leds-gpio: fix default state handling on OF platforms
[linux-2.6/mini2440.git] / ipc / sem.c
blob2f2a47959576964c200e7493464818da982c3e7a
1 /*
2 * linux/ipc/sem.c
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
62 * Lockless wakeup
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>
68 * namespaces support
69 * OpenVZ, SWsoft Inc.
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>
88 #include "util.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 *);
97 #ifdef CONFIG_PROC_FS
98 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
99 #endif
101 #define SEMMSL_FAST 256 /* 512 bytes on stack */
102 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
105 * linked list protection:
106 * sem_undo.id_next,
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;
124 ns->used_sems = 0;
125 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
128 #ifdef CONFIG_IPC_NS
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);
134 #endif
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
146 * is not held.
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);
152 if (IS_ERR(ipcp))
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,
159 int id)
161 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
163 if (IS_ERR(ipcp))
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);
172 ipc_rcu_putref(sma);
175 static inline void sem_getref_and_unlock(struct sem_array *sma)
177 ipc_rcu_getref(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);
184 ipc_rcu_putref(sma);
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
212 * races:
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).
225 #define IN_WAKEUP 1
228 * newary - Create a new semaphore set
229 * @ns: namespace
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)
237 int id;
238 int retval;
239 struct sem_array *sma;
240 int size;
241 key_t key = params->key;
242 int nsems = params->u.nsems;
243 int semflg = params->flg;
245 if (!nsems)
246 return -EINVAL;
247 if (ns->used_sems + nsems > ns->sc_semmns)
248 return -ENOSPC;
250 size = sizeof (*sma) + nsems * sizeof (struct sem);
251 sma = ipc_rcu_alloc(size);
252 if (!sma) {
253 return -ENOMEM;
255 memset (sma, 0, size);
257 sma->sem_perm.mode = (semflg & S_IRWXUGO);
258 sma->sem_perm.key = key;
260 sma->sem_perm.security = NULL;
261 retval = security_sem_alloc(sma);
262 if (retval) {
263 ipc_rcu_putref(sma);
264 return retval;
267 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
268 if (id < 0) {
269 security_sem_free(sma);
270 ipc_rcu_putref(sma);
271 return id;
273 ns->used_sems += nsems;
275 sma->sem_base = (struct sem *) &sma[1];
276 INIT_LIST_HEAD(&sma->sem_pending);
277 INIT_LIST_HEAD(&sma->list_id);
278 sma->sem_nsems = nsems;
279 sma->sem_ctime = get_seconds();
280 sem_unlock(sma);
282 return sma->sem_perm.id;
287 * Called with sem_ids.rw_mutex and ipcp locked.
289 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
291 struct sem_array *sma;
293 sma = container_of(ipcp, struct sem_array, sem_perm);
294 return security_sem_associate(sma, semflg);
298 * Called with sem_ids.rw_mutex and ipcp locked.
300 static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
301 struct ipc_params *params)
303 struct sem_array *sma;
305 sma = container_of(ipcp, struct sem_array, sem_perm);
306 if (params->u.nsems > sma->sem_nsems)
307 return -EINVAL;
309 return 0;
312 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
314 struct ipc_namespace *ns;
315 struct ipc_ops sem_ops;
316 struct ipc_params sem_params;
318 ns = current->nsproxy->ipc_ns;
320 if (nsems < 0 || nsems > ns->sc_semmsl)
321 return -EINVAL;
323 sem_ops.getnew = newary;
324 sem_ops.associate = sem_security;
325 sem_ops.more_checks = sem_more_checks;
327 sem_params.key = key;
328 sem_params.flg = semflg;
329 sem_params.u.nsems = nsems;
331 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
335 * Determine whether a sequence of semaphore operations would succeed
336 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
339 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
340 int nsops, struct sem_undo *un, int pid)
342 int result, sem_op;
343 struct sembuf *sop;
344 struct sem * curr;
346 for (sop = sops; sop < sops + nsops; sop++) {
347 curr = sma->sem_base + sop->sem_num;
348 sem_op = sop->sem_op;
349 result = curr->semval;
351 if (!sem_op && result)
352 goto would_block;
354 result += sem_op;
355 if (result < 0)
356 goto would_block;
357 if (result > SEMVMX)
358 goto out_of_range;
359 if (sop->sem_flg & SEM_UNDO) {
360 int undo = un->semadj[sop->sem_num] - sem_op;
362 * Exceeding the undo range is an error.
364 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
365 goto out_of_range;
367 curr->semval = result;
370 sop--;
371 while (sop >= sops) {
372 sma->sem_base[sop->sem_num].sempid = pid;
373 if (sop->sem_flg & SEM_UNDO)
374 un->semadj[sop->sem_num] -= sop->sem_op;
375 sop--;
378 sma->sem_otime = get_seconds();
379 return 0;
381 out_of_range:
382 result = -ERANGE;
383 goto undo;
385 would_block:
386 if (sop->sem_flg & IPC_NOWAIT)
387 result = -EAGAIN;
388 else
389 result = 1;
391 undo:
392 sop--;
393 while (sop >= sops) {
394 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
395 sop--;
398 return result;
401 /* Go through the pending queue for the indicated semaphore
402 * looking for tasks that can be completed.
404 static void update_queue (struct sem_array * sma)
406 int error;
407 struct sem_queue * q;
409 q = list_entry(sma->sem_pending.next, struct sem_queue, list);
410 while (&q->list != &sma->sem_pending) {
411 error = try_atomic_semop(sma, q->sops, q->nsops,
412 q->undo, q->pid);
414 /* Does q->sleeper still need to sleep? */
415 if (error <= 0) {
416 struct sem_queue *n;
419 * Continue scanning. The next operation
420 * that must be checked depends on the type of the
421 * completed operation:
422 * - if the operation modified the array, then
423 * restart from the head of the queue and
424 * check for threads that might be waiting
425 * for semaphore values to become 0.
426 * - if the operation didn't modify the array,
427 * then just continue.
428 * The order of list_del() and reading ->next
429 * is crucial: In the former case, the list_del()
430 * must be done first [because we might be the
431 * first entry in ->sem_pending], in the latter
432 * case the list_del() must be done last
433 * [because the list is invalid after the list_del()]
435 if (q->alter) {
436 list_del(&q->list);
437 n = list_entry(sma->sem_pending.next,
438 struct sem_queue, list);
439 } else {
440 n = list_entry(q->list.next, struct sem_queue,
441 list);
442 list_del(&q->list);
445 /* wake up the waiting thread */
446 q->status = IN_WAKEUP;
448 wake_up_process(q->sleeper);
449 /* hands-off: q will disappear immediately after
450 * writing q->status.
452 smp_wmb();
453 q->status = error;
454 q = n;
455 } else {
456 q = list_entry(q->list.next, struct sem_queue, list);
461 /* The following counts are associated to each semaphore:
462 * semncnt number of tasks waiting on semval being nonzero
463 * semzcnt number of tasks waiting on semval being zero
464 * This model assumes that a task waits on exactly one semaphore.
465 * Since semaphore operations are to be performed atomically, tasks actually
466 * wait on a whole sequence of semaphores simultaneously.
467 * The counts we return here are a rough approximation, but still
468 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
470 static int count_semncnt (struct sem_array * sma, ushort semnum)
472 int semncnt;
473 struct sem_queue * q;
475 semncnt = 0;
476 list_for_each_entry(q, &sma->sem_pending, list) {
477 struct sembuf * sops = q->sops;
478 int nsops = q->nsops;
479 int i;
480 for (i = 0; i < nsops; i++)
481 if (sops[i].sem_num == semnum
482 && (sops[i].sem_op < 0)
483 && !(sops[i].sem_flg & IPC_NOWAIT))
484 semncnt++;
486 return semncnt;
489 static int count_semzcnt (struct sem_array * sma, ushort semnum)
491 int semzcnt;
492 struct sem_queue * q;
494 semzcnt = 0;
495 list_for_each_entry(q, &sma->sem_pending, list) {
496 struct sembuf * sops = q->sops;
497 int nsops = q->nsops;
498 int i;
499 for (i = 0; i < nsops; i++)
500 if (sops[i].sem_num == semnum
501 && (sops[i].sem_op == 0)
502 && !(sops[i].sem_flg & IPC_NOWAIT))
503 semzcnt++;
505 return semzcnt;
508 static void free_un(struct rcu_head *head)
510 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
511 kfree(un);
514 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
515 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
516 * remains locked on exit.
518 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
520 struct sem_undo *un, *tu;
521 struct sem_queue *q, *tq;
522 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
524 /* Free the existing undo structures for this semaphore set. */
525 assert_spin_locked(&sma->sem_perm.lock);
526 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
527 list_del(&un->list_id);
528 spin_lock(&un->ulp->lock);
529 un->semid = -1;
530 list_del_rcu(&un->list_proc);
531 spin_unlock(&un->ulp->lock);
532 call_rcu(&un->rcu, free_un);
535 /* Wake up all pending processes and let them fail with EIDRM. */
536 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
537 list_del(&q->list);
539 q->status = IN_WAKEUP;
540 wake_up_process(q->sleeper); /* doesn't sleep */
541 smp_wmb();
542 q->status = -EIDRM; /* hands-off q */
545 /* Remove the semaphore set from the IDR */
546 sem_rmid(ns, sma);
547 sem_unlock(sma);
549 ns->used_sems -= sma->sem_nsems;
550 security_sem_free(sma);
551 ipc_rcu_putref(sma);
554 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
556 switch(version) {
557 case IPC_64:
558 return copy_to_user(buf, in, sizeof(*in));
559 case IPC_OLD:
561 struct semid_ds out;
563 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
565 out.sem_otime = in->sem_otime;
566 out.sem_ctime = in->sem_ctime;
567 out.sem_nsems = in->sem_nsems;
569 return copy_to_user(buf, &out, sizeof(out));
571 default:
572 return -EINVAL;
576 static int semctl_nolock(struct ipc_namespace *ns, int semid,
577 int cmd, int version, union semun arg)
579 int err = -EINVAL;
580 struct sem_array *sma;
582 switch(cmd) {
583 case IPC_INFO:
584 case SEM_INFO:
586 struct seminfo seminfo;
587 int max_id;
589 err = security_sem_semctl(NULL, cmd);
590 if (err)
591 return err;
593 memset(&seminfo,0,sizeof(seminfo));
594 seminfo.semmni = ns->sc_semmni;
595 seminfo.semmns = ns->sc_semmns;
596 seminfo.semmsl = ns->sc_semmsl;
597 seminfo.semopm = ns->sc_semopm;
598 seminfo.semvmx = SEMVMX;
599 seminfo.semmnu = SEMMNU;
600 seminfo.semmap = SEMMAP;
601 seminfo.semume = SEMUME;
602 down_read(&sem_ids(ns).rw_mutex);
603 if (cmd == SEM_INFO) {
604 seminfo.semusz = sem_ids(ns).in_use;
605 seminfo.semaem = ns->used_sems;
606 } else {
607 seminfo.semusz = SEMUSZ;
608 seminfo.semaem = SEMAEM;
610 max_id = ipc_get_maxid(&sem_ids(ns));
611 up_read(&sem_ids(ns).rw_mutex);
612 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
613 return -EFAULT;
614 return (max_id < 0) ? 0: max_id;
616 case IPC_STAT:
617 case SEM_STAT:
619 struct semid64_ds tbuf;
620 int id;
622 if (cmd == SEM_STAT) {
623 sma = sem_lock(ns, semid);
624 if (IS_ERR(sma))
625 return PTR_ERR(sma);
626 id = sma->sem_perm.id;
627 } else {
628 sma = sem_lock_check(ns, semid);
629 if (IS_ERR(sma))
630 return PTR_ERR(sma);
631 id = 0;
634 err = -EACCES;
635 if (ipcperms (&sma->sem_perm, S_IRUGO))
636 goto out_unlock;
638 err = security_sem_semctl(sma, cmd);
639 if (err)
640 goto out_unlock;
642 memset(&tbuf, 0, sizeof(tbuf));
644 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
645 tbuf.sem_otime = sma->sem_otime;
646 tbuf.sem_ctime = sma->sem_ctime;
647 tbuf.sem_nsems = sma->sem_nsems;
648 sem_unlock(sma);
649 if (copy_semid_to_user (arg.buf, &tbuf, version))
650 return -EFAULT;
651 return id;
653 default:
654 return -EINVAL;
656 return err;
657 out_unlock:
658 sem_unlock(sma);
659 return err;
662 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
663 int cmd, int version, union semun arg)
665 struct sem_array *sma;
666 struct sem* curr;
667 int err;
668 ushort fast_sem_io[SEMMSL_FAST];
669 ushort* sem_io = fast_sem_io;
670 int nsems;
672 sma = sem_lock_check(ns, semid);
673 if (IS_ERR(sma))
674 return PTR_ERR(sma);
676 nsems = sma->sem_nsems;
678 err = -EACCES;
679 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
680 goto out_unlock;
682 err = security_sem_semctl(sma, cmd);
683 if (err)
684 goto out_unlock;
686 err = -EACCES;
687 switch (cmd) {
688 case GETALL:
690 ushort __user *array = arg.array;
691 int i;
693 if(nsems > SEMMSL_FAST) {
694 sem_getref_and_unlock(sma);
696 sem_io = ipc_alloc(sizeof(ushort)*nsems);
697 if(sem_io == NULL) {
698 sem_putref(sma);
699 return -ENOMEM;
702 sem_lock_and_putref(sma);
703 if (sma->sem_perm.deleted) {
704 sem_unlock(sma);
705 err = -EIDRM;
706 goto out_free;
710 for (i = 0; i < sma->sem_nsems; i++)
711 sem_io[i] = sma->sem_base[i].semval;
712 sem_unlock(sma);
713 err = 0;
714 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
715 err = -EFAULT;
716 goto out_free;
718 case SETALL:
720 int i;
721 struct sem_undo *un;
723 sem_getref_and_unlock(sma);
725 if(nsems > SEMMSL_FAST) {
726 sem_io = ipc_alloc(sizeof(ushort)*nsems);
727 if(sem_io == NULL) {
728 sem_putref(sma);
729 return -ENOMEM;
733 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
734 sem_putref(sma);
735 err = -EFAULT;
736 goto out_free;
739 for (i = 0; i < nsems; i++) {
740 if (sem_io[i] > SEMVMX) {
741 sem_putref(sma);
742 err = -ERANGE;
743 goto out_free;
746 sem_lock_and_putref(sma);
747 if (sma->sem_perm.deleted) {
748 sem_unlock(sma);
749 err = -EIDRM;
750 goto out_free;
753 for (i = 0; i < nsems; i++)
754 sma->sem_base[i].semval = sem_io[i];
756 assert_spin_locked(&sma->sem_perm.lock);
757 list_for_each_entry(un, &sma->list_id, list_id) {
758 for (i = 0; i < nsems; i++)
759 un->semadj[i] = 0;
761 sma->sem_ctime = get_seconds();
762 /* maybe some queued-up processes were waiting for this */
763 update_queue(sma);
764 err = 0;
765 goto out_unlock;
767 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
769 err = -EINVAL;
770 if(semnum < 0 || semnum >= nsems)
771 goto out_unlock;
773 curr = &sma->sem_base[semnum];
775 switch (cmd) {
776 case GETVAL:
777 err = curr->semval;
778 goto out_unlock;
779 case GETPID:
780 err = curr->sempid;
781 goto out_unlock;
782 case GETNCNT:
783 err = count_semncnt(sma,semnum);
784 goto out_unlock;
785 case GETZCNT:
786 err = count_semzcnt(sma,semnum);
787 goto out_unlock;
788 case SETVAL:
790 int val = arg.val;
791 struct sem_undo *un;
793 err = -ERANGE;
794 if (val > SEMVMX || val < 0)
795 goto out_unlock;
797 assert_spin_locked(&sma->sem_perm.lock);
798 list_for_each_entry(un, &sma->list_id, list_id)
799 un->semadj[semnum] = 0;
801 curr->semval = val;
802 curr->sempid = task_tgid_vnr(current);
803 sma->sem_ctime = get_seconds();
804 /* maybe some queued-up processes were waiting for this */
805 update_queue(sma);
806 err = 0;
807 goto out_unlock;
810 out_unlock:
811 sem_unlock(sma);
812 out_free:
813 if(sem_io != fast_sem_io)
814 ipc_free(sem_io, sizeof(ushort)*nsems);
815 return err;
818 static inline unsigned long
819 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
821 switch(version) {
822 case IPC_64:
823 if (copy_from_user(out, buf, sizeof(*out)))
824 return -EFAULT;
825 return 0;
826 case IPC_OLD:
828 struct semid_ds tbuf_old;
830 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
831 return -EFAULT;
833 out->sem_perm.uid = tbuf_old.sem_perm.uid;
834 out->sem_perm.gid = tbuf_old.sem_perm.gid;
835 out->sem_perm.mode = tbuf_old.sem_perm.mode;
837 return 0;
839 default:
840 return -EINVAL;
845 * This function handles some semctl commands which require the rw_mutex
846 * to be held in write mode.
847 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
849 static int semctl_down(struct ipc_namespace *ns, int semid,
850 int cmd, int version, union semun arg)
852 struct sem_array *sma;
853 int err;
854 struct semid64_ds semid64;
855 struct kern_ipc_perm *ipcp;
857 if(cmd == IPC_SET) {
858 if (copy_semid_from_user(&semid64, arg.buf, version))
859 return -EFAULT;
862 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
863 if (IS_ERR(ipcp))
864 return PTR_ERR(ipcp);
866 sma = container_of(ipcp, struct sem_array, sem_perm);
868 err = security_sem_semctl(sma, cmd);
869 if (err)
870 goto out_unlock;
872 switch(cmd){
873 case IPC_RMID:
874 freeary(ns, ipcp);
875 goto out_up;
876 case IPC_SET:
877 ipc_update_perm(&semid64.sem_perm, ipcp);
878 sma->sem_ctime = get_seconds();
879 break;
880 default:
881 err = -EINVAL;
884 out_unlock:
885 sem_unlock(sma);
886 out_up:
887 up_write(&sem_ids(ns).rw_mutex);
888 return err;
891 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
893 int err = -EINVAL;
894 int version;
895 struct ipc_namespace *ns;
897 if (semid < 0)
898 return -EINVAL;
900 version = ipc_parse_version(&cmd);
901 ns = current->nsproxy->ipc_ns;
903 switch(cmd) {
904 case IPC_INFO:
905 case SEM_INFO:
906 case IPC_STAT:
907 case SEM_STAT:
908 err = semctl_nolock(ns, semid, cmd, version, arg);
909 return err;
910 case GETALL:
911 case GETVAL:
912 case GETPID:
913 case GETNCNT:
914 case GETZCNT:
915 case SETVAL:
916 case SETALL:
917 err = semctl_main(ns,semid,semnum,cmd,version,arg);
918 return err;
919 case IPC_RMID:
920 case IPC_SET:
921 err = semctl_down(ns, semid, cmd, version, arg);
922 return err;
923 default:
924 return -EINVAL;
927 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
928 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
930 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
932 SYSCALL_ALIAS(sys_semctl, SyS_semctl);
933 #endif
935 /* If the task doesn't already have a undo_list, then allocate one
936 * here. We guarantee there is only one thread using this undo list,
937 * and current is THE ONE
939 * If this allocation and assignment succeeds, but later
940 * portions of this code fail, there is no need to free the sem_undo_list.
941 * Just let it stay associated with the task, and it'll be freed later
942 * at exit time.
944 * This can block, so callers must hold no locks.
946 static inline int get_undo_list(struct sem_undo_list **undo_listp)
948 struct sem_undo_list *undo_list;
950 undo_list = current->sysvsem.undo_list;
951 if (!undo_list) {
952 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
953 if (undo_list == NULL)
954 return -ENOMEM;
955 spin_lock_init(&undo_list->lock);
956 atomic_set(&undo_list->refcnt, 1);
957 INIT_LIST_HEAD(&undo_list->list_proc);
959 current->sysvsem.undo_list = undo_list;
961 *undo_listp = undo_list;
962 return 0;
965 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
967 struct sem_undo *walk;
969 list_for_each_entry_rcu(walk, &ulp->list_proc, list_proc) {
970 if (walk->semid == semid)
971 return walk;
973 return NULL;
977 * find_alloc_undo - Lookup (and if not present create) undo array
978 * @ns: namespace
979 * @semid: semaphore array id
981 * The function looks up (and if not present creates) the undo structure.
982 * The size of the undo structure depends on the size of the semaphore
983 * array, thus the alloc path is not that straightforward.
984 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
985 * performs a rcu_read_lock().
987 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
989 struct sem_array *sma;
990 struct sem_undo_list *ulp;
991 struct sem_undo *un, *new;
992 int nsems;
993 int error;
995 error = get_undo_list(&ulp);
996 if (error)
997 return ERR_PTR(error);
999 rcu_read_lock();
1000 spin_lock(&ulp->lock);
1001 un = lookup_undo(ulp, semid);
1002 spin_unlock(&ulp->lock);
1003 if (likely(un!=NULL))
1004 goto out;
1005 rcu_read_unlock();
1007 /* no undo structure around - allocate one. */
1008 /* step 1: figure out the size of the semaphore array */
1009 sma = sem_lock_check(ns, semid);
1010 if (IS_ERR(sma))
1011 return ERR_PTR(PTR_ERR(sma));
1013 nsems = sma->sem_nsems;
1014 sem_getref_and_unlock(sma);
1016 /* step 2: allocate new undo structure */
1017 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1018 if (!new) {
1019 sem_putref(sma);
1020 return ERR_PTR(-ENOMEM);
1023 /* step 3: Acquire the lock on semaphore array */
1024 sem_lock_and_putref(sma);
1025 if (sma->sem_perm.deleted) {
1026 sem_unlock(sma);
1027 kfree(new);
1028 un = ERR_PTR(-EIDRM);
1029 goto out;
1031 spin_lock(&ulp->lock);
1034 * step 4: check for races: did someone else allocate the undo struct?
1036 un = lookup_undo(ulp, semid);
1037 if (un) {
1038 kfree(new);
1039 goto success;
1041 /* step 5: initialize & link new undo structure */
1042 new->semadj = (short *) &new[1];
1043 new->ulp = ulp;
1044 new->semid = semid;
1045 assert_spin_locked(&ulp->lock);
1046 list_add_rcu(&new->list_proc, &ulp->list_proc);
1047 assert_spin_locked(&sma->sem_perm.lock);
1048 list_add(&new->list_id, &sma->list_id);
1049 un = new;
1051 success:
1052 spin_unlock(&ulp->lock);
1053 rcu_read_lock();
1054 sem_unlock(sma);
1055 out:
1056 return un;
1059 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1060 unsigned, nsops, const struct timespec __user *, timeout)
1062 int error = -EINVAL;
1063 struct sem_array *sma;
1064 struct sembuf fast_sops[SEMOPM_FAST];
1065 struct sembuf* sops = fast_sops, *sop;
1066 struct sem_undo *un;
1067 int undos = 0, alter = 0, max;
1068 struct sem_queue queue;
1069 unsigned long jiffies_left = 0;
1070 struct ipc_namespace *ns;
1072 ns = current->nsproxy->ipc_ns;
1074 if (nsops < 1 || semid < 0)
1075 return -EINVAL;
1076 if (nsops > ns->sc_semopm)
1077 return -E2BIG;
1078 if(nsops > SEMOPM_FAST) {
1079 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1080 if(sops==NULL)
1081 return -ENOMEM;
1083 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1084 error=-EFAULT;
1085 goto out_free;
1087 if (timeout) {
1088 struct timespec _timeout;
1089 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1090 error = -EFAULT;
1091 goto out_free;
1093 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1094 _timeout.tv_nsec >= 1000000000L) {
1095 error = -EINVAL;
1096 goto out_free;
1098 jiffies_left = timespec_to_jiffies(&_timeout);
1100 max = 0;
1101 for (sop = sops; sop < sops + nsops; sop++) {
1102 if (sop->sem_num >= max)
1103 max = sop->sem_num;
1104 if (sop->sem_flg & SEM_UNDO)
1105 undos = 1;
1106 if (sop->sem_op != 0)
1107 alter = 1;
1110 if (undos) {
1111 un = find_alloc_undo(ns, semid);
1112 if (IS_ERR(un)) {
1113 error = PTR_ERR(un);
1114 goto out_free;
1116 } else
1117 un = NULL;
1119 sma = sem_lock_check(ns, semid);
1120 if (IS_ERR(sma)) {
1121 if (un)
1122 rcu_read_unlock();
1123 error = PTR_ERR(sma);
1124 goto out_free;
1128 * semid identifiers are not unique - find_alloc_undo may have
1129 * allocated an undo structure, it was invalidated by an RMID
1130 * and now a new array with received the same id. Check and fail.
1131 * This case can be detected checking un->semid. The existance of
1132 * "un" itself is guaranteed by rcu.
1134 error = -EIDRM;
1135 if (un) {
1136 if (un->semid == -1) {
1137 rcu_read_unlock();
1138 goto out_unlock_free;
1139 } else {
1141 * rcu lock can be released, "un" cannot disappear:
1142 * - sem_lock is acquired, thus IPC_RMID is
1143 * impossible.
1144 * - exit_sem is impossible, it always operates on
1145 * current (or a dead task).
1148 rcu_read_unlock();
1152 error = -EFBIG;
1153 if (max >= sma->sem_nsems)
1154 goto out_unlock_free;
1156 error = -EACCES;
1157 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1158 goto out_unlock_free;
1160 error = security_sem_semop(sma, sops, nsops, alter);
1161 if (error)
1162 goto out_unlock_free;
1164 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1165 if (error <= 0) {
1166 if (alter && error == 0)
1167 update_queue (sma);
1168 goto out_unlock_free;
1171 /* We need to sleep on this operation, so we put the current
1172 * task into the pending queue and go to sleep.
1175 queue.sops = sops;
1176 queue.nsops = nsops;
1177 queue.undo = un;
1178 queue.pid = task_tgid_vnr(current);
1179 queue.alter = alter;
1180 if (alter)
1181 list_add_tail(&queue.list, &sma->sem_pending);
1182 else
1183 list_add(&queue.list, &sma->sem_pending);
1185 queue.status = -EINTR;
1186 queue.sleeper = current;
1187 current->state = TASK_INTERRUPTIBLE;
1188 sem_unlock(sma);
1190 if (timeout)
1191 jiffies_left = schedule_timeout(jiffies_left);
1192 else
1193 schedule();
1195 error = queue.status;
1196 while(unlikely(error == IN_WAKEUP)) {
1197 cpu_relax();
1198 error = queue.status;
1201 if (error != -EINTR) {
1202 /* fast path: update_queue already obtained all requested
1203 * resources */
1204 goto out_free;
1207 sma = sem_lock(ns, semid);
1208 if (IS_ERR(sma)) {
1209 error = -EIDRM;
1210 goto out_free;
1214 * If queue.status != -EINTR we are woken up by another process
1216 error = queue.status;
1217 if (error != -EINTR) {
1218 goto out_unlock_free;
1222 * If an interrupt occurred we have to clean up the queue
1224 if (timeout && jiffies_left == 0)
1225 error = -EAGAIN;
1226 list_del(&queue.list);
1228 out_unlock_free:
1229 sem_unlock(sma);
1230 out_free:
1231 if(sops != fast_sops)
1232 kfree(sops);
1233 return error;
1236 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1237 unsigned, nsops)
1239 return sys_semtimedop(semid, tsops, nsops, NULL);
1242 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1243 * parent and child tasks.
1246 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1248 struct sem_undo_list *undo_list;
1249 int error;
1251 if (clone_flags & CLONE_SYSVSEM) {
1252 error = get_undo_list(&undo_list);
1253 if (error)
1254 return error;
1255 atomic_inc(&undo_list->refcnt);
1256 tsk->sysvsem.undo_list = undo_list;
1257 } else
1258 tsk->sysvsem.undo_list = NULL;
1260 return 0;
1264 * add semadj values to semaphores, free undo structures.
1265 * undo structures are not freed when semaphore arrays are destroyed
1266 * so some of them may be out of date.
1267 * IMPLEMENTATION NOTE: There is some confusion over whether the
1268 * set of adjustments that needs to be done should be done in an atomic
1269 * manner or not. That is, if we are attempting to decrement the semval
1270 * should we queue up and wait until we can do so legally?
1271 * The original implementation attempted to do this (queue and wait).
1272 * The current implementation does not do so. The POSIX standard
1273 * and SVID should be consulted to determine what behavior is mandated.
1275 void exit_sem(struct task_struct *tsk)
1277 struct sem_undo_list *ulp;
1279 ulp = tsk->sysvsem.undo_list;
1280 if (!ulp)
1281 return;
1282 tsk->sysvsem.undo_list = NULL;
1284 if (!atomic_dec_and_test(&ulp->refcnt))
1285 return;
1287 for (;;) {
1288 struct sem_array *sma;
1289 struct sem_undo *un;
1290 int semid;
1291 int i;
1293 rcu_read_lock();
1294 un = list_entry_rcu(ulp->list_proc.next,
1295 struct sem_undo, list_proc);
1296 if (&un->list_proc == &ulp->list_proc)
1297 semid = -1;
1298 else
1299 semid = un->semid;
1300 rcu_read_unlock();
1302 if (semid == -1)
1303 break;
1305 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1307 /* exit_sem raced with IPC_RMID, nothing to do */
1308 if (IS_ERR(sma))
1309 continue;
1311 un = lookup_undo(ulp, semid);
1312 if (un == NULL) {
1313 /* exit_sem raced with IPC_RMID+semget() that created
1314 * exactly the same semid. Nothing to do.
1316 sem_unlock(sma);
1317 continue;
1320 /* remove un from the linked lists */
1321 assert_spin_locked(&sma->sem_perm.lock);
1322 list_del(&un->list_id);
1324 spin_lock(&ulp->lock);
1325 list_del_rcu(&un->list_proc);
1326 spin_unlock(&ulp->lock);
1328 /* perform adjustments registered in un */
1329 for (i = 0; i < sma->sem_nsems; i++) {
1330 struct sem * semaphore = &sma->sem_base[i];
1331 if (un->semadj[i]) {
1332 semaphore->semval += un->semadj[i];
1334 * Range checks of the new semaphore value,
1335 * not defined by sus:
1336 * - Some unices ignore the undo entirely
1337 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1338 * - some cap the value (e.g. FreeBSD caps
1339 * at 0, but doesn't enforce SEMVMX)
1341 * Linux caps the semaphore value, both at 0
1342 * and at SEMVMX.
1344 * Manfred <manfred@colorfullife.com>
1346 if (semaphore->semval < 0)
1347 semaphore->semval = 0;
1348 if (semaphore->semval > SEMVMX)
1349 semaphore->semval = SEMVMX;
1350 semaphore->sempid = task_tgid_vnr(current);
1353 sma->sem_otime = get_seconds();
1354 /* maybe some queued-up processes were waiting for this */
1355 update_queue(sma);
1356 sem_unlock(sma);
1358 call_rcu(&un->rcu, free_un);
1360 kfree(ulp);
1363 #ifdef CONFIG_PROC_FS
1364 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1366 struct sem_array *sma = it;
1368 return seq_printf(s,
1369 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1370 sma->sem_perm.key,
1371 sma->sem_perm.id,
1372 sma->sem_perm.mode,
1373 sma->sem_nsems,
1374 sma->sem_perm.uid,
1375 sma->sem_perm.gid,
1376 sma->sem_perm.cuid,
1377 sma->sem_perm.cgid,
1378 sma->sem_otime,
1379 sma->sem_ctime);
1381 #endif