Linux v2.6.17-rc2
[linux-2.6/next.git] / ipc / sem.c
blob642659cd596b82f5aa3974e61925708cdc62f9dd
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 <alan@redhat.com>
62 * Lockless wakeup
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
66 #include <linux/config.h>
67 #include <linux/slab.h>
68 #include <linux/spinlock.h>
69 #include <linux/init.h>
70 #include <linux/proc_fs.h>
71 #include <linux/time.h>
72 #include <linux/smp_lock.h>
73 #include <linux/security.h>
74 #include <linux/syscalls.h>
75 #include <linux/audit.h>
76 #include <linux/capability.h>
77 #include <linux/seq_file.h>
78 #include <linux/mutex.h>
80 #include <asm/uaccess.h>
81 #include "util.h"
84 #define sem_lock(id) ((struct sem_array*)ipc_lock(&sem_ids,id))
85 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
86 #define sem_rmid(id) ((struct sem_array*)ipc_rmid(&sem_ids,id))
87 #define sem_checkid(sma, semid) \
88 ipc_checkid(&sem_ids,&sma->sem_perm,semid)
89 #define sem_buildid(id, seq) \
90 ipc_buildid(&sem_ids, id, seq)
91 static struct ipc_ids sem_ids;
93 static int newary (key_t, int, int);
94 static void freeary (struct sem_array *sma, int id);
95 #ifdef CONFIG_PROC_FS
96 static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
97 #endif
99 #define SEMMSL_FAST 256 /* 512 bytes on stack */
100 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
103 * linked list protection:
104 * sem_undo.id_next,
105 * sem_array.sem_pending{,last},
106 * sem_array.sem_undo: sem_lock() for read/write
107 * sem_undo.proc_next: only "current" is allowed to read/write that field.
111 int sem_ctls[4] = {SEMMSL, SEMMNS, SEMOPM, SEMMNI};
112 #define sc_semmsl (sem_ctls[0])
113 #define sc_semmns (sem_ctls[1])
114 #define sc_semopm (sem_ctls[2])
115 #define sc_semmni (sem_ctls[3])
117 static int used_sems;
119 void __init sem_init (void)
121 used_sems = 0;
122 ipc_init_ids(&sem_ids,sc_semmni);
123 ipc_init_proc_interface("sysvipc/sem",
124 " key semid perms nsems uid gid cuid cgid otime ctime\n",
125 &sem_ids,
126 sysvipc_sem_proc_show);
130 * Lockless wakeup algorithm:
131 * Without the check/retry algorithm a lockless wakeup is possible:
132 * - queue.status is initialized to -EINTR before blocking.
133 * - wakeup is performed by
134 * * unlinking the queue entry from sma->sem_pending
135 * * setting queue.status to IN_WAKEUP
136 * This is the notification for the blocked thread that a
137 * result value is imminent.
138 * * call wake_up_process
139 * * set queue.status to the final value.
140 * - the previously blocked thread checks queue.status:
141 * * if it's IN_WAKEUP, then it must wait until the value changes
142 * * if it's not -EINTR, then the operation was completed by
143 * update_queue. semtimedop can return queue.status without
144 * performing any operation on the sem array.
145 * * otherwise it must acquire the spinlock and check what's up.
147 * The two-stage algorithm is necessary to protect against the following
148 * races:
149 * - if queue.status is set after wake_up_process, then the woken up idle
150 * thread could race forward and try (and fail) to acquire sma->lock
151 * before update_queue had a chance to set queue.status
152 * - if queue.status is written before wake_up_process and if the
153 * blocked process is woken up by a signal between writing
154 * queue.status and the wake_up_process, then the woken up
155 * process could return from semtimedop and die by calling
156 * sys_exit before wake_up_process is called. Then wake_up_process
157 * will oops, because the task structure is already invalid.
158 * (yes, this happened on s390 with sysv msg).
161 #define IN_WAKEUP 1
163 static int newary (key_t key, int nsems, int semflg)
165 int id;
166 int retval;
167 struct sem_array *sma;
168 int size;
170 if (!nsems)
171 return -EINVAL;
172 if (used_sems + nsems > sc_semmns)
173 return -ENOSPC;
175 size = sizeof (*sma) + nsems * sizeof (struct sem);
176 sma = ipc_rcu_alloc(size);
177 if (!sma) {
178 return -ENOMEM;
180 memset (sma, 0, size);
182 sma->sem_perm.mode = (semflg & S_IRWXUGO);
183 sma->sem_perm.key = key;
185 sma->sem_perm.security = NULL;
186 retval = security_sem_alloc(sma);
187 if (retval) {
188 ipc_rcu_putref(sma);
189 return retval;
192 id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
193 if(id == -1) {
194 security_sem_free(sma);
195 ipc_rcu_putref(sma);
196 return -ENOSPC;
198 used_sems += nsems;
200 sma->sem_id = sem_buildid(id, sma->sem_perm.seq);
201 sma->sem_base = (struct sem *) &sma[1];
202 /* sma->sem_pending = NULL; */
203 sma->sem_pending_last = &sma->sem_pending;
204 /* sma->undo = NULL; */
205 sma->sem_nsems = nsems;
206 sma->sem_ctime = get_seconds();
207 sem_unlock(sma);
209 return sma->sem_id;
212 asmlinkage long sys_semget (key_t key, int nsems, int semflg)
214 int id, err = -EINVAL;
215 struct sem_array *sma;
217 if (nsems < 0 || nsems > sc_semmsl)
218 return -EINVAL;
219 mutex_lock(&sem_ids.mutex);
221 if (key == IPC_PRIVATE) {
222 err = newary(key, nsems, semflg);
223 } else if ((id = ipc_findkey(&sem_ids, key)) == -1) { /* key not used */
224 if (!(semflg & IPC_CREAT))
225 err = -ENOENT;
226 else
227 err = newary(key, nsems, semflg);
228 } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) {
229 err = -EEXIST;
230 } else {
231 sma = sem_lock(id);
232 BUG_ON(sma==NULL);
233 if (nsems > sma->sem_nsems)
234 err = -EINVAL;
235 else if (ipcperms(&sma->sem_perm, semflg))
236 err = -EACCES;
237 else {
238 int semid = sem_buildid(id, sma->sem_perm.seq);
239 err = security_sem_associate(sma, semflg);
240 if (!err)
241 err = semid;
243 sem_unlock(sma);
246 mutex_unlock(&sem_ids.mutex);
247 return err;
250 /* Manage the doubly linked list sma->sem_pending as a FIFO:
251 * insert new queue elements at the tail sma->sem_pending_last.
253 static inline void append_to_queue (struct sem_array * sma,
254 struct sem_queue * q)
256 *(q->prev = sma->sem_pending_last) = q;
257 *(sma->sem_pending_last = &q->next) = NULL;
260 static inline void prepend_to_queue (struct sem_array * sma,
261 struct sem_queue * q)
263 q->next = sma->sem_pending;
264 *(q->prev = &sma->sem_pending) = q;
265 if (q->next)
266 q->next->prev = &q->next;
267 else /* sma->sem_pending_last == &sma->sem_pending */
268 sma->sem_pending_last = &q->next;
271 static inline void remove_from_queue (struct sem_array * sma,
272 struct sem_queue * q)
274 *(q->prev) = q->next;
275 if (q->next)
276 q->next->prev = q->prev;
277 else /* sma->sem_pending_last == &q->next */
278 sma->sem_pending_last = q->prev;
279 q->prev = NULL; /* mark as removed */
283 * Determine whether a sequence of semaphore operations would succeed
284 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
287 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
288 int nsops, struct sem_undo *un, int pid)
290 int result, sem_op;
291 struct sembuf *sop;
292 struct sem * curr;
294 for (sop = sops; sop < sops + nsops; sop++) {
295 curr = sma->sem_base + sop->sem_num;
296 sem_op = sop->sem_op;
297 result = curr->semval;
299 if (!sem_op && result)
300 goto would_block;
302 result += sem_op;
303 if (result < 0)
304 goto would_block;
305 if (result > SEMVMX)
306 goto out_of_range;
307 if (sop->sem_flg & SEM_UNDO) {
308 int undo = un->semadj[sop->sem_num] - sem_op;
310 * Exceeding the undo range is an error.
312 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
313 goto out_of_range;
315 curr->semval = result;
318 sop--;
319 while (sop >= sops) {
320 sma->sem_base[sop->sem_num].sempid = pid;
321 if (sop->sem_flg & SEM_UNDO)
322 un->semadj[sop->sem_num] -= sop->sem_op;
323 sop--;
326 sma->sem_otime = get_seconds();
327 return 0;
329 out_of_range:
330 result = -ERANGE;
331 goto undo;
333 would_block:
334 if (sop->sem_flg & IPC_NOWAIT)
335 result = -EAGAIN;
336 else
337 result = 1;
339 undo:
340 sop--;
341 while (sop >= sops) {
342 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
343 sop--;
346 return result;
349 /* Go through the pending queue for the indicated semaphore
350 * looking for tasks that can be completed.
352 static void update_queue (struct sem_array * sma)
354 int error;
355 struct sem_queue * q;
357 q = sma->sem_pending;
358 while(q) {
359 error = try_atomic_semop(sma, q->sops, q->nsops,
360 q->undo, q->pid);
362 /* Does q->sleeper still need to sleep? */
363 if (error <= 0) {
364 struct sem_queue *n;
365 remove_from_queue(sma,q);
366 q->status = IN_WAKEUP;
368 * Continue scanning. The next operation
369 * that must be checked depends on the type of the
370 * completed operation:
371 * - if the operation modified the array, then
372 * restart from the head of the queue and
373 * check for threads that might be waiting
374 * for semaphore values to become 0.
375 * - if the operation didn't modify the array,
376 * then just continue.
378 if (q->alter)
379 n = sma->sem_pending;
380 else
381 n = q->next;
382 wake_up_process(q->sleeper);
383 /* hands-off: q will disappear immediately after
384 * writing q->status.
386 smp_wmb();
387 q->status = error;
388 q = n;
389 } else {
390 q = q->next;
395 /* The following counts are associated to each semaphore:
396 * semncnt number of tasks waiting on semval being nonzero
397 * semzcnt number of tasks waiting on semval being zero
398 * This model assumes that a task waits on exactly one semaphore.
399 * Since semaphore operations are to be performed atomically, tasks actually
400 * wait on a whole sequence of semaphores simultaneously.
401 * The counts we return here are a rough approximation, but still
402 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
404 static int count_semncnt (struct sem_array * sma, ushort semnum)
406 int semncnt;
407 struct sem_queue * q;
409 semncnt = 0;
410 for (q = sma->sem_pending; q; q = q->next) {
411 struct sembuf * sops = q->sops;
412 int nsops = q->nsops;
413 int i;
414 for (i = 0; i < nsops; i++)
415 if (sops[i].sem_num == semnum
416 && (sops[i].sem_op < 0)
417 && !(sops[i].sem_flg & IPC_NOWAIT))
418 semncnt++;
420 return semncnt;
422 static int count_semzcnt (struct sem_array * sma, ushort semnum)
424 int semzcnt;
425 struct sem_queue * q;
427 semzcnt = 0;
428 for (q = sma->sem_pending; q; q = q->next) {
429 struct sembuf * sops = q->sops;
430 int nsops = q->nsops;
431 int i;
432 for (i = 0; i < nsops; i++)
433 if (sops[i].sem_num == semnum
434 && (sops[i].sem_op == 0)
435 && !(sops[i].sem_flg & IPC_NOWAIT))
436 semzcnt++;
438 return semzcnt;
441 /* Free a semaphore set. freeary() is called with sem_ids.mutex locked and
442 * the spinlock for this semaphore set hold. sem_ids.mutex remains locked
443 * on exit.
445 static void freeary (struct sem_array *sma, int id)
447 struct sem_undo *un;
448 struct sem_queue *q;
449 int size;
451 /* Invalidate the existing undo structures for this semaphore set.
452 * (They will be freed without any further action in exit_sem()
453 * or during the next semop.)
455 for (un = sma->undo; un; un = un->id_next)
456 un->semid = -1;
458 /* Wake up all pending processes and let them fail with EIDRM. */
459 q = sma->sem_pending;
460 while(q) {
461 struct sem_queue *n;
462 /* lazy remove_from_queue: we are killing the whole queue */
463 q->prev = NULL;
464 n = q->next;
465 q->status = IN_WAKEUP;
466 wake_up_process(q->sleeper); /* doesn't sleep */
467 smp_wmb();
468 q->status = -EIDRM; /* hands-off q */
469 q = n;
472 /* Remove the semaphore set from the ID array*/
473 sma = sem_rmid(id);
474 sem_unlock(sma);
476 used_sems -= sma->sem_nsems;
477 size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
478 security_sem_free(sma);
479 ipc_rcu_putref(sma);
482 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
484 switch(version) {
485 case IPC_64:
486 return copy_to_user(buf, in, sizeof(*in));
487 case IPC_OLD:
489 struct semid_ds out;
491 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
493 out.sem_otime = in->sem_otime;
494 out.sem_ctime = in->sem_ctime;
495 out.sem_nsems = in->sem_nsems;
497 return copy_to_user(buf, &out, sizeof(out));
499 default:
500 return -EINVAL;
504 static int semctl_nolock(int semid, int semnum, int cmd, int version, union semun arg)
506 int err = -EINVAL;
507 struct sem_array *sma;
509 switch(cmd) {
510 case IPC_INFO:
511 case SEM_INFO:
513 struct seminfo seminfo;
514 int max_id;
516 err = security_sem_semctl(NULL, cmd);
517 if (err)
518 return err;
520 memset(&seminfo,0,sizeof(seminfo));
521 seminfo.semmni = sc_semmni;
522 seminfo.semmns = sc_semmns;
523 seminfo.semmsl = sc_semmsl;
524 seminfo.semopm = sc_semopm;
525 seminfo.semvmx = SEMVMX;
526 seminfo.semmnu = SEMMNU;
527 seminfo.semmap = SEMMAP;
528 seminfo.semume = SEMUME;
529 mutex_lock(&sem_ids.mutex);
530 if (cmd == SEM_INFO) {
531 seminfo.semusz = sem_ids.in_use;
532 seminfo.semaem = used_sems;
533 } else {
534 seminfo.semusz = SEMUSZ;
535 seminfo.semaem = SEMAEM;
537 max_id = sem_ids.max_id;
538 mutex_unlock(&sem_ids.mutex);
539 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
540 return -EFAULT;
541 return (max_id < 0) ? 0: max_id;
543 case SEM_STAT:
545 struct semid64_ds tbuf;
546 int id;
548 if(semid >= sem_ids.entries->size)
549 return -EINVAL;
551 memset(&tbuf,0,sizeof(tbuf));
553 sma = sem_lock(semid);
554 if(sma == NULL)
555 return -EINVAL;
557 err = -EACCES;
558 if (ipcperms (&sma->sem_perm, S_IRUGO))
559 goto out_unlock;
561 err = security_sem_semctl(sma, cmd);
562 if (err)
563 goto out_unlock;
565 id = sem_buildid(semid, sma->sem_perm.seq);
567 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
568 tbuf.sem_otime = sma->sem_otime;
569 tbuf.sem_ctime = sma->sem_ctime;
570 tbuf.sem_nsems = sma->sem_nsems;
571 sem_unlock(sma);
572 if (copy_semid_to_user (arg.buf, &tbuf, version))
573 return -EFAULT;
574 return id;
576 default:
577 return -EINVAL;
579 return err;
580 out_unlock:
581 sem_unlock(sma);
582 return err;
585 static int semctl_main(int semid, int semnum, int cmd, int version, union semun arg)
587 struct sem_array *sma;
588 struct sem* curr;
589 int err;
590 ushort fast_sem_io[SEMMSL_FAST];
591 ushort* sem_io = fast_sem_io;
592 int nsems;
594 sma = sem_lock(semid);
595 if(sma==NULL)
596 return -EINVAL;
598 nsems = sma->sem_nsems;
600 err=-EIDRM;
601 if (sem_checkid(sma,semid))
602 goto out_unlock;
604 err = -EACCES;
605 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
606 goto out_unlock;
608 err = security_sem_semctl(sma, cmd);
609 if (err)
610 goto out_unlock;
612 err = -EACCES;
613 switch (cmd) {
614 case GETALL:
616 ushort __user *array = arg.array;
617 int i;
619 if(nsems > SEMMSL_FAST) {
620 ipc_rcu_getref(sma);
621 sem_unlock(sma);
623 sem_io = ipc_alloc(sizeof(ushort)*nsems);
624 if(sem_io == NULL) {
625 ipc_lock_by_ptr(&sma->sem_perm);
626 ipc_rcu_putref(sma);
627 sem_unlock(sma);
628 return -ENOMEM;
631 ipc_lock_by_ptr(&sma->sem_perm);
632 ipc_rcu_putref(sma);
633 if (sma->sem_perm.deleted) {
634 sem_unlock(sma);
635 err = -EIDRM;
636 goto out_free;
640 for (i = 0; i < sma->sem_nsems; i++)
641 sem_io[i] = sma->sem_base[i].semval;
642 sem_unlock(sma);
643 err = 0;
644 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
645 err = -EFAULT;
646 goto out_free;
648 case SETALL:
650 int i;
651 struct sem_undo *un;
653 ipc_rcu_getref(sma);
654 sem_unlock(sma);
656 if(nsems > SEMMSL_FAST) {
657 sem_io = ipc_alloc(sizeof(ushort)*nsems);
658 if(sem_io == NULL) {
659 ipc_lock_by_ptr(&sma->sem_perm);
660 ipc_rcu_putref(sma);
661 sem_unlock(sma);
662 return -ENOMEM;
666 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
667 ipc_lock_by_ptr(&sma->sem_perm);
668 ipc_rcu_putref(sma);
669 sem_unlock(sma);
670 err = -EFAULT;
671 goto out_free;
674 for (i = 0; i < nsems; i++) {
675 if (sem_io[i] > SEMVMX) {
676 ipc_lock_by_ptr(&sma->sem_perm);
677 ipc_rcu_putref(sma);
678 sem_unlock(sma);
679 err = -ERANGE;
680 goto out_free;
683 ipc_lock_by_ptr(&sma->sem_perm);
684 ipc_rcu_putref(sma);
685 if (sma->sem_perm.deleted) {
686 sem_unlock(sma);
687 err = -EIDRM;
688 goto out_free;
691 for (i = 0; i < nsems; i++)
692 sma->sem_base[i].semval = sem_io[i];
693 for (un = sma->undo; un; un = un->id_next)
694 for (i = 0; i < nsems; i++)
695 un->semadj[i] = 0;
696 sma->sem_ctime = get_seconds();
697 /* maybe some queued-up processes were waiting for this */
698 update_queue(sma);
699 err = 0;
700 goto out_unlock;
702 case IPC_STAT:
704 struct semid64_ds tbuf;
705 memset(&tbuf,0,sizeof(tbuf));
706 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
707 tbuf.sem_otime = sma->sem_otime;
708 tbuf.sem_ctime = sma->sem_ctime;
709 tbuf.sem_nsems = sma->sem_nsems;
710 sem_unlock(sma);
711 if (copy_semid_to_user (arg.buf, &tbuf, version))
712 return -EFAULT;
713 return 0;
715 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
717 err = -EINVAL;
718 if(semnum < 0 || semnum >= nsems)
719 goto out_unlock;
721 curr = &sma->sem_base[semnum];
723 switch (cmd) {
724 case GETVAL:
725 err = curr->semval;
726 goto out_unlock;
727 case GETPID:
728 err = curr->sempid;
729 goto out_unlock;
730 case GETNCNT:
731 err = count_semncnt(sma,semnum);
732 goto out_unlock;
733 case GETZCNT:
734 err = count_semzcnt(sma,semnum);
735 goto out_unlock;
736 case SETVAL:
738 int val = arg.val;
739 struct sem_undo *un;
740 err = -ERANGE;
741 if (val > SEMVMX || val < 0)
742 goto out_unlock;
744 for (un = sma->undo; un; un = un->id_next)
745 un->semadj[semnum] = 0;
746 curr->semval = val;
747 curr->sempid = current->tgid;
748 sma->sem_ctime = get_seconds();
749 /* maybe some queued-up processes were waiting for this */
750 update_queue(sma);
751 err = 0;
752 goto out_unlock;
755 out_unlock:
756 sem_unlock(sma);
757 out_free:
758 if(sem_io != fast_sem_io)
759 ipc_free(sem_io, sizeof(ushort)*nsems);
760 return err;
763 struct sem_setbuf {
764 uid_t uid;
765 gid_t gid;
766 mode_t mode;
769 static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
771 switch(version) {
772 case IPC_64:
774 struct semid64_ds tbuf;
776 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
777 return -EFAULT;
779 out->uid = tbuf.sem_perm.uid;
780 out->gid = tbuf.sem_perm.gid;
781 out->mode = tbuf.sem_perm.mode;
783 return 0;
785 case IPC_OLD:
787 struct semid_ds tbuf_old;
789 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
790 return -EFAULT;
792 out->uid = tbuf_old.sem_perm.uid;
793 out->gid = tbuf_old.sem_perm.gid;
794 out->mode = tbuf_old.sem_perm.mode;
796 return 0;
798 default:
799 return -EINVAL;
803 static int semctl_down(int semid, int semnum, int cmd, int version, union semun arg)
805 struct sem_array *sma;
806 int err;
807 struct sem_setbuf setbuf;
808 struct kern_ipc_perm *ipcp;
810 if(cmd == IPC_SET) {
811 if(copy_semid_from_user (&setbuf, arg.buf, version))
812 return -EFAULT;
814 sma = sem_lock(semid);
815 if(sma==NULL)
816 return -EINVAL;
818 if (sem_checkid(sma,semid)) {
819 err=-EIDRM;
820 goto out_unlock;
822 ipcp = &sma->sem_perm;
823 if (current->euid != ipcp->cuid &&
824 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
825 err=-EPERM;
826 goto out_unlock;
829 err = security_sem_semctl(sma, cmd);
830 if (err)
831 goto out_unlock;
833 switch(cmd){
834 case IPC_RMID:
835 freeary(sma, semid);
836 err = 0;
837 break;
838 case IPC_SET:
839 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode, ipcp)))
840 goto out_unlock;
841 ipcp->uid = setbuf.uid;
842 ipcp->gid = setbuf.gid;
843 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
844 | (setbuf.mode & S_IRWXUGO);
845 sma->sem_ctime = get_seconds();
846 sem_unlock(sma);
847 err = 0;
848 break;
849 default:
850 sem_unlock(sma);
851 err = -EINVAL;
852 break;
854 return err;
856 out_unlock:
857 sem_unlock(sma);
858 return err;
861 asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
863 int err = -EINVAL;
864 int version;
866 if (semid < 0)
867 return -EINVAL;
869 version = ipc_parse_version(&cmd);
871 switch(cmd) {
872 case IPC_INFO:
873 case SEM_INFO:
874 case SEM_STAT:
875 err = semctl_nolock(semid,semnum,cmd,version,arg);
876 return err;
877 case GETALL:
878 case GETVAL:
879 case GETPID:
880 case GETNCNT:
881 case GETZCNT:
882 case IPC_STAT:
883 case SETVAL:
884 case SETALL:
885 err = semctl_main(semid,semnum,cmd,version,arg);
886 return err;
887 case IPC_RMID:
888 case IPC_SET:
889 mutex_lock(&sem_ids.mutex);
890 err = semctl_down(semid,semnum,cmd,version,arg);
891 mutex_unlock(&sem_ids.mutex);
892 return err;
893 default:
894 return -EINVAL;
898 static inline void lock_semundo(void)
900 struct sem_undo_list *undo_list;
902 undo_list = current->sysvsem.undo_list;
903 if (undo_list)
904 spin_lock(&undo_list->lock);
907 /* This code has an interaction with copy_semundo().
908 * Consider; two tasks are sharing the undo_list. task1
909 * acquires the undo_list lock in lock_semundo(). If task2 now
910 * exits before task1 releases the lock (by calling
911 * unlock_semundo()), then task1 will never call spin_unlock().
912 * This leave the sem_undo_list in a locked state. If task1 now creats task3
913 * and once again shares the sem_undo_list, the sem_undo_list will still be
914 * locked, and future SEM_UNDO operations will deadlock. This case is
915 * dealt with in copy_semundo() by having it reinitialize the spin lock when
916 * the refcnt goes from 1 to 2.
918 static inline void unlock_semundo(void)
920 struct sem_undo_list *undo_list;
922 undo_list = current->sysvsem.undo_list;
923 if (undo_list)
924 spin_unlock(&undo_list->lock);
928 /* If the task doesn't already have a undo_list, then allocate one
929 * here. We guarantee there is only one thread using this undo list,
930 * and current is THE ONE
932 * If this allocation and assignment succeeds, but later
933 * portions of this code fail, there is no need to free the sem_undo_list.
934 * Just let it stay associated with the task, and it'll be freed later
935 * at exit time.
937 * This can block, so callers must hold no locks.
939 static inline int get_undo_list(struct sem_undo_list **undo_listp)
941 struct sem_undo_list *undo_list;
942 int size;
944 undo_list = current->sysvsem.undo_list;
945 if (!undo_list) {
946 size = sizeof(struct sem_undo_list);
947 undo_list = (struct sem_undo_list *) kmalloc(size, GFP_KERNEL);
948 if (undo_list == NULL)
949 return -ENOMEM;
950 memset(undo_list, 0, size);
951 spin_lock_init(&undo_list->lock);
952 atomic_set(&undo_list->refcnt, 1);
953 current->sysvsem.undo_list = undo_list;
955 *undo_listp = undo_list;
956 return 0;
959 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
961 struct sem_undo **last, *un;
963 last = &ulp->proc_list;
964 un = *last;
965 while(un != NULL) {
966 if(un->semid==semid)
967 break;
968 if(un->semid==-1) {
969 *last=un->proc_next;
970 kfree(un);
971 } else {
972 last=&un->proc_next;
974 un=*last;
976 return un;
979 static struct sem_undo *find_undo(int semid)
981 struct sem_array *sma;
982 struct sem_undo_list *ulp;
983 struct sem_undo *un, *new;
984 int nsems;
985 int error;
987 error = get_undo_list(&ulp);
988 if (error)
989 return ERR_PTR(error);
991 lock_semundo();
992 un = lookup_undo(ulp, semid);
993 unlock_semundo();
994 if (likely(un!=NULL))
995 goto out;
997 /* no undo structure around - allocate one. */
998 sma = sem_lock(semid);
999 un = ERR_PTR(-EINVAL);
1000 if(sma==NULL)
1001 goto out;
1002 un = ERR_PTR(-EIDRM);
1003 if (sem_checkid(sma,semid)) {
1004 sem_unlock(sma);
1005 goto out;
1007 nsems = sma->sem_nsems;
1008 ipc_rcu_getref(sma);
1009 sem_unlock(sma);
1011 new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1012 if (!new) {
1013 ipc_lock_by_ptr(&sma->sem_perm);
1014 ipc_rcu_putref(sma);
1015 sem_unlock(sma);
1016 return ERR_PTR(-ENOMEM);
1018 memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems);
1019 new->semadj = (short *) &new[1];
1020 new->semid = semid;
1022 lock_semundo();
1023 un = lookup_undo(ulp, semid);
1024 if (un) {
1025 unlock_semundo();
1026 kfree(new);
1027 ipc_lock_by_ptr(&sma->sem_perm);
1028 ipc_rcu_putref(sma);
1029 sem_unlock(sma);
1030 goto out;
1032 ipc_lock_by_ptr(&sma->sem_perm);
1033 ipc_rcu_putref(sma);
1034 if (sma->sem_perm.deleted) {
1035 sem_unlock(sma);
1036 unlock_semundo();
1037 kfree(new);
1038 un = ERR_PTR(-EIDRM);
1039 goto out;
1041 new->proc_next = ulp->proc_list;
1042 ulp->proc_list = new;
1043 new->id_next = sma->undo;
1044 sma->undo = new;
1045 sem_unlock(sma);
1046 un = new;
1047 unlock_semundo();
1048 out:
1049 return un;
1052 asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1053 unsigned nsops, const struct timespec __user *timeout)
1055 int error = -EINVAL;
1056 struct sem_array *sma;
1057 struct sembuf fast_sops[SEMOPM_FAST];
1058 struct sembuf* sops = fast_sops, *sop;
1059 struct sem_undo *un;
1060 int undos = 0, alter = 0, max;
1061 struct sem_queue queue;
1062 unsigned long jiffies_left = 0;
1064 if (nsops < 1 || semid < 0)
1065 return -EINVAL;
1066 if (nsops > sc_semopm)
1067 return -E2BIG;
1068 if(nsops > SEMOPM_FAST) {
1069 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1070 if(sops==NULL)
1071 return -ENOMEM;
1073 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1074 error=-EFAULT;
1075 goto out_free;
1077 if (timeout) {
1078 struct timespec _timeout;
1079 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1080 error = -EFAULT;
1081 goto out_free;
1083 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1084 _timeout.tv_nsec >= 1000000000L) {
1085 error = -EINVAL;
1086 goto out_free;
1088 jiffies_left = timespec_to_jiffies(&_timeout);
1090 max = 0;
1091 for (sop = sops; sop < sops + nsops; sop++) {
1092 if (sop->sem_num >= max)
1093 max = sop->sem_num;
1094 if (sop->sem_flg & SEM_UNDO)
1095 undos = 1;
1096 if (sop->sem_op != 0)
1097 alter = 1;
1100 retry_undos:
1101 if (undos) {
1102 un = find_undo(semid);
1103 if (IS_ERR(un)) {
1104 error = PTR_ERR(un);
1105 goto out_free;
1107 } else
1108 un = NULL;
1110 sma = sem_lock(semid);
1111 error=-EINVAL;
1112 if(sma==NULL)
1113 goto out_free;
1114 error = -EIDRM;
1115 if (sem_checkid(sma,semid))
1116 goto out_unlock_free;
1118 * semid identifies are not unique - find_undo may have
1119 * allocated an undo structure, it was invalidated by an RMID
1120 * and now a new array with received the same id. Check and retry.
1122 if (un && un->semid == -1) {
1123 sem_unlock(sma);
1124 goto retry_undos;
1126 error = -EFBIG;
1127 if (max >= sma->sem_nsems)
1128 goto out_unlock_free;
1130 error = -EACCES;
1131 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1132 goto out_unlock_free;
1134 error = security_sem_semop(sma, sops, nsops, alter);
1135 if (error)
1136 goto out_unlock_free;
1138 error = try_atomic_semop (sma, sops, nsops, un, current->tgid);
1139 if (error <= 0) {
1140 if (alter && error == 0)
1141 update_queue (sma);
1142 goto out_unlock_free;
1145 /* We need to sleep on this operation, so we put the current
1146 * task into the pending queue and go to sleep.
1149 queue.sma = sma;
1150 queue.sops = sops;
1151 queue.nsops = nsops;
1152 queue.undo = un;
1153 queue.pid = current->tgid;
1154 queue.id = semid;
1155 queue.alter = alter;
1156 if (alter)
1157 append_to_queue(sma ,&queue);
1158 else
1159 prepend_to_queue(sma ,&queue);
1161 queue.status = -EINTR;
1162 queue.sleeper = current;
1163 current->state = TASK_INTERRUPTIBLE;
1164 sem_unlock(sma);
1166 if (timeout)
1167 jiffies_left = schedule_timeout(jiffies_left);
1168 else
1169 schedule();
1171 error = queue.status;
1172 while(unlikely(error == IN_WAKEUP)) {
1173 cpu_relax();
1174 error = queue.status;
1177 if (error != -EINTR) {
1178 /* fast path: update_queue already obtained all requested
1179 * resources */
1180 goto out_free;
1183 sma = sem_lock(semid);
1184 if(sma==NULL) {
1185 BUG_ON(queue.prev != NULL);
1186 error = -EIDRM;
1187 goto out_free;
1191 * If queue.status != -EINTR we are woken up by another process
1193 error = queue.status;
1194 if (error != -EINTR) {
1195 goto out_unlock_free;
1199 * If an interrupt occurred we have to clean up the queue
1201 if (timeout && jiffies_left == 0)
1202 error = -EAGAIN;
1203 remove_from_queue(sma,&queue);
1204 goto out_unlock_free;
1206 out_unlock_free:
1207 sem_unlock(sma);
1208 out_free:
1209 if(sops != fast_sops)
1210 kfree(sops);
1211 return error;
1214 asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1216 return sys_semtimedop(semid, tsops, nsops, NULL);
1219 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1220 * parent and child tasks.
1222 * See the notes above unlock_semundo() regarding the spin_lock_init()
1223 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1224 * because of the reasoning in the comment above unlock_semundo.
1227 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1229 struct sem_undo_list *undo_list;
1230 int error;
1232 if (clone_flags & CLONE_SYSVSEM) {
1233 error = get_undo_list(&undo_list);
1234 if (error)
1235 return error;
1236 atomic_inc(&undo_list->refcnt);
1237 tsk->sysvsem.undo_list = undo_list;
1238 } else
1239 tsk->sysvsem.undo_list = NULL;
1241 return 0;
1245 * add semadj values to semaphores, free undo structures.
1246 * undo structures are not freed when semaphore arrays are destroyed
1247 * so some of them may be out of date.
1248 * IMPLEMENTATION NOTE: There is some confusion over whether the
1249 * set of adjustments that needs to be done should be done in an atomic
1250 * manner or not. That is, if we are attempting to decrement the semval
1251 * should we queue up and wait until we can do so legally?
1252 * The original implementation attempted to do this (queue and wait).
1253 * The current implementation does not do so. The POSIX standard
1254 * and SVID should be consulted to determine what behavior is mandated.
1256 void exit_sem(struct task_struct *tsk)
1258 struct sem_undo_list *undo_list;
1259 struct sem_undo *u, **up;
1261 undo_list = tsk->sysvsem.undo_list;
1262 if (!undo_list)
1263 return;
1265 if (!atomic_dec_and_test(&undo_list->refcnt))
1266 return;
1268 /* There's no need to hold the semundo list lock, as current
1269 * is the last task exiting for this undo list.
1271 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1272 struct sem_array *sma;
1273 int nsems, i;
1274 struct sem_undo *un, **unp;
1275 int semid;
1277 semid = u->semid;
1279 if(semid == -1)
1280 continue;
1281 sma = sem_lock(semid);
1282 if (sma == NULL)
1283 continue;
1285 if (u->semid == -1)
1286 goto next_entry;
1288 BUG_ON(sem_checkid(sma,u->semid));
1290 /* remove u from the sma->undo list */
1291 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1292 if (u == un)
1293 goto found;
1295 printk ("exit_sem undo list error id=%d\n", u->semid);
1296 goto next_entry;
1297 found:
1298 *unp = un->id_next;
1299 /* perform adjustments registered in u */
1300 nsems = sma->sem_nsems;
1301 for (i = 0; i < nsems; i++) {
1302 struct sem * semaphore = &sma->sem_base[i];
1303 if (u->semadj[i]) {
1304 semaphore->semval += u->semadj[i];
1306 * Range checks of the new semaphore value,
1307 * not defined by sus:
1308 * - Some unices ignore the undo entirely
1309 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1310 * - some cap the value (e.g. FreeBSD caps
1311 * at 0, but doesn't enforce SEMVMX)
1313 * Linux caps the semaphore value, both at 0
1314 * and at SEMVMX.
1316 * Manfred <manfred@colorfullife.com>
1318 if (semaphore->semval < 0)
1319 semaphore->semval = 0;
1320 if (semaphore->semval > SEMVMX)
1321 semaphore->semval = SEMVMX;
1322 semaphore->sempid = current->tgid;
1325 sma->sem_otime = get_seconds();
1326 /* maybe some queued-up processes were waiting for this */
1327 update_queue(sma);
1328 next_entry:
1329 sem_unlock(sma);
1331 kfree(undo_list);
1334 #ifdef CONFIG_PROC_FS
1335 static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1337 struct sem_array *sma = it;
1339 return seq_printf(s,
1340 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1341 sma->sem_perm.key,
1342 sma->sem_id,
1343 sma->sem_perm.mode,
1344 sma->sem_nsems,
1345 sma->sem_perm.uid,
1346 sma->sem_perm.gid,
1347 sma->sem_perm.cuid,
1348 sma->sem_perm.cgid,
1349 sma->sem_otime,
1350 sma->sem_ctime);
1352 #endif