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 */
286 /* We need a full barrier after seting complex_mode:
287 * The write to complex_mode must be visible
288 * before we read the first sem->lock spinlock state.
290 smp_store_mb(sma->complex_mode, true);
292 for (i = 0; i < sma->sem_nsems; i++) {
293 sem = sma->sem_base + i;
294 spin_unlock_wait(&sem->lock);
296 ipc_smp_acquire__after_spin_is_unlocked();
300 * Try to leave the mode that disallows simple operations:
301 * Caller must own sem_perm.lock.
303 static void complexmode_tryleave(struct sem_array *sma)
305 if (sma->complex_count) {
306 /* Complex ops are sleeping.
307 * We must stay in complex mode
312 * Immediately after setting complex_mode to false,
313 * a simple op can start. Thus: all memory writes
314 * performed by the current operation must be visible
315 * before we set complex_mode to false.
317 smp_store_release(&sma->complex_mode, false);
320 #define SEM_GLOBAL_LOCK (-1)
322 * If the request contains only one semaphore operation, and there are
323 * no complex transactions pending, lock only the semaphore involved.
324 * Otherwise, lock the entire semaphore array, since we either have
325 * multiple semaphores in our own semops, or we need to look at
326 * semaphores from other pending complex operations.
328 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
334 /* Complex operation - acquire a full lock */
335 ipc_lock_object(&sma->sem_perm);
337 /* Prevent parallel simple ops */
338 complexmode_enter(sma);
339 return SEM_GLOBAL_LOCK;
343 * Only one semaphore affected - try to optimize locking.
344 * Optimized locking is possible if no complex operation
345 * is either enqueued or processed right now.
347 * Both facts are tracked by complex_mode.
349 sem = sma->sem_base + sops->sem_num;
352 * Initial check for complex_mode. Just an optimization,
353 * no locking, no memory barrier.
355 if (!sma->complex_mode) {
357 * It appears that no complex operation is around.
358 * Acquire the per-semaphore lock.
360 spin_lock(&sem->lock);
364 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
365 * A full barrier is required: the write of sem->lock
366 * must be visible before the read is executed
370 if (!smp_load_acquire(&sma->complex_mode)) {
371 /* fast path successful! */
372 return sops->sem_num;
374 spin_unlock(&sem->lock);
377 /* slow path: acquire the full lock */
378 ipc_lock_object(&sma->sem_perm);
380 if (sma->complex_count == 0) {
382 * There is no complex operation, thus we can switch
383 * back to the fast path.
385 spin_lock(&sem->lock);
386 ipc_unlock_object(&sma->sem_perm);
387 return sops->sem_num;
389 /* Not a false alarm, thus complete the sequence for a
392 complexmode_enter(sma);
393 return SEM_GLOBAL_LOCK;
397 static inline void sem_unlock(struct sem_array *sma, int locknum)
399 if (locknum == SEM_GLOBAL_LOCK) {
401 complexmode_tryleave(sma);
402 ipc_unlock_object(&sma->sem_perm);
404 struct sem *sem = sma->sem_base + locknum;
405 spin_unlock(&sem->lock);
410 * sem_lock_(check_) routines are called in the paths where the rwsem
413 * The caller holds the RCU read lock.
415 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
416 int id, struct sembuf *sops, int nsops, int *locknum)
418 struct kern_ipc_perm *ipcp;
419 struct sem_array *sma;
421 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
423 return ERR_CAST(ipcp);
425 sma = container_of(ipcp, struct sem_array, sem_perm);
426 *locknum = sem_lock(sma, sops, nsops);
428 /* ipc_rmid() may have already freed the ID while sem_lock
429 * was spinning: verify that the structure is still valid
431 if (ipc_valid_object(ipcp))
432 return container_of(ipcp, struct sem_array, sem_perm);
434 sem_unlock(sma, *locknum);
435 return ERR_PTR(-EINVAL);
438 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
440 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
443 return ERR_CAST(ipcp);
445 return container_of(ipcp, struct sem_array, sem_perm);
448 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
451 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
454 return ERR_CAST(ipcp);
456 return container_of(ipcp, struct sem_array, sem_perm);
459 static inline void sem_lock_and_putref(struct sem_array *sma)
461 sem_lock(sma, NULL, -1);
462 ipc_rcu_putref(sma, sem_rcu_free);
465 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
467 ipc_rmid(&sem_ids(ns), &s->sem_perm);
471 * Lockless wakeup algorithm:
472 * Without the check/retry algorithm a lockless wakeup is possible:
473 * - queue.status is initialized to -EINTR before blocking.
474 * - wakeup is performed by
475 * * unlinking the queue entry from the pending list
476 * * setting queue.status to IN_WAKEUP
477 * This is the notification for the blocked thread that a
478 * result value is imminent.
479 * * call wake_up_process
480 * * set queue.status to the final value.
481 * - the previously blocked thread checks queue.status:
482 * * if it's IN_WAKEUP, then it must wait until the value changes
483 * * if it's not -EINTR, then the operation was completed by
484 * update_queue. semtimedop can return queue.status without
485 * performing any operation on the sem array.
486 * * otherwise it must acquire the spinlock and check what's up.
488 * The two-stage algorithm is necessary to protect against the following
490 * - if queue.status is set after wake_up_process, then the woken up idle
491 * thread could race forward and try (and fail) to acquire sma->lock
492 * before update_queue had a chance to set queue.status
493 * - if queue.status is written before wake_up_process and if the
494 * blocked process is woken up by a signal between writing
495 * queue.status and the wake_up_process, then the woken up
496 * process could return from semtimedop and die by calling
497 * sys_exit before wake_up_process is called. Then wake_up_process
498 * will oops, because the task structure is already invalid.
499 * (yes, this happened on s390 with sysv msg).
505 * newary - Create a new semaphore set
507 * @params: ptr to the structure that contains key, semflg and nsems
509 * Called with sem_ids.rwsem held (as a writer)
511 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
515 struct sem_array *sma;
517 key_t key = params->key;
518 int nsems = params->u.nsems;
519 int semflg = params->flg;
524 if (ns->used_sems + nsems > ns->sc_semmns)
527 size = sizeof(*sma) + nsems * sizeof(struct sem);
528 sma = ipc_rcu_alloc(size);
532 memset(sma, 0, size);
534 sma->sem_perm.mode = (semflg & S_IRWXUGO);
535 sma->sem_perm.key = key;
537 sma->sem_perm.security = NULL;
538 retval = security_sem_alloc(sma);
540 ipc_rcu_putref(sma, ipc_rcu_free);
544 sma->sem_base = (struct sem *) &sma[1];
546 for (i = 0; i < nsems; i++) {
547 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
548 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
549 spin_lock_init(&sma->sem_base[i].lock);
552 sma->complex_count = 0;
553 sma->complex_mode = true; /* dropped by sem_unlock below */
554 INIT_LIST_HEAD(&sma->pending_alter);
555 INIT_LIST_HEAD(&sma->pending_const);
556 INIT_LIST_HEAD(&sma->list_id);
557 sma->sem_nsems = nsems;
558 sma->sem_ctime = get_seconds();
560 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
562 ipc_rcu_putref(sma, sem_rcu_free);
565 ns->used_sems += nsems;
570 return sma->sem_perm.id;
575 * Called with sem_ids.rwsem and ipcp locked.
577 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
579 struct sem_array *sma;
581 sma = container_of(ipcp, struct sem_array, sem_perm);
582 return security_sem_associate(sma, semflg);
586 * Called with sem_ids.rwsem and ipcp locked.
588 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
589 struct ipc_params *params)
591 struct sem_array *sma;
593 sma = container_of(ipcp, struct sem_array, sem_perm);
594 if (params->u.nsems > sma->sem_nsems)
600 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
602 struct ipc_namespace *ns;
603 static const struct ipc_ops sem_ops = {
605 .associate = sem_security,
606 .more_checks = sem_more_checks,
608 struct ipc_params sem_params;
610 ns = current->nsproxy->ipc_ns;
612 if (nsems < 0 || nsems > ns->sc_semmsl)
615 sem_params.key = key;
616 sem_params.flg = semflg;
617 sem_params.u.nsems = nsems;
619 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
623 * perform_atomic_semop - Perform (if possible) a semaphore operation
624 * @sma: semaphore array
625 * @q: struct sem_queue that describes the operation
627 * Returns 0 if the operation was possible.
628 * Returns 1 if the operation is impossible, the caller must sleep.
629 * Negative values are error codes.
631 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
633 int result, sem_op, nsops, pid;
643 for (sop = sops; sop < sops + nsops; sop++) {
644 curr = sma->sem_base + sop->sem_num;
645 sem_op = sop->sem_op;
646 result = curr->semval;
648 if (!sem_op && result)
657 if (sop->sem_flg & SEM_UNDO) {
658 int undo = un->semadj[sop->sem_num] - sem_op;
659 /* Exceeding the undo range is an error. */
660 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
662 un->semadj[sop->sem_num] = undo;
665 curr->semval = result;
670 while (sop >= sops) {
671 sma->sem_base[sop->sem_num].sempid = pid;
684 if (sop->sem_flg & IPC_NOWAIT)
691 while (sop >= sops) {
692 sem_op = sop->sem_op;
693 sma->sem_base[sop->sem_num].semval -= sem_op;
694 if (sop->sem_flg & SEM_UNDO)
695 un->semadj[sop->sem_num] += sem_op;
702 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
703 * @q: queue entry that must be signaled
704 * @error: Error value for the signal
706 * Prepare the wake-up of the queue entry q.
708 static void wake_up_sem_queue_prepare(struct list_head *pt,
709 struct sem_queue *q, int error)
711 #ifdef CONFIG_PREEMPT_RT_BASE
712 struct task_struct *p = q->sleeper;
718 if (list_empty(pt)) {
720 * Hold preempt off so that we don't get preempted and have the
721 * wakee busy-wait until we're scheduled back on.
725 q->status = IN_WAKEUP;
728 list_add_tail(&q->list, pt);
733 * wake_up_sem_queue_do - do the actual wake-up
734 * @pt: list of tasks to be woken up
736 * Do the actual wake-up.
737 * The function is called without any locks held, thus the semaphore array
738 * could be destroyed already and the tasks can disappear as soon as the
739 * status is set to the actual return code.
741 static void wake_up_sem_queue_do(struct list_head *pt)
743 #ifndef CONFIG_PREEMPT_RT_BASE
744 struct sem_queue *q, *t;
747 did_something = !list_empty(pt);
748 list_for_each_entry_safe(q, t, pt, list) {
749 wake_up_process(q->sleeper);
750 /* q can disappear immediately after writing q->status. */
759 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
763 sma->complex_count--;
766 /** check_restart(sma, q)
767 * @sma: semaphore array
768 * @q: the operation that just completed
770 * update_queue is O(N^2) when it restarts scanning the whole queue of
771 * waiting operations. Therefore this function checks if the restart is
772 * really necessary. It is called after a previously waiting operation
773 * modified the array.
774 * Note that wait-for-zero operations are handled without restart.
776 static int check_restart(struct sem_array *sma, struct sem_queue *q)
778 /* pending complex alter operations are too difficult to analyse */
779 if (!list_empty(&sma->pending_alter))
782 /* we were a sleeping complex operation. Too difficult */
786 /* It is impossible that someone waits for the new value:
787 * - complex operations always restart.
788 * - wait-for-zero are handled seperately.
789 * - q is a previously sleeping simple operation that
790 * altered the array. It must be a decrement, because
791 * simple increments never sleep.
792 * - If there are older (higher priority) decrements
793 * in the queue, then they have observed the original
794 * semval value and couldn't proceed. The operation
795 * decremented to value - thus they won't proceed either.
801 * wake_const_ops - wake up non-alter tasks
802 * @sma: semaphore array.
803 * @semnum: semaphore that was modified.
804 * @pt: list head for the tasks that must be woken up.
806 * wake_const_ops must be called after a semaphore in a semaphore array
807 * was set to 0. If complex const operations are pending, wake_const_ops must
808 * be called with semnum = -1, as well as with the number of each modified
810 * The tasks that must be woken up are added to @pt. The return code
811 * is stored in q->pid.
812 * The function returns 1 if at least one operation was completed successfully.
814 static int wake_const_ops(struct sem_array *sma, int semnum,
815 struct list_head *pt)
818 struct list_head *walk;
819 struct list_head *pending_list;
820 int semop_completed = 0;
823 pending_list = &sma->pending_const;
825 pending_list = &sma->sem_base[semnum].pending_const;
827 walk = pending_list->next;
828 while (walk != pending_list) {
831 q = container_of(walk, struct sem_queue, list);
834 error = perform_atomic_semop(sma, q);
837 /* operation completed, remove from queue & wakeup */
839 unlink_queue(sma, q);
841 wake_up_sem_queue_prepare(pt, q, error);
846 return semop_completed;
850 * do_smart_wakeup_zero - wakeup all wait for zero tasks
851 * @sma: semaphore array
852 * @sops: operations that were performed
853 * @nsops: number of operations
854 * @pt: list head of the tasks that must be woken up.
856 * Checks all required queue for wait-for-zero operations, based
857 * on the actual changes that were performed on the semaphore array.
858 * The function returns 1 if at least one operation was completed successfully.
860 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
861 int nsops, struct list_head *pt)
864 int semop_completed = 0;
867 /* first: the per-semaphore queues, if known */
869 for (i = 0; i < nsops; i++) {
870 int num = sops[i].sem_num;
872 if (sma->sem_base[num].semval == 0) {
874 semop_completed |= wake_const_ops(sma, num, pt);
879 * No sops means modified semaphores not known.
880 * Assume all were changed.
882 for (i = 0; i < sma->sem_nsems; i++) {
883 if (sma->sem_base[i].semval == 0) {
885 semop_completed |= wake_const_ops(sma, i, pt);
890 * If one of the modified semaphores got 0,
891 * then check the global queue, too.
894 semop_completed |= wake_const_ops(sma, -1, pt);
896 return semop_completed;
901 * update_queue - look for tasks that can be completed.
902 * @sma: semaphore array.
903 * @semnum: semaphore that was modified.
904 * @pt: list head for the tasks that must be woken up.
906 * update_queue must be called after a semaphore in a semaphore array
907 * was modified. If multiple semaphores were modified, update_queue must
908 * be called with semnum = -1, as well as with the number of each modified
910 * The tasks that must be woken up are added to @pt. The return code
911 * is stored in q->pid.
912 * The function internally checks if const operations can now succeed.
914 * The function return 1 if at least one semop was completed successfully.
916 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
919 struct list_head *walk;
920 struct list_head *pending_list;
921 int semop_completed = 0;
924 pending_list = &sma->pending_alter;
926 pending_list = &sma->sem_base[semnum].pending_alter;
929 walk = pending_list->next;
930 while (walk != pending_list) {
933 q = container_of(walk, struct sem_queue, list);
936 /* If we are scanning the single sop, per-semaphore list of
937 * one semaphore and that semaphore is 0, then it is not
938 * necessary to scan further: simple increments
939 * that affect only one entry succeed immediately and cannot
940 * be in the per semaphore pending queue, and decrements
941 * cannot be successful if the value is already 0.
943 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
946 error = perform_atomic_semop(sma, q);
948 /* Does q->sleeper still need to sleep? */
952 unlink_queue(sma, q);
958 do_smart_wakeup_zero(sma, q->sops, q->nsops, pt);
959 restart = check_restart(sma, q);
962 wake_up_sem_queue_prepare(pt, q, error);
966 return semop_completed;
970 * set_semotime - set sem_otime
971 * @sma: semaphore array
972 * @sops: operations that modified the array, may be NULL
974 * sem_otime is replicated to avoid cache line trashing.
975 * This function sets one instance to the current time.
977 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
980 sma->sem_base[0].sem_otime = get_seconds();
982 sma->sem_base[sops[0].sem_num].sem_otime =
988 * do_smart_update - optimized update_queue
989 * @sma: semaphore array
990 * @sops: operations that were performed
991 * @nsops: number of operations
992 * @otime: force setting otime
993 * @pt: list head of the tasks that must be woken up.
995 * do_smart_update() does the required calls to update_queue and wakeup_zero,
996 * based on the actual changes that were performed on the semaphore array.
997 * Note that the function does not do the actual wake-up: the caller is
998 * responsible for calling wake_up_sem_queue_do(@pt).
999 * It is safe to perform this call after dropping all locks.
1001 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
1002 int otime, struct list_head *pt)
1006 otime |= do_smart_wakeup_zero(sma, sops, nsops, pt);
1008 if (!list_empty(&sma->pending_alter)) {
1009 /* semaphore array uses the global queue - just process it. */
1010 otime |= update_queue(sma, -1, pt);
1014 * No sops, thus the modified semaphores are not
1017 for (i = 0; i < sma->sem_nsems; i++)
1018 otime |= update_queue(sma, i, pt);
1021 * Check the semaphores that were increased:
1022 * - No complex ops, thus all sleeping ops are
1024 * - if we decreased the value, then any sleeping
1025 * semaphore ops wont be able to run: If the
1026 * previous value was too small, then the new
1027 * value will be too small, too.
1029 for (i = 0; i < nsops; i++) {
1030 if (sops[i].sem_op > 0) {
1031 otime |= update_queue(sma,
1032 sops[i].sem_num, pt);
1038 set_semotime(sma, sops);
1042 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1044 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1047 struct sembuf *sop = q->blocking;
1050 * Linux always (since 0.99.10) reported a task as sleeping on all
1051 * semaphores. This violates SUS, therefore it was changed to the
1052 * standard compliant behavior.
1053 * Give the administrators a chance to notice that an application
1054 * might misbehave because it relies on the Linux behavior.
1056 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1057 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1058 current->comm, task_pid_nr(current));
1060 if (sop->sem_num != semnum)
1063 if (count_zero && sop->sem_op == 0)
1065 if (!count_zero && sop->sem_op < 0)
1071 /* The following counts are associated to each semaphore:
1072 * semncnt number of tasks waiting on semval being nonzero
1073 * semzcnt number of tasks waiting on semval being zero
1075 * Per definition, a task waits only on the semaphore of the first semop
1076 * that cannot proceed, even if additional operation would block, too.
1078 static int count_semcnt(struct sem_array *sma, ushort semnum,
1081 struct list_head *l;
1082 struct sem_queue *q;
1086 /* First: check the simple operations. They are easy to evaluate */
1088 l = &sma->sem_base[semnum].pending_const;
1090 l = &sma->sem_base[semnum].pending_alter;
1092 list_for_each_entry(q, l, list) {
1093 /* all task on a per-semaphore list sleep on exactly
1099 /* Then: check the complex operations. */
1100 list_for_each_entry(q, &sma->pending_alter, list) {
1101 semcnt += check_qop(sma, semnum, q, count_zero);
1104 list_for_each_entry(q, &sma->pending_const, list) {
1105 semcnt += check_qop(sma, semnum, q, count_zero);
1111 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1112 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1113 * remains locked on exit.
1115 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1117 struct sem_undo *un, *tu;
1118 struct sem_queue *q, *tq;
1119 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1120 struct list_head tasks;
1123 /* Free the existing undo structures for this semaphore set. */
1124 ipc_assert_locked_object(&sma->sem_perm);
1125 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1126 list_del(&un->list_id);
1127 spin_lock(&un->ulp->lock);
1129 list_del_rcu(&un->list_proc);
1130 spin_unlock(&un->ulp->lock);
1134 /* Wake up all pending processes and let them fail with EIDRM. */
1135 INIT_LIST_HEAD(&tasks);
1136 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1137 unlink_queue(sma, q);
1138 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1141 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1142 unlink_queue(sma, q);
1143 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1145 for (i = 0; i < sma->sem_nsems; i++) {
1146 struct sem *sem = sma->sem_base + i;
1147 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1148 unlink_queue(sma, q);
1149 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1151 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1152 unlink_queue(sma, q);
1153 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1157 /* Remove the semaphore set from the IDR */
1159 sem_unlock(sma, -1);
1162 wake_up_sem_queue_do(&tasks);
1163 ns->used_sems -= sma->sem_nsems;
1164 ipc_rcu_putref(sma, sem_rcu_free);
1167 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1171 return copy_to_user(buf, in, sizeof(*in));
1174 struct semid_ds out;
1176 memset(&out, 0, sizeof(out));
1178 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1180 out.sem_otime = in->sem_otime;
1181 out.sem_ctime = in->sem_ctime;
1182 out.sem_nsems = in->sem_nsems;
1184 return copy_to_user(buf, &out, sizeof(out));
1191 static time_t get_semotime(struct sem_array *sma)
1196 res = sma->sem_base[0].sem_otime;
1197 for (i = 1; i < sma->sem_nsems; i++) {
1198 time_t to = sma->sem_base[i].sem_otime;
1206 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1207 int cmd, int version, void __user *p)
1210 struct sem_array *sma;
1216 struct seminfo seminfo;
1219 err = security_sem_semctl(NULL, cmd);
1223 memset(&seminfo, 0, sizeof(seminfo));
1224 seminfo.semmni = ns->sc_semmni;
1225 seminfo.semmns = ns->sc_semmns;
1226 seminfo.semmsl = ns->sc_semmsl;
1227 seminfo.semopm = ns->sc_semopm;
1228 seminfo.semvmx = SEMVMX;
1229 seminfo.semmnu = SEMMNU;
1230 seminfo.semmap = SEMMAP;
1231 seminfo.semume = SEMUME;
1232 down_read(&sem_ids(ns).rwsem);
1233 if (cmd == SEM_INFO) {
1234 seminfo.semusz = sem_ids(ns).in_use;
1235 seminfo.semaem = ns->used_sems;
1237 seminfo.semusz = SEMUSZ;
1238 seminfo.semaem = SEMAEM;
1240 max_id = ipc_get_maxid(&sem_ids(ns));
1241 up_read(&sem_ids(ns).rwsem);
1242 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1244 return (max_id < 0) ? 0 : max_id;
1249 struct semid64_ds tbuf;
1252 memset(&tbuf, 0, sizeof(tbuf));
1255 if (cmd == SEM_STAT) {
1256 sma = sem_obtain_object(ns, semid);
1261 id = sma->sem_perm.id;
1263 sma = sem_obtain_object_check(ns, semid);
1271 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1274 err = security_sem_semctl(sma, cmd);
1278 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1279 tbuf.sem_otime = get_semotime(sma);
1280 tbuf.sem_ctime = sma->sem_ctime;
1281 tbuf.sem_nsems = sma->sem_nsems;
1283 if (copy_semid_to_user(p, &tbuf, version))
1295 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1298 struct sem_undo *un;
1299 struct sem_array *sma;
1302 struct list_head tasks;
1304 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1305 /* big-endian 64bit */
1308 /* 32bit or little-endian 64bit */
1312 if (val > SEMVMX || val < 0)
1315 INIT_LIST_HEAD(&tasks);
1318 sma = sem_obtain_object_check(ns, semid);
1321 return PTR_ERR(sma);
1324 if (semnum < 0 || semnum >= sma->sem_nsems) {
1330 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1335 err = security_sem_semctl(sma, SETVAL);
1341 sem_lock(sma, NULL, -1);
1343 if (!ipc_valid_object(&sma->sem_perm)) {
1344 sem_unlock(sma, -1);
1349 curr = &sma->sem_base[semnum];
1351 ipc_assert_locked_object(&sma->sem_perm);
1352 list_for_each_entry(un, &sma->list_id, list_id)
1353 un->semadj[semnum] = 0;
1356 curr->sempid = task_tgid_vnr(current);
1357 sma->sem_ctime = get_seconds();
1358 /* maybe some queued-up processes were waiting for this */
1359 do_smart_update(sma, NULL, 0, 0, &tasks);
1360 sem_unlock(sma, -1);
1362 wake_up_sem_queue_do(&tasks);
1366 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1367 int cmd, void __user *p)
1369 struct sem_array *sma;
1372 ushort fast_sem_io[SEMMSL_FAST];
1373 ushort *sem_io = fast_sem_io;
1374 struct list_head tasks;
1376 INIT_LIST_HEAD(&tasks);
1379 sma = sem_obtain_object_check(ns, semid);
1382 return PTR_ERR(sma);
1385 nsems = sma->sem_nsems;
1388 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1389 goto out_rcu_wakeup;
1391 err = security_sem_semctl(sma, cmd);
1393 goto out_rcu_wakeup;
1399 ushort __user *array = p;
1402 sem_lock(sma, NULL, -1);
1403 if (!ipc_valid_object(&sma->sem_perm)) {
1407 if (nsems > SEMMSL_FAST) {
1408 if (!ipc_rcu_getref(sma)) {
1412 sem_unlock(sma, -1);
1414 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1415 if (sem_io == NULL) {
1416 ipc_rcu_putref(sma, sem_rcu_free);
1421 sem_lock_and_putref(sma);
1422 if (!ipc_valid_object(&sma->sem_perm)) {
1427 for (i = 0; i < sma->sem_nsems; i++)
1428 sem_io[i] = sma->sem_base[i].semval;
1429 sem_unlock(sma, -1);
1432 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1439 struct sem_undo *un;
1441 if (!ipc_rcu_getref(sma)) {
1443 goto out_rcu_wakeup;
1447 if (nsems > SEMMSL_FAST) {
1448 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1449 if (sem_io == NULL) {
1450 ipc_rcu_putref(sma, sem_rcu_free);
1455 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1456 ipc_rcu_putref(sma, sem_rcu_free);
1461 for (i = 0; i < nsems; i++) {
1462 if (sem_io[i] > SEMVMX) {
1463 ipc_rcu_putref(sma, sem_rcu_free);
1469 sem_lock_and_putref(sma);
1470 if (!ipc_valid_object(&sma->sem_perm)) {
1475 for (i = 0; i < nsems; i++)
1476 sma->sem_base[i].semval = sem_io[i];
1478 ipc_assert_locked_object(&sma->sem_perm);
1479 list_for_each_entry(un, &sma->list_id, list_id) {
1480 for (i = 0; i < nsems; i++)
1483 sma->sem_ctime = get_seconds();
1484 /* maybe some queued-up processes were waiting for this */
1485 do_smart_update(sma, NULL, 0, 0, &tasks);
1489 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1492 if (semnum < 0 || semnum >= nsems)
1493 goto out_rcu_wakeup;
1495 sem_lock(sma, NULL, -1);
1496 if (!ipc_valid_object(&sma->sem_perm)) {
1500 curr = &sma->sem_base[semnum];
1510 err = count_semcnt(sma, semnum, 0);
1513 err = count_semcnt(sma, semnum, 1);
1518 sem_unlock(sma, -1);
1521 wake_up_sem_queue_do(&tasks);
1523 if (sem_io != fast_sem_io)
1524 ipc_free(sem_io, sizeof(ushort)*nsems);
1528 static inline unsigned long
1529 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1533 if (copy_from_user(out, buf, sizeof(*out)))
1538 struct semid_ds tbuf_old;
1540 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1543 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1544 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1545 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1555 * This function handles some semctl commands which require the rwsem
1556 * to be held in write mode.
1557 * NOTE: no locks must be held, the rwsem is taken inside this function.
1559 static int semctl_down(struct ipc_namespace *ns, int semid,
1560 int cmd, int version, void __user *p)
1562 struct sem_array *sma;
1564 struct semid64_ds semid64;
1565 struct kern_ipc_perm *ipcp;
1567 if (cmd == IPC_SET) {
1568 if (copy_semid_from_user(&semid64, p, version))
1572 down_write(&sem_ids(ns).rwsem);
1575 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1576 &semid64.sem_perm, 0);
1578 err = PTR_ERR(ipcp);
1582 sma = container_of(ipcp, struct sem_array, sem_perm);
1584 err = security_sem_semctl(sma, cmd);
1590 sem_lock(sma, NULL, -1);
1591 /* freeary unlocks the ipc object and rcu */
1595 sem_lock(sma, NULL, -1);
1596 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1599 sma->sem_ctime = get_seconds();
1607 sem_unlock(sma, -1);
1611 up_write(&sem_ids(ns).rwsem);
1615 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1618 struct ipc_namespace *ns;
1619 void __user *p = (void __user *)arg;
1624 version = ipc_parse_version(&cmd);
1625 ns = current->nsproxy->ipc_ns;
1632 return semctl_nolock(ns, semid, cmd, version, p);
1639 return semctl_main(ns, semid, semnum, cmd, p);
1641 return semctl_setval(ns, semid, semnum, arg);
1644 return semctl_down(ns, semid, cmd, version, p);
1650 /* If the task doesn't already have a undo_list, then allocate one
1651 * here. We guarantee there is only one thread using this undo list,
1652 * and current is THE ONE
1654 * If this allocation and assignment succeeds, but later
1655 * portions of this code fail, there is no need to free the sem_undo_list.
1656 * Just let it stay associated with the task, and it'll be freed later
1659 * This can block, so callers must hold no locks.
1661 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1663 struct sem_undo_list *undo_list;
1665 undo_list = current->sysvsem.undo_list;
1667 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1668 if (undo_list == NULL)
1670 spin_lock_init(&undo_list->lock);
1671 atomic_set(&undo_list->refcnt, 1);
1672 INIT_LIST_HEAD(&undo_list->list_proc);
1674 current->sysvsem.undo_list = undo_list;
1676 *undo_listp = undo_list;
1680 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1682 struct sem_undo *un;
1684 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1685 if (un->semid == semid)
1691 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1693 struct sem_undo *un;
1695 assert_spin_locked(&ulp->lock);
1697 un = __lookup_undo(ulp, semid);
1699 list_del_rcu(&un->list_proc);
1700 list_add_rcu(&un->list_proc, &ulp->list_proc);
1706 * find_alloc_undo - lookup (and if not present create) undo array
1708 * @semid: semaphore array id
1710 * The function looks up (and if not present creates) the undo structure.
1711 * The size of the undo structure depends on the size of the semaphore
1712 * array, thus the alloc path is not that straightforward.
1713 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1714 * performs a rcu_read_lock().
1716 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1718 struct sem_array *sma;
1719 struct sem_undo_list *ulp;
1720 struct sem_undo *un, *new;
1723 error = get_undo_list(&ulp);
1725 return ERR_PTR(error);
1728 spin_lock(&ulp->lock);
1729 un = lookup_undo(ulp, semid);
1730 spin_unlock(&ulp->lock);
1731 if (likely(un != NULL))
1734 /* no undo structure around - allocate one. */
1735 /* step 1: figure out the size of the semaphore array */
1736 sma = sem_obtain_object_check(ns, semid);
1739 return ERR_CAST(sma);
1742 nsems = sma->sem_nsems;
1743 if (!ipc_rcu_getref(sma)) {
1745 un = ERR_PTR(-EIDRM);
1750 /* step 2: allocate new undo structure */
1751 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1753 ipc_rcu_putref(sma, sem_rcu_free);
1754 return ERR_PTR(-ENOMEM);
1757 /* step 3: Acquire the lock on semaphore array */
1759 sem_lock_and_putref(sma);
1760 if (!ipc_valid_object(&sma->sem_perm)) {
1761 sem_unlock(sma, -1);
1764 un = ERR_PTR(-EIDRM);
1767 spin_lock(&ulp->lock);
1770 * step 4: check for races: did someone else allocate the undo struct?
1772 un = lookup_undo(ulp, semid);
1777 /* step 5: initialize & link new undo structure */
1778 new->semadj = (short *) &new[1];
1781 assert_spin_locked(&ulp->lock);
1782 list_add_rcu(&new->list_proc, &ulp->list_proc);
1783 ipc_assert_locked_object(&sma->sem_perm);
1784 list_add(&new->list_id, &sma->list_id);
1788 spin_unlock(&ulp->lock);
1789 sem_unlock(sma, -1);
1796 * get_queue_result - retrieve the result code from sem_queue
1797 * @q: Pointer to queue structure
1799 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1800 * q->status, then we must loop until the value is replaced with the final
1801 * value: This may happen if a task is woken up by an unrelated event (e.g.
1802 * signal) and in parallel the task is woken up by another task because it got
1803 * the requested semaphores.
1805 * The function can be called with or without holding the semaphore spinlock.
1807 static int get_queue_result(struct sem_queue *q)
1812 while (unlikely(error == IN_WAKEUP)) {
1820 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1821 unsigned, nsops, const struct timespec __user *, timeout)
1823 int error = -EINVAL;
1824 struct sem_array *sma;
1825 struct sembuf fast_sops[SEMOPM_FAST];
1826 struct sembuf *sops = fast_sops, *sop;
1827 struct sem_undo *un;
1828 int undos = 0, alter = 0, max, locknum;
1829 struct sem_queue queue;
1830 unsigned long jiffies_left = 0;
1831 struct ipc_namespace *ns;
1832 struct list_head tasks;
1834 ns = current->nsproxy->ipc_ns;
1836 if (nsops < 1 || semid < 0)
1838 if (nsops > ns->sc_semopm)
1840 if (nsops > SEMOPM_FAST) {
1841 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1845 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1850 struct timespec _timeout;
1851 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1855 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1856 _timeout.tv_nsec >= 1000000000L) {
1860 jiffies_left = timespec_to_jiffies(&_timeout);
1863 for (sop = sops; sop < sops + nsops; sop++) {
1864 if (sop->sem_num >= max)
1866 if (sop->sem_flg & SEM_UNDO)
1868 if (sop->sem_op != 0)
1872 INIT_LIST_HEAD(&tasks);
1875 /* On success, find_alloc_undo takes the rcu_read_lock */
1876 un = find_alloc_undo(ns, semid);
1878 error = PTR_ERR(un);
1886 sma = sem_obtain_object_check(ns, semid);
1889 error = PTR_ERR(sma);
1894 if (max >= sma->sem_nsems)
1895 goto out_rcu_wakeup;
1898 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1899 goto out_rcu_wakeup;
1901 error = security_sem_semop(sma, sops, nsops, alter);
1903 goto out_rcu_wakeup;
1906 locknum = sem_lock(sma, sops, nsops);
1908 * We eventually might perform the following check in a lockless
1909 * fashion, considering ipc_valid_object() locking constraints.
1910 * If nsops == 1 and there is no contention for sem_perm.lock, then
1911 * only a per-semaphore lock is held and it's OK to proceed with the
1912 * check below. More details on the fine grained locking scheme
1913 * entangled here and why it's RMID race safe on comments at sem_lock()
1915 if (!ipc_valid_object(&sma->sem_perm))
1916 goto out_unlock_free;
1918 * semid identifiers are not unique - find_alloc_undo may have
1919 * allocated an undo structure, it was invalidated by an RMID
1920 * and now a new array with received the same id. Check and fail.
1921 * This case can be detected checking un->semid. The existence of
1922 * "un" itself is guaranteed by rcu.
1924 if (un && un->semid == -1)
1925 goto out_unlock_free;
1928 queue.nsops = nsops;
1930 queue.pid = task_tgid_vnr(current);
1931 queue.alter = alter;
1933 error = perform_atomic_semop(sma, &queue);
1935 /* If the operation was successful, then do
1936 * the required updates.
1939 do_smart_update(sma, sops, nsops, 1, &tasks);
1941 set_semotime(sma, sops);
1944 goto out_unlock_free;
1946 /* We need to sleep on this operation, so we put the current
1947 * task into the pending queue and go to sleep.
1952 curr = &sma->sem_base[sops->sem_num];
1955 if (sma->complex_count) {
1956 list_add_tail(&queue.list,
1957 &sma->pending_alter);
1960 list_add_tail(&queue.list,
1961 &curr->pending_alter);
1964 list_add_tail(&queue.list, &curr->pending_const);
1967 if (!sma->complex_count)
1971 list_add_tail(&queue.list, &sma->pending_alter);
1973 list_add_tail(&queue.list, &sma->pending_const);
1975 sma->complex_count++;
1978 queue.status = -EINTR;
1979 queue.sleeper = current;
1982 __set_current_state(TASK_INTERRUPTIBLE);
1983 sem_unlock(sma, locknum);
1987 jiffies_left = schedule_timeout(jiffies_left);
1991 error = get_queue_result(&queue);
1993 if (error != -EINTR) {
1994 /* fast path: update_queue already obtained all requested
1996 * Perform a smp_mb(): User space could assume that semop()
1997 * is a memory barrier: Without the mb(), the cpu could
1998 * speculatively read in user space stale data that was
1999 * overwritten by the previous owner of the semaphore.
2007 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
2010 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
2012 error = get_queue_result(&queue);
2015 * Array removed? If yes, leave without sem_unlock().
2024 * If queue.status != -EINTR we are woken up by another process.
2025 * Leave without unlink_queue(), but with sem_unlock().
2027 if (error != -EINTR)
2028 goto out_unlock_free;
2031 * If an interrupt occurred we have to clean up the queue
2033 if (timeout && jiffies_left == 0)
2037 * If the wakeup was spurious, just retry
2039 if (error == -EINTR && !signal_pending(current))
2042 unlink_queue(sma, &queue);
2045 sem_unlock(sma, locknum);
2048 wake_up_sem_queue_do(&tasks);
2050 if (sops != fast_sops)
2055 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2058 return sys_semtimedop(semid, tsops, nsops, NULL);
2061 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2062 * parent and child tasks.
2065 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2067 struct sem_undo_list *undo_list;
2070 if (clone_flags & CLONE_SYSVSEM) {
2071 error = get_undo_list(&undo_list);
2074 atomic_inc(&undo_list->refcnt);
2075 tsk->sysvsem.undo_list = undo_list;
2077 tsk->sysvsem.undo_list = NULL;
2083 * add semadj values to semaphores, free undo structures.
2084 * undo structures are not freed when semaphore arrays are destroyed
2085 * so some of them may be out of date.
2086 * IMPLEMENTATION NOTE: There is some confusion over whether the
2087 * set of adjustments that needs to be done should be done in an atomic
2088 * manner or not. That is, if we are attempting to decrement the semval
2089 * should we queue up and wait until we can do so legally?
2090 * The original implementation attempted to do this (queue and wait).
2091 * The current implementation does not do so. The POSIX standard
2092 * and SVID should be consulted to determine what behavior is mandated.
2094 void exit_sem(struct task_struct *tsk)
2096 struct sem_undo_list *ulp;
2098 ulp = tsk->sysvsem.undo_list;
2101 tsk->sysvsem.undo_list = NULL;
2103 if (!atomic_dec_and_test(&ulp->refcnt))
2107 struct sem_array *sma;
2108 struct sem_undo *un;
2109 struct list_head tasks;
2113 un = list_entry_rcu(ulp->list_proc.next,
2114 struct sem_undo, list_proc);
2115 if (&un->list_proc == &ulp->list_proc) {
2117 * We must wait for freeary() before freeing this ulp,
2118 * in case we raced with last sem_undo. There is a small
2119 * possibility where we exit while freeary() didn't
2120 * finish unlocking sem_undo_list.
2122 spin_unlock_wait(&ulp->lock);
2126 spin_lock(&ulp->lock);
2128 spin_unlock(&ulp->lock);
2130 /* exit_sem raced with IPC_RMID, nothing to do */
2136 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2137 /* exit_sem raced with IPC_RMID, nothing to do */
2143 sem_lock(sma, NULL, -1);
2144 /* exit_sem raced with IPC_RMID, nothing to do */
2145 if (!ipc_valid_object(&sma->sem_perm)) {
2146 sem_unlock(sma, -1);
2150 un = __lookup_undo(ulp, semid);
2152 /* exit_sem raced with IPC_RMID+semget() that created
2153 * exactly the same semid. Nothing to do.
2155 sem_unlock(sma, -1);
2160 /* remove un from the linked lists */
2161 ipc_assert_locked_object(&sma->sem_perm);
2162 list_del(&un->list_id);
2164 /* we are the last process using this ulp, acquiring ulp->lock
2165 * isn't required. Besides that, we are also protected against
2166 * IPC_RMID as we hold sma->sem_perm lock now
2168 list_del_rcu(&un->list_proc);
2170 /* perform adjustments registered in un */
2171 for (i = 0; i < sma->sem_nsems; i++) {
2172 struct sem *semaphore = &sma->sem_base[i];
2173 if (un->semadj[i]) {
2174 semaphore->semval += un->semadj[i];
2176 * Range checks of the new semaphore value,
2177 * not defined by sus:
2178 * - Some unices ignore the undo entirely
2179 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2180 * - some cap the value (e.g. FreeBSD caps
2181 * at 0, but doesn't enforce SEMVMX)
2183 * Linux caps the semaphore value, both at 0
2186 * Manfred <manfred@colorfullife.com>
2188 if (semaphore->semval < 0)
2189 semaphore->semval = 0;
2190 if (semaphore->semval > SEMVMX)
2191 semaphore->semval = SEMVMX;
2192 semaphore->sempid = task_tgid_vnr(current);
2195 /* maybe some queued-up processes were waiting for this */
2196 INIT_LIST_HEAD(&tasks);
2197 do_smart_update(sma, NULL, 0, 1, &tasks);
2198 sem_unlock(sma, -1);
2200 wake_up_sem_queue_do(&tasks);
2207 #ifdef CONFIG_PROC_FS
2208 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2210 struct user_namespace *user_ns = seq_user_ns(s);
2211 struct sem_array *sma = it;
2215 * The proc interface isn't aware of sem_lock(), it calls
2216 * ipc_lock_object() directly (in sysvipc_find_ipc).
2217 * In order to stay compatible with sem_lock(), we must
2218 * enter / leave complex_mode.
2220 complexmode_enter(sma);
2222 sem_otime = get_semotime(sma);
2225 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2230 from_kuid_munged(user_ns, sma->sem_perm.uid),
2231 from_kgid_munged(user_ns, sma->sem_perm.gid),
2232 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2233 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2237 complexmode_tryleave(sma);