3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
22 * Pavel Emelianov <xemul@openvz.org>
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semcnt()
51 * - the task that performs a successful semop() scans the list of all
52 * sleeping tasks and completes any pending operations that can be fulfilled.
53 * Semaphores are actively given to waiting tasks (necessary for FIFO).
54 * (see update_queue())
55 * - To improve the scalability, the actual wake-up calls are performed after
56 * dropping all locks. (see wake_up_sem_queue_prepare(),
57 * wake_up_sem_queue_do())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - The synchronizations between wake-ups due to a timeout/signal and a
63 * wake-up due to a completed semaphore operation is achieved by using an
64 * intermediate state (IN_WAKEUP).
65 * - UNDO values are stored in an array (one per process and per
66 * semaphore array, lazily allocated). For backwards compatibility, multiple
67 * modes for the UNDO variables are supported (per process, per thread)
68 * (see copy_semundo, CLONE_SYSVSEM)
69 * - There are two lists of the pending operations: a per-array list
70 * and per-semaphore list (stored in the array). This allows to achieve FIFO
71 * ordering without always scanning all pending operations.
72 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
75 #include <linux/slab.h>
76 #include <linux/spinlock.h>
77 #include <linux/init.h>
78 #include <linux/proc_fs.h>
79 #include <linux/time.h>
80 #include <linux/security.h>
81 #include <linux/syscalls.h>
82 #include <linux/audit.h>
83 #include <linux/capability.h>
84 #include <linux/seq_file.h>
85 #include <linux/rwsem.h>
86 #include <linux/nsproxy.h>
87 #include <linux/ipc_namespace.h>
89 #include <linux/uaccess.h>
92 /* One semaphore structure for each semaphore in the system. */
94 int semval; /* current value */
95 int sempid; /* pid of last operation */
96 spinlock_t lock; /* spinlock for fine-grained semtimedop */
97 struct list_head pending_alter; /* pending single-sop operations */
98 /* that alter the semaphore */
99 struct list_head pending_const; /* pending single-sop operations */
100 /* that do not alter the semaphore*/
101 time_t sem_otime; /* candidate for sem_otime */
102 } ____cacheline_aligned_in_smp;
104 /* One queue for each sleeping process in the system. */
106 struct list_head list; /* queue of pending operations */
107 struct task_struct *sleeper; /* this process */
108 struct sem_undo *undo; /* undo structure */
109 int pid; /* process id of requesting process */
110 int status; /* completion status of operation */
111 struct sembuf *sops; /* array of pending operations */
112 struct sembuf *blocking; /* the operation that blocked */
113 int nsops; /* number of operations */
114 int alter; /* does *sops alter the array? */
117 /* Each task has a list of undo requests. They are executed automatically
118 * when the process exits.
121 struct list_head list_proc; /* per-process list: *
122 * all undos from one process
124 struct rcu_head rcu; /* rcu struct for sem_undo */
125 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
126 struct list_head list_id; /* per semaphore array list:
127 * all undos for one array */
128 int semid; /* semaphore set identifier */
129 short *semadj; /* array of adjustments */
130 /* one per semaphore */
133 /* sem_undo_list controls shared access to the list of sem_undo structures
134 * that may be shared among all a CLONE_SYSVSEM task group.
136 struct sem_undo_list {
139 struct list_head list_proc;
143 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
145 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
147 static int newary(struct ipc_namespace *, struct ipc_params *);
148 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
149 #ifdef CONFIG_PROC_FS
150 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
153 #define SEMMSL_FAST 256 /* 512 bytes on stack */
154 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
159 * sem_array.complex_count,
160 * sem_array.pending{_alter,_cont},
161 * sem_array.sem_undo: global sem_lock() for read/write
162 * sem_undo.proc_next: only "current" is allowed to read/write that field.
164 * sem_array.sem_base[i].pending_{const,alter}:
165 * global or semaphore sem_lock() for read/write
168 #define sc_semmsl sem_ctls[0]
169 #define sc_semmns sem_ctls[1]
170 #define sc_semopm sem_ctls[2]
171 #define sc_semmni sem_ctls[3]
173 void sem_init_ns(struct ipc_namespace *ns)
175 ns->sc_semmsl = SEMMSL;
176 ns->sc_semmns = SEMMNS;
177 ns->sc_semopm = SEMOPM;
178 ns->sc_semmni = SEMMNI;
180 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
184 void sem_exit_ns(struct ipc_namespace *ns)
186 free_ipcs(ns, &sem_ids(ns), freeary);
187 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
191 void __init sem_init(void)
193 sem_init_ns(&init_ipc_ns);
194 ipc_init_proc_interface("sysvipc/sem",
195 " key semid perms nsems uid gid cuid cgid otime ctime\n",
196 IPC_SEM_IDS, sysvipc_sem_proc_show);
200 * unmerge_queues - unmerge queues, if possible.
201 * @sma: semaphore array
203 * The function unmerges the wait queues if complex_count is 0.
204 * It must be called prior to dropping the global semaphore array lock.
206 static void unmerge_queues(struct sem_array *sma)
208 struct sem_queue *q, *tq;
210 /* complex operations still around? */
211 if (sma->complex_count)
214 * We will switch back to simple mode.
215 * Move all pending operation back into the per-semaphore
218 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
220 curr = &sma->sem_base[q->sops[0].sem_num];
222 list_add_tail(&q->list, &curr->pending_alter);
224 INIT_LIST_HEAD(&sma->pending_alter);
228 * merge_queues - merge single semop queues into global queue
229 * @sma: semaphore array
231 * This function merges all per-semaphore queues into the global queue.
232 * It is necessary to achieve FIFO ordering for the pending single-sop
233 * operations when a multi-semop operation must sleep.
234 * Only the alter operations must be moved, the const operations can stay.
236 static void merge_queues(struct sem_array *sma)
239 for (i = 0; i < sma->sem_nsems; i++) {
240 struct sem *sem = sma->sem_base + i;
242 list_splice_init(&sem->pending_alter, &sma->pending_alter);
246 static void sem_rcu_free(struct rcu_head *head)
248 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
249 struct sem_array *sma = ipc_rcu_to_struct(p);
251 security_sem_free(sma);
256 * Wait until all currently ongoing simple ops have completed.
257 * Caller must own sem_perm.lock.
258 * New simple ops cannot start, because simple ops first check
259 * that sem_perm.lock is free.
260 * that a) sem_perm.lock is free and b) complex_count is 0.
262 static void sem_wait_array(struct sem_array *sma)
267 if (sma->complex_count) {
268 /* The thread that increased sma->complex_count waited on
269 * all sem->lock locks. Thus we don't need to wait again.
274 for (i = 0; i < sma->sem_nsems; i++) {
275 sem = sma->sem_base + i;
276 spin_unlock_wait(&sem->lock);
281 * If the request contains only one semaphore operation, and there are
282 * no complex transactions pending, lock only the semaphore involved.
283 * Otherwise, lock the entire semaphore array, since we either have
284 * multiple semaphores in our own semops, or we need to look at
285 * semaphores from other pending complex operations.
287 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
293 /* Complex operation - acquire a full lock */
294 ipc_lock_object(&sma->sem_perm);
296 /* And wait until all simple ops that are processed
297 * right now have dropped their locks.
304 * Only one semaphore affected - try to optimize locking.
306 * - optimized locking is possible if no complex operation
307 * is either enqueued or processed right now.
308 * - The test for enqueued complex ops is simple:
309 * sma->complex_count != 0
310 * - Testing for complex ops that are processed right now is
311 * a bit more difficult. Complex ops acquire the full lock
312 * and first wait that the running simple ops have completed.
314 * Thus: If we own a simple lock and the global lock is free
315 * and complex_count is now 0, then it will stay 0 and
316 * thus just locking sem->lock is sufficient.
318 sem = sma->sem_base + sops->sem_num;
320 if (sma->complex_count == 0) {
322 * It appears that no complex operation is around.
323 * Acquire the per-semaphore lock.
325 spin_lock(&sem->lock);
327 /* Then check that the global lock is free */
328 if (!spin_is_locked(&sma->sem_perm.lock)) {
330 * The ipc object lock check must be visible on all
331 * cores before rechecking the complex count. Otherwise
332 * we can race with another thread that does:
334 * spin_unlock(sem_perm.lock);
339 * Now repeat the test of complex_count:
340 * It can't change anymore until we drop sem->lock.
341 * Thus: if is now 0, then it will stay 0.
343 if (sma->complex_count == 0) {
344 /* fast path successful! */
345 return sops->sem_num;
348 spin_unlock(&sem->lock);
351 /* slow path: acquire the full lock */
352 ipc_lock_object(&sma->sem_perm);
354 if (sma->complex_count == 0) {
356 * There is no complex operation, thus we can switch
357 * back to the fast path.
359 spin_lock(&sem->lock);
360 ipc_unlock_object(&sma->sem_perm);
361 return sops->sem_num;
363 /* Not a false alarm, thus complete the sequence for a
371 static inline void sem_unlock(struct sem_array *sma, int locknum)
375 ipc_unlock_object(&sma->sem_perm);
377 struct sem *sem = sma->sem_base + locknum;
378 spin_unlock(&sem->lock);
383 * sem_lock_(check_) routines are called in the paths where the rwsem
386 * The caller holds the RCU read lock.
388 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
389 int id, struct sembuf *sops, int nsops, int *locknum)
391 struct kern_ipc_perm *ipcp;
392 struct sem_array *sma;
394 ipcp = ipc_obtain_object(&sem_ids(ns), id);
396 return ERR_CAST(ipcp);
398 sma = container_of(ipcp, struct sem_array, sem_perm);
399 *locknum = sem_lock(sma, sops, nsops);
401 /* ipc_rmid() may have already freed the ID while sem_lock
402 * was spinning: verify that the structure is still valid
404 if (ipc_valid_object(ipcp))
405 return container_of(ipcp, struct sem_array, sem_perm);
407 sem_unlock(sma, *locknum);
408 return ERR_PTR(-EINVAL);
411 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
413 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
416 return ERR_CAST(ipcp);
418 return container_of(ipcp, struct sem_array, sem_perm);
421 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
424 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
427 return ERR_CAST(ipcp);
429 return container_of(ipcp, struct sem_array, sem_perm);
432 static inline void sem_lock_and_putref(struct sem_array *sma)
434 sem_lock(sma, NULL, -1);
435 ipc_rcu_putref(sma, ipc_rcu_free);
438 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
440 ipc_rmid(&sem_ids(ns), &s->sem_perm);
444 * Lockless wakeup algorithm:
445 * Without the check/retry algorithm a lockless wakeup is possible:
446 * - queue.status is initialized to -EINTR before blocking.
447 * - wakeup is performed by
448 * * unlinking the queue entry from the pending list
449 * * setting queue.status to IN_WAKEUP
450 * This is the notification for the blocked thread that a
451 * result value is imminent.
452 * * call wake_up_process
453 * * set queue.status to the final value.
454 * - the previously blocked thread checks queue.status:
455 * * if it's IN_WAKEUP, then it must wait until the value changes
456 * * if it's not -EINTR, then the operation was completed by
457 * update_queue. semtimedop can return queue.status without
458 * performing any operation on the sem array.
459 * * otherwise it must acquire the spinlock and check what's up.
461 * The two-stage algorithm is necessary to protect against the following
463 * - if queue.status is set after wake_up_process, then the woken up idle
464 * thread could race forward and try (and fail) to acquire sma->lock
465 * before update_queue had a chance to set queue.status
466 * - if queue.status is written before wake_up_process and if the
467 * blocked process is woken up by a signal between writing
468 * queue.status and the wake_up_process, then the woken up
469 * process could return from semtimedop and die by calling
470 * sys_exit before wake_up_process is called. Then wake_up_process
471 * will oops, because the task structure is already invalid.
472 * (yes, this happened on s390 with sysv msg).
478 * newary - Create a new semaphore set
480 * @params: ptr to the structure that contains key, semflg and nsems
482 * Called with sem_ids.rwsem held (as a writer)
484 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
488 struct sem_array *sma;
490 key_t key = params->key;
491 int nsems = params->u.nsems;
492 int semflg = params->flg;
497 if (ns->used_sems + nsems > ns->sc_semmns)
500 size = sizeof(*sma) + nsems * sizeof(struct sem);
501 sma = ipc_rcu_alloc(size);
505 memset(sma, 0, size);
507 sma->sem_perm.mode = (semflg & S_IRWXUGO);
508 sma->sem_perm.key = key;
510 sma->sem_perm.security = NULL;
511 retval = security_sem_alloc(sma);
513 ipc_rcu_putref(sma, ipc_rcu_free);
517 sma->sem_base = (struct sem *) &sma[1];
519 for (i = 0; i < nsems; i++) {
520 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
521 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
522 spin_lock_init(&sma->sem_base[i].lock);
525 sma->complex_count = 0;
526 INIT_LIST_HEAD(&sma->pending_alter);
527 INIT_LIST_HEAD(&sma->pending_const);
528 INIT_LIST_HEAD(&sma->list_id);
529 sma->sem_nsems = nsems;
530 sma->sem_ctime = get_seconds();
532 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
534 ipc_rcu_putref(sma, sem_rcu_free);
537 ns->used_sems += nsems;
542 return sma->sem_perm.id;
547 * Called with sem_ids.rwsem and ipcp locked.
549 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
551 struct sem_array *sma;
553 sma = container_of(ipcp, struct sem_array, sem_perm);
554 return security_sem_associate(sma, semflg);
558 * Called with sem_ids.rwsem and ipcp locked.
560 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
561 struct ipc_params *params)
563 struct sem_array *sma;
565 sma = container_of(ipcp, struct sem_array, sem_perm);
566 if (params->u.nsems > sma->sem_nsems)
572 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
574 struct ipc_namespace *ns;
575 static const struct ipc_ops sem_ops = {
577 .associate = sem_security,
578 .more_checks = sem_more_checks,
580 struct ipc_params sem_params;
582 ns = current->nsproxy->ipc_ns;
584 if (nsems < 0 || nsems > ns->sc_semmsl)
587 sem_params.key = key;
588 sem_params.flg = semflg;
589 sem_params.u.nsems = nsems;
591 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
595 * perform_atomic_semop - Perform (if possible) a semaphore operation
596 * @sma: semaphore array
597 * @q: struct sem_queue that describes the operation
599 * Returns 0 if the operation was possible.
600 * Returns 1 if the operation is impossible, the caller must sleep.
601 * Negative values are error codes.
603 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
605 int result, sem_op, nsops, pid;
615 for (sop = sops; sop < sops + nsops; sop++) {
616 curr = sma->sem_base + sop->sem_num;
617 sem_op = sop->sem_op;
618 result = curr->semval;
620 if (!sem_op && result)
629 if (sop->sem_flg & SEM_UNDO) {
630 int undo = un->semadj[sop->sem_num] - sem_op;
631 /* Exceeding the undo range is an error. */
632 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
634 un->semadj[sop->sem_num] = undo;
637 curr->semval = result;
642 while (sop >= sops) {
643 sma->sem_base[sop->sem_num].sempid = pid;
656 if (sop->sem_flg & IPC_NOWAIT)
663 while (sop >= sops) {
664 sem_op = sop->sem_op;
665 sma->sem_base[sop->sem_num].semval -= sem_op;
666 if (sop->sem_flg & SEM_UNDO)
667 un->semadj[sop->sem_num] += sem_op;
674 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
675 * @q: queue entry that must be signaled
676 * @error: Error value for the signal
678 * Prepare the wake-up of the queue entry q.
680 static void wake_up_sem_queue_prepare(struct list_head *pt,
681 struct sem_queue *q, int error)
683 if (list_empty(pt)) {
685 * Hold preempt off so that we don't get preempted and have the
686 * wakee busy-wait until we're scheduled back on.
690 q->status = IN_WAKEUP;
693 list_add_tail(&q->list, pt);
697 * wake_up_sem_queue_do - do the actual wake-up
698 * @pt: list of tasks to be woken up
700 * Do the actual wake-up.
701 * The function is called without any locks held, thus the semaphore array
702 * could be destroyed already and the tasks can disappear as soon as the
703 * status is set to the actual return code.
705 static void wake_up_sem_queue_do(struct list_head *pt)
707 struct sem_queue *q, *t;
710 did_something = !list_empty(pt);
711 list_for_each_entry_safe(q, t, pt, list) {
712 wake_up_process(q->sleeper);
713 /* q can disappear immediately after writing q->status. */
721 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
725 sma->complex_count--;
728 /** check_restart(sma, q)
729 * @sma: semaphore array
730 * @q: the operation that just completed
732 * update_queue is O(N^2) when it restarts scanning the whole queue of
733 * waiting operations. Therefore this function checks if the restart is
734 * really necessary. It is called after a previously waiting operation
735 * modified the array.
736 * Note that wait-for-zero operations are handled without restart.
738 static int check_restart(struct sem_array *sma, struct sem_queue *q)
740 /* pending complex alter operations are too difficult to analyse */
741 if (!list_empty(&sma->pending_alter))
744 /* we were a sleeping complex operation. Too difficult */
748 /* It is impossible that someone waits for the new value:
749 * - complex operations always restart.
750 * - wait-for-zero are handled seperately.
751 * - q is a previously sleeping simple operation that
752 * altered the array. It must be a decrement, because
753 * simple increments never sleep.
754 * - If there are older (higher priority) decrements
755 * in the queue, then they have observed the original
756 * semval value and couldn't proceed. The operation
757 * decremented to value - thus they won't proceed either.
763 * wake_const_ops - wake up non-alter tasks
764 * @sma: semaphore array.
765 * @semnum: semaphore that was modified.
766 * @pt: list head for the tasks that must be woken up.
768 * wake_const_ops must be called after a semaphore in a semaphore array
769 * was set to 0. If complex const operations are pending, wake_const_ops must
770 * be called with semnum = -1, as well as with the number of each modified
772 * The tasks that must be woken up are added to @pt. The return code
773 * is stored in q->pid.
774 * The function returns 1 if at least one operation was completed successfully.
776 static int wake_const_ops(struct sem_array *sma, int semnum,
777 struct list_head *pt)
780 struct list_head *walk;
781 struct list_head *pending_list;
782 int semop_completed = 0;
785 pending_list = &sma->pending_const;
787 pending_list = &sma->sem_base[semnum].pending_const;
789 walk = pending_list->next;
790 while (walk != pending_list) {
793 q = container_of(walk, struct sem_queue, list);
796 error = perform_atomic_semop(sma, q);
799 /* operation completed, remove from queue & wakeup */
801 unlink_queue(sma, q);
803 wake_up_sem_queue_prepare(pt, q, error);
808 return semop_completed;
812 * do_smart_wakeup_zero - wakeup all wait for zero tasks
813 * @sma: semaphore array
814 * @sops: operations that were performed
815 * @nsops: number of operations
816 * @pt: list head of the tasks that must be woken up.
818 * Checks all required queue for wait-for-zero operations, based
819 * on the actual changes that were performed on the semaphore array.
820 * The function returns 1 if at least one operation was completed successfully.
822 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
823 int nsops, struct list_head *pt)
826 int semop_completed = 0;
829 /* first: the per-semaphore queues, if known */
831 for (i = 0; i < nsops; i++) {
832 int num = sops[i].sem_num;
834 if (sma->sem_base[num].semval == 0) {
836 semop_completed |= wake_const_ops(sma, num, pt);
841 * No sops means modified semaphores not known.
842 * Assume all were changed.
844 for (i = 0; i < sma->sem_nsems; i++) {
845 if (sma->sem_base[i].semval == 0) {
847 semop_completed |= wake_const_ops(sma, i, pt);
852 * If one of the modified semaphores got 0,
853 * then check the global queue, too.
856 semop_completed |= wake_const_ops(sma, -1, pt);
858 return semop_completed;
863 * update_queue - look for tasks that can be completed.
864 * @sma: semaphore array.
865 * @semnum: semaphore that was modified.
866 * @pt: list head for the tasks that must be woken up.
868 * update_queue must be called after a semaphore in a semaphore array
869 * was modified. If multiple semaphores were modified, update_queue must
870 * be called with semnum = -1, as well as with the number of each modified
872 * The tasks that must be woken up are added to @pt. The return code
873 * is stored in q->pid.
874 * The function internally checks if const operations can now succeed.
876 * The function return 1 if at least one semop was completed successfully.
878 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
881 struct list_head *walk;
882 struct list_head *pending_list;
883 int semop_completed = 0;
886 pending_list = &sma->pending_alter;
888 pending_list = &sma->sem_base[semnum].pending_alter;
891 walk = pending_list->next;
892 while (walk != pending_list) {
895 q = container_of(walk, struct sem_queue, list);
898 /* If we are scanning the single sop, per-semaphore list of
899 * one semaphore and that semaphore is 0, then it is not
900 * necessary to scan further: simple increments
901 * that affect only one entry succeed immediately and cannot
902 * be in the per semaphore pending queue, and decrements
903 * cannot be successful if the value is already 0.
905 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
908 error = perform_atomic_semop(sma, q);
910 /* Does q->sleeper still need to sleep? */
914 unlink_queue(sma, q);
920 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
921 restart = check_restart(sma, q);
924 wake_up_sem_queue_prepare(pt, q, error);
928 return semop_completed;
932 * set_semotime - set sem_otime
933 * @sma: semaphore array
934 * @sops: operations that modified the array, may be NULL
936 * sem_otime is replicated to avoid cache line trashing.
937 * This function sets one instance to the current time.
939 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
942 sma->sem_base[0].sem_otime = get_seconds();
944 sma->sem_base[sops[0].sem_num].sem_otime =
950 * do_smart_update - optimized update_queue
951 * @sma: semaphore array
952 * @sops: operations that were performed
953 * @nsops: number of operations
954 * @otime: force setting otime
955 * @pt: list head of the tasks that must be woken up.
957 * do_smart_update() does the required calls to update_queue and wakeup_zero,
958 * based on the actual changes that were performed on the semaphore array.
959 * Note that the function does not do the actual wake-up: the caller is
960 * responsible for calling wake_up_sem_queue_do(@pt).
961 * It is safe to perform this call after dropping all locks.
963 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
964 int otime, struct list_head *pt)
968 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
970 if (!list_empty(&sma->pending_alter)) {
971 /* semaphore array uses the global queue - just process it. */
972 otime |= update_queue(sma, -1, pt);
976 * No sops, thus the modified semaphores are not
979 for (i = 0; i < sma->sem_nsems; i++)
980 otime |= update_queue(sma, i, pt);
983 * Check the semaphores that were increased:
984 * - No complex ops, thus all sleeping ops are
986 * - if we decreased the value, then any sleeping
987 * semaphore ops wont be able to run: If the
988 * previous value was too small, then the new
989 * value will be too small, too.
991 for (i = 0; i < nsops; i++) {
992 if (sops[i].sem_op > 0) {
993 otime |= update_queue(sma,
994 sops[i].sem_num, pt);
1000 set_semotime(sma, sops);
1004 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1006 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1009 struct sembuf *sop = q->blocking;
1012 * Linux always (since 0.99.10) reported a task as sleeping on all
1013 * semaphores. This violates SUS, therefore it was changed to the
1014 * standard compliant behavior.
1015 * Give the administrators a chance to notice that an application
1016 * might misbehave because it relies on the Linux behavior.
1018 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1019 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1020 current->comm, task_pid_nr(current));
1022 if (sop->sem_num != semnum)
1025 if (count_zero && sop->sem_op == 0)
1027 if (!count_zero && sop->sem_op < 0)
1033 /* The following counts are associated to each semaphore:
1034 * semncnt number of tasks waiting on semval being nonzero
1035 * semzcnt number of tasks waiting on semval being zero
1037 * Per definition, a task waits only on the semaphore of the first semop
1038 * that cannot proceed, even if additional operation would block, too.
1040 static int count_semcnt(struct sem_array *sma, ushort semnum,
1043 struct list_head *l;
1044 struct sem_queue *q;
1048 /* First: check the simple operations. They are easy to evaluate */
1050 l = &sma->sem_base[semnum].pending_const;
1052 l = &sma->sem_base[semnum].pending_alter;
1054 list_for_each_entry(q, l, list) {
1055 /* all task on a per-semaphore list sleep on exactly
1061 /* Then: check the complex operations. */
1062 list_for_each_entry(q, &sma->pending_alter, list) {
1063 semcnt += check_qop(sma, semnum, q, count_zero);
1066 list_for_each_entry(q, &sma->pending_const, list) {
1067 semcnt += check_qop(sma, semnum, q, count_zero);
1073 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1074 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1075 * remains locked on exit.
1077 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1079 struct sem_undo *un, *tu;
1080 struct sem_queue *q, *tq;
1081 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1082 struct list_head tasks;
1085 /* Free the existing undo structures for this semaphore set. */
1086 ipc_assert_locked_object(&sma->sem_perm);
1087 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1088 list_del(&un->list_id);
1089 spin_lock(&un->ulp->lock);
1091 list_del_rcu(&un->list_proc);
1092 spin_unlock(&un->ulp->lock);
1096 /* Wake up all pending processes and let them fail with EIDRM. */
1097 INIT_LIST_HEAD(&tasks);
1098 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1099 unlink_queue(sma, q);
1100 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1103 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1104 unlink_queue(sma, q);
1105 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1107 for (i = 0; i < sma->sem_nsems; i++) {
1108 struct sem *sem = sma->sem_base + i;
1109 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1110 unlink_queue(sma, q);
1111 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1113 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1114 unlink_queue(sma, q);
1115 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1119 /* Remove the semaphore set from the IDR */
1121 sem_unlock(sma, -1);
1124 wake_up_sem_queue_do(&tasks);
1125 ns->used_sems -= sma->sem_nsems;
1126 ipc_rcu_putref(sma, sem_rcu_free);
1129 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1133 return copy_to_user(buf, in, sizeof(*in));
1136 struct semid_ds out;
1138 memset(&out, 0, sizeof(out));
1140 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1142 out.sem_otime = in->sem_otime;
1143 out.sem_ctime = in->sem_ctime;
1144 out.sem_nsems = in->sem_nsems;
1146 return copy_to_user(buf, &out, sizeof(out));
1153 static time_t get_semotime(struct sem_array *sma)
1158 res = sma->sem_base[0].sem_otime;
1159 for (i = 1; i < sma->sem_nsems; i++) {
1160 time_t to = sma->sem_base[i].sem_otime;
1168 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1169 int cmd, int version, void __user *p)
1172 struct sem_array *sma;
1178 struct seminfo seminfo;
1181 err = security_sem_semctl(NULL, cmd);
1185 memset(&seminfo, 0, sizeof(seminfo));
1186 seminfo.semmni = ns->sc_semmni;
1187 seminfo.semmns = ns->sc_semmns;
1188 seminfo.semmsl = ns->sc_semmsl;
1189 seminfo.semopm = ns->sc_semopm;
1190 seminfo.semvmx = SEMVMX;
1191 seminfo.semmnu = SEMMNU;
1192 seminfo.semmap = SEMMAP;
1193 seminfo.semume = SEMUME;
1194 down_read(&sem_ids(ns).rwsem);
1195 if (cmd == SEM_INFO) {
1196 seminfo.semusz = sem_ids(ns).in_use;
1197 seminfo.semaem = ns->used_sems;
1199 seminfo.semusz = SEMUSZ;
1200 seminfo.semaem = SEMAEM;
1202 max_id = ipc_get_maxid(&sem_ids(ns));
1203 up_read(&sem_ids(ns).rwsem);
1204 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1206 return (max_id < 0) ? 0 : max_id;
1211 struct semid64_ds tbuf;
1214 memset(&tbuf, 0, sizeof(tbuf));
1217 if (cmd == SEM_STAT) {
1218 sma = sem_obtain_object(ns, semid);
1223 id = sma->sem_perm.id;
1225 sma = sem_obtain_object_check(ns, semid);
1233 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1236 err = security_sem_semctl(sma, cmd);
1240 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1241 tbuf.sem_otime = get_semotime(sma);
1242 tbuf.sem_ctime = sma->sem_ctime;
1243 tbuf.sem_nsems = sma->sem_nsems;
1245 if (copy_semid_to_user(p, &tbuf, version))
1257 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1260 struct sem_undo *un;
1261 struct sem_array *sma;
1264 struct list_head tasks;
1266 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1267 /* big-endian 64bit */
1270 /* 32bit or little-endian 64bit */
1274 if (val > SEMVMX || val < 0)
1277 INIT_LIST_HEAD(&tasks);
1280 sma = sem_obtain_object_check(ns, semid);
1283 return PTR_ERR(sma);
1286 if (semnum < 0 || semnum >= sma->sem_nsems) {
1292 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1297 err = security_sem_semctl(sma, SETVAL);
1303 sem_lock(sma, NULL, -1);
1305 if (!ipc_valid_object(&sma->sem_perm)) {
1306 sem_unlock(sma, -1);
1311 curr = &sma->sem_base[semnum];
1313 ipc_assert_locked_object(&sma->sem_perm);
1314 list_for_each_entry(un, &sma->list_id, list_id)
1315 un->semadj[semnum] = 0;
1318 curr->sempid = task_tgid_vnr(current);
1319 sma->sem_ctime = get_seconds();
1320 /* maybe some queued-up processes were waiting for this */
1321 do_smart_update(sma, NULL, 0, 0, &tasks);
1322 sem_unlock(sma, -1);
1324 wake_up_sem_queue_do(&tasks);
1328 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1329 int cmd, void __user *p)
1331 struct sem_array *sma;
1334 ushort fast_sem_io[SEMMSL_FAST];
1335 ushort *sem_io = fast_sem_io;
1336 struct list_head tasks;
1338 INIT_LIST_HEAD(&tasks);
1341 sma = sem_obtain_object_check(ns, semid);
1344 return PTR_ERR(sma);
1347 nsems = sma->sem_nsems;
1350 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1351 goto out_rcu_wakeup;
1353 err = security_sem_semctl(sma, cmd);
1355 goto out_rcu_wakeup;
1361 ushort __user *array = p;
1364 sem_lock(sma, NULL, -1);
1365 if (!ipc_valid_object(&sma->sem_perm)) {
1369 if (nsems > SEMMSL_FAST) {
1370 if (!ipc_rcu_getref(sma)) {
1374 sem_unlock(sma, -1);
1376 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1377 if (sem_io == NULL) {
1378 ipc_rcu_putref(sma, ipc_rcu_free);
1383 sem_lock_and_putref(sma);
1384 if (!ipc_valid_object(&sma->sem_perm)) {
1389 for (i = 0; i < sma->sem_nsems; i++)
1390 sem_io[i] = sma->sem_base[i].semval;
1391 sem_unlock(sma, -1);
1394 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1401 struct sem_undo *un;
1403 if (!ipc_rcu_getref(sma)) {
1405 goto out_rcu_wakeup;
1409 if (nsems > SEMMSL_FAST) {
1410 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1411 if (sem_io == NULL) {
1412 ipc_rcu_putref(sma, ipc_rcu_free);
1417 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1418 ipc_rcu_putref(sma, ipc_rcu_free);
1423 for (i = 0; i < nsems; i++) {
1424 if (sem_io[i] > SEMVMX) {
1425 ipc_rcu_putref(sma, ipc_rcu_free);
1431 sem_lock_and_putref(sma);
1432 if (!ipc_valid_object(&sma->sem_perm)) {
1437 for (i = 0; i < nsems; i++)
1438 sma->sem_base[i].semval = sem_io[i];
1440 ipc_assert_locked_object(&sma->sem_perm);
1441 list_for_each_entry(un, &sma->list_id, list_id) {
1442 for (i = 0; i < nsems; i++)
1445 sma->sem_ctime = get_seconds();
1446 /* maybe some queued-up processes were waiting for this */
1447 do_smart_update(sma, NULL, 0, 0, &tasks);
1451 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1454 if (semnum < 0 || semnum >= nsems)
1455 goto out_rcu_wakeup;
1457 sem_lock(sma, NULL, -1);
1458 if (!ipc_valid_object(&sma->sem_perm)) {
1462 curr = &sma->sem_base[semnum];
1472 err = count_semcnt(sma, semnum, 0);
1475 err = count_semcnt(sma, semnum, 1);
1480 sem_unlock(sma, -1);
1483 wake_up_sem_queue_do(&tasks);
1485 if (sem_io != fast_sem_io)
1486 ipc_free(sem_io, sizeof(ushort)*nsems);
1490 static inline unsigned long
1491 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1495 if (copy_from_user(out, buf, sizeof(*out)))
1500 struct semid_ds tbuf_old;
1502 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1505 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1506 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1507 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1517 * This function handles some semctl commands which require the rwsem
1518 * to be held in write mode.
1519 * NOTE: no locks must be held, the rwsem is taken inside this function.
1521 static int semctl_down(struct ipc_namespace *ns, int semid,
1522 int cmd, int version, void __user *p)
1524 struct sem_array *sma;
1526 struct semid64_ds semid64;
1527 struct kern_ipc_perm *ipcp;
1529 if (cmd == IPC_SET) {
1530 if (copy_semid_from_user(&semid64, p, version))
1534 down_write(&sem_ids(ns).rwsem);
1537 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1538 &semid64.sem_perm, 0);
1540 err = PTR_ERR(ipcp);
1544 sma = container_of(ipcp, struct sem_array, sem_perm);
1546 err = security_sem_semctl(sma, cmd);
1552 sem_lock(sma, NULL, -1);
1553 /* freeary unlocks the ipc object and rcu */
1557 sem_lock(sma, NULL, -1);
1558 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1561 sma->sem_ctime = get_seconds();
1569 sem_unlock(sma, -1);
1573 up_write(&sem_ids(ns).rwsem);
1577 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1580 struct ipc_namespace *ns;
1581 void __user *p = (void __user *)arg;
1586 version = ipc_parse_version(&cmd);
1587 ns = current->nsproxy->ipc_ns;
1594 return semctl_nolock(ns, semid, cmd, version, p);
1601 return semctl_main(ns, semid, semnum, cmd, p);
1603 return semctl_setval(ns, semid, semnum, arg);
1606 return semctl_down(ns, semid, cmd, version, p);
1612 /* If the task doesn't already have a undo_list, then allocate one
1613 * here. We guarantee there is only one thread using this undo list,
1614 * and current is THE ONE
1616 * If this allocation and assignment succeeds, but later
1617 * portions of this code fail, there is no need to free the sem_undo_list.
1618 * Just let it stay associated with the task, and it'll be freed later
1621 * This can block, so callers must hold no locks.
1623 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1625 struct sem_undo_list *undo_list;
1627 undo_list = current->sysvsem.undo_list;
1629 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1630 if (undo_list == NULL)
1632 spin_lock_init(&undo_list->lock);
1633 atomic_set(&undo_list->refcnt, 1);
1634 INIT_LIST_HEAD(&undo_list->list_proc);
1636 current->sysvsem.undo_list = undo_list;
1638 *undo_listp = undo_list;
1642 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1644 struct sem_undo *un;
1646 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1647 if (un->semid == semid)
1653 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1655 struct sem_undo *un;
1657 assert_spin_locked(&ulp->lock);
1659 un = __lookup_undo(ulp, semid);
1661 list_del_rcu(&un->list_proc);
1662 list_add_rcu(&un->list_proc, &ulp->list_proc);
1668 * find_alloc_undo - lookup (and if not present create) undo array
1670 * @semid: semaphore array id
1672 * The function looks up (and if not present creates) the undo structure.
1673 * The size of the undo structure depends on the size of the semaphore
1674 * array, thus the alloc path is not that straightforward.
1675 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1676 * performs a rcu_read_lock().
1678 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1680 struct sem_array *sma;
1681 struct sem_undo_list *ulp;
1682 struct sem_undo *un, *new;
1685 error = get_undo_list(&ulp);
1687 return ERR_PTR(error);
1690 spin_lock(&ulp->lock);
1691 un = lookup_undo(ulp, semid);
1692 spin_unlock(&ulp->lock);
1693 if (likely(un != NULL))
1696 /* no undo structure around - allocate one. */
1697 /* step 1: figure out the size of the semaphore array */
1698 sma = sem_obtain_object_check(ns, semid);
1701 return ERR_CAST(sma);
1704 nsems = sma->sem_nsems;
1705 if (!ipc_rcu_getref(sma)) {
1707 un = ERR_PTR(-EIDRM);
1712 /* step 2: allocate new undo structure */
1713 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1715 ipc_rcu_putref(sma, ipc_rcu_free);
1716 return ERR_PTR(-ENOMEM);
1719 /* step 3: Acquire the lock on semaphore array */
1721 sem_lock_and_putref(sma);
1722 if (!ipc_valid_object(&sma->sem_perm)) {
1723 sem_unlock(sma, -1);
1726 un = ERR_PTR(-EIDRM);
1729 spin_lock(&ulp->lock);
1732 * step 4: check for races: did someone else allocate the undo struct?
1734 un = lookup_undo(ulp, semid);
1739 /* step 5: initialize & link new undo structure */
1740 new->semadj = (short *) &new[1];
1743 assert_spin_locked(&ulp->lock);
1744 list_add_rcu(&new->list_proc, &ulp->list_proc);
1745 ipc_assert_locked_object(&sma->sem_perm);
1746 list_add(&new->list_id, &sma->list_id);
1750 spin_unlock(&ulp->lock);
1751 sem_unlock(sma, -1);
1758 * get_queue_result - retrieve the result code from sem_queue
1759 * @q: Pointer to queue structure
1761 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1762 * q->status, then we must loop until the value is replaced with the final
1763 * value: This may happen if a task is woken up by an unrelated event (e.g.
1764 * signal) and in parallel the task is woken up by another task because it got
1765 * the requested semaphores.
1767 * The function can be called with or without holding the semaphore spinlock.
1769 static int get_queue_result(struct sem_queue *q)
1774 while (unlikely(error == IN_WAKEUP)) {
1782 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1783 unsigned, nsops, const struct timespec __user *, timeout)
1785 int error = -EINVAL;
1786 struct sem_array *sma;
1787 struct sembuf fast_sops[SEMOPM_FAST];
1788 struct sembuf *sops = fast_sops, *sop;
1789 struct sem_undo *un;
1790 int undos = 0, alter = 0, max, locknum;
1791 struct sem_queue queue;
1792 unsigned long jiffies_left = 0;
1793 struct ipc_namespace *ns;
1794 struct list_head tasks;
1796 ns = current->nsproxy->ipc_ns;
1798 if (nsops < 1 || semid < 0)
1800 if (nsops > ns->sc_semopm)
1802 if (nsops > SEMOPM_FAST) {
1803 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1807 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1812 struct timespec _timeout;
1813 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1817 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1818 _timeout.tv_nsec >= 1000000000L) {
1822 jiffies_left = timespec_to_jiffies(&_timeout);
1825 for (sop = sops; sop < sops + nsops; sop++) {
1826 if (sop->sem_num >= max)
1828 if (sop->sem_flg & SEM_UNDO)
1830 if (sop->sem_op != 0)
1834 INIT_LIST_HEAD(&tasks);
1837 /* On success, find_alloc_undo takes the rcu_read_lock */
1838 un = find_alloc_undo(ns, semid);
1840 error = PTR_ERR(un);
1848 sma = sem_obtain_object_check(ns, semid);
1851 error = PTR_ERR(sma);
1856 if (max >= sma->sem_nsems)
1857 goto out_rcu_wakeup;
1860 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1861 goto out_rcu_wakeup;
1863 error = security_sem_semop(sma, sops, nsops, alter);
1865 goto out_rcu_wakeup;
1868 locknum = sem_lock(sma, sops, nsops);
1870 * We eventually might perform the following check in a lockless
1871 * fashion, considering ipc_valid_object() locking constraints.
1872 * If nsops == 1 and there is no contention for sem_perm.lock, then
1873 * only a per-semaphore lock is held and it's OK to proceed with the
1874 * check below. More details on the fine grained locking scheme
1875 * entangled here and why it's RMID race safe on comments at sem_lock()
1877 if (!ipc_valid_object(&sma->sem_perm))
1878 goto out_unlock_free;
1880 * semid identifiers are not unique - find_alloc_undo may have
1881 * allocated an undo structure, it was invalidated by an RMID
1882 * and now a new array with received the same id. Check and fail.
1883 * This case can be detected checking un->semid. The existence of
1884 * "un" itself is guaranteed by rcu.
1886 if (un && un->semid == -1)
1887 goto out_unlock_free;
1890 queue.nsops = nsops;
1892 queue.pid = task_tgid_vnr(current);
1893 queue.alter = alter;
1895 error = perform_atomic_semop(sma, &queue);
1897 /* If the operation was successful, then do
1898 * the required updates.
1901 do_smart_update(sma, sops, nsops, 1, &tasks);
1903 set_semotime(sma, sops);
1906 goto out_unlock_free;
1908 /* We need to sleep on this operation, so we put the current
1909 * task into the pending queue and go to sleep.
1914 curr = &sma->sem_base[sops->sem_num];
1917 if (sma->complex_count) {
1918 list_add_tail(&queue.list,
1919 &sma->pending_alter);
1922 list_add_tail(&queue.list,
1923 &curr->pending_alter);
1926 list_add_tail(&queue.list, &curr->pending_const);
1929 if (!sma->complex_count)
1933 list_add_tail(&queue.list, &sma->pending_alter);
1935 list_add_tail(&queue.list, &sma->pending_const);
1937 sma->complex_count++;
1940 queue.status = -EINTR;
1941 queue.sleeper = current;
1944 __set_current_state(TASK_INTERRUPTIBLE);
1945 sem_unlock(sma, locknum);
1949 jiffies_left = schedule_timeout(jiffies_left);
1953 error = get_queue_result(&queue);
1955 if (error != -EINTR) {
1956 /* fast path: update_queue already obtained all requested
1958 * Perform a smp_mb(): User space could assume that semop()
1959 * is a memory barrier: Without the mb(), the cpu could
1960 * speculatively read in user space stale data that was
1961 * overwritten by the previous owner of the semaphore.
1969 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1972 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1974 error = get_queue_result(&queue);
1977 * Array removed? If yes, leave without sem_unlock().
1986 * If queue.status != -EINTR we are woken up by another process.
1987 * Leave without unlink_queue(), but with sem_unlock().
1989 if (error != -EINTR)
1990 goto out_unlock_free;
1993 * If an interrupt occurred we have to clean up the queue
1995 if (timeout && jiffies_left == 0)
1999 * If the wakeup was spurious, just retry
2001 if (error == -EINTR && !signal_pending(current))
2004 unlink_queue(sma, &queue);
2007 sem_unlock(sma, locknum);
2010 wake_up_sem_queue_do(&tasks);
2012 if (sops != fast_sops)
2017 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2020 return sys_semtimedop(semid, tsops, nsops, NULL);
2023 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2024 * parent and child tasks.
2027 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2029 struct sem_undo_list *undo_list;
2032 if (clone_flags & CLONE_SYSVSEM) {
2033 error = get_undo_list(&undo_list);
2036 atomic_inc(&undo_list->refcnt);
2037 tsk->sysvsem.undo_list = undo_list;
2039 tsk->sysvsem.undo_list = NULL;
2045 * add semadj values to semaphores, free undo structures.
2046 * undo structures are not freed when semaphore arrays are destroyed
2047 * so some of them may be out of date.
2048 * IMPLEMENTATION NOTE: There is some confusion over whether the
2049 * set of adjustments that needs to be done should be done in an atomic
2050 * manner or not. That is, if we are attempting to decrement the semval
2051 * should we queue up and wait until we can do so legally?
2052 * The original implementation attempted to do this (queue and wait).
2053 * The current implementation does not do so. The POSIX standard
2054 * and SVID should be consulted to determine what behavior is mandated.
2056 void exit_sem(struct task_struct *tsk)
2058 struct sem_undo_list *ulp;
2060 ulp = tsk->sysvsem.undo_list;
2063 tsk->sysvsem.undo_list = NULL;
2065 if (!atomic_dec_and_test(&ulp->refcnt))
2069 struct sem_array *sma;
2070 struct sem_undo *un;
2071 struct list_head tasks;
2075 un = list_entry_rcu(ulp->list_proc.next,
2076 struct sem_undo, list_proc);
2077 if (&un->list_proc == &ulp->list_proc)
2087 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
2088 /* exit_sem raced with IPC_RMID, nothing to do */
2094 sem_lock(sma, NULL, -1);
2095 /* exit_sem raced with IPC_RMID, nothing to do */
2096 if (!ipc_valid_object(&sma->sem_perm)) {
2097 sem_unlock(sma, -1);
2101 un = __lookup_undo(ulp, semid);
2103 /* exit_sem raced with IPC_RMID+semget() that created
2104 * exactly the same semid. Nothing to do.
2106 sem_unlock(sma, -1);
2111 /* remove un from the linked lists */
2112 ipc_assert_locked_object(&sma->sem_perm);
2113 list_del(&un->list_id);
2115 spin_lock(&ulp->lock);
2116 list_del_rcu(&un->list_proc);
2117 spin_unlock(&ulp->lock);
2119 /* perform adjustments registered in un */
2120 for (i = 0; i < sma->sem_nsems; i++) {
2121 struct sem *semaphore = &sma->sem_base[i];
2122 if (un->semadj[i]) {
2123 semaphore->semval += un->semadj[i];
2125 * Range checks of the new semaphore value,
2126 * not defined by sus:
2127 * - Some unices ignore the undo entirely
2128 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2129 * - some cap the value (e.g. FreeBSD caps
2130 * at 0, but doesn't enforce SEMVMX)
2132 * Linux caps the semaphore value, both at 0
2135 * Manfred <manfred@colorfullife.com>
2137 if (semaphore->semval < 0)
2138 semaphore->semval = 0;
2139 if (semaphore->semval > SEMVMX)
2140 semaphore->semval = SEMVMX;
2141 semaphore->sempid = task_tgid_vnr(current);
2144 /* maybe some queued-up processes were waiting for this */
2145 INIT_LIST_HEAD(&tasks);
2146 do_smart_update(sma, NULL, 0, 1, &tasks);
2147 sem_unlock(sma, -1);
2149 wake_up_sem_queue_do(&tasks);
2156 #ifdef CONFIG_PROC_FS
2157 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2159 struct user_namespace *user_ns = seq_user_ns(s);
2160 struct sem_array *sma = it;
2164 * The proc interface isn't aware of sem_lock(), it calls
2165 * ipc_lock_object() directly (in sysvipc_find_ipc).
2166 * In order to stay compatible with sem_lock(), we must wait until
2167 * all simple semop() calls have left their critical regions.
2169 sem_wait_array(sma);
2171 sem_otime = get_semotime(sma);
2174 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2179 from_kuid_munged(user_ns, sma->sem_perm.uid),
2180 from_kgid_munged(user_ns, sma->sem_perm.gid),
2181 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2182 from_kgid_munged(user_ns, sma->sem_perm.cgid),