blob: 5ad7ac0ed60d0dc4fa46ce06964c3cd2c6f55b8c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
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.
32 *
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.
51 *
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.
55 *
56 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
57 *
58 * SMP-threaded, sysctl's added
59 * (c) 1999 Manfred Spraul <manfreds@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>
64 */
65
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 <asm/uaccess.h>
77#include "util.h"
78
79
80#define sem_lock(id) ((struct sem_array*)ipc_lock(&sem_ids,id))
81#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
82#define sem_rmid(id) ((struct sem_array*)ipc_rmid(&sem_ids,id))
83#define sem_checkid(sma, semid) \
84 ipc_checkid(&sem_ids,&sma->sem_perm,semid)
85#define sem_buildid(id, seq) \
86 ipc_buildid(&sem_ids, id, seq)
87static struct ipc_ids sem_ids;
88
89static int newary (key_t, int, int);
90static void freeary (struct sem_array *sma, int id);
91#ifdef CONFIG_PROC_FS
92static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data);
93#endif
94
95#define SEMMSL_FAST 256 /* 512 bytes on stack */
96#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
97
98/*
99 * linked list protection:
100 * sem_undo.id_next,
101 * sem_array.sem_pending{,last},
102 * sem_array.sem_undo: sem_lock() for read/write
103 * sem_undo.proc_next: only "current" is allowed to read/write that field.
104 *
105 */
106
107int sem_ctls[4] = {SEMMSL, SEMMNS, SEMOPM, SEMMNI};
108#define sc_semmsl (sem_ctls[0])
109#define sc_semmns (sem_ctls[1])
110#define sc_semopm (sem_ctls[2])
111#define sc_semmni (sem_ctls[3])
112
113static int used_sems;
114
115void __init sem_init (void)
116{
117 used_sems = 0;
118 ipc_init_ids(&sem_ids,sc_semmni);
119
120#ifdef CONFIG_PROC_FS
121 create_proc_read_entry("sysvipc/sem", 0, NULL, sysvipc_sem_read_proc, NULL);
122#endif
123}
124
125/*
126 * Lockless wakeup algorithm:
127 * Without the check/retry algorithm a lockless wakeup is possible:
128 * - queue.status is initialized to -EINTR before blocking.
129 * - wakeup is performed by
130 * * unlinking the queue entry from sma->sem_pending
131 * * setting queue.status to IN_WAKEUP
132 * This is the notification for the blocked thread that a
133 * result value is imminent.
134 * * call wake_up_process
135 * * set queue.status to the final value.
136 * - the previously blocked thread checks queue.status:
137 * * if it's IN_WAKEUP, then it must wait until the value changes
138 * * if it's not -EINTR, then the operation was completed by
139 * update_queue. semtimedop can return queue.status without
140 * performing any operation on the semaphore array.
141 * * otherwise it must acquire the spinlock and check what's up.
142 *
143 * The two-stage algorithm is necessary to protect against the following
144 * races:
145 * - if queue.status is set after wake_up_process, then the woken up idle
146 * thread could race forward and try (and fail) to acquire sma->lock
147 * before update_queue had a chance to set queue.status
148 * - if queue.status is written before wake_up_process and if the
149 * blocked process is woken up by a signal between writing
150 * queue.status and the wake_up_process, then the woken up
151 * process could return from semtimedop and die by calling
152 * sys_exit before wake_up_process is called. Then wake_up_process
153 * will oops, because the task structure is already invalid.
154 * (yes, this happened on s390 with sysv msg).
155 *
156 */
157#define IN_WAKEUP 1
158
159static int newary (key_t key, int nsems, int semflg)
160{
161 int id;
162 int retval;
163 struct sem_array *sma;
164 int size;
165
166 if (!nsems)
167 return -EINVAL;
168 if (used_sems + nsems > sc_semmns)
169 return -ENOSPC;
170
171 size = sizeof (*sma) + nsems * sizeof (struct sem);
172 sma = ipc_rcu_alloc(size);
173 if (!sma) {
174 return -ENOMEM;
175 }
176 memset (sma, 0, size);
177
178 sma->sem_perm.mode = (semflg & S_IRWXUGO);
179 sma->sem_perm.key = key;
180
181 sma->sem_perm.security = NULL;
182 retval = security_sem_alloc(sma);
183 if (retval) {
184 ipc_rcu_putref(sma);
185 return retval;
186 }
187
188 id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni);
189 if(id == -1) {
190 security_sem_free(sma);
191 ipc_rcu_putref(sma);
192 return -ENOSPC;
193 }
194 used_sems += nsems;
195
196 sma->sem_base = (struct sem *) &sma[1];
197 /* sma->sem_pending = NULL; */
198 sma->sem_pending_last = &sma->sem_pending;
199 /* sma->undo = NULL; */
200 sma->sem_nsems = nsems;
201 sma->sem_ctime = get_seconds();
202 sem_unlock(sma);
203
204 return sem_buildid(id, sma->sem_perm.seq);
205}
206
207asmlinkage long sys_semget (key_t key, int nsems, int semflg)
208{
209 int id, err = -EINVAL;
210 struct sem_array *sma;
211
212 if (nsems < 0 || nsems > sc_semmsl)
213 return -EINVAL;
214 down(&sem_ids.sem);
215
216 if (key == IPC_PRIVATE) {
217 err = newary(key, nsems, semflg);
218 } else if ((id = ipc_findkey(&sem_ids, key)) == -1) { /* key not used */
219 if (!(semflg & IPC_CREAT))
220 err = -ENOENT;
221 else
222 err = newary(key, nsems, semflg);
223 } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) {
224 err = -EEXIST;
225 } else {
226 sma = sem_lock(id);
227 if(sma==NULL)
228 BUG();
229 if (nsems > sma->sem_nsems)
230 err = -EINVAL;
231 else if (ipcperms(&sma->sem_perm, semflg))
232 err = -EACCES;
233 else {
234 int semid = sem_buildid(id, sma->sem_perm.seq);
235 err = security_sem_associate(sma, semflg);
236 if (!err)
237 err = semid;
238 }
239 sem_unlock(sma);
240 }
241
242 up(&sem_ids.sem);
243 return err;
244}
245
246/* Manage the doubly linked list sma->sem_pending as a FIFO:
247 * insert new queue elements at the tail sma->sem_pending_last.
248 */
249static inline void append_to_queue (struct sem_array * sma,
250 struct sem_queue * q)
251{
252 *(q->prev = sma->sem_pending_last) = q;
253 *(sma->sem_pending_last = &q->next) = NULL;
254}
255
256static inline void prepend_to_queue (struct sem_array * sma,
257 struct sem_queue * q)
258{
259 q->next = sma->sem_pending;
260 *(q->prev = &sma->sem_pending) = q;
261 if (q->next)
262 q->next->prev = &q->next;
263 else /* sma->sem_pending_last == &sma->sem_pending */
264 sma->sem_pending_last = &q->next;
265}
266
267static inline void remove_from_queue (struct sem_array * sma,
268 struct sem_queue * q)
269{
270 *(q->prev) = q->next;
271 if (q->next)
272 q->next->prev = q->prev;
273 else /* sma->sem_pending_last == &q->next */
274 sma->sem_pending_last = q->prev;
275 q->prev = NULL; /* mark as removed */
276}
277
278/*
279 * Determine whether a sequence of semaphore operations would succeed
280 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
281 */
282
283static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
284 int nsops, struct sem_undo *un, int pid)
285{
286 int result, sem_op;
287 struct sembuf *sop;
288 struct sem * curr;
289
290 for (sop = sops; sop < sops + nsops; sop++) {
291 curr = sma->sem_base + sop->sem_num;
292 sem_op = sop->sem_op;
293 result = curr->semval;
294
295 if (!sem_op && result)
296 goto would_block;
297
298 result += sem_op;
299 if (result < 0)
300 goto would_block;
301 if (result > SEMVMX)
302 goto out_of_range;
303 if (sop->sem_flg & SEM_UNDO) {
304 int undo = un->semadj[sop->sem_num] - sem_op;
305 /*
306 * Exceeding the undo range is an error.
307 */
308 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
309 goto out_of_range;
310 }
311 curr->semval = result;
312 }
313
314 sop--;
315 while (sop >= sops) {
316 sma->sem_base[sop->sem_num].sempid = pid;
317 if (sop->sem_flg & SEM_UNDO)
318 un->semadj[sop->sem_num] -= sop->sem_op;
319 sop--;
320 }
321
322 sma->sem_otime = get_seconds();
323 return 0;
324
325out_of_range:
326 result = -ERANGE;
327 goto undo;
328
329would_block:
330 if (sop->sem_flg & IPC_NOWAIT)
331 result = -EAGAIN;
332 else
333 result = 1;
334
335undo:
336 sop--;
337 while (sop >= sops) {
338 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
339 sop--;
340 }
341
342 return result;
343}
344
345/* Go through the pending queue for the indicated semaphore
346 * looking for tasks that can be completed.
347 */
348static void update_queue (struct sem_array * sma)
349{
350 int error;
351 struct sem_queue * q;
352
353 q = sma->sem_pending;
354 while(q) {
355 error = try_atomic_semop(sma, q->sops, q->nsops,
356 q->undo, q->pid);
357
358 /* Does q->sleeper still need to sleep? */
359 if (error <= 0) {
360 struct sem_queue *n;
361 remove_from_queue(sma,q);
362 q->status = IN_WAKEUP;
363 /*
364 * Continue scanning. The next operation
365 * that must be checked depends on the type of the
366 * completed operation:
367 * - if the operation modified the array, then
368 * restart from the head of the queue and
369 * check for threads that might be waiting
370 * for semaphore values to become 0.
371 * - if the operation didn't modify the array,
372 * then just continue.
373 */
374 if (q->alter)
375 n = sma->sem_pending;
376 else
377 n = q->next;
378 wake_up_process(q->sleeper);
379 /* hands-off: q will disappear immediately after
380 * writing q->status.
381 */
382 q->status = error;
383 q = n;
384 } else {
385 q = q->next;
386 }
387 }
388}
389
390/* The following counts are associated to each semaphore:
391 * semncnt number of tasks waiting on semval being nonzero
392 * semzcnt number of tasks waiting on semval being zero
393 * This model assumes that a task waits on exactly one semaphore.
394 * Since semaphore operations are to be performed atomically, tasks actually
395 * wait on a whole sequence of semaphores simultaneously.
396 * The counts we return here are a rough approximation, but still
397 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
398 */
399static int count_semncnt (struct sem_array * sma, ushort semnum)
400{
401 int semncnt;
402 struct sem_queue * q;
403
404 semncnt = 0;
405 for (q = sma->sem_pending; q; q = q->next) {
406 struct sembuf * sops = q->sops;
407 int nsops = q->nsops;
408 int i;
409 for (i = 0; i < nsops; i++)
410 if (sops[i].sem_num == semnum
411 && (sops[i].sem_op < 0)
412 && !(sops[i].sem_flg & IPC_NOWAIT))
413 semncnt++;
414 }
415 return semncnt;
416}
417static int count_semzcnt (struct sem_array * sma, ushort semnum)
418{
419 int semzcnt;
420 struct sem_queue * q;
421
422 semzcnt = 0;
423 for (q = sma->sem_pending; q; q = q->next) {
424 struct sembuf * sops = q->sops;
425 int nsops = q->nsops;
426 int i;
427 for (i = 0; i < nsops; i++)
428 if (sops[i].sem_num == semnum
429 && (sops[i].sem_op == 0)
430 && !(sops[i].sem_flg & IPC_NOWAIT))
431 semzcnt++;
432 }
433 return semzcnt;
434}
435
436/* Free a semaphore set. freeary() is called with sem_ids.sem down and
437 * the spinlock for this semaphore set hold. sem_ids.sem remains locked
438 * on exit.
439 */
440static void freeary (struct sem_array *sma, int id)
441{
442 struct sem_undo *un;
443 struct sem_queue *q;
444 int size;
445
446 /* Invalidate the existing undo structures for this semaphore set.
447 * (They will be freed without any further action in exit_sem()
448 * or during the next semop.)
449 */
450 for (un = sma->undo; un; un = un->id_next)
451 un->semid = -1;
452
453 /* Wake up all pending processes and let them fail with EIDRM. */
454 q = sma->sem_pending;
455 while(q) {
456 struct sem_queue *n;
457 /* lazy remove_from_queue: we are killing the whole queue */
458 q->prev = NULL;
459 n = q->next;
460 q->status = IN_WAKEUP;
461 wake_up_process(q->sleeper); /* doesn't sleep */
462 q->status = -EIDRM; /* hands-off q */
463 q = n;
464 }
465
466 /* Remove the semaphore set from the ID array*/
467 sma = sem_rmid(id);
468 sem_unlock(sma);
469
470 used_sems -= sma->sem_nsems;
471 size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem);
472 security_sem_free(sma);
473 ipc_rcu_putref(sma);
474}
475
476static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
477{
478 switch(version) {
479 case IPC_64:
480 return copy_to_user(buf, in, sizeof(*in));
481 case IPC_OLD:
482 {
483 struct semid_ds out;
484
485 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
486
487 out.sem_otime = in->sem_otime;
488 out.sem_ctime = in->sem_ctime;
489 out.sem_nsems = in->sem_nsems;
490
491 return copy_to_user(buf, &out, sizeof(out));
492 }
493 default:
494 return -EINVAL;
495 }
496}
497
498static int semctl_nolock(int semid, int semnum, int cmd, int version, union semun arg)
499{
500 int err = -EINVAL;
501 struct sem_array *sma;
502
503 switch(cmd) {
504 case IPC_INFO:
505 case SEM_INFO:
506 {
507 struct seminfo seminfo;
508 int max_id;
509
510 err = security_sem_semctl(NULL, cmd);
511 if (err)
512 return err;
513
514 memset(&seminfo,0,sizeof(seminfo));
515 seminfo.semmni = sc_semmni;
516 seminfo.semmns = sc_semmns;
517 seminfo.semmsl = sc_semmsl;
518 seminfo.semopm = sc_semopm;
519 seminfo.semvmx = SEMVMX;
520 seminfo.semmnu = SEMMNU;
521 seminfo.semmap = SEMMAP;
522 seminfo.semume = SEMUME;
523 down(&sem_ids.sem);
524 if (cmd == SEM_INFO) {
525 seminfo.semusz = sem_ids.in_use;
526 seminfo.semaem = used_sems;
527 } else {
528 seminfo.semusz = SEMUSZ;
529 seminfo.semaem = SEMAEM;
530 }
531 max_id = sem_ids.max_id;
532 up(&sem_ids.sem);
533 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
534 return -EFAULT;
535 return (max_id < 0) ? 0: max_id;
536 }
537 case SEM_STAT:
538 {
539 struct semid64_ds tbuf;
540 int id;
541
542 if(semid >= sem_ids.entries->size)
543 return -EINVAL;
544
545 memset(&tbuf,0,sizeof(tbuf));
546
547 sma = sem_lock(semid);
548 if(sma == NULL)
549 return -EINVAL;
550
551 err = -EACCES;
552 if (ipcperms (&sma->sem_perm, S_IRUGO))
553 goto out_unlock;
554
555 err = security_sem_semctl(sma, cmd);
556 if (err)
557 goto out_unlock;
558
559 id = sem_buildid(semid, sma->sem_perm.seq);
560
561 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
562 tbuf.sem_otime = sma->sem_otime;
563 tbuf.sem_ctime = sma->sem_ctime;
564 tbuf.sem_nsems = sma->sem_nsems;
565 sem_unlock(sma);
566 if (copy_semid_to_user (arg.buf, &tbuf, version))
567 return -EFAULT;
568 return id;
569 }
570 default:
571 return -EINVAL;
572 }
573 return err;
574out_unlock:
575 sem_unlock(sma);
576 return err;
577}
578
579static int semctl_main(int semid, int semnum, int cmd, int version, union semun arg)
580{
581 struct sem_array *sma;
582 struct sem* curr;
583 int err;
584 ushort fast_sem_io[SEMMSL_FAST];
585 ushort* sem_io = fast_sem_io;
586 int nsems;
587
588 sma = sem_lock(semid);
589 if(sma==NULL)
590 return -EINVAL;
591
592 nsems = sma->sem_nsems;
593
594 err=-EIDRM;
595 if (sem_checkid(sma,semid))
596 goto out_unlock;
597
598 err = -EACCES;
599 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
600 goto out_unlock;
601
602 err = security_sem_semctl(sma, cmd);
603 if (err)
604 goto out_unlock;
605
606 err = -EACCES;
607 switch (cmd) {
608 case GETALL:
609 {
610 ushort __user *array = arg.array;
611 int i;
612
613 if(nsems > SEMMSL_FAST) {
614 ipc_rcu_getref(sma);
615 sem_unlock(sma);
616
617 sem_io = ipc_alloc(sizeof(ushort)*nsems);
618 if(sem_io == NULL) {
619 ipc_lock_by_ptr(&sma->sem_perm);
620 ipc_rcu_putref(sma);
621 sem_unlock(sma);
622 return -ENOMEM;
623 }
624
625 ipc_lock_by_ptr(&sma->sem_perm);
626 ipc_rcu_putref(sma);
627 if (sma->sem_perm.deleted) {
628 sem_unlock(sma);
629 err = -EIDRM;
630 goto out_free;
631 }
632 }
633
634 for (i = 0; i < sma->sem_nsems; i++)
635 sem_io[i] = sma->sem_base[i].semval;
636 sem_unlock(sma);
637 err = 0;
638 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
639 err = -EFAULT;
640 goto out_free;
641 }
642 case SETALL:
643 {
644 int i;
645 struct sem_undo *un;
646
647 ipc_rcu_getref(sma);
648 sem_unlock(sma);
649
650 if(nsems > SEMMSL_FAST) {
651 sem_io = ipc_alloc(sizeof(ushort)*nsems);
652 if(sem_io == NULL) {
653 ipc_lock_by_ptr(&sma->sem_perm);
654 ipc_rcu_putref(sma);
655 sem_unlock(sma);
656 return -ENOMEM;
657 }
658 }
659
660 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
661 ipc_lock_by_ptr(&sma->sem_perm);
662 ipc_rcu_putref(sma);
663 sem_unlock(sma);
664 err = -EFAULT;
665 goto out_free;
666 }
667
668 for (i = 0; i < nsems; i++) {
669 if (sem_io[i] > SEMVMX) {
670 ipc_lock_by_ptr(&sma->sem_perm);
671 ipc_rcu_putref(sma);
672 sem_unlock(sma);
673 err = -ERANGE;
674 goto out_free;
675 }
676 }
677 ipc_lock_by_ptr(&sma->sem_perm);
678 ipc_rcu_putref(sma);
679 if (sma->sem_perm.deleted) {
680 sem_unlock(sma);
681 err = -EIDRM;
682 goto out_free;
683 }
684
685 for (i = 0; i < nsems; i++)
686 sma->sem_base[i].semval = sem_io[i];
687 for (un = sma->undo; un; un = un->id_next)
688 for (i = 0; i < nsems; i++)
689 un->semadj[i] = 0;
690 sma->sem_ctime = get_seconds();
691 /* maybe some queued-up processes were waiting for this */
692 update_queue(sma);
693 err = 0;
694 goto out_unlock;
695 }
696 case IPC_STAT:
697 {
698 struct semid64_ds tbuf;
699 memset(&tbuf,0,sizeof(tbuf));
700 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
701 tbuf.sem_otime = sma->sem_otime;
702 tbuf.sem_ctime = sma->sem_ctime;
703 tbuf.sem_nsems = sma->sem_nsems;
704 sem_unlock(sma);
705 if (copy_semid_to_user (arg.buf, &tbuf, version))
706 return -EFAULT;
707 return 0;
708 }
709 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
710 }
711 err = -EINVAL;
712 if(semnum < 0 || semnum >= nsems)
713 goto out_unlock;
714
715 curr = &sma->sem_base[semnum];
716
717 switch (cmd) {
718 case GETVAL:
719 err = curr->semval;
720 goto out_unlock;
721 case GETPID:
722 err = curr->sempid;
723 goto out_unlock;
724 case GETNCNT:
725 err = count_semncnt(sma,semnum);
726 goto out_unlock;
727 case GETZCNT:
728 err = count_semzcnt(sma,semnum);
729 goto out_unlock;
730 case SETVAL:
731 {
732 int val = arg.val;
733 struct sem_undo *un;
734 err = -ERANGE;
735 if (val > SEMVMX || val < 0)
736 goto out_unlock;
737
738 for (un = sma->undo; un; un = un->id_next)
739 un->semadj[semnum] = 0;
740 curr->semval = val;
741 curr->sempid = current->tgid;
742 sma->sem_ctime = get_seconds();
743 /* maybe some queued-up processes were waiting for this */
744 update_queue(sma);
745 err = 0;
746 goto out_unlock;
747 }
748 }
749out_unlock:
750 sem_unlock(sma);
751out_free:
752 if(sem_io != fast_sem_io)
753 ipc_free(sem_io, sizeof(ushort)*nsems);
754 return err;
755}
756
757struct sem_setbuf {
758 uid_t uid;
759 gid_t gid;
760 mode_t mode;
761};
762
763static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
764{
765 switch(version) {
766 case IPC_64:
767 {
768 struct semid64_ds tbuf;
769
770 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
771 return -EFAULT;
772
773 out->uid = tbuf.sem_perm.uid;
774 out->gid = tbuf.sem_perm.gid;
775 out->mode = tbuf.sem_perm.mode;
776
777 return 0;
778 }
779 case IPC_OLD:
780 {
781 struct semid_ds tbuf_old;
782
783 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
784 return -EFAULT;
785
786 out->uid = tbuf_old.sem_perm.uid;
787 out->gid = tbuf_old.sem_perm.gid;
788 out->mode = tbuf_old.sem_perm.mode;
789
790 return 0;
791 }
792 default:
793 return -EINVAL;
794 }
795}
796
797static int semctl_down(int semid, int semnum, int cmd, int version, union semun arg)
798{
799 struct sem_array *sma;
800 int err;
801 struct sem_setbuf setbuf;
802 struct kern_ipc_perm *ipcp;
803
804 if(cmd == IPC_SET) {
805 if(copy_semid_from_user (&setbuf, arg.buf, version))
806 return -EFAULT;
807 if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode)))
808 return err;
809 }
810 sma = sem_lock(semid);
811 if(sma==NULL)
812 return -EINVAL;
813
814 if (sem_checkid(sma,semid)) {
815 err=-EIDRM;
816 goto out_unlock;
817 }
818 ipcp = &sma->sem_perm;
819
820 if (current->euid != ipcp->cuid &&
821 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
822 err=-EPERM;
823 goto out_unlock;
824 }
825
826 err = security_sem_semctl(sma, cmd);
827 if (err)
828 goto out_unlock;
829
830 switch(cmd){
831 case IPC_RMID:
832 freeary(sma, semid);
833 err = 0;
834 break;
835 case IPC_SET:
836 ipcp->uid = setbuf.uid;
837 ipcp->gid = setbuf.gid;
838 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
839 | (setbuf.mode & S_IRWXUGO);
840 sma->sem_ctime = get_seconds();
841 sem_unlock(sma);
842 err = 0;
843 break;
844 default:
845 sem_unlock(sma);
846 err = -EINVAL;
847 break;
848 }
849 return err;
850
851out_unlock:
852 sem_unlock(sma);
853 return err;
854}
855
856asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
857{
858 int err = -EINVAL;
859 int version;
860
861 if (semid < 0)
862 return -EINVAL;
863
864 version = ipc_parse_version(&cmd);
865
866 switch(cmd) {
867 case IPC_INFO:
868 case SEM_INFO:
869 case SEM_STAT:
870 err = semctl_nolock(semid,semnum,cmd,version,arg);
871 return err;
872 case GETALL:
873 case GETVAL:
874 case GETPID:
875 case GETNCNT:
876 case GETZCNT:
877 case IPC_STAT:
878 case SETVAL:
879 case SETALL:
880 err = semctl_main(semid,semnum,cmd,version,arg);
881 return err;
882 case IPC_RMID:
883 case IPC_SET:
884 down(&sem_ids.sem);
885 err = semctl_down(semid,semnum,cmd,version,arg);
886 up(&sem_ids.sem);
887 return err;
888 default:
889 return -EINVAL;
890 }
891}
892
893static inline void lock_semundo(void)
894{
895 struct sem_undo_list *undo_list;
896
897 undo_list = current->sysvsem.undo_list;
898 if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
899 spin_lock(&undo_list->lock);
900}
901
902/* This code has an interaction with copy_semundo().
903 * Consider; two tasks are sharing the undo_list. task1
904 * acquires the undo_list lock in lock_semundo(). If task2 now
905 * exits before task1 releases the lock (by calling
906 * unlock_semundo()), then task1 will never call spin_unlock().
907 * This leave the sem_undo_list in a locked state. If task1 now creats task3
908 * and once again shares the sem_undo_list, the sem_undo_list will still be
909 * locked, and future SEM_UNDO operations will deadlock. This case is
910 * dealt with in copy_semundo() by having it reinitialize the spin lock when
911 * the refcnt goes from 1 to 2.
912 */
913static inline void unlock_semundo(void)
914{
915 struct sem_undo_list *undo_list;
916
917 undo_list = current->sysvsem.undo_list;
918 if ((undo_list != NULL) && (atomic_read(&undo_list->refcnt) != 1))
919 spin_unlock(&undo_list->lock);
920}
921
922
923/* If the task doesn't already have a undo_list, then allocate one
924 * here. We guarantee there is only one thread using this undo list,
925 * and current is THE ONE
926 *
927 * If this allocation and assignment succeeds, but later
928 * portions of this code fail, there is no need to free the sem_undo_list.
929 * Just let it stay associated with the task, and it'll be freed later
930 * at exit time.
931 *
932 * This can block, so callers must hold no locks.
933 */
934static inline int get_undo_list(struct sem_undo_list **undo_listp)
935{
936 struct sem_undo_list *undo_list;
937 int size;
938
939 undo_list = current->sysvsem.undo_list;
940 if (!undo_list) {
941 size = sizeof(struct sem_undo_list);
942 undo_list = (struct sem_undo_list *) kmalloc(size, GFP_KERNEL);
943 if (undo_list == NULL)
944 return -ENOMEM;
945 memset(undo_list, 0, size);
946 /* don't initialize unodhd->lock here. It's done
947 * in copy_semundo() instead.
948 */
949 atomic_set(&undo_list->refcnt, 1);
950 current->sysvsem.undo_list = undo_list;
951 }
952 *undo_listp = undo_list;
953 return 0;
954}
955
956static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
957{
958 struct sem_undo **last, *un;
959
960 last = &ulp->proc_list;
961 un = *last;
962 while(un != NULL) {
963 if(un->semid==semid)
964 break;
965 if(un->semid==-1) {
966 *last=un->proc_next;
967 kfree(un);
968 } else {
969 last=&un->proc_next;
970 }
971 un=*last;
972 }
973 return un;
974}
975
976static struct sem_undo *find_undo(int semid)
977{
978 struct sem_array *sma;
979 struct sem_undo_list *ulp;
980 struct sem_undo *un, *new;
981 int nsems;
982 int error;
983
984 error = get_undo_list(&ulp);
985 if (error)
986 return ERR_PTR(error);
987
988 lock_semundo();
989 un = lookup_undo(ulp, semid);
990 unlock_semundo();
991 if (likely(un!=NULL))
992 goto out;
993
994 /* no undo structure around - allocate one. */
995 sma = sem_lock(semid);
996 un = ERR_PTR(-EINVAL);
997 if(sma==NULL)
998 goto out;
999 un = ERR_PTR(-EIDRM);
1000 if (sem_checkid(sma,semid)) {
1001 sem_unlock(sma);
1002 goto out;
1003 }
1004 nsems = sma->sem_nsems;
1005 ipc_rcu_getref(sma);
1006 sem_unlock(sma);
1007
1008 new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1009 if (!new) {
1010 ipc_lock_by_ptr(&sma->sem_perm);
1011 ipc_rcu_putref(sma);
1012 sem_unlock(sma);
1013 return ERR_PTR(-ENOMEM);
1014 }
1015 memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems);
1016 new->semadj = (short *) &new[1];
1017 new->semid = semid;
1018
1019 lock_semundo();
1020 un = lookup_undo(ulp, semid);
1021 if (un) {
1022 unlock_semundo();
1023 kfree(new);
1024 ipc_lock_by_ptr(&sma->sem_perm);
1025 ipc_rcu_putref(sma);
1026 sem_unlock(sma);
1027 goto out;
1028 }
1029 ipc_lock_by_ptr(&sma->sem_perm);
1030 ipc_rcu_putref(sma);
1031 if (sma->sem_perm.deleted) {
1032 sem_unlock(sma);
1033 unlock_semundo();
1034 kfree(new);
1035 un = ERR_PTR(-EIDRM);
1036 goto out;
1037 }
1038 new->proc_next = ulp->proc_list;
1039 ulp->proc_list = new;
1040 new->id_next = sma->undo;
1041 sma->undo = new;
1042 sem_unlock(sma);
1043 un = new;
1044 unlock_semundo();
1045out:
1046 return un;
1047}
1048
1049asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1050 unsigned nsops, const struct timespec __user *timeout)
1051{
1052 int error = -EINVAL;
1053 struct sem_array *sma;
1054 struct sembuf fast_sops[SEMOPM_FAST];
1055 struct sembuf* sops = fast_sops, *sop;
1056 struct sem_undo *un;
1057 int undos = 0, decrease = 0, alter = 0, max;
1058 struct sem_queue queue;
1059 unsigned long jiffies_left = 0;
1060
1061 if (nsops < 1 || semid < 0)
1062 return -EINVAL;
1063 if (nsops > sc_semopm)
1064 return -E2BIG;
1065 if(nsops > SEMOPM_FAST) {
1066 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1067 if(sops==NULL)
1068 return -ENOMEM;
1069 }
1070 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1071 error=-EFAULT;
1072 goto out_free;
1073 }
1074 if (timeout) {
1075 struct timespec _timeout;
1076 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1077 error = -EFAULT;
1078 goto out_free;
1079 }
1080 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1081 _timeout.tv_nsec >= 1000000000L) {
1082 error = -EINVAL;
1083 goto out_free;
1084 }
1085 jiffies_left = timespec_to_jiffies(&_timeout);
1086 }
1087 max = 0;
1088 for (sop = sops; sop < sops + nsops; sop++) {
1089 if (sop->sem_num >= max)
1090 max = sop->sem_num;
1091 if (sop->sem_flg & SEM_UNDO)
1092 undos++;
1093 if (sop->sem_op < 0)
1094 decrease = 1;
1095 if (sop->sem_op > 0)
1096 alter = 1;
1097 }
1098 alter |= decrease;
1099
1100retry_undos:
1101 if (undos) {
1102 un = find_undo(semid);
1103 if (IS_ERR(un)) {
1104 error = PTR_ERR(un);
1105 goto out_free;
1106 }
1107 } else
1108 un = NULL;
1109
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;
1117 /*
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.
1121 */
1122 if (un && un->semid == -1) {
1123 sem_unlock(sma);
1124 goto retry_undos;
1125 }
1126 error = -EFBIG;
1127 if (max >= sma->sem_nsems)
1128 goto out_unlock_free;
1129
1130 error = -EACCES;
1131 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1132 goto out_unlock_free;
1133
1134 error = security_sem_semop(sma, sops, nsops, alter);
1135 if (error)
1136 goto out_unlock_free;
1137
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;
1143 }
1144
1145 /* We need to sleep on this operation, so we put the current
1146 * task into the pending queue and go to sleep.
1147 */
1148
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);
1160
1161 queue.status = -EINTR;
1162 queue.sleeper = current;
1163 current->state = TASK_INTERRUPTIBLE;
1164 sem_unlock(sma);
1165
1166 if (timeout)
1167 jiffies_left = schedule_timeout(jiffies_left);
1168 else
1169 schedule();
1170
1171 error = queue.status;
1172 while(unlikely(error == IN_WAKEUP)) {
1173 cpu_relax();
1174 error = queue.status;
1175 }
1176
1177 if (error != -EINTR) {
1178 /* fast path: update_queue already obtained all requested
1179 * resources */
1180 goto out_free;
1181 }
1182
1183 sma = sem_lock(semid);
1184 if(sma==NULL) {
1185 if(queue.prev != NULL)
1186 BUG();
1187 error = -EIDRM;
1188 goto out_free;
1189 }
1190
1191 /*
1192 * If queue.status != -EINTR we are woken up by another process
1193 */
1194 error = queue.status;
1195 if (error != -EINTR) {
1196 goto out_unlock_free;
1197 }
1198
1199 /*
1200 * If an interrupt occurred we have to clean up the queue
1201 */
1202 if (timeout && jiffies_left == 0)
1203 error = -EAGAIN;
1204 remove_from_queue(sma,&queue);
1205 goto out_unlock_free;
1206
1207out_unlock_free:
1208 sem_unlock(sma);
1209out_free:
1210 if(sops != fast_sops)
1211 kfree(sops);
1212 return error;
1213}
1214
1215asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1216{
1217 return sys_semtimedop(semid, tsops, nsops, NULL);
1218}
1219
1220/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1221 * parent and child tasks.
1222 *
1223 * See the notes above unlock_semundo() regarding the spin_lock_init()
1224 * in this code. Initialize the undo_list->lock here instead of get_undo_list()
1225 * because of the reasoning in the comment above unlock_semundo.
1226 */
1227
1228int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1229{
1230 struct sem_undo_list *undo_list;
1231 int error;
1232
1233 if (clone_flags & CLONE_SYSVSEM) {
1234 error = get_undo_list(&undo_list);
1235 if (error)
1236 return error;
1237 if (atomic_read(&undo_list->refcnt) == 1)
1238 spin_lock_init(&undo_list->lock);
1239 atomic_inc(&undo_list->refcnt);
1240 tsk->sysvsem.undo_list = undo_list;
1241 } else
1242 tsk->sysvsem.undo_list = NULL;
1243
1244 return 0;
1245}
1246
1247/*
1248 * add semadj values to semaphores, free undo structures.
1249 * undo structures are not freed when semaphore arrays are destroyed
1250 * so some of them may be out of date.
1251 * IMPLEMENTATION NOTE: There is some confusion over whether the
1252 * set of adjustments that needs to be done should be done in an atomic
1253 * manner or not. That is, if we are attempting to decrement the semval
1254 * should we queue up and wait until we can do so legally?
1255 * The original implementation attempted to do this (queue and wait).
1256 * The current implementation does not do so. The POSIX standard
1257 * and SVID should be consulted to determine what behavior is mandated.
1258 */
1259void exit_sem(struct task_struct *tsk)
1260{
1261 struct sem_undo_list *undo_list;
1262 struct sem_undo *u, **up;
1263
1264 undo_list = tsk->sysvsem.undo_list;
1265 if (!undo_list)
1266 return;
1267
1268 if (!atomic_dec_and_test(&undo_list->refcnt))
1269 return;
1270
1271 /* There's no need to hold the semundo list lock, as current
1272 * is the last task exiting for this undo list.
1273 */
1274 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1275 struct sem_array *sma;
1276 int nsems, i;
1277 struct sem_undo *un, **unp;
1278 int semid;
1279
1280 semid = u->semid;
1281
1282 if(semid == -1)
1283 continue;
1284 sma = sem_lock(semid);
1285 if (sma == NULL)
1286 continue;
1287
1288 if (u->semid == -1)
1289 goto next_entry;
1290
1291 BUG_ON(sem_checkid(sma,u->semid));
1292
1293 /* remove u from the sma->undo list */
1294 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1295 if (u == un)
1296 goto found;
1297 }
1298 printk ("exit_sem undo list error id=%d\n", u->semid);
1299 goto next_entry;
1300found:
1301 *unp = un->id_next;
1302 /* perform adjustments registered in u */
1303 nsems = sma->sem_nsems;
1304 for (i = 0; i < nsems; i++) {
1305 struct sem * sem = &sma->sem_base[i];
1306 if (u->semadj[i]) {
1307 sem->semval += u->semadj[i];
1308 /*
1309 * Range checks of the new semaphore value,
1310 * not defined by sus:
1311 * - Some unices ignore the undo entirely
1312 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1313 * - some cap the value (e.g. FreeBSD caps
1314 * at 0, but doesn't enforce SEMVMX)
1315 *
1316 * Linux caps the semaphore value, both at 0
1317 * and at SEMVMX.
1318 *
1319 * Manfred <manfred@colorfullife.com>
1320 */
1321 if (sem->semval < 0)
1322 sem->semval = 0;
1323 if (sem->semval > SEMVMX)
1324 sem->semval = SEMVMX;
1325 sem->sempid = current->tgid;
1326 }
1327 }
1328 sma->sem_otime = get_seconds();
1329 /* maybe some queued-up processes were waiting for this */
1330 update_queue(sma);
1331next_entry:
1332 sem_unlock(sma);
1333 }
1334 kfree(undo_list);
1335}
1336
1337#ifdef CONFIG_PROC_FS
1338static int sysvipc_sem_read_proc(char *buffer, char **start, off_t offset, int length, int *eof, void *data)
1339{
1340 off_t pos = 0;
1341 off_t begin = 0;
1342 int i, len = 0;
1343
1344 len += sprintf(buffer, " key semid perms nsems uid gid cuid cgid otime ctime\n");
1345 down(&sem_ids.sem);
1346
1347 for(i = 0; i <= sem_ids.max_id; i++) {
1348 struct sem_array *sma;
1349 sma = sem_lock(i);
1350 if(sma) {
1351 len += sprintf(buffer + len, "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1352 sma->sem_perm.key,
1353 sem_buildid(i,sma->sem_perm.seq),
1354 sma->sem_perm.mode,
1355 sma->sem_nsems,
1356 sma->sem_perm.uid,
1357 sma->sem_perm.gid,
1358 sma->sem_perm.cuid,
1359 sma->sem_perm.cgid,
1360 sma->sem_otime,
1361 sma->sem_ctime);
1362 sem_unlock(sma);
1363
1364 pos += len;
1365 if(pos < offset) {
1366 len = 0;
1367 begin = pos;
1368 }
1369 if(pos > offset + length)
1370 goto done;
1371 }
1372 }
1373 *eof = 1;
1374done:
1375 up(&sem_ids.sem);
1376 *start = buffer + (offset - begin);
1377 len -= (offset - begin);
1378 if(len > length)
1379 len = length;
1380 if(len < 0)
1381 len = 0;
1382 return len;
1383}
1384#endif