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 */
158 * a) global sem_lock() for read/write
160 * sem_array.complex_count,
161 * sem_array.complex_mode
162 * sem_array.pending{_alter,_const},
165 * b) global or semaphore sem_lock() for read/write:
166 * sem_array.sem_base[i].pending_{const,alter}:
167 * sem_array.complex_mode (for read)
170 * sem_undo_list.list_proc:
171 * * undo_list->lock for write
175 #define sc_semmsl sem_ctls[0]
176 #define sc_semmns sem_ctls[1]
177 #define sc_semopm sem_ctls[2]
178 #define sc_semmni sem_ctls[3]
180 void sem_init_ns(struct ipc_namespace *ns)
182 ns->sc_semmsl = SEMMSL;
183 ns->sc_semmns = SEMMNS;
184 ns->sc_semopm = SEMOPM;
185 ns->sc_semmni = SEMMNI;
187 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
191 void sem_exit_ns(struct ipc_namespace *ns)
193 free_ipcs(ns, &sem_ids(ns), freeary);
194 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
198 void __init sem_init(void)
200 sem_init_ns(&init_ipc_ns);
201 ipc_init_proc_interface("sysvipc/sem",
202 " key semid perms nsems uid gid cuid cgid otime ctime\n",
203 IPC_SEM_IDS, sysvipc_sem_proc_show);
207 * unmerge_queues - unmerge queues, if possible.
208 * @sma: semaphore array
210 * The function unmerges the wait queues if complex_count is 0.
211 * It must be called prior to dropping the global semaphore array lock.
213 static void unmerge_queues(struct sem_array *sma)
215 struct sem_queue *q, *tq;
217 /* complex operations still around? */
218 if (sma->complex_count)
221 * We will switch back to simple mode.
222 * Move all pending operation back into the per-semaphore
225 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
227 curr = &sma->sem_base[q->sops[0].sem_num];
229 list_add_tail(&q->list, &curr->pending_alter);
231 INIT_LIST_HEAD(&sma->pending_alter);
235 * merge_queues - merge single semop queues into global queue
236 * @sma: semaphore array
238 * This function merges all per-semaphore queues into the global queue.
239 * It is necessary to achieve FIFO ordering for the pending single-sop
240 * operations when a multi-semop operation must sleep.
241 * Only the alter operations must be moved, the const operations can stay.
243 static void merge_queues(struct sem_array *sma)
246 for (i = 0; i < sma->sem_nsems; i++) {
247 struct sem *sem = sma->sem_base + i;
249 list_splice_init(&sem->pending_alter, &sma->pending_alter);
253 static void sem_rcu_free(struct rcu_head *head)
255 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
256 struct sem_array *sma = ipc_rcu_to_struct(p);
258 security_sem_free(sma);
263 * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they
264 * are only control barriers.
265 * The code must pair with spin_unlock(&sem->lock) or
266 * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient.
268 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
270 #define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb()
273 * Enter the mode suitable for non-simple operations:
274 * Caller must own sem_perm.lock.
276 static void complexmode_enter(struct sem_array *sma)
281 if (sma->complex_mode) {
282 /* We are already in complex_mode. Nothing to do */
285 WRITE_ONCE(sma->complex_mode, true);
287 /* We need a full barrier:
288 * The write to complex_mode must be visible
289 * before we read the first sem->lock spinlock state.
293 for (i = 0; i < sma->sem_nsems; i++) {
294 sem = sma->sem_base + i;
295 spin_unlock_wait(&sem->lock);
297 ipc_smp_acquire__after_spin_is_unlocked();
301 * Try to leave the mode that disallows simple operations:
302 * Caller must own sem_perm.lock.
304 static void complexmode_tryleave(struct sem_array *sma)
306 if (sma->complex_count) {
307 /* Complex ops are sleeping.
308 * We must stay in complex mode
313 * Immediately after setting complex_mode to false,
314 * a simple op can start. Thus: all memory writes
315 * performed by the current operation must be visible
316 * before we set complex_mode to false.
320 WRITE_ONCE(sma->complex_mode, false);
324 * If the request contains only one semaphore operation, and there are
325 * no complex transactions pending, lock only the semaphore involved.
326 * Otherwise, lock the entire semaphore array, since we either have
327 * multiple semaphores in our own semops, or we need to look at
328 * semaphores from other pending complex operations.
330 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
336 /* Complex operation - acquire a full lock */
337 ipc_lock_object(&sma->sem_perm);
339 /* Prevent parallel simple ops */
340 complexmode_enter(sma);
345 * Only one semaphore affected - try to optimize locking.
346 * Optimized locking is possible if no complex operation
347 * is either enqueued or processed right now.
349 * Both facts are tracked by complex_mode.
351 sem = sma->sem_base + sops->sem_num;
354 * Initial check for complex_mode. Just an optimization,
357 if (!READ_ONCE(sma->complex_mode)) {
359 * It appears that no complex operation is around.
360 * Acquire the per-semaphore lock.
362 spin_lock(&sem->lock);
364 /* Now repeat the test for complex_mode.
365 * A memory barrier is provided by the spin_lock()
368 if (!READ_ONCE(sma->complex_mode)) {
369 /* fast path successful! */
370 return sops->sem_num;
372 spin_unlock(&sem->lock);
375 /* slow path: acquire the full lock */
376 ipc_lock_object(&sma->sem_perm);
378 if (sma->complex_count == 0) {
380 * There is no complex operation, thus we can switch
381 * back to the fast path.
383 spin_lock(&sem->lock);
384 ipc_unlock_object(&sma->sem_perm);
385 return sops->sem_num;
387 /* Not a false alarm, thus complete the sequence for a
390 complexmode_enter(sma);
395 static inline void sem_unlock(struct sem_array *sma, int locknum)
399 complexmode_tryleave(sma);
400 ipc_unlock_object(&sma->sem_perm);
402 struct sem *sem = sma->sem_base + locknum;
403 spin_unlock(&sem->lock);
408 * sem_lock_(check_) routines are called in the paths where the rwsem
411 * The caller holds the RCU read lock.
413 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
414 int id, struct sembuf *sops, int nsops, int *locknum)
416 struct kern_ipc_perm *ipcp;
417 struct sem_array *sma;
419 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
421 return ERR_CAST(ipcp);
423 sma = container_of(ipcp, struct sem_array, sem_perm);
424 *locknum = sem_lock(sma, sops, nsops);
426 /* ipc_rmid() may have already freed the ID while sem_lock
427 * was spinning: verify that the structure is still valid
429 if (ipc_valid_object(ipcp))
430 return container_of(ipcp, struct sem_array, sem_perm);
432 sem_unlock(sma, *locknum);
433 return ERR_PTR(-EINVAL);
436 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
438 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
441 return ERR_CAST(ipcp);
443 return container_of(ipcp, struct sem_array, sem_perm);
446 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
449 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
452 return ERR_CAST(ipcp);
454 return container_of(ipcp, struct sem_array, sem_perm);
457 static inline void sem_lock_and_putref(struct sem_array *sma)
459 sem_lock(sma, NULL, -1);
460 ipc_rcu_putref(sma, ipc_rcu_free);
463 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
465 ipc_rmid(&sem_ids(ns), &s->sem_perm);
469 * Lockless wakeup algorithm:
470 * Without the check/retry algorithm a lockless wakeup is possible:
471 * - queue.status is initialized to -EINTR before blocking.
472 * - wakeup is performed by
473 * * unlinking the queue entry from the pending list
474 * * setting queue.status to IN_WAKEUP
475 * This is the notification for the blocked thread that a
476 * result value is imminent.
477 * * call wake_up_process
478 * * set queue.status to the final value.
479 * - the previously blocked thread checks queue.status:
480 * * if it's IN_WAKEUP, then it must wait until the value changes
481 * * if it's not -EINTR, then the operation was completed by
482 * update_queue. semtimedop can return queue.status without
483 * performing any operation on the sem array.
484 * * otherwise it must acquire the spinlock and check what's up.
486 * The two-stage algorithm is necessary to protect against the following
488 * - if queue.status is set after wake_up_process, then the woken up idle
489 * thread could race forward and try (and fail) to acquire sma->lock
490 * before update_queue had a chance to set queue.status
491 * - if queue.status is written before wake_up_process and if the
492 * blocked process is woken up by a signal between writing
493 * queue.status and the wake_up_process, then the woken up
494 * process could return from semtimedop and die by calling
495 * sys_exit before wake_up_process is called. Then wake_up_process
496 * will oops, because the task structure is already invalid.
497 * (yes, this happened on s390 with sysv msg).
503 * newary - Create a new semaphore set
505 * @params: ptr to the structure that contains key, semflg and nsems
507 * Called with sem_ids.rwsem held (as a writer)
509 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
513 struct sem_array *sma;
515 key_t key = params->key;
516 int nsems = params->u.nsems;
517 int semflg = params->flg;
522 if (ns->used_sems + nsems > ns->sc_semmns)
525 size = sizeof(*sma) + nsems * sizeof(struct sem);
526 sma = ipc_rcu_alloc(size);
530 memset(sma, 0, size);
532 sma->sem_perm.mode = (semflg & S_IRWXUGO);
533 sma->sem_perm.key = key;
535 sma->sem_perm.security = NULL;
536 retval = security_sem_alloc(sma);
538 ipc_rcu_putref(sma, ipc_rcu_free);
542 sma->sem_base = (struct sem *) &sma[1];
544 for (i = 0; i < nsems; i++) {
545 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
546 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
547 spin_lock_init(&sma->sem_base[i].lock);
550 sma->complex_count = 0;
551 sma->complex_mode = true; /* dropped by sem_unlock below */
552 INIT_LIST_HEAD(&sma->pending_alter);
553 INIT_LIST_HEAD(&sma->pending_const);
554 INIT_LIST_HEAD(&sma->list_id);
555 sma->sem_nsems = nsems;
556 sma->sem_ctime = get_seconds();
558 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
560 ipc_rcu_putref(sma, sem_rcu_free);
563 ns->used_sems += nsems;
568 return sma->sem_perm.id;
573 * Called with sem_ids.rwsem and ipcp locked.
575 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
577 struct sem_array *sma;
579 sma = container_of(ipcp, struct sem_array, sem_perm);
580 return security_sem_associate(sma, semflg);
584 * Called with sem_ids.rwsem and ipcp locked.
586 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
587 struct ipc_params *params)
589 struct sem_array *sma;
591 sma = container_of(ipcp, struct sem_array, sem_perm);
592 if (params->u.nsems > sma->sem_nsems)
598 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
600 struct ipc_namespace *ns;
601 static const struct ipc_ops sem_ops = {
603 .associate = sem_security,
604 .more_checks = sem_more_checks,
606 struct ipc_params sem_params;
608 ns = current->nsproxy->ipc_ns;
610 if (nsems < 0 || nsems > ns->sc_semmsl)
613 sem_params.key = key;
614 sem_params.flg = semflg;
615 sem_params.u.nsems = nsems;
617 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
621 * perform_atomic_semop - Perform (if possible) a semaphore operation
622 * @sma: semaphore array
623 * @q: struct sem_queue that describes the operation
625 * Returns 0 if the operation was possible.
626 * Returns 1 if the operation is impossible, the caller must sleep.
627 * Negative values are error codes.
629 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
631 int result, sem_op, nsops, pid;
641 for (sop = sops; sop < sops + nsops; sop++) {
642 curr = sma->sem_base + sop->sem_num;
643 sem_op = sop->sem_op;
644 result = curr->semval;
646 if (!sem_op && result)
655 if (sop->sem_flg & SEM_UNDO) {
656 int undo = un->semadj[sop->sem_num] - sem_op;
657 /* Exceeding the undo range is an error. */
658 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
660 un->semadj[sop->sem_num] = undo;
663 curr->semval = result;
668 while (sop >= sops) {
669 sma->sem_base[sop->sem_num].sempid = pid;
682 if (sop->sem_flg & IPC_NOWAIT)
689 while (sop >= sops) {
690 sem_op = sop->sem_op;
691 sma->sem_base[sop->sem_num].semval -= sem_op;
692 if (sop->sem_flg & SEM_UNDO)
693 un->semadj[sop->sem_num] += sem_op;
700 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
701 * @q: queue entry that must be signaled
702 * @error: Error value for the signal
704 * Prepare the wake-up of the queue entry q.
706 static void wake_up_sem_queue_prepare(struct list_head *pt,
707 struct sem_queue *q, int error)
709 if (list_empty(pt)) {
711 * Hold preempt off so that we don't get preempted and have the
712 * wakee busy-wait until we're scheduled back on.
716 q->status = IN_WAKEUP;
719 list_add_tail(&q->list, pt);
723 * wake_up_sem_queue_do - do the actual wake-up
724 * @pt: list of tasks to be woken up
726 * Do the actual wake-up.
727 * The function is called without any locks held, thus the semaphore array
728 * could be destroyed already and the tasks can disappear as soon as the
729 * status is set to the actual return code.
731 static void wake_up_sem_queue_do(struct list_head *pt)
733 struct sem_queue *q, *t;
736 did_something = !list_empty(pt);
737 list_for_each_entry_safe(q, t, pt, list) {
738 wake_up_process(q->sleeper);
739 /* q can disappear immediately after writing q->status. */
747 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
751 sma->complex_count--;
754 /** check_restart(sma, q)
755 * @sma: semaphore array
756 * @q: the operation that just completed
758 * update_queue is O(N^2) when it restarts scanning the whole queue of
759 * waiting operations. Therefore this function checks if the restart is
760 * really necessary. It is called after a previously waiting operation
761 * modified the array.
762 * Note that wait-for-zero operations are handled without restart.
764 static int check_restart(struct sem_array *sma, struct sem_queue *q)
766 /* pending complex alter operations are too difficult to analyse */
767 if (!list_empty(&sma->pending_alter))
770 /* we were a sleeping complex operation. Too difficult */
774 /* It is impossible that someone waits for the new value:
775 * - complex operations always restart.
776 * - wait-for-zero are handled seperately.
777 * - q is a previously sleeping simple operation that
778 * altered the array. It must be a decrement, because
779 * simple increments never sleep.
780 * - If there are older (higher priority) decrements
781 * in the queue, then they have observed the original
782 * semval value and couldn't proceed. The operation
783 * decremented to value - thus they won't proceed either.
789 * wake_const_ops - wake up non-alter tasks
790 * @sma: semaphore array.
791 * @semnum: semaphore that was modified.
792 * @pt: list head for the tasks that must be woken up.
794 * wake_const_ops must be called after a semaphore in a semaphore array
795 * was set to 0. If complex const operations are pending, wake_const_ops must
796 * be called with semnum = -1, as well as with the number of each modified
798 * The tasks that must be woken up are added to @pt. The return code
799 * is stored in q->pid.
800 * The function returns 1 if at least one operation was completed successfully.
802 static int wake_const_ops(struct sem_array *sma, int semnum,
803 struct list_head *pt)
806 struct list_head *walk;
807 struct list_head *pending_list;
808 int semop_completed = 0;
811 pending_list = &sma->pending_const;
813 pending_list = &sma->sem_base[semnum].pending_const;
815 walk = pending_list->next;
816 while (walk != pending_list) {
819 q = container_of(walk, struct sem_queue, list);
822 error = perform_atomic_semop(sma, q);
825 /* operation completed, remove from queue & wakeup */
827 unlink_queue(sma, q);
829 wake_up_sem_queue_prepare(pt, q, error);
834 return semop_completed;
838 * do_smart_wakeup_zero - wakeup all wait for zero tasks
839 * @sma: semaphore array
840 * @sops: operations that were performed
841 * @nsops: number of operations
842 * @pt: list head of the tasks that must be woken up.
844 * Checks all required queue for wait-for-zero operations, based
845 * on the actual changes that were performed on the semaphore array.
846 * The function returns 1 if at least one operation was completed successfully.
848 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
849 int nsops, struct list_head *pt)
852 int semop_completed = 0;
855 /* first: the per-semaphore queues, if known */
857 for (i = 0; i < nsops; i++) {
858 int num = sops[i].sem_num;
860 if (sma->sem_base[num].semval == 0) {
862 semop_completed |= wake_const_ops(sma, num, pt);
867 * No sops means modified semaphores not known.
868 * Assume all were changed.
870 for (i = 0; i < sma->sem_nsems; i++) {
871 if (sma->sem_base[i].semval == 0) {
873 semop_completed |= wake_const_ops(sma, i, pt);
878 * If one of the modified semaphores got 0,
879 * then check the global queue, too.
882 semop_completed |= wake_const_ops(sma, -1, pt);
884 return semop_completed;
889 * update_queue - look for tasks that can be completed.
890 * @sma: semaphore array.
891 * @semnum: semaphore that was modified.
892 * @pt: list head for the tasks that must be woken up.
894 * update_queue must be called after a semaphore in a semaphore array
895 * was modified. If multiple semaphores were modified, update_queue must
896 * be called with semnum = -1, as well as with the number of each modified
898 * The tasks that must be woken up are added to @pt. The return code
899 * is stored in q->pid.
900 * The function internally checks if const operations can now succeed.
902 * The function return 1 if at least one semop was completed successfully.
904 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
907 struct list_head *walk;
908 struct list_head *pending_list;
909 int semop_completed = 0;
912 pending_list = &sma->pending_alter;
914 pending_list = &sma->sem_base[semnum].pending_alter;
917 walk = pending_list->next;
918 while (walk != pending_list) {
921 q = container_of(walk, struct sem_queue, list);
924 /* If we are scanning the single sop, per-semaphore list of
925 * one semaphore and that semaphore is 0, then it is not
926 * necessary to scan further: simple increments
927 * that affect only one entry succeed immediately and cannot
928 * be in the per semaphore pending queue, and decrements
929 * cannot be successful if the value is already 0.
931 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
934 error = perform_atomic_semop(sma, q);
936 /* Does q->sleeper still need to sleep? */
940 unlink_queue(sma, q);
946 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
947 restart = check_restart(sma, q);
950 wake_up_sem_queue_prepare(pt, q, error);
954 return semop_completed;
958 * set_semotime - set sem_otime
959 * @sma: semaphore array
960 * @sops: operations that modified the array, may be NULL
962 * sem_otime is replicated to avoid cache line trashing.
963 * This function sets one instance to the current time.
965 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
968 sma->sem_base[0].sem_otime = get_seconds();
970 sma->sem_base[sops[0].sem_num].sem_otime =
976 * do_smart_update - optimized update_queue
977 * @sma: semaphore array
978 * @sops: operations that were performed
979 * @nsops: number of operations
980 * @otime: force setting otime
981 * @pt: list head of the tasks that must be woken up.
983 * do_smart_update() does the required calls to update_queue and wakeup_zero,
984 * based on the actual changes that were performed on the semaphore array.
985 * Note that the function does not do the actual wake-up: the caller is
986 * responsible for calling wake_up_sem_queue_do(@pt).
987 * It is safe to perform this call after dropping all locks.
989 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
990 int otime, struct list_head *pt)
994 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
996 if (!list_empty(&sma->pending_alter)) {
997 /* semaphore array uses the global queue - just process it. */
998 otime |= update_queue(sma, -1, pt);
1002 * No sops, thus the modified semaphores are not
1005 for (i = 0; i < sma->sem_nsems; i++)
1006 otime |= update_queue(sma, i, pt);
1009 * Check the semaphores that were increased:
1010 * - No complex ops, thus all sleeping ops are
1012 * - if we decreased the value, then any sleeping
1013 * semaphore ops wont be able to run: If the
1014 * previous value was too small, then the new
1015 * value will be too small, too.
1017 for (i = 0; i < nsops; i++) {
1018 if (sops[i].sem_op > 0) {
1019 otime |= update_queue(sma,
1020 sops[i].sem_num, pt);
1026 set_semotime(sma, sops);
1030 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1032 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1035 struct sembuf *sop = q->blocking;
1038 * Linux always (since 0.99.10) reported a task as sleeping on all
1039 * semaphores. This violates SUS, therefore it was changed to the
1040 * standard compliant behavior.
1041 * Give the administrators a chance to notice that an application
1042 * might misbehave because it relies on the Linux behavior.
1044 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1045 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1046 current->comm, task_pid_nr(current));
1048 if (sop->sem_num != semnum)
1051 if (count_zero && sop->sem_op == 0)
1053 if (!count_zero && sop->sem_op < 0)
1059 /* The following counts are associated to each semaphore:
1060 * semncnt number of tasks waiting on semval being nonzero
1061 * semzcnt number of tasks waiting on semval being zero
1063 * Per definition, a task waits only on the semaphore of the first semop
1064 * that cannot proceed, even if additional operation would block, too.
1066 static int count_semcnt(struct sem_array *sma, ushort semnum,
1069 struct list_head *l;
1070 struct sem_queue *q;
1074 /* First: check the simple operations. They are easy to evaluate */
1076 l = &sma->sem_base[semnum].pending_const;
1078 l = &sma->sem_base[semnum].pending_alter;
1080 list_for_each_entry(q, l, list) {
1081 /* all task on a per-semaphore list sleep on exactly
1087 /* Then: check the complex operations. */
1088 list_for_each_entry(q, &sma->pending_alter, list) {
1089 semcnt += check_qop(sma, semnum, q, count_zero);
1092 list_for_each_entry(q, &sma->pending_const, list) {
1093 semcnt += check_qop(sma, semnum, q, count_zero);
1099 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1100 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1101 * remains locked on exit.
1103 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1105 struct sem_undo *un, *tu;
1106 struct sem_queue *q, *tq;
1107 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1108 struct list_head tasks;
1111 /* Free the existing undo structures for this semaphore set. */
1112 ipc_assert_locked_object(&sma->sem_perm);
1113 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1114 list_del(&un->list_id);
1115 spin_lock(&un->ulp->lock);
1117 list_del_rcu(&un->list_proc);
1118 spin_unlock(&un->ulp->lock);
1122 /* Wake up all pending processes and let them fail with EIDRM. */
1123 INIT_LIST_HEAD(&tasks);
1124 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1125 unlink_queue(sma, q);
1126 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1129 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1130 unlink_queue(sma, q);
1131 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1133 for (i = 0; i < sma->sem_nsems; i++) {
1134 struct sem *sem = sma->sem_base + i;
1135 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1136 unlink_queue(sma, q);
1137 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1139 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1140 unlink_queue(sma, q);
1141 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1145 /* Remove the semaphore set from the IDR */
1147 sem_unlock(sma, -1);
1150 wake_up_sem_queue_do(&tasks);
1151 ns->used_sems -= sma->sem_nsems;
1152 ipc_rcu_putref(sma, sem_rcu_free);
1155 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1159 return copy_to_user(buf, in, sizeof(*in));
1162 struct semid_ds out;
1164 memset(&out, 0, sizeof(out));
1166 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1168 out.sem_otime = in->sem_otime;
1169 out.sem_ctime = in->sem_ctime;
1170 out.sem_nsems = in->sem_nsems;
1172 return copy_to_user(buf, &out, sizeof(out));
1179 static time_t get_semotime(struct sem_array *sma)
1184 res = sma->sem_base[0].sem_otime;
1185 for (i = 1; i < sma->sem_nsems; i++) {
1186 time_t to = sma->sem_base[i].sem_otime;
1194 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1195 int cmd, int version, void __user *p)
1198 struct sem_array *sma;
1204 struct seminfo seminfo;
1207 err = security_sem_semctl(NULL, cmd);
1211 memset(&seminfo, 0, sizeof(seminfo));
1212 seminfo.semmni = ns->sc_semmni;
1213 seminfo.semmns = ns->sc_semmns;
1214 seminfo.semmsl = ns->sc_semmsl;
1215 seminfo.semopm = ns->sc_semopm;
1216 seminfo.semvmx = SEMVMX;
1217 seminfo.semmnu = SEMMNU;
1218 seminfo.semmap = SEMMAP;
1219 seminfo.semume = SEMUME;
1220 down_read(&sem_ids(ns).rwsem);
1221 if (cmd == SEM_INFO) {
1222 seminfo.semusz = sem_ids(ns).in_use;
1223 seminfo.semaem = ns->used_sems;
1225 seminfo.semusz = SEMUSZ;
1226 seminfo.semaem = SEMAEM;
1228 max_id = ipc_get_maxid(&sem_ids(ns));
1229 up_read(&sem_ids(ns).rwsem);
1230 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1232 return (max_id < 0) ? 0 : max_id;
1237 struct semid64_ds tbuf;
1240 memset(&tbuf, 0, sizeof(tbuf));
1243 if (cmd == SEM_STAT) {
1244 sma = sem_obtain_object(ns, semid);
1249 id = sma->sem_perm.id;
1251 sma = sem_obtain_object_check(ns, semid);
1259 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1262 err = security_sem_semctl(sma, cmd);
1266 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1267 tbuf.sem_otime = get_semotime(sma);
1268 tbuf.sem_ctime = sma->sem_ctime;
1269 tbuf.sem_nsems = sma->sem_nsems;
1271 if (copy_semid_to_user(p, &tbuf, version))
1283 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1286 struct sem_undo *un;
1287 struct sem_array *sma;
1290 struct list_head tasks;
1292 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1293 /* big-endian 64bit */
1296 /* 32bit or little-endian 64bit */
1300 if (val > SEMVMX || val < 0)
1303 INIT_LIST_HEAD(&tasks);
1306 sma = sem_obtain_object_check(ns, semid);
1309 return PTR_ERR(sma);
1312 if (semnum < 0 || semnum >= sma->sem_nsems) {
1318 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1323 err = security_sem_semctl(sma, SETVAL);
1329 sem_lock(sma, NULL, -1);
1331 if (!ipc_valid_object(&sma->sem_perm)) {
1332 sem_unlock(sma, -1);
1337 curr = &sma->sem_base[semnum];
1339 ipc_assert_locked_object(&sma->sem_perm);
1340 list_for_each_entry(un, &sma->list_id, list_id)
1341 un->semadj[semnum] = 0;
1344 curr->sempid = task_tgid_vnr(current);
1345 sma->sem_ctime = get_seconds();
1346 /* maybe some queued-up processes were waiting for this */
1347 do_smart_update(sma, NULL, 0, 0, &tasks);
1348 sem_unlock(sma, -1);
1350 wake_up_sem_queue_do(&tasks);
1354 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1355 int cmd, void __user *p)
1357 struct sem_array *sma;
1360 ushort fast_sem_io[SEMMSL_FAST];
1361 ushort *sem_io = fast_sem_io;
1362 struct list_head tasks;
1364 INIT_LIST_HEAD(&tasks);
1367 sma = sem_obtain_object_check(ns, semid);
1370 return PTR_ERR(sma);
1373 nsems = sma->sem_nsems;
1376 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1377 goto out_rcu_wakeup;
1379 err = security_sem_semctl(sma, cmd);
1381 goto out_rcu_wakeup;
1387 ushort __user *array = p;
1390 sem_lock(sma, NULL, -1);
1391 if (!ipc_valid_object(&sma->sem_perm)) {
1395 if (nsems > SEMMSL_FAST) {
1396 if (!ipc_rcu_getref(sma)) {
1400 sem_unlock(sma, -1);
1402 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1403 if (sem_io == NULL) {
1404 ipc_rcu_putref(sma, ipc_rcu_free);
1409 sem_lock_and_putref(sma);
1410 if (!ipc_valid_object(&sma->sem_perm)) {
1415 for (i = 0; i < sma->sem_nsems; i++)
1416 sem_io[i] = sma->sem_base[i].semval;
1417 sem_unlock(sma, -1);
1420 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1427 struct sem_undo *un;
1429 if (!ipc_rcu_getref(sma)) {
1431 goto out_rcu_wakeup;
1435 if (nsems > SEMMSL_FAST) {
1436 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1437 if (sem_io == NULL) {
1438 ipc_rcu_putref(sma, ipc_rcu_free);
1443 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1444 ipc_rcu_putref(sma, ipc_rcu_free);
1449 for (i = 0; i < nsems; i++) {
1450 if (sem_io[i] > SEMVMX) {
1451 ipc_rcu_putref(sma, ipc_rcu_free);
1457 sem_lock_and_putref(sma);
1458 if (!ipc_valid_object(&sma->sem_perm)) {
1463 for (i = 0; i < nsems; i++)
1464 sma->sem_base[i].semval = sem_io[i];
1466 ipc_assert_locked_object(&sma->sem_perm);
1467 list_for_each_entry(un, &sma->list_id, list_id) {
1468 for (i = 0; i < nsems; i++)
1471 sma->sem_ctime = get_seconds();
1472 /* maybe some queued-up processes were waiting for this */
1473 do_smart_update(sma, NULL, 0, 0, &tasks);
1477 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1480 if (semnum < 0 || semnum >= nsems)
1481 goto out_rcu_wakeup;
1483 sem_lock(sma, NULL, -1);
1484 if (!ipc_valid_object(&sma->sem_perm)) {
1488 curr = &sma->sem_base[semnum];
1498 err = count_semcnt(sma, semnum, 0);
1501 err = count_semcnt(sma, semnum, 1);
1506 sem_unlock(sma, -1);
1509 wake_up_sem_queue_do(&tasks);
1511 if (sem_io != fast_sem_io)
1516 static inline unsigned long
1517 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1521 if (copy_from_user(out, buf, sizeof(*out)))
1526 struct semid_ds tbuf_old;
1528 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1531 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1532 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1533 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1543 * This function handles some semctl commands which require the rwsem
1544 * to be held in write mode.
1545 * NOTE: no locks must be held, the rwsem is taken inside this function.
1547 static int semctl_down(struct ipc_namespace *ns, int semid,
1548 int cmd, int version, void __user *p)
1550 struct sem_array *sma;
1552 struct semid64_ds semid64;
1553 struct kern_ipc_perm *ipcp;
1555 if (cmd == IPC_SET) {
1556 if (copy_semid_from_user(&semid64, p, version))
1560 down_write(&sem_ids(ns).rwsem);
1563 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1564 &semid64.sem_perm, 0);
1566 err = PTR_ERR(ipcp);
1570 sma = container_of(ipcp, struct sem_array, sem_perm);
1572 err = security_sem_semctl(sma, cmd);
1578 sem_lock(sma, NULL, -1);
1579 /* freeary unlocks the ipc object and rcu */
1583 sem_lock(sma, NULL, -1);
1584 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1587 sma->sem_ctime = get_seconds();
1595 sem_unlock(sma, -1);
1599 up_write(&sem_ids(ns).rwsem);
1603 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1606 struct ipc_namespace *ns;
1607 void __user *p = (void __user *)arg;
1612 version = ipc_parse_version(&cmd);
1613 ns = current->nsproxy->ipc_ns;
1620 return semctl_nolock(ns, semid, cmd, version, p);
1627 return semctl_main(ns, semid, semnum, cmd, p);
1629 return semctl_setval(ns, semid, semnum, arg);
1632 return semctl_down(ns, semid, cmd, version, p);
1638 /* If the task doesn't already have a undo_list, then allocate one
1639 * here. We guarantee there is only one thread using this undo list,
1640 * and current is THE ONE
1642 * If this allocation and assignment succeeds, but later
1643 * portions of this code fail, there is no need to free the sem_undo_list.
1644 * Just let it stay associated with the task, and it'll be freed later
1647 * This can block, so callers must hold no locks.
1649 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1651 struct sem_undo_list *undo_list;
1653 undo_list = current->sysvsem.undo_list;
1655 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1656 if (undo_list == NULL)
1658 spin_lock_init(&undo_list->lock);
1659 atomic_set(&undo_list->refcnt, 1);
1660 INIT_LIST_HEAD(&undo_list->list_proc);
1662 current->sysvsem.undo_list = undo_list;
1664 *undo_listp = undo_list;
1668 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1670 struct sem_undo *un;
1672 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1673 if (un->semid == semid)
1679 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1681 struct sem_undo *un;
1683 assert_spin_locked(&ulp->lock);
1685 un = __lookup_undo(ulp, semid);
1687 list_del_rcu(&un->list_proc);
1688 list_add_rcu(&un->list_proc, &ulp->list_proc);
1694 * find_alloc_undo - lookup (and if not present create) undo array
1696 * @semid: semaphore array id
1698 * The function looks up (and if not present creates) the undo structure.
1699 * The size of the undo structure depends on the size of the semaphore
1700 * array, thus the alloc path is not that straightforward.
1701 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1702 * performs a rcu_read_lock().
1704 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1706 struct sem_array *sma;
1707 struct sem_undo_list *ulp;
1708 struct sem_undo *un, *new;
1711 error = get_undo_list(&ulp);
1713 return ERR_PTR(error);
1716 spin_lock(&ulp->lock);
1717 un = lookup_undo(ulp, semid);
1718 spin_unlock(&ulp->lock);
1719 if (likely(un != NULL))
1722 /* no undo structure around - allocate one. */
1723 /* step 1: figure out the size of the semaphore array */
1724 sma = sem_obtain_object_check(ns, semid);
1727 return ERR_CAST(sma);
1730 nsems = sma->sem_nsems;
1731 if (!ipc_rcu_getref(sma)) {
1733 un = ERR_PTR(-EIDRM);
1738 /* step 2: allocate new undo structure */
1739 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1741 ipc_rcu_putref(sma, ipc_rcu_free);
1742 return ERR_PTR(-ENOMEM);
1745 /* step 3: Acquire the lock on semaphore array */
1747 sem_lock_and_putref(sma);
1748 if (!ipc_valid_object(&sma->sem_perm)) {
1749 sem_unlock(sma, -1);
1752 un = ERR_PTR(-EIDRM);
1755 spin_lock(&ulp->lock);
1758 * step 4: check for races: did someone else allocate the undo struct?
1760 un = lookup_undo(ulp, semid);
1765 /* step 5: initialize & link new undo structure */
1766 new->semadj = (short *) &new[1];
1769 assert_spin_locked(&ulp->lock);
1770 list_add_rcu(&new->list_proc, &ulp->list_proc);
1771 ipc_assert_locked_object(&sma->sem_perm);
1772 list_add(&new->list_id, &sma->list_id);
1776 spin_unlock(&ulp->lock);
1777 sem_unlock(sma, -1);
1784 * get_queue_result - retrieve the result code from sem_queue
1785 * @q: Pointer to queue structure
1787 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1788 * q->status, then we must loop until the value is replaced with the final
1789 * value: This may happen if a task is woken up by an unrelated event (e.g.
1790 * signal) and in parallel the task is woken up by another task because it got
1791 * the requested semaphores.
1793 * The function can be called with or without holding the semaphore spinlock.
1795 static int get_queue_result(struct sem_queue *q)
1800 while (unlikely(error == IN_WAKEUP)) {
1808 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1809 unsigned, nsops, const struct timespec __user *, timeout)
1811 int error = -EINVAL;
1812 struct sem_array *sma;
1813 struct sembuf fast_sops[SEMOPM_FAST];
1814 struct sembuf *sops = fast_sops, *sop;
1815 struct sem_undo *un;
1816 int undos = 0, alter = 0, max, locknum;
1817 struct sem_queue queue;
1818 unsigned long jiffies_left = 0;
1819 struct ipc_namespace *ns;
1820 struct list_head tasks;
1822 ns = current->nsproxy->ipc_ns;
1824 if (nsops < 1 || semid < 0)
1826 if (nsops > ns->sc_semopm)
1828 if (nsops > SEMOPM_FAST) {
1829 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1833 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1838 struct timespec _timeout;
1839 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1843 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1844 _timeout.tv_nsec >= 1000000000L) {
1848 jiffies_left = timespec_to_jiffies(&_timeout);
1851 for (sop = sops; sop < sops + nsops; sop++) {
1852 if (sop->sem_num >= max)
1854 if (sop->sem_flg & SEM_UNDO)
1856 if (sop->sem_op != 0)
1860 INIT_LIST_HEAD(&tasks);
1863 /* On success, find_alloc_undo takes the rcu_read_lock */
1864 un = find_alloc_undo(ns, semid);
1866 error = PTR_ERR(un);
1874 sma = sem_obtain_object_check(ns, semid);
1877 error = PTR_ERR(sma);
1882 if (max >= sma->sem_nsems)
1883 goto out_rcu_wakeup;
1886 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1887 goto out_rcu_wakeup;
1889 error = security_sem_semop(sma, sops, nsops, alter);
1891 goto out_rcu_wakeup;
1894 locknum = sem_lock(sma, sops, nsops);
1896 * We eventually might perform the following check in a lockless
1897 * fashion, considering ipc_valid_object() locking constraints.
1898 * If nsops == 1 and there is no contention for sem_perm.lock, then
1899 * only a per-semaphore lock is held and it's OK to proceed with the
1900 * check below. More details on the fine grained locking scheme
1901 * entangled here and why it's RMID race safe on comments at sem_lock()
1903 if (!ipc_valid_object(&sma->sem_perm))
1904 goto out_unlock_free;
1906 * semid identifiers are not unique - find_alloc_undo may have
1907 * allocated an undo structure, it was invalidated by an RMID
1908 * and now a new array with received the same id. Check and fail.
1909 * This case can be detected checking un->semid. The existence of
1910 * "un" itself is guaranteed by rcu.
1912 if (un && un->semid == -1)
1913 goto out_unlock_free;
1916 queue.nsops = nsops;
1918 queue.pid = task_tgid_vnr(current);
1919 queue.alter = alter;
1921 error = perform_atomic_semop(sma, &queue);
1923 /* If the operation was successful, then do
1924 * the required updates.
1927 do_smart_update(sma, sops, nsops, 1, &tasks);
1929 set_semotime(sma, sops);
1932 goto out_unlock_free;
1934 /* We need to sleep on this operation, so we put the current
1935 * task into the pending queue and go to sleep.
1940 curr = &sma->sem_base[sops->sem_num];
1943 if (sma->complex_count) {
1944 list_add_tail(&queue.list,
1945 &sma->pending_alter);
1948 list_add_tail(&queue.list,
1949 &curr->pending_alter);
1952 list_add_tail(&queue.list, &curr->pending_const);
1955 if (!sma->complex_count)
1959 list_add_tail(&queue.list, &sma->pending_alter);
1961 list_add_tail(&queue.list, &sma->pending_const);
1963 sma->complex_count++;
1966 queue.status = -EINTR;
1967 queue.sleeper = current;
1970 __set_current_state(TASK_INTERRUPTIBLE);
1971 sem_unlock(sma, locknum);
1975 jiffies_left = schedule_timeout(jiffies_left);
1979 error = get_queue_result(&queue);
1981 if (error != -EINTR) {
1982 /* fast path: update_queue already obtained all requested
1984 * Perform a smp_mb(): User space could assume that semop()
1985 * is a memory barrier: Without the mb(), the cpu could
1986 * speculatively read in user space stale data that was
1987 * overwritten by the previous owner of the semaphore.
1995 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1998 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
2000 error = get_queue_result(&queue);
2003 * Array removed? If yes, leave without sem_unlock().
2012 * If queue.status != -EINTR we are woken up by another process.
2013 * Leave without unlink_queue(), but with sem_unlock().
2015 if (error != -EINTR)
2016 goto out_unlock_free;
2019 * If an interrupt occurred we have to clean up the queue
2021 if (timeout && jiffies_left == 0)
2025 * If the wakeup was spurious, just retry
2027 if (error == -EINTR && !signal_pending(current))
2030 unlink_queue(sma, &queue);
2033 sem_unlock(sma, locknum);
2036 wake_up_sem_queue_do(&tasks);
2038 if (sops != fast_sops)
2043 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2046 return sys_semtimedop(semid, tsops, nsops, NULL);
2049 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2050 * parent and child tasks.
2053 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2055 struct sem_undo_list *undo_list;
2058 if (clone_flags & CLONE_SYSVSEM) {
2059 error = get_undo_list(&undo_list);
2062 atomic_inc(&undo_list->refcnt);
2063 tsk->sysvsem.undo_list = undo_list;
2065 tsk->sysvsem.undo_list = NULL;
2071 * add semadj values to semaphores, free undo structures.
2072 * undo structures are not freed when semaphore arrays are destroyed
2073 * so some of them may be out of date.
2074 * IMPLEMENTATION NOTE: There is some confusion over whether the
2075 * set of adjustments that needs to be done should be done in an atomic
2076 * manner or not. That is, if we are attempting to decrement the semval
2077 * should we queue up and wait until we can do so legally?
2078 * The original implementation attempted to do this (queue and wait).
2079 * The current implementation does not do so. The POSIX standard
2080 * and SVID should be consulted to determine what behavior is mandated.
2082 void exit_sem(struct task_struct *tsk)
2084 struct sem_undo_list *ulp;
2086 ulp = tsk->sysvsem.undo_list;
2089 tsk->sysvsem.undo_list = NULL;
2091 if (!atomic_dec_and_test(&ulp->refcnt))
2095 struct sem_array *sma;
2096 struct sem_undo *un;
2097 struct list_head tasks;
2101 un = list_entry_rcu(ulp->list_proc.next,
2102 struct sem_undo, list_proc);
2103 if (&un->list_proc == &ulp->list_proc) {
2105 * We must wait for freeary() before freeing this ulp,
2106 * in case we raced with last sem_undo. There is a small
2107 * possibility where we exit while freeary() didn't
2108 * finish unlocking sem_undo_list.
2110 spin_unlock_wait(&ulp->lock);
2114 spin_lock(&ulp->lock);
2116 spin_unlock(&ulp->lock);
2118 /* exit_sem raced with IPC_RMID, nothing to do */
2124 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2125 /* exit_sem raced with IPC_RMID, nothing to do */
2131 sem_lock(sma, NULL, -1);
2132 /* exit_sem raced with IPC_RMID, nothing to do */
2133 if (!ipc_valid_object(&sma->sem_perm)) {
2134 sem_unlock(sma, -1);
2138 un = __lookup_undo(ulp, semid);
2140 /* exit_sem raced with IPC_RMID+semget() that created
2141 * exactly the same semid. Nothing to do.
2143 sem_unlock(sma, -1);
2148 /* remove un from the linked lists */
2149 ipc_assert_locked_object(&sma->sem_perm);
2150 list_del(&un->list_id);
2152 /* we are the last process using this ulp, acquiring ulp->lock
2153 * isn't required. Besides that, we are also protected against
2154 * IPC_RMID as we hold sma->sem_perm lock now
2156 list_del_rcu(&un->list_proc);
2158 /* perform adjustments registered in un */
2159 for (i = 0; i < sma->sem_nsems; i++) {
2160 struct sem *semaphore = &sma->sem_base[i];
2161 if (un->semadj[i]) {
2162 semaphore->semval += un->semadj[i];
2164 * Range checks of the new semaphore value,
2165 * not defined by sus:
2166 * - Some unices ignore the undo entirely
2167 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2168 * - some cap the value (e.g. FreeBSD caps
2169 * at 0, but doesn't enforce SEMVMX)
2171 * Linux caps the semaphore value, both at 0
2174 * Manfred <manfred@colorfullife.com>
2176 if (semaphore->semval < 0)
2177 semaphore->semval = 0;
2178 if (semaphore->semval > SEMVMX)
2179 semaphore->semval = SEMVMX;
2180 semaphore->sempid = task_tgid_vnr(current);
2183 /* maybe some queued-up processes were waiting for this */
2184 INIT_LIST_HEAD(&tasks);
2185 do_smart_update(sma, NULL, 0, 1, &tasks);
2186 sem_unlock(sma, -1);
2188 wake_up_sem_queue_do(&tasks);
2195 #ifdef CONFIG_PROC_FS
2196 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2198 struct user_namespace *user_ns = seq_user_ns(s);
2199 struct sem_array *sma = it;
2203 * The proc interface isn't aware of sem_lock(), it calls
2204 * ipc_lock_object() directly (in sysvipc_find_ipc).
2205 * In order to stay compatible with sem_lock(), we must
2206 * enter / leave complex_mode.
2208 complexmode_enter(sma);
2210 sem_otime = get_semotime(sma);
2213 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2218 from_kuid_munged(user_ns, sma->sem_perm.uid),
2219 from_kgid_munged(user_ns, sma->sem_perm.gid),
2220 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2221 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2225 complexmode_tryleave(sma);