3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semcnt()
51 * - the task that performs a successful semop() scans the list of all
52 * sleeping tasks and completes any pending operations that can be fulfilled.
53 * Semaphores are actively given to waiting tasks (necessary for FIFO).
54 * (see update_queue())
55 * - To improve the scalability, the actual wake-up calls are performed after
56 * dropping all locks. (see wake_up_sem_queue_prepare(),
57 * wake_up_sem_queue_do())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - The synchronizations between wake-ups due to a timeout/signal and a
63 * wake-up due to a completed semaphore operation is achieved by using an
64 * intermediate state (IN_WAKEUP).
65 * - UNDO values are stored in an array (one per process and per
66 * semaphore array, lazily allocated). For backwards compatibility, multiple
67 * modes for the UNDO variables are supported (per process, per thread)
68 * (see copy_semundo, CLONE_SYSVSEM)
69 * - There are two lists of the pending operations: a per-array list
70 * and per-semaphore list (stored in the array). This allows to achieve FIFO
71 * ordering without always scanning all pending operations.
72 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
75 #include <linux/slab.h>
76 #include <linux/spinlock.h>
77 #include <linux/init.h>
78 #include <linux/proc_fs.h>
79 #include <linux/time.h>
80 #include <linux/security.h>
81 #include <linux/syscalls.h>
82 #include <linux/audit.h>
83 #include <linux/capability.h>
84 #include <linux/seq_file.h>
85 #include <linux/rwsem.h>
86 #include <linux/nsproxy.h>
87 #include <linux/ipc_namespace.h>
89 #include <linux/uaccess.h>
92 /* One semaphore structure for each semaphore in the system. */
94 int semval; /* current value */
95 int sempid; /* pid of last operation */
96 spinlock_t lock; /* spinlock for fine-grained semtimedop */
97 struct list_head pending_alter; /* pending single-sop operations */
98 /* that alter the semaphore */
99 struct list_head pending_const; /* pending single-sop operations */
100 /* that do not alter the semaphore*/
101 time_t sem_otime; /* candidate for sem_otime */
102 } ____cacheline_aligned_in_smp;
104 /* One queue for each sleeping process in the system. */
106 struct list_head list; /* queue of pending operations */
107 struct task_struct *sleeper; /* this process */
108 struct sem_undo *undo; /* undo structure */
109 int pid; /* process id of requesting process */
110 int status; /* completion status of operation */
111 struct sembuf *sops; /* array of pending operations */
112 struct sembuf *blocking; /* the operation that blocked */
113 int nsops; /* number of operations */
114 int alter; /* does *sops alter the array? */
117 /* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
121 struct list_head list_proc; /* per-process list: *
122 * all undos from one process
124 struct rcu_head rcu; /* rcu struct for sem_undo */
125 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
126 struct list_head list_id; /* per semaphore array list:
127 * all undos for one array */
128 int semid; /* semaphore set identifier */
129 short *semadj; /* array of adjustments */
130 /* one per semaphore */
133 /* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
136 struct sem_undo_list {
139 struct list_head list_proc;
143 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
145 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
147 static int newary(struct ipc_namespace *, struct ipc_params *);
148 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
149 #ifdef CONFIG_PROC_FS
150 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
153 #define SEMMSL_FAST 256 /* 512 bytes on stack */
154 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
159 * sem_array.complex_count,
160 * sem_array.pending{_alter,_cont},
161 * sem_array.sem_undo: global sem_lock() for read/write
162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
168 #define sc_semmsl sem_ctls[0]
169 #define sc_semmns sem_ctls[1]
170 #define sc_semopm sem_ctls[2]
171 #define sc_semmni sem_ctls[3]
173 void sem_init_ns(struct ipc_namespace *ns)
175 ns->sc_semmsl = SEMMSL;
176 ns->sc_semmns = SEMMNS;
177 ns->sc_semopm = SEMOPM;
178 ns->sc_semmni = SEMMNI;
180 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
184 void sem_exit_ns(struct ipc_namespace *ns)
186 free_ipcs(ns, &sem_ids(ns), freeary);
187 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
191 void __init sem_init(void)
193 sem_init_ns(&init_ipc_ns);
194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
196 IPC_SEM_IDS, sysvipc_sem_proc_show);
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
206 static void unmerge_queues(struct sem_array *sma)
208 struct sem_queue *q, *tq;
210 /* complex operations still around? */
211 if (sma->complex_count)
214 * We will switch back to simple mode.
215 * Move all pending operation back into the per-semaphore
218 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
220 curr = &sma->sem_base[q->sops[0].sem_num];
222 list_add_tail(&q->list, &curr->pending_alter);
224 INIT_LIST_HEAD(&sma->pending_alter);
228 * merge_queues - merge single semop queues into global queue
229 * @sma: semaphore array
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
236 static void merge_queues(struct sem_array *sma)
239 for (i = 0; i < sma->sem_nsems; i++) {
240 struct sem *sem = sma->sem_base + i;
242 list_splice_init(&sem->pending_alter, &sma->pending_alter);
246 static void sem_rcu_free(struct rcu_head *head)
248 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249 struct sem_array *sma = ipc_rcu_to_struct(p);
251 security_sem_free(sma);
256 * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they
257 * are only control barriers.
258 * The code must pair with spin_unlock(&sem->lock) or
259 * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient.
261 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
263 #define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb()
266 * Wait until all currently ongoing simple ops have completed.
267 * Caller must own sem_perm.lock.
268 * New simple ops cannot start, because simple ops first check
269 * that sem_perm.lock is free.
270 * that a) sem_perm.lock is free and b) complex_count is 0.
272 static void sem_wait_array(struct sem_array *sma)
277 if (sma->complex_count) {
278 /* The thread that increased sma->complex_count waited on
279 * all sem->lock locks. Thus we don't need to wait again.
284 for (i = 0; i < sma->sem_nsems; i++) {
285 sem = sma->sem_base + i;
286 spin_unlock_wait(&sem->lock);
288 ipc_smp_acquire__after_spin_is_unlocked();
292 * If the request contains only one semaphore operation, and there are
293 * no complex transactions pending, lock only the semaphore involved.
294 * Otherwise, lock the entire semaphore array, since we either have
295 * multiple semaphores in our own semops, or we need to look at
296 * semaphores from other pending complex operations.
298 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
304 /* Complex operation - acquire a full lock */
305 ipc_lock_object(&sma->sem_perm);
307 /* And wait until all simple ops that are processed
308 * right now have dropped their locks.
315 * Only one semaphore affected - try to optimize locking.
317 * - optimized locking is possible if no complex operation
318 * is either enqueued or processed right now.
319 * - The test for enqueued complex ops is simple:
320 * sma->complex_count != 0
321 * - Testing for complex ops that are processed right now is
322 * a bit more difficult. Complex ops acquire the full lock
323 * and first wait that the running simple ops have completed.
325 * Thus: If we own a simple lock and the global lock is free
326 * and complex_count is now 0, then it will stay 0 and
327 * thus just locking sem->lock is sufficient.
329 sem = sma->sem_base + sops->sem_num;
331 if (sma->complex_count == 0) {
333 * It appears that no complex operation is around.
334 * Acquire the per-semaphore lock.
336 spin_lock(&sem->lock);
338 /* Then check that the global lock is free */
339 if (!spin_is_locked(&sma->sem_perm.lock)) {
341 * We need a memory barrier with acquire semantics,
342 * otherwise we can race with another thread that does:
344 * spin_unlock(sem_perm.lock);
346 ipc_smp_acquire__after_spin_is_unlocked();
349 * Now repeat the test of complex_count:
350 * It can't change anymore until we drop sem->lock.
351 * Thus: if is now 0, then it will stay 0.
353 if (sma->complex_count == 0) {
354 /* fast path successful! */
355 return sops->sem_num;
358 spin_unlock(&sem->lock);
361 /* slow path: acquire the full lock */
362 ipc_lock_object(&sma->sem_perm);
364 if (sma->complex_count == 0) {
366 * There is no complex operation, thus we can switch
367 * back to the fast path.
369 spin_lock(&sem->lock);
370 ipc_unlock_object(&sma->sem_perm);
371 return sops->sem_num;
373 /* Not a false alarm, thus complete the sequence for a
381 static inline void sem_unlock(struct sem_array *sma, int locknum)
385 ipc_unlock_object(&sma->sem_perm);
387 struct sem *sem = sma->sem_base + locknum;
388 spin_unlock(&sem->lock);
393 * sem_lock_(check_) routines are called in the paths where the rwsem
396 * The caller holds the RCU read lock.
398 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
399 int id, struct sembuf *sops, int nsops, int *locknum)
401 struct kern_ipc_perm *ipcp;
402 struct sem_array *sma;
404 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
406 return ERR_CAST(ipcp);
408 sma = container_of(ipcp, struct sem_array, sem_perm);
409 *locknum = sem_lock(sma, sops, nsops);
411 /* ipc_rmid() may have already freed the ID while sem_lock
412 * was spinning: verify that the structure is still valid
414 if (ipc_valid_object(ipcp))
415 return container_of(ipcp, struct sem_array, sem_perm);
417 sem_unlock(sma, *locknum);
418 return ERR_PTR(-EINVAL);
421 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
423 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
426 return ERR_CAST(ipcp);
428 return container_of(ipcp, struct sem_array, sem_perm);
431 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
434 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
437 return ERR_CAST(ipcp);
439 return container_of(ipcp, struct sem_array, sem_perm);
442 static inline void sem_lock_and_putref(struct sem_array *sma)
444 sem_lock(sma, NULL, -1);
445 ipc_rcu_putref(sma, ipc_rcu_free);
448 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
450 ipc_rmid(&sem_ids(ns), &s->sem_perm);
454 * Lockless wakeup algorithm:
455 * Without the check/retry algorithm a lockless wakeup is possible:
456 * - queue.status is initialized to -EINTR before blocking.
457 * - wakeup is performed by
458 * * unlinking the queue entry from the pending list
459 * * setting queue.status to IN_WAKEUP
460 * This is the notification for the blocked thread that a
461 * result value is imminent.
462 * * call wake_up_process
463 * * set queue.status to the final value.
464 * - the previously blocked thread checks queue.status:
465 * * if it's IN_WAKEUP, then it must wait until the value changes
466 * * if it's not -EINTR, then the operation was completed by
467 * update_queue. semtimedop can return queue.status without
468 * performing any operation on the sem array.
469 * * otherwise it must acquire the spinlock and check what's up.
471 * The two-stage algorithm is necessary to protect against the following
473 * - if queue.status is set after wake_up_process, then the woken up idle
474 * thread could race forward and try (and fail) to acquire sma->lock
475 * before update_queue had a chance to set queue.status
476 * - if queue.status is written before wake_up_process and if the
477 * blocked process is woken up by a signal between writing
478 * queue.status and the wake_up_process, then the woken up
479 * process could return from semtimedop and die by calling
480 * sys_exit before wake_up_process is called. Then wake_up_process
481 * will oops, because the task structure is already invalid.
482 * (yes, this happened on s390 with sysv msg).
488 * newary - Create a new semaphore set
490 * @params: ptr to the structure that contains key, semflg and nsems
492 * Called with sem_ids.rwsem held (as a writer)
494 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
498 struct sem_array *sma;
500 key_t key = params->key;
501 int nsems = params->u.nsems;
502 int semflg = params->flg;
507 if (ns->used_sems + nsems > ns->sc_semmns)
510 size = sizeof(*sma) + nsems * sizeof(struct sem);
511 sma = ipc_rcu_alloc(size);
515 memset(sma, 0, size);
517 sma->sem_perm.mode = (semflg & S_IRWXUGO);
518 sma->sem_perm.key = key;
520 sma->sem_perm.security = NULL;
521 retval = security_sem_alloc(sma);
523 ipc_rcu_putref(sma, ipc_rcu_free);
527 sma->sem_base = (struct sem *) &sma[1];
529 for (i = 0; i < nsems; i++) {
530 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
531 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
532 spin_lock_init(&sma->sem_base[i].lock);
535 sma->complex_count = 0;
536 INIT_LIST_HEAD(&sma->pending_alter);
537 INIT_LIST_HEAD(&sma->pending_const);
538 INIT_LIST_HEAD(&sma->list_id);
539 sma->sem_nsems = nsems;
540 sma->sem_ctime = get_seconds();
542 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
544 ipc_rcu_putref(sma, sem_rcu_free);
547 ns->used_sems += nsems;
552 return sma->sem_perm.id;
557 * Called with sem_ids.rwsem and ipcp locked.
559 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
561 struct sem_array *sma;
563 sma = container_of(ipcp, struct sem_array, sem_perm);
564 return security_sem_associate(sma, semflg);
568 * Called with sem_ids.rwsem and ipcp locked.
570 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
571 struct ipc_params *params)
573 struct sem_array *sma;
575 sma = container_of(ipcp, struct sem_array, sem_perm);
576 if (params->u.nsems > sma->sem_nsems)
582 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
584 struct ipc_namespace *ns;
585 static const struct ipc_ops sem_ops = {
587 .associate = sem_security,
588 .more_checks = sem_more_checks,
590 struct ipc_params sem_params;
592 ns = current->nsproxy->ipc_ns;
594 if (nsems < 0 || nsems > ns->sc_semmsl)
597 sem_params.key = key;
598 sem_params.flg = semflg;
599 sem_params.u.nsems = nsems;
601 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
605 * perform_atomic_semop - Perform (if possible) a semaphore operation
606 * @sma: semaphore array
607 * @q: struct sem_queue that describes the operation
609 * Returns 0 if the operation was possible.
610 * Returns 1 if the operation is impossible, the caller must sleep.
611 * Negative values are error codes.
613 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
615 int result, sem_op, nsops, pid;
625 for (sop = sops; sop < sops + nsops; sop++) {
626 curr = sma->sem_base + sop->sem_num;
627 sem_op = sop->sem_op;
628 result = curr->semval;
630 if (!sem_op && result)
639 if (sop->sem_flg & SEM_UNDO) {
640 int undo = un->semadj[sop->sem_num] - sem_op;
641 /* Exceeding the undo range is an error. */
642 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
644 un->semadj[sop->sem_num] = undo;
647 curr->semval = result;
652 while (sop >= sops) {
653 sma->sem_base[sop->sem_num].sempid = pid;
666 if (sop->sem_flg & IPC_NOWAIT)
673 while (sop >= sops) {
674 sem_op = sop->sem_op;
675 sma->sem_base[sop->sem_num].semval -= sem_op;
676 if (sop->sem_flg & SEM_UNDO)
677 un->semadj[sop->sem_num] += sem_op;
684 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
685 * @q: queue entry that must be signaled
686 * @error: Error value for the signal
688 * Prepare the wake-up of the queue entry q.
690 static void wake_up_sem_queue_prepare(struct list_head *pt,
691 struct sem_queue *q, int error)
693 if (list_empty(pt)) {
695 * Hold preempt off so that we don't get preempted and have the
696 * wakee busy-wait until we're scheduled back on.
700 q->status = IN_WAKEUP;
703 list_add_tail(&q->list, pt);
707 * wake_up_sem_queue_do - do the actual wake-up
708 * @pt: list of tasks to be woken up
710 * Do the actual wake-up.
711 * The function is called without any locks held, thus the semaphore array
712 * could be destroyed already and the tasks can disappear as soon as the
713 * status is set to the actual return code.
715 static void wake_up_sem_queue_do(struct list_head *pt)
717 struct sem_queue *q, *t;
720 did_something = !list_empty(pt);
721 list_for_each_entry_safe(q, t, pt, list) {
722 wake_up_process(q->sleeper);
723 /* q can disappear immediately after writing q->status. */
731 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
735 sma->complex_count--;
738 /** check_restart(sma, q)
739 * @sma: semaphore array
740 * @q: the operation that just completed
742 * update_queue is O(N^2) when it restarts scanning the whole queue of
743 * waiting operations. Therefore this function checks if the restart is
744 * really necessary. It is called after a previously waiting operation
745 * modified the array.
746 * Note that wait-for-zero operations are handled without restart.
748 static int check_restart(struct sem_array *sma, struct sem_queue *q)
750 /* pending complex alter operations are too difficult to analyse */
751 if (!list_empty(&sma->pending_alter))
754 /* we were a sleeping complex operation. Too difficult */
758 /* It is impossible that someone waits for the new value:
759 * - complex operations always restart.
760 * - wait-for-zero are handled seperately.
761 * - q is a previously sleeping simple operation that
762 * altered the array. It must be a decrement, because
763 * simple increments never sleep.
764 * - If there are older (higher priority) decrements
765 * in the queue, then they have observed the original
766 * semval value and couldn't proceed. The operation
767 * decremented to value - thus they won't proceed either.
773 * wake_const_ops - wake up non-alter tasks
774 * @sma: semaphore array.
775 * @semnum: semaphore that was modified.
776 * @pt: list head for the tasks that must be woken up.
778 * wake_const_ops must be called after a semaphore in a semaphore array
779 * was set to 0. If complex const operations are pending, wake_const_ops must
780 * be called with semnum = -1, as well as with the number of each modified
782 * The tasks that must be woken up are added to @pt. The return code
783 * is stored in q->pid.
784 * The function returns 1 if at least one operation was completed successfully.
786 static int wake_const_ops(struct sem_array *sma, int semnum,
787 struct list_head *pt)
790 struct list_head *walk;
791 struct list_head *pending_list;
792 int semop_completed = 0;
795 pending_list = &sma->pending_const;
797 pending_list = &sma->sem_base[semnum].pending_const;
799 walk = pending_list->next;
800 while (walk != pending_list) {
803 q = container_of(walk, struct sem_queue, list);
806 error = perform_atomic_semop(sma, q);
809 /* operation completed, remove from queue & wakeup */
811 unlink_queue(sma, q);
813 wake_up_sem_queue_prepare(pt, q, error);
818 return semop_completed;
822 * do_smart_wakeup_zero - wakeup all wait for zero tasks
823 * @sma: semaphore array
824 * @sops: operations that were performed
825 * @nsops: number of operations
826 * @pt: list head of the tasks that must be woken up.
828 * Checks all required queue for wait-for-zero operations, based
829 * on the actual changes that were performed on the semaphore array.
830 * The function returns 1 if at least one operation was completed successfully.
832 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
833 int nsops, struct list_head *pt)
836 int semop_completed = 0;
839 /* first: the per-semaphore queues, if known */
841 for (i = 0; i < nsops; i++) {
842 int num = sops[i].sem_num;
844 if (sma->sem_base[num].semval == 0) {
846 semop_completed |= wake_const_ops(sma, num, pt);
851 * No sops means modified semaphores not known.
852 * Assume all were changed.
854 for (i = 0; i < sma->sem_nsems; i++) {
855 if (sma->sem_base[i].semval == 0) {
857 semop_completed |= wake_const_ops(sma, i, pt);
862 * If one of the modified semaphores got 0,
863 * then check the global queue, too.
866 semop_completed |= wake_const_ops(sma, -1, pt);
868 return semop_completed;
873 * update_queue - look for tasks that can be completed.
874 * @sma: semaphore array.
875 * @semnum: semaphore that was modified.
876 * @pt: list head for the tasks that must be woken up.
878 * update_queue must be called after a semaphore in a semaphore array
879 * was modified. If multiple semaphores were modified, update_queue must
880 * be called with semnum = -1, as well as with the number of each modified
882 * The tasks that must be woken up are added to @pt. The return code
883 * is stored in q->pid.
884 * The function internally checks if const operations can now succeed.
886 * The function return 1 if at least one semop was completed successfully.
888 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
891 struct list_head *walk;
892 struct list_head *pending_list;
893 int semop_completed = 0;
896 pending_list = &sma->pending_alter;
898 pending_list = &sma->sem_base[semnum].pending_alter;
901 walk = pending_list->next;
902 while (walk != pending_list) {
905 q = container_of(walk, struct sem_queue, list);
908 /* If we are scanning the single sop, per-semaphore list of
909 * one semaphore and that semaphore is 0, then it is not
910 * necessary to scan further: simple increments
911 * that affect only one entry succeed immediately and cannot
912 * be in the per semaphore pending queue, and decrements
913 * cannot be successful if the value is already 0.
915 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
918 error = perform_atomic_semop(sma, q);
920 /* Does q->sleeper still need to sleep? */
924 unlink_queue(sma, q);
930 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
931 restart = check_restart(sma, q);
934 wake_up_sem_queue_prepare(pt, q, error);
938 return semop_completed;
942 * set_semotime - set sem_otime
943 * @sma: semaphore array
944 * @sops: operations that modified the array, may be NULL
946 * sem_otime is replicated to avoid cache line trashing.
947 * This function sets one instance to the current time.
949 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
952 sma->sem_base[0].sem_otime = get_seconds();
954 sma->sem_base[sops[0].sem_num].sem_otime =
960 * do_smart_update - optimized update_queue
961 * @sma: semaphore array
962 * @sops: operations that were performed
963 * @nsops: number of operations
964 * @otime: force setting otime
965 * @pt: list head of the tasks that must be woken up.
967 * do_smart_update() does the required calls to update_queue and wakeup_zero,
968 * based on the actual changes that were performed on the semaphore array.
969 * Note that the function does not do the actual wake-up: the caller is
970 * responsible for calling wake_up_sem_queue_do(@pt).
971 * It is safe to perform this call after dropping all locks.
973 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
974 int otime, struct list_head *pt)
978 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
980 if (!list_empty(&sma->pending_alter)) {
981 /* semaphore array uses the global queue - just process it. */
982 otime |= update_queue(sma, -1, pt);
986 * No sops, thus the modified semaphores are not
989 for (i = 0; i < sma->sem_nsems; i++)
990 otime |= update_queue(sma, i, pt);
993 * Check the semaphores that were increased:
994 * - No complex ops, thus all sleeping ops are
996 * - if we decreased the value, then any sleeping
997 * semaphore ops wont be able to run: If the
998 * previous value was too small, then the new
999 * value will be too small, too.
1001 for (i = 0; i < nsops; i++) {
1002 if (sops[i].sem_op > 0) {
1003 otime |= update_queue(sma,
1004 sops[i].sem_num, pt);
1010 set_semotime(sma, sops);
1014 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1016 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1019 struct sembuf *sop = q->blocking;
1022 * Linux always (since 0.99.10) reported a task as sleeping on all
1023 * semaphores. This violates SUS, therefore it was changed to the
1024 * standard compliant behavior.
1025 * Give the administrators a chance to notice that an application
1026 * might misbehave because it relies on the Linux behavior.
1028 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1029 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1030 current->comm, task_pid_nr(current));
1032 if (sop->sem_num != semnum)
1035 if (count_zero && sop->sem_op == 0)
1037 if (!count_zero && sop->sem_op < 0)
1043 /* The following counts are associated to each semaphore:
1044 * semncnt number of tasks waiting on semval being nonzero
1045 * semzcnt number of tasks waiting on semval being zero
1047 * Per definition, a task waits only on the semaphore of the first semop
1048 * that cannot proceed, even if additional operation would block, too.
1050 static int count_semcnt(struct sem_array *sma, ushort semnum,
1053 struct list_head *l;
1054 struct sem_queue *q;
1058 /* First: check the simple operations. They are easy to evaluate */
1060 l = &sma->sem_base[semnum].pending_const;
1062 l = &sma->sem_base[semnum].pending_alter;
1064 list_for_each_entry(q, l, list) {
1065 /* all task on a per-semaphore list sleep on exactly
1071 /* Then: check the complex operations. */
1072 list_for_each_entry(q, &sma->pending_alter, list) {
1073 semcnt += check_qop(sma, semnum, q, count_zero);
1076 list_for_each_entry(q, &sma->pending_const, list) {
1077 semcnt += check_qop(sma, semnum, q, count_zero);
1083 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1084 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1085 * remains locked on exit.
1087 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1089 struct sem_undo *un, *tu;
1090 struct sem_queue *q, *tq;
1091 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1092 struct list_head tasks;
1095 /* Free the existing undo structures for this semaphore set. */
1096 ipc_assert_locked_object(&sma->sem_perm);
1097 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1098 list_del(&un->list_id);
1099 spin_lock(&un->ulp->lock);
1101 list_del_rcu(&un->list_proc);
1102 spin_unlock(&un->ulp->lock);
1106 /* Wake up all pending processes and let them fail with EIDRM. */
1107 INIT_LIST_HEAD(&tasks);
1108 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1109 unlink_queue(sma, q);
1110 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1113 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1114 unlink_queue(sma, q);
1115 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1117 for (i = 0; i < sma->sem_nsems; i++) {
1118 struct sem *sem = sma->sem_base + i;
1119 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1120 unlink_queue(sma, q);
1121 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1123 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1124 unlink_queue(sma, q);
1125 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1129 /* Remove the semaphore set from the IDR */
1131 sem_unlock(sma, -1);
1134 wake_up_sem_queue_do(&tasks);
1135 ns->used_sems -= sma->sem_nsems;
1136 ipc_rcu_putref(sma, sem_rcu_free);
1139 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1143 return copy_to_user(buf, in, sizeof(*in));
1146 struct semid_ds out;
1148 memset(&out, 0, sizeof(out));
1150 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1152 out.sem_otime = in->sem_otime;
1153 out.sem_ctime = in->sem_ctime;
1154 out.sem_nsems = in->sem_nsems;
1156 return copy_to_user(buf, &out, sizeof(out));
1163 static time_t get_semotime(struct sem_array *sma)
1168 res = sma->sem_base[0].sem_otime;
1169 for (i = 1; i < sma->sem_nsems; i++) {
1170 time_t to = sma->sem_base[i].sem_otime;
1178 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1179 int cmd, int version, void __user *p)
1182 struct sem_array *sma;
1188 struct seminfo seminfo;
1191 err = security_sem_semctl(NULL, cmd);
1195 memset(&seminfo, 0, sizeof(seminfo));
1196 seminfo.semmni = ns->sc_semmni;
1197 seminfo.semmns = ns->sc_semmns;
1198 seminfo.semmsl = ns->sc_semmsl;
1199 seminfo.semopm = ns->sc_semopm;
1200 seminfo.semvmx = SEMVMX;
1201 seminfo.semmnu = SEMMNU;
1202 seminfo.semmap = SEMMAP;
1203 seminfo.semume = SEMUME;
1204 down_read(&sem_ids(ns).rwsem);
1205 if (cmd == SEM_INFO) {
1206 seminfo.semusz = sem_ids(ns).in_use;
1207 seminfo.semaem = ns->used_sems;
1209 seminfo.semusz = SEMUSZ;
1210 seminfo.semaem = SEMAEM;
1212 max_id = ipc_get_maxid(&sem_ids(ns));
1213 up_read(&sem_ids(ns).rwsem);
1214 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1216 return (max_id < 0) ? 0 : max_id;
1221 struct semid64_ds tbuf;
1224 memset(&tbuf, 0, sizeof(tbuf));
1227 if (cmd == SEM_STAT) {
1228 sma = sem_obtain_object(ns, semid);
1233 id = sma->sem_perm.id;
1235 sma = sem_obtain_object_check(ns, semid);
1243 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1246 err = security_sem_semctl(sma, cmd);
1250 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1251 tbuf.sem_otime = get_semotime(sma);
1252 tbuf.sem_ctime = sma->sem_ctime;
1253 tbuf.sem_nsems = sma->sem_nsems;
1255 if (copy_semid_to_user(p, &tbuf, version))
1267 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1270 struct sem_undo *un;
1271 struct sem_array *sma;
1274 struct list_head tasks;
1276 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1277 /* big-endian 64bit */
1280 /* 32bit or little-endian 64bit */
1284 if (val > SEMVMX || val < 0)
1287 INIT_LIST_HEAD(&tasks);
1290 sma = sem_obtain_object_check(ns, semid);
1293 return PTR_ERR(sma);
1296 if (semnum < 0 || semnum >= sma->sem_nsems) {
1302 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1307 err = security_sem_semctl(sma, SETVAL);
1313 sem_lock(sma, NULL, -1);
1315 if (!ipc_valid_object(&sma->sem_perm)) {
1316 sem_unlock(sma, -1);
1321 curr = &sma->sem_base[semnum];
1323 ipc_assert_locked_object(&sma->sem_perm);
1324 list_for_each_entry(un, &sma->list_id, list_id)
1325 un->semadj[semnum] = 0;
1328 curr->sempid = task_tgid_vnr(current);
1329 sma->sem_ctime = get_seconds();
1330 /* maybe some queued-up processes were waiting for this */
1331 do_smart_update(sma, NULL, 0, 0, &tasks);
1332 sem_unlock(sma, -1);
1334 wake_up_sem_queue_do(&tasks);
1338 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1339 int cmd, void __user *p)
1341 struct sem_array *sma;
1344 ushort fast_sem_io[SEMMSL_FAST];
1345 ushort *sem_io = fast_sem_io;
1346 struct list_head tasks;
1348 INIT_LIST_HEAD(&tasks);
1351 sma = sem_obtain_object_check(ns, semid);
1354 return PTR_ERR(sma);
1357 nsems = sma->sem_nsems;
1360 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1361 goto out_rcu_wakeup;
1363 err = security_sem_semctl(sma, cmd);
1365 goto out_rcu_wakeup;
1371 ushort __user *array = p;
1374 sem_lock(sma, NULL, -1);
1375 if (!ipc_valid_object(&sma->sem_perm)) {
1379 if (nsems > SEMMSL_FAST) {
1380 if (!ipc_rcu_getref(sma)) {
1384 sem_unlock(sma, -1);
1386 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1387 if (sem_io == NULL) {
1388 ipc_rcu_putref(sma, ipc_rcu_free);
1393 sem_lock_and_putref(sma);
1394 if (!ipc_valid_object(&sma->sem_perm)) {
1399 for (i = 0; i < sma->sem_nsems; i++)
1400 sem_io[i] = sma->sem_base[i].semval;
1401 sem_unlock(sma, -1);
1404 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1411 struct sem_undo *un;
1413 if (!ipc_rcu_getref(sma)) {
1415 goto out_rcu_wakeup;
1419 if (nsems > SEMMSL_FAST) {
1420 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1421 if (sem_io == NULL) {
1422 ipc_rcu_putref(sma, ipc_rcu_free);
1427 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1428 ipc_rcu_putref(sma, ipc_rcu_free);
1433 for (i = 0; i < nsems; i++) {
1434 if (sem_io[i] > SEMVMX) {
1435 ipc_rcu_putref(sma, ipc_rcu_free);
1441 sem_lock_and_putref(sma);
1442 if (!ipc_valid_object(&sma->sem_perm)) {
1447 for (i = 0; i < nsems; i++)
1448 sma->sem_base[i].semval = sem_io[i];
1450 ipc_assert_locked_object(&sma->sem_perm);
1451 list_for_each_entry(un, &sma->list_id, list_id) {
1452 for (i = 0; i < nsems; i++)
1455 sma->sem_ctime = get_seconds();
1456 /* maybe some queued-up processes were waiting for this */
1457 do_smart_update(sma, NULL, 0, 0, &tasks);
1461 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1464 if (semnum < 0 || semnum >= nsems)
1465 goto out_rcu_wakeup;
1467 sem_lock(sma, NULL, -1);
1468 if (!ipc_valid_object(&sma->sem_perm)) {
1472 curr = &sma->sem_base[semnum];
1482 err = count_semcnt(sma, semnum, 0);
1485 err = count_semcnt(sma, semnum, 1);
1490 sem_unlock(sma, -1);
1493 wake_up_sem_queue_do(&tasks);
1495 if (sem_io != fast_sem_io)
1500 static inline unsigned long
1501 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1505 if (copy_from_user(out, buf, sizeof(*out)))
1510 struct semid_ds tbuf_old;
1512 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1515 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1516 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1517 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1527 * This function handles some semctl commands which require the rwsem
1528 * to be held in write mode.
1529 * NOTE: no locks must be held, the rwsem is taken inside this function.
1531 static int semctl_down(struct ipc_namespace *ns, int semid,
1532 int cmd, int version, void __user *p)
1534 struct sem_array *sma;
1536 struct semid64_ds semid64;
1537 struct kern_ipc_perm *ipcp;
1539 if (cmd == IPC_SET) {
1540 if (copy_semid_from_user(&semid64, p, version))
1544 down_write(&sem_ids(ns).rwsem);
1547 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1548 &semid64.sem_perm, 0);
1550 err = PTR_ERR(ipcp);
1554 sma = container_of(ipcp, struct sem_array, sem_perm);
1556 err = security_sem_semctl(sma, cmd);
1562 sem_lock(sma, NULL, -1);
1563 /* freeary unlocks the ipc object and rcu */
1567 sem_lock(sma, NULL, -1);
1568 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1571 sma->sem_ctime = get_seconds();
1579 sem_unlock(sma, -1);
1583 up_write(&sem_ids(ns).rwsem);
1587 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1590 struct ipc_namespace *ns;
1591 void __user *p = (void __user *)arg;
1596 version = ipc_parse_version(&cmd);
1597 ns = current->nsproxy->ipc_ns;
1604 return semctl_nolock(ns, semid, cmd, version, p);
1611 return semctl_main(ns, semid, semnum, cmd, p);
1613 return semctl_setval(ns, semid, semnum, arg);
1616 return semctl_down(ns, semid, cmd, version, p);
1622 /* If the task doesn't already have a undo_list, then allocate one
1623 * here. We guarantee there is only one thread using this undo list,
1624 * and current is THE ONE
1626 * If this allocation and assignment succeeds, but later
1627 * portions of this code fail, there is no need to free the sem_undo_list.
1628 * Just let it stay associated with the task, and it'll be freed later
1631 * This can block, so callers must hold no locks.
1633 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1635 struct sem_undo_list *undo_list;
1637 undo_list = current->sysvsem.undo_list;
1639 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1640 if (undo_list == NULL)
1642 spin_lock_init(&undo_list->lock);
1643 atomic_set(&undo_list->refcnt, 1);
1644 INIT_LIST_HEAD(&undo_list->list_proc);
1646 current->sysvsem.undo_list = undo_list;
1648 *undo_listp = undo_list;
1652 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1654 struct sem_undo *un;
1656 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1657 if (un->semid == semid)
1663 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1665 struct sem_undo *un;
1667 assert_spin_locked(&ulp->lock);
1669 un = __lookup_undo(ulp, semid);
1671 list_del_rcu(&un->list_proc);
1672 list_add_rcu(&un->list_proc, &ulp->list_proc);
1678 * find_alloc_undo - lookup (and if not present create) undo array
1680 * @semid: semaphore array id
1682 * The function looks up (and if not present creates) the undo structure.
1683 * The size of the undo structure depends on the size of the semaphore
1684 * array, thus the alloc path is not that straightforward.
1685 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1686 * performs a rcu_read_lock().
1688 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1690 struct sem_array *sma;
1691 struct sem_undo_list *ulp;
1692 struct sem_undo *un, *new;
1695 error = get_undo_list(&ulp);
1697 return ERR_PTR(error);
1700 spin_lock(&ulp->lock);
1701 un = lookup_undo(ulp, semid);
1702 spin_unlock(&ulp->lock);
1703 if (likely(un != NULL))
1706 /* no undo structure around - allocate one. */
1707 /* step 1: figure out the size of the semaphore array */
1708 sma = sem_obtain_object_check(ns, semid);
1711 return ERR_CAST(sma);
1714 nsems = sma->sem_nsems;
1715 if (!ipc_rcu_getref(sma)) {
1717 un = ERR_PTR(-EIDRM);
1722 /* step 2: allocate new undo structure */
1723 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1725 ipc_rcu_putref(sma, ipc_rcu_free);
1726 return ERR_PTR(-ENOMEM);
1729 /* step 3: Acquire the lock on semaphore array */
1731 sem_lock_and_putref(sma);
1732 if (!ipc_valid_object(&sma->sem_perm)) {
1733 sem_unlock(sma, -1);
1736 un = ERR_PTR(-EIDRM);
1739 spin_lock(&ulp->lock);
1742 * step 4: check for races: did someone else allocate the undo struct?
1744 un = lookup_undo(ulp, semid);
1749 /* step 5: initialize & link new undo structure */
1750 new->semadj = (short *) &new[1];
1753 assert_spin_locked(&ulp->lock);
1754 list_add_rcu(&new->list_proc, &ulp->list_proc);
1755 ipc_assert_locked_object(&sma->sem_perm);
1756 list_add(&new->list_id, &sma->list_id);
1760 spin_unlock(&ulp->lock);
1761 sem_unlock(sma, -1);
1768 * get_queue_result - retrieve the result code from sem_queue
1769 * @q: Pointer to queue structure
1771 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1772 * q->status, then we must loop until the value is replaced with the final
1773 * value: This may happen if a task is woken up by an unrelated event (e.g.
1774 * signal) and in parallel the task is woken up by another task because it got
1775 * the requested semaphores.
1777 * The function can be called with or without holding the semaphore spinlock.
1779 static int get_queue_result(struct sem_queue *q)
1784 while (unlikely(error == IN_WAKEUP)) {
1792 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1793 unsigned, nsops, const struct timespec __user *, timeout)
1795 int error = -EINVAL;
1796 struct sem_array *sma;
1797 struct sembuf fast_sops[SEMOPM_FAST];
1798 struct sembuf *sops = fast_sops, *sop;
1799 struct sem_undo *un;
1800 int undos = 0, alter = 0, max, locknum;
1801 struct sem_queue queue;
1802 unsigned long jiffies_left = 0;
1803 struct ipc_namespace *ns;
1804 struct list_head tasks;
1806 ns = current->nsproxy->ipc_ns;
1808 if (nsops < 1 || semid < 0)
1810 if (nsops > ns->sc_semopm)
1812 if (nsops > SEMOPM_FAST) {
1813 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1817 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1822 struct timespec _timeout;
1823 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1827 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1828 _timeout.tv_nsec >= 1000000000L) {
1832 jiffies_left = timespec_to_jiffies(&_timeout);
1835 for (sop = sops; sop < sops + nsops; sop++) {
1836 if (sop->sem_num >= max)
1838 if (sop->sem_flg & SEM_UNDO)
1840 if (sop->sem_op != 0)
1844 INIT_LIST_HEAD(&tasks);
1847 /* On success, find_alloc_undo takes the rcu_read_lock */
1848 un = find_alloc_undo(ns, semid);
1850 error = PTR_ERR(un);
1858 sma = sem_obtain_object_check(ns, semid);
1861 error = PTR_ERR(sma);
1866 if (max >= sma->sem_nsems)
1867 goto out_rcu_wakeup;
1870 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1871 goto out_rcu_wakeup;
1873 error = security_sem_semop(sma, sops, nsops, alter);
1875 goto out_rcu_wakeup;
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;
1905 error = perform_atomic_semop(sma, &queue);
1907 /* If the operation was successful, then do
1908 * the required updates.
1911 do_smart_update(sma, sops, nsops, 1, &tasks);
1913 set_semotime(sma, sops);
1916 goto out_unlock_free;
1918 /* We need to sleep on this operation, so we put the current
1919 * task into the pending queue and go to sleep.
1924 curr = &sma->sem_base[sops->sem_num];
1927 if (sma->complex_count) {
1928 list_add_tail(&queue.list,
1929 &sma->pending_alter);
1932 list_add_tail(&queue.list,
1933 &curr->pending_alter);
1936 list_add_tail(&queue.list, &curr->pending_const);
1939 if (!sma->complex_count)
1943 list_add_tail(&queue.list, &sma->pending_alter);
1945 list_add_tail(&queue.list, &sma->pending_const);
1947 sma->complex_count++;
1950 queue.status = -EINTR;
1951 queue.sleeper = current;
1954 __set_current_state(TASK_INTERRUPTIBLE);
1955 sem_unlock(sma, locknum);
1959 jiffies_left = schedule_timeout(jiffies_left);
1963 error = get_queue_result(&queue);
1965 if (error != -EINTR) {
1966 /* fast path: update_queue already obtained all requested
1968 * Perform a smp_mb(): User space could assume that semop()
1969 * is a memory barrier: Without the mb(), the cpu could
1970 * speculatively read in user space stale data that was
1971 * overwritten by the previous owner of the semaphore.
1979 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1982 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1984 error = get_queue_result(&queue);
1987 * Array removed? If yes, leave without sem_unlock().
1996 * If queue.status != -EINTR we are woken up by another process.
1997 * Leave without unlink_queue(), but with sem_unlock().
1999 if (error != -EINTR)
2000 goto out_unlock_free;
2003 * If an interrupt occurred we have to clean up the queue
2005 if (timeout && jiffies_left == 0)
2009 * If the wakeup was spurious, just retry
2011 if (error == -EINTR && !signal_pending(current))
2014 unlink_queue(sma, &queue);
2017 sem_unlock(sma, locknum);
2020 wake_up_sem_queue_do(&tasks);
2022 if (sops != fast_sops)
2027 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2030 return sys_semtimedop(semid, tsops, nsops, NULL);
2033 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2034 * parent and child tasks.
2037 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2039 struct sem_undo_list *undo_list;
2042 if (clone_flags & CLONE_SYSVSEM) {
2043 error = get_undo_list(&undo_list);
2046 atomic_inc(&undo_list->refcnt);
2047 tsk->sysvsem.undo_list = undo_list;
2049 tsk->sysvsem.undo_list = NULL;
2055 * add semadj values to semaphores, free undo structures.
2056 * undo structures are not freed when semaphore arrays are destroyed
2057 * so some of them may be out of date.
2058 * IMPLEMENTATION NOTE: There is some confusion over whether the
2059 * set of adjustments that needs to be done should be done in an atomic
2060 * manner or not. That is, if we are attempting to decrement the semval
2061 * should we queue up and wait until we can do so legally?
2062 * The original implementation attempted to do this (queue and wait).
2063 * The current implementation does not do so. The POSIX standard
2064 * and SVID should be consulted to determine what behavior is mandated.
2066 void exit_sem(struct task_struct *tsk)
2068 struct sem_undo_list *ulp;
2070 ulp = tsk->sysvsem.undo_list;
2073 tsk->sysvsem.undo_list = NULL;
2075 if (!atomic_dec_and_test(&ulp->refcnt))
2079 struct sem_array *sma;
2080 struct sem_undo *un;
2081 struct list_head tasks;
2085 un = list_entry_rcu(ulp->list_proc.next,
2086 struct sem_undo, list_proc);
2087 if (&un->list_proc == &ulp->list_proc) {
2089 * We must wait for freeary() before freeing this ulp,
2090 * in case we raced with last sem_undo. There is a small
2091 * possibility where we exit while freeary() didn't
2092 * finish unlocking sem_undo_list.
2094 spin_unlock_wait(&ulp->lock);
2098 spin_lock(&ulp->lock);
2100 spin_unlock(&ulp->lock);
2102 /* exit_sem raced with IPC_RMID, nothing to do */
2108 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2109 /* exit_sem raced with IPC_RMID, nothing to do */
2115 sem_lock(sma, NULL, -1);
2116 /* exit_sem raced with IPC_RMID, nothing to do */
2117 if (!ipc_valid_object(&sma->sem_perm)) {
2118 sem_unlock(sma, -1);
2122 un = __lookup_undo(ulp, semid);
2124 /* exit_sem raced with IPC_RMID+semget() that created
2125 * exactly the same semid. Nothing to do.
2127 sem_unlock(sma, -1);
2132 /* remove un from the linked lists */
2133 ipc_assert_locked_object(&sma->sem_perm);
2134 list_del(&un->list_id);
2136 /* we are the last process using this ulp, acquiring ulp->lock
2137 * isn't required. Besides that, we are also protected against
2138 * IPC_RMID as we hold sma->sem_perm lock now
2140 list_del_rcu(&un->list_proc);
2142 /* perform adjustments registered in un */
2143 for (i = 0; i < sma->sem_nsems; i++) {
2144 struct sem *semaphore = &sma->sem_base[i];
2145 if (un->semadj[i]) {
2146 semaphore->semval += un->semadj[i];
2148 * Range checks of the new semaphore value,
2149 * not defined by sus:
2150 * - Some unices ignore the undo entirely
2151 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2152 * - some cap the value (e.g. FreeBSD caps
2153 * at 0, but doesn't enforce SEMVMX)
2155 * Linux caps the semaphore value, both at 0
2158 * Manfred <manfred@colorfullife.com>
2160 if (semaphore->semval < 0)
2161 semaphore->semval = 0;
2162 if (semaphore->semval > SEMVMX)
2163 semaphore->semval = SEMVMX;
2164 semaphore->sempid = task_tgid_vnr(current);
2167 /* maybe some queued-up processes were waiting for this */
2168 INIT_LIST_HEAD(&tasks);
2169 do_smart_update(sma, NULL, 0, 1, &tasks);
2170 sem_unlock(sma, -1);
2172 wake_up_sem_queue_do(&tasks);
2179 #ifdef CONFIG_PROC_FS
2180 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2182 struct user_namespace *user_ns = seq_user_ns(s);
2183 struct sem_array *sma = it;
2187 * The proc interface isn't aware of sem_lock(), it calls
2188 * ipc_lock_object() directly (in sysvipc_find_ipc).
2189 * In order to stay compatible with sem_lock(), we must wait until
2190 * all simple semop() calls have left their critical regions.
2192 sem_wait_array(sma);
2194 sem_otime = get_semotime(sma);
2197 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2202 from_kuid_munged(user_ns, sma->sem_perm.uid),
2203 from_kgid_munged(user_ns, sma->sem_perm.gid),
2204 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2205 from_kgid_munged(user_ns, sma->sem_perm.cgid),