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 * The caller holds the RCU read lock.
275 static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns,
276 int id, struct sembuf *sops, int nsops, int *locknum)
278 struct kern_ipc_perm *ipcp;
279 struct sem_array *sma;
281 ipcp = ipc_obtain_object(&sem_ids(ns), id);
283 return ERR_CAST(ipcp);
285 sma = container_of(ipcp, struct sem_array, sem_perm);
286 *locknum = sem_lock(sma, sops, nsops);
288 /* ipc_rmid() may have already freed the ID while sem_lock
289 * was spinning: verify that the structure is still valid
292 return container_of(ipcp, struct sem_array, sem_perm);
294 sem_unlock(sma, *locknum);
295 return ERR_PTR(-EINVAL);
298 static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id)
300 struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id);
303 return ERR_CAST(ipcp);
305 return container_of(ipcp, struct sem_array, sem_perm);
308 static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns,
311 struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id);
314 return ERR_CAST(ipcp);
316 return container_of(ipcp, struct sem_array, sem_perm);
319 static inline void sem_lock_and_putref(struct sem_array *sma)
321 sem_lock(sma, NULL, -1);
325 static inline void sem_putref(struct sem_array *sma)
330 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
332 ipc_rmid(&sem_ids(ns), &s->sem_perm);
336 * Lockless wakeup algorithm:
337 * Without the check/retry algorithm a lockless wakeup is possible:
338 * - queue.status is initialized to -EINTR before blocking.
339 * - wakeup is performed by
340 * * unlinking the queue entry from sma->sem_pending
341 * * setting queue.status to IN_WAKEUP
342 * This is the notification for the blocked thread that a
343 * result value is imminent.
344 * * call wake_up_process
345 * * set queue.status to the final value.
346 * - the previously blocked thread checks queue.status:
347 * * if it's IN_WAKEUP, then it must wait until the value changes
348 * * if it's not -EINTR, then the operation was completed by
349 * update_queue. semtimedop can return queue.status without
350 * performing any operation on the sem array.
351 * * otherwise it must acquire the spinlock and check what's up.
353 * The two-stage algorithm is necessary to protect against the following
355 * - if queue.status is set after wake_up_process, then the woken up idle
356 * thread could race forward and try (and fail) to acquire sma->lock
357 * before update_queue had a chance to set queue.status
358 * - if queue.status is written before wake_up_process and if the
359 * blocked process is woken up by a signal between writing
360 * queue.status and the wake_up_process, then the woken up
361 * process could return from semtimedop and die by calling
362 * sys_exit before wake_up_process is called. Then wake_up_process
363 * will oops, because the task structure is already invalid.
364 * (yes, this happened on s390 with sysv msg).
370 * newary - Create a new semaphore set
372 * @params: ptr to the structure that contains key, semflg and nsems
374 * Called with sem_ids.rw_mutex held (as a writer)
377 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
381 struct sem_array *sma;
383 key_t key = params->key;
384 int nsems = params->u.nsems;
385 int semflg = params->flg;
390 if (ns->used_sems + nsems > ns->sc_semmns)
393 size = sizeof (*sma) + nsems * sizeof (struct sem);
394 sma = ipc_rcu_alloc(size);
398 memset (sma, 0, size);
400 sma->sem_perm.mode = (semflg & S_IRWXUGO);
401 sma->sem_perm.key = key;
403 sma->sem_perm.security = NULL;
404 retval = security_sem_alloc(sma);
410 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
412 security_sem_free(sma);
416 ns->used_sems += nsems;
418 sma->sem_base = (struct sem *) &sma[1];
420 for (i = 0; i < nsems; i++) {
421 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
422 spin_lock_init(&sma->sem_base[i].lock);
425 sma->complex_count = 0;
426 INIT_LIST_HEAD(&sma->sem_pending);
427 INIT_LIST_HEAD(&sma->list_id);
428 sma->sem_nsems = nsems;
429 sma->sem_ctime = get_seconds();
433 return sma->sem_perm.id;
438 * Called with sem_ids.rw_mutex and ipcp locked.
440 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
442 struct sem_array *sma;
444 sma = container_of(ipcp, struct sem_array, sem_perm);
445 return security_sem_associate(sma, semflg);
449 * Called with sem_ids.rw_mutex and ipcp locked.
451 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
452 struct ipc_params *params)
454 struct sem_array *sma;
456 sma = container_of(ipcp, struct sem_array, sem_perm);
457 if (params->u.nsems > sma->sem_nsems)
463 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
465 struct ipc_namespace *ns;
466 struct ipc_ops sem_ops;
467 struct ipc_params sem_params;
469 ns = current->nsproxy->ipc_ns;
471 if (nsems < 0 || nsems > ns->sc_semmsl)
474 sem_ops.getnew = newary;
475 sem_ops.associate = sem_security;
476 sem_ops.more_checks = sem_more_checks;
478 sem_params.key = key;
479 sem_params.flg = semflg;
480 sem_params.u.nsems = nsems;
482 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
486 * Determine whether a sequence of semaphore operations would succeed
487 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
490 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
491 int nsops, struct sem_undo *un, int pid)
497 for (sop = sops; sop < sops + nsops; sop++) {
498 curr = sma->sem_base + sop->sem_num;
499 sem_op = sop->sem_op;
500 result = curr->semval;
502 if (!sem_op && result)
510 if (sop->sem_flg & SEM_UNDO) {
511 int undo = un->semadj[sop->sem_num] - sem_op;
513 * Exceeding the undo range is an error.
515 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
518 curr->semval = result;
522 while (sop >= sops) {
523 sma->sem_base[sop->sem_num].sempid = pid;
524 if (sop->sem_flg & SEM_UNDO)
525 un->semadj[sop->sem_num] -= sop->sem_op;
536 if (sop->sem_flg & IPC_NOWAIT)
543 while (sop >= sops) {
544 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
551 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
552 * @q: queue entry that must be signaled
553 * @error: Error value for the signal
555 * Prepare the wake-up of the queue entry q.
557 static void wake_up_sem_queue_prepare(struct list_head *pt,
558 struct sem_queue *q, int error)
560 if (list_empty(pt)) {
562 * Hold preempt off so that we don't get preempted and have the
563 * wakee busy-wait until we're scheduled back on.
567 q->status = IN_WAKEUP;
570 list_add_tail(&q->list, pt);
574 * wake_up_sem_queue_do(pt) - do the actual wake-up
575 * @pt: list of tasks to be woken up
577 * Do the actual wake-up.
578 * The function is called without any locks held, thus the semaphore array
579 * could be destroyed already and the tasks can disappear as soon as the
580 * status is set to the actual return code.
582 static void wake_up_sem_queue_do(struct list_head *pt)
584 struct sem_queue *q, *t;
587 did_something = !list_empty(pt);
588 list_for_each_entry_safe(q, t, pt, list) {
589 wake_up_process(q->sleeper);
590 /* q can disappear immediately after writing q->status. */
598 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
602 sma->complex_count--;
605 /** check_restart(sma, q)
606 * @sma: semaphore array
607 * @q: the operation that just completed
609 * update_queue is O(N^2) when it restarts scanning the whole queue of
610 * waiting operations. Therefore this function checks if the restart is
611 * really necessary. It is called after a previously waiting operation
614 static int check_restart(struct sem_array *sma, struct sem_queue *q)
619 /* if the operation didn't modify the array, then no restart */
623 /* pending complex operations are too difficult to analyse */
624 if (sma->complex_count)
627 /* we were a sleeping complex operation. Too difficult */
631 curr = sma->sem_base + q->sops[0].sem_num;
633 /* No-one waits on this queue */
634 if (list_empty(&curr->sem_pending))
637 /* the new semaphore value */
639 /* It is impossible that someone waits for the new value:
640 * - q is a previously sleeping simple operation that
641 * altered the array. It must be a decrement, because
642 * simple increments never sleep.
643 * - The value is not 0, thus wait-for-zero won't proceed.
644 * - If there are older (higher priority) decrements
645 * in the queue, then they have observed the original
646 * semval value and couldn't proceed. The operation
647 * decremented to value - thus they won't proceed either.
649 BUG_ON(q->sops[0].sem_op >= 0);
653 * semval is 0. Check if there are wait-for-zero semops.
654 * They must be the first entries in the per-semaphore queue
656 h = list_first_entry(&curr->sem_pending, struct sem_queue, list);
657 BUG_ON(h->nsops != 1);
658 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
660 /* Yes, there is a wait-for-zero semop. Restart */
661 if (h->sops[0].sem_op == 0)
664 /* Again - no-one is waiting for the new value. */
670 * update_queue(sma, semnum): Look for tasks that can be completed.
671 * @sma: semaphore array.
672 * @semnum: semaphore that was modified.
673 * @pt: list head for the tasks that must be woken up.
675 * update_queue must be called after a semaphore in a semaphore array
676 * was modified. If multiple semaphores were modified, update_queue must
677 * be called with semnum = -1, as well as with the number of each modified
679 * The tasks that must be woken up are added to @pt. The return code
680 * is stored in q->pid.
681 * The function return 1 if at least one semop was completed successfully.
683 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
686 struct list_head *walk;
687 struct list_head *pending_list;
688 int semop_completed = 0;
691 pending_list = &sma->sem_pending;
693 pending_list = &sma->sem_base[semnum].sem_pending;
696 walk = pending_list->next;
697 while (walk != pending_list) {
700 q = container_of(walk, struct sem_queue, list);
703 /* If we are scanning the single sop, per-semaphore list of
704 * one semaphore and that semaphore is 0, then it is not
705 * necessary to scan the "alter" entries: simple increments
706 * that affect only one entry succeed immediately and cannot
707 * be in the per semaphore pending queue, and decrements
708 * cannot be successful if the value is already 0.
710 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
714 error = try_atomic_semop(sma, q->sops, q->nsops,
717 /* Does q->sleeper still need to sleep? */
721 unlink_queue(sma, q);
727 restart = check_restart(sma, q);
730 wake_up_sem_queue_prepare(pt, q, error);
734 return semop_completed;
738 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
739 * @sma: semaphore array
740 * @sops: operations that were performed
741 * @nsops: number of operations
742 * @otime: force setting otime
743 * @pt: list head of the tasks that must be woken up.
745 * do_smart_update() does the required called to update_queue, based on the
746 * actual changes that were performed on the semaphore array.
747 * Note that the function does not do the actual wake-up: the caller is
748 * responsible for calling wake_up_sem_queue_do(@pt).
749 * It is safe to perform this call after dropping all locks.
751 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
752 int otime, struct list_head *pt)
756 if (sma->complex_count || sops == NULL) {
757 if (update_queue(sma, -1, pt))
762 /* No semops; something special is going on. */
763 for (i = 0; i < sma->sem_nsems; i++) {
764 if (update_queue(sma, i, pt))
770 /* Check the semaphores that were modified. */
771 for (i = 0; i < nsops; i++) {
772 if (sops[i].sem_op > 0 ||
773 (sops[i].sem_op < 0 &&
774 sma->sem_base[sops[i].sem_num].semval == 0))
775 if (update_queue(sma, sops[i].sem_num, pt))
780 sma->sem_otime = get_seconds();
784 /* The following counts are associated to each semaphore:
785 * semncnt number of tasks waiting on semval being nonzero
786 * semzcnt number of tasks waiting on semval being zero
787 * This model assumes that a task waits on exactly one semaphore.
788 * Since semaphore operations are to be performed atomically, tasks actually
789 * wait on a whole sequence of semaphores simultaneously.
790 * The counts we return here are a rough approximation, but still
791 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
793 static int count_semncnt (struct sem_array * sma, ushort semnum)
796 struct sem_queue * q;
799 list_for_each_entry(q, &sma->sem_base[semnum].sem_pending, list) {
800 struct sembuf * sops = q->sops;
801 BUG_ON(sops->sem_num != semnum);
802 if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT))
806 list_for_each_entry(q, &sma->sem_pending, list) {
807 struct sembuf * sops = q->sops;
808 int nsops = q->nsops;
810 for (i = 0; i < nsops; i++)
811 if (sops[i].sem_num == semnum
812 && (sops[i].sem_op < 0)
813 && !(sops[i].sem_flg & IPC_NOWAIT))
819 static int count_semzcnt (struct sem_array * sma, ushort semnum)
822 struct sem_queue * q;
825 list_for_each_entry(q, &sma->sem_base[semnum].sem_pending, list) {
826 struct sembuf * sops = q->sops;
827 BUG_ON(sops->sem_num != semnum);
828 if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT))
832 list_for_each_entry(q, &sma->sem_pending, list) {
833 struct sembuf * sops = q->sops;
834 int nsops = q->nsops;
836 for (i = 0; i < nsops; i++)
837 if (sops[i].sem_num == semnum
838 && (sops[i].sem_op == 0)
839 && !(sops[i].sem_flg & IPC_NOWAIT))
845 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
846 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
847 * remains locked on exit.
849 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
851 struct sem_undo *un, *tu;
852 struct sem_queue *q, *tq;
853 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
854 struct list_head tasks;
857 /* Free the existing undo structures for this semaphore set. */
858 assert_spin_locked(&sma->sem_perm.lock);
859 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
860 list_del(&un->list_id);
861 spin_lock(&un->ulp->lock);
863 list_del_rcu(&un->list_proc);
864 spin_unlock(&un->ulp->lock);
868 /* Wake up all pending processes and let them fail with EIDRM. */
869 INIT_LIST_HEAD(&tasks);
870 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
871 unlink_queue(sma, q);
872 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
874 for (i = 0; i < sma->sem_nsems; i++) {
875 struct sem *sem = sma->sem_base + i;
876 list_for_each_entry_safe(q, tq, &sem->sem_pending, list) {
877 unlink_queue(sma, q);
878 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
882 /* Remove the semaphore set from the IDR */
887 wake_up_sem_queue_do(&tasks);
888 ns->used_sems -= sma->sem_nsems;
889 security_sem_free(sma);
893 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
897 return copy_to_user(buf, in, sizeof(*in));
902 memset(&out, 0, sizeof(out));
904 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
906 out.sem_otime = in->sem_otime;
907 out.sem_ctime = in->sem_ctime;
908 out.sem_nsems = in->sem_nsems;
910 return copy_to_user(buf, &out, sizeof(out));
917 static int semctl_nolock(struct ipc_namespace *ns, int semid,
918 int cmd, int version, void __user *p)
921 struct sem_array *sma;
927 struct seminfo seminfo;
930 err = security_sem_semctl(NULL, cmd);
934 memset(&seminfo,0,sizeof(seminfo));
935 seminfo.semmni = ns->sc_semmni;
936 seminfo.semmns = ns->sc_semmns;
937 seminfo.semmsl = ns->sc_semmsl;
938 seminfo.semopm = ns->sc_semopm;
939 seminfo.semvmx = SEMVMX;
940 seminfo.semmnu = SEMMNU;
941 seminfo.semmap = SEMMAP;
942 seminfo.semume = SEMUME;
943 down_read(&sem_ids(ns).rw_mutex);
944 if (cmd == SEM_INFO) {
945 seminfo.semusz = sem_ids(ns).in_use;
946 seminfo.semaem = ns->used_sems;
948 seminfo.semusz = SEMUSZ;
949 seminfo.semaem = SEMAEM;
951 max_id = ipc_get_maxid(&sem_ids(ns));
952 up_read(&sem_ids(ns).rw_mutex);
953 if (copy_to_user(p, &seminfo, sizeof(struct seminfo)))
955 return (max_id < 0) ? 0: max_id;
960 struct semid64_ds tbuf;
963 memset(&tbuf, 0, sizeof(tbuf));
966 if (cmd == SEM_STAT) {
967 sma = sem_obtain_object(ns, semid);
972 id = sma->sem_perm.id;
974 sma = sem_obtain_object_check(ns, semid);
982 if (ipcperms(ns, &sma->sem_perm, S_IRUGO))
985 err = security_sem_semctl(sma, cmd);
989 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
990 tbuf.sem_otime = sma->sem_otime;
991 tbuf.sem_ctime = sma->sem_ctime;
992 tbuf.sem_nsems = sma->sem_nsems;
994 if (copy_semid_to_user(p, &tbuf, version))
1006 static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum,
1009 struct sem_undo *un;
1010 struct sem_array *sma;
1013 struct list_head tasks;
1015 #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN)
1016 /* big-endian 64bit */
1019 /* 32bit or little-endian 64bit */
1023 if (val > SEMVMX || val < 0)
1026 INIT_LIST_HEAD(&tasks);
1029 sma = sem_obtain_object_check(ns, semid);
1032 return PTR_ERR(sma);
1035 if (semnum < 0 || semnum >= sma->sem_nsems) {
1041 if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) {
1046 err = security_sem_semctl(sma, SETVAL);
1052 sem_lock(sma, NULL, -1);
1054 curr = &sma->sem_base[semnum];
1056 assert_spin_locked(&sma->sem_perm.lock);
1057 list_for_each_entry(un, &sma->list_id, list_id)
1058 un->semadj[semnum] = 0;
1061 curr->sempid = task_tgid_vnr(current);
1062 sma->sem_ctime = get_seconds();
1063 /* maybe some queued-up processes were waiting for this */
1064 do_smart_update(sma, NULL, 0, 0, &tasks);
1065 sem_unlock(sma, -1);
1067 wake_up_sem_queue_do(&tasks);
1071 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
1072 int cmd, void __user *p)
1074 struct sem_array *sma;
1077 ushort fast_sem_io[SEMMSL_FAST];
1078 ushort* sem_io = fast_sem_io;
1079 struct list_head tasks;
1081 INIT_LIST_HEAD(&tasks);
1084 sma = sem_obtain_object_check(ns, semid);
1087 return PTR_ERR(sma);
1090 nsems = sma->sem_nsems;
1093 if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO))
1094 goto out_rcu_wakeup;
1096 err = security_sem_semctl(sma, cmd);
1098 goto out_rcu_wakeup;
1104 ushort __user *array = p;
1107 sem_lock(sma, NULL, -1);
1108 if(nsems > SEMMSL_FAST) {
1109 if (!ipc_rcu_getref(sma)) {
1110 sem_unlock(sma, -1);
1115 sem_unlock(sma, -1);
1117 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1118 if(sem_io == NULL) {
1124 sem_lock_and_putref(sma);
1125 if (sma->sem_perm.deleted) {
1126 sem_unlock(sma, -1);
1132 for (i = 0; i < sma->sem_nsems; i++)
1133 sem_io[i] = sma->sem_base[i].semval;
1134 sem_unlock(sma, -1);
1137 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
1144 struct sem_undo *un;
1146 if (!ipc_rcu_getref(sma)) {
1152 if(nsems > SEMMSL_FAST) {
1153 sem_io = ipc_alloc(sizeof(ushort)*nsems);
1154 if(sem_io == NULL) {
1160 if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) {
1166 for (i = 0; i < nsems; i++) {
1167 if (sem_io[i] > SEMVMX) {
1174 sem_lock_and_putref(sma);
1175 if (sma->sem_perm.deleted) {
1176 sem_unlock(sma, -1);
1182 for (i = 0; i < nsems; i++)
1183 sma->sem_base[i].semval = sem_io[i];
1185 assert_spin_locked(&sma->sem_perm.lock);
1186 list_for_each_entry(un, &sma->list_id, list_id) {
1187 for (i = 0; i < nsems; i++)
1190 sma->sem_ctime = get_seconds();
1191 /* maybe some queued-up processes were waiting for this */
1192 do_smart_update(sma, NULL, 0, 0, &tasks);
1196 /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */
1199 if (semnum < 0 || semnum >= nsems)
1200 goto out_rcu_wakeup;
1202 sem_lock(sma, NULL, -1);
1203 curr = &sma->sem_base[semnum];
1213 err = count_semncnt(sma,semnum);
1216 err = count_semzcnt(sma,semnum);
1221 sem_unlock(sma, -1);
1224 wake_up_sem_queue_do(&tasks);
1226 if(sem_io != fast_sem_io)
1227 ipc_free(sem_io, sizeof(ushort)*nsems);
1231 static inline unsigned long
1232 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1236 if (copy_from_user(out, buf, sizeof(*out)))
1241 struct semid_ds tbuf_old;
1243 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1246 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1247 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1248 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1258 * This function handles some semctl commands which require the rw_mutex
1259 * to be held in write mode.
1260 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1262 static int semctl_down(struct ipc_namespace *ns, int semid,
1263 int cmd, int version, void __user *p)
1265 struct sem_array *sma;
1267 struct semid64_ds semid64;
1268 struct kern_ipc_perm *ipcp;
1270 if(cmd == IPC_SET) {
1271 if (copy_semid_from_user(&semid64, p, version))
1275 ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd,
1276 &semid64.sem_perm, 0);
1278 return PTR_ERR(ipcp);
1280 sma = container_of(ipcp, struct sem_array, sem_perm);
1282 err = security_sem_semctl(sma, cmd);
1290 sem_lock(sma, NULL, -1);
1294 sem_lock(sma, NULL, -1);
1295 err = ipc_update_perm(&semid64.sem_perm, ipcp);
1298 sma->sem_ctime = get_seconds();
1307 sem_unlock(sma, -1);
1310 up_write(&sem_ids(ns).rw_mutex);
1314 SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg)
1317 struct ipc_namespace *ns;
1318 void __user *p = (void __user *)arg;
1323 version = ipc_parse_version(&cmd);
1324 ns = current->nsproxy->ipc_ns;
1331 return semctl_nolock(ns, semid, cmd, version, p);
1338 return semctl_main(ns, semid, semnum, cmd, p);
1340 return semctl_setval(ns, semid, semnum, arg);
1343 return semctl_down(ns, semid, cmd, version, p);
1349 /* If the task doesn't already have a undo_list, then allocate one
1350 * here. We guarantee there is only one thread using this undo list,
1351 * and current is THE ONE
1353 * If this allocation and assignment succeeds, but later
1354 * portions of this code fail, there is no need to free the sem_undo_list.
1355 * Just let it stay associated with the task, and it'll be freed later
1358 * This can block, so callers must hold no locks.
1360 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1362 struct sem_undo_list *undo_list;
1364 undo_list = current->sysvsem.undo_list;
1366 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1367 if (undo_list == NULL)
1369 spin_lock_init(&undo_list->lock);
1370 atomic_set(&undo_list->refcnt, 1);
1371 INIT_LIST_HEAD(&undo_list->list_proc);
1373 current->sysvsem.undo_list = undo_list;
1375 *undo_listp = undo_list;
1379 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1381 struct sem_undo *un;
1383 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1384 if (un->semid == semid)
1390 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1392 struct sem_undo *un;
1394 assert_spin_locked(&ulp->lock);
1396 un = __lookup_undo(ulp, semid);
1398 list_del_rcu(&un->list_proc);
1399 list_add_rcu(&un->list_proc, &ulp->list_proc);
1405 * find_alloc_undo - Lookup (and if not present create) undo array
1407 * @semid: semaphore array id
1409 * The function looks up (and if not present creates) the undo structure.
1410 * The size of the undo structure depends on the size of the semaphore
1411 * array, thus the alloc path is not that straightforward.
1412 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1413 * performs a rcu_read_lock().
1415 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1417 struct sem_array *sma;
1418 struct sem_undo_list *ulp;
1419 struct sem_undo *un, *new;
1422 error = get_undo_list(&ulp);
1424 return ERR_PTR(error);
1427 spin_lock(&ulp->lock);
1428 un = lookup_undo(ulp, semid);
1429 spin_unlock(&ulp->lock);
1430 if (likely(un!=NULL))
1433 /* no undo structure around - allocate one. */
1434 /* step 1: figure out the size of the semaphore array */
1435 sma = sem_obtain_object_check(ns, semid);
1438 return ERR_CAST(sma);
1441 nsems = sma->sem_nsems;
1442 if (!ipc_rcu_getref(sma)) {
1444 un = ERR_PTR(-EIDRM);
1449 /* step 2: allocate new undo structure */
1450 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1453 return ERR_PTR(-ENOMEM);
1456 /* step 3: Acquire the lock on semaphore array */
1458 sem_lock_and_putref(sma);
1459 if (sma->sem_perm.deleted) {
1460 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);
1488 sem_unlock(sma, -1);
1495 * get_queue_result - Retrieve the result code from sem_queue
1496 * @q: Pointer to queue structure
1498 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1499 * q->status, then we must loop until the value is replaced with the final
1500 * value: This may happen if a task is woken up by an unrelated event (e.g.
1501 * signal) and in parallel the task is woken up by another task because it got
1502 * the requested semaphores.
1504 * The function can be called with or without holding the semaphore spinlock.
1506 static int get_queue_result(struct sem_queue *q)
1511 while (unlikely(error == IN_WAKEUP)) {
1520 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1521 unsigned, nsops, const struct timespec __user *, timeout)
1523 int error = -EINVAL;
1524 struct sem_array *sma;
1525 struct sembuf fast_sops[SEMOPM_FAST];
1526 struct sembuf* sops = fast_sops, *sop;
1527 struct sem_undo *un;
1528 int undos = 0, alter = 0, max, locknum;
1529 struct sem_queue queue;
1530 unsigned long jiffies_left = 0;
1531 struct ipc_namespace *ns;
1532 struct list_head tasks;
1534 ns = current->nsproxy->ipc_ns;
1536 if (nsops < 1 || semid < 0)
1538 if (nsops > ns->sc_semopm)
1540 if(nsops > SEMOPM_FAST) {
1541 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1545 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1550 struct timespec _timeout;
1551 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1555 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1556 _timeout.tv_nsec >= 1000000000L) {
1560 jiffies_left = timespec_to_jiffies(&_timeout);
1563 for (sop = sops; sop < sops + nsops; sop++) {
1564 if (sop->sem_num >= max)
1566 if (sop->sem_flg & SEM_UNDO)
1568 if (sop->sem_op != 0)
1572 INIT_LIST_HEAD(&tasks);
1575 /* On success, find_alloc_undo takes the rcu_read_lock */
1576 un = find_alloc_undo(ns, semid);
1578 error = PTR_ERR(un);
1586 sma = sem_obtain_object_check(ns, semid);
1589 error = PTR_ERR(sma);
1594 if (max >= sma->sem_nsems)
1595 goto out_rcu_wakeup;
1598 if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1599 goto out_rcu_wakeup;
1601 error = security_sem_semop(sma, sops, nsops, alter);
1603 goto out_rcu_wakeup;
1606 * semid identifiers are not unique - find_alloc_undo may have
1607 * allocated an undo structure, it was invalidated by an RMID
1608 * and now a new array with received the same id. Check and fail.
1609 * This case can be detected checking un->semid. The existence of
1610 * "un" itself is guaranteed by rcu.
1613 locknum = sem_lock(sma, sops, nsops);
1614 if (un && un->semid == -1)
1615 goto out_unlock_free;
1617 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1619 if (alter && error == 0)
1620 do_smart_update(sma, sops, nsops, 1, &tasks);
1622 goto out_unlock_free;
1625 /* We need to sleep on this operation, so we put the current
1626 * task into the pending queue and go to sleep.
1630 queue.nsops = nsops;
1632 queue.pid = task_tgid_vnr(current);
1633 queue.alter = alter;
1637 curr = &sma->sem_base[sops->sem_num];
1640 list_add_tail(&queue.list, &curr->sem_pending);
1642 list_add(&queue.list, &curr->sem_pending);
1645 list_add_tail(&queue.list, &sma->sem_pending);
1647 list_add(&queue.list, &sma->sem_pending);
1648 sma->complex_count++;
1651 queue.status = -EINTR;
1652 queue.sleeper = current;
1655 current->state = TASK_INTERRUPTIBLE;
1656 sem_unlock(sma, locknum);
1660 jiffies_left = schedule_timeout(jiffies_left);
1664 error = get_queue_result(&queue);
1666 if (error != -EINTR) {
1667 /* fast path: update_queue already obtained all requested
1669 * Perform a smp_mb(): User space could assume that semop()
1670 * is a memory barrier: Without the mb(), the cpu could
1671 * speculatively read in user space stale data that was
1672 * overwritten by the previous owner of the semaphore.
1680 sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum);
1683 * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing.
1685 error = get_queue_result(&queue);
1688 * Array removed? If yes, leave without sem_unlock().
1697 * If queue.status != -EINTR we are woken up by another process.
1698 * Leave without unlink_queue(), but with sem_unlock().
1701 if (error != -EINTR) {
1702 goto out_unlock_free;
1706 * If an interrupt occurred we have to clean up the queue
1708 if (timeout && jiffies_left == 0)
1712 * If the wakeup was spurious, just retry
1714 if (error == -EINTR && !signal_pending(current))
1717 unlink_queue(sma, &queue);
1720 sem_unlock(sma, locknum);
1723 wake_up_sem_queue_do(&tasks);
1725 if(sops != fast_sops)
1730 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1733 return sys_semtimedop(semid, tsops, nsops, NULL);
1736 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1737 * parent and child tasks.
1740 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1742 struct sem_undo_list *undo_list;
1745 if (clone_flags & CLONE_SYSVSEM) {
1746 error = get_undo_list(&undo_list);
1749 atomic_inc(&undo_list->refcnt);
1750 tsk->sysvsem.undo_list = undo_list;
1752 tsk->sysvsem.undo_list = NULL;
1758 * add semadj values to semaphores, free undo structures.
1759 * undo structures are not freed when semaphore arrays are destroyed
1760 * so some of them may be out of date.
1761 * IMPLEMENTATION NOTE: There is some confusion over whether the
1762 * set of adjustments that needs to be done should be done in an atomic
1763 * manner or not. That is, if we are attempting to decrement the semval
1764 * should we queue up and wait until we can do so legally?
1765 * The original implementation attempted to do this (queue and wait).
1766 * The current implementation does not do so. The POSIX standard
1767 * and SVID should be consulted to determine what behavior is mandated.
1769 void exit_sem(struct task_struct *tsk)
1771 struct sem_undo_list *ulp;
1773 ulp = tsk->sysvsem.undo_list;
1776 tsk->sysvsem.undo_list = NULL;
1778 if (!atomic_dec_and_test(&ulp->refcnt))
1782 struct sem_array *sma;
1783 struct sem_undo *un;
1784 struct list_head tasks;
1788 un = list_entry_rcu(ulp->list_proc.next,
1789 struct sem_undo, list_proc);
1790 if (&un->list_proc == &ulp->list_proc)
1800 sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid);
1801 /* exit_sem raced with IPC_RMID, nothing to do */
1807 sem_lock(sma, NULL, -1);
1808 un = __lookup_undo(ulp, semid);
1810 /* exit_sem raced with IPC_RMID+semget() that created
1811 * exactly the same semid. Nothing to do.
1813 sem_unlock(sma, -1);
1818 /* remove un from the linked lists */
1819 assert_spin_locked(&sma->sem_perm.lock);
1820 list_del(&un->list_id);
1822 spin_lock(&ulp->lock);
1823 list_del_rcu(&un->list_proc);
1824 spin_unlock(&ulp->lock);
1826 /* perform adjustments registered in un */
1827 for (i = 0; i < sma->sem_nsems; i++) {
1828 struct sem * semaphore = &sma->sem_base[i];
1829 if (un->semadj[i]) {
1830 semaphore->semval += un->semadj[i];
1832 * Range checks of the new semaphore value,
1833 * not defined by sus:
1834 * - Some unices ignore the undo entirely
1835 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1836 * - some cap the value (e.g. FreeBSD caps
1837 * at 0, but doesn't enforce SEMVMX)
1839 * Linux caps the semaphore value, both at 0
1842 * Manfred <manfred@colorfullife.com>
1844 if (semaphore->semval < 0)
1845 semaphore->semval = 0;
1846 if (semaphore->semval > SEMVMX)
1847 semaphore->semval = SEMVMX;
1848 semaphore->sempid = task_tgid_vnr(current);
1851 /* maybe some queued-up processes were waiting for this */
1852 INIT_LIST_HEAD(&tasks);
1853 do_smart_update(sma, NULL, 0, 1, &tasks);
1854 sem_unlock(sma, -1);
1856 wake_up_sem_queue_do(&tasks);
1863 #ifdef CONFIG_PROC_FS
1864 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1866 struct user_namespace *user_ns = seq_user_ns(s);
1867 struct sem_array *sma = it;
1869 return seq_printf(s,
1870 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
1875 from_kuid_munged(user_ns, sma->sem_perm.uid),
1876 from_kgid_munged(user_ns, sma->sem_perm.gid),
1877 from_kuid_munged(user_ns, sma->sem_perm.cuid),
1878 from_kgid_munged(user_ns, sma->sem_perm.cgid),