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 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
15 * Further wakeup optimizations, documentation
16 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
23 * Pavel Emelianov <xemul@openvz.org>
25 * Implementation notes: (May 2010)
26 * This file implements System V semaphores.
28 * User space visible behavior:
29 * - FIFO ordering for semop() operations (just FIFO, not starvation
31 * - multiple semaphore operations that alter the same semaphore in
32 * one semop() are handled.
33 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
35 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
36 * - undo adjustments at process exit are limited to 0..SEMVMX.
37 * - namespace are supported.
38 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
39 * to /proc/sys/kernel/sem.
40 * - statistics about the usage are reported in /proc/sysvipc/sem.
44 * - all global variables are read-mostly.
45 * - semop() calls and semctl(RMID) are synchronized by RCU.
46 * - most operations do write operations (actually: spin_lock calls) to
47 * the per-semaphore array structure.
48 * Thus: Perfect SMP scaling between independent semaphore arrays.
49 * If multiple semaphores in one array are used, then cache line
50 * trashing on the semaphore array spinlock will limit the scaling.
51 * - semncnt and semzcnt are calculated on demand in count_semcnt()
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 * - 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 * - UNDO values are stored in an array (one per process and per
63 * semaphore array, lazily allocated). For backwards compatibility, multiple
64 * modes for the UNDO variables are supported (per process, per thread)
65 * (see copy_semundo, CLONE_SYSVSEM)
66 * - There are two lists of the pending operations: a per-array list
67 * and per-semaphore list (stored in the array). This allows to achieve FIFO
68 * ordering without always scanning all pending operations.
69 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
72 #include <linux/slab.h>
73 #include <linux/spinlock.h>
74 #include <linux/init.h>
75 #include <linux/proc_fs.h>
76 #include <linux/time.h>
77 #include <linux/security.h>
78 #include <linux/syscalls.h>
79 #include <linux/audit.h>
80 #include <linux/capability.h>
81 #include <linux/seq_file.h>
82 #include <linux/rwsem.h>
83 #include <linux/nsproxy.h>
84 #include <linux/ipc_namespace.h>
85 #include <linux/sched/wake_q.h>
87 #include <linux/uaccess.h>
91 /* One queue for each sleeping process in the system. */
93 struct list_head list
; /* queue of pending operations */
94 struct task_struct
*sleeper
; /* this process */
95 struct sem_undo
*undo
; /* undo structure */
96 int pid
; /* process id of requesting process */
97 int status
; /* completion status of operation */
98 struct sembuf
*sops
; /* array of pending operations */
99 struct sembuf
*blocking
; /* the operation that blocked */
100 int nsops
; /* number of operations */
101 bool alter
; /* does *sops alter the array? */
102 bool dupsop
; /* sops on more than one sem_num */
105 /* Each task has a list of undo requests. They are executed automatically
106 * when the process exits.
109 struct list_head list_proc
; /* per-process list: *
110 * all undos from one process
112 struct rcu_head rcu
; /* rcu struct for sem_undo */
113 struct sem_undo_list
*ulp
; /* back ptr to sem_undo_list */
114 struct list_head list_id
; /* per semaphore array list:
115 * all undos for one array */
116 int semid
; /* semaphore set identifier */
117 short *semadj
; /* array of adjustments */
118 /* one per semaphore */
121 /* sem_undo_list controls shared access to the list of sem_undo structures
122 * that may be shared among all a CLONE_SYSVSEM task group.
124 struct sem_undo_list
{
127 struct list_head list_proc
;
131 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
133 static int newary(struct ipc_namespace
*, struct ipc_params
*);
134 static void freeary(struct ipc_namespace
*, struct kern_ipc_perm
*);
135 #ifdef CONFIG_PROC_FS
136 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
);
139 #define SEMMSL_FAST 256 /* 512 bytes on stack */
140 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
143 * Switching from the mode suitable for simple ops
144 * to the mode for complex ops is costly. Therefore:
145 * use some hysteresis
147 #define USE_GLOBAL_LOCK_HYSTERESIS 10
151 * a) global sem_lock() for read/write
153 * sem_array.complex_count,
154 * sem_array.pending{_alter,_const},
157 * b) global or semaphore sem_lock() for read/write:
158 * sem_array.sems[i].pending_{const,alter}:
161 * sem_undo_list.list_proc:
162 * * undo_list->lock for write
165 * * global sem_lock() for write
166 * * either local or global sem_lock() for read.
169 * Most ordering is enforced by using spin_lock() and spin_unlock().
170 * The special case is use_global_lock:
171 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
172 * using smp_store_release().
173 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
174 * smp_load_acquire().
175 * Setting it from 0 to non-zero must be ordered with regards to
176 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
177 * is inside a spin_lock() and after a write from 0 to non-zero a
178 * spin_lock()+spin_unlock() is done.
181 #define sc_semmsl sem_ctls[0]
182 #define sc_semmns sem_ctls[1]
183 #define sc_semopm sem_ctls[2]
184 #define sc_semmni sem_ctls[3]
186 int sem_init_ns(struct ipc_namespace
*ns
)
188 ns
->sc_semmsl
= SEMMSL
;
189 ns
->sc_semmns
= SEMMNS
;
190 ns
->sc_semopm
= SEMOPM
;
191 ns
->sc_semmni
= SEMMNI
;
193 return ipc_init_ids(&ns
->ids
[IPC_SEM_IDS
]);
197 void sem_exit_ns(struct ipc_namespace
*ns
)
199 free_ipcs(ns
, &sem_ids(ns
), freeary
);
200 idr_destroy(&ns
->ids
[IPC_SEM_IDS
].ipcs_idr
);
201 rhashtable_destroy(&ns
->ids
[IPC_SEM_IDS
].key_ht
);
205 int __init
sem_init(void)
207 const int err
= sem_init_ns(&init_ipc_ns
);
209 ipc_init_proc_interface("sysvipc/sem",
210 " key semid perms nsems uid gid cuid cgid otime ctime\n",
211 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
216 * unmerge_queues - unmerge queues, if possible.
217 * @sma: semaphore array
219 * The function unmerges the wait queues if complex_count is 0.
220 * It must be called prior to dropping the global semaphore array lock.
222 static void unmerge_queues(struct sem_array
*sma
)
224 struct sem_queue
*q
, *tq
;
226 /* complex operations still around? */
227 if (sma
->complex_count
)
230 * We will switch back to simple mode.
231 * Move all pending operation back into the per-semaphore
234 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
236 curr
= &sma
->sems
[q
->sops
[0].sem_num
];
238 list_add_tail(&q
->list
, &curr
->pending_alter
);
240 INIT_LIST_HEAD(&sma
->pending_alter
);
244 * merge_queues - merge single semop queues into global queue
245 * @sma: semaphore array
247 * This function merges all per-semaphore queues into the global queue.
248 * It is necessary to achieve FIFO ordering for the pending single-sop
249 * operations when a multi-semop operation must sleep.
250 * Only the alter operations must be moved, the const operations can stay.
252 static void merge_queues(struct sem_array
*sma
)
255 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
256 struct sem
*sem
= &sma
->sems
[i
];
258 list_splice_init(&sem
->pending_alter
, &sma
->pending_alter
);
262 static void sem_rcu_free(struct rcu_head
*head
)
264 struct kern_ipc_perm
*p
= container_of(head
, struct kern_ipc_perm
, rcu
);
265 struct sem_array
*sma
= container_of(p
, struct sem_array
, sem_perm
);
267 security_sem_free(sma
);
272 * Enter the mode suitable for non-simple operations:
273 * Caller must own sem_perm.lock.
275 static void complexmode_enter(struct sem_array
*sma
)
280 if (sma
->use_global_lock
> 0) {
282 * We are already in global lock mode.
283 * Nothing to do, just reset the
284 * counter until we return to simple mode.
286 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
289 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
291 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
293 spin_lock(&sem
->lock
);
294 spin_unlock(&sem
->lock
);
299 * Try to leave the mode that disallows simple operations:
300 * Caller must own sem_perm.lock.
302 static void complexmode_tryleave(struct sem_array
*sma
)
304 if (sma
->complex_count
) {
305 /* Complex ops are sleeping.
306 * We must stay in complex mode
310 if (sma
->use_global_lock
== 1) {
312 * Immediately after setting use_global_lock to 0,
313 * a simple op can start. Thus: all memory writes
314 * performed by the current operation must be visible
315 * before we set use_global_lock to 0.
317 smp_store_release(&sma
->use_global_lock
, 0);
319 sma
->use_global_lock
--;
323 #define SEM_GLOBAL_LOCK (-1)
325 * If the request contains only one semaphore operation, and there are
326 * no complex transactions pending, lock only the semaphore involved.
327 * Otherwise, lock the entire semaphore array, since we either have
328 * multiple semaphores in our own semops, or we need to look at
329 * semaphores from other pending complex operations.
331 static inline int sem_lock(struct sem_array
*sma
, struct sembuf
*sops
,
337 /* Complex operation - acquire a full lock */
338 ipc_lock_object(&sma
->sem_perm
);
340 /* Prevent parallel simple ops */
341 complexmode_enter(sma
);
342 return SEM_GLOBAL_LOCK
;
346 * Only one semaphore affected - try to optimize locking.
347 * Optimized locking is possible if no complex operation
348 * is either enqueued or processed right now.
350 * Both facts are tracked by use_global_mode.
352 sem
= &sma
->sems
[sops
->sem_num
];
355 * Initial check for use_global_lock. Just an optimization,
356 * no locking, no memory barrier.
358 if (!sma
->use_global_lock
) {
360 * It appears that no complex operation is around.
361 * Acquire the per-semaphore lock.
363 spin_lock(&sem
->lock
);
365 /* pairs with smp_store_release() */
366 if (!smp_load_acquire(&sma
->use_global_lock
)) {
367 /* fast path successful! */
368 return sops
->sem_num
;
370 spin_unlock(&sem
->lock
);
373 /* slow path: acquire the full lock */
374 ipc_lock_object(&sma
->sem_perm
);
376 if (sma
->use_global_lock
== 0) {
378 * The use_global_lock mode ended while we waited for
379 * sma->sem_perm.lock. Thus we must switch to locking
381 * Unlike in the fast path, there is no need to recheck
382 * sma->use_global_lock after we have acquired sem->lock:
383 * We own sma->sem_perm.lock, thus use_global_lock cannot
386 spin_lock(&sem
->lock
);
388 ipc_unlock_object(&sma
->sem_perm
);
389 return sops
->sem_num
;
392 * Not a false alarm, thus continue to use the global lock
393 * mode. No need for complexmode_enter(), this was done by
394 * the caller that has set use_global_mode to non-zero.
396 return SEM_GLOBAL_LOCK
;
400 static inline void sem_unlock(struct sem_array
*sma
, int locknum
)
402 if (locknum
== SEM_GLOBAL_LOCK
) {
404 complexmode_tryleave(sma
);
405 ipc_unlock_object(&sma
->sem_perm
);
407 struct sem
*sem
= &sma
->sems
[locknum
];
408 spin_unlock(&sem
->lock
);
413 * sem_lock_(check_) routines are called in the paths where the rwsem
416 * The caller holds the RCU read lock.
418 static inline struct sem_array
*sem_obtain_object(struct ipc_namespace
*ns
, int id
)
420 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&sem_ids(ns
), id
);
423 return ERR_CAST(ipcp
);
425 return container_of(ipcp
, struct sem_array
, sem_perm
);
428 static inline struct sem_array
*sem_obtain_object_check(struct ipc_namespace
*ns
,
431 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&sem_ids(ns
), id
);
434 return ERR_CAST(ipcp
);
436 return container_of(ipcp
, struct sem_array
, sem_perm
);
439 static inline void sem_lock_and_putref(struct sem_array
*sma
)
441 sem_lock(sma
, NULL
, -1);
442 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
445 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
447 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
450 static struct sem_array
*sem_alloc(size_t nsems
)
452 struct sem_array
*sma
;
455 if (nsems
> (INT_MAX
- sizeof(*sma
)) / sizeof(sma
->sems
[0]))
458 size
= sizeof(*sma
) + nsems
* sizeof(sma
->sems
[0]);
459 sma
= kvmalloc(size
, GFP_KERNEL
);
463 memset(sma
, 0, size
);
469 * newary - Create a new semaphore set
471 * @params: ptr to the structure that contains key, semflg and nsems
473 * Called with sem_ids.rwsem held (as a writer)
475 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
478 struct sem_array
*sma
;
479 key_t key
= params
->key
;
480 int nsems
= params
->u
.nsems
;
481 int semflg
= params
->flg
;
486 if (ns
->used_sems
+ nsems
> ns
->sc_semmns
)
489 sma
= sem_alloc(nsems
);
493 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
494 sma
->sem_perm
.key
= key
;
496 sma
->sem_perm
.security
= NULL
;
497 retval
= security_sem_alloc(sma
);
503 for (i
= 0; i
< nsems
; i
++) {
504 INIT_LIST_HEAD(&sma
->sems
[i
].pending_alter
);
505 INIT_LIST_HEAD(&sma
->sems
[i
].pending_const
);
506 spin_lock_init(&sma
->sems
[i
].lock
);
509 sma
->complex_count
= 0;
510 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
511 INIT_LIST_HEAD(&sma
->pending_alter
);
512 INIT_LIST_HEAD(&sma
->pending_const
);
513 INIT_LIST_HEAD(&sma
->list_id
);
514 sma
->sem_nsems
= nsems
;
515 sma
->sem_ctime
= ktime_get_real_seconds();
517 retval
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
519 call_rcu(&sma
->sem_perm
.rcu
, sem_rcu_free
);
522 ns
->used_sems
+= nsems
;
527 return sma
->sem_perm
.id
;
532 * Called with sem_ids.rwsem and ipcp locked.
534 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
536 struct sem_array
*sma
;
538 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
539 return security_sem_associate(sma
, semflg
);
543 * Called with sem_ids.rwsem and ipcp locked.
545 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
546 struct ipc_params
*params
)
548 struct sem_array
*sma
;
550 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
551 if (params
->u
.nsems
> sma
->sem_nsems
)
557 SYSCALL_DEFINE3(semget
, key_t
, key
, int, nsems
, int, semflg
)
559 struct ipc_namespace
*ns
;
560 static const struct ipc_ops sem_ops
= {
562 .associate
= sem_security
,
563 .more_checks
= sem_more_checks
,
565 struct ipc_params sem_params
;
567 ns
= current
->nsproxy
->ipc_ns
;
569 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
572 sem_params
.key
= key
;
573 sem_params
.flg
= semflg
;
574 sem_params
.u
.nsems
= nsems
;
576 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
580 * perform_atomic_semop[_slow] - Attempt to perform semaphore
581 * operations on a given array.
582 * @sma: semaphore array
583 * @q: struct sem_queue that describes the operation
585 * Caller blocking are as follows, based the value
586 * indicated by the semaphore operation (sem_op):
588 * (1) >0 never blocks.
589 * (2) 0 (wait-for-zero operation): semval is non-zero.
590 * (3) <0 attempting to decrement semval to a value smaller than zero.
592 * Returns 0 if the operation was possible.
593 * Returns 1 if the operation is impossible, the caller must sleep.
594 * Returns <0 for error codes.
596 static int perform_atomic_semop_slow(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
->sems
[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
->sems
[sop
->sem_num
].sempid
= pid
;
649 if (sop
->sem_flg
& IPC_NOWAIT
)
656 while (sop
>= sops
) {
657 sem_op
= sop
->sem_op
;
658 sma
->sems
[sop
->sem_num
].semval
-= sem_op
;
659 if (sop
->sem_flg
& SEM_UNDO
)
660 un
->semadj
[sop
->sem_num
] += sem_op
;
667 static int perform_atomic_semop(struct sem_array
*sma
, struct sem_queue
*q
)
669 int result
, sem_op
, nsops
;
679 if (unlikely(q
->dupsop
))
680 return perform_atomic_semop_slow(sma
, q
);
683 * We scan the semaphore set twice, first to ensure that the entire
684 * operation can succeed, therefore avoiding any pointless writes
685 * to shared memory and having to undo such changes in order to block
686 * until the operations can go through.
688 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
689 curr
= &sma
->sems
[sop
->sem_num
];
690 sem_op
= sop
->sem_op
;
691 result
= curr
->semval
;
693 if (!sem_op
&& result
)
694 goto would_block
; /* wait-for-zero */
703 if (sop
->sem_flg
& SEM_UNDO
) {
704 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
706 /* Exceeding the undo range is an error. */
707 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
712 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
713 curr
= &sma
->sems
[sop
->sem_num
];
714 sem_op
= sop
->sem_op
;
715 result
= curr
->semval
;
717 if (sop
->sem_flg
& SEM_UNDO
) {
718 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
720 un
->semadj
[sop
->sem_num
] = undo
;
722 curr
->semval
+= sem_op
;
723 curr
->sempid
= q
->pid
;
730 return sop
->sem_flg
& IPC_NOWAIT
? -EAGAIN
: 1;
733 static inline void wake_up_sem_queue_prepare(struct sem_queue
*q
, int error
,
734 struct wake_q_head
*wake_q
)
736 wake_q_add(wake_q
, q
->sleeper
);
738 * Rely on the above implicit barrier, such that we can
739 * ensure that we hold reference to the task before setting
740 * q->status. Otherwise we could race with do_exit if the
741 * task is awoken by an external event before calling
744 WRITE_ONCE(q
->status
, error
);
747 static void unlink_queue(struct sem_array
*sma
, struct sem_queue
*q
)
751 sma
->complex_count
--;
754 /** check_restart(sma, q)
755 * @sma: semaphore array
756 * @q: the operation that just completed
758 * update_queue is O(N^2) when it restarts scanning the whole queue of
759 * waiting operations. Therefore this function checks if the restart is
760 * really necessary. It is called after a previously waiting operation
761 * modified the array.
762 * Note that wait-for-zero operations are handled without restart.
764 static inline int check_restart(struct sem_array
*sma
, struct sem_queue
*q
)
766 /* pending complex alter operations are too difficult to analyse */
767 if (!list_empty(&sma
->pending_alter
))
770 /* we were a sleeping complex operation. Too difficult */
774 /* It is impossible that someone waits for the new value:
775 * - complex operations always restart.
776 * - wait-for-zero are handled seperately.
777 * - q is a previously sleeping simple operation that
778 * altered the array. It must be a decrement, because
779 * simple increments never sleep.
780 * - If there are older (higher priority) decrements
781 * in the queue, then they have observed the original
782 * semval value and couldn't proceed. The operation
783 * decremented to value - thus they won't proceed either.
789 * wake_const_ops - wake up non-alter tasks
790 * @sma: semaphore array.
791 * @semnum: semaphore that was modified.
792 * @wake_q: lockless wake-queue head.
794 * wake_const_ops must be called after a semaphore in a semaphore array
795 * was set to 0. If complex const operations are pending, wake_const_ops must
796 * be called with semnum = -1, as well as with the number of each modified
798 * The tasks that must be woken up are added to @wake_q. The return code
799 * is stored in q->pid.
800 * The function returns 1 if at least one operation was completed successfully.
802 static int wake_const_ops(struct sem_array
*sma
, int semnum
,
803 struct wake_q_head
*wake_q
)
805 struct sem_queue
*q
, *tmp
;
806 struct list_head
*pending_list
;
807 int semop_completed
= 0;
810 pending_list
= &sma
->pending_const
;
812 pending_list
= &sma
->sems
[semnum
].pending_const
;
814 list_for_each_entry_safe(q
, tmp
, pending_list
, list
) {
815 int error
= perform_atomic_semop(sma
, q
);
819 /* operation completed, remove from queue & wakeup */
820 unlink_queue(sma
, q
);
822 wake_up_sem_queue_prepare(q
, error
, wake_q
);
827 return semop_completed
;
831 * do_smart_wakeup_zero - wakeup all wait for zero tasks
832 * @sma: semaphore array
833 * @sops: operations that were performed
834 * @nsops: number of operations
835 * @wake_q: lockless wake-queue head
837 * Checks all required queue for wait-for-zero operations, based
838 * on the actual changes that were performed on the semaphore array.
839 * The function returns 1 if at least one operation was completed successfully.
841 static int do_smart_wakeup_zero(struct sem_array
*sma
, struct sembuf
*sops
,
842 int nsops
, struct wake_q_head
*wake_q
)
845 int semop_completed
= 0;
848 /* first: the per-semaphore queues, if known */
850 for (i
= 0; i
< nsops
; i
++) {
851 int num
= sops
[i
].sem_num
;
853 if (sma
->sems
[num
].semval
== 0) {
855 semop_completed
|= wake_const_ops(sma
, num
, wake_q
);
860 * No sops means modified semaphores not known.
861 * Assume all were changed.
863 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
864 if (sma
->sems
[i
].semval
== 0) {
866 semop_completed
|= wake_const_ops(sma
, i
, wake_q
);
871 * If one of the modified semaphores got 0,
872 * then check the global queue, too.
875 semop_completed
|= wake_const_ops(sma
, -1, wake_q
);
877 return semop_completed
;
882 * update_queue - look for tasks that can be completed.
883 * @sma: semaphore array.
884 * @semnum: semaphore that was modified.
885 * @wake_q: lockless wake-queue head.
887 * update_queue must be called after a semaphore in a semaphore array
888 * was modified. If multiple semaphores were modified, update_queue must
889 * be called with semnum = -1, as well as with the number of each modified
891 * The tasks that must be woken up are added to @wake_q. The return code
892 * is stored in q->pid.
893 * The function internally checks if const operations can now succeed.
895 * The function return 1 if at least one semop was completed successfully.
897 static int update_queue(struct sem_array
*sma
, int semnum
, struct wake_q_head
*wake_q
)
899 struct sem_queue
*q
, *tmp
;
900 struct list_head
*pending_list
;
901 int semop_completed
= 0;
904 pending_list
= &sma
->pending_alter
;
906 pending_list
= &sma
->sems
[semnum
].pending_alter
;
909 list_for_each_entry_safe(q
, tmp
, pending_list
, list
) {
912 /* If we are scanning the single sop, per-semaphore list of
913 * one semaphore and that semaphore is 0, then it is not
914 * necessary to scan further: simple increments
915 * that affect only one entry succeed immediately and cannot
916 * be in the per semaphore pending queue, and decrements
917 * cannot be successful if the value is already 0.
919 if (semnum
!= -1 && sma
->sems
[semnum
].semval
== 0)
922 error
= perform_atomic_semop(sma
, q
);
924 /* Does q->sleeper still need to sleep? */
928 unlink_queue(sma
, q
);
934 do_smart_wakeup_zero(sma
, q
->sops
, q
->nsops
, wake_q
);
935 restart
= check_restart(sma
, q
);
938 wake_up_sem_queue_prepare(q
, error
, wake_q
);
942 return semop_completed
;
946 * set_semotime - set sem_otime
947 * @sma: semaphore array
948 * @sops: operations that modified the array, may be NULL
950 * sem_otime is replicated to avoid cache line trashing.
951 * This function sets one instance to the current time.
953 static void set_semotime(struct sem_array
*sma
, struct sembuf
*sops
)
956 sma
->sems
[0].sem_otime
= get_seconds();
958 sma
->sems
[sops
[0].sem_num
].sem_otime
=
964 * do_smart_update - optimized update_queue
965 * @sma: semaphore array
966 * @sops: operations that were performed
967 * @nsops: number of operations
968 * @otime: force setting otime
969 * @wake_q: lockless wake-queue head
971 * do_smart_update() does the required calls to update_queue and wakeup_zero,
972 * based on the actual changes that were performed on the semaphore array.
973 * Note that the function does not do the actual wake-up: the caller is
974 * responsible for calling wake_up_q().
975 * It is safe to perform this call after dropping all locks.
977 static void do_smart_update(struct sem_array
*sma
, struct sembuf
*sops
, int nsops
,
978 int otime
, struct wake_q_head
*wake_q
)
982 otime
|= do_smart_wakeup_zero(sma
, sops
, nsops
, wake_q
);
984 if (!list_empty(&sma
->pending_alter
)) {
985 /* semaphore array uses the global queue - just process it. */
986 otime
|= update_queue(sma
, -1, wake_q
);
990 * No sops, thus the modified semaphores are not
993 for (i
= 0; i
< sma
->sem_nsems
; i
++)
994 otime
|= update_queue(sma
, i
, wake_q
);
997 * Check the semaphores that were increased:
998 * - No complex ops, thus all sleeping ops are
1000 * - if we decreased the value, then any sleeping
1001 * semaphore ops wont be able to run: If the
1002 * previous value was too small, then the new
1003 * value will be too small, too.
1005 for (i
= 0; i
< nsops
; i
++) {
1006 if (sops
[i
].sem_op
> 0) {
1007 otime
|= update_queue(sma
,
1008 sops
[i
].sem_num
, wake_q
);
1014 set_semotime(sma
, sops
);
1018 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1020 static int check_qop(struct sem_array
*sma
, int semnum
, struct sem_queue
*q
,
1023 struct sembuf
*sop
= q
->blocking
;
1026 * Linux always (since 0.99.10) reported a task as sleeping on all
1027 * semaphores. This violates SUS, therefore it was changed to the
1028 * standard compliant behavior.
1029 * Give the administrators a chance to notice that an application
1030 * might misbehave because it relies on the Linux behavior.
1032 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1033 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1034 current
->comm
, task_pid_nr(current
));
1036 if (sop
->sem_num
!= semnum
)
1039 if (count_zero
&& sop
->sem_op
== 0)
1041 if (!count_zero
&& sop
->sem_op
< 0)
1047 /* The following counts are associated to each semaphore:
1048 * semncnt number of tasks waiting on semval being nonzero
1049 * semzcnt number of tasks waiting on semval being zero
1051 * Per definition, a task waits only on the semaphore of the first semop
1052 * that cannot proceed, even if additional operation would block, too.
1054 static int count_semcnt(struct sem_array
*sma
, ushort semnum
,
1057 struct list_head
*l
;
1058 struct sem_queue
*q
;
1062 /* First: check the simple operations. They are easy to evaluate */
1064 l
= &sma
->sems
[semnum
].pending_const
;
1066 l
= &sma
->sems
[semnum
].pending_alter
;
1068 list_for_each_entry(q
, l
, list
) {
1069 /* all task on a per-semaphore list sleep on exactly
1075 /* Then: check the complex operations. */
1076 list_for_each_entry(q
, &sma
->pending_alter
, list
) {
1077 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1080 list_for_each_entry(q
, &sma
->pending_const
, list
) {
1081 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1087 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1088 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1089 * remains locked on exit.
1091 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
1093 struct sem_undo
*un
, *tu
;
1094 struct sem_queue
*q
, *tq
;
1095 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1097 DEFINE_WAKE_Q(wake_q
);
1099 /* Free the existing undo structures for this semaphore set. */
1100 ipc_assert_locked_object(&sma
->sem_perm
);
1101 list_for_each_entry_safe(un
, tu
, &sma
->list_id
, list_id
) {
1102 list_del(&un
->list_id
);
1103 spin_lock(&un
->ulp
->lock
);
1105 list_del_rcu(&un
->list_proc
);
1106 spin_unlock(&un
->ulp
->lock
);
1110 /* Wake up all pending processes and let them fail with EIDRM. */
1111 list_for_each_entry_safe(q
, tq
, &sma
->pending_const
, list
) {
1112 unlink_queue(sma
, q
);
1113 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1116 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
1117 unlink_queue(sma
, q
);
1118 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1120 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
1121 struct sem
*sem
= &sma
->sems
[i
];
1122 list_for_each_entry_safe(q
, tq
, &sem
->pending_const
, list
) {
1123 unlink_queue(sma
, q
);
1124 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1126 list_for_each_entry_safe(q
, tq
, &sem
->pending_alter
, list
) {
1127 unlink_queue(sma
, q
);
1128 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1132 /* Remove the semaphore set from the IDR */
1134 sem_unlock(sma
, -1);
1138 ns
->used_sems
-= sma
->sem_nsems
;
1139 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
1142 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
1146 return copy_to_user(buf
, in
, sizeof(*in
));
1149 struct semid_ds out
;
1151 memset(&out
, 0, sizeof(out
));
1153 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
1155 out
.sem_otime
= in
->sem_otime
;
1156 out
.sem_ctime
= in
->sem_ctime
;
1157 out
.sem_nsems
= in
->sem_nsems
;
1159 return copy_to_user(buf
, &out
, sizeof(out
));
1166 static time64_t
get_semotime(struct sem_array
*sma
)
1171 res
= sma
->sems
[0].sem_otime
;
1172 for (i
= 1; i
< sma
->sem_nsems
; i
++) {
1173 time64_t to
= sma
->sems
[i
].sem_otime
;
1181 static int semctl_stat(struct ipc_namespace
*ns
, int semid
,
1182 int cmd
, struct semid64_ds
*semid64
)
1184 struct sem_array
*sma
;
1188 memset(semid64
, 0, sizeof(*semid64
));
1191 if (cmd
== SEM_STAT
) {
1192 sma
= sem_obtain_object(ns
, semid
);
1197 id
= sma
->sem_perm
.id
;
1199 sma
= sem_obtain_object_check(ns
, semid
);
1207 if (ipcperms(ns
, &sma
->sem_perm
, S_IRUGO
))
1210 err
= security_sem_semctl(sma
, cmd
);
1214 kernel_to_ipc64_perm(&sma
->sem_perm
, &semid64
->sem_perm
);
1215 semid64
->sem_otime
= get_semotime(sma
);
1216 semid64
->sem_ctime
= sma
->sem_ctime
;
1217 semid64
->sem_nsems
= sma
->sem_nsems
;
1226 static int semctl_info(struct ipc_namespace
*ns
, int semid
,
1227 int cmd
, void __user
*p
)
1229 struct seminfo seminfo
;
1233 err
= security_sem_semctl(NULL
, cmd
);
1237 memset(&seminfo
, 0, sizeof(seminfo
));
1238 seminfo
.semmni
= ns
->sc_semmni
;
1239 seminfo
.semmns
= ns
->sc_semmns
;
1240 seminfo
.semmsl
= ns
->sc_semmsl
;
1241 seminfo
.semopm
= ns
->sc_semopm
;
1242 seminfo
.semvmx
= SEMVMX
;
1243 seminfo
.semmnu
= SEMMNU
;
1244 seminfo
.semmap
= SEMMAP
;
1245 seminfo
.semume
= SEMUME
;
1246 down_read(&sem_ids(ns
).rwsem
);
1247 if (cmd
== SEM_INFO
) {
1248 seminfo
.semusz
= sem_ids(ns
).in_use
;
1249 seminfo
.semaem
= ns
->used_sems
;
1251 seminfo
.semusz
= SEMUSZ
;
1252 seminfo
.semaem
= SEMAEM
;
1254 max_id
= ipc_get_maxid(&sem_ids(ns
));
1255 up_read(&sem_ids(ns
).rwsem
);
1256 if (copy_to_user(p
, &seminfo
, sizeof(struct seminfo
)))
1258 return (max_id
< 0) ? 0 : max_id
;
1261 static int semctl_setval(struct ipc_namespace
*ns
, int semid
, int semnum
,
1264 struct sem_undo
*un
;
1265 struct sem_array
*sma
;
1268 DEFINE_WAKE_Q(wake_q
);
1270 if (val
> SEMVMX
|| val
< 0)
1274 sma
= sem_obtain_object_check(ns
, semid
);
1277 return PTR_ERR(sma
);
1280 if (semnum
< 0 || semnum
>= sma
->sem_nsems
) {
1286 if (ipcperms(ns
, &sma
->sem_perm
, S_IWUGO
)) {
1291 err
= security_sem_semctl(sma
, SETVAL
);
1297 sem_lock(sma
, NULL
, -1);
1299 if (!ipc_valid_object(&sma
->sem_perm
)) {
1300 sem_unlock(sma
, -1);
1305 curr
= &sma
->sems
[semnum
];
1307 ipc_assert_locked_object(&sma
->sem_perm
);
1308 list_for_each_entry(un
, &sma
->list_id
, list_id
)
1309 un
->semadj
[semnum
] = 0;
1312 curr
->sempid
= task_tgid_vnr(current
);
1313 sma
->sem_ctime
= ktime_get_real_seconds();
1314 /* maybe some queued-up processes were waiting for this */
1315 do_smart_update(sma
, NULL
, 0, 0, &wake_q
);
1316 sem_unlock(sma
, -1);
1322 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
1323 int cmd
, void __user
*p
)
1325 struct sem_array
*sma
;
1328 ushort fast_sem_io
[SEMMSL_FAST
];
1329 ushort
*sem_io
= fast_sem_io
;
1330 DEFINE_WAKE_Q(wake_q
);
1333 sma
= sem_obtain_object_check(ns
, semid
);
1336 return PTR_ERR(sma
);
1339 nsems
= sma
->sem_nsems
;
1342 if (ipcperms(ns
, &sma
->sem_perm
, cmd
== SETALL
? S_IWUGO
: S_IRUGO
))
1343 goto out_rcu_wakeup
;
1345 err
= security_sem_semctl(sma
, cmd
);
1347 goto out_rcu_wakeup
;
1353 ushort __user
*array
= p
;
1356 sem_lock(sma
, NULL
, -1);
1357 if (!ipc_valid_object(&sma
->sem_perm
)) {
1361 if (nsems
> SEMMSL_FAST
) {
1362 if (!ipc_rcu_getref(&sma
->sem_perm
)) {
1366 sem_unlock(sma
, -1);
1368 sem_io
= kvmalloc_array(nsems
, sizeof(ushort
),
1370 if (sem_io
== NULL
) {
1371 ipc_rcu_putref(&sma
->sem_perm
, sem_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
->sems
[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
->sem_perm
)) {
1398 goto out_rcu_wakeup
;
1402 if (nsems
> SEMMSL_FAST
) {
1403 sem_io
= kvmalloc_array(nsems
, sizeof(ushort
),
1405 if (sem_io
== NULL
) {
1406 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
1411 if (copy_from_user(sem_io
, p
, nsems
*sizeof(ushort
))) {
1412 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
1417 for (i
= 0; i
< nsems
; i
++) {
1418 if (sem_io
[i
] > SEMVMX
) {
1419 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
1425 sem_lock_and_putref(sma
);
1426 if (!ipc_valid_object(&sma
->sem_perm
)) {
1431 for (i
= 0; i
< nsems
; i
++) {
1432 sma
->sems
[i
].semval
= sem_io
[i
];
1433 sma
->sems
[i
].sempid
= task_tgid_vnr(current
);
1436 ipc_assert_locked_object(&sma
->sem_perm
);
1437 list_for_each_entry(un
, &sma
->list_id
, list_id
) {
1438 for (i
= 0; i
< nsems
; i
++)
1441 sma
->sem_ctime
= ktime_get_real_seconds();
1442 /* maybe some queued-up processes were waiting for this */
1443 do_smart_update(sma
, NULL
, 0, 0, &wake_q
);
1447 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1450 if (semnum
< 0 || semnum
>= nsems
)
1451 goto out_rcu_wakeup
;
1453 sem_lock(sma
, NULL
, -1);
1454 if (!ipc_valid_object(&sma
->sem_perm
)) {
1458 curr
= &sma
->sems
[semnum
];
1468 err
= count_semcnt(sma
, semnum
, 0);
1471 err
= count_semcnt(sma
, semnum
, 1);
1476 sem_unlock(sma
, -1);
1481 if (sem_io
!= fast_sem_io
)
1486 static inline unsigned long
1487 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
1491 if (copy_from_user(out
, buf
, sizeof(*out
)))
1496 struct semid_ds tbuf_old
;
1498 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
1501 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
1502 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
1503 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
1513 * This function handles some semctl commands which require the rwsem
1514 * to be held in write mode.
1515 * NOTE: no locks must be held, the rwsem is taken inside this function.
1517 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
1518 int cmd
, struct semid64_ds
*semid64
)
1520 struct sem_array
*sma
;
1522 struct kern_ipc_perm
*ipcp
;
1524 down_write(&sem_ids(ns
).rwsem
);
1527 ipcp
= ipcctl_pre_down_nolock(ns
, &sem_ids(ns
), semid
, cmd
,
1528 &semid64
->sem_perm
, 0);
1530 err
= PTR_ERR(ipcp
);
1534 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1536 err
= security_sem_semctl(sma
, cmd
);
1542 sem_lock(sma
, NULL
, -1);
1543 /* freeary unlocks the ipc object and rcu */
1547 sem_lock(sma
, NULL
, -1);
1548 err
= ipc_update_perm(&semid64
->sem_perm
, ipcp
);
1551 sma
->sem_ctime
= ktime_get_real_seconds();
1559 sem_unlock(sma
, -1);
1563 up_write(&sem_ids(ns
).rwsem
);
1567 SYSCALL_DEFINE4(semctl
, int, semid
, int, semnum
, int, cmd
, unsigned long, arg
)
1570 struct ipc_namespace
*ns
;
1571 void __user
*p
= (void __user
*)arg
;
1572 struct semid64_ds semid64
;
1578 version
= ipc_parse_version(&cmd
);
1579 ns
= current
->nsproxy
->ipc_ns
;
1584 return semctl_info(ns
, semid
, cmd
, p
);
1587 err
= semctl_stat(ns
, semid
, cmd
, &semid64
);
1590 if (copy_semid_to_user(p
, &semid64
, version
))
1599 return semctl_main(ns
, semid
, semnum
, cmd
, p
);
1602 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1603 /* big-endian 64bit */
1606 /* 32bit or little-endian 64bit */
1609 return semctl_setval(ns
, semid
, semnum
, val
);
1612 if (copy_semid_from_user(&semid64
, p
, version
))
1615 return semctl_down(ns
, semid
, cmd
, &semid64
);
1621 #ifdef CONFIG_COMPAT
1623 struct compat_semid_ds
{
1624 struct compat_ipc_perm sem_perm
;
1625 compat_time_t sem_otime
;
1626 compat_time_t sem_ctime
;
1627 compat_uptr_t sem_base
;
1628 compat_uptr_t sem_pending
;
1629 compat_uptr_t sem_pending_last
;
1631 unsigned short sem_nsems
;
1634 static int copy_compat_semid_from_user(struct semid64_ds
*out
, void __user
*buf
,
1637 memset(out
, 0, sizeof(*out
));
1638 if (version
== IPC_64
) {
1639 struct compat_semid64_ds
*p
= buf
;
1640 return get_compat_ipc64_perm(&out
->sem_perm
, &p
->sem_perm
);
1642 struct compat_semid_ds
*p
= buf
;
1643 return get_compat_ipc_perm(&out
->sem_perm
, &p
->sem_perm
);
1647 static int copy_compat_semid_to_user(void __user
*buf
, struct semid64_ds
*in
,
1650 if (version
== IPC_64
) {
1651 struct compat_semid64_ds v
;
1652 memset(&v
, 0, sizeof(v
));
1653 to_compat_ipc64_perm(&v
.sem_perm
, &in
->sem_perm
);
1654 v
.sem_otime
= in
->sem_otime
;
1655 v
.sem_ctime
= in
->sem_ctime
;
1656 v
.sem_nsems
= in
->sem_nsems
;
1657 return copy_to_user(buf
, &v
, sizeof(v
));
1659 struct compat_semid_ds v
;
1660 memset(&v
, 0, sizeof(v
));
1661 to_compat_ipc_perm(&v
.sem_perm
, &in
->sem_perm
);
1662 v
.sem_otime
= in
->sem_otime
;
1663 v
.sem_ctime
= in
->sem_ctime
;
1664 v
.sem_nsems
= in
->sem_nsems
;
1665 return copy_to_user(buf
, &v
, sizeof(v
));
1669 COMPAT_SYSCALL_DEFINE4(semctl
, int, semid
, int, semnum
, int, cmd
, int, arg
)
1671 void __user
*p
= compat_ptr(arg
);
1672 struct ipc_namespace
*ns
;
1673 struct semid64_ds semid64
;
1674 int version
= compat_ipc_parse_version(&cmd
);
1677 ns
= current
->nsproxy
->ipc_ns
;
1682 switch (cmd
& (~IPC_64
)) {
1685 return semctl_info(ns
, semid
, cmd
, p
);
1688 err
= semctl_stat(ns
, semid
, cmd
, &semid64
);
1691 if (copy_compat_semid_to_user(p
, &semid64
, version
))
1700 return semctl_main(ns
, semid
, semnum
, cmd
, p
);
1702 return semctl_setval(ns
, semid
, semnum
, arg
);
1704 if (copy_compat_semid_from_user(&semid64
, p
, version
))
1708 return semctl_down(ns
, semid
, cmd
, &semid64
);
1715 /* If the task doesn't already have a undo_list, then allocate one
1716 * here. We guarantee there is only one thread using this undo list,
1717 * and current is THE ONE
1719 * If this allocation and assignment succeeds, but later
1720 * portions of this code fail, there is no need to free the sem_undo_list.
1721 * Just let it stay associated with the task, and it'll be freed later
1724 * This can block, so callers must hold no locks.
1726 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
1728 struct sem_undo_list
*undo_list
;
1730 undo_list
= current
->sysvsem
.undo_list
;
1732 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
1733 if (undo_list
== NULL
)
1735 spin_lock_init(&undo_list
->lock
);
1736 refcount_set(&undo_list
->refcnt
, 1);
1737 INIT_LIST_HEAD(&undo_list
->list_proc
);
1739 current
->sysvsem
.undo_list
= undo_list
;
1741 *undo_listp
= undo_list
;
1745 static struct sem_undo
*__lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1747 struct sem_undo
*un
;
1749 list_for_each_entry_rcu(un
, &ulp
->list_proc
, list_proc
) {
1750 if (un
->semid
== semid
)
1756 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1758 struct sem_undo
*un
;
1760 assert_spin_locked(&ulp
->lock
);
1762 un
= __lookup_undo(ulp
, semid
);
1764 list_del_rcu(&un
->list_proc
);
1765 list_add_rcu(&un
->list_proc
, &ulp
->list_proc
);
1771 * find_alloc_undo - lookup (and if not present create) undo array
1773 * @semid: semaphore array id
1775 * The function looks up (and if not present creates) the undo structure.
1776 * The size of the undo structure depends on the size of the semaphore
1777 * array, thus the alloc path is not that straightforward.
1778 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1779 * performs a rcu_read_lock().
1781 static struct sem_undo
*find_alloc_undo(struct ipc_namespace
*ns
, int semid
)
1783 struct sem_array
*sma
;
1784 struct sem_undo_list
*ulp
;
1785 struct sem_undo
*un
, *new;
1788 error
= get_undo_list(&ulp
);
1790 return ERR_PTR(error
);
1793 spin_lock(&ulp
->lock
);
1794 un
= lookup_undo(ulp
, semid
);
1795 spin_unlock(&ulp
->lock
);
1796 if (likely(un
!= NULL
))
1799 /* no undo structure around - allocate one. */
1800 /* step 1: figure out the size of the semaphore array */
1801 sma
= sem_obtain_object_check(ns
, semid
);
1804 return ERR_CAST(sma
);
1807 nsems
= sma
->sem_nsems
;
1808 if (!ipc_rcu_getref(&sma
->sem_perm
)) {
1810 un
= ERR_PTR(-EIDRM
);
1815 /* step 2: allocate new undo structure */
1816 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1818 ipc_rcu_putref(&sma
->sem_perm
, sem_rcu_free
);
1819 return ERR_PTR(-ENOMEM
);
1822 /* step 3: Acquire the lock on semaphore array */
1824 sem_lock_and_putref(sma
);
1825 if (!ipc_valid_object(&sma
->sem_perm
)) {
1826 sem_unlock(sma
, -1);
1829 un
= ERR_PTR(-EIDRM
);
1832 spin_lock(&ulp
->lock
);
1835 * step 4: check for races: did someone else allocate the undo struct?
1837 un
= lookup_undo(ulp
, semid
);
1842 /* step 5: initialize & link new undo structure */
1843 new->semadj
= (short *) &new[1];
1846 assert_spin_locked(&ulp
->lock
);
1847 list_add_rcu(&new->list_proc
, &ulp
->list_proc
);
1848 ipc_assert_locked_object(&sma
->sem_perm
);
1849 list_add(&new->list_id
, &sma
->list_id
);
1853 spin_unlock(&ulp
->lock
);
1854 sem_unlock(sma
, -1);
1859 static long do_semtimedop(int semid
, struct sembuf __user
*tsops
,
1860 unsigned nsops
, const struct timespec64
*timeout
)
1862 int error
= -EINVAL
;
1863 struct sem_array
*sma
;
1864 struct sembuf fast_sops
[SEMOPM_FAST
];
1865 struct sembuf
*sops
= fast_sops
, *sop
;
1866 struct sem_undo
*un
;
1868 bool undos
= false, alter
= false, dupsop
= false;
1869 struct sem_queue queue
;
1870 unsigned long dup
= 0, jiffies_left
= 0;
1871 struct ipc_namespace
*ns
;
1873 ns
= current
->nsproxy
->ipc_ns
;
1875 if (nsops
< 1 || semid
< 0)
1877 if (nsops
> ns
->sc_semopm
)
1879 if (nsops
> SEMOPM_FAST
) {
1880 sops
= kvmalloc(sizeof(*sops
)*nsops
, GFP_KERNEL
);
1885 if (copy_from_user(sops
, tsops
, nsops
* sizeof(*tsops
))) {
1891 if (timeout
->tv_sec
< 0 || timeout
->tv_nsec
< 0 ||
1892 timeout
->tv_nsec
>= 1000000000L) {
1896 jiffies_left
= timespec64_to_jiffies(timeout
);
1900 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1901 unsigned long mask
= 1ULL << ((sop
->sem_num
) % BITS_PER_LONG
);
1903 if (sop
->sem_num
>= max
)
1905 if (sop
->sem_flg
& SEM_UNDO
)
1909 * There was a previous alter access that appears
1910 * to have accessed the same semaphore, thus use
1911 * the dupsop logic. "appears", because the detection
1912 * can only check % BITS_PER_LONG.
1916 if (sop
->sem_op
!= 0) {
1923 /* On success, find_alloc_undo takes the rcu_read_lock */
1924 un
= find_alloc_undo(ns
, semid
);
1926 error
= PTR_ERR(un
);
1934 sma
= sem_obtain_object_check(ns
, semid
);
1937 error
= PTR_ERR(sma
);
1942 if (max
>= sma
->sem_nsems
) {
1948 if (ipcperms(ns
, &sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
)) {
1953 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1960 locknum
= sem_lock(sma
, sops
, nsops
);
1962 * We eventually might perform the following check in a lockless
1963 * fashion, considering ipc_valid_object() locking constraints.
1964 * If nsops == 1 and there is no contention for sem_perm.lock, then
1965 * only a per-semaphore lock is held and it's OK to proceed with the
1966 * check below. More details on the fine grained locking scheme
1967 * entangled here and why it's RMID race safe on comments at sem_lock()
1969 if (!ipc_valid_object(&sma
->sem_perm
))
1970 goto out_unlock_free
;
1972 * semid identifiers are not unique - find_alloc_undo may have
1973 * allocated an undo structure, it was invalidated by an RMID
1974 * and now a new array with received the same id. Check and fail.
1975 * This case can be detected checking un->semid. The existence of
1976 * "un" itself is guaranteed by rcu.
1978 if (un
&& un
->semid
== -1)
1979 goto out_unlock_free
;
1982 queue
.nsops
= nsops
;
1984 queue
.pid
= task_tgid_vnr(current
);
1985 queue
.alter
= alter
;
1986 queue
.dupsop
= dupsop
;
1988 error
= perform_atomic_semop(sma
, &queue
);
1989 if (error
== 0) { /* non-blocking succesfull path */
1990 DEFINE_WAKE_Q(wake_q
);
1993 * If the operation was successful, then do
1994 * the required updates.
1997 do_smart_update(sma
, sops
, nsops
, 1, &wake_q
);
1999 set_semotime(sma
, sops
);
2001 sem_unlock(sma
, locknum
);
2007 if (error
< 0) /* non-blocking error path */
2008 goto out_unlock_free
;
2011 * We need to sleep on this operation, so we put the current
2012 * task into the pending queue and go to sleep.
2016 curr
= &sma
->sems
[sops
->sem_num
];
2019 if (sma
->complex_count
) {
2020 list_add_tail(&queue
.list
,
2021 &sma
->pending_alter
);
2024 list_add_tail(&queue
.list
,
2025 &curr
->pending_alter
);
2028 list_add_tail(&queue
.list
, &curr
->pending_const
);
2031 if (!sma
->complex_count
)
2035 list_add_tail(&queue
.list
, &sma
->pending_alter
);
2037 list_add_tail(&queue
.list
, &sma
->pending_const
);
2039 sma
->complex_count
++;
2043 queue
.status
= -EINTR
;
2044 queue
.sleeper
= current
;
2046 __set_current_state(TASK_INTERRUPTIBLE
);
2047 sem_unlock(sma
, locknum
);
2051 jiffies_left
= schedule_timeout(jiffies_left
);
2056 * fastpath: the semop has completed, either successfully or
2057 * not, from the syscall pov, is quite irrelevant to us at this
2058 * point; we're done.
2060 * We _do_ care, nonetheless, about being awoken by a signal or
2061 * spuriously. The queue.status is checked again in the
2062 * slowpath (aka after taking sem_lock), such that we can detect
2063 * scenarios where we were awakened externally, during the
2064 * window between wake_q_add() and wake_up_q().
2066 error
= READ_ONCE(queue
.status
);
2067 if (error
!= -EINTR
) {
2069 * User space could assume that semop() is a memory
2070 * barrier: Without the mb(), the cpu could
2071 * speculatively read in userspace stale data that was
2072 * overwritten by the previous owner of the semaphore.
2079 locknum
= sem_lock(sma
, sops
, nsops
);
2081 if (!ipc_valid_object(&sma
->sem_perm
))
2082 goto out_unlock_free
;
2084 error
= READ_ONCE(queue
.status
);
2087 * If queue.status != -EINTR we are woken up by another process.
2088 * Leave without unlink_queue(), but with sem_unlock().
2090 if (error
!= -EINTR
)
2091 goto out_unlock_free
;
2094 * If an interrupt occurred we have to clean up the queue.
2096 if (timeout
&& jiffies_left
== 0)
2098 } while (error
== -EINTR
&& !signal_pending(current
)); /* spurious */
2100 unlink_queue(sma
, &queue
);
2103 sem_unlock(sma
, locknum
);
2106 if (sops
!= fast_sops
)
2111 SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsops
,
2112 unsigned, nsops
, const struct timespec __user
*, timeout
)
2115 struct timespec64 ts
;
2116 if (get_timespec64(&ts
, timeout
))
2118 return do_semtimedop(semid
, tsops
, nsops
, &ts
);
2120 return do_semtimedop(semid
, tsops
, nsops
, NULL
);
2123 #ifdef CONFIG_COMPAT
2124 COMPAT_SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsems
,
2126 const struct compat_timespec __user
*, timeout
)
2129 struct timespec64 ts
;
2130 if (compat_get_timespec64(&ts
, timeout
))
2132 return do_semtimedop(semid
, tsems
, nsops
, &ts
);
2134 return do_semtimedop(semid
, tsems
, nsops
, NULL
);
2138 SYSCALL_DEFINE3(semop
, int, semid
, struct sembuf __user
*, tsops
,
2141 return do_semtimedop(semid
, tsops
, nsops
, NULL
);
2144 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2145 * parent and child tasks.
2148 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
2150 struct sem_undo_list
*undo_list
;
2153 if (clone_flags
& CLONE_SYSVSEM
) {
2154 error
= get_undo_list(&undo_list
);
2157 refcount_inc(&undo_list
->refcnt
);
2158 tsk
->sysvsem
.undo_list
= undo_list
;
2160 tsk
->sysvsem
.undo_list
= NULL
;
2166 * add semadj values to semaphores, free undo structures.
2167 * undo structures are not freed when semaphore arrays are destroyed
2168 * so some of them may be out of date.
2169 * IMPLEMENTATION NOTE: There is some confusion over whether the
2170 * set of adjustments that needs to be done should be done in an atomic
2171 * manner or not. That is, if we are attempting to decrement the semval
2172 * should we queue up and wait until we can do so legally?
2173 * The original implementation attempted to do this (queue and wait).
2174 * The current implementation does not do so. The POSIX standard
2175 * and SVID should be consulted to determine what behavior is mandated.
2177 void exit_sem(struct task_struct
*tsk
)
2179 struct sem_undo_list
*ulp
;
2181 ulp
= tsk
->sysvsem
.undo_list
;
2184 tsk
->sysvsem
.undo_list
= NULL
;
2186 if (!refcount_dec_and_test(&ulp
->refcnt
))
2190 struct sem_array
*sma
;
2191 struct sem_undo
*un
;
2193 DEFINE_WAKE_Q(wake_q
);
2198 un
= list_entry_rcu(ulp
->list_proc
.next
,
2199 struct sem_undo
, list_proc
);
2200 if (&un
->list_proc
== &ulp
->list_proc
) {
2202 * We must wait for freeary() before freeing this ulp,
2203 * in case we raced with last sem_undo. There is a small
2204 * possibility where we exit while freeary() didn't
2205 * finish unlocking sem_undo_list.
2207 spin_lock(&ulp
->lock
);
2208 spin_unlock(&ulp
->lock
);
2212 spin_lock(&ulp
->lock
);
2214 spin_unlock(&ulp
->lock
);
2216 /* exit_sem raced with IPC_RMID, nothing to do */
2222 sma
= sem_obtain_object_check(tsk
->nsproxy
->ipc_ns
, semid
);
2223 /* exit_sem raced with IPC_RMID, nothing to do */
2229 sem_lock(sma
, NULL
, -1);
2230 /* exit_sem raced with IPC_RMID, nothing to do */
2231 if (!ipc_valid_object(&sma
->sem_perm
)) {
2232 sem_unlock(sma
, -1);
2236 un
= __lookup_undo(ulp
, semid
);
2238 /* exit_sem raced with IPC_RMID+semget() that created
2239 * exactly the same semid. Nothing to do.
2241 sem_unlock(sma
, -1);
2246 /* remove un from the linked lists */
2247 ipc_assert_locked_object(&sma
->sem_perm
);
2248 list_del(&un
->list_id
);
2250 /* we are the last process using this ulp, acquiring ulp->lock
2251 * isn't required. Besides that, we are also protected against
2252 * IPC_RMID as we hold sma->sem_perm lock now
2254 list_del_rcu(&un
->list_proc
);
2256 /* perform adjustments registered in un */
2257 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
2258 struct sem
*semaphore
= &sma
->sems
[i
];
2259 if (un
->semadj
[i
]) {
2260 semaphore
->semval
+= un
->semadj
[i
];
2262 * Range checks of the new semaphore value,
2263 * not defined by sus:
2264 * - Some unices ignore the undo entirely
2265 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2266 * - some cap the value (e.g. FreeBSD caps
2267 * at 0, but doesn't enforce SEMVMX)
2269 * Linux caps the semaphore value, both at 0
2272 * Manfred <manfred@colorfullife.com>
2274 if (semaphore
->semval
< 0)
2275 semaphore
->semval
= 0;
2276 if (semaphore
->semval
> SEMVMX
)
2277 semaphore
->semval
= SEMVMX
;
2278 semaphore
->sempid
= task_tgid_vnr(current
);
2281 /* maybe some queued-up processes were waiting for this */
2282 do_smart_update(sma
, NULL
, 0, 1, &wake_q
);
2283 sem_unlock(sma
, -1);
2292 #ifdef CONFIG_PROC_FS
2293 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
2295 struct user_namespace
*user_ns
= seq_user_ns(s
);
2296 struct kern_ipc_perm
*ipcp
= it
;
2297 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
2301 * The proc interface isn't aware of sem_lock(), it calls
2302 * ipc_lock_object() directly (in sysvipc_find_ipc).
2303 * In order to stay compatible with sem_lock(), we must
2304 * enter / leave complex_mode.
2306 complexmode_enter(sma
);
2308 sem_otime
= get_semotime(sma
);
2311 "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n",
2316 from_kuid_munged(user_ns
, sma
->sem_perm
.uid
),
2317 from_kgid_munged(user_ns
, sma
->sem_perm
.gid
),
2318 from_kuid_munged(user_ns
, sma
->sem_perm
.cuid
),
2319 from_kgid_munged(user_ns
, sma
->sem_perm
.cgid
),
2323 complexmode_tryleave(sma
);