3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
8 * SMP-threaded, sysctl's added
9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
10 * Enforced range limit on SEM_UNDO
11 * (c) 2001 Red Hat Inc
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
14 * (c) 2016 Davidlohr Bueso <dave@stgolabs.net>
15 * Further wakeup optimizations, documentation
16 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
18 * support for audit of ipc object properties and permission changes
19 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
23 * Pavel Emelianov <xemul@openvz.org>
25 * Implementation notes: (May 2010)
26 * This file implements System V semaphores.
28 * User space visible behavior:
29 * - FIFO ordering for semop() operations (just FIFO, not starvation
31 * - multiple semaphore operations that alter the same semaphore in
32 * one semop() are handled.
33 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
35 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
36 * - undo adjustments at process exit are limited to 0..SEMVMX.
37 * - namespace are supported.
38 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
39 * to /proc/sys/kernel/sem.
40 * - statistics about the usage are reported in /proc/sysvipc/sem.
44 * - all global variables are read-mostly.
45 * - semop() calls and semctl(RMID) are synchronized by RCU.
46 * - most operations do write operations (actually: spin_lock calls) to
47 * the per-semaphore array structure.
48 * Thus: Perfect SMP scaling between independent semaphore arrays.
49 * If multiple semaphores in one array are used, then cache line
50 * trashing on the semaphore array spinlock will limit the scaling.
51 * - semncnt and semzcnt are calculated on demand in count_semcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare())
58 * - All work is done by the waker, the woken up task does not have to do
59 * anything - not even acquiring a lock or dropping a refcount.
60 * - A woken up task may not even touch the semaphore array anymore, it may
61 * have been destroyed already by a semctl(RMID).
62 * - UNDO values are stored in an array (one per process and per
63 * semaphore array, lazily allocated). For backwards compatibility, multiple
64 * modes for the UNDO variables are supported (per process, per thread)
65 * (see copy_semundo, CLONE_SYSVSEM)
66 * - There are two lists of the pending operations: a per-array list
67 * and per-semaphore list (stored in the array). This allows to achieve FIFO
68 * ordering without always scanning all pending operations.
69 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
72 #include <linux/slab.h>
73 #include <linux/spinlock.h>
74 #include <linux/init.h>
75 #include <linux/proc_fs.h>
76 #include <linux/time.h>
77 #include <linux/security.h>
78 #include <linux/syscalls.h>
79 #include <linux/audit.h>
80 #include <linux/capability.h>
81 #include <linux/seq_file.h>
82 #include <linux/rwsem.h>
83 #include <linux/nsproxy.h>
84 #include <linux/ipc_namespace.h>
86 #include <linux/uaccess.h>
89 /* One semaphore structure for each semaphore in the system. */
91 int semval; /* current value */
93 * PID of the process that last modified the semaphore. For
94 * Linux, specifically these are:
96 * - semctl, via SETVAL and SETALL.
97 * - at task exit when performing undo adjustments (see exit_sem).
100 spinlock_t lock; /* spinlock for fine-grained semtimedop */
101 struct list_head pending_alter; /* pending single-sop operations */
102 /* that alter the semaphore */
103 struct list_head pending_const; /* pending single-sop operations */
104 /* that do not alter the semaphore*/
105 time_t sem_otime; /* candidate for sem_otime */
106 } ____cacheline_aligned_in_smp;
108 /* One queue for each sleeping process in the system. */
110 struct list_head list; /* queue of pending operations */
111 struct task_struct *sleeper; /* this process */
112 struct sem_undo *undo; /* undo structure */
113 int pid; /* process id of requesting process */
114 int status; /* completion status of operation */
115 struct sembuf *sops; /* array of pending operations */
116 struct sembuf *blocking; /* the operation that blocked */
117 int nsops; /* number of operations */
118 bool alter; /* does *sops alter the array? */
119 bool dupsop; /* sops on more than one sem_num */
122 /* Each task has a list of undo requests. They are executed automatically
123 * when the process exits.
126 struct list_head list_proc; /* per-process list: *
127 * all undos from one process
129 struct rcu_head rcu; /* rcu struct for sem_undo */
130 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
131 struct list_head list_id; /* per semaphore array list:
132 * all undos for one array */
133 int semid; /* semaphore set identifier */
134 short *semadj; /* array of adjustments */
135 /* one per semaphore */
138 /* sem_undo_list controls shared access to the list of sem_undo structures
139 * that may be shared among all a CLONE_SYSVSEM task group.
141 struct sem_undo_list {
144 struct list_head list_proc;
148 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
150 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
152 static int newary(struct ipc_namespace *, struct ipc_params *);
153 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
154 #ifdef CONFIG_PROC_FS
155 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
158 #define SEMMSL_FAST 256 /* 512 bytes on stack */
159 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
163 * a) global sem_lock() for read/write
165 * sem_array.complex_count,
166 * sem_array.complex_mode
167 * sem_array.pending{_alter,_const},
170 * b) global or semaphore sem_lock() for read/write:
171 * sem_array.sem_base[i].pending_{const,alter}:
172 * sem_array.complex_mode (for read)
175 * sem_undo_list.list_proc:
176 * * undo_list->lock for write
180 #define sc_semmsl sem_ctls[0]
181 #define sc_semmns sem_ctls[1]
182 #define sc_semopm sem_ctls[2]
183 #define sc_semmni sem_ctls[3]
185 void sem_init_ns(struct ipc_namespace *ns)
187 ns->sc_semmsl = SEMMSL;
188 ns->sc_semmns = SEMMNS;
189 ns->sc_semopm = SEMOPM;
190 ns->sc_semmni = SEMMNI;
192 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
196 void sem_exit_ns(struct ipc_namespace *ns)
198 free_ipcs(ns, &sem_ids(ns), freeary);
199 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
203 void __init sem_init(void)
205 sem_init_ns(&init_ipc_ns);
206 ipc_init_proc_interface("sysvipc/sem",
207 " key semid perms nsems uid gid cuid cgid otime ctime\n",
208 IPC_SEM_IDS, sysvipc_sem_proc_show);
212 * unmerge_queues - unmerge queues, if possible.
213 * @sma: semaphore array
215 * The function unmerges the wait queues if complex_count is 0.
216 * It must be called prior to dropping the global semaphore array lock.
218 static void unmerge_queues(struct sem_array *sma)
220 struct sem_queue *q, *tq;
222 /* complex operations still around? */
223 if (sma->complex_count)
226 * We will switch back to simple mode.
227 * Move all pending operation back into the per-semaphore
230 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
232 curr = &sma->sem_base[q->sops[0].sem_num];
234 list_add_tail(&q->list, &curr->pending_alter);
236 INIT_LIST_HEAD(&sma->pending_alter);
240 * merge_queues - merge single semop queues into global queue
241 * @sma: semaphore array
243 * This function merges all per-semaphore queues into the global queue.
244 * It is necessary to achieve FIFO ordering for the pending single-sop
245 * operations when a multi-semop operation must sleep.
246 * Only the alter operations must be moved, the const operations can stay.
248 static void merge_queues(struct sem_array *sma)
251 for (i = 0; i < sma->sem_nsems; i++) {
252 struct sem *sem = sma->sem_base + i;
254 list_splice_init(&sem->pending_alter, &sma->pending_alter);
258 static void sem_rcu_free(struct rcu_head *head)
260 struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu);
261 struct sem_array *sma = ipc_rcu_to_struct(p);
263 security_sem_free(sma);
268 * Enter the mode suitable for non-simple operations:
269 * Caller must own sem_perm.lock.
271 static void complexmode_enter(struct sem_array *sma)
276 if (sma->complex_mode) {
277 /* We are already in complex_mode. Nothing to do */
281 /* We need a full barrier after seting complex_mode:
282 * The write to complex_mode must be visible
283 * before we read the first sem->lock spinlock state.
285 smp_store_mb(sma->complex_mode, true);
287 for (i = 0; i < sma->sem_nsems; i++) {
288 sem = sma->sem_base + i;
289 spin_unlock_wait(&sem->lock);
292 * spin_unlock_wait() is not a memory barriers, it is only a
293 * control barrier. The code must pair with spin_unlock(&sem->lock),
294 * thus just the control barrier is insufficient.
296 * smp_rmb() is sufficient, as writes cannot pass the control barrier.
302 * Try to leave the mode that disallows simple operations:
303 * Caller must own sem_perm.lock.
305 static void complexmode_tryleave(struct sem_array *sma)
307 if (sma->complex_count) {
308 /* Complex ops are sleeping.
309 * We must stay in complex mode
314 * Immediately after setting complex_mode to false,
315 * a simple op can start. Thus: all memory writes
316 * performed by the current operation must be visible
317 * before we set complex_mode to false.
319 smp_store_release(&sma->complex_mode, false);
322 #define SEM_GLOBAL_LOCK (-1)
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);
341 return SEM_GLOBAL_LOCK;
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,
355 * no locking, no memory barrier.
357 if (!sma->complex_mode) {
359 * It appears that no complex operation is around.
360 * Acquire the per-semaphore lock.
362 spin_lock(&sem->lock);
366 * ("powerpc: Add smp_mb() to arch_spin_is_locked()"):
367 * A full barrier is required: the write of sem->lock
368 * must be visible before the read is executed
372 if (!smp_load_acquire(&sma->complex_mode)) {
373 /* fast path successful! */
374 return sops->sem_num;
376 spin_unlock(&sem->lock);
379 /* slow path: acquire the full lock */
380 ipc_lock_object(&sma->sem_perm);
382 if (sma->complex_count == 0) {
384 * There is no complex operation, thus we can switch
385 * back to the fast path.
387 spin_lock(&sem->lock);
388 ipc_unlock_object(&sma->sem_perm);
389 return sops->sem_num;
391 /* Not a false alarm, thus complete the sequence for a
394 complexmode_enter(sma);
395 return SEM_GLOBAL_LOCK;
399 static inline void sem_unlock(struct sem_array *sma, int locknum)
401 if (locknum == SEM_GLOBAL_LOCK) {
403 complexmode_tryleave(sma);
404 ipc_unlock_object(&sma->sem_perm);
406 struct sem *sem = sma->sem_base + locknum;
407 spin_unlock(&sem->lock);
412 * sem_lock_(check_) routines are called in the paths where the rwsem
415 * The caller holds the RCU read lock.
417 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
418 int id, struct sembuf *sops, int nsops, int *locknum)
420 struct kern_ipc_perm *ipcp;
421 struct sem_array *sma;
423 ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
425 return ERR_CAST(ipcp);
427 sma = container_of(ipcp, struct sem_array, sem_perm);
428 *locknum = sem_lock(sma, sops, nsops);
430 /* ipc_rmid() may have already freed the ID while sem_lock
431 * was spinning: verify that the structure is still valid
433 if (ipc_valid_object(ipcp))
434 return container_of(ipcp, struct sem_array, sem_perm);
436 sem_unlock(sma, *locknum);
437 return ERR_PTR(-EINVAL);
440 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
442 struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id);
445 return ERR_CAST(ipcp);
447 return container_of(ipcp, struct sem_array, sem_perm);
450 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
453 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
456 return ERR_CAST(ipcp);
458 return container_of(ipcp, struct sem_array, sem_perm);
461 static inline void sem_lock_and_putref(struct sem_array *sma)
463 sem_lock(sma, NULL, -1);
464 ipc_rcu_putref(sma, sem_rcu_free);
467 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
469 ipc_rmid(&sem_ids(ns), &s->sem_perm);
473 * newary - Create a new semaphore set
475 * @params: ptr to the structure that contains key, semflg and nsems
477 * Called with sem_ids.rwsem held (as a writer)
479 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
483 struct sem_array *sma;
485 key_t key = params->key;
486 int nsems = params->u.nsems;
487 int semflg = params->flg;
492 if (ns->used_sems + nsems > ns->sc_semmns)
495 size = sizeof(*sma) + nsems * sizeof(struct sem);
496 sma = ipc_rcu_alloc(size);
500 memset(sma, 0, size);
502 sma->sem_perm.mode = (semflg & S_IRWXUGO);
503 sma->sem_perm.key = key;
505 sma->sem_perm.security = NULL;
506 retval = security_sem_alloc(sma);
508 ipc_rcu_putref(sma, ipc_rcu_free);
512 sma->sem_base = (struct sem *) &sma[1];
514 for (i = 0; i < nsems; i++) {
515 INIT_LIST_HEAD(&sma->sem_base[i].pending_alter);
516 INIT_LIST_HEAD(&sma->sem_base[i].pending_const);
517 spin_lock_init(&sma->sem_base[i].lock);
520 sma->complex_count = 0;
521 sma->complex_mode = true; /* dropped by sem_unlock below */
522 INIT_LIST_HEAD(&sma->pending_alter);
523 INIT_LIST_HEAD(&sma->pending_const);
524 INIT_LIST_HEAD(&sma->list_id);
525 sma->sem_nsems = nsems;
526 sma->sem_ctime = get_seconds();
528 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
530 ipc_rcu_putref(sma, sem_rcu_free);
533 ns->used_sems += nsems;
538 return sma->sem_perm.id;
543 * Called with sem_ids.rwsem and ipcp locked.
545 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
547 struct sem_array *sma;
549 sma = container_of(ipcp, struct sem_array, sem_perm);
550 return security_sem_associate(sma, semflg);
554 * Called with sem_ids.rwsem and ipcp locked.
556 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
557 struct ipc_params *params)
559 struct sem_array *sma;
561 sma = container_of(ipcp, struct sem_array, sem_perm);
562 if (params->u.nsems > sma->sem_nsems)
568 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
570 struct ipc_namespace *ns;
571 static const struct ipc_ops sem_ops = {
573 .associate = sem_security,
574 .more_checks = sem_more_checks,
576 struct ipc_params sem_params;
578 ns = current->nsproxy->ipc_ns;
580 if (nsems < 0 || nsems > ns->sc_semmsl)
583 sem_params.key = key;
584 sem_params.flg = semflg;
585 sem_params.u.nsems = nsems;
587 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
591 * perform_atomic_semop[_slow] - Attempt to perform semaphore
592 * operations on a given array.
593 * @sma: semaphore array
594 * @q: struct sem_queue that describes the operation
596 * Caller blocking are as follows, based the value
597 * indicated by the semaphore operation (sem_op):
599 * (1) >0 never blocks.
600 * (2) 0 (wait-for-zero operation): semval is non-zero.
601 * (3) <0 attempting to decrement semval to a value smaller than zero.
603 * Returns 0 if the operation was possible.
604 * Returns 1 if the operation is impossible, the caller must sleep.
605 * Returns <0 for error codes.
607 static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q)
609 int result, sem_op, nsops, pid;
619 for (sop = sops; sop < sops + nsops; sop++) {
620 curr = sma->sem_base + sop->sem_num;
621 sem_op = sop->sem_op;
622 result = curr->semval;
624 if (!sem_op && result)
633 if (sop->sem_flg & SEM_UNDO) {
634 int undo = un->semadj[sop->sem_num] - sem_op;
635 /* Exceeding the undo range is an error. */
636 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
638 un->semadj[sop->sem_num] = undo;
641 curr->semval = result;
646 while (sop >= sops) {
647 sma->sem_base[sop->sem_num].sempid = pid;
660 if (sop->sem_flg & IPC_NOWAIT)
667 while (sop >= sops) {
668 sem_op = sop->sem_op;
669 sma->sem_base[sop->sem_num].semval -= sem_op;
670 if (sop->sem_flg & SEM_UNDO)
671 un->semadj[sop->sem_num] += sem_op;
678 static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q)
680 int result, sem_op, nsops;
690 if (unlikely(q->dupsop))
691 return perform_atomic_semop_slow(sma, q);
694 * We scan the semaphore set twice, first to ensure that the entire
695 * operation can succeed, therefore avoiding any pointless writes
696 * to shared memory and having to undo such changes in order to block
697 * until the operations can go through.
699 for (sop = sops; sop < sops + nsops; sop++) {
700 curr = sma->sem_base + sop->sem_num;
701 sem_op = sop->sem_op;
702 result = curr->semval;
704 if (!sem_op && result)
705 goto would_block; /* wait-for-zero */
714 if (sop->sem_flg & SEM_UNDO) {
715 int undo = un->semadj[sop->sem_num] - sem_op;
717 /* Exceeding the undo range is an error. */
718 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
723 for (sop = sops; sop < sops + nsops; sop++) {
724 curr = sma->sem_base + sop->sem_num;
725 sem_op = sop->sem_op;
726 result = curr->semval;
728 if (sop->sem_flg & SEM_UNDO) {
729 int undo = un->semadj[sop->sem_num] - sem_op;
731 un->semadj[sop->sem_num] = undo;
733 curr->semval += sem_op;
734 curr->sempid = q->pid;
741 return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1;
744 static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error,
745 struct wake_q_head *wake_q)
747 wake_q_add(wake_q, q->sleeper);
749 * Rely on the above implicit barrier, such that we can
750 * ensure that we hold reference to the task before setting
751 * q->status. Otherwise we could race with do_exit if the
752 * task is awoken by an external event before calling
755 WRITE_ONCE(q->status, error);
758 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
762 sma->complex_count--;
765 /** check_restart(sma, q)
766 * @sma: semaphore array
767 * @q: the operation that just completed
769 * update_queue is O(N^2) when it restarts scanning the whole queue of
770 * waiting operations. Therefore this function checks if the restart is
771 * really necessary. It is called after a previously waiting operation
772 * modified the array.
773 * Note that wait-for-zero operations are handled without restart.
775 static inline int check_restart(struct sem_array *sma, struct sem_queue *q)
777 /* pending complex alter operations are too difficult to analyse */
778 if (!list_empty(&sma->pending_alter))
781 /* we were a sleeping complex operation. Too difficult */
785 /* It is impossible that someone waits for the new value:
786 * - complex operations always restart.
787 * - wait-for-zero are handled seperately.
788 * - q is a previously sleeping simple operation that
789 * altered the array. It must be a decrement, because
790 * simple increments never sleep.
791 * - If there are older (higher priority) decrements
792 * in the queue, then they have observed the original
793 * semval value and couldn't proceed. The operation
794 * decremented to value - thus they won't proceed either.
800 * wake_const_ops - wake up non-alter tasks
801 * @sma: semaphore array.
802 * @semnum: semaphore that was modified.
803 * @wake_q: lockless wake-queue head.
805 * wake_const_ops must be called after a semaphore in a semaphore array
806 * was set to 0. If complex const operations are pending, wake_const_ops must
807 * be called with semnum = -1, as well as with the number of each modified
809 * The tasks that must be woken up are added to @wake_q. The return code
810 * is stored in q->pid.
811 * The function returns 1 if at least one operation was completed successfully.
813 static int wake_const_ops(struct sem_array *sma, int semnum,
814 struct wake_q_head *wake_q)
816 struct sem_queue *q, *tmp;
817 struct list_head *pending_list;
818 int semop_completed = 0;
821 pending_list = &sma->pending_const;
823 pending_list = &sma->sem_base[semnum].pending_const;
825 list_for_each_entry_safe(q, tmp, pending_list, list) {
826 int error = perform_atomic_semop(sma, q);
830 /* operation completed, remove from queue & wakeup */
831 unlink_queue(sma, q);
833 wake_up_sem_queue_prepare(q, error, wake_q);
838 return semop_completed;
842 * do_smart_wakeup_zero - wakeup all wait for zero tasks
843 * @sma: semaphore array
844 * @sops: operations that were performed
845 * @nsops: number of operations
846 * @wake_q: lockless wake-queue head
848 * Checks all required queue for wait-for-zero operations, based
849 * on the actual changes that were performed on the semaphore array.
850 * The function returns 1 if at least one operation was completed successfully.
852 static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops,
853 int nsops, struct wake_q_head *wake_q)
856 int semop_completed = 0;
859 /* first: the per-semaphore queues, if known */
861 for (i = 0; i < nsops; i++) {
862 int num = sops[i].sem_num;
864 if (sma->sem_base[num].semval == 0) {
866 semop_completed |= wake_const_ops(sma, num, wake_q);
871 * No sops means modified semaphores not known.
872 * Assume all were changed.
874 for (i = 0; i < sma->sem_nsems; i++) {
875 if (sma->sem_base[i].semval == 0) {
877 semop_completed |= wake_const_ops(sma, i, wake_q);
882 * If one of the modified semaphores got 0,
883 * then check the global queue, too.
886 semop_completed |= wake_const_ops(sma, -1, wake_q);
888 return semop_completed;
893 * update_queue - look for tasks that can be completed.
894 * @sma: semaphore array.
895 * @semnum: semaphore that was modified.
896 * @wake_q: lockless wake-queue head.
898 * update_queue must be called after a semaphore in a semaphore array
899 * was modified. If multiple semaphores were modified, update_queue must
900 * be called with semnum = -1, as well as with the number of each modified
902 * The tasks that must be woken up are added to @wake_q. The return code
903 * is stored in q->pid.
904 * The function internally checks if const operations can now succeed.
906 * The function return 1 if at least one semop was completed successfully.
908 static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q)
910 struct sem_queue *q, *tmp;
911 struct list_head *pending_list;
912 int semop_completed = 0;
915 pending_list = &sma->pending_alter;
917 pending_list = &sma->sem_base[semnum].pending_alter;
920 list_for_each_entry_safe(q, tmp, pending_list, list) {
923 /* If we are scanning the single sop, per-semaphore list of
924 * one semaphore and that semaphore is 0, then it is not
925 * necessary to scan further: simple increments
926 * that affect only one entry succeed immediately and cannot
927 * be in the per semaphore pending queue, and decrements
928 * cannot be successful if the value is already 0.
930 if (semnum != -1 && sma->sem_base[semnum].semval == 0)
933 error = perform_atomic_semop(sma, q);
935 /* Does q->sleeper still need to sleep? */
939 unlink_queue(sma, q);
945 do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q);
946 restart = check_restart(sma, q);
949 wake_up_sem_queue_prepare(q, error, wake_q);
953 return semop_completed;
957 * set_semotime - set sem_otime
958 * @sma: semaphore array
959 * @sops: operations that modified the array, may be NULL
961 * sem_otime is replicated to avoid cache line trashing.
962 * This function sets one instance to the current time.
964 static void set_semotime(struct sem_array *sma, struct sembuf *sops)
967 sma->sem_base[0].sem_otime = get_seconds();
969 sma->sem_base[sops[0].sem_num].sem_otime =
975 * do_smart_update - optimized update_queue
976 * @sma: semaphore array
977 * @sops: operations that were performed
978 * @nsops: number of operations
979 * @otime: force setting otime
980 * @wake_q: lockless wake-queue head
982 * do_smart_update() does the required calls to update_queue and wakeup_zero,
983 * based on the actual changes that were performed on the semaphore array.
984 * Note that the function does not do the actual wake-up: the caller is
985 * responsible for calling wake_up_q().
986 * It is safe to perform this call after dropping all locks.
988 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
989 int otime, struct wake_q_head *wake_q)
993 otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q);
995 if (!list_empty(&sma->pending_alter)) {
996 /* semaphore array uses the global queue - just process it. */
997 otime |= update_queue(sma, -1, wake_q);
1001 * No sops, thus the modified semaphores are not
1004 for (i = 0; i < sma->sem_nsems; i++)
1005 otime |= update_queue(sma, i, wake_q);
1008 * Check the semaphores that were increased:
1009 * - No complex ops, thus all sleeping ops are
1011 * - if we decreased the value, then any sleeping
1012 * semaphore ops wont be able to run: If the
1013 * previous value was too small, then the new
1014 * value will be too small, too.
1016 for (i = 0; i < nsops; i++) {
1017 if (sops[i].sem_op > 0) {
1018 otime |= update_queue(sma,
1019 sops[i].sem_num, wake_q);
1025 set_semotime(sma, sops);
1029 * check_qop: Test if a queued operation sleeps on the semaphore semnum
1031 static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q,
1034 struct sembuf *sop = q->blocking;
1037 * Linux always (since 0.99.10) reported a task as sleeping on all
1038 * semaphores. This violates SUS, therefore it was changed to the
1039 * standard compliant behavior.
1040 * Give the administrators a chance to notice that an application
1041 * might misbehave because it relies on the Linux behavior.
1043 pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n"
1044 "The task %s (%d) triggered the difference, watch for misbehavior.\n",
1045 current->comm, task_pid_nr(current));
1047 if (sop->sem_num != semnum)
1050 if (count_zero && sop->sem_op == 0)
1052 if (!count_zero && sop->sem_op < 0)
1058 /* The following counts are associated to each semaphore:
1059 * semncnt number of tasks waiting on semval being nonzero
1060 * semzcnt number of tasks waiting on semval being zero
1062 * Per definition, a task waits only on the semaphore of the first semop
1063 * that cannot proceed, even if additional operation would block, too.
1065 static int count_semcnt(struct sem_array *sma, ushort semnum,
1068 struct list_head *l;
1069 struct sem_queue *q;
1073 /* First: check the simple operations. They are easy to evaluate */
1075 l = &sma->sem_base[semnum].pending_const;
1077 l = &sma->sem_base[semnum].pending_alter;
1079 list_for_each_entry(q, l, list) {
1080 /* all task on a per-semaphore list sleep on exactly
1086 /* Then: check the complex operations. */
1087 list_for_each_entry(q, &sma->pending_alter, list) {
1088 semcnt += check_qop(sma, semnum, q, count_zero);
1091 list_for_each_entry(q, &sma->pending_const, list) {
1092 semcnt += check_qop(sma, semnum, q, count_zero);
1098 /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked
1099 * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem
1100 * remains locked on exit.
1102 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1104 struct sem_undo *un, *tu;
1105 struct sem_queue *q, *tq;
1106 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
1108 DEFINE_WAKE_Q(wake_q);
1110 /* Free the existing undo structures for this semaphore set. */
1111 ipc_assert_locked_object(&sma->sem_perm);
1112 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
1113 list_del(&un->list_id);
1114 spin_lock(&un->ulp->lock);
1116 list_del_rcu(&un->list_proc);
1117 spin_unlock(&un->ulp->lock);
1121 /* Wake up all pending processes and let them fail with EIDRM. */
1122 list_for_each_entry_safe(q, tq, &sma->pending_const, list) {
1123 unlink_queue(sma, q);
1124 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1127 list_for_each_entry_safe(q, tq, &sma->pending_alter, list) {
1128 unlink_queue(sma, q);
1129 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1131 for (i = 0; i < sma->sem_nsems; i++) {
1132 struct sem *sem = sma->sem_base + i;
1133 list_for_each_entry_safe(q, tq, &sem->pending_const, list) {
1134 unlink_queue(sma, q);
1135 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1137 list_for_each_entry_safe(q, tq, &sem->pending_alter, list) {
1138 unlink_queue(sma, q);
1139 wake_up_sem_queue_prepare(q, -EIDRM, &wake_q);
1143 /* Remove the semaphore set from the IDR */
1145 sem_unlock(sma, -1);
1149 ns->used_sems -= sma->sem_nsems;
1150 ipc_rcu_putref(sma, sem_rcu_free);
1153 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
1157 return copy_to_user(buf, in, sizeof(*in));
1160 struct semid_ds out;
1162 memset(&out, 0, sizeof(out));
1164 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
1166 out.sem_otime = in->sem_otime;
1167 out.sem_ctime = in->sem_ctime;
1168 out.sem_nsems = in->sem_nsems;
1170 return copy_to_user(buf, &out, sizeof(out));
1177 static time_t get_semotime(struct sem_array *sma)
1182 res = sma->sem_base[0].sem_otime;
1183 for (i = 1; i < sma->sem_nsems; i++) {
1184 time_t to = sma->sem_base[i].sem_otime;
1192 static int semctl_nolock(struct ipc_namespace *ns, int semid,
1193 int cmd, int version, void __user *p)
1196 struct sem_array *sma;
1202 struct seminfo seminfo;
1205 err = security_sem_semctl(NULL, cmd);
1209 memset(&seminfo, 0, sizeof(seminfo));
1210 seminfo.semmni = ns->sc_semmni;
1211 seminfo.semmns = ns->sc_semmns;
1212 seminfo.semmsl = ns->sc_semmsl;
1213 seminfo.semopm = ns->sc_semopm;
1214 seminfo.semvmx = SEMVMX;
1215 seminfo.semmnu = SEMMNU;
1216 seminfo.semmap = SEMMAP;
1217 seminfo.semume = SEMUME;
1218 down_read(&sem_ids(ns).rwsem);
1219 if (cmd == SEM_INFO) {
1220 seminfo.semusz = sem_ids(ns).in_use;
1221 seminfo.semaem = ns->used_sems;
1223 seminfo.semusz = SEMUSZ;
1224 seminfo.semaem = SEMAEM;
1226 max_id = ipc_get_maxid(&sem_ids(ns));
1227 up_read(&sem_ids(ns).rwsem);
1228 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
1230 return (max_id < 0) ? 0 : max_id;
1235 struct semid64_ds tbuf;
1238 memset(&tbuf, 0, sizeof(tbuf));
1241 if (cmd == SEM_STAT) {
1242 sma = sem_obtain_object(ns, semid);
1247 id = sma->sem_perm.id;
1249 sma = sem_obtain_object_check(ns, semid);
1257 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
1260 err = security_sem_semctl(sma, cmd);
1264 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
1265 tbuf.sem_otime = get_semotime(sma);
1266 tbuf.sem_ctime = sma->sem_ctime;
1267 tbuf.sem_nsems = sma->sem_nsems;
1269 if (copy_semid_to_user(p, &tbuf, version))
1281 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1284 struct sem_undo *un;
1285 struct sem_array *sma;
1288 DEFINE_WAKE_Q(wake_q);
1290 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1291 /* big-endian 64bit */
1294 /* 32bit or little-endian 64bit */
1298 if (val > SEMVMX || val < 0)
1302 sma = sem_obtain_object_check(ns, semid);
1305 return PTR_ERR(sma);
1308 if (semnum < 0 || semnum >= sma->sem_nsems) {
1314 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1319 err = security_sem_semctl(sma, SETVAL);
1325 sem_lock(sma, NULL, -1);
1327 if (!ipc_valid_object(&sma->sem_perm)) {
1328 sem_unlock(sma, -1);
1333 curr = &sma->sem_base[semnum];
1335 ipc_assert_locked_object(&sma->sem_perm);
1336 list_for_each_entry(un, &sma->list_id, list_id)
1337 un->semadj[semnum] = 0;
1340 curr->sempid = task_tgid_vnr(current);
1341 sma->sem_ctime = get_seconds();
1342 /* maybe some queued-up processes were waiting for this */
1343 do_smart_update(sma, NULL, 0, 0, &wake_q);
1344 sem_unlock(sma, -1);
1350 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1351 int cmd, void __user *p)
1353 struct sem_array *sma;
1356 ushort fast_sem_io[SEMMSL_FAST];
1357 ushort *sem_io = fast_sem_io;
1358 DEFINE_WAKE_Q(wake_q);
1361 sma = sem_obtain_object_check(ns, semid);
1364 return PTR_ERR(sma);
1367 nsems = sma->sem_nsems;
1370 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1371 goto out_rcu_wakeup;
1373 err = security_sem_semctl(sma, cmd);
1375 goto out_rcu_wakeup;
1381 ushort __user *array = p;
1384 sem_lock(sma, NULL, -1);
1385 if (!ipc_valid_object(&sma->sem_perm)) {
1389 if (nsems > SEMMSL_FAST) {
1390 if (!ipc_rcu_getref(sma)) {
1394 sem_unlock(sma, -1);
1396 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1397 if (sem_io == NULL) {
1398 ipc_rcu_putref(sma, sem_rcu_free);
1403 sem_lock_and_putref(sma);
1404 if (!ipc_valid_object(&sma->sem_perm)) {
1409 for (i = 0; i < sma->sem_nsems; i++)
1410 sem_io[i] = sma->sem_base[i].semval;
1411 sem_unlock(sma, -1);
1414 if (copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1421 struct sem_undo *un;
1423 if (!ipc_rcu_getref(sma)) {
1425 goto out_rcu_wakeup;
1429 if (nsems > SEMMSL_FAST) {
1430 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1431 if (sem_io == NULL) {
1432 ipc_rcu_putref(sma, sem_rcu_free);
1437 if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) {
1438 ipc_rcu_putref(sma, sem_rcu_free);
1443 for (i = 0; i < nsems; i++) {
1444 if (sem_io[i] > SEMVMX) {
1445 ipc_rcu_putref(sma, sem_rcu_free);
1451 sem_lock_and_putref(sma);
1452 if (!ipc_valid_object(&sma->sem_perm)) {
1457 for (i = 0; i < nsems; i++) {
1458 sma->sem_base[i].semval = sem_io[i];
1459 sma->sem_base[i].sempid = task_tgid_vnr(current);
1462 ipc_assert_locked_object(&sma->sem_perm);
1463 list_for_each_entry(un, &sma->list_id, list_id) {
1464 for (i = 0; i < nsems; i++)
1467 sma->sem_ctime = get_seconds();
1468 /* maybe some queued-up processes were waiting for this */
1469 do_smart_update(sma, NULL, 0, 0, &wake_q);
1473 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1476 if (semnum < 0 || semnum >= nsems)
1477 goto out_rcu_wakeup;
1479 sem_lock(sma, NULL, -1);
1480 if (!ipc_valid_object(&sma->sem_perm)) {
1484 curr = &sma->sem_base[semnum];
1494 err = count_semcnt(sma, semnum, 0);
1497 err = count_semcnt(sma, semnum, 1);
1502 sem_unlock(sma, -1);
1507 if (sem_io != fast_sem_io)
1512 static inline unsigned long
1513 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1517 if (copy_from_user(out, buf, sizeof(*out)))
1522 struct semid_ds tbuf_old;
1524 if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1527 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1528 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1529 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1539 * This function handles some semctl commands which require the rwsem
1540 * to be held in write mode.
1541 * NOTE: no locks must be held, the rwsem is taken inside this function.
1543 static int semctl_down(struct ipc_namespace *ns, int semid,
1544 int cmd, int version, void __user *p)
1546 struct sem_array *sma;
1548 struct semid64_ds semid64;
1549 struct kern_ipc_perm *ipcp;
1551 if (cmd == IPC_SET) {
1552 if (copy_semid_from_user(&semid64, p, version))
1556 down_write(&sem_ids(ns).rwsem);
1559 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1560 &semid64.sem_perm, 0);
1562 err = PTR_ERR(ipcp);
1566 sma = container_of(ipcp, struct sem_array, sem_perm);
1568 err = security_sem_semctl(sma, cmd);
1574 sem_lock(sma, NULL, -1);
1575 /* freeary unlocks the ipc object and rcu */
1579 sem_lock(sma, NULL, -1);
1580 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1583 sma->sem_ctime = get_seconds();
1591 sem_unlock(sma, -1);
1595 up_write(&sem_ids(ns).rwsem);
1599 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1602 struct ipc_namespace *ns;
1603 void __user *p = (void __user *)arg;
1608 version = ipc_parse_version(&cmd);
1609 ns = current->nsproxy->ipc_ns;
1616 return semctl_nolock(ns, semid, cmd, version, p);
1623 return semctl_main(ns, semid, semnum, cmd, p);
1625 return semctl_setval(ns, semid, semnum, arg);
1628 return semctl_down(ns, semid, cmd, version, p);
1634 /* If the task doesn't already have a undo_list, then allocate one
1635 * here. We guarantee there is only one thread using this undo list,
1636 * and current is THE ONE
1638 * If this allocation and assignment succeeds, but later
1639 * portions of this code fail, there is no need to free the sem_undo_list.
1640 * Just let it stay associated with the task, and it'll be freed later
1643 * This can block, so callers must hold no locks.
1645 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1647 struct sem_undo_list *undo_list;
1649 undo_list = current->sysvsem.undo_list;
1651 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1652 if (undo_list == NULL)
1654 spin_lock_init(&undo_list->lock);
1655 atomic_set(&undo_list->refcnt, 1);
1656 INIT_LIST_HEAD(&undo_list->list_proc);
1658 current->sysvsem.undo_list = undo_list;
1660 *undo_listp = undo_list;
1664 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1666 struct sem_undo *un;
1668 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1669 if (un->semid == semid)
1675 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1677 struct sem_undo *un;
1679 assert_spin_locked(&ulp->lock);
1681 un = __lookup_undo(ulp, semid);
1683 list_del_rcu(&un->list_proc);
1684 list_add_rcu(&un->list_proc, &ulp->list_proc);
1690 * find_alloc_undo - lookup (and if not present create) undo array
1692 * @semid: semaphore array id
1694 * The function looks up (and if not present creates) the undo structure.
1695 * The size of the undo structure depends on the size of the semaphore
1696 * array, thus the alloc path is not that straightforward.
1697 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1698 * performs a rcu_read_lock().
1700 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1702 struct sem_array *sma;
1703 struct sem_undo_list *ulp;
1704 struct sem_undo *un, *new;
1707 error = get_undo_list(&ulp);
1709 return ERR_PTR(error);
1712 spin_lock(&ulp->lock);
1713 un = lookup_undo(ulp, semid);
1714 spin_unlock(&ulp->lock);
1715 if (likely(un != NULL))
1718 /* no undo structure around - allocate one. */
1719 /* step 1: figure out the size of the semaphore array */
1720 sma = sem_obtain_object_check(ns, semid);
1723 return ERR_CAST(sma);
1726 nsems = sma->sem_nsems;
1727 if (!ipc_rcu_getref(sma)) {
1729 un = ERR_PTR(-EIDRM);
1734 /* step 2: allocate new undo structure */
1735 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1737 ipc_rcu_putref(sma, sem_rcu_free);
1738 return ERR_PTR(-ENOMEM);
1741 /* step 3: Acquire the lock on semaphore array */
1743 sem_lock_and_putref(sma);
1744 if (!ipc_valid_object(&sma->sem_perm)) {
1745 sem_unlock(sma, -1);
1748 un = ERR_PTR(-EIDRM);
1751 spin_lock(&ulp->lock);
1754 * step 4: check for races: did someone else allocate the undo struct?
1756 un = lookup_undo(ulp, semid);
1761 /* step 5: initialize & link new undo structure */
1762 new->semadj = (short *) &new[1];
1765 assert_spin_locked(&ulp->lock);
1766 list_add_rcu(&new->list_proc, &ulp->list_proc);
1767 ipc_assert_locked_object(&sma->sem_perm);
1768 list_add(&new->list_id, &sma->list_id);
1772 spin_unlock(&ulp->lock);
1773 sem_unlock(sma, -1);
1778 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1779 unsigned, nsops, const struct timespec __user *, timeout)
1781 int error = -EINVAL;
1782 struct sem_array *sma;
1783 struct sembuf fast_sops[SEMOPM_FAST];
1784 struct sembuf *sops = fast_sops, *sop;
1785 struct sem_undo *un;
1787 bool undos = false, alter = false, dupsop = false;
1788 struct sem_queue queue;
1789 unsigned long dup = 0, jiffies_left = 0;
1790 struct ipc_namespace *ns;
1792 ns = current->nsproxy->ipc_ns;
1794 if (nsops < 1 || semid < 0)
1796 if (nsops > ns->sc_semopm)
1798 if (nsops > SEMOPM_FAST) {
1799 sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL);
1804 if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) {
1810 struct timespec _timeout;
1811 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1815 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1816 _timeout.tv_nsec >= 1000000000L) {
1820 jiffies_left = timespec_to_jiffies(&_timeout);
1824 for (sop = sops; sop < sops + nsops; sop++) {
1825 unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG);
1827 if (sop->sem_num >= max)
1829 if (sop->sem_flg & SEM_UNDO)
1833 * There was a previous alter access that appears
1834 * to have accessed the same semaphore, thus use
1835 * the dupsop logic. "appears", because the detection
1836 * can only check % BITS_PER_LONG.
1840 if (sop->sem_op != 0) {
1847 /* On success, find_alloc_undo takes the rcu_read_lock */
1848 un = find_alloc_undo(ns, semid);
1850 error = PTR_ERR(un);
1858 sma = sem_obtain_object_check(ns, semid);
1861 error = PTR_ERR(sma);
1866 if (max >= sma->sem_nsems) {
1872 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1877 error = security_sem_semop(sma, sops, nsops, alter);
1884 locknum = sem_lock(sma, sops, nsops);
1886 * We eventually might perform the following check in a lockless
1887 * fashion, considering ipc_valid_object() locking constraints.
1888 * If nsops == 1 and there is no contention for sem_perm.lock, then
1889 * only a per-semaphore lock is held and it's OK to proceed with the
1890 * check below. More details on the fine grained locking scheme
1891 * entangled here and why it's RMID race safe on comments at sem_lock()
1893 if (!ipc_valid_object(&sma->sem_perm))
1894 goto out_unlock_free;
1896 * semid identifiers are not unique - find_alloc_undo may have
1897 * allocated an undo structure, it was invalidated by an RMID
1898 * and now a new array with received the same id. Check and fail.
1899 * This case can be detected checking un->semid. The existence of
1900 * "un" itself is guaranteed by rcu.
1902 if (un && un->semid == -1)
1903 goto out_unlock_free;
1906 queue.nsops = nsops;
1908 queue.pid = task_tgid_vnr(current);
1909 queue.alter = alter;
1910 queue.dupsop = dupsop;
1912 error = perform_atomic_semop(sma, &queue);
1913 if (error == 0) { /* non-blocking succesfull path */
1914 DEFINE_WAKE_Q(wake_q);
1917 * If the operation was successful, then do
1918 * the required updates.
1921 do_smart_update(sma, sops, nsops, 1, &wake_q);
1923 set_semotime(sma, sops);
1925 sem_unlock(sma, locknum);
1931 if (error < 0) /* non-blocking error path */
1932 goto out_unlock_free;
1935 * We need to sleep on this operation, so we put the current
1936 * 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++;
1967 queue.status = -EINTR;
1968 queue.sleeper = current;
1970 __set_current_state(TASK_INTERRUPTIBLE);
1971 sem_unlock(sma, locknum);
1975 jiffies_left = schedule_timeout(jiffies_left);
1980 * fastpath: the semop has completed, either successfully or
1981 * not, from the syscall pov, is quite irrelevant to us at this
1982 * point; we're done.
1984 * We _do_ care, nonetheless, about being awoken by a signal or
1985 * spuriously. The queue.status is checked again in the
1986 * slowpath (aka after taking sem_lock), such that we can detect
1987 * scenarios where we were awakened externally, during the
1988 * window between wake_q_add() and wake_up_q().
1990 error = READ_ONCE(queue.status);
1991 if (error != -EINTR) {
1993 * User space could assume that semop() is a memory
1994 * barrier: Without the mb(), the cpu could
1995 * speculatively read in userspace stale data that was
1996 * overwritten by the previous owner of the semaphore.
2003 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
2004 error = READ_ONCE(queue.status);
2007 * Array removed? If yes, leave without sem_unlock().
2015 * If queue.status != -EINTR we are woken up by another process.
2016 * Leave without unlink_queue(), but with sem_unlock().
2018 if (error != -EINTR)
2019 goto out_unlock_free;
2022 * If an interrupt occurred we have to clean up the queue.
2024 if (timeout && jiffies_left == 0)
2026 } while (error == -EINTR && !signal_pending(current)); /* spurious */
2028 unlink_queue(sma, &queue);
2031 sem_unlock(sma, locknum);
2034 if (sops != fast_sops)
2039 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
2042 return sys_semtimedop(semid, tsops, nsops, NULL);
2045 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
2046 * parent and child tasks.
2049 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
2051 struct sem_undo_list *undo_list;
2054 if (clone_flags & CLONE_SYSVSEM) {
2055 error = get_undo_list(&undo_list);
2058 atomic_inc(&undo_list->refcnt);
2059 tsk->sysvsem.undo_list = undo_list;
2061 tsk->sysvsem.undo_list = NULL;
2067 * add semadj values to semaphores, free undo structures.
2068 * undo structures are not freed when semaphore arrays are destroyed
2069 * so some of them may be out of date.
2070 * IMPLEMENTATION NOTE: There is some confusion over whether the
2071 * set of adjustments that needs to be done should be done in an atomic
2072 * manner or not. That is, if we are attempting to decrement the semval
2073 * should we queue up and wait until we can do so legally?
2074 * The original implementation attempted to do this (queue and wait).
2075 * The current implementation does not do so. The POSIX standard
2076 * and SVID should be consulted to determine what behavior is mandated.
2078 void exit_sem(struct task_struct *tsk)
2080 struct sem_undo_list *ulp;
2082 ulp = tsk->sysvsem.undo_list;
2085 tsk->sysvsem.undo_list = NULL;
2087 if (!atomic_dec_and_test(&ulp->refcnt))
2091 struct sem_array *sma;
2092 struct sem_undo *un;
2094 DEFINE_WAKE_Q(wake_q);
2099 un = list_entry_rcu(ulp->list_proc.next,
2100 struct sem_undo, list_proc);
2101 if (&un->list_proc == &ulp->list_proc) {
2103 * We must wait for freeary() before freeing this ulp,
2104 * in case we raced with last sem_undo. There is a small
2105 * possibility where we exit while freeary() didn't
2106 * finish unlocking sem_undo_list.
2108 spin_unlock_wait(&ulp->lock);
2112 spin_lock(&ulp->lock);
2114 spin_unlock(&ulp->lock);
2116 /* exit_sem raced with IPC_RMID, nothing to do */
2122 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid);
2123 /* exit_sem raced with IPC_RMID, nothing to do */
2129 sem_lock(sma, NULL, -1);
2130 /* exit_sem raced with IPC_RMID, nothing to do */
2131 if (!ipc_valid_object(&sma->sem_perm)) {
2132 sem_unlock(sma, -1);
2136 un = __lookup_undo(ulp, semid);
2138 /* exit_sem raced with IPC_RMID+semget() that created
2139 * exactly the same semid. Nothing to do.
2141 sem_unlock(sma, -1);
2146 /* remove un from the linked lists */
2147 ipc_assert_locked_object(&sma->sem_perm);
2148 list_del(&un->list_id);
2150 /* we are the last process using this ulp, acquiring ulp->lock
2151 * isn't required. Besides that, we are also protected against
2152 * IPC_RMID as we hold sma->sem_perm lock now
2154 list_del_rcu(&un->list_proc);
2156 /* perform adjustments registered in un */
2157 for (i = 0; i < sma->sem_nsems; i++) {
2158 struct sem *semaphore = &sma->sem_base[i];
2159 if (un->semadj[i]) {
2160 semaphore->semval += un->semadj[i];
2162 * Range checks of the new semaphore value,
2163 * not defined by sus:
2164 * - Some unices ignore the undo entirely
2165 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
2166 * - some cap the value (e.g. FreeBSD caps
2167 * at 0, but doesn't enforce SEMVMX)
2169 * Linux caps the semaphore value, both at 0
2172 * Manfred <manfred@colorfullife.com>
2174 if (semaphore->semval < 0)
2175 semaphore->semval = 0;
2176 if (semaphore->semval > SEMVMX)
2177 semaphore->semval = SEMVMX;
2178 semaphore->sempid = task_tgid_vnr(current);
2181 /* maybe some queued-up processes were waiting for this */
2182 do_smart_update(sma, NULL, 0, 1, &wake_q);
2183 sem_unlock(sma, -1);
2192 #ifdef CONFIG_PROC_FS
2193 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
2195 struct user_namespace *user_ns = seq_user_ns(s);
2196 struct sem_array *sma = it;
2200 * The proc interface isn't aware of sem_lock(), it calls
2201 * ipc_lock_object() directly (in sysvipc_find_ipc).
2202 * In order to stay compatible with sem_lock(), we must
2203 * enter / leave complex_mode.
2205 complexmode_enter(sma);
2207 sem_otime = get_semotime(sma);
2210 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
2215 from_kuid_munged(user_ns, sma->sem_perm.uid),
2216 from_kgid_munged(user_ns, sma->sem_perm.gid),
2217 from_kuid_munged(user_ns, sma->sem_perm.cuid),
2218 from_kgid_munged(user_ns, sma->sem_perm.cgid),
2222 complexmode_tryleave(sma);