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>
86 #include <linux/uaccess.h>
89 /* One semaphore structure for each semaphore in the system. */
91 int semval
; /* current value */
93 * PID of the process that last modified the semaphore. For
94 * Linux, specifically these are:
96 * - semctl, via SETVAL and SETALL.
97 * - at task exit when performing undo adjustments (see exit_sem).
100 spinlock_t lock
; /* spinlock for fine-grained semtimedop */
101 struct list_head pending_alter
; /* pending single-sop operations */
102 /* that alter the semaphore */
103 struct list_head pending_const
; /* pending single-sop operations */
104 /* that do not alter the semaphore*/
105 time_t sem_otime
; /* candidate for sem_otime */
106 } ____cacheline_aligned_in_smp
;
108 /* One queue for each sleeping process in the system. */
110 struct list_head list
; /* queue of pending operations */
111 struct task_struct
*sleeper
; /* this process */
112 struct sem_undo
*undo
; /* undo structure */
113 int pid
; /* process id of requesting process */
114 int status
; /* completion status of operation */
115 struct sembuf
*sops
; /* array of pending operations */
116 struct sembuf
*blocking
; /* the operation that blocked */
117 int nsops
; /* number of operations */
118 bool alter
; /* does *sops alter the array? */
119 bool dupsop
; /* sops on more than one sem_num */
122 /* Each task has a list of undo requests. They are executed automatically
123 * when the process exits.
126 struct list_head list_proc
; /* per-process list: *
127 * all undos from one process
129 struct rcu_head rcu
; /* rcu struct for sem_undo */
130 struct sem_undo_list
*ulp
; /* back ptr to sem_undo_list */
131 struct list_head list_id
; /* per semaphore array list:
132 * all undos for one array */
133 int semid
; /* semaphore set identifier */
134 short *semadj
; /* array of adjustments */
135 /* one per semaphore */
138 /* sem_undo_list controls shared access to the list of sem_undo structures
139 * that may be shared among all a CLONE_SYSVSEM task group.
141 struct sem_undo_list
{
144 struct list_head list_proc
;
148 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
150 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
152 static int newary(struct ipc_namespace
*, struct ipc_params
*);
153 static void freeary(struct ipc_namespace
*, struct kern_ipc_perm
*);
154 #ifdef CONFIG_PROC_FS
155 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
);
158 #define SEMMSL_FAST 256 /* 512 bytes on stack */
159 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
162 * Switching from the mode suitable for simple ops
163 * to the mode for complex ops is costly. Therefore:
164 * use some hysteresis
166 #define USE_GLOBAL_LOCK_HYSTERESIS 10
170 * a) global sem_lock() for read/write
172 * sem_array.complex_count,
173 * sem_array.pending{_alter,_const},
176 * b) global or semaphore sem_lock() for read/write:
177 * sem_array.sem_base[i].pending_{const,alter}:
180 * sem_undo_list.list_proc:
181 * * undo_list->lock for write
184 * * global sem_lock() for write
185 * * either local or global sem_lock() for read.
188 * Most ordering is enforced by using spin_lock() and spin_unlock().
189 * The special case is use_global_lock:
190 * Setting it from non-zero to 0 is a RELEASE, this is ensured by
191 * using smp_store_release().
192 * Testing if it is non-zero is an ACQUIRE, this is ensured by using
193 * smp_load_acquire().
194 * Setting it from 0 to non-zero must be ordered with regards to
195 * this smp_load_acquire(), this is guaranteed because the smp_load_acquire()
196 * is inside a spin_lock() and after a write from 0 to non-zero a
197 * spin_lock()+spin_unlock() is done.
200 #define sc_semmsl sem_ctls[0]
201 #define sc_semmns sem_ctls[1]
202 #define sc_semopm sem_ctls[2]
203 #define sc_semmni sem_ctls[3]
205 void sem_init_ns(struct ipc_namespace
*ns
)
207 ns
->sc_semmsl
= SEMMSL
;
208 ns
->sc_semmns
= SEMMNS
;
209 ns
->sc_semopm
= SEMOPM
;
210 ns
->sc_semmni
= SEMMNI
;
212 ipc_init_ids(&ns
->ids
[IPC_SEM_IDS
]);
216 void sem_exit_ns(struct ipc_namespace
*ns
)
218 free_ipcs(ns
, &sem_ids(ns
), freeary
);
219 idr_destroy(&ns
->ids
[IPC_SEM_IDS
].ipcs_idr
);
223 void __init
sem_init(void)
225 sem_init_ns(&init_ipc_ns
);
226 ipc_init_proc_interface("sysvipc/sem",
227 " key semid perms nsems uid gid cuid cgid otime ctime\n",
228 IPC_SEM_IDS
, sysvipc_sem_proc_show
);
232 * unmerge_queues - unmerge queues, if possible.
233 * @sma: semaphore array
235 * The function unmerges the wait queues if complex_count is 0.
236 * It must be called prior to dropping the global semaphore array lock.
238 static void unmerge_queues(struct sem_array
*sma
)
240 struct sem_queue
*q
, *tq
;
242 /* complex operations still around? */
243 if (sma
->complex_count
)
246 * We will switch back to simple mode.
247 * Move all pending operation back into the per-semaphore
250 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
252 curr
= &sma
->sem_base
[q
->sops
[0].sem_num
];
254 list_add_tail(&q
->list
, &curr
->pending_alter
);
256 INIT_LIST_HEAD(&sma
->pending_alter
);
260 * merge_queues - merge single semop queues into global queue
261 * @sma: semaphore array
263 * This function merges all per-semaphore queues into the global queue.
264 * It is necessary to achieve FIFO ordering for the pending single-sop
265 * operations when a multi-semop operation must sleep.
266 * Only the alter operations must be moved, the const operations can stay.
268 static void merge_queues(struct sem_array
*sma
)
271 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
272 struct sem
*sem
= sma
->sem_base
+ i
;
274 list_splice_init(&sem
->pending_alter
, &sma
->pending_alter
);
278 static void sem_rcu_free(struct rcu_head
*head
)
280 struct ipc_rcu
*p
= container_of(head
, struct ipc_rcu
, rcu
);
281 struct sem_array
*sma
= ipc_rcu_to_struct(p
);
283 security_sem_free(sma
);
288 * Enter the mode suitable for non-simple operations:
289 * Caller must own sem_perm.lock.
291 static void complexmode_enter(struct sem_array
*sma
)
296 if (sma
->use_global_lock
> 0) {
298 * We are already in global lock mode.
299 * Nothing to do, just reset the
300 * counter until we return to simple mode.
302 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
305 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
307 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
308 sem
= sma
->sem_base
+ i
;
309 spin_lock(&sem
->lock
);
310 spin_unlock(&sem
->lock
);
315 * Try to leave the mode that disallows simple operations:
316 * Caller must own sem_perm.lock.
318 static void complexmode_tryleave(struct sem_array
*sma
)
320 if (sma
->complex_count
) {
321 /* Complex ops are sleeping.
322 * We must stay in complex mode
326 if (sma
->use_global_lock
== 1) {
328 * Immediately after setting use_global_lock to 0,
329 * a simple op can start. Thus: all memory writes
330 * performed by the current operation must be visible
331 * before we set use_global_lock to 0.
333 smp_store_release(&sma
->use_global_lock
, 0);
335 sma
->use_global_lock
--;
339 #define SEM_GLOBAL_LOCK (-1)
341 * If the request contains only one semaphore operation, and there are
342 * no complex transactions pending, lock only the semaphore involved.
343 * Otherwise, lock the entire semaphore array, since we either have
344 * multiple semaphores in our own semops, or we need to look at
345 * semaphores from other pending complex operations.
347 static inline int sem_lock(struct sem_array
*sma
, struct sembuf
*sops
,
353 /* Complex operation - acquire a full lock */
354 ipc_lock_object(&sma
->sem_perm
);
356 /* Prevent parallel simple ops */
357 complexmode_enter(sma
);
358 return SEM_GLOBAL_LOCK
;
362 * Only one semaphore affected - try to optimize locking.
363 * Optimized locking is possible if no complex operation
364 * is either enqueued or processed right now.
366 * Both facts are tracked by use_global_mode.
368 sem
= sma
->sem_base
+ sops
->sem_num
;
371 * Initial check for use_global_lock. Just an optimization,
372 * no locking, no memory barrier.
374 if (!sma
->use_global_lock
) {
376 * It appears that no complex operation is around.
377 * Acquire the per-semaphore lock.
379 spin_lock(&sem
->lock
);
381 /* pairs with smp_store_release() */
382 if (!smp_load_acquire(&sma
->use_global_lock
)) {
383 /* fast path successful! */
384 return sops
->sem_num
;
386 spin_unlock(&sem
->lock
);
389 /* slow path: acquire the full lock */
390 ipc_lock_object(&sma
->sem_perm
);
392 if (sma
->use_global_lock
== 0) {
394 * The use_global_lock mode ended while we waited for
395 * sma->sem_perm.lock. Thus we must switch to locking
397 * Unlike in the fast path, there is no need to recheck
398 * sma->use_global_lock after we have acquired sem->lock:
399 * We own sma->sem_perm.lock, thus use_global_lock cannot
402 spin_lock(&sem
->lock
);
404 ipc_unlock_object(&sma
->sem_perm
);
405 return sops
->sem_num
;
408 * Not a false alarm, thus continue to use the global lock
409 * mode. No need for complexmode_enter(), this was done by
410 * the caller that has set use_global_mode to non-zero.
412 return SEM_GLOBAL_LOCK
;
416 static inline void sem_unlock(struct sem_array
*sma
, int locknum
)
418 if (locknum
== SEM_GLOBAL_LOCK
) {
420 complexmode_tryleave(sma
);
421 ipc_unlock_object(&sma
->sem_perm
);
423 struct sem
*sem
= sma
->sem_base
+ locknum
;
424 spin_unlock(&sem
->lock
);
429 * sem_lock_(check_) routines are called in the paths where the rwsem
432 * The caller holds the RCU read lock.
434 static inline struct sem_array
*sem_obtain_object(struct ipc_namespace
*ns
, int id
)
436 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_idr(&sem_ids(ns
), id
);
439 return ERR_CAST(ipcp
);
441 return container_of(ipcp
, struct sem_array
, sem_perm
);
444 static inline struct sem_array
*sem_obtain_object_check(struct ipc_namespace
*ns
,
447 struct kern_ipc_perm
*ipcp
= ipc_obtain_object_check(&sem_ids(ns
), id
);
450 return ERR_CAST(ipcp
);
452 return container_of(ipcp
, struct sem_array
, sem_perm
);
455 static inline void sem_lock_and_putref(struct sem_array
*sma
)
457 sem_lock(sma
, NULL
, -1);
458 ipc_rcu_putref(sma
, sem_rcu_free
);
461 static inline void sem_rmid(struct ipc_namespace
*ns
, struct sem_array
*s
)
463 ipc_rmid(&sem_ids(ns
), &s
->sem_perm
);
467 * newary - Create a new semaphore set
469 * @params: ptr to the structure that contains key, semflg and nsems
471 * Called with sem_ids.rwsem held (as a writer)
473 static int newary(struct ipc_namespace
*ns
, struct ipc_params
*params
)
477 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 size
= sizeof(*sma
) + nsems
* sizeof(struct sem
);
490 sma
= ipc_rcu_alloc(size
);
494 memset(sma
, 0, size
);
496 sma
->sem_perm
.mode
= (semflg
& S_IRWXUGO
);
497 sma
->sem_perm
.key
= key
;
499 sma
->sem_perm
.security
= NULL
;
500 retval
= security_sem_alloc(sma
);
502 ipc_rcu_putref(sma
, ipc_rcu_free
);
506 sma
->sem_base
= (struct sem
*) &sma
[1];
508 for (i
= 0; i
< nsems
; i
++) {
509 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_alter
);
510 INIT_LIST_HEAD(&sma
->sem_base
[i
].pending_const
);
511 spin_lock_init(&sma
->sem_base
[i
].lock
);
514 sma
->complex_count
= 0;
515 sma
->use_global_lock
= USE_GLOBAL_LOCK_HYSTERESIS
;
516 INIT_LIST_HEAD(&sma
->pending_alter
);
517 INIT_LIST_HEAD(&sma
->pending_const
);
518 INIT_LIST_HEAD(&sma
->list_id
);
519 sma
->sem_nsems
= nsems
;
520 sma
->sem_ctime
= get_seconds();
522 id
= ipc_addid(&sem_ids(ns
), &sma
->sem_perm
, ns
->sc_semmni
);
524 ipc_rcu_putref(sma
, sem_rcu_free
);
527 ns
->used_sems
+= nsems
;
532 return sma
->sem_perm
.id
;
537 * Called with sem_ids.rwsem and ipcp locked.
539 static inline int sem_security(struct kern_ipc_perm
*ipcp
, int semflg
)
541 struct sem_array
*sma
;
543 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
544 return security_sem_associate(sma
, semflg
);
548 * Called with sem_ids.rwsem and ipcp locked.
550 static inline int sem_more_checks(struct kern_ipc_perm
*ipcp
,
551 struct ipc_params
*params
)
553 struct sem_array
*sma
;
555 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
556 if (params
->u
.nsems
> sma
->sem_nsems
)
562 SYSCALL_DEFINE3(semget
, key_t
, key
, int, nsems
, int, semflg
)
564 struct ipc_namespace
*ns
;
565 static const struct ipc_ops sem_ops
= {
567 .associate
= sem_security
,
568 .more_checks
= sem_more_checks
,
570 struct ipc_params sem_params
;
572 ns
= current
->nsproxy
->ipc_ns
;
574 if (nsems
< 0 || nsems
> ns
->sc_semmsl
)
577 sem_params
.key
= key
;
578 sem_params
.flg
= semflg
;
579 sem_params
.u
.nsems
= nsems
;
581 return ipcget(ns
, &sem_ids(ns
), &sem_ops
, &sem_params
);
585 * perform_atomic_semop[_slow] - Attempt to perform semaphore
586 * operations on a given array.
587 * @sma: semaphore array
588 * @q: struct sem_queue that describes the operation
590 * Caller blocking are as follows, based the value
591 * indicated by the semaphore operation (sem_op):
593 * (1) >0 never blocks.
594 * (2) 0 (wait-for-zero operation): semval is non-zero.
595 * (3) <0 attempting to decrement semval to a value smaller than zero.
597 * Returns 0 if the operation was possible.
598 * Returns 1 if the operation is impossible, the caller must sleep.
599 * Returns <0 for error codes.
601 static int perform_atomic_semop_slow(struct sem_array
*sma
, struct sem_queue
*q
)
603 int result
, sem_op
, nsops
, pid
;
613 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
614 curr
= sma
->sem_base
+ sop
->sem_num
;
615 sem_op
= sop
->sem_op
;
616 result
= curr
->semval
;
618 if (!sem_op
&& result
)
627 if (sop
->sem_flg
& SEM_UNDO
) {
628 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
629 /* Exceeding the undo range is an error. */
630 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
632 un
->semadj
[sop
->sem_num
] = undo
;
635 curr
->semval
= result
;
640 while (sop
>= sops
) {
641 sma
->sem_base
[sop
->sem_num
].sempid
= pid
;
654 if (sop
->sem_flg
& IPC_NOWAIT
)
661 while (sop
>= sops
) {
662 sem_op
= sop
->sem_op
;
663 sma
->sem_base
[sop
->sem_num
].semval
-= sem_op
;
664 if (sop
->sem_flg
& SEM_UNDO
)
665 un
->semadj
[sop
->sem_num
] += sem_op
;
672 static int perform_atomic_semop(struct sem_array
*sma
, struct sem_queue
*q
)
674 int result
, sem_op
, nsops
;
684 if (unlikely(q
->dupsop
))
685 return perform_atomic_semop_slow(sma
, q
);
688 * We scan the semaphore set twice, first to ensure that the entire
689 * operation can succeed, therefore avoiding any pointless writes
690 * to shared memory and having to undo such changes in order to block
691 * until the operations can go through.
693 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
694 curr
= sma
->sem_base
+ sop
->sem_num
;
695 sem_op
= sop
->sem_op
;
696 result
= curr
->semval
;
698 if (!sem_op
&& result
)
699 goto would_block
; /* wait-for-zero */
708 if (sop
->sem_flg
& SEM_UNDO
) {
709 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
711 /* Exceeding the undo range is an error. */
712 if (undo
< (-SEMAEM
- 1) || undo
> SEMAEM
)
717 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
718 curr
= sma
->sem_base
+ sop
->sem_num
;
719 sem_op
= sop
->sem_op
;
720 result
= curr
->semval
;
722 if (sop
->sem_flg
& SEM_UNDO
) {
723 int undo
= un
->semadj
[sop
->sem_num
] - sem_op
;
725 un
->semadj
[sop
->sem_num
] = undo
;
727 curr
->semval
+= sem_op
;
728 curr
->sempid
= q
->pid
;
735 return sop
->sem_flg
& IPC_NOWAIT
? -EAGAIN
: 1;
738 static inline void wake_up_sem_queue_prepare(struct sem_queue
*q
, int error
,
739 struct wake_q_head
*wake_q
)
741 wake_q_add(wake_q
, q
->sleeper
);
743 * Rely on the above implicit barrier, such that we can
744 * ensure that we hold reference to the task before setting
745 * q->status. Otherwise we could race with do_exit if the
746 * task is awoken by an external event before calling
749 WRITE_ONCE(q
->status
, error
);
752 static void unlink_queue(struct sem_array
*sma
, struct sem_queue
*q
)
756 sma
->complex_count
--;
759 /** check_restart(sma, q)
760 * @sma: semaphore array
761 * @q: the operation that just completed
763 * update_queue is O(N^2) when it restarts scanning the whole queue of
764 * waiting operations. Therefore this function checks if the restart is
765 * really necessary. It is called after a previously waiting operation
766 * modified the array.
767 * Note that wait-for-zero operations are handled without restart.
769 static inline int check_restart(struct sem_array
*sma
, struct sem_queue
*q
)
771 /* pending complex alter operations are too difficult to analyse */
772 if (!list_empty(&sma
->pending_alter
))
775 /* we were a sleeping complex operation. Too difficult */
779 /* It is impossible that someone waits for the new value:
780 * - complex operations always restart.
781 * - wait-for-zero are handled seperately.
782 * - q is a previously sleeping simple operation that
783 * altered the array. It must be a decrement, because
784 * simple increments never sleep.
785 * - If there are older (higher priority) decrements
786 * in the queue, then they have observed the original
787 * semval value and couldn't proceed. The operation
788 * decremented to value - thus they won't proceed either.
794 * wake_const_ops - wake up non-alter tasks
795 * @sma: semaphore array.
796 * @semnum: semaphore that was modified.
797 * @wake_q: lockless wake-queue head.
799 * wake_const_ops must be called after a semaphore in a semaphore array
800 * was set to 0. If complex const operations are pending, wake_const_ops must
801 * be called with semnum = -1, as well as with the number of each modified
803 * The tasks that must be woken up are added to @wake_q. The return code
804 * is stored in q->pid.
805 * The function returns 1 if at least one operation was completed successfully.
807 static int wake_const_ops(struct sem_array
*sma
, int semnum
,
808 struct wake_q_head
*wake_q
)
810 struct sem_queue
*q
, *tmp
;
811 struct list_head
*pending_list
;
812 int semop_completed
= 0;
815 pending_list
= &sma
->pending_const
;
817 pending_list
= &sma
->sem_base
[semnum
].pending_const
;
819 list_for_each_entry_safe(q
, tmp
, pending_list
, list
) {
820 int error
= perform_atomic_semop(sma
, q
);
824 /* operation completed, remove from queue & wakeup */
825 unlink_queue(sma
, q
);
827 wake_up_sem_queue_prepare(q
, error
, wake_q
);
832 return semop_completed
;
836 * do_smart_wakeup_zero - wakeup all wait for zero tasks
837 * @sma: semaphore array
838 * @sops: operations that were performed
839 * @nsops: number of operations
840 * @wake_q: lockless wake-queue head
842 * Checks all required queue for wait-for-zero operations, based
843 * on the actual changes that were performed on the semaphore array.
844 * The function returns 1 if at least one operation was completed successfully.
846 static int do_smart_wakeup_zero(struct sem_array
*sma
, struct sembuf
*sops
,
847 int nsops
, struct wake_q_head
*wake_q
)
850 int semop_completed
= 0;
853 /* first: the per-semaphore queues, if known */
855 for (i
= 0; i
< nsops
; i
++) {
856 int num
= sops
[i
].sem_num
;
858 if (sma
->sem_base
[num
].semval
== 0) {
860 semop_completed
|= wake_const_ops(sma
, num
, wake_q
);
865 * No sops means modified semaphores not known.
866 * Assume all were changed.
868 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
869 if (sma
->sem_base
[i
].semval
== 0) {
871 semop_completed
|= wake_const_ops(sma
, i
, wake_q
);
876 * If one of the modified semaphores got 0,
877 * then check the global queue, too.
880 semop_completed
|= wake_const_ops(sma
, -1, wake_q
);
882 return semop_completed
;
887 * update_queue - look for tasks that can be completed.
888 * @sma: semaphore array.
889 * @semnum: semaphore that was modified.
890 * @wake_q: lockless wake-queue head.
892 * update_queue must be called after a semaphore in a semaphore array
893 * was modified. If multiple semaphores were modified, update_queue must
894 * be called with semnum = -1, as well as with the number of each modified
896 * The tasks that must be woken up are added to @wake_q. The return code
897 * is stored in q->pid.
898 * The function internally checks if const operations can now succeed.
900 * The function return 1 if at least one semop was completed successfully.
902 static int update_queue(struct sem_array
*sma
, int semnum
, struct wake_q_head
*wake_q
)
904 struct sem_queue
*q
, *tmp
;
905 struct list_head
*pending_list
;
906 int semop_completed
= 0;
909 pending_list
= &sma
->pending_alter
;
911 pending_list
= &sma
->sem_base
[semnum
].pending_alter
;
914 list_for_each_entry_safe(q
, tmp
, pending_list
, list
) {
917 /* If we are scanning the single sop, per-semaphore list of
918 * one semaphore and that semaphore is 0, then it is not
919 * necessary to scan further: simple increments
920 * that affect only one entry succeed immediately and cannot
921 * be in the per semaphore pending queue, and decrements
922 * cannot be successful if the value is already 0.
924 if (semnum
!= -1 && sma
->sem_base
[semnum
].semval
== 0)
927 error
= perform_atomic_semop(sma
, q
);
929 /* Does q->sleeper still need to sleep? */
933 unlink_queue(sma
, q
);
939 do_smart_wakeup_zero(sma
, q
->sops
, q
->nsops
, wake_q
);
940 restart
= check_restart(sma
, q
);
943 wake_up_sem_queue_prepare(q
, error
, wake_q
);
947 return semop_completed
;
951 * set_semotime - set sem_otime
952 * @sma: semaphore array
953 * @sops: operations that modified the array, may be NULL
955 * sem_otime is replicated to avoid cache line trashing.
956 * This function sets one instance to the current time.
958 static void set_semotime(struct sem_array
*sma
, struct sembuf
*sops
)
961 sma
->sem_base
[0].sem_otime
= get_seconds();
963 sma
->sem_base
[sops
[0].sem_num
].sem_otime
=
969 * do_smart_update - optimized update_queue
970 * @sma: semaphore array
971 * @sops: operations that were performed
972 * @nsops: number of operations
973 * @otime: force setting otime
974 * @wake_q: lockless wake-queue head
976 * do_smart_update() does the required calls to update_queue and wakeup_zero,
977 * based on the actual changes that were performed on the semaphore array.
978 * Note that the function does not do the actual wake-up: the caller is
979 * responsible for calling wake_up_q().
980 * It is safe to perform this call after dropping all locks.
982 static void do_smart_update(struct sem_array
*sma
, struct sembuf
*sops
, int nsops
,
983 int otime
, struct wake_q_head
*wake_q
)
987 otime
|= do_smart_wakeup_zero(sma
, sops
, nsops
, wake_q
);
989 if (!list_empty(&sma
->pending_alter
)) {
990 /* semaphore array uses the global queue - just process it. */
991 otime
|= update_queue(sma
, -1, wake_q
);
995 * No sops, thus the modified semaphores are not
998 for (i
= 0; i
< sma
->sem_nsems
; i
++)
999 otime
|= update_queue(sma
, i
, wake_q
);
1002 * Check the semaphores that were increased:
1003 * - No complex ops, thus all sleeping ops are
1005 * - if we decreased the value, then any sleeping
1006 * semaphore ops wont be able to run: If the
1007 * previous value was too small, then the new
1008 * value will be too small, too.
1010 for (i
= 0; i
< nsops
; i
++) {
1011 if (sops
[i
].sem_op
> 0) {
1012 otime
|= update_queue(sma
,
1013 sops
[i
].sem_num
, wake_q
);
1019 set_semotime(sma
, sops
);
1023 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1025 static int check_qop(struct sem_array
*sma
, int semnum
, struct sem_queue
*q
,
1028 struct sembuf
*sop
= q
->blocking
;
1031 * Linux always (since 0.99.10) reported a task as sleeping on all
1032 * semaphores. This violates SUS, therefore it was changed to the
1033 * standard compliant behavior.
1034 * Give the administrators a chance to notice that an application
1035 * might misbehave because it relies on the Linux behavior.
1037 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1038 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1039 current
->comm
, task_pid_nr(current
));
1041 if (sop
->sem_num
!= semnum
)
1044 if (count_zero
&& sop
->sem_op
== 0)
1046 if (!count_zero
&& sop
->sem_op
< 0)
1052 /* The following counts are associated to each semaphore:
1053 * semncnt number of tasks waiting on semval being nonzero
1054 * semzcnt number of tasks waiting on semval being zero
1056 * Per definition, a task waits only on the semaphore of the first semop
1057 * that cannot proceed, even if additional operation would block, too.
1059 static int count_semcnt(struct sem_array
*sma
, ushort semnum
,
1062 struct list_head
*l
;
1063 struct sem_queue
*q
;
1067 /* First: check the simple operations. They are easy to evaluate */
1069 l
= &sma
->sem_base
[semnum
].pending_const
;
1071 l
= &sma
->sem_base
[semnum
].pending_alter
;
1073 list_for_each_entry(q
, l
, list
) {
1074 /* all task on a per-semaphore list sleep on exactly
1080 /* Then: check the complex operations. */
1081 list_for_each_entry(q
, &sma
->pending_alter
, list
) {
1082 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1085 list_for_each_entry(q
, &sma
->pending_const
, list
) {
1086 semcnt
+= check_qop(sma
, semnum
, q
, count_zero
);
1092 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1093 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1094 * remains locked on exit.
1096 static void freeary(struct ipc_namespace
*ns
, struct kern_ipc_perm
*ipcp
)
1098 struct sem_undo
*un
, *tu
;
1099 struct sem_queue
*q
, *tq
;
1100 struct sem_array
*sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1102 DEFINE_WAKE_Q(wake_q
);
1104 /* Free the existing undo structures for this semaphore set. */
1105 ipc_assert_locked_object(&sma
->sem_perm
);
1106 list_for_each_entry_safe(un
, tu
, &sma
->list_id
, list_id
) {
1107 list_del(&un
->list_id
);
1108 spin_lock(&un
->ulp
->lock
);
1110 list_del_rcu(&un
->list_proc
);
1111 spin_unlock(&un
->ulp
->lock
);
1115 /* Wake up all pending processes and let them fail with EIDRM. */
1116 list_for_each_entry_safe(q
, tq
, &sma
->pending_const
, list
) {
1117 unlink_queue(sma
, q
);
1118 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1121 list_for_each_entry_safe(q
, tq
, &sma
->pending_alter
, list
) {
1122 unlink_queue(sma
, q
);
1123 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1125 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
1126 struct sem
*sem
= sma
->sem_base
+ i
;
1127 list_for_each_entry_safe(q
, tq
, &sem
->pending_const
, list
) {
1128 unlink_queue(sma
, q
);
1129 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1131 list_for_each_entry_safe(q
, tq
, &sem
->pending_alter
, list
) {
1132 unlink_queue(sma
, q
);
1133 wake_up_sem_queue_prepare(q
, -EIDRM
, &wake_q
);
1137 /* Remove the semaphore set from the IDR */
1139 sem_unlock(sma
, -1);
1143 ns
->used_sems
-= sma
->sem_nsems
;
1144 ipc_rcu_putref(sma
, sem_rcu_free
);
1147 static unsigned long copy_semid_to_user(void __user
*buf
, struct semid64_ds
*in
, int version
)
1151 return copy_to_user(buf
, in
, sizeof(*in
));
1154 struct semid_ds out
;
1156 memset(&out
, 0, sizeof(out
));
1158 ipc64_perm_to_ipc_perm(&in
->sem_perm
, &out
.sem_perm
);
1160 out
.sem_otime
= in
->sem_otime
;
1161 out
.sem_ctime
= in
->sem_ctime
;
1162 out
.sem_nsems
= in
->sem_nsems
;
1164 return copy_to_user(buf
, &out
, sizeof(out
));
1171 static time_t get_semotime(struct sem_array
*sma
)
1176 res
= sma
->sem_base
[0].sem_otime
;
1177 for (i
= 1; i
< sma
->sem_nsems
; i
++) {
1178 time_t to
= sma
->sem_base
[i
].sem_otime
;
1186 static int semctl_nolock(struct ipc_namespace
*ns
, int semid
,
1187 int cmd
, int version
, void __user
*p
)
1190 struct sem_array
*sma
;
1196 struct seminfo seminfo
;
1199 err
= security_sem_semctl(NULL
, cmd
);
1203 memset(&seminfo
, 0, sizeof(seminfo
));
1204 seminfo
.semmni
= ns
->sc_semmni
;
1205 seminfo
.semmns
= ns
->sc_semmns
;
1206 seminfo
.semmsl
= ns
->sc_semmsl
;
1207 seminfo
.semopm
= ns
->sc_semopm
;
1208 seminfo
.semvmx
= SEMVMX
;
1209 seminfo
.semmnu
= SEMMNU
;
1210 seminfo
.semmap
= SEMMAP
;
1211 seminfo
.semume
= SEMUME
;
1212 down_read(&sem_ids(ns
).rwsem
);
1213 if (cmd
== SEM_INFO
) {
1214 seminfo
.semusz
= sem_ids(ns
).in_use
;
1215 seminfo
.semaem
= ns
->used_sems
;
1217 seminfo
.semusz
= SEMUSZ
;
1218 seminfo
.semaem
= SEMAEM
;
1220 max_id
= ipc_get_maxid(&sem_ids(ns
));
1221 up_read(&sem_ids(ns
).rwsem
);
1222 if (copy_to_user(p
, &seminfo
, sizeof(struct seminfo
)))
1224 return (max_id
< 0) ? 0 : max_id
;
1229 struct semid64_ds tbuf
;
1232 memset(&tbuf
, 0, sizeof(tbuf
));
1235 if (cmd
== SEM_STAT
) {
1236 sma
= sem_obtain_object(ns
, semid
);
1241 id
= sma
->sem_perm
.id
;
1243 sma
= sem_obtain_object_check(ns
, semid
);
1251 if (ipcperms(ns
, &sma
->sem_perm
, S_IRUGO
))
1254 err
= security_sem_semctl(sma
, cmd
);
1258 kernel_to_ipc64_perm(&sma
->sem_perm
, &tbuf
.sem_perm
);
1259 tbuf
.sem_otime
= get_semotime(sma
);
1260 tbuf
.sem_ctime
= sma
->sem_ctime
;
1261 tbuf
.sem_nsems
= sma
->sem_nsems
;
1263 if (copy_semid_to_user(p
, &tbuf
, version
))
1275 static int semctl_setval(struct ipc_namespace
*ns
, int semid
, int semnum
,
1278 struct sem_undo
*un
;
1279 struct sem_array
*sma
;
1282 DEFINE_WAKE_Q(wake_q
);
1284 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1285 /* big-endian 64bit */
1288 /* 32bit or little-endian 64bit */
1292 if (val
> SEMVMX
|| val
< 0)
1296 sma
= sem_obtain_object_check(ns
, semid
);
1299 return PTR_ERR(sma
);
1302 if (semnum
< 0 || semnum
>= sma
->sem_nsems
) {
1308 if (ipcperms(ns
, &sma
->sem_perm
, S_IWUGO
)) {
1313 err
= security_sem_semctl(sma
, SETVAL
);
1319 sem_lock(sma
, NULL
, -1);
1321 if (!ipc_valid_object(&sma
->sem_perm
)) {
1322 sem_unlock(sma
, -1);
1327 curr
= &sma
->sem_base
[semnum
];
1329 ipc_assert_locked_object(&sma
->sem_perm
);
1330 list_for_each_entry(un
, &sma
->list_id
, list_id
)
1331 un
->semadj
[semnum
] = 0;
1334 curr
->sempid
= task_tgid_vnr(current
);
1335 sma
->sem_ctime
= get_seconds();
1336 /* maybe some queued-up processes were waiting for this */
1337 do_smart_update(sma
, NULL
, 0, 0, &wake_q
);
1338 sem_unlock(sma
, -1);
1344 static int semctl_main(struct ipc_namespace
*ns
, int semid
, int semnum
,
1345 int cmd
, void __user
*p
)
1347 struct sem_array
*sma
;
1350 ushort fast_sem_io
[SEMMSL_FAST
];
1351 ushort
*sem_io
= fast_sem_io
;
1352 DEFINE_WAKE_Q(wake_q
);
1355 sma
= sem_obtain_object_check(ns
, semid
);
1358 return PTR_ERR(sma
);
1361 nsems
= sma
->sem_nsems
;
1364 if (ipcperms(ns
, &sma
->sem_perm
, cmd
== SETALL
? S_IWUGO
: S_IRUGO
))
1365 goto out_rcu_wakeup
;
1367 err
= security_sem_semctl(sma
, cmd
);
1369 goto out_rcu_wakeup
;
1375 ushort __user
*array
= p
;
1378 sem_lock(sma
, NULL
, -1);
1379 if (!ipc_valid_object(&sma
->sem_perm
)) {
1383 if (nsems
> SEMMSL_FAST
) {
1384 if (!ipc_rcu_getref(sma
)) {
1388 sem_unlock(sma
, -1);
1390 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1391 if (sem_io
== NULL
) {
1392 ipc_rcu_putref(sma
, sem_rcu_free
);
1397 sem_lock_and_putref(sma
);
1398 if (!ipc_valid_object(&sma
->sem_perm
)) {
1403 for (i
= 0; i
< sma
->sem_nsems
; i
++)
1404 sem_io
[i
] = sma
->sem_base
[i
].semval
;
1405 sem_unlock(sma
, -1);
1408 if (copy_to_user(array
, sem_io
, nsems
*sizeof(ushort
)))
1415 struct sem_undo
*un
;
1417 if (!ipc_rcu_getref(sma
)) {
1419 goto out_rcu_wakeup
;
1423 if (nsems
> SEMMSL_FAST
) {
1424 sem_io
= ipc_alloc(sizeof(ushort
)*nsems
);
1425 if (sem_io
== NULL
) {
1426 ipc_rcu_putref(sma
, sem_rcu_free
);
1431 if (copy_from_user(sem_io
, p
, nsems
*sizeof(ushort
))) {
1432 ipc_rcu_putref(sma
, sem_rcu_free
);
1437 for (i
= 0; i
< nsems
; i
++) {
1438 if (sem_io
[i
] > SEMVMX
) {
1439 ipc_rcu_putref(sma
, sem_rcu_free
);
1445 sem_lock_and_putref(sma
);
1446 if (!ipc_valid_object(&sma
->sem_perm
)) {
1451 for (i
= 0; i
< nsems
; i
++) {
1452 sma
->sem_base
[i
].semval
= sem_io
[i
];
1453 sma
->sem_base
[i
].sempid
= task_tgid_vnr(current
);
1456 ipc_assert_locked_object(&sma
->sem_perm
);
1457 list_for_each_entry(un
, &sma
->list_id
, list_id
) {
1458 for (i
= 0; i
< nsems
; i
++)
1461 sma
->sem_ctime
= get_seconds();
1462 /* maybe some queued-up processes were waiting for this */
1463 do_smart_update(sma
, NULL
, 0, 0, &wake_q
);
1467 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1470 if (semnum
< 0 || semnum
>= nsems
)
1471 goto out_rcu_wakeup
;
1473 sem_lock(sma
, NULL
, -1);
1474 if (!ipc_valid_object(&sma
->sem_perm
)) {
1478 curr
= &sma
->sem_base
[semnum
];
1488 err
= count_semcnt(sma
, semnum
, 0);
1491 err
= count_semcnt(sma
, semnum
, 1);
1496 sem_unlock(sma
, -1);
1501 if (sem_io
!= fast_sem_io
)
1506 static inline unsigned long
1507 copy_semid_from_user(struct semid64_ds
*out
, void __user
*buf
, int version
)
1511 if (copy_from_user(out
, buf
, sizeof(*out
)))
1516 struct semid_ds tbuf_old
;
1518 if (copy_from_user(&tbuf_old
, buf
, sizeof(tbuf_old
)))
1521 out
->sem_perm
.uid
= tbuf_old
.sem_perm
.uid
;
1522 out
->sem_perm
.gid
= tbuf_old
.sem_perm
.gid
;
1523 out
->sem_perm
.mode
= tbuf_old
.sem_perm
.mode
;
1533 * This function handles some semctl commands which require the rwsem
1534 * to be held in write mode.
1535 * NOTE: no locks must be held, the rwsem is taken inside this function.
1537 static int semctl_down(struct ipc_namespace
*ns
, int semid
,
1538 int cmd
, int version
, void __user
*p
)
1540 struct sem_array
*sma
;
1542 struct semid64_ds semid64
;
1543 struct kern_ipc_perm
*ipcp
;
1545 if (cmd
== IPC_SET
) {
1546 if (copy_semid_from_user(&semid64
, p
, version
))
1550 down_write(&sem_ids(ns
).rwsem
);
1553 ipcp
= ipcctl_pre_down_nolock(ns
, &sem_ids(ns
), semid
, cmd
,
1554 &semid64
.sem_perm
, 0);
1556 err
= PTR_ERR(ipcp
);
1560 sma
= container_of(ipcp
, struct sem_array
, sem_perm
);
1562 err
= security_sem_semctl(sma
, cmd
);
1568 sem_lock(sma
, NULL
, -1);
1569 /* freeary unlocks the ipc object and rcu */
1573 sem_lock(sma
, NULL
, -1);
1574 err
= ipc_update_perm(&semid64
.sem_perm
, ipcp
);
1577 sma
->sem_ctime
= get_seconds();
1585 sem_unlock(sma
, -1);
1589 up_write(&sem_ids(ns
).rwsem
);
1593 SYSCALL_DEFINE4(semctl
, int, semid
, int, semnum
, int, cmd
, unsigned long, arg
)
1596 struct ipc_namespace
*ns
;
1597 void __user
*p
= (void __user
*)arg
;
1602 version
= ipc_parse_version(&cmd
);
1603 ns
= current
->nsproxy
->ipc_ns
;
1610 return semctl_nolock(ns
, semid
, cmd
, version
, p
);
1617 return semctl_main(ns
, semid
, semnum
, cmd
, p
);
1619 return semctl_setval(ns
, semid
, semnum
, arg
);
1622 return semctl_down(ns
, semid
, cmd
, version
, p
);
1628 /* If the task doesn't already have a undo_list, then allocate one
1629 * here. We guarantee there is only one thread using this undo list,
1630 * and current is THE ONE
1632 * If this allocation and assignment succeeds, but later
1633 * portions of this code fail, there is no need to free the sem_undo_list.
1634 * Just let it stay associated with the task, and it'll be freed later
1637 * This can block, so callers must hold no locks.
1639 static inline int get_undo_list(struct sem_undo_list
**undo_listp
)
1641 struct sem_undo_list
*undo_list
;
1643 undo_list
= current
->sysvsem
.undo_list
;
1645 undo_list
= kzalloc(sizeof(*undo_list
), GFP_KERNEL
);
1646 if (undo_list
== NULL
)
1648 spin_lock_init(&undo_list
->lock
);
1649 atomic_set(&undo_list
->refcnt
, 1);
1650 INIT_LIST_HEAD(&undo_list
->list_proc
);
1652 current
->sysvsem
.undo_list
= undo_list
;
1654 *undo_listp
= undo_list
;
1658 static struct sem_undo
*__lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1660 struct sem_undo
*un
;
1662 list_for_each_entry_rcu(un
, &ulp
->list_proc
, list_proc
) {
1663 if (un
->semid
== semid
)
1669 static struct sem_undo
*lookup_undo(struct sem_undo_list
*ulp
, int semid
)
1671 struct sem_undo
*un
;
1673 assert_spin_locked(&ulp
->lock
);
1675 un
= __lookup_undo(ulp
, semid
);
1677 list_del_rcu(&un
->list_proc
);
1678 list_add_rcu(&un
->list_proc
, &ulp
->list_proc
);
1684 * find_alloc_undo - lookup (and if not present create) undo array
1686 * @semid: semaphore array id
1688 * The function looks up (and if not present creates) the undo structure.
1689 * The size of the undo structure depends on the size of the semaphore
1690 * array, thus the alloc path is not that straightforward.
1691 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1692 * performs a rcu_read_lock().
1694 static struct sem_undo
*find_alloc_undo(struct ipc_namespace
*ns
, int semid
)
1696 struct sem_array
*sma
;
1697 struct sem_undo_list
*ulp
;
1698 struct sem_undo
*un
, *new;
1701 error
= get_undo_list(&ulp
);
1703 return ERR_PTR(error
);
1706 spin_lock(&ulp
->lock
);
1707 un
= lookup_undo(ulp
, semid
);
1708 spin_unlock(&ulp
->lock
);
1709 if (likely(un
!= NULL
))
1712 /* no undo structure around - allocate one. */
1713 /* step 1: figure out the size of the semaphore array */
1714 sma
= sem_obtain_object_check(ns
, semid
);
1717 return ERR_CAST(sma
);
1720 nsems
= sma
->sem_nsems
;
1721 if (!ipc_rcu_getref(sma
)) {
1723 un
= ERR_PTR(-EIDRM
);
1728 /* step 2: allocate new undo structure */
1729 new = kzalloc(sizeof(struct sem_undo
) + sizeof(short)*nsems
, GFP_KERNEL
);
1731 ipc_rcu_putref(sma
, sem_rcu_free
);
1732 return ERR_PTR(-ENOMEM
);
1735 /* step 3: Acquire the lock on semaphore array */
1737 sem_lock_and_putref(sma
);
1738 if (!ipc_valid_object(&sma
->sem_perm
)) {
1739 sem_unlock(sma
, -1);
1742 un
= ERR_PTR(-EIDRM
);
1745 spin_lock(&ulp
->lock
);
1748 * step 4: check for races: did someone else allocate the undo struct?
1750 un
= lookup_undo(ulp
, semid
);
1755 /* step 5: initialize & link new undo structure */
1756 new->semadj
= (short *) &new[1];
1759 assert_spin_locked(&ulp
->lock
);
1760 list_add_rcu(&new->list_proc
, &ulp
->list_proc
);
1761 ipc_assert_locked_object(&sma
->sem_perm
);
1762 list_add(&new->list_id
, &sma
->list_id
);
1766 spin_unlock(&ulp
->lock
);
1767 sem_unlock(sma
, -1);
1772 SYSCALL_DEFINE4(semtimedop
, int, semid
, struct sembuf __user
*, tsops
,
1773 unsigned, nsops
, const struct timespec __user
*, timeout
)
1775 int error
= -EINVAL
;
1776 struct sem_array
*sma
;
1777 struct sembuf fast_sops
[SEMOPM_FAST
];
1778 struct sembuf
*sops
= fast_sops
, *sop
;
1779 struct sem_undo
*un
;
1781 bool undos
= false, alter
= false, dupsop
= false;
1782 struct sem_queue queue
;
1783 unsigned long dup
= 0, jiffies_left
= 0;
1784 struct ipc_namespace
*ns
;
1786 ns
= current
->nsproxy
->ipc_ns
;
1788 if (nsops
< 1 || semid
< 0)
1790 if (nsops
> ns
->sc_semopm
)
1792 if (nsops
> SEMOPM_FAST
) {
1793 sops
= kmalloc(sizeof(*sops
)*nsops
, GFP_KERNEL
);
1798 if (copy_from_user(sops
, tsops
, nsops
* sizeof(*tsops
))) {
1804 struct timespec _timeout
;
1805 if (copy_from_user(&_timeout
, timeout
, sizeof(*timeout
))) {
1809 if (_timeout
.tv_sec
< 0 || _timeout
.tv_nsec
< 0 ||
1810 _timeout
.tv_nsec
>= 1000000000L) {
1814 jiffies_left
= timespec_to_jiffies(&_timeout
);
1818 for (sop
= sops
; sop
< sops
+ nsops
; sop
++) {
1819 unsigned long mask
= 1ULL << ((sop
->sem_num
) % BITS_PER_LONG
);
1821 if (sop
->sem_num
>= max
)
1823 if (sop
->sem_flg
& SEM_UNDO
)
1827 * There was a previous alter access that appears
1828 * to have accessed the same semaphore, thus use
1829 * the dupsop logic. "appears", because the detection
1830 * can only check % BITS_PER_LONG.
1834 if (sop
->sem_op
!= 0) {
1841 /* On success, find_alloc_undo takes the rcu_read_lock */
1842 un
= find_alloc_undo(ns
, semid
);
1844 error
= PTR_ERR(un
);
1852 sma
= sem_obtain_object_check(ns
, semid
);
1855 error
= PTR_ERR(sma
);
1860 if (max
>= sma
->sem_nsems
) {
1866 if (ipcperms(ns
, &sma
->sem_perm
, alter
? S_IWUGO
: S_IRUGO
)) {
1871 error
= security_sem_semop(sma
, sops
, nsops
, alter
);
1878 locknum
= sem_lock(sma
, sops
, nsops
);
1880 * We eventually might perform the following check in a lockless
1881 * fashion, considering ipc_valid_object() locking constraints.
1882 * If nsops == 1 and there is no contention for sem_perm.lock, then
1883 * only a per-semaphore lock is held and it's OK to proceed with the
1884 * check below. More details on the fine grained locking scheme
1885 * entangled here and why it's RMID race safe on comments at sem_lock()
1887 if (!ipc_valid_object(&sma
->sem_perm
))
1888 goto out_unlock_free
;
1890 * semid identifiers are not unique - find_alloc_undo may have
1891 * allocated an undo structure, it was invalidated by an RMID
1892 * and now a new array with received the same id. Check and fail.
1893 * This case can be detected checking un->semid. The existence of
1894 * "un" itself is guaranteed by rcu.
1896 if (un
&& un
->semid
== -1)
1897 goto out_unlock_free
;
1900 queue
.nsops
= nsops
;
1902 queue
.pid
= task_tgid_vnr(current
);
1903 queue
.alter
= alter
;
1904 queue
.dupsop
= dupsop
;
1906 error
= perform_atomic_semop(sma
, &queue
);
1907 if (error
== 0) { /* non-blocking succesfull path */
1908 DEFINE_WAKE_Q(wake_q
);
1911 * If the operation was successful, then do
1912 * the required updates.
1915 do_smart_update(sma
, sops
, nsops
, 1, &wake_q
);
1917 set_semotime(sma
, sops
);
1919 sem_unlock(sma
, locknum
);
1925 if (error
< 0) /* non-blocking error path */
1926 goto out_unlock_free
;
1929 * We need to sleep on this operation, so we put the current
1930 * task into the pending queue and go to sleep.
1934 curr
= &sma
->sem_base
[sops
->sem_num
];
1937 if (sma
->complex_count
) {
1938 list_add_tail(&queue
.list
,
1939 &sma
->pending_alter
);
1942 list_add_tail(&queue
.list
,
1943 &curr
->pending_alter
);
1946 list_add_tail(&queue
.list
, &curr
->pending_const
);
1949 if (!sma
->complex_count
)
1953 list_add_tail(&queue
.list
, &sma
->pending_alter
);
1955 list_add_tail(&queue
.list
, &sma
->pending_const
);
1957 sma
->complex_count
++;
1961 queue
.status
= -EINTR
;
1962 queue
.sleeper
= current
;
1964 __set_current_state(TASK_INTERRUPTIBLE
);
1965 sem_unlock(sma
, locknum
);
1969 jiffies_left
= schedule_timeout(jiffies_left
);
1974 * fastpath: the semop has completed, either successfully or
1975 * not, from the syscall pov, is quite irrelevant to us at this
1976 * point; we're done.
1978 * We _do_ care, nonetheless, about being awoken by a signal or
1979 * spuriously. The queue.status is checked again in the
1980 * slowpath (aka after taking sem_lock), such that we can detect
1981 * scenarios where we were awakened externally, during the
1982 * window between wake_q_add() and wake_up_q().
1984 error
= READ_ONCE(queue
.status
);
1985 if (error
!= -EINTR
) {
1987 * User space could assume that semop() is a memory
1988 * barrier: Without the mb(), the cpu could
1989 * speculatively read in userspace stale data that was
1990 * overwritten by the previous owner of the semaphore.
1997 locknum
= sem_lock(sma
, sops
, nsops
);
1999 if (!ipc_valid_object(&sma
->sem_perm
))
2000 goto out_unlock_free
;
2002 error
= READ_ONCE(queue
.status
);
2005 * If queue.status != -EINTR we are woken up by another process.
2006 * Leave without unlink_queue(), but with sem_unlock().
2008 if (error
!= -EINTR
)
2009 goto out_unlock_free
;
2012 * If an interrupt occurred we have to clean up the queue.
2014 if (timeout
&& jiffies_left
== 0)
2016 } while (error
== -EINTR
&& !signal_pending(current
)); /* spurious */
2018 unlink_queue(sma
, &queue
);
2021 sem_unlock(sma
, locknum
);
2024 if (sops
!= fast_sops
)
2029 SYSCALL_DEFINE3(semop
, int, semid
, struct sembuf __user
*, tsops
,
2032 return sys_semtimedop(semid
, tsops
, nsops
, NULL
);
2035 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2036 * parent and child tasks.
2039 int copy_semundo(unsigned long clone_flags
, struct task_struct
*tsk
)
2041 struct sem_undo_list
*undo_list
;
2044 if (clone_flags
& CLONE_SYSVSEM
) {
2045 error
= get_undo_list(&undo_list
);
2048 atomic_inc(&undo_list
->refcnt
);
2049 tsk
->sysvsem
.undo_list
= undo_list
;
2051 tsk
->sysvsem
.undo_list
= NULL
;
2057 * add semadj values to semaphores, free undo structures.
2058 * undo structures are not freed when semaphore arrays are destroyed
2059 * so some of them may be out of date.
2060 * IMPLEMENTATION NOTE: There is some confusion over whether the
2061 * set of adjustments that needs to be done should be done in an atomic
2062 * manner or not. That is, if we are attempting to decrement the semval
2063 * should we queue up and wait until we can do so legally?
2064 * The original implementation attempted to do this (queue and wait).
2065 * The current implementation does not do so. The POSIX standard
2066 * and SVID should be consulted to determine what behavior is mandated.
2068 void exit_sem(struct task_struct
*tsk
)
2070 struct sem_undo_list
*ulp
;
2072 ulp
= tsk
->sysvsem
.undo_list
;
2075 tsk
->sysvsem
.undo_list
= NULL
;
2077 if (!atomic_dec_and_test(&ulp
->refcnt
))
2081 struct sem_array
*sma
;
2082 struct sem_undo
*un
;
2084 DEFINE_WAKE_Q(wake_q
);
2089 un
= list_entry_rcu(ulp
->list_proc
.next
,
2090 struct sem_undo
, list_proc
);
2091 if (&un
->list_proc
== &ulp
->list_proc
) {
2093 * We must wait for freeary() before freeing this ulp,
2094 * in case we raced with last sem_undo. There is a small
2095 * possibility where we exit while freeary() didn't
2096 * finish unlocking sem_undo_list.
2098 spin_unlock_wait(&ulp
->lock
);
2102 spin_lock(&ulp
->lock
);
2104 spin_unlock(&ulp
->lock
);
2106 /* exit_sem raced with IPC_RMID, nothing to do */
2112 sma
= sem_obtain_object_check(tsk
->nsproxy
->ipc_ns
, semid
);
2113 /* exit_sem raced with IPC_RMID, nothing to do */
2119 sem_lock(sma
, NULL
, -1);
2120 /* exit_sem raced with IPC_RMID, nothing to do */
2121 if (!ipc_valid_object(&sma
->sem_perm
)) {
2122 sem_unlock(sma
, -1);
2126 un
= __lookup_undo(ulp
, semid
);
2128 /* exit_sem raced with IPC_RMID+semget() that created
2129 * exactly the same semid. Nothing to do.
2131 sem_unlock(sma
, -1);
2136 /* remove un from the linked lists */
2137 ipc_assert_locked_object(&sma
->sem_perm
);
2138 list_del(&un
->list_id
);
2140 /* we are the last process using this ulp, acquiring ulp->lock
2141 * isn't required. Besides that, we are also protected against
2142 * IPC_RMID as we hold sma->sem_perm lock now
2144 list_del_rcu(&un
->list_proc
);
2146 /* perform adjustments registered in un */
2147 for (i
= 0; i
< sma
->sem_nsems
; i
++) {
2148 struct sem
*semaphore
= &sma
->sem_base
[i
];
2149 if (un
->semadj
[i
]) {
2150 semaphore
->semval
+= un
->semadj
[i
];
2152 * Range checks of the new semaphore value,
2153 * not defined by sus:
2154 * - Some unices ignore the undo entirely
2155 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2156 * - some cap the value (e.g. FreeBSD caps
2157 * at 0, but doesn't enforce SEMVMX)
2159 * Linux caps the semaphore value, both at 0
2162 * Manfred <manfred@colorfullife.com>
2164 if (semaphore
->semval
< 0)
2165 semaphore
->semval
= 0;
2166 if (semaphore
->semval
> SEMVMX
)
2167 semaphore
->semval
= SEMVMX
;
2168 semaphore
->sempid
= task_tgid_vnr(current
);
2171 /* maybe some queued-up processes were waiting for this */
2172 do_smart_update(sma
, NULL
, 0, 1, &wake_q
);
2173 sem_unlock(sma
, -1);
2182 #ifdef CONFIG_PROC_FS
2183 static int sysvipc_sem_proc_show(struct seq_file
*s
, void *it
)
2185 struct user_namespace
*user_ns
= seq_user_ns(s
);
2186 struct sem_array
*sma
= it
;
2190 * The proc interface isn't aware of sem_lock(), it calls
2191 * ipc_lock_object() directly (in sysvipc_find_ipc).
2192 * In order to stay compatible with sem_lock(), we must
2193 * enter / leave complex_mode.
2195 complexmode_enter(sma
);
2197 sem_otime
= get_semotime(sma
);
2200 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2205 from_kuid_munged(user_ns
, sma
->sem_perm
.uid
),
2206 from_kgid_munged(user_ns
, sma
->sem_perm
.gid
),
2207 from_kuid_munged(user_ns
, sma
->sem_perm
.cuid
),
2208 from_kgid_munged(user_ns
, sma
->sem_perm
.cgid
),
2212 complexmode_tryleave(sma
);