3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
6 * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995):
7 * This code underwent a massive rewrite in order to solve some problems
8 * with the original code. In particular the original code failed to
9 * wake up processes that were waiting for semval to go to 0 if the
10 * value went to 0 and was then incremented rapidly enough. In solving
11 * this problem I have also modified the implementation so that it
12 * processes pending operations in a FIFO manner, thus give a guarantee
13 * that processes waiting for a lock on the semaphore won't starve
14 * unless another locking process fails to unlock.
15 * In addition the following two changes in behavior have been introduced:
16 * - The original implementation of semop returned the value
17 * last semaphore element examined on success. This does not
18 * match the manual page specifications, and effectively
19 * allows the user to read the semaphore even if they do not
20 * have read permissions. The implementation now returns 0
21 * on success as stated in the manual page.
22 * - There is some confusion over whether the set of undo adjustments
23 * to be performed at exit should be done in an atomic manner.
24 * That is, if we are attempting to decrement the semval should we queue
25 * up and wait until we can do so legally?
26 * The original implementation attempted to do this.
27 * The current implementation does not do so. This is because I don't
28 * think it is the right thing (TM) to do, and because I couldn't
29 * see a clean way to get the old behavior with the new design.
30 * The POSIX standard and SVID should be consulted to determine
31 * what behavior is mandated.
33 * Further notes on refinement (Christoph Rohland, December 1998):
34 * - The POSIX standard says, that the undo adjustments simply should
35 * redo. So the current implementation is o.K.
36 * - The previous code had two flaws:
37 * 1) It actively gave the semaphore to the next waiting process
38 * sleeping on the semaphore. Since this process did not have the
39 * cpu this led to many unnecessary context switches and bad
40 * performance. Now we only check which process should be able to
41 * get the semaphore and if this process wants to reduce some
42 * semaphore value we simply wake it up without doing the
43 * operation. So it has to try to get it later. Thus e.g. the
44 * running process may reacquire the semaphore during the current
45 * time slice. If it only waits for zero or increases the semaphore,
46 * we do the operation in advance and wake it up.
47 * 2) It did not wake up all zero waiting processes. We try to do
48 * better but only get the semops right which only wait for zero or
49 * increase. If there are decrement operations in the operations
50 * array we do the same as before.
52 * With the incarnation of O(1) scheduler, it becomes unnecessary to perform
53 * check/retry algorithm for waking up blocked processes as the new scheduler
54 * is better at handling thread switch than the old one.
56 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
58 * SMP-threaded, sysctl's added
59 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
60 * Enforced range limit on SEM_UNDO
61 * (c) 2001 Red Hat Inc
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
65 * support for audit of ipc object properties and permission changes
66 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
70 * Pavel Emelianov <xemul@openvz.org>
73 #include <linux/slab.h>
74 #include <linux/spinlock.h>
75 #include <linux/init.h>
76 #include <linux/proc_fs.h>
77 #include <linux/time.h>
78 #include <linux/security.h>
79 #include <linux/syscalls.h>
80 #include <linux/audit.h>
81 #include <linux/capability.h>
82 #include <linux/seq_file.h>
83 #include <linux/rwsem.h>
84 #include <linux/nsproxy.h>
85 #include <linux/ipc_namespace.h>
87 #include <asm/uaccess.h>
90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
95 static int newary(struct ipc_namespace *, struct ipc_params *);
96 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
98 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
101 #define SEMMSL_FAST 256 /* 512 bytes on stack */
102 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
105 * linked list protection:
107 * sem_array.sem_pending{,last},
108 * sem_array.sem_undo: sem_lock() for read/write
109 * sem_undo.proc_next: only "current" is allowed to read/write that field.
113 #define sc_semmsl sem_ctls[0]
114 #define sc_semmns sem_ctls[1]
115 #define sc_semopm sem_ctls[2]
116 #define sc_semmni sem_ctls[3]
118 void sem_init_ns(struct ipc_namespace *ns)
120 ns->sc_semmsl = SEMMSL;
121 ns->sc_semmns = SEMMNS;
122 ns->sc_semopm = SEMOPM;
123 ns->sc_semmni = SEMMNI;
125 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
129 void sem_exit_ns(struct ipc_namespace *ns)
131 free_ipcs(ns, &sem_ids(ns), freeary);
132 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
136 void __init sem_init (void)
138 sem_init_ns(&init_ipc_ns);
139 ipc_init_proc_interface("sysvipc/sem",
140 " key semid perms nsems uid gid cuid cgid otime ctime\n",
141 IPC_SEM_IDS, sysvipc_sem_proc_show);
145 * sem_lock_(check_) routines are called in the paths where the rw_mutex
148 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
150 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
153 return (struct sem_array *)ipcp;
155 return container_of(ipcp, struct sem_array, sem_perm);
158 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
161 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
164 return (struct sem_array *)ipcp;
166 return container_of(ipcp, struct sem_array, sem_perm);
169 static inline void sem_lock_and_putref(struct sem_array *sma)
171 ipc_lock_by_ptr(&sma->sem_perm);
175 static inline void sem_getref_and_unlock(struct sem_array *sma)
178 ipc_unlock(&(sma)->sem_perm);
181 static inline void sem_putref(struct sem_array *sma)
183 ipc_lock_by_ptr(&sma->sem_perm);
185 ipc_unlock(&(sma)->sem_perm);
188 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
190 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194 * Lockless wakeup algorithm:
195 * Without the check/retry algorithm a lockless wakeup is possible:
196 * - queue.status is initialized to -EINTR before blocking.
197 * - wakeup is performed by
198 * * unlinking the queue entry from sma->sem_pending
199 * * setting queue.status to IN_WAKEUP
200 * This is the notification for the blocked thread that a
201 * result value is imminent.
202 * * call wake_up_process
203 * * set queue.status to the final value.
204 * - the previously blocked thread checks queue.status:
205 * * if it's IN_WAKEUP, then it must wait until the value changes
206 * * if it's not -EINTR, then the operation was completed by
207 * update_queue. semtimedop can return queue.status without
208 * performing any operation on the sem array.
209 * * otherwise it must acquire the spinlock and check what's up.
211 * The two-stage algorithm is necessary to protect against the following
213 * - if queue.status is set after wake_up_process, then the woken up idle
214 * thread could race forward and try (and fail) to acquire sma->lock
215 * before update_queue had a chance to set queue.status
216 * - if queue.status is written before wake_up_process and if the
217 * blocked process is woken up by a signal between writing
218 * queue.status and the wake_up_process, then the woken up
219 * process could return from semtimedop and die by calling
220 * sys_exit before wake_up_process is called. Then wake_up_process
221 * will oops, because the task structure is already invalid.
222 * (yes, this happened on s390 with sysv msg).
228 * newary - Create a new semaphore set
230 * @params: ptr to the structure that contains key, semflg and nsems
232 * Called with sem_ids.rw_mutex held (as a writer)
235 static int newary(struct ipc_namespace *ns, struct ipc_params *params)
239 struct sem_array *sma;
241 key_t key = params->key;
242 int nsems = params->u.nsems;
243 int semflg = params->flg;
248 if (ns->used_sems + nsems > ns->sc_semmns)
251 size = sizeof (*sma) + nsems * sizeof (struct sem);
252 sma = ipc_rcu_alloc(size);
256 memset (sma, 0, size);
258 sma->sem_perm.mode = (semflg & S_IRWXUGO);
259 sma->sem_perm.key = key;
261 sma->sem_perm.security = NULL;
262 retval = security_sem_alloc(sma);
268 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
270 security_sem_free(sma);
274 ns->used_sems += nsems;
276 sma->sem_base = (struct sem *) &sma[1];
278 for (i = 0; i < nsems; i++)
279 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
281 sma->complex_count = 0;
282 INIT_LIST_HEAD(&sma->sem_pending);
283 INIT_LIST_HEAD(&sma->list_id);
284 sma->sem_nsems = nsems;
285 sma->sem_ctime = get_seconds();
288 return sma->sem_perm.id;
293 * Called with sem_ids.rw_mutex and ipcp locked.
295 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
297 struct sem_array *sma;
299 sma = container_of(ipcp, struct sem_array, sem_perm);
300 return security_sem_associate(sma, semflg);
304 * Called with sem_ids.rw_mutex and ipcp locked.
306 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
307 struct ipc_params *params)
309 struct sem_array *sma;
311 sma = container_of(ipcp, struct sem_array, sem_perm);
312 if (params->u.nsems > sma->sem_nsems)
318 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
320 struct ipc_namespace *ns;
321 struct ipc_ops sem_ops;
322 struct ipc_params sem_params;
324 ns = current->nsproxy->ipc_ns;
326 if (nsems < 0 || nsems > ns->sc_semmsl)
329 sem_ops.getnew = newary;
330 sem_ops.associate = sem_security;
331 sem_ops.more_checks = sem_more_checks;
333 sem_params.key = key;
334 sem_params.flg = semflg;
335 sem_params.u.nsems = nsems;
337 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
341 * Determine whether a sequence of semaphore operations would succeed
342 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
345 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
346 int nsops, struct sem_undo *un, int pid)
352 for (sop = sops; sop < sops + nsops; sop++) {
353 curr = sma->sem_base + sop->sem_num;
354 sem_op = sop->sem_op;
355 result = curr->semval;
357 if (!sem_op && result)
365 if (sop->sem_flg & SEM_UNDO) {
366 int undo = un->semadj[sop->sem_num] - sem_op;
368 * Exceeding the undo range is an error.
370 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
373 curr->semval = result;
377 while (sop >= sops) {
378 sma->sem_base[sop->sem_num].sempid = pid;
379 if (sop->sem_flg & SEM_UNDO)
380 un->semadj[sop->sem_num] -= sop->sem_op;
391 if (sop->sem_flg & IPC_NOWAIT)
398 while (sop >= sops) {
399 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
406 /** wake_up_sem_queue_prepare(q, error): Prepare wake-up
407 * @q: queue entry that must be signaled
408 * @error: Error value for the signal
410 * Prepare the wake-up of the queue entry q.
412 static void wake_up_sem_queue_prepare(struct list_head *pt,
413 struct sem_queue *q, int error)
415 if (list_empty(pt)) {
417 * Hold preempt off so that we don't get preempted and have the
418 * wakee busy-wait until we're scheduled back on.
422 q->status = IN_WAKEUP;
425 list_add_tail(&q->simple_list, pt);
429 * wake_up_sem_queue_do(pt) - do the actual wake-up
430 * @pt: list of tasks to be woken up
432 * Do the actual wake-up.
433 * The function is called without any locks held, thus the semaphore array
434 * could be destroyed already and the tasks can disappear as soon as the
435 * status is set to the actual return code.
437 static void wake_up_sem_queue_do(struct list_head *pt)
439 struct sem_queue *q, *t;
442 did_something = !list_empty(pt);
443 list_for_each_entry_safe(q, t, pt, simple_list) {
444 wake_up_process(q->sleeper);
445 /* q can disappear immediately after writing q->status. */
453 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
457 list_del(&q->simple_list);
459 sma->complex_count--;
462 /** check_restart(sma, q)
463 * @sma: semaphore array
464 * @q: the operation that just completed
466 * update_queue is O(N^2) when it restarts scanning the whole queue of
467 * waiting operations. Therefore this function checks if the restart is
468 * really necessary. It is called after a previously waiting operation
471 static int check_restart(struct sem_array *sma, struct sem_queue *q)
476 /* if the operation didn't modify the array, then no restart */
480 /* pending complex operations are too difficult to analyse */
481 if (sma->complex_count)
484 /* we were a sleeping complex operation. Too difficult */
488 curr = sma->sem_base + q->sops[0].sem_num;
490 /* No-one waits on this queue */
491 if (list_empty(&curr->sem_pending))
494 /* the new semaphore value */
496 /* It is impossible that someone waits for the new value:
497 * - q is a previously sleeping simple operation that
498 * altered the array. It must be a decrement, because
499 * simple increments never sleep.
500 * - The value is not 0, thus wait-for-zero won't proceed.
501 * - If there are older (higher priority) decrements
502 * in the queue, then they have observed the original
503 * semval value and couldn't proceed. The operation
504 * decremented to value - thus they won't proceed either.
506 BUG_ON(q->sops[0].sem_op >= 0);
510 * semval is 0. Check if there are wait-for-zero semops.
511 * They must be the first entries in the per-semaphore simple queue
513 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
514 BUG_ON(h->nsops != 1);
515 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
517 /* Yes, there is a wait-for-zero semop. Restart */
518 if (h->sops[0].sem_op == 0)
521 /* Again - no-one is waiting for the new value. */
527 * update_queue(sma, semnum): Look for tasks that can be completed.
528 * @sma: semaphore array.
529 * @semnum: semaphore that was modified.
530 * @pt: list head for the tasks that must be woken up.
532 * update_queue must be called after a semaphore in a semaphore array
533 * was modified. If multiple semaphore were modified, then @semnum
535 * The tasks that must be woken up are added to @pt. The return code
536 * is stored in q->pid.
537 * The function return 1 if at least one semop was completed successfully.
539 static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
542 struct list_head *walk;
543 struct list_head *pending_list;
545 int semop_completed = 0;
547 /* if there are complex operations around, then knowing the semaphore
548 * that was modified doesn't help us. Assume that multiple semaphores
551 if (sma->complex_count)
555 pending_list = &sma->sem_pending;
556 offset = offsetof(struct sem_queue, list);
558 pending_list = &sma->sem_base[semnum].sem_pending;
559 offset = offsetof(struct sem_queue, simple_list);
563 walk = pending_list->next;
564 while (walk != pending_list) {
567 q = (struct sem_queue *)((char *)walk - offset);
570 /* If we are scanning the single sop, per-semaphore list of
571 * one semaphore and that semaphore is 0, then it is not
572 * necessary to scan the "alter" entries: simple increments
573 * that affect only one entry succeed immediately and cannot
574 * be in the per semaphore pending queue, and decrements
575 * cannot be successful if the value is already 0.
577 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
581 error = try_atomic_semop(sma, q->sops, q->nsops,
584 /* Does q->sleeper still need to sleep? */
588 unlink_queue(sma, q);
594 restart = check_restart(sma, q);
597 wake_up_sem_queue_prepare(pt, q, error);
601 return semop_completed;
605 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
606 * @sma: semaphore array
607 * @sops: operations that were performed
608 * @nsops: number of operations
609 * @otime: force setting otime
610 * @pt: list head of the tasks that must be woken up.
612 * do_smart_update() does the required called to update_queue, based on the
613 * actual changes that were performed on the semaphore array.
614 * Note that the function does not do the actual wake-up: the caller is
615 * responsible for calling wake_up_sem_queue_do(@pt).
616 * It is safe to perform this call after dropping all locks.
618 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
619 int otime, struct list_head *pt)
623 if (sma->complex_count || sops == NULL) {
624 if (update_queue(sma, -1, pt))
629 for (i = 0; i < nsops; i++) {
630 if (sops[i].sem_op > 0 ||
631 (sops[i].sem_op < 0 &&
632 sma->sem_base[sops[i].sem_num].semval == 0))
633 if (update_queue(sma, sops[i].sem_num, pt))
638 sma->sem_otime = get_seconds();
642 /* The following counts are associated to each semaphore:
643 * semncnt number of tasks waiting on semval being nonzero
644 * semzcnt number of tasks waiting on semval being zero
645 * This model assumes that a task waits on exactly one semaphore.
646 * Since semaphore operations are to be performed atomically, tasks actually
647 * wait on a whole sequence of semaphores simultaneously.
648 * The counts we return here are a rough approximation, but still
649 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
651 static int count_semncnt (struct sem_array * sma, ushort semnum)
654 struct sem_queue * q;
657 list_for_each_entry(q, &sma->sem_pending, list) {
658 struct sembuf * sops = q->sops;
659 int nsops = q->nsops;
661 for (i = 0; i < nsops; i++)
662 if (sops[i].sem_num == semnum
663 && (sops[i].sem_op < 0)
664 && !(sops[i].sem_flg & IPC_NOWAIT))
670 static int count_semzcnt (struct sem_array * sma, ushort semnum)
673 struct sem_queue * q;
676 list_for_each_entry(q, &sma->sem_pending, list) {
677 struct sembuf * sops = q->sops;
678 int nsops = q->nsops;
680 for (i = 0; i < nsops; i++)
681 if (sops[i].sem_num == semnum
682 && (sops[i].sem_op == 0)
683 && !(sops[i].sem_flg & IPC_NOWAIT))
689 static void free_un(struct rcu_head *head)
691 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
695 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
696 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
697 * remains locked on exit.
699 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
701 struct sem_undo *un, *tu;
702 struct sem_queue *q, *tq;
703 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
704 struct list_head tasks;
706 /* Free the existing undo structures for this semaphore set. */
707 assert_spin_locked(&sma->sem_perm.lock);
708 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
709 list_del(&un->list_id);
710 spin_lock(&un->ulp->lock);
712 list_del_rcu(&un->list_proc);
713 spin_unlock(&un->ulp->lock);
714 call_rcu(&un->rcu, free_un);
717 /* Wake up all pending processes and let them fail with EIDRM. */
718 INIT_LIST_HEAD(&tasks);
719 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
720 unlink_queue(sma, q);
721 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
724 /* Remove the semaphore set from the IDR */
728 wake_up_sem_queue_do(&tasks);
729 ns->used_sems -= sma->sem_nsems;
730 security_sem_free(sma);
734 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
738 return copy_to_user(buf, in, sizeof(*in));
743 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
745 out.sem_otime = in->sem_otime;
746 out.sem_ctime = in->sem_ctime;
747 out.sem_nsems = in->sem_nsems;
749 return copy_to_user(buf, &out, sizeof(out));
756 static int semctl_nolock(struct ipc_namespace *ns, int semid,
757 int cmd, int version, union semun arg)
760 struct sem_array *sma;
766 struct seminfo seminfo;
769 err = security_sem_semctl(NULL, cmd);
773 memset(&seminfo,0,sizeof(seminfo));
774 seminfo.semmni = ns->sc_semmni;
775 seminfo.semmns = ns->sc_semmns;
776 seminfo.semmsl = ns->sc_semmsl;
777 seminfo.semopm = ns->sc_semopm;
778 seminfo.semvmx = SEMVMX;
779 seminfo.semmnu = SEMMNU;
780 seminfo.semmap = SEMMAP;
781 seminfo.semume = SEMUME;
782 down_read(&sem_ids(ns).rw_mutex);
783 if (cmd == SEM_INFO) {
784 seminfo.semusz = sem_ids(ns).in_use;
785 seminfo.semaem = ns->used_sems;
787 seminfo.semusz = SEMUSZ;
788 seminfo.semaem = SEMAEM;
790 max_id = ipc_get_maxid(&sem_ids(ns));
791 up_read(&sem_ids(ns).rw_mutex);
792 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
794 return (max_id < 0) ? 0: max_id;
799 struct semid64_ds tbuf;
802 if (cmd == SEM_STAT) {
803 sma = sem_lock(ns, semid);
806 id = sma->sem_perm.id;
808 sma = sem_lock_check(ns, semid);
815 if (ipcperms (&sma->sem_perm, S_IRUGO))
818 err = security_sem_semctl(sma, cmd);
822 memset(&tbuf, 0, sizeof(tbuf));
824 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
825 tbuf.sem_otime = sma->sem_otime;
826 tbuf.sem_ctime = sma->sem_ctime;
827 tbuf.sem_nsems = sma->sem_nsems;
829 if (copy_semid_to_user (arg.buf, &tbuf, version))
841 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
842 int cmd, int version, union semun arg)
844 struct sem_array *sma;
847 ushort fast_sem_io[SEMMSL_FAST];
848 ushort* sem_io = fast_sem_io;
850 struct list_head tasks;
852 sma = sem_lock_check(ns, semid);
856 INIT_LIST_HEAD(&tasks);
857 nsems = sma->sem_nsems;
860 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
863 err = security_sem_semctl(sma, cmd);
871 ushort __user *array = arg.array;
874 if(nsems > SEMMSL_FAST) {
875 sem_getref_and_unlock(sma);
877 sem_io = ipc_alloc(sizeof(ushort)*nsems);
883 sem_lock_and_putref(sma);
884 if (sma->sem_perm.deleted) {
891 for (i = 0; i < sma->sem_nsems; i++)
892 sem_io[i] = sma->sem_base[i].semval;
895 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
904 sem_getref_and_unlock(sma);
906 if(nsems > SEMMSL_FAST) {
907 sem_io = ipc_alloc(sizeof(ushort)*nsems);
914 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
920 for (i = 0; i < nsems; i++) {
921 if (sem_io[i] > SEMVMX) {
927 sem_lock_and_putref(sma);
928 if (sma->sem_perm.deleted) {
934 for (i = 0; i < nsems; i++)
935 sma->sem_base[i].semval = sem_io[i];
937 assert_spin_locked(&sma->sem_perm.lock);
938 list_for_each_entry(un, &sma->list_id, list_id) {
939 for (i = 0; i < nsems; i++)
942 sma->sem_ctime = get_seconds();
943 /* maybe some queued-up processes were waiting for this */
944 do_smart_update(sma, NULL, 0, 0, &tasks);
948 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
951 if(semnum < 0 || semnum >= nsems)
954 curr = &sma->sem_base[semnum];
964 err = count_semncnt(sma,semnum);
967 err = count_semzcnt(sma,semnum);
975 if (val > SEMVMX || val < 0)
978 assert_spin_locked(&sma->sem_perm.lock);
979 list_for_each_entry(un, &sma->list_id, list_id)
980 un->semadj[semnum] = 0;
983 curr->sempid = task_tgid_vnr(current);
984 sma->sem_ctime = get_seconds();
985 /* maybe some queued-up processes were waiting for this */
986 do_smart_update(sma, NULL, 0, 0, &tasks);
993 wake_up_sem_queue_do(&tasks);
996 if(sem_io != fast_sem_io)
997 ipc_free(sem_io, sizeof(ushort)*nsems);
1001 static inline unsigned long
1002 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1006 if (copy_from_user(out, buf, sizeof(*out)))
1011 struct semid_ds tbuf_old;
1013 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1016 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1017 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1018 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1028 * This function handles some semctl commands which require the rw_mutex
1029 * to be held in write mode.
1030 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1032 static int semctl_down(struct ipc_namespace *ns, int semid,
1033 int cmd, int version, union semun arg)
1035 struct sem_array *sma;
1037 struct semid64_ds semid64;
1038 struct kern_ipc_perm *ipcp;
1040 if(cmd == IPC_SET) {
1041 if (copy_semid_from_user(&semid64, arg.buf, version))
1045 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
1047 return PTR_ERR(ipcp);
1049 sma = container_of(ipcp, struct sem_array, sem_perm);
1051 err = security_sem_semctl(sma, cmd);
1060 ipc_update_perm(&semid64.sem_perm, ipcp);
1061 sma->sem_ctime = get_seconds();
1070 up_write(&sem_ids(ns).rw_mutex);
1074 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1078 struct ipc_namespace *ns;
1083 version = ipc_parse_version(&cmd);
1084 ns = current->nsproxy->ipc_ns;
1091 err = semctl_nolock(ns, semid, cmd, version, arg);
1100 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1104 err = semctl_down(ns, semid, cmd, version, arg);
1110 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1111 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1113 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1115 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1118 /* If the task doesn't already have a undo_list, then allocate one
1119 * here. We guarantee there is only one thread using this undo list,
1120 * and current is THE ONE
1122 * If this allocation and assignment succeeds, but later
1123 * portions of this code fail, there is no need to free the sem_undo_list.
1124 * Just let it stay associated with the task, and it'll be freed later
1127 * This can block, so callers must hold no locks.
1129 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1131 struct sem_undo_list *undo_list;
1133 undo_list = current->sysvsem.undo_list;
1135 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1136 if (undo_list == NULL)
1138 spin_lock_init(&undo_list->lock);
1139 atomic_set(&undo_list->refcnt, 1);
1140 INIT_LIST_HEAD(&undo_list->list_proc);
1142 current->sysvsem.undo_list = undo_list;
1144 *undo_listp = undo_list;
1148 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1150 struct sem_undo *un;
1152 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1153 if (un->semid == semid)
1159 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1161 struct sem_undo *un;
1163 assert_spin_locked(&ulp->lock);
1165 un = __lookup_undo(ulp, semid);
1167 list_del_rcu(&un->list_proc);
1168 list_add_rcu(&un->list_proc, &ulp->list_proc);
1174 * find_alloc_undo - Lookup (and if not present create) undo array
1176 * @semid: semaphore array id
1178 * The function looks up (and if not present creates) the undo structure.
1179 * The size of the undo structure depends on the size of the semaphore
1180 * array, thus the alloc path is not that straightforward.
1181 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1182 * performs a rcu_read_lock().
1184 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1186 struct sem_array *sma;
1187 struct sem_undo_list *ulp;
1188 struct sem_undo *un, *new;
1192 error = get_undo_list(&ulp);
1194 return ERR_PTR(error);
1197 spin_lock(&ulp->lock);
1198 un = lookup_undo(ulp, semid);
1199 spin_unlock(&ulp->lock);
1200 if (likely(un!=NULL))
1204 /* no undo structure around - allocate one. */
1205 /* step 1: figure out the size of the semaphore array */
1206 sma = sem_lock_check(ns, semid);
1208 return ERR_PTR(PTR_ERR(sma));
1210 nsems = sma->sem_nsems;
1211 sem_getref_and_unlock(sma);
1213 /* step 2: allocate new undo structure */
1214 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1217 return ERR_PTR(-ENOMEM);
1220 /* step 3: Acquire the lock on semaphore array */
1221 sem_lock_and_putref(sma);
1222 if (sma->sem_perm.deleted) {
1225 un = ERR_PTR(-EIDRM);
1228 spin_lock(&ulp->lock);
1231 * step 4: check for races: did someone else allocate the undo struct?
1233 un = lookup_undo(ulp, semid);
1238 /* step 5: initialize & link new undo structure */
1239 new->semadj = (short *) &new[1];
1242 assert_spin_locked(&ulp->lock);
1243 list_add_rcu(&new->list_proc, &ulp->list_proc);
1244 assert_spin_locked(&sma->sem_perm.lock);
1245 list_add(&new->list_id, &sma->list_id);
1249 spin_unlock(&ulp->lock);
1256 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1257 unsigned, nsops, const struct timespec __user *, timeout)
1259 int error = -EINVAL;
1260 struct sem_array *sma;
1261 struct sembuf fast_sops[SEMOPM_FAST];
1262 struct sembuf* sops = fast_sops, *sop;
1263 struct sem_undo *un;
1264 int undos = 0, alter = 0, max;
1265 struct sem_queue queue;
1266 unsigned long jiffies_left = 0;
1267 struct ipc_namespace *ns;
1268 struct list_head tasks;
1270 ns = current->nsproxy->ipc_ns;
1272 if (nsops < 1 || semid < 0)
1274 if (nsops > ns->sc_semopm)
1276 if(nsops > SEMOPM_FAST) {
1277 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1281 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1286 struct timespec _timeout;
1287 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1291 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1292 _timeout.tv_nsec >= 1000000000L) {
1296 jiffies_left = timespec_to_jiffies(&_timeout);
1299 for (sop = sops; sop < sops + nsops; sop++) {
1300 if (sop->sem_num >= max)
1302 if (sop->sem_flg & SEM_UNDO)
1304 if (sop->sem_op != 0)
1309 un = find_alloc_undo(ns, semid);
1311 error = PTR_ERR(un);
1317 INIT_LIST_HEAD(&tasks);
1319 sma = sem_lock_check(ns, semid);
1323 error = PTR_ERR(sma);
1328 * semid identifiers are not unique - find_alloc_undo may have
1329 * allocated an undo structure, it was invalidated by an RMID
1330 * and now a new array with received the same id. Check and fail.
1331 * This case can be detected checking un->semid. The existance of
1332 * "un" itself is guaranteed by rcu.
1336 if (un->semid == -1) {
1338 goto out_unlock_free;
1341 * rcu lock can be released, "un" cannot disappear:
1342 * - sem_lock is acquired, thus IPC_RMID is
1344 * - exit_sem is impossible, it always operates on
1345 * current (or a dead task).
1353 if (max >= sma->sem_nsems)
1354 goto out_unlock_free;
1357 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1358 goto out_unlock_free;
1360 error = security_sem_semop(sma, sops, nsops, alter);
1362 goto out_unlock_free;
1364 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1366 if (alter && error == 0)
1367 do_smart_update(sma, sops, nsops, 1, &tasks);
1369 goto out_unlock_free;
1372 /* We need to sleep on this operation, so we put the current
1373 * task into the pending queue and go to sleep.
1377 queue.nsops = nsops;
1379 queue.pid = task_tgid_vnr(current);
1380 queue.alter = alter;
1382 list_add_tail(&queue.list, &sma->sem_pending);
1384 list_add(&queue.list, &sma->sem_pending);
1388 curr = &sma->sem_base[sops->sem_num];
1391 list_add_tail(&queue.simple_list, &curr->sem_pending);
1393 list_add(&queue.simple_list, &curr->sem_pending);
1395 INIT_LIST_HEAD(&queue.simple_list);
1396 sma->complex_count++;
1399 queue.status = -EINTR;
1400 queue.sleeper = current;
1401 current->state = TASK_INTERRUPTIBLE;
1405 jiffies_left = schedule_timeout(jiffies_left);
1409 error = queue.status;
1410 while(unlikely(error == IN_WAKEUP)) {
1412 error = queue.status;
1415 if (error != -EINTR) {
1416 /* fast path: update_queue already obtained all requested
1421 sma = sem_lock(ns, semid);
1428 * If queue.status != -EINTR we are woken up by another process
1430 error = queue.status;
1431 if (error != -EINTR) {
1432 goto out_unlock_free;
1436 * If an interrupt occurred we have to clean up the queue
1438 if (timeout && jiffies_left == 0)
1440 unlink_queue(sma, &queue);
1445 wake_up_sem_queue_do(&tasks);
1447 if(sops != fast_sops)
1452 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1455 return sys_semtimedop(semid, tsops, nsops, NULL);
1458 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1459 * parent and child tasks.
1462 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1464 struct sem_undo_list *undo_list;
1467 if (clone_flags & CLONE_SYSVSEM) {
1468 error = get_undo_list(&undo_list);
1471 atomic_inc(&undo_list->refcnt);
1472 tsk->sysvsem.undo_list = undo_list;
1474 tsk->sysvsem.undo_list = NULL;
1480 * add semadj values to semaphores, free undo structures.
1481 * undo structures are not freed when semaphore arrays are destroyed
1482 * so some of them may be out of date.
1483 * IMPLEMENTATION NOTE: There is some confusion over whether the
1484 * set of adjustments that needs to be done should be done in an atomic
1485 * manner or not. That is, if we are attempting to decrement the semval
1486 * should we queue up and wait until we can do so legally?
1487 * The original implementation attempted to do this (queue and wait).
1488 * The current implementation does not do so. The POSIX standard
1489 * and SVID should be consulted to determine what behavior is mandated.
1491 void exit_sem(struct task_struct *tsk)
1493 struct sem_undo_list *ulp;
1495 ulp = tsk->sysvsem.undo_list;
1498 tsk->sysvsem.undo_list = NULL;
1500 if (!atomic_dec_and_test(&ulp->refcnt))
1504 struct sem_array *sma;
1505 struct sem_undo *un;
1506 struct list_head tasks;
1511 un = list_entry_rcu(ulp->list_proc.next,
1512 struct sem_undo, list_proc);
1513 if (&un->list_proc == &ulp->list_proc)
1522 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1524 /* exit_sem raced with IPC_RMID, nothing to do */
1528 un = __lookup_undo(ulp, semid);
1530 /* exit_sem raced with IPC_RMID+semget() that created
1531 * exactly the same semid. Nothing to do.
1537 /* remove un from the linked lists */
1538 assert_spin_locked(&sma->sem_perm.lock);
1539 list_del(&un->list_id);
1541 spin_lock(&ulp->lock);
1542 list_del_rcu(&un->list_proc);
1543 spin_unlock(&ulp->lock);
1545 /* perform adjustments registered in un */
1546 for (i = 0; i < sma->sem_nsems; i++) {
1547 struct sem * semaphore = &sma->sem_base[i];
1548 if (un->semadj[i]) {
1549 semaphore->semval += un->semadj[i];
1551 * Range checks of the new semaphore value,
1552 * not defined by sus:
1553 * - Some unices ignore the undo entirely
1554 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1555 * - some cap the value (e.g. FreeBSD caps
1556 * at 0, but doesn't enforce SEMVMX)
1558 * Linux caps the semaphore value, both at 0
1561 * Manfred <manfred@colorfullife.com>
1563 if (semaphore->semval < 0)
1564 semaphore->semval = 0;
1565 if (semaphore->semval > SEMVMX)
1566 semaphore->semval = SEMVMX;
1567 semaphore->sempid = task_tgid_vnr(current);
1570 /* maybe some queued-up processes were waiting for this */
1571 INIT_LIST_HEAD(&tasks);
1572 do_smart_update(sma, NULL, 0, 1, &tasks);
1574 wake_up_sem_queue_do(&tasks);
1576 call_rcu(&un->rcu, free_un);
1581 #ifdef CONFIG_PROC_FS
1582 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1584 struct sem_array *sma = it;
1586 return seq_printf(s,
1587 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",