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);
270 * sem_lock_(check_) routines are called in the paths where the rw_mutex
273 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
274 int id, struct sembuf *sops, int nsops, int *locknum)
276 struct kern_ipc_perm *ipcp;
277 struct sem_array *sma;
280 ipcp = ipc_obtain_object(&sem_ids(ns), id);
282 sma = ERR_CAST(ipcp);
286 sma = container_of(ipcp, struct sem_array, sem_perm);
287 *locknum = sem_lock(sma, sops, nsops);
289 /* ipc_rmid() may have already freed the ID while sem_lock
290 * was spinning: verify that the structure is still valid
293 return container_of(ipcp, struct sem_array, sem_perm);
295 sem_unlock(sma, *locknum);
296 sma = ERR_PTR(-EINVAL);
302 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
304 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
307 return ERR_CAST(ipcp);
309 return container_of(ipcp, struct sem_array, sem_perm);
312 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
315 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
318 return ERR_CAST(ipcp);
320 return container_of(ipcp, struct sem_array, sem_perm);
323 static inline void sem_lock_and_putref(struct sem_array *sma)
325 sem_lock(sma, NULL, -1);
329 static inline void sem_putref(struct sem_array *sma)
334 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
336 ipc_rmid(&sem_ids(ns), &s->sem_perm);
340 * Lockless wakeup algorithm:
341 * Without the check/retry algorithm a lockless wakeup is possible:
342 * - queue.status is initialized to -EINTR before blocking.
343 * - wakeup is performed by
344 * * unlinking the queue entry from sma->sem_pending
345 * * setting queue.status to IN_WAKEUP
346 * This is the notification for the blocked thread that a
347 * result value is imminent.
348 * * call wake_up_process
349 * * set queue.status to the final value.
350 * - the previously blocked thread checks queue.status:
351 * * if it's IN_WAKEUP, then it must wait until the value changes
352 * * if it's not -EINTR, then the operation was completed by
353 * update_queue. semtimedop can return queue.status without
354 * performing any operation on the sem array.
355 * * otherwise it must acquire the spinlock and check what's up.
357 * The two-stage algorithm is necessary to protect against the following
359 * - if queue.status is set after wake_up_process, then the woken up idle
360 * thread could race forward and try (and fail) to acquire sma->lock
361 * before update_queue had a chance to set queue.status
362 * - if queue.status is written before wake_up_process and if the
363 * blocked process is woken up by a signal between writing
364 * queue.status and the wake_up_process, then the woken up
365 * process could return from semtimedop and die by calling
366 * sys_exit before wake_up_process is called. Then wake_up_process
367 * will oops, because the task structure is already invalid.
368 * (yes, this happened on s390 with sysv msg).
374 * newary - Create a new semaphore set
376 * @params: ptr to the structure that contains key, semflg and nsems
378 * Called with sem_ids.rw_mutex held (as a writer)
381 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
385 struct sem_array *sma;
387 key_t key = params->key;
388 int nsems = params->u.nsems;
389 int semflg = params->flg;
394 if (ns->used_sems + nsems > ns->sc_semmns)
397 size = sizeof (*sma) + nsems * sizeof (struct sem);
398 sma = ipc_rcu_alloc(size);
402 memset (sma, 0, size);
404 sma->sem_perm.mode = (semflg & S_IRWXUGO);
405 sma->sem_perm.key = key;
407 sma->sem_perm.security = NULL;
408 retval = security_sem_alloc(sma);
414 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
416 security_sem_free(sma);
420 ns->used_sems += nsems;
422 sma->sem_base = (struct sem *) &sma[1];
424 for (i = 0; i < nsems; i++) {
425 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
426 spin_lock_init(&sma->sem_base[i].lock);
429 sma->complex_count = 0;
430 INIT_LIST_HEAD(&sma->sem_pending);
431 INIT_LIST_HEAD(&sma->list_id);
432 sma->sem_nsems = nsems;
433 sma->sem_ctime = get_seconds();
437 return sma->sem_perm.id;
442 * Called with sem_ids.rw_mutex and ipcp locked.
444 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
446 struct sem_array *sma;
448 sma = container_of(ipcp, struct sem_array, sem_perm);
449 return security_sem_associate(sma, semflg);
453 * Called with sem_ids.rw_mutex and ipcp locked.
455 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
456 struct ipc_params *params)
458 struct sem_array *sma;
460 sma = container_of(ipcp, struct sem_array, sem_perm);
461 if (params->u.nsems > sma->sem_nsems)
467 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
469 struct ipc_namespace *ns;
470 struct ipc_ops sem_ops;
471 struct ipc_params sem_params;
473 ns = current->nsproxy->ipc_ns;
475 if (nsems < 0 || nsems > ns->sc_semmsl)
478 sem_ops.getnew = newary;
479 sem_ops.associate = sem_security;
480 sem_ops.more_checks = sem_more_checks;
482 sem_params.key = key;
483 sem_params.flg = semflg;
484 sem_params.u.nsems = nsems;
486 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
490 * Determine whether a sequence of semaphore operations would succeed
491 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
494 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
495 int nsops, struct sem_undo *un, int pid)
501 for (sop = sops; sop < sops + nsops; sop++) {
502 curr = sma->sem_base + sop->sem_num;
503 sem_op = sop->sem_op;
504 result = curr->semval;
506 if (!sem_op && result)
514 if (sop->sem_flg & SEM_UNDO) {
515 int undo = un->semadj[sop->sem_num] - sem_op;
517 * Exceeding the undo range is an error.
519 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
522 curr->semval = result;
526 while (sop >= sops) {
527 sma->sem_base[sop->sem_num].sempid = pid;
528 if (sop->sem_flg & SEM_UNDO)
529 un->semadj[sop->sem_num] -= sop->sem_op;
540 if (sop->sem_flg & IPC_NOWAIT)
547 while (sop >= sops) {
548 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
555 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
556 * @q: queue entry that must be signaled
557 * @error: Error value for the signal
559 * Prepare the wake-up of the queue entry q.
561 static void wake_up_sem_queue_prepare(struct list_head *pt,
562 struct sem_queue *q, int error)
564 if (list_empty(pt)) {
566 * Hold preempt off so that we don't get preempted and have the
567 * wakee busy-wait until we're scheduled back on.
571 q->status = IN_WAKEUP;
574 list_add_tail(&q->list, pt);
578 * wake_up_sem_queue_do(pt) - do the actual wake-up
579 * @pt: list of tasks to be woken up
581 * Do the actual wake-up.
582 * The function is called without any locks held, thus the semaphore array
583 * could be destroyed already and the tasks can disappear as soon as the
584 * status is set to the actual return code.
586 static void wake_up_sem_queue_do(struct list_head *pt)
588 struct sem_queue *q, *t;
591 did_something = !list_empty(pt);
592 list_for_each_entry_safe(q, t, pt, list) {
593 wake_up_process(q->sleeper);
594 /* q can disappear immediately after writing q->status. */
602 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
606 sma->complex_count--;
609 /** check_restart(sma, q)
610 * @sma: semaphore array
611 * @q: the operation that just completed
613 * update_queue is O(N^2) when it restarts scanning the whole queue of
614 * waiting operations. Therefore this function checks if the restart is
615 * really necessary. It is called after a previously waiting operation
618 static int check_restart(struct sem_array *sma, struct sem_queue *q)
623 /* if the operation didn't modify the array, then no restart */
627 /* pending complex operations are too difficult to analyse */
628 if (sma->complex_count)
631 /* we were a sleeping complex operation. Too difficult */
635 curr = sma->sem_base + q->sops[0].sem_num;
637 /* No-one waits on this queue */
638 if (list_empty(&curr->sem_pending))
641 /* the new semaphore value */
643 /* It is impossible that someone waits for the new value:
644 * - q is a previously sleeping simple operation that
645 * altered the array. It must be a decrement, because
646 * simple increments never sleep.
647 * - The value is not 0, thus wait-for-zero won't proceed.
648 * - If there are older (higher priority) decrements
649 * in the queue, then they have observed the original
650 * semval value and couldn't proceed. The operation
651 * decremented to value - thus they won't proceed either.
653 BUG_ON(q->sops[0].sem_op >= 0);
657 * semval is 0. Check if there are wait-for-zero semops.
658 * They must be the first entries in the per-semaphore queue
660 h = list_first_entry(&curr->sem_pending, struct sem_queue, list);
661 BUG_ON(h->nsops != 1);
662 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
664 /* Yes, there is a wait-for-zero semop. Restart */
665 if (h->sops[0].sem_op == 0)
668 /* Again - no-one is waiting for the new value. */
674 * update_queue(sma, semnum): Look for tasks that can be completed.
675 * @sma: semaphore array.
676 * @semnum: semaphore that was modified.
677 * @pt: list head for the tasks that must be woken up.
679 * update_queue must be called after a semaphore in a semaphore array
680 * was modified. If multiple semaphores were modified, update_queue must
681 * be called with semnum = -1, as well as with the number of each modified
683 * The tasks that must be woken up are added to @pt. The return code
684 * is stored in q->pid.
685 * The function return 1 if at least one semop was completed successfully.
687 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
690 struct list_head *walk;
691 struct list_head *pending_list;
692 int semop_completed = 0;
695 pending_list = &sma->sem_pending;
697 pending_list = &sma->sem_base[semnum].sem_pending;
700 walk = pending_list->next;
701 while (walk != pending_list) {
704 q = container_of(walk, struct sem_queue, list);
707 /* If we are scanning the single sop, per-semaphore list of
708 * one semaphore and that semaphore is 0, then it is not
709 * necessary to scan the "alter" entries: simple increments
710 * that affect only one entry succeed immediately and cannot
711 * be in the per semaphore pending queue, and decrements
712 * cannot be successful if the value is already 0.
714 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
718 error = try_atomic_semop(sma, q->sops, q->nsops,
721 /* Does q->sleeper still need to sleep? */
725 unlink_queue(sma, q);
731 restart = check_restart(sma, q);
734 wake_up_sem_queue_prepare(pt, q, error);
738 return semop_completed;
742 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
743 * @sma: semaphore array
744 * @sops: operations that were performed
745 * @nsops: number of operations
746 * @otime: force setting otime
747 * @pt: list head of the tasks that must be woken up.
749 * do_smart_update() does the required called to update_queue, based on the
750 * actual changes that were performed on the semaphore array.
751 * Note that the function does not do the actual wake-up: the caller is
752 * responsible for calling wake_up_sem_queue_do(@pt).
753 * It is safe to perform this call after dropping all locks.
755 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
756 int otime, struct list_head *pt)
760 if (sma->complex_count || sops == NULL) {
761 if (update_queue(sma, -1, pt))
766 /* No semops; something special is going on. */
767 for (i = 0; i < sma->sem_nsems; i++) {
768 if (update_queue(sma, i, pt))
774 /* Check the semaphores that were modified. */
775 for (i = 0; i < nsops; i++) {
776 if (sops[i].sem_op > 0 ||
777 (sops[i].sem_op < 0 &&
778 sma->sem_base[sops[i].sem_num].semval == 0))
779 if (update_queue(sma, sops[i].sem_num, pt))
784 sma->sem_otime = get_seconds();
788 /* The following counts are associated to each semaphore:
789 * semncnt number of tasks waiting on semval being nonzero
790 * semzcnt number of tasks waiting on semval being zero
791 * This model assumes that a task waits on exactly one semaphore.
792 * Since semaphore operations are to be performed atomically, tasks actually
793 * wait on a whole sequence of semaphores simultaneously.
794 * The counts we return here are a rough approximation, but still
795 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
797 static int count_semncnt (struct sem_array * sma, ushort semnum)
800 struct sem_queue * q;
803 list_for_each_entry(q, &sma->sem_pending, list) {
804 struct sembuf * sops = q->sops;
805 int nsops = q->nsops;
807 for (i = 0; i < nsops; i++)
808 if (sops[i].sem_num == semnum
809 && (sops[i].sem_op < 0)
810 && !(sops[i].sem_flg & IPC_NOWAIT))
816 static int count_semzcnt (struct sem_array * sma, ushort semnum)
819 struct sem_queue * q;
822 list_for_each_entry(q, &sma->sem_pending, list) {
823 struct sembuf * sops = q->sops;
824 int nsops = q->nsops;
826 for (i = 0; i < nsops; i++)
827 if (sops[i].sem_num == semnum
828 && (sops[i].sem_op == 0)
829 && !(sops[i].sem_flg & IPC_NOWAIT))
835 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
836 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
837 * remains locked on exit.
839 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
841 struct sem_undo *un, *tu;
842 struct sem_queue *q, *tq;
843 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
844 struct list_head tasks;
847 /* Free the existing undo structures for this semaphore set. */
848 assert_spin_locked(&sma->sem_perm.lock);
849 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
850 list_del(&un->list_id);
851 spin_lock(&un->ulp->lock);
853 list_del_rcu(&un->list_proc);
854 spin_unlock(&un->ulp->lock);
858 /* Wake up all pending processes and let them fail with EIDRM. */
859 INIT_LIST_HEAD(&tasks);
860 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
861 unlink_queue(sma, q);
862 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
864 for (i = 0; i < sma->sem_nsems; i++) {
865 struct sem *sem = sma->sem_base + i;
866 list_for_each_entry_safe(q, tq, &sem->sem_pending, list) {
867 unlink_queue(sma, q);
868 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
872 /* Remove the semaphore set from the IDR */
877 wake_up_sem_queue_do(&tasks);
878 ns->used_sems -= sma->sem_nsems;
879 security_sem_free(sma);
883 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
887 return copy_to_user(buf, in, sizeof(*in));
892 memset(&out, 0, sizeof(out));
894 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
896 out.sem_otime = in->sem_otime;
897 out.sem_ctime = in->sem_ctime;
898 out.sem_nsems = in->sem_nsems;
900 return copy_to_user(buf, &out, sizeof(out));
907 static int semctl_nolock(struct ipc_namespace *ns, int semid,
908 int cmd, int version, void __user *p)
911 struct sem_array *sma;
917 struct seminfo seminfo;
920 err = security_sem_semctl(NULL, cmd);
924 memset(&seminfo,0,sizeof(seminfo));
925 seminfo.semmni = ns->sc_semmni;
926 seminfo.semmns = ns->sc_semmns;
927 seminfo.semmsl = ns->sc_semmsl;
928 seminfo.semopm = ns->sc_semopm;
929 seminfo.semvmx = SEMVMX;
930 seminfo.semmnu = SEMMNU;
931 seminfo.semmap = SEMMAP;
932 seminfo.semume = SEMUME;
933 down_read(&sem_ids(ns).rw_mutex);
934 if (cmd == SEM_INFO) {
935 seminfo.semusz = sem_ids(ns).in_use;
936 seminfo.semaem = ns->used_sems;
938 seminfo.semusz = SEMUSZ;
939 seminfo.semaem = SEMAEM;
941 max_id = ipc_get_maxid(&sem_ids(ns));
942 up_read(&sem_ids(ns).rw_mutex);
943 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
945 return (max_id < 0) ? 0: max_id;
950 struct semid64_ds tbuf;
953 memset(&tbuf, 0, sizeof(tbuf));
955 if (cmd == SEM_STAT) {
957 sma = sem_obtain_object(ns, semid);
962 id = sma->sem_perm.id;
965 sma = sem_obtain_object_check(ns, semid);
973 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
976 err = security_sem_semctl(sma, cmd);
980 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
981 tbuf.sem_otime = sma->sem_otime;
982 tbuf.sem_ctime = sma->sem_ctime;
983 tbuf.sem_nsems = sma->sem_nsems;
985 if (copy_semid_to_user(p, &tbuf, version))
997 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1000 struct sem_undo *un;
1001 struct sem_array *sma;
1004 struct list_head tasks;
1006 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1007 /* big-endian 64bit */
1010 /* 32bit or little-endian 64bit */
1014 if (val > SEMVMX || val < 0)
1017 INIT_LIST_HEAD(&tasks);
1020 sma = sem_obtain_object_check(ns, semid);
1023 return PTR_ERR(sma);
1026 if (semnum < 0 || semnum >= sma->sem_nsems) {
1032 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1037 err = security_sem_semctl(sma, SETVAL);
1043 sem_lock(sma, NULL, -1);
1045 curr = &sma->sem_base[semnum];
1047 assert_spin_locked(&sma->sem_perm.lock);
1048 list_for_each_entry(un, &sma->list_id, list_id)
1049 un->semadj[semnum] = 0;
1052 curr->sempid = task_tgid_vnr(current);
1053 sma->sem_ctime = get_seconds();
1054 /* maybe some queued-up processes were waiting for this */
1055 do_smart_update(sma, NULL, 0, 0, &tasks);
1056 sem_unlock(sma, -1);
1058 wake_up_sem_queue_do(&tasks);
1062 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1063 int cmd, void __user *p)
1065 struct sem_array *sma;
1068 ushort fast_sem_io[SEMMSL_FAST];
1069 ushort* sem_io = fast_sem_io;
1070 struct list_head tasks;
1072 INIT_LIST_HEAD(&tasks);
1075 sma = sem_obtain_object_check(ns, semid);
1078 return PTR_ERR(sma);
1081 nsems = sma->sem_nsems;
1084 if (ipcperms(ns, &sma->sem_perm,
1085 cmd == SETALL ? S_IWUGO : S_IRUGO)) {
1090 err = security_sem_semctl(sma, cmd);
1100 ushort __user *array = p;
1103 sem_lock(sma, NULL, -1);
1104 if(nsems > SEMMSL_FAST) {
1105 if (!ipc_rcu_getref(sma)) {
1106 sem_unlock(sma, -1);
1111 sem_unlock(sma, -1);
1113 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1114 if(sem_io == NULL) {
1120 sem_lock_and_putref(sma);
1121 if (sma->sem_perm.deleted) {
1122 sem_unlock(sma, -1);
1128 for (i = 0; i < sma->sem_nsems; i++)
1129 sem_io[i] = sma->sem_base[i].semval;
1130 sem_unlock(sma, -1);
1133 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1140 struct sem_undo *un;
1142 if (!ipc_rcu_getref(sma)) {
1148 if(nsems > SEMMSL_FAST) {
1149 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1150 if(sem_io == NULL) {
1156 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1162 for (i = 0; i < nsems; i++) {
1163 if (sem_io[i] > SEMVMX) {
1170 sem_lock_and_putref(sma);
1171 if (sma->sem_perm.deleted) {
1172 sem_unlock(sma, -1);
1178 for (i = 0; i < nsems; i++)
1179 sma->sem_base[i].semval = sem_io[i];
1181 assert_spin_locked(&sma->sem_perm.lock);
1182 list_for_each_entry(un, &sma->list_id, list_id) {
1183 for (i = 0; i < nsems; i++)
1186 sma->sem_ctime = get_seconds();
1187 /* maybe some queued-up processes were waiting for this */
1188 do_smart_update(sma, NULL, 0, 0, &tasks);
1192 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1195 if (semnum < 0 || semnum >= nsems) {
1200 sem_lock(sma, NULL, -1);
1201 curr = &sma->sem_base[semnum];
1211 err = count_semncnt(sma,semnum);
1214 err = count_semzcnt(sma,semnum);
1219 sem_unlock(sma, -1);
1222 wake_up_sem_queue_do(&tasks);
1224 if(sem_io != fast_sem_io)
1225 ipc_free(sem_io, sizeof(ushort)*nsems);
1229 static inline unsigned long
1230 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1234 if (copy_from_user(out, buf, sizeof(*out)))
1239 struct semid_ds tbuf_old;
1241 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1244 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1245 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1246 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1256 * This function handles some semctl commands which require the rw_mutex
1257 * to be held in write mode.
1258 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1260 static int semctl_down(struct ipc_namespace *ns, int semid,
1261 int cmd, int version, void __user *p)
1263 struct sem_array *sma;
1265 struct semid64_ds semid64;
1266 struct kern_ipc_perm *ipcp;
1268 if(cmd == IPC_SET) {
1269 if (copy_semid_from_user(&semid64, p, version))
1273 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1274 &semid64.sem_perm, 0);
1276 return PTR_ERR(ipcp);
1278 sma = container_of(ipcp, struct sem_array, sem_perm);
1280 err = security_sem_semctl(sma, cmd);
1288 sem_lock(sma, NULL, -1);
1292 sem_lock(sma, NULL, -1);
1293 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1296 sma->sem_ctime = get_seconds();
1305 sem_unlock(sma, -1);
1308 up_write(&sem_ids(ns).rw_mutex);
1312 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1315 struct ipc_namespace *ns;
1316 void __user *p = (void __user *)arg;
1321 version = ipc_parse_version(&cmd);
1322 ns = current->nsproxy->ipc_ns;
1329 return semctl_nolock(ns, semid, cmd, version, p);
1336 return semctl_main(ns, semid, semnum, cmd, p);
1338 return semctl_setval(ns, semid, semnum, arg);
1341 return semctl_down(ns, semid, cmd, version, p);
1347 /* If the task doesn't already have a undo_list, then allocate one
1348 * here. We guarantee there is only one thread using this undo list,
1349 * and current is THE ONE
1351 * If this allocation and assignment succeeds, but later
1352 * portions of this code fail, there is no need to free the sem_undo_list.
1353 * Just let it stay associated with the task, and it'll be freed later
1356 * This can block, so callers must hold no locks.
1358 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1360 struct sem_undo_list *undo_list;
1362 undo_list = current->sysvsem.undo_list;
1364 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1365 if (undo_list == NULL)
1367 spin_lock_init(&undo_list->lock);
1368 atomic_set(&undo_list->refcnt, 1);
1369 INIT_LIST_HEAD(&undo_list->list_proc);
1371 current->sysvsem.undo_list = undo_list;
1373 *undo_listp = undo_list;
1377 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1379 struct sem_undo *un;
1381 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1382 if (un->semid == semid)
1388 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1390 struct sem_undo *un;
1392 assert_spin_locked(&ulp->lock);
1394 un = __lookup_undo(ulp, semid);
1396 list_del_rcu(&un->list_proc);
1397 list_add_rcu(&un->list_proc, &ulp->list_proc);
1403 * find_alloc_undo - Lookup (and if not present create) undo array
1405 * @semid: semaphore array id
1407 * The function looks up (and if not present creates) the undo structure.
1408 * The size of the undo structure depends on the size of the semaphore
1409 * array, thus the alloc path is not that straightforward.
1410 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1411 * performs a rcu_read_lock().
1413 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1415 struct sem_array *sma;
1416 struct sem_undo_list *ulp;
1417 struct sem_undo *un, *new;
1420 error = get_undo_list(&ulp);
1422 return ERR_PTR(error);
1425 spin_lock(&ulp->lock);
1426 un = lookup_undo(ulp, semid);
1427 spin_unlock(&ulp->lock);
1428 if (likely(un!=NULL))
1431 /* no undo structure around - allocate one. */
1432 /* step 1: figure out the size of the semaphore array */
1433 sma = sem_obtain_object_check(ns, semid);
1436 return ERR_CAST(sma);
1439 nsems = sma->sem_nsems;
1440 if (!ipc_rcu_getref(sma)) {
1442 un = ERR_PTR(-EIDRM);
1447 /* step 2: allocate new undo structure */
1448 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1451 return ERR_PTR(-ENOMEM);
1454 /* step 3: Acquire the lock on semaphore array */
1456 sem_lock_and_putref(sma);
1457 if (sma->sem_perm.deleted) {
1458 sem_unlock(sma, -1);
1461 un = ERR_PTR(-EIDRM);
1464 spin_lock(&ulp->lock);
1467 * step 4: check for races: did someone else allocate the undo struct?
1469 un = lookup_undo(ulp, semid);
1474 /* step 5: initialize & link new undo structure */
1475 new->semadj = (short *) &new[1];
1478 assert_spin_locked(&ulp->lock);
1479 list_add_rcu(&new->list_proc, &ulp->list_proc);
1480 assert_spin_locked(&sma->sem_perm.lock);
1481 list_add(&new->list_id, &sma->list_id);
1485 spin_unlock(&ulp->lock);
1486 sem_unlock(sma, -1);
1493 * get_queue_result - Retrieve the result code from sem_queue
1494 * @q: Pointer to queue structure
1496 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1497 * q->status, then we must loop until the value is replaced with the final
1498 * value: This may happen if a task is woken up by an unrelated event (e.g.
1499 * signal) and in parallel the task is woken up by another task because it got
1500 * the requested semaphores.
1502 * The function can be called with or without holding the semaphore spinlock.
1504 static int get_queue_result(struct sem_queue *q)
1509 while (unlikely(error == IN_WAKEUP)) {
1518 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1519 unsigned, nsops, const struct timespec __user *, timeout)
1521 int error = -EINVAL;
1522 struct sem_array *sma;
1523 struct sembuf fast_sops[SEMOPM_FAST];
1524 struct sembuf* sops = fast_sops, *sop;
1525 struct sem_undo *un;
1526 int undos = 0, alter = 0, max, locknum;
1527 struct sem_queue queue;
1528 unsigned long jiffies_left = 0;
1529 struct ipc_namespace *ns;
1530 struct list_head tasks;
1532 ns = current->nsproxy->ipc_ns;
1534 if (nsops < 1 || semid < 0)
1536 if (nsops > ns->sc_semopm)
1538 if(nsops > SEMOPM_FAST) {
1539 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1543 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1548 struct timespec _timeout;
1549 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1553 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1554 _timeout.tv_nsec >= 1000000000L) {
1558 jiffies_left = timespec_to_jiffies(&_timeout);
1561 for (sop = sops; sop < sops + nsops; sop++) {
1562 if (sop->sem_num >= max)
1564 if (sop->sem_flg & SEM_UNDO)
1566 if (sop->sem_op != 0)
1570 INIT_LIST_HEAD(&tasks);
1573 /* On success, find_alloc_undo takes the rcu_read_lock */
1574 un = find_alloc_undo(ns, semid);
1576 error = PTR_ERR(un);
1584 sma = sem_obtain_object_check(ns, semid);
1587 error = PTR_ERR(sma);
1592 if (max >= sma->sem_nsems) {
1598 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) {
1603 error = security_sem_semop(sma, sops, nsops, alter);
1610 * semid identifiers are not unique - find_alloc_undo may have
1611 * allocated an undo structure, it was invalidated by an RMID
1612 * and now a new array with received the same id. Check and fail.
1613 * This case can be detected checking un->semid. The existence of
1614 * "un" itself is guaranteed by rcu.
1617 locknum = sem_lock(sma, sops, nsops);
1618 if (un && un->semid == -1)
1619 goto out_unlock_free;
1621 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1623 if (alter && error == 0)
1624 do_smart_update(sma, sops, nsops, 1, &tasks);
1626 goto out_unlock_free;
1629 /* We need to sleep on this operation, so we put the current
1630 * task into the pending queue and go to sleep.
1634 queue.nsops = nsops;
1636 queue.pid = task_tgid_vnr(current);
1637 queue.alter = alter;
1641 curr = &sma->sem_base[sops->sem_num];
1644 list_add_tail(&queue.list, &curr->sem_pending);
1646 list_add(&queue.list, &curr->sem_pending);
1649 list_add_tail(&queue.list, &sma->sem_pending);
1651 list_add(&queue.list, &sma->sem_pending);
1652 sma->complex_count++;
1655 queue.status = -EINTR;
1656 queue.sleeper = current;
1659 current->state = TASK_INTERRUPTIBLE;
1660 sem_unlock(sma, locknum);
1664 jiffies_left = schedule_timeout(jiffies_left);
1668 error = get_queue_result(&queue);
1670 if (error != -EINTR) {
1671 /* fast path: update_queue already obtained all requested
1673 * Perform a smp_mb(): User space could assume that semop()
1674 * is a memory barrier: Without the mb(), the cpu could
1675 * speculatively read in user space stale data that was
1676 * overwritten by the previous owner of the semaphore.
1683 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1686 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1688 error = get_queue_result(&queue);
1691 * Array removed? If yes, leave without sem_unlock().
1699 * If queue.status != -EINTR we are woken up by another process.
1700 * Leave without unlink_queue(), but with sem_unlock().
1703 if (error != -EINTR) {
1704 goto out_unlock_free;
1708 * If an interrupt occurred we have to clean up the queue
1710 if (timeout && jiffies_left == 0)
1714 * If the wakeup was spurious, just retry
1716 if (error == -EINTR && !signal_pending(current))
1719 unlink_queue(sma, &queue);
1722 sem_unlock(sma, locknum);
1725 wake_up_sem_queue_do(&tasks);
1727 if(sops != fast_sops)
1732 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1735 return sys_semtimedop(semid, tsops, nsops, NULL);
1738 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1739 * parent and child tasks.
1742 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1744 struct sem_undo_list *undo_list;
1747 if (clone_flags & CLONE_SYSVSEM) {
1748 error = get_undo_list(&undo_list);
1751 atomic_inc(&undo_list->refcnt);
1752 tsk->sysvsem.undo_list = undo_list;
1754 tsk->sysvsem.undo_list = NULL;
1760 * add semadj values to semaphores, free undo structures.
1761 * undo structures are not freed when semaphore arrays are destroyed
1762 * so some of them may be out of date.
1763 * IMPLEMENTATION NOTE: There is some confusion over whether the
1764 * set of adjustments that needs to be done should be done in an atomic
1765 * manner or not. That is, if we are attempting to decrement the semval
1766 * should we queue up and wait until we can do so legally?
1767 * The original implementation attempted to do this (queue and wait).
1768 * The current implementation does not do so. The POSIX standard
1769 * and SVID should be consulted to determine what behavior is mandated.
1771 void exit_sem(struct task_struct *tsk)
1773 struct sem_undo_list *ulp;
1775 ulp = tsk->sysvsem.undo_list;
1778 tsk->sysvsem.undo_list = NULL;
1780 if (!atomic_dec_and_test(&ulp->refcnt))
1784 struct sem_array *sma;
1785 struct sem_undo *un;
1786 struct list_head tasks;
1790 un = list_entry_rcu(ulp->list_proc.next,
1791 struct sem_undo, list_proc);
1792 if (&un->list_proc == &ulp->list_proc)
1802 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
1803 /* exit_sem raced with IPC_RMID, nothing to do */
1809 sem_lock(sma, NULL, -1);
1810 un = __lookup_undo(ulp, semid);
1812 /* exit_sem raced with IPC_RMID+semget() that created
1813 * exactly the same semid. Nothing to do.
1815 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);
1858 wake_up_sem_queue_do(&tasks);
1865 #ifdef CONFIG_PROC_FS
1866 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1868 struct user_namespace *user_ns = seq_user_ns(s);
1869 struct sem_array *sma = it;
1871 return seq_printf(s,
1872 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1877 from_kuid_munged(user_ns, sma->sem_perm.uid),
1878 from_kgid_munged(user_ns, sma->sem_perm.gid),
1879 from_kuid_munged(user_ns, sma->sem_perm.cuid),
1880 from_kgid_munged(user_ns, sma->sem_perm.cgid),