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;
384 sma->sem_otime = get_seconds();
392 if (sop->sem_flg & IPC_NOWAIT)
399 while (sop >= sops) {
400 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
408 * Wake up a process waiting on the sem queue with a given error.
409 * The queue is invalid (may not be accessed) after the function returns.
411 static void wake_up_sem_queue(struct sem_queue *q, int error)
414 * Hold preempt off so that we don't get preempted and have the
415 * wakee busy-wait until we're scheduled back on. We're holding
416 * locks here so it may not strictly be needed, however if the
417 * locks become preemptible then this prevents such a problem.
420 q->status = IN_WAKEUP;
421 wake_up_process(q->sleeper);
422 /* hands-off: q can disappear immediately after writing q->status. */
428 static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
432 list_del(&q->simple_list);
434 sma->complex_count--;
437 /** check_restart(sma, q)
438 * @sma: semaphore array
439 * @q: the operation that just completed
441 * update_queue is O(N^2) when it restarts scanning the whole queue of
442 * waiting operations. Therefore this function checks if the restart is
443 * really necessary. It is called after a previously waiting operation
446 static int check_restart(struct sem_array *sma, struct sem_queue *q)
451 /* if the operation didn't modify the array, then no restart */
455 /* pending complex operations are too difficult to analyse */
456 if (sma->complex_count)
459 /* we were a sleeping complex operation. Too difficult */
463 curr = sma->sem_base + q->sops[0].sem_num;
465 /* No-one waits on this queue */
466 if (list_empty(&curr->sem_pending))
469 /* the new semaphore value */
471 /* It is impossible that someone waits for the new value:
472 * - q is a previously sleeping simple operation that
473 * altered the array. It must be a decrement, because
474 * simple increments never sleep.
475 * - The value is not 0, thus wait-for-zero won't proceed.
476 * - If there are older (higher priority) decrements
477 * in the queue, then they have observed the original
478 * semval value and couldn't proceed. The operation
479 * decremented to value - thus they won't proceed either.
481 BUG_ON(q->sops[0].sem_op >= 0);
485 * semval is 0. Check if there are wait-for-zero semops.
486 * They must be the first entries in the per-semaphore simple queue
488 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
489 BUG_ON(h->nsops != 1);
490 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
492 /* Yes, there is a wait-for-zero semop. Restart */
493 if (h->sops[0].sem_op == 0)
496 /* Again - no-one is waiting for the new value. */
502 * update_queue(sma, semnum): Look for tasks that can be completed.
503 * @sma: semaphore array.
504 * @semnum: semaphore that was modified.
506 * update_queue must be called after a semaphore in a semaphore array
507 * was modified. If multiple semaphore were modified, then @semnum
510 static void update_queue(struct sem_array *sma, int semnum)
513 struct list_head *walk;
514 struct list_head *pending_list;
517 /* if there are complex operations around, then knowing the semaphore
518 * that was modified doesn't help us. Assume that multiple semaphores
521 if (sma->complex_count)
525 pending_list = &sma->sem_pending;
526 offset = offsetof(struct sem_queue, list);
528 pending_list = &sma->sem_base[semnum].sem_pending;
529 offset = offsetof(struct sem_queue, simple_list);
533 walk = pending_list->next;
534 while (walk != pending_list) {
537 q = (struct sem_queue *)((char *)walk - offset);
540 /* If we are scanning the single sop, per-semaphore list of
541 * one semaphore and that semaphore is 0, then it is not
542 * necessary to scan the "alter" entries: simple increments
543 * that affect only one entry succeed immediately and cannot
544 * be in the per semaphore pending queue, and decrements
545 * cannot be successful if the value is already 0.
547 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
551 error = try_atomic_semop(sma, q->sops, q->nsops,
554 /* Does q->sleeper still need to sleep? */
558 unlink_queue(sma, q);
563 restart = check_restart(sma, q);
565 wake_up_sem_queue(q, error);
571 /** do_smart_update(sma, sops, nsops): Optimized update_queue
572 * @sma: semaphore array
573 * @sops: operations that were performed
574 * @nsops: number of operations
576 * do_smart_update() does the required called to update_queue, based on the
577 * actual changes that were performed on the semaphore array.
579 static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops)
583 if (sma->complex_count || sops == NULL) {
584 update_queue(sma, -1);
588 for (i = 0; i < nsops; i++) {
589 if (sops[i].sem_op > 0 ||
590 (sops[i].sem_op < 0 &&
591 sma->sem_base[sops[i].sem_num].semval == 0))
592 update_queue(sma, sops[i].sem_num);
597 /* The following counts are associated to each semaphore:
598 * semncnt number of tasks waiting on semval being nonzero
599 * semzcnt number of tasks waiting on semval being zero
600 * This model assumes that a task waits on exactly one semaphore.
601 * Since semaphore operations are to be performed atomically, tasks actually
602 * wait on a whole sequence of semaphores simultaneously.
603 * The counts we return here are a rough approximation, but still
604 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
606 static int count_semncnt (struct sem_array * sma, ushort semnum)
609 struct sem_queue * q;
612 list_for_each_entry(q, &sma->sem_pending, list) {
613 struct sembuf * sops = q->sops;
614 int nsops = q->nsops;
616 for (i = 0; i < nsops; i++)
617 if (sops[i].sem_num == semnum
618 && (sops[i].sem_op < 0)
619 && !(sops[i].sem_flg & IPC_NOWAIT))
625 static int count_semzcnt (struct sem_array * sma, ushort semnum)
628 struct sem_queue * q;
631 list_for_each_entry(q, &sma->sem_pending, list) {
632 struct sembuf * sops = q->sops;
633 int nsops = q->nsops;
635 for (i = 0; i < nsops; i++)
636 if (sops[i].sem_num == semnum
637 && (sops[i].sem_op == 0)
638 && !(sops[i].sem_flg & IPC_NOWAIT))
644 static void free_un(struct rcu_head *head)
646 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
650 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
651 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
652 * remains locked on exit.
654 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
656 struct sem_undo *un, *tu;
657 struct sem_queue *q, *tq;
658 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
660 /* Free the existing undo structures for this semaphore set. */
661 assert_spin_locked(&sma->sem_perm.lock);
662 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
663 list_del(&un->list_id);
664 spin_lock(&un->ulp->lock);
666 list_del_rcu(&un->list_proc);
667 spin_unlock(&un->ulp->lock);
668 call_rcu(&un->rcu, free_un);
671 /* Wake up all pending processes and let them fail with EIDRM. */
672 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
673 unlink_queue(sma, q);
674 wake_up_sem_queue(q, -EIDRM);
677 /* Remove the semaphore set from the IDR */
681 ns->used_sems -= sma->sem_nsems;
682 security_sem_free(sma);
686 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
690 return copy_to_user(buf, in, sizeof(*in));
695 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
697 out.sem_otime = in->sem_otime;
698 out.sem_ctime = in->sem_ctime;
699 out.sem_nsems = in->sem_nsems;
701 return copy_to_user(buf, &out, sizeof(out));
708 static int semctl_nolock(struct ipc_namespace *ns, int semid,
709 int cmd, int version, union semun arg)
712 struct sem_array *sma;
718 struct seminfo seminfo;
721 err = security_sem_semctl(NULL, cmd);
725 memset(&seminfo,0,sizeof(seminfo));
726 seminfo.semmni = ns->sc_semmni;
727 seminfo.semmns = ns->sc_semmns;
728 seminfo.semmsl = ns->sc_semmsl;
729 seminfo.semopm = ns->sc_semopm;
730 seminfo.semvmx = SEMVMX;
731 seminfo.semmnu = SEMMNU;
732 seminfo.semmap = SEMMAP;
733 seminfo.semume = SEMUME;
734 down_read(&sem_ids(ns).rw_mutex);
735 if (cmd == SEM_INFO) {
736 seminfo.semusz = sem_ids(ns).in_use;
737 seminfo.semaem = ns->used_sems;
739 seminfo.semusz = SEMUSZ;
740 seminfo.semaem = SEMAEM;
742 max_id = ipc_get_maxid(&sem_ids(ns));
743 up_read(&sem_ids(ns).rw_mutex);
744 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
746 return (max_id < 0) ? 0: max_id;
751 struct semid64_ds tbuf;
754 if (cmd == SEM_STAT) {
755 sma = sem_lock(ns, semid);
758 id = sma->sem_perm.id;
760 sma = sem_lock_check(ns, semid);
767 if (ipcperms (&sma->sem_perm, S_IRUGO))
770 err = security_sem_semctl(sma, cmd);
774 memset(&tbuf, 0, sizeof(tbuf));
776 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
777 tbuf.sem_otime = sma->sem_otime;
778 tbuf.sem_ctime = sma->sem_ctime;
779 tbuf.sem_nsems = sma->sem_nsems;
781 if (copy_semid_to_user (arg.buf, &tbuf, version))
793 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
794 int cmd, int version, union semun arg)
796 struct sem_array *sma;
799 ushort fast_sem_io[SEMMSL_FAST];
800 ushort* sem_io = fast_sem_io;
803 sma = sem_lock_check(ns, semid);
807 nsems = sma->sem_nsems;
810 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
813 err = security_sem_semctl(sma, cmd);
821 ushort __user *array = arg.array;
824 if(nsems > SEMMSL_FAST) {
825 sem_getref_and_unlock(sma);
827 sem_io = ipc_alloc(sizeof(ushort)*nsems);
833 sem_lock_and_putref(sma);
834 if (sma->sem_perm.deleted) {
841 for (i = 0; i < sma->sem_nsems; i++)
842 sem_io[i] = sma->sem_base[i].semval;
845 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
854 sem_getref_and_unlock(sma);
856 if(nsems > SEMMSL_FAST) {
857 sem_io = ipc_alloc(sizeof(ushort)*nsems);
864 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
870 for (i = 0; i < nsems; i++) {
871 if (sem_io[i] > SEMVMX) {
877 sem_lock_and_putref(sma);
878 if (sma->sem_perm.deleted) {
884 for (i = 0; i < nsems; i++)
885 sma->sem_base[i].semval = sem_io[i];
887 assert_spin_locked(&sma->sem_perm.lock);
888 list_for_each_entry(un, &sma->list_id, list_id) {
889 for (i = 0; i < nsems; i++)
892 sma->sem_ctime = get_seconds();
893 /* maybe some queued-up processes were waiting for this */
894 update_queue(sma, -1);
898 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
901 if(semnum < 0 || semnum >= nsems)
904 curr = &sma->sem_base[semnum];
914 err = count_semncnt(sma,semnum);
917 err = count_semzcnt(sma,semnum);
925 if (val > SEMVMX || val < 0)
928 assert_spin_locked(&sma->sem_perm.lock);
929 list_for_each_entry(un, &sma->list_id, list_id)
930 un->semadj[semnum] = 0;
933 curr->sempid = task_tgid_vnr(current);
934 sma->sem_ctime = get_seconds();
935 /* maybe some queued-up processes were waiting for this */
936 update_queue(sma, semnum);
944 if(sem_io != fast_sem_io)
945 ipc_free(sem_io, sizeof(ushort)*nsems);
949 static inline unsigned long
950 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
954 if (copy_from_user(out, buf, sizeof(*out)))
959 struct semid_ds tbuf_old;
961 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
964 out->sem_perm.uid = tbuf_old.sem_perm.uid;
965 out->sem_perm.gid = tbuf_old.sem_perm.gid;
966 out->sem_perm.mode = tbuf_old.sem_perm.mode;
976 * This function handles some semctl commands which require the rw_mutex
977 * to be held in write mode.
978 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
980 static int semctl_down(struct ipc_namespace *ns, int semid,
981 int cmd, int version, union semun arg)
983 struct sem_array *sma;
985 struct semid64_ds semid64;
986 struct kern_ipc_perm *ipcp;
989 if (copy_semid_from_user(&semid64, arg.buf, version))
993 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
995 return PTR_ERR(ipcp);
997 sma = container_of(ipcp, struct sem_array, sem_perm);
999 err = security_sem_semctl(sma, cmd);
1008 ipc_update_perm(&semid64.sem_perm, ipcp);
1009 sma->sem_ctime = get_seconds();
1018 up_write(&sem_ids(ns).rw_mutex);
1022 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1026 struct ipc_namespace *ns;
1031 version = ipc_parse_version(&cmd);
1032 ns = current->nsproxy->ipc_ns;
1039 err = semctl_nolock(ns, semid, cmd, version, arg);
1048 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1052 err = semctl_down(ns, semid, cmd, version, arg);
1058 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1059 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1061 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1063 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1066 /* If the task doesn't already have a undo_list, then allocate one
1067 * here. We guarantee there is only one thread using this undo list,
1068 * and current is THE ONE
1070 * If this allocation and assignment succeeds, but later
1071 * portions of this code fail, there is no need to free the sem_undo_list.
1072 * Just let it stay associated with the task, and it'll be freed later
1075 * This can block, so callers must hold no locks.
1077 static inline int get_undo_list(struct sem_undo_list **undo_listp)
1079 struct sem_undo_list *undo_list;
1081 undo_list = current->sysvsem.undo_list;
1083 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1084 if (undo_list == NULL)
1086 spin_lock_init(&undo_list->lock);
1087 atomic_set(&undo_list->refcnt, 1);
1088 INIT_LIST_HEAD(&undo_list->list_proc);
1090 current->sysvsem.undo_list = undo_list;
1092 *undo_listp = undo_list;
1096 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1098 struct sem_undo *un;
1100 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1101 if (un->semid == semid)
1107 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1109 struct sem_undo *un;
1111 assert_spin_locked(&ulp->lock);
1113 un = __lookup_undo(ulp, semid);
1115 list_del_rcu(&un->list_proc);
1116 list_add_rcu(&un->list_proc, &ulp->list_proc);
1122 * find_alloc_undo - Lookup (and if not present create) undo array
1124 * @semid: semaphore array id
1126 * The function looks up (and if not present creates) the undo structure.
1127 * The size of the undo structure depends on the size of the semaphore
1128 * array, thus the alloc path is not that straightforward.
1129 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1130 * performs a rcu_read_lock().
1132 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1134 struct sem_array *sma;
1135 struct sem_undo_list *ulp;
1136 struct sem_undo *un, *new;
1140 error = get_undo_list(&ulp);
1142 return ERR_PTR(error);
1145 spin_lock(&ulp->lock);
1146 un = lookup_undo(ulp, semid);
1147 spin_unlock(&ulp->lock);
1148 if (likely(un!=NULL))
1152 /* no undo structure around - allocate one. */
1153 /* step 1: figure out the size of the semaphore array */
1154 sma = sem_lock_check(ns, semid);
1156 return ERR_PTR(PTR_ERR(sma));
1158 nsems = sma->sem_nsems;
1159 sem_getref_and_unlock(sma);
1161 /* step 2: allocate new undo structure */
1162 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1165 return ERR_PTR(-ENOMEM);
1168 /* step 3: Acquire the lock on semaphore array */
1169 sem_lock_and_putref(sma);
1170 if (sma->sem_perm.deleted) {
1173 un = ERR_PTR(-EIDRM);
1176 spin_lock(&ulp->lock);
1179 * step 4: check for races: did someone else allocate the undo struct?
1181 un = lookup_undo(ulp, semid);
1186 /* step 5: initialize & link new undo structure */
1187 new->semadj = (short *) &new[1];
1190 assert_spin_locked(&ulp->lock);
1191 list_add_rcu(&new->list_proc, &ulp->list_proc);
1192 assert_spin_locked(&sma->sem_perm.lock);
1193 list_add(&new->list_id, &sma->list_id);
1197 spin_unlock(&ulp->lock);
1204 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1205 unsigned, nsops, const struct timespec __user *, timeout)
1207 int error = -EINVAL;
1208 struct sem_array *sma;
1209 struct sembuf fast_sops[SEMOPM_FAST];
1210 struct sembuf* sops = fast_sops, *sop;
1211 struct sem_undo *un;
1212 int undos = 0, alter = 0, max;
1213 struct sem_queue queue;
1214 unsigned long jiffies_left = 0;
1215 struct ipc_namespace *ns;
1217 ns = current->nsproxy->ipc_ns;
1219 if (nsops < 1 || semid < 0)
1221 if (nsops > ns->sc_semopm)
1223 if(nsops > SEMOPM_FAST) {
1224 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1228 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1233 struct timespec _timeout;
1234 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1238 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1239 _timeout.tv_nsec >= 1000000000L) {
1243 jiffies_left = timespec_to_jiffies(&_timeout);
1246 for (sop = sops; sop < sops + nsops; sop++) {
1247 if (sop->sem_num >= max)
1249 if (sop->sem_flg & SEM_UNDO)
1251 if (sop->sem_op != 0)
1256 un = find_alloc_undo(ns, semid);
1258 error = PTR_ERR(un);
1264 sma = sem_lock_check(ns, semid);
1268 error = PTR_ERR(sma);
1273 * semid identifiers are not unique - find_alloc_undo may have
1274 * allocated an undo structure, it was invalidated by an RMID
1275 * and now a new array with received the same id. Check and fail.
1276 * This case can be detected checking un->semid. The existance of
1277 * "un" itself is guaranteed by rcu.
1281 if (un->semid == -1) {
1283 goto out_unlock_free;
1286 * rcu lock can be released, "un" cannot disappear:
1287 * - sem_lock is acquired, thus IPC_RMID is
1289 * - exit_sem is impossible, it always operates on
1290 * current (or a dead task).
1298 if (max >= sma->sem_nsems)
1299 goto out_unlock_free;
1302 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1303 goto out_unlock_free;
1305 error = security_sem_semop(sma, sops, nsops, alter);
1307 goto out_unlock_free;
1309 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1311 if (alter && error == 0)
1312 do_smart_update(sma, sops, nsops);
1314 goto out_unlock_free;
1317 /* We need to sleep on this operation, so we put the current
1318 * task into the pending queue and go to sleep.
1322 queue.nsops = nsops;
1324 queue.pid = task_tgid_vnr(current);
1325 queue.alter = alter;
1327 list_add_tail(&queue.list, &sma->sem_pending);
1329 list_add(&queue.list, &sma->sem_pending);
1333 curr = &sma->sem_base[sops->sem_num];
1336 list_add_tail(&queue.simple_list, &curr->sem_pending);
1338 list_add(&queue.simple_list, &curr->sem_pending);
1340 INIT_LIST_HEAD(&queue.simple_list);
1341 sma->complex_count++;
1344 queue.status = -EINTR;
1345 queue.sleeper = current;
1346 current->state = TASK_INTERRUPTIBLE;
1350 jiffies_left = schedule_timeout(jiffies_left);
1354 error = queue.status;
1355 while(unlikely(error == IN_WAKEUP)) {
1357 error = queue.status;
1360 if (error != -EINTR) {
1361 /* fast path: update_queue already obtained all requested
1366 sma = sem_lock(ns, semid);
1373 * If queue.status != -EINTR we are woken up by another process
1375 error = queue.status;
1376 if (error != -EINTR) {
1377 goto out_unlock_free;
1381 * If an interrupt occurred we have to clean up the queue
1383 if (timeout && jiffies_left == 0)
1385 unlink_queue(sma, &queue);
1390 if(sops != fast_sops)
1395 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1398 return sys_semtimedop(semid, tsops, nsops, NULL);
1401 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1402 * parent and child tasks.
1405 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1407 struct sem_undo_list *undo_list;
1410 if (clone_flags & CLONE_SYSVSEM) {
1411 error = get_undo_list(&undo_list);
1414 atomic_inc(&undo_list->refcnt);
1415 tsk->sysvsem.undo_list = undo_list;
1417 tsk->sysvsem.undo_list = NULL;
1423 * add semadj values to semaphores, free undo structures.
1424 * undo structures are not freed when semaphore arrays are destroyed
1425 * so some of them may be out of date.
1426 * IMPLEMENTATION NOTE: There is some confusion over whether the
1427 * set of adjustments that needs to be done should be done in an atomic
1428 * manner or not. That is, if we are attempting to decrement the semval
1429 * should we queue up and wait until we can do so legally?
1430 * The original implementation attempted to do this (queue and wait).
1431 * The current implementation does not do so. The POSIX standard
1432 * and SVID should be consulted to determine what behavior is mandated.
1434 void exit_sem(struct task_struct *tsk)
1436 struct sem_undo_list *ulp;
1438 ulp = tsk->sysvsem.undo_list;
1441 tsk->sysvsem.undo_list = NULL;
1443 if (!atomic_dec_and_test(&ulp->refcnt))
1447 struct sem_array *sma;
1448 struct sem_undo *un;
1453 un = list_entry_rcu(ulp->list_proc.next,
1454 struct sem_undo, list_proc);
1455 if (&un->list_proc == &ulp->list_proc)
1464 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1466 /* exit_sem raced with IPC_RMID, nothing to do */
1470 un = __lookup_undo(ulp, semid);
1472 /* exit_sem raced with IPC_RMID+semget() that created
1473 * exactly the same semid. Nothing to do.
1479 /* remove un from the linked lists */
1480 assert_spin_locked(&sma->sem_perm.lock);
1481 list_del(&un->list_id);
1483 spin_lock(&ulp->lock);
1484 list_del_rcu(&un->list_proc);
1485 spin_unlock(&ulp->lock);
1487 /* perform adjustments registered in un */
1488 for (i = 0; i < sma->sem_nsems; i++) {
1489 struct sem * semaphore = &sma->sem_base[i];
1490 if (un->semadj[i]) {
1491 semaphore->semval += un->semadj[i];
1493 * Range checks of the new semaphore value,
1494 * not defined by sus:
1495 * - Some unices ignore the undo entirely
1496 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1497 * - some cap the value (e.g. FreeBSD caps
1498 * at 0, but doesn't enforce SEMVMX)
1500 * Linux caps the semaphore value, both at 0
1503 * Manfred <manfred@colorfullife.com>
1505 if (semaphore->semval < 0)
1506 semaphore->semval = 0;
1507 if (semaphore->semval > SEMVMX)
1508 semaphore->semval = SEMVMX;
1509 semaphore->sempid = task_tgid_vnr(current);
1512 sma->sem_otime = get_seconds();
1513 /* maybe some queued-up processes were waiting for this */
1514 update_queue(sma, -1);
1517 call_rcu(&un->rcu, free_un);
1522 #ifdef CONFIG_PROC_FS
1523 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1525 struct sem_array *sma = it;
1527 return seq_printf(s,
1528 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",