast_sops
[SEMOPM_FAST
];
1781 struct sembuf
*sops
= fast_sops
, *sop
;
1782 struct sem_undo
*un
;
1783 int undos
= 0, alter
= 0, max
, locknum
;
1784 struct sem_queue queue
;
1785 unsigned long jiffies_left
= 0;
1786 struct ipc_namespace
*ns
;
1787 struct list_head tasks
;
1789 ns
= current
->nsproxy
->ipc_ns
;
1791 if (nsops
< 1 || semid
< 0)
1793 if (nsops
> ns
->sc_semopm
)
1795 if (nsops
> SEMOPM_FAST
) {
1796 sops
= kmalloc(sizeof(*sops
)*nsops
, GFP_KERNEL
);
1800 if (copy_from_user(sops
, tsops
, nsops
* sizeof(*tsops
))) {
1805 struct timespec _timeout
;
1806 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1810 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1811 _timeout
.tv_nsec
>= 1000000000L) {
1815 jiffies_left
= timespec_to_jiffies(&_timeout
);
1818 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1819 if (sop
->sem_num
>= max
)
1821 if (sop
->sem_flg
& SEM_UNDO
)
1823 if (sop
->sem_op
!= 0)
1827 INIT_LIST_HEAD(&tasks
);
1830 /* On success, find_alloc_undo takes the rcu_read_lock */
1831 un
= find_alloc_undo(ns
, semid
);
1833 error
= PTR_ERR(un
);
1841 sma
= sem_obtain_object_check(ns
, semid
);
1844 error
= PTR_ERR(sma
);
1849 if (max
>= sma
->sem_nsems
)
1850 goto out_rcu_wakeup
;
1853 if (ipcperms(ns
, &sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
))
1854 goto out_rcu_wakeup
;
1856 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1858 goto out_rcu_wakeup
;
1861 locknum
= sem_lock(sma
, sops
, nsops
);
1863 * We eventually might perform the following check in a lockless
1864 * fashion, considering ipc_valid_object() locking constraints.
1865 * If nsops == 1 and there is no contention for sem_perm.lock, then
1866 * only a per-semaphore lock is held and it's OK to proceed with the
1867 * check below. More details on the fine grained locking scheme
1868 * entangled here and why it's RMID race safe on comments at sem_lock()
1870 if (!ipc_valid_object(&sma
->sem_perm
))
1871 goto out_unlock_free
;
1873 * semid identifiers are not unique - find_alloc_undo may have
1874 * allocated an undo structure, it was invalidated by an RMID
1875 * and now a new array with received the same id. Check and fail.
1876 * This case can be detected checking un->semid. The existence of
1877 * "un" itself is guaranteed by rcu.
1879 if (un
&& un
->semid
== -1)
1880 goto out_unlock_free
;
1883 queue
.nsops
= nsops
;
1885 queue
.pid
= task_tgid_vnr(current
);
1886 queue
.alter
= alter
;
1888 error
= perform_atomic_semop(sma
, &queue
);
1890 /* If the operation was successful, then do
1891 * the required updates.
1894 do_smart_update(sma
, sops
, nsops
, 1, &tasks
);
1896 set_semotime(sma
, sops
);
1899 goto out_unlock_free
;
1901 /* We need to sleep on this operation, so we put the current
1902 * task into the pending queue and go to sleep.
1907 curr
= &sma
->sem_base
[sops
->sem_num
];
1910 if (sma
->complex_count
) {
1911 list_add_tail(&queue
.list
,
1912 &sma
->pending_alter
);
1915 list_add_tail(&queue
.list
,
1916 &curr
->pending_alter
);
1919 list_add_tail(&queue
.list
, &curr
->pending_const
);
1922 if (!sma
->complex_count
)
1926 list_add_tail(&queue
.list
, &sma
->pending_alter
);
1928 list_add_tail(&queue
.list
, &sma
->pending_const
);
1930 sma
->complex_count
++;
1933 queue
.status
= -EINTR
;
1934 queue
.sleeper
= current
;
1937 current
->state
= TASK_INTERRUPTIBLE
;
1938 sem_unlock(sma
, locknum
);
1942 jiffies_left
= schedule_timeout(jiffies_left
);
1946 error
= get_queue_result(&queue
);
1948 if (error
!= -EINTR
) {
1949 /* fast path: update_queue already obtained all requested
1951 * Perform a smp_mb(): User space could assume that semop()
1952 * is a memory barrier: Without the mb(), the cpu could
1953 * speculatively read in user space stale data that was
1954 * overwritten by the previous owner of the semaphore.
1962 sma
= sem_obtain_lock(ns
, semid
, sops
, nsops
, &locknum
);
1965 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1967 error
= get_queue_result(&queue
);
1970 * Array removed? If yes, leave without sem_unlock().
1979 * If queue.status != -EINTR we are woken up by another process.
1980 * Leave without unlink_queue(), but with sem_unlock().
1982 if (error
!= -EINTR
)
1983 goto out_unlock_free
;
1986 * If an interrupt occurred we have to clean up the queue
1988 if (timeout
&& jiffies_left
== 0)
1992 * If the wakeup was spurious, just retry
1994 if (error
== -EINTR
&& !signal_pending(current
))
1997 unlink_queue(sma
, &queue
);
2000 sem_unlock(sma
, locknum
);
2003 wake_up_sem_queue_do(&tasks
);
2005 if (sops
!= fast_sops
)
2010 SYSCALL_DEFINE3(semop
, int, semid
, struct sembuf __user
*, tsops
,
2013 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
2016 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2017 * parent and child tasks.
2020 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
2022 struct sem_undo_list
*undo_list
;
2025 if (clone_flags
& CLONE_SYSVSEM
) {
2026 error
= get_undo_list(&undo_list
);
2029 atomic_inc(&undo_list
->refcnt
);
2030 tsk
->sysvsem
.undo_list
= undo_list
;
2032 tsk
->sysvsem
.undo_list
= NULL
;
2038 * add semadj values to semaphores, free undo structures.
2039 * undo structures are not freed when semaphore arrays are destroyed
2040 * so some of them may be out of date.
2041 * IMPLEMENTATION NOTE: There is some confusion over whether the
2042 * set of adjustments that needs to be done should be done in an atomic
2043 * manner or not. That is, if we are attempting to decrement the semval
2044 * should we queue up and wait until we can do so legally?
2045 * The original implementation attempted to do this (queue and wait).
2046 * The current implementation does not do so. The POSIX standard
2047 * and SVID should be consulted to determine what behavior is mandated.
2049 void exit_sem(struct task_struct
*tsk
)
2051 struct sem_undo_list
*ulp
;
2053 ulp
= tsk
->sysvsem
.undo_list
;
2056 tsk
->sysvsem
.undo_list
= NULL
;
2058 if (!atomic_dec_and_test(&ulp
->refcnt
))
2062 struct sem_array
*sma
;
2063 struct sem_undo
*un
;
2064 struct list_head tasks
;
2068 un
= list_entry_rcu(ulp
->list_proc
.next
,
2069 struct sem_undo
, list_proc
);
2070 if (&un
->list_proc
== &ulp
->list_proc
)
2080 sma
= sem_obtain_object_check(tsk
->nsproxy
->ipc_ns
, un
->semid
);
2081 /* exit_sem raced with IPC_RMID, nothing to do */
2087 sem_lock(sma
, NULL
, -1);
2088 /* exit_sem raced with IPC_RMID, nothing to do */
2089 if (!ipc_valid_object(&sma
->sem_perm
)) {
2090 sem_unlock(sma
, -1);
2094 un
= __lookup_undo(ulp
, semid
);
2096 /* exit_sem raced with IPC_RMID+semget() that created
2097 * exactly the same semid. Nothing to do.
2099 sem_unlock(sma
, -1);
2104 /* remove un from the linked lists */
2105 ipc_assert_locked_object(&sma
->sem_perm
);
2106 list_del(&un
->list_id
);
2108 spin_lock(&ulp
->lock
);
2109 list_del_rcu(&un
->list_proc
);
2110 spin_unlock(&ulp
->lock
);
2112 /* perform adjustments registered in un */
2113 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
2114 struct sem
*semaphore
= &sma
->sem_base
[i
];
2115 if (un
->semadj
[i
]) {
2116 semaphore
->semval
+= un
->semadj
[i
];
2118 * Range checks of the new semaphore value,
2119 * not defined by sus:
2120 * - Some unices ignore the undo entirely
2121 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2122 * - some cap the value (e.g. FreeBSD caps
2123 * at 0, but doesn't enforce SEMVMX)
2125 * Linux caps the semaphore value, both at 0
2128 * Manfred <manfred@colorfullife.com>
2130 if (semaphore
->semval
< 0)
2131 semaphore
->semval
= 0;
2132 if (semaphore
->semval
> SEMVMX
)
2133 semaphore
->semval
= SEMVMX
;
2134 semaphore
->sempid
= task_tgid_vnr(current
);
2137 /* maybe some queued-up processes were waiting for this */
2138 INIT_LIST_HEAD(&tasks
);
2139 do_smart_update(sma
, NULL
, 0, 1, &tasks
);
2140 sem_unlock(sma
, -1);
2142 wake_up_sem_queue_do(&tasks
);
2149 #ifdef CONFIG_PROC_FS
2150 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
2152 struct user_namespace
*user_ns
= seq_user_ns(s
);
2153 struct sem_array
*sma
= it
;
2157 * The proc interface isn't aware of sem_lock(), it calls
2158 * ipc_lock_object() directly (in sysvipc_find_ipc).
2159 * In order to stay compatible with sem_lock(), we must wait until
2160 * all simple semop() calls have left their critical regions.
2162 sem_wait_array(sma
);
2164 sem_otime
= get_semotime(sma
);
2166 return seq_printf(s
,
2167 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2172 from_kuid_munged(user_ns
, sma
->sem_perm
.uid
),
2173 from_kgid_munged(user_ns
, sma
->sem_perm
.gid
),
2174 from_kuid_munged(user_ns
, sma
->sem_perm
.cuid
),
2175 from_kgid_munged(user_ns
, sma
->sem_perm
.cgid
),