]> bbs.cooldavid.org Git - net-next-2.6.git/blame - ipc/sem.c
ipc/sem.c: bugfix for semop() not reporting successful operation
[net-next-2.6.git] / ipc / sem.c
CommitLineData
1da177e4
LT
1/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
1da177e4
LT
6 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
7 *
8 * SMP-threaded, sysctl's added
624dffcb 9 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
1da177e4 10 * Enforced range limit on SEM_UNDO
046c6884 11 * (c) 2001 Red Hat Inc
1da177e4
LT
12 * Lockless wakeup
13 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
c5cf6359
MS
14 * Further wakeup optimizations, documentation
15 * (c) 2010 Manfred Spraul <manfred@colorfullife.com>
073115d6
SG
16 *
17 * support for audit of ipc object properties and permission changes
18 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
e3893534
KK
19 *
20 * namespaces support
21 * OpenVZ, SWsoft Inc.
22 * Pavel Emelianov <xemul@openvz.org>
c5cf6359
MS
23 *
24 * Implementation notes: (May 2010)
25 * This file implements System V semaphores.
26 *
27 * User space visible behavior:
28 * - FIFO ordering for semop() operations (just FIFO, not starvation
29 * protection)
30 * - multiple semaphore operations that alter the same semaphore in
31 * one semop() are handled.
32 * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and
33 * SETALL calls.
34 * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO.
35 * - undo adjustments at process exit are limited to 0..SEMVMX.
36 * - namespace are supported.
37 * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing
38 * to /proc/sys/kernel/sem.
39 * - statistics about the usage are reported in /proc/sysvipc/sem.
40 *
41 * Internals:
42 * - scalability:
43 * - all global variables are read-mostly.
44 * - semop() calls and semctl(RMID) are synchronized by RCU.
45 * - most operations do write operations (actually: spin_lock calls) to
46 * the per-semaphore array structure.
47 * Thus: Perfect SMP scaling between independent semaphore arrays.
48 * If multiple semaphores in one array are used, then cache line
49 * trashing on the semaphore array spinlock will limit the scaling.
50 * - semncnt and semzcnt are calculated on demand in count_semncnt() and
51 * count_semzcnt()
52 * - the task that performs a successful semop() scans the list of all
53 * sleeping tasks and completes any pending operations that can be fulfilled.
54 * Semaphores are actively given to waiting tasks (necessary for FIFO).
55 * (see update_queue())
56 * - To improve the scalability, the actual wake-up calls are performed after
57 * dropping all locks. (see wake_up_sem_queue_prepare(),
58 * wake_up_sem_queue_do())
59 * - All work is done by the waker, the woken up task does not have to do
60 * anything - not even acquiring a lock or dropping a refcount.
61 * - A woken up task may not even touch the semaphore array anymore, it may
62 * have been destroyed already by a semctl(RMID).
63 * - The synchronizations between wake-ups due to a timeout/signal and a
64 * wake-up due to a completed semaphore operation is achieved by using an
65 * intermediate state (IN_WAKEUP).
66 * - UNDO values are stored in an array (one per process and per
67 * semaphore array, lazily allocated). For backwards compatibility, multiple
68 * modes for the UNDO variables are supported (per process, per thread)
69 * (see copy_semundo, CLONE_SYSVSEM)
70 * - There are two lists of the pending operations: a per-array list
71 * and per-semaphore list (stored in the array). This allows to achieve FIFO
72 * ordering without always scanning all pending operations.
73 * The worst-case behavior is nevertheless O(N^2) for N wakeups.
1da177e4
LT
74 */
75
1da177e4
LT
76#include <linux/slab.h>
77#include <linux/spinlock.h>
78#include <linux/init.h>
79#include <linux/proc_fs.h>
80#include <linux/time.h>
1da177e4
LT
81#include <linux/security.h>
82#include <linux/syscalls.h>
83#include <linux/audit.h>
c59ede7b 84#include <linux/capability.h>
19b4946c 85#include <linux/seq_file.h>
3e148c79 86#include <linux/rwsem.h>
e3893534 87#include <linux/nsproxy.h>
ae5e1b22 88#include <linux/ipc_namespace.h>
5f921ae9 89
1da177e4
LT
90#include <asm/uaccess.h>
91#include "util.h"
92
ed2ddbf8 93#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
e3893534 94
e3893534 95#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
1b531f21 96#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
1da177e4 97
7748dbfa 98static int newary(struct ipc_namespace *, struct ipc_params *);
01b8b07a 99static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
1da177e4 100#ifdef CONFIG_PROC_FS
19b4946c 101static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
1da177e4
LT
102#endif
103
104#define SEMMSL_FAST 256 /* 512 bytes on stack */
105#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
106
107/*
108 * linked list protection:
109 * sem_undo.id_next,
110 * sem_array.sem_pending{,last},
111 * sem_array.sem_undo: sem_lock() for read/write
112 * sem_undo.proc_next: only "current" is allowed to read/write that field.
113 *
114 */
115
e3893534
KK
116#define sc_semmsl sem_ctls[0]
117#define sc_semmns sem_ctls[1]
118#define sc_semopm sem_ctls[2]
119#define sc_semmni sem_ctls[3]
120
ed2ddbf8 121void sem_init_ns(struct ipc_namespace *ns)
e3893534 122{
e3893534
KK
123 ns->sc_semmsl = SEMMSL;
124 ns->sc_semmns = SEMMNS;
125 ns->sc_semopm = SEMOPM;
126 ns->sc_semmni = SEMMNI;
127 ns->used_sems = 0;
ed2ddbf8 128 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
e3893534
KK
129}
130
ae5e1b22 131#ifdef CONFIG_IPC_NS
e3893534
KK
132void sem_exit_ns(struct ipc_namespace *ns)
133{
01b8b07a 134 free_ipcs(ns, &sem_ids(ns), freeary);
7d6feeb2 135 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr);
e3893534 136}
ae5e1b22 137#endif
1da177e4
LT
138
139void __init sem_init (void)
140{
ed2ddbf8 141 sem_init_ns(&init_ipc_ns);
19b4946c
MW
142 ipc_init_proc_interface("sysvipc/sem",
143 " key semid perms nsems uid gid cuid cgid otime ctime\n",
e3893534 144 IPC_SEM_IDS, sysvipc_sem_proc_show);
1da177e4
LT
145}
146
3e148c79
ND
147/*
148 * sem_lock_(check_) routines are called in the paths where the rw_mutex
149 * is not held.
150 */
023a5355
ND
151static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
152{
03f02c76
ND
153 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
154
b1ed88b4
PP
155 if (IS_ERR(ipcp))
156 return (struct sem_array *)ipcp;
157
03f02c76 158 return container_of(ipcp, struct sem_array, sem_perm);
023a5355
ND
159}
160
161static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
162 int id)
163{
03f02c76
ND
164 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
165
b1ed88b4
PP
166 if (IS_ERR(ipcp))
167 return (struct sem_array *)ipcp;
168
03f02c76 169 return container_of(ipcp, struct sem_array, sem_perm);
023a5355
ND
170}
171
6ff37972
PP
172static inline void sem_lock_and_putref(struct sem_array *sma)
173{
174 ipc_lock_by_ptr(&sma->sem_perm);
175 ipc_rcu_putref(sma);
176}
177
178static inline void sem_getref_and_unlock(struct sem_array *sma)
179{
180 ipc_rcu_getref(sma);
181 ipc_unlock(&(sma)->sem_perm);
182}
183
184static inline void sem_putref(struct sem_array *sma)
185{
186 ipc_lock_by_ptr(&sma->sem_perm);
187 ipc_rcu_putref(sma);
188 ipc_unlock(&(sma)->sem_perm);
189}
190
7ca7e564
ND
191static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
192{
193 ipc_rmid(&sem_ids(ns), &s->sem_perm);
194}
195
1da177e4
LT
196/*
197 * Lockless wakeup algorithm:
198 * Without the check/retry algorithm a lockless wakeup is possible:
199 * - queue.status is initialized to -EINTR before blocking.
200 * - wakeup is performed by
201 * * unlinking the queue entry from sma->sem_pending
202 * * setting queue.status to IN_WAKEUP
203 * This is the notification for the blocked thread that a
204 * result value is imminent.
205 * * call wake_up_process
206 * * set queue.status to the final value.
207 * - the previously blocked thread checks queue.status:
208 * * if it's IN_WAKEUP, then it must wait until the value changes
209 * * if it's not -EINTR, then the operation was completed by
210 * update_queue. semtimedop can return queue.status without
5f921ae9 211 * performing any operation on the sem array.
1da177e4
LT
212 * * otherwise it must acquire the spinlock and check what's up.
213 *
214 * The two-stage algorithm is necessary to protect against the following
215 * races:
216 * - if queue.status is set after wake_up_process, then the woken up idle
217 * thread could race forward and try (and fail) to acquire sma->lock
218 * before update_queue had a chance to set queue.status
219 * - if queue.status is written before wake_up_process and if the
220 * blocked process is woken up by a signal between writing
221 * queue.status and the wake_up_process, then the woken up
222 * process could return from semtimedop and die by calling
223 * sys_exit before wake_up_process is called. Then wake_up_process
224 * will oops, because the task structure is already invalid.
225 * (yes, this happened on s390 with sysv msg).
226 *
227 */
228#define IN_WAKEUP 1
229
f4566f04
ND
230/**
231 * newary - Create a new semaphore set
232 * @ns: namespace
233 * @params: ptr to the structure that contains key, semflg and nsems
234 *
3e148c79 235 * Called with sem_ids.rw_mutex held (as a writer)
f4566f04
ND
236 */
237
7748dbfa 238static int newary(struct ipc_namespace *ns, struct ipc_params *params)
1da177e4
LT
239{
240 int id;
241 int retval;
242 struct sem_array *sma;
243 int size;
7748dbfa
ND
244 key_t key = params->key;
245 int nsems = params->u.nsems;
246 int semflg = params->flg;
b97e820f 247 int i;
1da177e4
LT
248
249 if (!nsems)
250 return -EINVAL;
e3893534 251 if (ns->used_sems + nsems > ns->sc_semmns)
1da177e4
LT
252 return -ENOSPC;
253
254 size = sizeof (*sma) + nsems * sizeof (struct sem);
255 sma = ipc_rcu_alloc(size);
256 if (!sma) {
257 return -ENOMEM;
258 }
259 memset (sma, 0, size);
260
261 sma->sem_perm.mode = (semflg & S_IRWXUGO);
262 sma->sem_perm.key = key;
263
264 sma->sem_perm.security = NULL;
265 retval = security_sem_alloc(sma);
266 if (retval) {
267 ipc_rcu_putref(sma);
268 return retval;
269 }
270
e3893534 271 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
283bb7fa 272 if (id < 0) {
1da177e4
LT
273 security_sem_free(sma);
274 ipc_rcu_putref(sma);
283bb7fa 275 return id;
1da177e4 276 }
e3893534 277 ns->used_sems += nsems;
1da177e4
LT
278
279 sma->sem_base = (struct sem *) &sma[1];
b97e820f
MS
280
281 for (i = 0; i < nsems; i++)
282 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending);
283
284 sma->complex_count = 0;
a1193f8e 285 INIT_LIST_HEAD(&sma->sem_pending);
4daa28f6 286 INIT_LIST_HEAD(&sma->list_id);
1da177e4
LT
287 sma->sem_nsems = nsems;
288 sma->sem_ctime = get_seconds();
289 sem_unlock(sma);
290
7ca7e564 291 return sma->sem_perm.id;
1da177e4
LT
292}
293
7748dbfa 294
f4566f04 295/*
3e148c79 296 * Called with sem_ids.rw_mutex and ipcp locked.
f4566f04 297 */
03f02c76 298static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
7748dbfa 299{
03f02c76
ND
300 struct sem_array *sma;
301
302 sma = container_of(ipcp, struct sem_array, sem_perm);
303 return security_sem_associate(sma, semflg);
7748dbfa
ND
304}
305
f4566f04 306/*
3e148c79 307 * Called with sem_ids.rw_mutex and ipcp locked.
f4566f04 308 */
03f02c76
ND
309static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
310 struct ipc_params *params)
7748dbfa 311{
03f02c76
ND
312 struct sem_array *sma;
313
314 sma = container_of(ipcp, struct sem_array, sem_perm);
315 if (params->u.nsems > sma->sem_nsems)
7748dbfa
ND
316 return -EINVAL;
317
318 return 0;
319}
320
d5460c99 321SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg)
1da177e4 322{
e3893534 323 struct ipc_namespace *ns;
7748dbfa
ND
324 struct ipc_ops sem_ops;
325 struct ipc_params sem_params;
e3893534
KK
326
327 ns = current->nsproxy->ipc_ns;
1da177e4 328
e3893534 329 if (nsems < 0 || nsems > ns->sc_semmsl)
1da177e4 330 return -EINVAL;
7ca7e564 331
7748dbfa
ND
332 sem_ops.getnew = newary;
333 sem_ops.associate = sem_security;
334 sem_ops.more_checks = sem_more_checks;
335
336 sem_params.key = key;
337 sem_params.flg = semflg;
338 sem_params.u.nsems = nsems;
1da177e4 339
7748dbfa 340 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
1da177e4
LT
341}
342
1da177e4
LT
343/*
344 * Determine whether a sequence of semaphore operations would succeed
345 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
346 */
347
348static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
349 int nsops, struct sem_undo *un, int pid)
350{
351 int result, sem_op;
352 struct sembuf *sop;
353 struct sem * curr;
354
355 for (sop = sops; sop < sops + nsops; sop++) {
356 curr = sma->sem_base + sop->sem_num;
357 sem_op = sop->sem_op;
358 result = curr->semval;
359
360 if (!sem_op && result)
361 goto would_block;
362
363 result += sem_op;
364 if (result < 0)
365 goto would_block;
366 if (result > SEMVMX)
367 goto out_of_range;
368 if (sop->sem_flg & SEM_UNDO) {
369 int undo = un->semadj[sop->sem_num] - sem_op;
370 /*
371 * Exceeding the undo range is an error.
372 */
373 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
374 goto out_of_range;
375 }
376 curr->semval = result;
377 }
378
379 sop--;
380 while (sop >= sops) {
381 sma->sem_base[sop->sem_num].sempid = pid;
382 if (sop->sem_flg & SEM_UNDO)
383 un->semadj[sop->sem_num] -= sop->sem_op;
384 sop--;
385 }
386
1da177e4
LT
387 return 0;
388
389out_of_range:
390 result = -ERANGE;
391 goto undo;
392
393would_block:
394 if (sop->sem_flg & IPC_NOWAIT)
395 result = -EAGAIN;
396 else
397 result = 1;
398
399undo:
400 sop--;
401 while (sop >= sops) {
402 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
403 sop--;
404 }
405
406 return result;
407}
408
0a2b9d4c
MS
409/** wake_up_sem_queue_prepare(q, error): Prepare wake-up
410 * @q: queue entry that must be signaled
411 * @error: Error value for the signal
412 *
413 * Prepare the wake-up of the queue entry q.
d4212093 414 */
0a2b9d4c
MS
415static void wake_up_sem_queue_prepare(struct list_head *pt,
416 struct sem_queue *q, int error)
d4212093 417{
0a2b9d4c
MS
418 if (list_empty(pt)) {
419 /*
420 * Hold preempt off so that we don't get preempted and have the
421 * wakee busy-wait until we're scheduled back on.
422 */
423 preempt_disable();
424 }
d4212093 425 q->status = IN_WAKEUP;
0a2b9d4c
MS
426 q->pid = error;
427
428 list_add_tail(&q->simple_list, pt);
429}
430
431/**
432 * wake_up_sem_queue_do(pt) - do the actual wake-up
433 * @pt: list of tasks to be woken up
434 *
435 * Do the actual wake-up.
436 * The function is called without any locks held, thus the semaphore array
437 * could be destroyed already and the tasks can disappear as soon as the
438 * status is set to the actual return code.
439 */
440static void wake_up_sem_queue_do(struct list_head *pt)
441{
442 struct sem_queue *q, *t;
443 int did_something;
444
445 did_something = !list_empty(pt);
446 list_for_each_entry_safe(q, t, pt, simple_list) {
447 wake_up_process(q->sleeper);
448 /* q can disappear immediately after writing q->status. */
449 smp_wmb();
450 q->status = q->pid;
451 }
452 if (did_something)
453 preempt_enable();
d4212093
NP
454}
455
b97e820f
MS
456static void unlink_queue(struct sem_array *sma, struct sem_queue *q)
457{
458 list_del(&q->list);
459 if (q->nsops == 1)
460 list_del(&q->simple_list);
461 else
462 sma->complex_count--;
463}
464
fd5db422
MS
465/** check_restart(sma, q)
466 * @sma: semaphore array
467 * @q: the operation that just completed
468 *
469 * update_queue is O(N^2) when it restarts scanning the whole queue of
470 * waiting operations. Therefore this function checks if the restart is
471 * really necessary. It is called after a previously waiting operation
472 * was completed.
473 */
474static int check_restart(struct sem_array *sma, struct sem_queue *q)
475{
476 struct sem *curr;
477 struct sem_queue *h;
478
479 /* if the operation didn't modify the array, then no restart */
480 if (q->alter == 0)
481 return 0;
482
483 /* pending complex operations are too difficult to analyse */
484 if (sma->complex_count)
485 return 1;
486
487 /* we were a sleeping complex operation. Too difficult */
488 if (q->nsops > 1)
489 return 1;
490
491 curr = sma->sem_base + q->sops[0].sem_num;
492
493 /* No-one waits on this queue */
494 if (list_empty(&curr->sem_pending))
495 return 0;
496
497 /* the new semaphore value */
498 if (curr->semval) {
499 /* It is impossible that someone waits for the new value:
500 * - q is a previously sleeping simple operation that
501 * altered the array. It must be a decrement, because
502 * simple increments never sleep.
503 * - The value is not 0, thus wait-for-zero won't proceed.
504 * - If there are older (higher priority) decrements
505 * in the queue, then they have observed the original
506 * semval value and couldn't proceed. The operation
507 * decremented to value - thus they won't proceed either.
508 */
509 BUG_ON(q->sops[0].sem_op >= 0);
510 return 0;
511 }
512 /*
513 * semval is 0. Check if there are wait-for-zero semops.
514 * They must be the first entries in the per-semaphore simple queue
515 */
516 h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list);
517 BUG_ON(h->nsops != 1);
518 BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num);
519
520 /* Yes, there is a wait-for-zero semop. Restart */
521 if (h->sops[0].sem_op == 0)
522 return 1;
523
524 /* Again - no-one is waiting for the new value. */
525 return 0;
526}
527
636c6be8
MS
528
529/**
530 * update_queue(sma, semnum): Look for tasks that can be completed.
531 * @sma: semaphore array.
532 * @semnum: semaphore that was modified.
0a2b9d4c 533 * @pt: list head for the tasks that must be woken up.
636c6be8
MS
534 *
535 * update_queue must be called after a semaphore in a semaphore array
536 * was modified. If multiple semaphore were modified, then @semnum
537 * must be set to -1.
0a2b9d4c
MS
538 * The tasks that must be woken up are added to @pt. The return code
539 * is stored in q->pid.
540 * The function return 1 if at least one semop was completed successfully.
1da177e4 541 */
0a2b9d4c 542static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt)
1da177e4 543{
636c6be8
MS
544 struct sem_queue *q;
545 struct list_head *walk;
546 struct list_head *pending_list;
547 int offset;
0a2b9d4c 548 int semop_completed = 0;
636c6be8
MS
549
550 /* if there are complex operations around, then knowing the semaphore
551 * that was modified doesn't help us. Assume that multiple semaphores
552 * were modified.
553 */
554 if (sma->complex_count)
555 semnum = -1;
556
557 if (semnum == -1) {
558 pending_list = &sma->sem_pending;
559 offset = offsetof(struct sem_queue, list);
560 } else {
561 pending_list = &sma->sem_base[semnum].sem_pending;
562 offset = offsetof(struct sem_queue, simple_list);
563 }
9cad200c
NP
564
565again:
636c6be8
MS
566 walk = pending_list->next;
567 while (walk != pending_list) {
fd5db422 568 int error, restart;
636c6be8
MS
569
570 q = (struct sem_queue *)((char *)walk - offset);
571 walk = walk->next;
1da177e4 572
d987f8b2
MS
573 /* If we are scanning the single sop, per-semaphore list of
574 * one semaphore and that semaphore is 0, then it is not
575 * necessary to scan the "alter" entries: simple increments
576 * that affect only one entry succeed immediately and cannot
577 * be in the per semaphore pending queue, and decrements
578 * cannot be successful if the value is already 0.
579 */
580 if (semnum != -1 && sma->sem_base[semnum].semval == 0 &&
581 q->alter)
582 break;
583
1da177e4
LT
584 error = try_atomic_semop(sma, q->sops, q->nsops,
585 q->undo, q->pid);
586
587 /* Does q->sleeper still need to sleep? */
9cad200c
NP
588 if (error > 0)
589 continue;
590
b97e820f 591 unlink_queue(sma, q);
9cad200c 592
0a2b9d4c 593 if (error) {
fd5db422 594 restart = 0;
0a2b9d4c
MS
595 } else {
596 semop_completed = 1;
fd5db422 597 restart = check_restart(sma, q);
0a2b9d4c 598 }
fd5db422 599
0a2b9d4c 600 wake_up_sem_queue_prepare(pt, q, error);
fd5db422 601 if (restart)
9cad200c 602 goto again;
1da177e4 603 }
0a2b9d4c 604 return semop_completed;
1da177e4
LT
605}
606
0a2b9d4c
MS
607/**
608 * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue
fd5db422
MS
609 * @sma: semaphore array
610 * @sops: operations that were performed
611 * @nsops: number of operations
0a2b9d4c
MS
612 * @otime: force setting otime
613 * @pt: list head of the tasks that must be woken up.
fd5db422
MS
614 *
615 * do_smart_update() does the required called to update_queue, based on the
616 * actual changes that were performed on the semaphore array.
0a2b9d4c
MS
617 * Note that the function does not do the actual wake-up: the caller is
618 * responsible for calling wake_up_sem_queue_do(@pt).
619 * It is safe to perform this call after dropping all locks.
fd5db422 620 */
0a2b9d4c
MS
621static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops,
622 int otime, struct list_head *pt)
fd5db422
MS
623{
624 int i;
625
626 if (sma->complex_count || sops == NULL) {
0a2b9d4c
MS
627 if (update_queue(sma, -1, pt))
628 otime = 1;
629 goto done;
fd5db422
MS
630 }
631
632 for (i = 0; i < nsops; i++) {
633 if (sops[i].sem_op > 0 ||
634 (sops[i].sem_op < 0 &&
635 sma->sem_base[sops[i].sem_num].semval == 0))
0a2b9d4c
MS
636 if (update_queue(sma, sops[i].sem_num, pt))
637 otime = 1;
fd5db422 638 }
0a2b9d4c
MS
639done:
640 if (otime)
641 sma->sem_otime = get_seconds();
fd5db422
MS
642}
643
644
1da177e4
LT
645/* The following counts are associated to each semaphore:
646 * semncnt number of tasks waiting on semval being nonzero
647 * semzcnt number of tasks waiting on semval being zero
648 * This model assumes that a task waits on exactly one semaphore.
649 * Since semaphore operations are to be performed atomically, tasks actually
650 * wait on a whole sequence of semaphores simultaneously.
651 * The counts we return here are a rough approximation, but still
652 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
653 */
654static int count_semncnt (struct sem_array * sma, ushort semnum)
655{
656 int semncnt;
657 struct sem_queue * q;
658
659 semncnt = 0;
a1193f8e 660 list_for_each_entry(q, &sma->sem_pending, list) {
1da177e4
LT
661 struct sembuf * sops = q->sops;
662 int nsops = q->nsops;
663 int i;
664 for (i = 0; i < nsops; i++)
665 if (sops[i].sem_num == semnum
666 && (sops[i].sem_op < 0)
667 && !(sops[i].sem_flg & IPC_NOWAIT))
668 semncnt++;
669 }
670 return semncnt;
671}
a1193f8e 672
1da177e4
LT
673static int count_semzcnt (struct sem_array * sma, ushort semnum)
674{
675 int semzcnt;
676 struct sem_queue * q;
677
678 semzcnt = 0;
a1193f8e 679 list_for_each_entry(q, &sma->sem_pending, list) {
1da177e4
LT
680 struct sembuf * sops = q->sops;
681 int nsops = q->nsops;
682 int i;
683 for (i = 0; i < nsops; i++)
684 if (sops[i].sem_num == semnum
685 && (sops[i].sem_op == 0)
686 && !(sops[i].sem_flg & IPC_NOWAIT))
687 semzcnt++;
688 }
689 return semzcnt;
690}
691
6d97e234 692static void free_un(struct rcu_head *head)
380af1b3
MS
693{
694 struct sem_undo *un = container_of(head, struct sem_undo, rcu);
695 kfree(un);
696}
697
3e148c79
ND
698/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
699 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
700 * remains locked on exit.
1da177e4 701 */
01b8b07a 702static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
1da177e4 703{
380af1b3
MS
704 struct sem_undo *un, *tu;
705 struct sem_queue *q, *tq;
01b8b07a 706 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
0a2b9d4c 707 struct list_head tasks;
1da177e4 708
380af1b3 709 /* Free the existing undo structures for this semaphore set. */
4daa28f6 710 assert_spin_locked(&sma->sem_perm.lock);
380af1b3
MS
711 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) {
712 list_del(&un->list_id);
713 spin_lock(&un->ulp->lock);
1da177e4 714 un->semid = -1;
380af1b3
MS
715 list_del_rcu(&un->list_proc);
716 spin_unlock(&un->ulp->lock);
717 call_rcu(&un->rcu, free_un);
718 }
1da177e4
LT
719
720 /* Wake up all pending processes and let them fail with EIDRM. */
0a2b9d4c 721 INIT_LIST_HEAD(&tasks);
380af1b3 722 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) {
b97e820f 723 unlink_queue(sma, q);
0a2b9d4c 724 wake_up_sem_queue_prepare(&tasks, q, -EIDRM);
1da177e4
LT
725 }
726
7ca7e564
ND
727 /* Remove the semaphore set from the IDR */
728 sem_rmid(ns, sma);
1da177e4
LT
729 sem_unlock(sma);
730
0a2b9d4c 731 wake_up_sem_queue_do(&tasks);
e3893534 732 ns->used_sems -= sma->sem_nsems;
1da177e4
LT
733 security_sem_free(sma);
734 ipc_rcu_putref(sma);
735}
736
737static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
738{
739 switch(version) {
740 case IPC_64:
741 return copy_to_user(buf, in, sizeof(*in));
742 case IPC_OLD:
743 {
744 struct semid_ds out;
745
746 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
747
748 out.sem_otime = in->sem_otime;
749 out.sem_ctime = in->sem_ctime;
750 out.sem_nsems = in->sem_nsems;
751
752 return copy_to_user(buf, &out, sizeof(out));
753 }
754 default:
755 return -EINVAL;
756 }
757}
758
4b9fcb0e
PP
759static int semctl_nolock(struct ipc_namespace *ns, int semid,
760 int cmd, int version, union semun arg)
1da177e4 761{
e5cc9c7b 762 int err;
1da177e4
LT
763 struct sem_array *sma;
764
765 switch(cmd) {
766 case IPC_INFO:
767 case SEM_INFO:
768 {
769 struct seminfo seminfo;
770 int max_id;
771
772 err = security_sem_semctl(NULL, cmd);
773 if (err)
774 return err;
775
776 memset(&seminfo,0,sizeof(seminfo));
e3893534
KK
777 seminfo.semmni = ns->sc_semmni;
778 seminfo.semmns = ns->sc_semmns;
779 seminfo.semmsl = ns->sc_semmsl;
780 seminfo.semopm = ns->sc_semopm;
1da177e4
LT
781 seminfo.semvmx = SEMVMX;
782 seminfo.semmnu = SEMMNU;
783 seminfo.semmap = SEMMAP;
784 seminfo.semume = SEMUME;
3e148c79 785 down_read(&sem_ids(ns).rw_mutex);
1da177e4 786 if (cmd == SEM_INFO) {
e3893534
KK
787 seminfo.semusz = sem_ids(ns).in_use;
788 seminfo.semaem = ns->used_sems;
1da177e4
LT
789 } else {
790 seminfo.semusz = SEMUSZ;
791 seminfo.semaem = SEMAEM;
792 }
7ca7e564 793 max_id = ipc_get_maxid(&sem_ids(ns));
3e148c79 794 up_read(&sem_ids(ns).rw_mutex);
1da177e4
LT
795 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
796 return -EFAULT;
797 return (max_id < 0) ? 0: max_id;
798 }
4b9fcb0e 799 case IPC_STAT:
1da177e4
LT
800 case SEM_STAT:
801 {
802 struct semid64_ds tbuf;
803 int id;
804
4b9fcb0e
PP
805 if (cmd == SEM_STAT) {
806 sma = sem_lock(ns, semid);
807 if (IS_ERR(sma))
808 return PTR_ERR(sma);
809 id = sma->sem_perm.id;
810 } else {
811 sma = sem_lock_check(ns, semid);
812 if (IS_ERR(sma))
813 return PTR_ERR(sma);
814 id = 0;
815 }
1da177e4
LT
816
817 err = -EACCES;
818 if (ipcperms (&sma->sem_perm, S_IRUGO))
819 goto out_unlock;
820
821 err = security_sem_semctl(sma, cmd);
822 if (err)
823 goto out_unlock;
824
023a5355
ND
825 memset(&tbuf, 0, sizeof(tbuf));
826
1da177e4
LT
827 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
828 tbuf.sem_otime = sma->sem_otime;
829 tbuf.sem_ctime = sma->sem_ctime;
830 tbuf.sem_nsems = sma->sem_nsems;
831 sem_unlock(sma);
832 if (copy_semid_to_user (arg.buf, &tbuf, version))
833 return -EFAULT;
834 return id;
835 }
836 default:
837 return -EINVAL;
838 }
1da177e4
LT
839out_unlock:
840 sem_unlock(sma);
841 return err;
842}
843
e3893534
KK
844static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
845 int cmd, int version, union semun arg)
1da177e4
LT
846{
847 struct sem_array *sma;
848 struct sem* curr;
849 int err;
850 ushort fast_sem_io[SEMMSL_FAST];
851 ushort* sem_io = fast_sem_io;
852 int nsems;
0a2b9d4c 853 struct list_head tasks;
1da177e4 854
023a5355
ND
855 sma = sem_lock_check(ns, semid);
856 if (IS_ERR(sma))
857 return PTR_ERR(sma);
1da177e4 858
0a2b9d4c 859 INIT_LIST_HEAD(&tasks);
1da177e4
LT
860 nsems = sma->sem_nsems;
861
1da177e4
LT
862 err = -EACCES;
863 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
864 goto out_unlock;
865
866 err = security_sem_semctl(sma, cmd);
867 if (err)
868 goto out_unlock;
869
870 err = -EACCES;
871 switch (cmd) {
872 case GETALL:
873 {
874 ushort __user *array = arg.array;
875 int i;
876
877 if(nsems > SEMMSL_FAST) {
6ff37972 878 sem_getref_and_unlock(sma);
1da177e4
LT
879
880 sem_io = ipc_alloc(sizeof(ushort)*nsems);
881 if(sem_io == NULL) {
6ff37972 882 sem_putref(sma);
1da177e4
LT
883 return -ENOMEM;
884 }
885
6ff37972 886 sem_lock_and_putref(sma);
1da177e4
LT
887 if (sma->sem_perm.deleted) {
888 sem_unlock(sma);
889 err = -EIDRM;
890 goto out_free;
891 }
892 }
893
894 for (i = 0; i < sma->sem_nsems; i++)
895 sem_io[i] = sma->sem_base[i].semval;
896 sem_unlock(sma);
897 err = 0;
898 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
899 err = -EFAULT;
900 goto out_free;
901 }
902 case SETALL:
903 {
904 int i;
905 struct sem_undo *un;
906
6ff37972 907 sem_getref_and_unlock(sma);
1da177e4
LT
908
909 if(nsems > SEMMSL_FAST) {
910 sem_io = ipc_alloc(sizeof(ushort)*nsems);
911 if(sem_io == NULL) {
6ff37972 912 sem_putref(sma);
1da177e4
LT
913 return -ENOMEM;
914 }
915 }
916
917 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
6ff37972 918 sem_putref(sma);
1da177e4
LT
919 err = -EFAULT;
920 goto out_free;
921 }
922
923 for (i = 0; i < nsems; i++) {
924 if (sem_io[i] > SEMVMX) {
6ff37972 925 sem_putref(sma);
1da177e4
LT
926 err = -ERANGE;
927 goto out_free;
928 }
929 }
6ff37972 930 sem_lock_and_putref(sma);
1da177e4
LT
931 if (sma->sem_perm.deleted) {
932 sem_unlock(sma);
933 err = -EIDRM;
934 goto out_free;
935 }
936
937 for (i = 0; i < nsems; i++)
938 sma->sem_base[i].semval = sem_io[i];
4daa28f6
MS
939
940 assert_spin_locked(&sma->sem_perm.lock);
941 list_for_each_entry(un, &sma->list_id, list_id) {
1da177e4
LT
942 for (i = 0; i < nsems; i++)
943 un->semadj[i] = 0;
4daa28f6 944 }
1da177e4
LT
945 sma->sem_ctime = get_seconds();
946 /* maybe some queued-up processes were waiting for this */
0a2b9d4c 947 do_smart_update(sma, NULL, 0, 0, &tasks);
1da177e4
LT
948 err = 0;
949 goto out_unlock;
950 }
1da177e4
LT
951 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
952 }
953 err = -EINVAL;
954 if(semnum < 0 || semnum >= nsems)
955 goto out_unlock;
956
957 curr = &sma->sem_base[semnum];
958
959 switch (cmd) {
960 case GETVAL:
961 err = curr->semval;
962 goto out_unlock;
963 case GETPID:
964 err = curr->sempid;
965 goto out_unlock;
966 case GETNCNT:
967 err = count_semncnt(sma,semnum);
968 goto out_unlock;
969 case GETZCNT:
970 err = count_semzcnt(sma,semnum);
971 goto out_unlock;
972 case SETVAL:
973 {
974 int val = arg.val;
975 struct sem_undo *un;
4daa28f6 976
1da177e4
LT
977 err = -ERANGE;
978 if (val > SEMVMX || val < 0)
979 goto out_unlock;
980
4daa28f6
MS
981 assert_spin_locked(&sma->sem_perm.lock);
982 list_for_each_entry(un, &sma->list_id, list_id)
1da177e4 983 un->semadj[semnum] = 0;
4daa28f6 984
1da177e4 985 curr->semval = val;
b488893a 986 curr->sempid = task_tgid_vnr(current);
1da177e4
LT
987 sma->sem_ctime = get_seconds();
988 /* maybe some queued-up processes were waiting for this */
0a2b9d4c 989 do_smart_update(sma, NULL, 0, 0, &tasks);
1da177e4
LT
990 err = 0;
991 goto out_unlock;
992 }
993 }
994out_unlock:
995 sem_unlock(sma);
0a2b9d4c
MS
996 wake_up_sem_queue_do(&tasks);
997
1da177e4
LT
998out_free:
999 if(sem_io != fast_sem_io)
1000 ipc_free(sem_io, sizeof(ushort)*nsems);
1001 return err;
1002}
1003
016d7132
PP
1004static inline unsigned long
1005copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1da177e4
LT
1006{
1007 switch(version) {
1008 case IPC_64:
016d7132 1009 if (copy_from_user(out, buf, sizeof(*out)))
1da177e4 1010 return -EFAULT;
1da177e4 1011 return 0;
1da177e4
LT
1012 case IPC_OLD:
1013 {
1014 struct semid_ds tbuf_old;
1015
1016 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1017 return -EFAULT;
1018
016d7132
PP
1019 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1020 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1021 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1da177e4
LT
1022
1023 return 0;
1024 }
1025 default:
1026 return -EINVAL;
1027 }
1028}
1029
522bb2a2
PP
1030/*
1031 * This function handles some semctl commands which require the rw_mutex
1032 * to be held in write mode.
1033 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1034 */
21a4826a
PP
1035static int semctl_down(struct ipc_namespace *ns, int semid,
1036 int cmd, int version, union semun arg)
1da177e4
LT
1037{
1038 struct sem_array *sma;
1039 int err;
016d7132 1040 struct semid64_ds semid64;
1da177e4
LT
1041 struct kern_ipc_perm *ipcp;
1042
1043 if(cmd == IPC_SET) {
016d7132 1044 if (copy_semid_from_user(&semid64, arg.buf, version))
1da177e4 1045 return -EFAULT;
1da177e4 1046 }
073115d6 1047
a5f75e7f
PP
1048 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
1049 if (IS_ERR(ipcp))
1050 return PTR_ERR(ipcp);
073115d6 1051
a5f75e7f 1052 sma = container_of(ipcp, struct sem_array, sem_perm);
1da177e4
LT
1053
1054 err = security_sem_semctl(sma, cmd);
1055 if (err)
1056 goto out_unlock;
1057
1058 switch(cmd){
1059 case IPC_RMID:
01b8b07a 1060 freeary(ns, ipcp);
522bb2a2 1061 goto out_up;
1da177e4 1062 case IPC_SET:
8f4a3809 1063 ipc_update_perm(&semid64.sem_perm, ipcp);
1da177e4 1064 sma->sem_ctime = get_seconds();
1da177e4
LT
1065 break;
1066 default:
1da177e4 1067 err = -EINVAL;
1da177e4 1068 }
1da177e4
LT
1069
1070out_unlock:
1071 sem_unlock(sma);
522bb2a2
PP
1072out_up:
1073 up_write(&sem_ids(ns).rw_mutex);
1da177e4
LT
1074 return err;
1075}
1076
6673e0c3 1077SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1da177e4
LT
1078{
1079 int err = -EINVAL;
1080 int version;
e3893534 1081 struct ipc_namespace *ns;
1da177e4
LT
1082
1083 if (semid < 0)
1084 return -EINVAL;
1085
1086 version = ipc_parse_version(&cmd);
e3893534 1087 ns = current->nsproxy->ipc_ns;
1da177e4
LT
1088
1089 switch(cmd) {
1090 case IPC_INFO:
1091 case SEM_INFO:
4b9fcb0e 1092 case IPC_STAT:
1da177e4 1093 case SEM_STAT:
4b9fcb0e 1094 err = semctl_nolock(ns, semid, cmd, version, arg);
1da177e4
LT
1095 return err;
1096 case GETALL:
1097 case GETVAL:
1098 case GETPID:
1099 case GETNCNT:
1100 case GETZCNT:
1da177e4
LT
1101 case SETVAL:
1102 case SETALL:
e3893534 1103 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1da177e4
LT
1104 return err;
1105 case IPC_RMID:
1106 case IPC_SET:
21a4826a 1107 err = semctl_down(ns, semid, cmd, version, arg);
1da177e4
LT
1108 return err;
1109 default:
1110 return -EINVAL;
1111 }
1112}
6673e0c3
HC
1113#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1114asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1115{
1116 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1117}
1118SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1119#endif
1da177e4 1120
1da177e4
LT
1121/* If the task doesn't already have a undo_list, then allocate one
1122 * here. We guarantee there is only one thread using this undo list,
1123 * and current is THE ONE
1124 *
1125 * If this allocation and assignment succeeds, but later
1126 * portions of this code fail, there is no need to free the sem_undo_list.
1127 * Just let it stay associated with the task, and it'll be freed later
1128 * at exit time.
1129 *
1130 * This can block, so callers must hold no locks.
1131 */
1132static inline int get_undo_list(struct sem_undo_list **undo_listp)
1133{
1134 struct sem_undo_list *undo_list;
1da177e4
LT
1135
1136 undo_list = current->sysvsem.undo_list;
1137 if (!undo_list) {
2453a306 1138 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1da177e4
LT
1139 if (undo_list == NULL)
1140 return -ENOMEM;
00a5dfdb 1141 spin_lock_init(&undo_list->lock);
1da177e4 1142 atomic_set(&undo_list->refcnt, 1);
4daa28f6
MS
1143 INIT_LIST_HEAD(&undo_list->list_proc);
1144
1da177e4
LT
1145 current->sysvsem.undo_list = undo_list;
1146 }
1147 *undo_listp = undo_list;
1148 return 0;
1149}
1150
bf17bb71 1151static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1da177e4 1152{
bf17bb71 1153 struct sem_undo *un;
4daa28f6 1154
bf17bb71
NP
1155 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1156 if (un->semid == semid)
1157 return un;
1da177e4 1158 }
4daa28f6 1159 return NULL;
1da177e4
LT
1160}
1161
bf17bb71
NP
1162static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1163{
1164 struct sem_undo *un;
1165
1166 assert_spin_locked(&ulp->lock);
1167
1168 un = __lookup_undo(ulp, semid);
1169 if (un) {
1170 list_del_rcu(&un->list_proc);
1171 list_add_rcu(&un->list_proc, &ulp->list_proc);
1172 }
1173 return un;
1174}
1175
4daa28f6
MS
1176/**
1177 * find_alloc_undo - Lookup (and if not present create) undo array
1178 * @ns: namespace
1179 * @semid: semaphore array id
1180 *
1181 * The function looks up (and if not present creates) the undo structure.
1182 * The size of the undo structure depends on the size of the semaphore
1183 * array, thus the alloc path is not that straightforward.
380af1b3
MS
1184 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1185 * performs a rcu_read_lock().
4daa28f6
MS
1186 */
1187static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1da177e4
LT
1188{
1189 struct sem_array *sma;
1190 struct sem_undo_list *ulp;
1191 struct sem_undo *un, *new;
1192 int nsems;
1193 int error;
1194
1195 error = get_undo_list(&ulp);
1196 if (error)
1197 return ERR_PTR(error);
1198
380af1b3 1199 rcu_read_lock();
c530c6ac 1200 spin_lock(&ulp->lock);
1da177e4 1201 un = lookup_undo(ulp, semid);
c530c6ac 1202 spin_unlock(&ulp->lock);
1da177e4
LT
1203 if (likely(un!=NULL))
1204 goto out;
380af1b3 1205 rcu_read_unlock();
1da177e4
LT
1206
1207 /* no undo structure around - allocate one. */
4daa28f6 1208 /* step 1: figure out the size of the semaphore array */
023a5355
ND
1209 sma = sem_lock_check(ns, semid);
1210 if (IS_ERR(sma))
4de85cd6 1211 return ERR_CAST(sma);
023a5355 1212
1da177e4 1213 nsems = sma->sem_nsems;
6ff37972 1214 sem_getref_and_unlock(sma);
1da177e4 1215
4daa28f6 1216 /* step 2: allocate new undo structure */
4668edc3 1217 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1da177e4 1218 if (!new) {
6ff37972 1219 sem_putref(sma);
1da177e4
LT
1220 return ERR_PTR(-ENOMEM);
1221 }
1da177e4 1222
380af1b3 1223 /* step 3: Acquire the lock on semaphore array */
6ff37972 1224 sem_lock_and_putref(sma);
1da177e4
LT
1225 if (sma->sem_perm.deleted) {
1226 sem_unlock(sma);
1da177e4
LT
1227 kfree(new);
1228 un = ERR_PTR(-EIDRM);
1229 goto out;
1230 }
380af1b3
MS
1231 spin_lock(&ulp->lock);
1232
1233 /*
1234 * step 4: check for races: did someone else allocate the undo struct?
1235 */
1236 un = lookup_undo(ulp, semid);
1237 if (un) {
1238 kfree(new);
1239 goto success;
1240 }
4daa28f6
MS
1241 /* step 5: initialize & link new undo structure */
1242 new->semadj = (short *) &new[1];
380af1b3 1243 new->ulp = ulp;
4daa28f6
MS
1244 new->semid = semid;
1245 assert_spin_locked(&ulp->lock);
380af1b3 1246 list_add_rcu(&new->list_proc, &ulp->list_proc);
4daa28f6
MS
1247 assert_spin_locked(&sma->sem_perm.lock);
1248 list_add(&new->list_id, &sma->list_id);
380af1b3 1249 un = new;
4daa28f6 1250
380af1b3 1251success:
c530c6ac 1252 spin_unlock(&ulp->lock);
380af1b3
MS
1253 rcu_read_lock();
1254 sem_unlock(sma);
1da177e4
LT
1255out:
1256 return un;
1257}
1258
c61284e9
MS
1259
1260/**
1261 * get_queue_result - Retrieve the result code from sem_queue
1262 * @q: Pointer to queue structure
1263 *
1264 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1265 * q->status, then we must loop until the value is replaced with the final
1266 * value: This may happen if a task is woken up by an unrelated event (e.g.
1267 * signal) and in parallel the task is woken up by another task because it got
1268 * the requested semaphores.
1269 *
1270 * The function can be called with or without holding the semaphore spinlock.
1271 */
1272static int get_queue_result(struct sem_queue *q)
1273{
1274 int error;
1275
1276 error = q->status;
1277 while (unlikely(error == IN_WAKEUP)) {
1278 cpu_relax();
1279 error = q->status;
1280 }
1281
1282 return error;
1283}
1284
1285
d5460c99
HC
1286SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1287 unsigned, nsops, const struct timespec __user *, timeout)
1da177e4
LT
1288{
1289 int error = -EINVAL;
1290 struct sem_array *sma;
1291 struct sembuf fast_sops[SEMOPM_FAST];
1292 struct sembuf* sops = fast_sops, *sop;
1293 struct sem_undo *un;
b78755ab 1294 int undos = 0, alter = 0, max;
1da177e4
LT
1295 struct sem_queue queue;
1296 unsigned long jiffies_left = 0;
e3893534 1297 struct ipc_namespace *ns;
0a2b9d4c 1298 struct list_head tasks;
e3893534
KK
1299
1300 ns = current->nsproxy->ipc_ns;
1da177e4
LT
1301
1302 if (nsops < 1 || semid < 0)
1303 return -EINVAL;
e3893534 1304 if (nsops > ns->sc_semopm)
1da177e4
LT
1305 return -E2BIG;
1306 if(nsops > SEMOPM_FAST) {
1307 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1308 if(sops==NULL)
1309 return -ENOMEM;
1310 }
1311 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1312 error=-EFAULT;
1313 goto out_free;
1314 }
1315 if (timeout) {
1316 struct timespec _timeout;
1317 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1318 error = -EFAULT;
1319 goto out_free;
1320 }
1321 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1322 _timeout.tv_nsec >= 1000000000L) {
1323 error = -EINVAL;
1324 goto out_free;
1325 }
1326 jiffies_left = timespec_to_jiffies(&_timeout);
1327 }
1328 max = 0;
1329 for (sop = sops; sop < sops + nsops; sop++) {
1330 if (sop->sem_num >= max)
1331 max = sop->sem_num;
1332 if (sop->sem_flg & SEM_UNDO)
b78755ab
MS
1333 undos = 1;
1334 if (sop->sem_op != 0)
1da177e4
LT
1335 alter = 1;
1336 }
1da177e4 1337
1da177e4 1338 if (undos) {
4daa28f6 1339 un = find_alloc_undo(ns, semid);
1da177e4
LT
1340 if (IS_ERR(un)) {
1341 error = PTR_ERR(un);
1342 goto out_free;
1343 }
1344 } else
1345 un = NULL;
1346
0a2b9d4c
MS
1347 INIT_LIST_HEAD(&tasks);
1348
023a5355
ND
1349 sma = sem_lock_check(ns, semid);
1350 if (IS_ERR(sma)) {
380af1b3
MS
1351 if (un)
1352 rcu_read_unlock();
023a5355 1353 error = PTR_ERR(sma);
1da177e4 1354 goto out_free;
023a5355
ND
1355 }
1356
1da177e4 1357 /*
4daa28f6 1358 * semid identifiers are not unique - find_alloc_undo may have
1da177e4 1359 * allocated an undo structure, it was invalidated by an RMID
4daa28f6 1360 * and now a new array with received the same id. Check and fail.
380af1b3
MS
1361 * This case can be detected checking un->semid. The existance of
1362 * "un" itself is guaranteed by rcu.
1da177e4 1363 */
4daa28f6 1364 error = -EIDRM;
380af1b3
MS
1365 if (un) {
1366 if (un->semid == -1) {
1367 rcu_read_unlock();
1368 goto out_unlock_free;
1369 } else {
1370 /*
1371 * rcu lock can be released, "un" cannot disappear:
1372 * - sem_lock is acquired, thus IPC_RMID is
1373 * impossible.
1374 * - exit_sem is impossible, it always operates on
1375 * current (or a dead task).
1376 */
1377
1378 rcu_read_unlock();
1379 }
1380 }
4daa28f6 1381
1da177e4
LT
1382 error = -EFBIG;
1383 if (max >= sma->sem_nsems)
1384 goto out_unlock_free;
1385
1386 error = -EACCES;
1387 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1388 goto out_unlock_free;
1389
1390 error = security_sem_semop(sma, sops, nsops, alter);
1391 if (error)
1392 goto out_unlock_free;
1393
b488893a 1394 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1da177e4
LT
1395 if (error <= 0) {
1396 if (alter && error == 0)
0a2b9d4c 1397 do_smart_update(sma, sops, nsops, 1, &tasks);
636c6be8 1398
1da177e4
LT
1399 goto out_unlock_free;
1400 }
1401
1402 /* We need to sleep on this operation, so we put the current
1403 * task into the pending queue and go to sleep.
1404 */
1405
1da177e4
LT
1406 queue.sops = sops;
1407 queue.nsops = nsops;
1408 queue.undo = un;
b488893a 1409 queue.pid = task_tgid_vnr(current);
1da177e4
LT
1410 queue.alter = alter;
1411 if (alter)
a1193f8e 1412 list_add_tail(&queue.list, &sma->sem_pending);
1da177e4 1413 else
a1193f8e 1414 list_add(&queue.list, &sma->sem_pending);
1da177e4 1415
b97e820f
MS
1416 if (nsops == 1) {
1417 struct sem *curr;
1418 curr = &sma->sem_base[sops->sem_num];
1419
1420 if (alter)
1421 list_add_tail(&queue.simple_list, &curr->sem_pending);
1422 else
1423 list_add(&queue.simple_list, &curr->sem_pending);
1424 } else {
1425 INIT_LIST_HEAD(&queue.simple_list);
1426 sma->complex_count++;
1427 }
1428
1da177e4
LT
1429 queue.status = -EINTR;
1430 queue.sleeper = current;
1431 current->state = TASK_INTERRUPTIBLE;
1432 sem_unlock(sma);
1433
1434 if (timeout)
1435 jiffies_left = schedule_timeout(jiffies_left);
1436 else
1437 schedule();
1438
c61284e9 1439 error = get_queue_result(&queue);
1da177e4
LT
1440
1441 if (error != -EINTR) {
1442 /* fast path: update_queue already obtained all requested
c61284e9
MS
1443 * resources.
1444 * Perform a smp_mb(): User space could assume that semop()
1445 * is a memory barrier: Without the mb(), the cpu could
1446 * speculatively read in user space stale data that was
1447 * overwritten by the previous owner of the semaphore.
1448 */
1449 smp_mb();
1450
1da177e4
LT
1451 goto out_free;
1452 }
1453
e3893534 1454 sma = sem_lock(ns, semid);
023a5355 1455 if (IS_ERR(sma)) {
1da177e4
LT
1456 error = -EIDRM;
1457 goto out_free;
1458 }
1459
c61284e9
MS
1460 error = get_queue_result(&queue);
1461
1da177e4
LT
1462 /*
1463 * If queue.status != -EINTR we are woken up by another process
1464 */
c61284e9 1465
1da177e4
LT
1466 if (error != -EINTR) {
1467 goto out_unlock_free;
1468 }
1469
1470 /*
1471 * If an interrupt occurred we have to clean up the queue
1472 */
1473 if (timeout && jiffies_left == 0)
1474 error = -EAGAIN;
b97e820f 1475 unlink_queue(sma, &queue);
1da177e4
LT
1476
1477out_unlock_free:
1478 sem_unlock(sma);
0a2b9d4c
MS
1479
1480 wake_up_sem_queue_do(&tasks);
1da177e4
LT
1481out_free:
1482 if(sops != fast_sops)
1483 kfree(sops);
1484 return error;
1485}
1486
d5460c99
HC
1487SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1488 unsigned, nsops)
1da177e4
LT
1489{
1490 return sys_semtimedop(semid, tsops, nsops, NULL);
1491}
1492
1493/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1494 * parent and child tasks.
1da177e4
LT
1495 */
1496
1497int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1498{
1499 struct sem_undo_list *undo_list;
1500 int error;
1501
1502 if (clone_flags & CLONE_SYSVSEM) {
1503 error = get_undo_list(&undo_list);
1504 if (error)
1505 return error;
1da177e4
LT
1506 atomic_inc(&undo_list->refcnt);
1507 tsk->sysvsem.undo_list = undo_list;
1508 } else
1509 tsk->sysvsem.undo_list = NULL;
1510
1511 return 0;
1512}
1513
1514/*
1515 * add semadj values to semaphores, free undo structures.
1516 * undo structures are not freed when semaphore arrays are destroyed
1517 * so some of them may be out of date.
1518 * IMPLEMENTATION NOTE: There is some confusion over whether the
1519 * set of adjustments that needs to be done should be done in an atomic
1520 * manner or not. That is, if we are attempting to decrement the semval
1521 * should we queue up and wait until we can do so legally?
1522 * The original implementation attempted to do this (queue and wait).
1523 * The current implementation does not do so. The POSIX standard
1524 * and SVID should be consulted to determine what behavior is mandated.
1525 */
1526void exit_sem(struct task_struct *tsk)
1527{
4daa28f6 1528 struct sem_undo_list *ulp;
1da177e4 1529
4daa28f6
MS
1530 ulp = tsk->sysvsem.undo_list;
1531 if (!ulp)
1da177e4 1532 return;
9edff4ab 1533 tsk->sysvsem.undo_list = NULL;
1da177e4 1534
4daa28f6 1535 if (!atomic_dec_and_test(&ulp->refcnt))
1da177e4
LT
1536 return;
1537
380af1b3 1538 for (;;) {
1da177e4 1539 struct sem_array *sma;
380af1b3 1540 struct sem_undo *un;
0a2b9d4c 1541 struct list_head tasks;
380af1b3 1542 int semid;
4daa28f6
MS
1543 int i;
1544
380af1b3 1545 rcu_read_lock();
05725f7e
JP
1546 un = list_entry_rcu(ulp->list_proc.next,
1547 struct sem_undo, list_proc);
380af1b3
MS
1548 if (&un->list_proc == &ulp->list_proc)
1549 semid = -1;
1550 else
1551 semid = un->semid;
1552 rcu_read_unlock();
4daa28f6 1553
380af1b3
MS
1554 if (semid == -1)
1555 break;
1da177e4 1556
380af1b3 1557 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1da177e4 1558
380af1b3
MS
1559 /* exit_sem raced with IPC_RMID, nothing to do */
1560 if (IS_ERR(sma))
1561 continue;
1da177e4 1562
bf17bb71 1563 un = __lookup_undo(ulp, semid);
380af1b3
MS
1564 if (un == NULL) {
1565 /* exit_sem raced with IPC_RMID+semget() that created
1566 * exactly the same semid. Nothing to do.
1567 */
1568 sem_unlock(sma);
1569 continue;
1570 }
1571
1572 /* remove un from the linked lists */
4daa28f6
MS
1573 assert_spin_locked(&sma->sem_perm.lock);
1574 list_del(&un->list_id);
1575
380af1b3
MS
1576 spin_lock(&ulp->lock);
1577 list_del_rcu(&un->list_proc);
1578 spin_unlock(&ulp->lock);
1579
4daa28f6
MS
1580 /* perform adjustments registered in un */
1581 for (i = 0; i < sma->sem_nsems; i++) {
5f921ae9 1582 struct sem * semaphore = &sma->sem_base[i];
4daa28f6
MS
1583 if (un->semadj[i]) {
1584 semaphore->semval += un->semadj[i];
1da177e4
LT
1585 /*
1586 * Range checks of the new semaphore value,
1587 * not defined by sus:
1588 * - Some unices ignore the undo entirely
1589 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1590 * - some cap the value (e.g. FreeBSD caps
1591 * at 0, but doesn't enforce SEMVMX)
1592 *
1593 * Linux caps the semaphore value, both at 0
1594 * and at SEMVMX.
1595 *
1596 * Manfred <manfred@colorfullife.com>
1597 */
5f921ae9
IM
1598 if (semaphore->semval < 0)
1599 semaphore->semval = 0;
1600 if (semaphore->semval > SEMVMX)
1601 semaphore->semval = SEMVMX;
b488893a 1602 semaphore->sempid = task_tgid_vnr(current);
1da177e4
LT
1603 }
1604 }
1da177e4 1605 /* maybe some queued-up processes were waiting for this */
0a2b9d4c
MS
1606 INIT_LIST_HEAD(&tasks);
1607 do_smart_update(sma, NULL, 0, 1, &tasks);
1da177e4 1608 sem_unlock(sma);
0a2b9d4c 1609 wake_up_sem_queue_do(&tasks);
380af1b3
MS
1610
1611 call_rcu(&un->rcu, free_un);
1da177e4 1612 }
4daa28f6 1613 kfree(ulp);
1da177e4
LT
1614}
1615
1616#ifdef CONFIG_PROC_FS
19b4946c 1617static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1da177e4 1618{
19b4946c
MW
1619 struct sem_array *sma = it;
1620
1621 return seq_printf(s,
b97e820f 1622 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
19b4946c 1623 sma->sem_perm.key,
7ca7e564 1624 sma->sem_perm.id,
19b4946c
MW
1625 sma->sem_perm.mode,
1626 sma->sem_nsems,
1627 sma->sem_perm.uid,
1628 sma->sem_perm.gid,
1629 sma->sem_perm.cuid,
1630 sma->sem_perm.cgid,
1631 sma->sem_otime,
1632 sma->sem_ctime);
1da177e4
LT
1633}
1634#endif