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_semncnt() and
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 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
76 #include <linux/slab.h>
77 #include <linux/spinlock.h>
78 #include <linux/init.h>
79 #include <linux/proc_fs.h>
80 #include <linux/time.h>
81 #include <linux/security.h>
82 #include <linux/syscalls.h>
83 #include <linux/audit.h>
84 #include <linux/capability.h>
85 #include <linux/seq_file.h>
86 #include <linux/rwsem.h>
87 #include <linux/nsproxy.h>
88 #include <linux/ipc_namespace.h>
90 #include <asm/uaccess.h>
93 /* One semaphore structure for each semaphore in the system. */
95 int semval; /* current value */
96 int sempid; /* pid of last operation */
97 spinlock_t lock; /* spinlock for fine-grained semtimedop */
98 struct list_head sem_pending; /* pending single-sop operations */
101 /* One queue for each sleeping process in the system. */
103 struct list_head list; /* queue of pending operations */
104 struct task_struct *sleeper; /* this process */
105 struct sem_undo *undo; /* undo structure */
106 int pid; /* process id of requesting process */
107 int status; /* completion status of operation */
108 struct sembuf *sops; /* array of pending operations */
109 int nsops; /* number of operations */
110 int alter; /* does *sops alter the array? */
113 /* Each task has a list of undo requests. They are executed automatically
114 * when the process exits.
117 struct list_head list_proc; /* per-process list: *
118 * all undos from one process
120 struct rcu_head rcu; /* rcu struct for sem_undo */
121 struct sem_undo_list *ulp; /* back ptr to sem_undo_list */
122 struct list_head list_id; /* per semaphore array list:
123 * all undos for one array */
124 int semid; /* semaphore set identifier */
125 short *semadj; /* array of adjustments */
126 /* one per semaphore */
129 /* sem_undo_list controls shared access to the list of sem_undo structures
130 * that may be shared among all a CLONE_SYSVSEM task group.
132 struct sem_undo_list {
135 struct list_head list_proc;
139 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
141 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
143 static int newary(struct ipc_namespace *, struct ipc_params *);
144 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
145 #ifdef CONFIG_PROC_FS
146 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
149 #define SEMMSL_FAST 256 /* 512 bytes on stack */
150 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
153 * linked list protection:
155 * sem_array.sem_pending{,last},
156 * sem_array.sem_undo: sem_lock() for read/write
157 * sem_undo.proc_next: only "current" is allowed to read/write that field.
161 #define sc_semmsl sem_ctls[0]
162 #define sc_semmns sem_ctls[1]
163 #define sc_semopm sem_ctls[2]
164 #define sc_semmni sem_ctls[3]
166 void sem_init_ns(struct ipc_namespace *ns)
168 ns->sc_semmsl = SEMMSL;
169 ns->sc_semmns = SEMMNS;
170 ns->sc_semopm = SEMOPM;
171 ns->sc_semmni = SEMMNI;
173 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
177 void sem_exit_ns(struct ipc_namespace *ns)
179 free_ipcs(ns, &sem_ids(ns), freeary);
180 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
184 void __init sem_init (void)
186 sem_init_ns(&init_ipc_ns);
187 ipc_init_proc_interface("sysvipc/sem",
188 " key semid perms nsems uid gid cuid cgid otime ctime\n",
189 IPC_SEM_IDS, sysvipc_sem_proc_show);
193 * If the request contains only one semaphore operation, and there are
194 * no complex transactions pending, lock only the semaphore involved.
195 * Otherwise, lock the entire semaphore array, since we either have
196 * multiple semaphores in our own semops, or we need to look at
197 * semaphores from other pending complex operations.
199 * Carefully guard against sma->complex_count changing between zero
200 * and non-zero while we are spinning for the lock. The value of
201 * sma->complex_count cannot change while we are holding the lock,
202 * so sem_unlock should be fine.
204 * The global lock path checks that all the local locks have been released,
205 * checking each local lock once. This means that the local lock paths
206 * cannot start their critical sections while the global lock is held.
208 static inline int sem_lock(struct sem_array *sma, struct sembuf *sops,
213 if (nsops == 1 && !sma->complex_count) {
214 struct sem *sem = sma->sem_base + sops->sem_num;
216 /* Lock just the semaphore we are interested in. */
217 spin_lock(&sem->lock);
220 * If sma->complex_count was set while we were spinning,
221 * we may need to look at things we did not lock here.
223 if (unlikely(sma->complex_count)) {
224 spin_unlock(&sem->lock);
229 * Another process is holding the global lock on the
230 * sem_array; we cannot enter our critical section,
231 * but have to wait for the global lock to be released.
233 if (unlikely(spin_is_locked(&sma->sem_perm.lock))) {
234 spin_unlock(&sem->lock);
235 spin_unlock_wait(&sma->sem_perm.lock);
239 locknum = sops->sem_num;
243 * Lock the semaphore array, and wait for all of the
244 * individual semaphore locks to go away. The code
245 * above ensures no new single-lock holders will enter
246 * their critical section while the array lock is held.
249 spin_lock(&sma->sem_perm.lock);
250 for (i = 0; i < sma->sem_nsems; i++) {
251 struct sem *sem = sma->sem_base + i;
252 spin_unlock_wait(&sem->lock);
259 static inline void sem_unlock(struct sem_array *sma, int locknum)
262 spin_unlock(&sma->sem_perm.lock);
264 struct sem *sem = sma->sem_base + locknum;
265 spin_unlock(&sem->lock);
271 * sem_lock_(check_) routines are called in the paths where the rw_mutex
274 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
275 int id, struct sembuf *sops, int nsops, int *locknum)
277 struct kern_ipc_perm *ipcp;
278 struct sem_array *sma;
281 ipcp = ipc_obtain_object(&sem_ids(ns), id);
283 sma = ERR_CAST(ipcp);
287 sma = container_of(ipcp, struct sem_array, sem_perm);
288 *locknum = sem_lock(sma, sops, nsops);
290 /* ipc_rmid() may have already freed the ID while sem_lock
291 * was spinning: verify that the structure is still valid
294 return container_of(ipcp, struct sem_array, sem_perm);
296 sem_unlock(sma, *locknum);
297 sma = ERR_PTR(-EINVAL);
303 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
305 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
308 return ERR_CAST(ipcp);
310 return container_of(ipcp, struct sem_array, sem_perm);
313 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
316 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
319 return ERR_CAST(ipcp);
321 return container_of(ipcp, struct sem_array, sem_perm);
324 static inline void sem_lock_and_putref(struct sem_array *sma)
327 sem_lock(sma, NULL, -1);
331 static inline void sem_getref_and_unlock(struct sem_array *sma)
333 WARN_ON_ONCE(!ipc_rcu_getref(sma));
337 static inline void sem_putref(struct sem_array *sma)
339 sem_lock_and_putref(sma);
344 * Call inside the rcu read section.
346 static inline void sem_getref(struct sem_array *sma)
348 sem_lock(sma, NULL, -1);
349 WARN_ON_ONCE(!ipc_rcu_getref(sma));
353 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
355 ipc_rmid(&sem_ids(ns), &s->sem_perm);
359 * Lockless wakeup algorithm:
360 * Without the check/retry algorithm a lockless wakeup is possible:
361 * - queue.status is initialized to -EINTR before blocking.
362 * - wakeup is performed by
363 * * unlinking the queue entry from sma->sem_pending
364 * * setting queue.status to IN_WAKEUP
365 * This is the notification for the blocked thread that a
366 * result value is imminent.
367 * * call wake_up_process
368 * * set queue.status to the final value.
369 * - the previously blocked thread checks queue.status:
370 * * if it's IN_WAKEUP, then it must wait until the value changes
371 * * if it's not -EINTR, then the operation was completed by
372 * update_queue. semtimedop can return queue.status without
373 * performing any operation on the sem array.
374 * * otherwise it must acquire the spinlock and check what's up.
376 * The two-stage algorithm is necessary to protect against the following
378 * - if queue.status is set after wake_up_process, then the woken up idle
379 * thread could race forward and try (and fail) to acquire sma->lock
380 * before update_queue had a chance to set queue.status
381 * - if queue.status is written before wake_up_process and if the
382 * blocked process is woken up by a signal between writing
383 * queue.status and the wake_up_process, then the woken up
384 * process could return from semtimedop and die by calling
385 * sys_exit before wake_up_process is called. Then wake_up_process
386 * will oops, because the task structure is already invalid.
387 * (yes, this happened on s390 with sysv msg).
393 * newary - Create a new semaphore set
395 * @params: ptr to the structure that contains key, semflg and nsems
397 * Called with sem_ids.rw_mutex held (as a writer)
400 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
404 struct sem_array *sma;
406 key_t key = params->key;
407 int nsems = params->u.nsems;
408 int semflg = params->flg;
413 if (ns->used_sems + nsems > ns->sc_semmns)
416 size = sizeof (*sma) + nsems * sizeof (struct sem);
417 sma = ipc_rcu_alloc(size);
421 memset (sma, 0, size);
423 sma->sem_perm.mode = (semflg & S_IRWXUGO);
424 sma->sem_perm.key = key;
426 sma->sem_perm.security = NULL;
427 retval = security_sem_alloc(sma);
433 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
435 security_sem_free(sma);
439 ns->used_sems += nsems;
441 sma->sem_base = (struct sem *) &sma[1];
443 for (i = 0; i < nsems; i++) {
444 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
445 spin_lock_init(&sma->sem_base[i].lock);
448 sma->complex_count = 0;
449 INIT_LIST_HEAD(&sma->sem_pending);
450 INIT_LIST_HEAD(&sma->list_id);
451 sma->sem_nsems = nsems;
452 sma->sem_ctime = get_seconds();
455 return sma->sem_perm.id;
460 * Called with sem_ids.rw_mutex and ipcp locked.
462 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
464 struct sem_array *sma;
466 sma = container_of(ipcp, struct sem_array, sem_perm);
467 return security_sem_associate(sma, semflg);
471 * Called with sem_ids.rw_mutex and ipcp locked.
473 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
474 struct ipc_params *params)
476 struct sem_array *sma;
478 sma = container_of(ipcp, struct sem_array, sem_perm);
479 if (params->u.nsems > sma->sem_nsems)
485 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
487 struct ipc_namespace *ns;
488 struct ipc_ops sem_ops;
489 struct ipc_params sem_params;
491 ns = current->nsproxy->ipc_ns;
493 if (nsems < 0 || nsems > ns->sc_semmsl)
496 sem_ops.getnew = newary;
497 sem_ops.associate = sem_security;
498 sem_ops.more_checks = sem_more_checks;
500 sem_params.key = key;
501 sem_params.flg = semflg;
502 sem_params.u.nsems = nsems;
504 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
508 * Determine whether a sequence of semaphore operations would succeed
509 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
512 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
513 int nsops, struct sem_undo *un, int pid)
519 for (sop = sops; sop < sops + nsops; sop++) {
520 curr = sma->sem_base + sop->sem_num;
521 sem_op = sop->sem_op;
522 result = curr->semval;
524 if (!sem_op && result)
532 if (sop->sem_flg & SEM_UNDO) {
533 int undo = un->semadj[sop->sem_num] - sem_op;
535 * Exceeding the undo range is an error.
537 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
540 curr->semval = result;
544 while (sop >= sops) {
545 sma->sem_base[sop->sem_num].sempid = pid;
546 if (sop->sem_flg & SEM_UNDO)
547 un->semadj[sop->sem_num] -= sop->sem_op;
558 if (sop->sem_flg & IPC_NOWAIT)
565 while (sop >= sops) {
566 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
573 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
574 * @q: queue entry that must be signaled
575 * @error: Error value for the signal
577 * Prepare the wake-up of the queue entry q.
579 static void wake_up_sem_queue_prepare(struct list_head *pt,
580 struct sem_queue *q, int error)
582 if (list_empty(pt)) {
584 * Hold preempt off so that we don't get preempted and have the
585 * wakee busy-wait until we're scheduled back on.
589 q->status = IN_WAKEUP;
592 list_add_tail(&q->list, pt);
596 * wake_up_sem_queue_do(pt) - do the actual wake-up
597 * @pt: list of tasks to be woken up
599 * Do the actual wake-up.
600 * The function is called without any locks held, thus the semaphore array
601 * could be destroyed already and the tasks can disappear as soon as the
602 * status is set to the actual return code.
604 static void wake_up_sem_queue_do(struct list_head *pt)
606 struct sem_queue *q, *t;
609 did_something = !list_empty(pt);
610 list_for_each_entry_safe(q, t, pt, list) {
611 wake_up_process(q->sleeper);
612 /* q can disappear immediately after writing q->status. */
620 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
624 sma->complex_count--;
627 /** check_restart(sma, q)
628 * @sma: semaphore array
629 * @q: the operation that just completed
631 * update_queue is O(N^2) when it restarts scanning the whole queue of
632 * waiting operations. Therefore this function checks if the restart is
633 * really necessary. It is called after a previously waiting operation
636 static int check_restart(struct sem_array *sma, struct sem_queue *q)
641 /* if the operation didn't modify the array, then no restart */
645 /* pending complex operations are too difficult to analyse */
646 if (sma->complex_count)
649 /* we were a sleeping complex operation. Too difficult */
653 curr = sma->sem_base + q->sops[0].sem_num;
655 /* No-one waits on this queue */
656 if (list_empty(&curr->sem_pending))
659 /* the new semaphore value */
661 /* It is impossible that someone waits for the new value:
662 * - q is a previously sleeping simple operation that
663 * altered the array. It must be a decrement, because
664 * simple increments never sleep.
665 * - The value is not 0, thus wait-for-zero won't proceed.
666 * - If there are older (higher priority) decrements
667 * in the queue, then they have observed the original
668 * semval value and couldn't proceed. The operation
669 * decremented to value - thus they won't proceed either.
671 BUG_ON(q->sops[0].sem_op >= 0);
675 * semval is 0. Check if there are wait-for-zero semops.
676 * They must be the first entries in the per-semaphore queue
678 h = list_first_entry(&curr->sem_pending, struct sem_queue, list);
679 BUG_ON(h->nsops != 1);
680 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
682 /* Yes, there is a wait-for-zero semop. Restart */
683 if (h->sops[0].sem_op == 0)
686 /* Again - no-one is waiting for the new value. */
692 * update_queue(sma, semnum): Look for tasks that can be completed.
693 * @sma: semaphore array.
694 * @semnum: semaphore that was modified.
695 * @pt: list head for the tasks that must be woken up.
697 * update_queue must be called after a semaphore in a semaphore array
698 * was modified. If multiple semaphores were modified, update_queue must
699 * be called with semnum = -1, as well as with the number of each modified
701 * The tasks that must be woken up are added to @pt. The return code
702 * is stored in q->pid.
703 * The function return 1 if at least one semop was completed successfully.
705 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
708 struct list_head *walk;
709 struct list_head *pending_list;
710 int semop_completed = 0;
713 pending_list = &sma->sem_pending;
715 pending_list = &sma->sem_base[semnum].sem_pending;
718 walk = pending_list->next;
719 while (walk != pending_list) {
722 q = container_of(walk, struct sem_queue, list);
725 /* If we are scanning the single sop, per-semaphore list of
726 * one semaphore and that semaphore is 0, then it is not
727 * necessary to scan the "alter" entries: simple increments
728 * that affect only one entry succeed immediately and cannot
729 * be in the per semaphore pending queue, and decrements
730 * cannot be successful if the value is already 0.
732 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
736 error = try_atomic_semop(sma, q->sops, q->nsops,
739 /* Does q->sleeper still need to sleep? */
743 unlink_queue(sma, q);
749 restart = check_restart(sma, q);
752 wake_up_sem_queue_prepare(pt, q, error);
756 return semop_completed;
760 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
761 * @sma: semaphore array
762 * @sops: operations that were performed
763 * @nsops: number of operations
764 * @otime: force setting otime
765 * @pt: list head of the tasks that must be woken up.
767 * do_smart_update() does the required called to update_queue, based on the
768 * actual changes that were performed on the semaphore array.
769 * Note that the function does not do the actual wake-up: the caller is
770 * responsible for calling wake_up_sem_queue_do(@pt).
771 * It is safe to perform this call after dropping all locks.
773 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
774 int otime, struct list_head *pt)
778 if (sma->complex_count || sops == NULL) {
779 if (update_queue(sma, -1, pt))
784 /* No semops; something special is going on. */
785 for (i = 0; i < sma->sem_nsems; i++) {
786 if (update_queue(sma, i, pt))
792 /* Check the semaphores that were modified. */
793 for (i = 0; i < nsops; i++) {
794 if (sops[i].sem_op > 0 ||
795 (sops[i].sem_op < 0 &&
796 sma->sem_base[sops[i].sem_num].semval == 0))
797 if (update_queue(sma, sops[i].sem_num, pt))
802 sma->sem_otime = get_seconds();
806 /* The following counts are associated to each semaphore:
807 * semncnt number of tasks waiting on semval being nonzero
808 * semzcnt number of tasks waiting on semval being zero
809 * This model assumes that a task waits on exactly one semaphore.
810 * Since semaphore operations are to be performed atomically, tasks actually
811 * wait on a whole sequence of semaphores simultaneously.
812 * The counts we return here are a rough approximation, but still
813 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
815 static int count_semncnt (struct sem_array * sma, ushort semnum)
818 struct sem_queue * q;
821 list_for_each_entry(q, &sma->sem_pending, list) {
822 struct sembuf * sops = q->sops;
823 int nsops = q->nsops;
825 for (i = 0; i < nsops; i++)
826 if (sops[i].sem_num == semnum
827 && (sops[i].sem_op < 0)
828 && !(sops[i].sem_flg & IPC_NOWAIT))
834 static int count_semzcnt (struct sem_array * sma, ushort semnum)
837 struct sem_queue * q;
840 list_for_each_entry(q, &sma->sem_pending, list) {
841 struct sembuf * sops = q->sops;
842 int nsops = q->nsops;
844 for (i = 0; i < nsops; i++)
845 if (sops[i].sem_num == semnum
846 && (sops[i].sem_op == 0)
847 && !(sops[i].sem_flg & IPC_NOWAIT))
853 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
854 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
855 * remains locked on exit.
857 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
859 struct sem_undo *un, *tu;
860 struct sem_queue *q, *tq;
861 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
862 struct list_head tasks;
865 /* Free the existing undo structures for this semaphore set. */
866 assert_spin_locked(&sma->sem_perm.lock);
867 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
868 list_del(&un->list_id);
869 spin_lock(&un->ulp->lock);
871 list_del_rcu(&un->list_proc);
872 spin_unlock(&un->ulp->lock);
876 /* Wake up all pending processes and let them fail with EIDRM. */
877 INIT_LIST_HEAD(&tasks);
878 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
879 unlink_queue(sma, q);
880 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
882 for (i = 0; i < sma->sem_nsems; i++) {
883 struct sem *sem = sma->sem_base + i;
884 list_for_each_entry_safe(q, tq, &sem->sem_pending, list) {
885 unlink_queue(sma, q);
886 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
890 /* Remove the semaphore set from the IDR */
894 wake_up_sem_queue_do(&tasks);
895 ns->used_sems -= sma->sem_nsems;
896 security_sem_free(sma);
900 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
904 return copy_to_user(buf, in, sizeof(*in));
909 memset(&out, 0, sizeof(out));
911 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
913 out.sem_otime = in->sem_otime;
914 out.sem_ctime = in->sem_ctime;
915 out.sem_nsems = in->sem_nsems;
917 return copy_to_user(buf, &out, sizeof(out));
924 static int semctl_nolock(struct ipc_namespace *ns, int semid,
925 int cmd, int version, void __user *p)
928 struct sem_array *sma;
934 struct seminfo seminfo;
937 err = security_sem_semctl(NULL, cmd);
941 memset(&seminfo,0,sizeof(seminfo));
942 seminfo.semmni = ns->sc_semmni;
943 seminfo.semmns = ns->sc_semmns;
944 seminfo.semmsl = ns->sc_semmsl;
945 seminfo.semopm = ns->sc_semopm;
946 seminfo.semvmx = SEMVMX;
947 seminfo.semmnu = SEMMNU;
948 seminfo.semmap = SEMMAP;
949 seminfo.semume = SEMUME;
950 down_read(&sem_ids(ns).rw_mutex);
951 if (cmd == SEM_INFO) {
952 seminfo.semusz = sem_ids(ns).in_use;
953 seminfo.semaem = ns->used_sems;
955 seminfo.semusz = SEMUSZ;
956 seminfo.semaem = SEMAEM;
958 max_id = ipc_get_maxid(&sem_ids(ns));
959 up_read(&sem_ids(ns).rw_mutex);
960 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
962 return (max_id < 0) ? 0: max_id;
967 struct semid64_ds tbuf;
970 memset(&tbuf, 0, sizeof(tbuf));
972 if (cmd == SEM_STAT) {
974 sma = sem_obtain_object(ns, semid);
979 id = sma->sem_perm.id;
982 sma = sem_obtain_object_check(ns, semid);
990 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
993 err = security_sem_semctl(sma, cmd);
997 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
998 tbuf.sem_otime = sma->sem_otime;
999 tbuf.sem_ctime = sma->sem_ctime;
1000 tbuf.sem_nsems = sma->sem_nsems;
1002 if (copy_semid_to_user(p, &tbuf, version))
1014 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1017 struct sem_undo *un;
1018 struct sem_array *sma;
1021 struct list_head tasks;
1023 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1024 /* big-endian 64bit */
1027 /* 32bit or little-endian 64bit */
1031 if (val > SEMVMX || val < 0)
1034 INIT_LIST_HEAD(&tasks);
1037 sma = sem_obtain_object_check(ns, semid);
1040 return PTR_ERR(sma);
1043 if (semnum < 0 || semnum >= sma->sem_nsems) {
1049 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1054 err = security_sem_semctl(sma, SETVAL);
1060 sem_lock(sma, NULL, -1);
1062 curr = &sma->sem_base[semnum];
1064 assert_spin_locked(&sma->sem_perm.lock);
1065 list_for_each_entry(un, &sma->list_id, list_id)
1066 un->semadj[semnum] = 0;
1069 curr->sempid = task_tgid_vnr(current);
1070 sma->sem_ctime = get_seconds();
1071 /* maybe some queued-up processes were waiting for this */
1072 do_smart_update(sma, NULL, 0, 0, &tasks);
1073 sem_unlock(sma, -1);
1074 wake_up_sem_queue_do(&tasks);
1078 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1079 int cmd, void __user *p)
1081 struct sem_array *sma;
1084 ushort fast_sem_io[SEMMSL_FAST];
1085 ushort* sem_io = fast_sem_io;
1086 struct list_head tasks;
1088 INIT_LIST_HEAD(&tasks);
1091 sma = sem_obtain_object_check(ns, semid);
1094 return PTR_ERR(sma);
1097 nsems = sma->sem_nsems;
1100 if (ipcperms(ns, &sma->sem_perm,
1101 cmd == SETALL ? S_IWUGO : S_IRUGO)) {
1106 err = security_sem_semctl(sma, cmd);
1116 ushort __user *array = p;
1119 if(nsems > SEMMSL_FAST) {
1122 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1123 if(sem_io == NULL) {
1128 sem_lock_and_putref(sma);
1129 if (sma->sem_perm.deleted) {
1130 sem_unlock(sma, -1);
1135 sem_lock(sma, NULL, -1);
1137 for (i = 0; i < sma->sem_nsems; i++)
1138 sem_io[i] = sma->sem_base[i].semval;
1139 sem_unlock(sma, -1);
1141 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1148 struct sem_undo *un;
1150 if (!ipc_rcu_getref(sma)) {
1156 if(nsems > SEMMSL_FAST) {
1157 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1158 if(sem_io == NULL) {
1164 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1170 for (i = 0; i < nsems; i++) {
1171 if (sem_io[i] > SEMVMX) {
1177 sem_lock_and_putref(sma);
1178 if (sma->sem_perm.deleted) {
1179 sem_unlock(sma, -1);
1184 for (i = 0; i < nsems; i++)
1185 sma->sem_base[i].semval = sem_io[i];
1187 assert_spin_locked(&sma->sem_perm.lock);
1188 list_for_each_entry(un, &sma->list_id, list_id) {
1189 for (i = 0; i < nsems; i++)
1192 sma->sem_ctime = get_seconds();
1193 /* maybe some queued-up processes were waiting for this */
1194 do_smart_update(sma, NULL, 0, 0, &tasks);
1198 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1201 if (semnum < 0 || semnum >= nsems) {
1206 sem_lock(sma, NULL, -1);
1207 curr = &sma->sem_base[semnum];
1217 err = count_semncnt(sma,semnum);
1220 err = count_semzcnt(sma,semnum);
1225 sem_unlock(sma, -1);
1227 wake_up_sem_queue_do(&tasks);
1229 if(sem_io != fast_sem_io)
1230 ipc_free(sem_io, sizeof(ushort)*nsems);
1234 static inline unsigned long
1235 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1239 if (copy_from_user(out, buf, sizeof(*out)))
1244 struct semid_ds tbuf_old;
1246 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1249 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1250 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1251 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1261 * This function handles some semctl commands which require the rw_mutex
1262 * to be held in write mode.
1263 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1265 static int semctl_down(struct ipc_namespace *ns, int semid,
1266 int cmd, int version, void __user *p)
1268 struct sem_array *sma;
1270 struct semid64_ds semid64;
1271 struct kern_ipc_perm *ipcp;
1273 if(cmd == IPC_SET) {
1274 if (copy_semid_from_user(&semid64, p, version))
1278 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1279 &semid64.sem_perm, 0);
1281 return PTR_ERR(ipcp);
1283 sma = container_of(ipcp, struct sem_array, sem_perm);
1285 err = security_sem_semctl(sma, cmd);
1293 sem_lock(sma, NULL, -1);
1297 sem_lock(sma, NULL, -1);
1298 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1301 sma->sem_ctime = get_seconds();
1310 sem_unlock(sma, -1);
1312 up_write(&sem_ids(ns).rw_mutex);
1316 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1319 struct ipc_namespace *ns;
1320 void __user *p = (void __user *)arg;
1325 version = ipc_parse_version(&cmd);
1326 ns = current->nsproxy->ipc_ns;
1333 return semctl_nolock(ns, semid, cmd, version, p);
1340 return semctl_main(ns, semid, semnum, cmd, p);
1342 return semctl_setval(ns, semid, semnum, arg);
1345 return semctl_down(ns, semid, cmd, version, p);
1351 /* If the task doesn't already have a undo_list, then allocate one
1352 * here. We guarantee there is only one thread using this undo list,
1353 * and current is THE ONE
1355 * If this allocation and assignment succeeds, but later
1356 * portions of this code fail, there is no need to free the sem_undo_list.
1357 * Just let it stay associated with the task, and it'll be freed later
1360 * This can block, so callers must hold no locks.
1362 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1364 struct sem_undo_list *undo_list;
1366 undo_list = current->sysvsem.undo_list;
1368 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1369 if (undo_list == NULL)
1371 spin_lock_init(&undo_list->lock);
1372 atomic_set(&undo_list->refcnt, 1);
1373 INIT_LIST_HEAD(&undo_list->list_proc);
1375 current->sysvsem.undo_list = undo_list;
1377 *undo_listp = undo_list;
1381 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1383 struct sem_undo *un;
1385 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1386 if (un->semid == semid)
1392 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1394 struct sem_undo *un;
1396 assert_spin_locked(&ulp->lock);
1398 un = __lookup_undo(ulp, semid);
1400 list_del_rcu(&un->list_proc);
1401 list_add_rcu(&un->list_proc, &ulp->list_proc);
1407 * find_alloc_undo - Lookup (and if not present create) undo array
1409 * @semid: semaphore array id
1411 * The function looks up (and if not present creates) the undo structure.
1412 * The size of the undo structure depends on the size of the semaphore
1413 * array, thus the alloc path is not that straightforward.
1414 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1415 * performs a rcu_read_lock().
1417 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1419 struct sem_array *sma;
1420 struct sem_undo_list *ulp;
1421 struct sem_undo *un, *new;
1424 error = get_undo_list(&ulp);
1426 return ERR_PTR(error);
1429 spin_lock(&ulp->lock);
1430 un = lookup_undo(ulp, semid);
1431 spin_unlock(&ulp->lock);
1432 if (likely(un!=NULL))
1435 /* no undo structure around - allocate one. */
1436 /* step 1: figure out the size of the semaphore array */
1437 sma = sem_obtain_object_check(ns, semid);
1440 return ERR_CAST(sma);
1443 nsems = sma->sem_nsems;
1444 if (!ipc_rcu_getref(sma)) {
1446 un = ERR_PTR(-EIDRM);
1451 /* step 2: allocate new undo structure */
1452 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1455 return ERR_PTR(-ENOMEM);
1458 /* step 3: Acquire the lock on semaphore array */
1459 sem_lock_and_putref(sma);
1460 if (sma->sem_perm.deleted) {
1461 sem_unlock(sma, -1);
1463 un = ERR_PTR(-EIDRM);
1466 spin_lock(&ulp->lock);
1469 * step 4: check for races: did someone else allocate the undo struct?
1471 un = lookup_undo(ulp, semid);
1476 /* step 5: initialize & link new undo structure */
1477 new->semadj = (short *) &new[1];
1480 assert_spin_locked(&ulp->lock);
1481 list_add_rcu(&new->list_proc, &ulp->list_proc);
1482 assert_spin_locked(&sma->sem_perm.lock);
1483 list_add(&new->list_id, &sma->list_id);
1487 spin_unlock(&ulp->lock);
1489 sem_unlock(sma, -1);
1496 * get_queue_result - Retrieve the result code from sem_queue
1497 * @q: Pointer to queue structure
1499 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1500 * q->status, then we must loop until the value is replaced with the final
1501 * value: This may happen if a task is woken up by an unrelated event (e.g.
1502 * signal) and in parallel the task is woken up by another task because it got
1503 * the requested semaphores.
1505 * The function can be called with or without holding the semaphore spinlock.
1507 static int get_queue_result(struct sem_queue *q)
1512 while (unlikely(error == IN_WAKEUP)) {
1521 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1522 unsigned, nsops, const struct timespec __user *, timeout)
1524 int error = -EINVAL;
1525 struct sem_array *sma;
1526 struct sembuf fast_sops[SEMOPM_FAST];
1527 struct sembuf* sops = fast_sops, *sop;
1528 struct sem_undo *un;
1529 int undos = 0, alter = 0, max, locknum;
1530 struct sem_queue queue;
1531 unsigned long jiffies_left = 0;
1532 struct ipc_namespace *ns;
1533 struct list_head tasks;
1535 ns = current->nsproxy->ipc_ns;
1537 if (nsops < 1 || semid < 0)
1539 if (nsops > ns->sc_semopm)
1541 if(nsops > SEMOPM_FAST) {
1542 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1546 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1551 struct timespec _timeout;
1552 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1556 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1557 _timeout.tv_nsec >= 1000000000L) {
1561 jiffies_left = timespec_to_jiffies(&_timeout);
1564 for (sop = sops; sop < sops + nsops; sop++) {
1565 if (sop->sem_num >= max)
1567 if (sop->sem_flg & SEM_UNDO)
1569 if (sop->sem_op != 0)
1573 INIT_LIST_HEAD(&tasks);
1576 /* On success, find_alloc_undo takes the rcu_read_lock */
1577 un = find_alloc_undo(ns, semid);
1579 error = PTR_ERR(un);
1587 sma = sem_obtain_object_check(ns, semid);
1590 error = PTR_ERR(sma);
1595 if (max >= sma->sem_nsems) {
1601 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1606 error = security_sem_semop(sma, sops, nsops, alter);
1613 * semid identifiers are not unique - find_alloc_undo may have
1614 * allocated an undo structure, it was invalidated by an RMID
1615 * and now a new array with received the same id. Check and fail.
1616 * This case can be detected checking un->semid. The existence of
1617 * "un" itself is guaranteed by rcu.
1620 locknum = sem_lock(sma, sops, nsops);
1621 if (un && un->semid == -1)
1622 goto out_unlock_free;
1624 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1626 if (alter && error == 0)
1627 do_smart_update(sma, sops, nsops, 1, &tasks);
1629 goto out_unlock_free;
1632 /* We need to sleep on this operation, so we put the current
1633 * task into the pending queue and go to sleep.
1637 queue.nsops = nsops;
1639 queue.pid = task_tgid_vnr(current);
1640 queue.alter = alter;
1644 curr = &sma->sem_base[sops->sem_num];
1647 list_add_tail(&queue.list, &curr->sem_pending);
1649 list_add(&queue.list, &curr->sem_pending);
1652 list_add_tail(&queue.list, &sma->sem_pending);
1654 list_add(&queue.list, &sma->sem_pending);
1655 sma->complex_count++;
1658 queue.status = -EINTR;
1659 queue.sleeper = current;
1662 current->state = TASK_INTERRUPTIBLE;
1663 sem_unlock(sma, locknum);
1666 jiffies_left = schedule_timeout(jiffies_left);
1670 error = get_queue_result(&queue);
1672 if (error != -EINTR) {
1673 /* fast path: update_queue already obtained all requested
1675 * Perform a smp_mb(): User space could assume that semop()
1676 * is a memory barrier: Without the mb(), the cpu could
1677 * speculatively read in user space stale data that was
1678 * overwritten by the previous owner of the semaphore.
1685 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1688 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1690 error = get_queue_result(&queue);
1693 * Array removed? If yes, leave without sem_unlock().
1701 * If queue.status != -EINTR we are woken up by another process.
1702 * Leave without unlink_queue(), but with sem_unlock().
1705 if (error != -EINTR) {
1706 goto out_unlock_free;
1710 * If an interrupt occurred we have to clean up the queue
1712 if (timeout && jiffies_left == 0)
1716 * If the wakeup was spurious, just retry
1718 if (error == -EINTR && !signal_pending(current))
1721 unlink_queue(sma, &queue);
1724 sem_unlock(sma, locknum);
1726 wake_up_sem_queue_do(&tasks);
1728 if(sops != fast_sops)
1733 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1736 return sys_semtimedop(semid, tsops, nsops, NULL);
1739 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1740 * parent and child tasks.
1743 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1745 struct sem_undo_list *undo_list;
1748 if (clone_flags & CLONE_SYSVSEM) {
1749 error = get_undo_list(&undo_list);
1752 atomic_inc(&undo_list->refcnt);
1753 tsk->sysvsem.undo_list = undo_list;
1755 tsk->sysvsem.undo_list = NULL;
1761 * add semadj values to semaphores, free undo structures.
1762 * undo structures are not freed when semaphore arrays are destroyed
1763 * so some of them may be out of date.
1764 * IMPLEMENTATION NOTE: There is some confusion over whether the
1765 * set of adjustments that needs to be done should be done in an atomic
1766 * manner or not. That is, if we are attempting to decrement the semval
1767 * should we queue up and wait until we can do so legally?
1768 * The original implementation attempted to do this (queue and wait).
1769 * The current implementation does not do so. The POSIX standard
1770 * and SVID should be consulted to determine what behavior is mandated.
1772 void exit_sem(struct task_struct *tsk)
1774 struct sem_undo_list *ulp;
1776 ulp = tsk->sysvsem.undo_list;
1779 tsk->sysvsem.undo_list = NULL;
1781 if (!atomic_dec_and_test(&ulp->refcnt))
1785 struct sem_array *sma;
1786 struct sem_undo *un;
1787 struct list_head tasks;
1791 un = list_entry_rcu(ulp->list_proc.next,
1792 struct sem_undo, list_proc);
1793 if (&un->list_proc == &ulp->list_proc)
1803 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
1804 /* exit_sem raced with IPC_RMID, nothing to do */
1810 sem_lock(sma, NULL, -1);
1811 un = __lookup_undo(ulp, semid);
1813 /* exit_sem raced with IPC_RMID+semget() that created
1814 * exactly the same semid. Nothing to do.
1816 sem_unlock(sma, -1);
1820 /* remove un from the linked lists */
1821 assert_spin_locked(&sma->sem_perm.lock);
1822 list_del(&un->list_id);
1824 spin_lock(&ulp->lock);
1825 list_del_rcu(&un->list_proc);
1826 spin_unlock(&ulp->lock);
1828 /* perform adjustments registered in un */
1829 for (i = 0; i < sma->sem_nsems; i++) {
1830 struct sem * semaphore = &sma->sem_base[i];
1831 if (un->semadj[i]) {
1832 semaphore->semval += un->semadj[i];
1834 * Range checks of the new semaphore value,
1835 * not defined by sus:
1836 * - Some unices ignore the undo entirely
1837 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1838 * - some cap the value (e.g. FreeBSD caps
1839 * at 0, but doesn't enforce SEMVMX)
1841 * Linux caps the semaphore value, both at 0
1844 * Manfred <manfred@colorfullife.com>
1846 if (semaphore->semval < 0)
1847 semaphore->semval = 0;
1848 if (semaphore->semval > SEMVMX)
1849 semaphore->semval = SEMVMX;
1850 semaphore->sempid = task_tgid_vnr(current);
1853 /* maybe some queued-up processes were waiting for this */
1854 INIT_LIST_HEAD(&tasks);
1855 do_smart_update(sma, NULL, 0, 1, &tasks);
1856 sem_unlock(sma, -1);
1857 wake_up_sem_queue_do(&tasks);
1864 #ifdef CONFIG_PROC_FS
1865 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1867 struct user_namespace *user_ns = seq_user_ns(s);
1868 struct sem_array *sma = it;
1870 return seq_printf(s,
1871 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1876 from_kuid_munged(user_ns, sma->sem_perm.uid),
1877 from_kgid_munged(user_ns, sma->sem_perm.gid),
1878 from_kuid_munged(user_ns, sma->sem_perm.cuid),
1879 from_kgid_munged(user_ns, sma->sem_perm.cgid),