]> bbs.cooldavid.org Git - net-next-2.6.git/blame - ipc/sem.c
Net: rxrpc: Makefile: Remove deprecated kbuild goal definitions
[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
982f7c2b
DR
746 memset(&out, 0, sizeof(out));
747
1da177e4
LT
748 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
749
750 out.sem_otime = in->sem_otime;
751 out.sem_ctime = in->sem_ctime;
752 out.sem_nsems = in->sem_nsems;
753
754 return copy_to_user(buf, &out, sizeof(out));
755 }
756 default:
757 return -EINVAL;
758 }
759}
760
4b9fcb0e
PP
761static int semctl_nolock(struct ipc_namespace *ns, int semid,
762 int cmd, int version, union semun arg)
1da177e4 763{
e5cc9c7b 764 int err;
1da177e4
LT
765 struct sem_array *sma;
766
767 switch(cmd) {
768 case IPC_INFO:
769 case SEM_INFO:
770 {
771 struct seminfo seminfo;
772 int max_id;
773
774 err = security_sem_semctl(NULL, cmd);
775 if (err)
776 return err;
777
778 memset(&seminfo,0,sizeof(seminfo));
e3893534
KK
779 seminfo.semmni = ns->sc_semmni;
780 seminfo.semmns = ns->sc_semmns;
781 seminfo.semmsl = ns->sc_semmsl;
782 seminfo.semopm = ns->sc_semopm;
1da177e4
LT
783 seminfo.semvmx = SEMVMX;
784 seminfo.semmnu = SEMMNU;
785 seminfo.semmap = SEMMAP;
786 seminfo.semume = SEMUME;
3e148c79 787 down_read(&sem_ids(ns).rw_mutex);
1da177e4 788 if (cmd == SEM_INFO) {
e3893534
KK
789 seminfo.semusz = sem_ids(ns).in_use;
790 seminfo.semaem = ns->used_sems;
1da177e4
LT
791 } else {
792 seminfo.semusz = SEMUSZ;
793 seminfo.semaem = SEMAEM;
794 }
7ca7e564 795 max_id = ipc_get_maxid(&sem_ids(ns));
3e148c79 796 up_read(&sem_ids(ns).rw_mutex);
1da177e4
LT
797 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
798 return -EFAULT;
799 return (max_id < 0) ? 0: max_id;
800 }
4b9fcb0e 801 case IPC_STAT:
1da177e4
LT
802 case SEM_STAT:
803 {
804 struct semid64_ds tbuf;
805 int id;
806
4b9fcb0e
PP
807 if (cmd == SEM_STAT) {
808 sma = sem_lock(ns, semid);
809 if (IS_ERR(sma))
810 return PTR_ERR(sma);
811 id = sma->sem_perm.id;
812 } else {
813 sma = sem_lock_check(ns, semid);
814 if (IS_ERR(sma))
815 return PTR_ERR(sma);
816 id = 0;
817 }
1da177e4
LT
818
819 err = -EACCES;
820 if (ipcperms (&sma->sem_perm, S_IRUGO))
821 goto out_unlock;
822
823 err = security_sem_semctl(sma, cmd);
824 if (err)
825 goto out_unlock;
826
023a5355
ND
827 memset(&tbuf, 0, sizeof(tbuf));
828
1da177e4
LT
829 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
830 tbuf.sem_otime = sma->sem_otime;
831 tbuf.sem_ctime = sma->sem_ctime;
832 tbuf.sem_nsems = sma->sem_nsems;
833 sem_unlock(sma);
834 if (copy_semid_to_user (arg.buf, &tbuf, version))
835 return -EFAULT;
836 return id;
837 }
838 default:
839 return -EINVAL;
840 }
1da177e4
LT
841out_unlock:
842 sem_unlock(sma);
843 return err;
844}
845
e3893534
KK
846static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
847 int cmd, int version, union semun arg)
1da177e4
LT
848{
849 struct sem_array *sma;
850 struct sem* curr;
851 int err;
852 ushort fast_sem_io[SEMMSL_FAST];
853 ushort* sem_io = fast_sem_io;
854 int nsems;
0a2b9d4c 855 struct list_head tasks;
1da177e4 856
023a5355
ND
857 sma = sem_lock_check(ns, semid);
858 if (IS_ERR(sma))
859 return PTR_ERR(sma);
1da177e4 860
0a2b9d4c 861 INIT_LIST_HEAD(&tasks);
1da177e4
LT
862 nsems = sma->sem_nsems;
863
1da177e4
LT
864 err = -EACCES;
865 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
866 goto out_unlock;
867
868 err = security_sem_semctl(sma, cmd);
869 if (err)
870 goto out_unlock;
871
872 err = -EACCES;
873 switch (cmd) {
874 case GETALL:
875 {
876 ushort __user *array = arg.array;
877 int i;
878
879 if(nsems > SEMMSL_FAST) {
6ff37972 880 sem_getref_and_unlock(sma);
1da177e4
LT
881
882 sem_io = ipc_alloc(sizeof(ushort)*nsems);
883 if(sem_io == NULL) {
6ff37972 884 sem_putref(sma);
1da177e4
LT
885 return -ENOMEM;
886 }
887
6ff37972 888 sem_lock_and_putref(sma);
1da177e4
LT
889 if (sma->sem_perm.deleted) {
890 sem_unlock(sma);
891 err = -EIDRM;
892 goto out_free;
893 }
894 }
895
896 for (i = 0; i < sma->sem_nsems; i++)
897 sem_io[i] = sma->sem_base[i].semval;
898 sem_unlock(sma);
899 err = 0;
900 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
901 err = -EFAULT;
902 goto out_free;
903 }
904 case SETALL:
905 {
906 int i;
907 struct sem_undo *un;
908
6ff37972 909 sem_getref_and_unlock(sma);
1da177e4
LT
910
911 if(nsems > SEMMSL_FAST) {
912 sem_io = ipc_alloc(sizeof(ushort)*nsems);
913 if(sem_io == NULL) {
6ff37972 914 sem_putref(sma);
1da177e4
LT
915 return -ENOMEM;
916 }
917 }
918
919 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
6ff37972 920 sem_putref(sma);
1da177e4
LT
921 err = -EFAULT;
922 goto out_free;
923 }
924
925 for (i = 0; i < nsems; i++) {
926 if (sem_io[i] > SEMVMX) {
6ff37972 927 sem_putref(sma);
1da177e4
LT
928 err = -ERANGE;
929 goto out_free;
930 }
931 }
6ff37972 932 sem_lock_and_putref(sma);
1da177e4
LT
933 if (sma->sem_perm.deleted) {
934 sem_unlock(sma);
935 err = -EIDRM;
936 goto out_free;
937 }
938
939 for (i = 0; i < nsems; i++)
940 sma->sem_base[i].semval = sem_io[i];
4daa28f6
MS
941
942 assert_spin_locked(&sma->sem_perm.lock);
943 list_for_each_entry(un, &sma->list_id, list_id) {
1da177e4
LT
944 for (i = 0; i < nsems; i++)
945 un->semadj[i] = 0;
4daa28f6 946 }
1da177e4
LT
947 sma->sem_ctime = get_seconds();
948 /* maybe some queued-up processes were waiting for this */
0a2b9d4c 949 do_smart_update(sma, NULL, 0, 0, &tasks);
1da177e4
LT
950 err = 0;
951 goto out_unlock;
952 }
1da177e4
LT
953 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
954 }
955 err = -EINVAL;
956 if(semnum < 0 || semnum >= nsems)
957 goto out_unlock;
958
959 curr = &sma->sem_base[semnum];
960
961 switch (cmd) {
962 case GETVAL:
963 err = curr->semval;
964 goto out_unlock;
965 case GETPID:
966 err = curr->sempid;
967 goto out_unlock;
968 case GETNCNT:
969 err = count_semncnt(sma,semnum);
970 goto out_unlock;
971 case GETZCNT:
972 err = count_semzcnt(sma,semnum);
973 goto out_unlock;
974 case SETVAL:
975 {
976 int val = arg.val;
977 struct sem_undo *un;
4daa28f6 978
1da177e4
LT
979 err = -ERANGE;
980 if (val > SEMVMX || val < 0)
981 goto out_unlock;
982
4daa28f6
MS
983 assert_spin_locked(&sma->sem_perm.lock);
984 list_for_each_entry(un, &sma->list_id, list_id)
1da177e4 985 un->semadj[semnum] = 0;
4daa28f6 986
1da177e4 987 curr->semval = val;
b488893a 988 curr->sempid = task_tgid_vnr(current);
1da177e4
LT
989 sma->sem_ctime = get_seconds();
990 /* maybe some queued-up processes were waiting for this */
0a2b9d4c 991 do_smart_update(sma, NULL, 0, 0, &tasks);
1da177e4
LT
992 err = 0;
993 goto out_unlock;
994 }
995 }
996out_unlock:
997 sem_unlock(sma);
0a2b9d4c
MS
998 wake_up_sem_queue_do(&tasks);
999
1da177e4
LT
1000out_free:
1001 if(sem_io != fast_sem_io)
1002 ipc_free(sem_io, sizeof(ushort)*nsems);
1003 return err;
1004}
1005
016d7132
PP
1006static inline unsigned long
1007copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version)
1da177e4
LT
1008{
1009 switch(version) {
1010 case IPC_64:
016d7132 1011 if (copy_from_user(out, buf, sizeof(*out)))
1da177e4 1012 return -EFAULT;
1da177e4 1013 return 0;
1da177e4
LT
1014 case IPC_OLD:
1015 {
1016 struct semid_ds tbuf_old;
1017
1018 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
1019 return -EFAULT;
1020
016d7132
PP
1021 out->sem_perm.uid = tbuf_old.sem_perm.uid;
1022 out->sem_perm.gid = tbuf_old.sem_perm.gid;
1023 out->sem_perm.mode = tbuf_old.sem_perm.mode;
1da177e4
LT
1024
1025 return 0;
1026 }
1027 default:
1028 return -EINVAL;
1029 }
1030}
1031
522bb2a2
PP
1032/*
1033 * This function handles some semctl commands which require the rw_mutex
1034 * to be held in write mode.
1035 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
1036 */
21a4826a
PP
1037static int semctl_down(struct ipc_namespace *ns, int semid,
1038 int cmd, int version, union semun arg)
1da177e4
LT
1039{
1040 struct sem_array *sma;
1041 int err;
016d7132 1042 struct semid64_ds semid64;
1da177e4
LT
1043 struct kern_ipc_perm *ipcp;
1044
1045 if(cmd == IPC_SET) {
016d7132 1046 if (copy_semid_from_user(&semid64, arg.buf, version))
1da177e4 1047 return -EFAULT;
1da177e4 1048 }
073115d6 1049
a5f75e7f
PP
1050 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0);
1051 if (IS_ERR(ipcp))
1052 return PTR_ERR(ipcp);
073115d6 1053
a5f75e7f 1054 sma = container_of(ipcp, struct sem_array, sem_perm);
1da177e4
LT
1055
1056 err = security_sem_semctl(sma, cmd);
1057 if (err)
1058 goto out_unlock;
1059
1060 switch(cmd){
1061 case IPC_RMID:
01b8b07a 1062 freeary(ns, ipcp);
522bb2a2 1063 goto out_up;
1da177e4 1064 case IPC_SET:
8f4a3809 1065 ipc_update_perm(&semid64.sem_perm, ipcp);
1da177e4 1066 sma->sem_ctime = get_seconds();
1da177e4
LT
1067 break;
1068 default:
1da177e4 1069 err = -EINVAL;
1da177e4 1070 }
1da177e4
LT
1071
1072out_unlock:
1073 sem_unlock(sma);
522bb2a2
PP
1074out_up:
1075 up_write(&sem_ids(ns).rw_mutex);
1da177e4
LT
1076 return err;
1077}
1078
6673e0c3 1079SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg)
1da177e4
LT
1080{
1081 int err = -EINVAL;
1082 int version;
e3893534 1083 struct ipc_namespace *ns;
1da177e4
LT
1084
1085 if (semid < 0)
1086 return -EINVAL;
1087
1088 version = ipc_parse_version(&cmd);
e3893534 1089 ns = current->nsproxy->ipc_ns;
1da177e4
LT
1090
1091 switch(cmd) {
1092 case IPC_INFO:
1093 case SEM_INFO:
4b9fcb0e 1094 case IPC_STAT:
1da177e4 1095 case SEM_STAT:
4b9fcb0e 1096 err = semctl_nolock(ns, semid, cmd, version, arg);
1da177e4
LT
1097 return err;
1098 case GETALL:
1099 case GETVAL:
1100 case GETPID:
1101 case GETNCNT:
1102 case GETZCNT:
1da177e4
LT
1103 case SETVAL:
1104 case SETALL:
e3893534 1105 err = semctl_main(ns,semid,semnum,cmd,version,arg);
1da177e4
LT
1106 return err;
1107 case IPC_RMID:
1108 case IPC_SET:
21a4826a 1109 err = semctl_down(ns, semid, cmd, version, arg);
1da177e4
LT
1110 return err;
1111 default:
1112 return -EINVAL;
1113 }
1114}
6673e0c3
HC
1115#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS
1116asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg)
1117{
1118 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg);
1119}
1120SYSCALL_ALIAS(sys_semctl, SyS_semctl);
1121#endif
1da177e4 1122
1da177e4
LT
1123/* If the task doesn't already have a undo_list, then allocate one
1124 * here. We guarantee there is only one thread using this undo list,
1125 * and current is THE ONE
1126 *
1127 * If this allocation and assignment succeeds, but later
1128 * portions of this code fail, there is no need to free the sem_undo_list.
1129 * Just let it stay associated with the task, and it'll be freed later
1130 * at exit time.
1131 *
1132 * This can block, so callers must hold no locks.
1133 */
1134static inline int get_undo_list(struct sem_undo_list **undo_listp)
1135{
1136 struct sem_undo_list *undo_list;
1da177e4
LT
1137
1138 undo_list = current->sysvsem.undo_list;
1139 if (!undo_list) {
2453a306 1140 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
1da177e4
LT
1141 if (undo_list == NULL)
1142 return -ENOMEM;
00a5dfdb 1143 spin_lock_init(&undo_list->lock);
1da177e4 1144 atomic_set(&undo_list->refcnt, 1);
4daa28f6
MS
1145 INIT_LIST_HEAD(&undo_list->list_proc);
1146
1da177e4
LT
1147 current->sysvsem.undo_list = undo_list;
1148 }
1149 *undo_listp = undo_list;
1150 return 0;
1151}
1152
bf17bb71 1153static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid)
1da177e4 1154{
bf17bb71 1155 struct sem_undo *un;
4daa28f6 1156
bf17bb71
NP
1157 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) {
1158 if (un->semid == semid)
1159 return un;
1da177e4 1160 }
4daa28f6 1161 return NULL;
1da177e4
LT
1162}
1163
bf17bb71
NP
1164static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1165{
1166 struct sem_undo *un;
1167
1168 assert_spin_locked(&ulp->lock);
1169
1170 un = __lookup_undo(ulp, semid);
1171 if (un) {
1172 list_del_rcu(&un->list_proc);
1173 list_add_rcu(&un->list_proc, &ulp->list_proc);
1174 }
1175 return un;
1176}
1177
4daa28f6
MS
1178/**
1179 * find_alloc_undo - Lookup (and if not present create) undo array
1180 * @ns: namespace
1181 * @semid: semaphore array id
1182 *
1183 * The function looks up (and if not present creates) the undo structure.
1184 * The size of the undo structure depends on the size of the semaphore
1185 * array, thus the alloc path is not that straightforward.
380af1b3
MS
1186 * Lifetime-rules: sem_undo is rcu-protected, on success, the function
1187 * performs a rcu_read_lock().
4daa28f6
MS
1188 */
1189static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid)
1da177e4
LT
1190{
1191 struct sem_array *sma;
1192 struct sem_undo_list *ulp;
1193 struct sem_undo *un, *new;
1194 int nsems;
1195 int error;
1196
1197 error = get_undo_list(&ulp);
1198 if (error)
1199 return ERR_PTR(error);
1200
380af1b3 1201 rcu_read_lock();
c530c6ac 1202 spin_lock(&ulp->lock);
1da177e4 1203 un = lookup_undo(ulp, semid);
c530c6ac 1204 spin_unlock(&ulp->lock);
1da177e4
LT
1205 if (likely(un!=NULL))
1206 goto out;
380af1b3 1207 rcu_read_unlock();
1da177e4
LT
1208
1209 /* no undo structure around - allocate one. */
4daa28f6 1210 /* step 1: figure out the size of the semaphore array */
023a5355
ND
1211 sma = sem_lock_check(ns, semid);
1212 if (IS_ERR(sma))
4de85cd6 1213 return ERR_CAST(sma);
023a5355 1214
1da177e4 1215 nsems = sma->sem_nsems;
6ff37972 1216 sem_getref_and_unlock(sma);
1da177e4 1217
4daa28f6 1218 /* step 2: allocate new undo structure */
4668edc3 1219 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
1da177e4 1220 if (!new) {
6ff37972 1221 sem_putref(sma);
1da177e4
LT
1222 return ERR_PTR(-ENOMEM);
1223 }
1da177e4 1224
380af1b3 1225 /* step 3: Acquire the lock on semaphore array */
6ff37972 1226 sem_lock_and_putref(sma);
1da177e4
LT
1227 if (sma->sem_perm.deleted) {
1228 sem_unlock(sma);
1da177e4
LT
1229 kfree(new);
1230 un = ERR_PTR(-EIDRM);
1231 goto out;
1232 }
380af1b3
MS
1233 spin_lock(&ulp->lock);
1234
1235 /*
1236 * step 4: check for races: did someone else allocate the undo struct?
1237 */
1238 un = lookup_undo(ulp, semid);
1239 if (un) {
1240 kfree(new);
1241 goto success;
1242 }
4daa28f6
MS
1243 /* step 5: initialize & link new undo structure */
1244 new->semadj = (short *) &new[1];
380af1b3 1245 new->ulp = ulp;
4daa28f6
MS
1246 new->semid = semid;
1247 assert_spin_locked(&ulp->lock);
380af1b3 1248 list_add_rcu(&new->list_proc, &ulp->list_proc);
4daa28f6
MS
1249 assert_spin_locked(&sma->sem_perm.lock);
1250 list_add(&new->list_id, &sma->list_id);
380af1b3 1251 un = new;
4daa28f6 1252
380af1b3 1253success:
c530c6ac 1254 spin_unlock(&ulp->lock);
380af1b3
MS
1255 rcu_read_lock();
1256 sem_unlock(sma);
1da177e4
LT
1257out:
1258 return un;
1259}
1260
c61284e9
MS
1261
1262/**
1263 * get_queue_result - Retrieve the result code from sem_queue
1264 * @q: Pointer to queue structure
1265 *
1266 * Retrieve the return code from the pending queue. If IN_WAKEUP is found in
1267 * q->status, then we must loop until the value is replaced with the final
1268 * value: This may happen if a task is woken up by an unrelated event (e.g.
1269 * signal) and in parallel the task is woken up by another task because it got
1270 * the requested semaphores.
1271 *
1272 * The function can be called with or without holding the semaphore spinlock.
1273 */
1274static int get_queue_result(struct sem_queue *q)
1275{
1276 int error;
1277
1278 error = q->status;
1279 while (unlikely(error == IN_WAKEUP)) {
1280 cpu_relax();
1281 error = q->status;
1282 }
1283
1284 return error;
1285}
1286
1287
d5460c99
HC
1288SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops,
1289 unsigned, nsops, const struct timespec __user *, timeout)
1da177e4
LT
1290{
1291 int error = -EINVAL;
1292 struct sem_array *sma;
1293 struct sembuf fast_sops[SEMOPM_FAST];
1294 struct sembuf* sops = fast_sops, *sop;
1295 struct sem_undo *un;
b78755ab 1296 int undos = 0, alter = 0, max;
1da177e4
LT
1297 struct sem_queue queue;
1298 unsigned long jiffies_left = 0;
e3893534 1299 struct ipc_namespace *ns;
0a2b9d4c 1300 struct list_head tasks;
e3893534
KK
1301
1302 ns = current->nsproxy->ipc_ns;
1da177e4
LT
1303
1304 if (nsops < 1 || semid < 0)
1305 return -EINVAL;
e3893534 1306 if (nsops > ns->sc_semopm)
1da177e4
LT
1307 return -E2BIG;
1308 if(nsops > SEMOPM_FAST) {
1309 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1310 if(sops==NULL)
1311 return -ENOMEM;
1312 }
1313 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1314 error=-EFAULT;
1315 goto out_free;
1316 }
1317 if (timeout) {
1318 struct timespec _timeout;
1319 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1320 error = -EFAULT;
1321 goto out_free;
1322 }
1323 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1324 _timeout.tv_nsec >= 1000000000L) {
1325 error = -EINVAL;
1326 goto out_free;
1327 }
1328 jiffies_left = timespec_to_jiffies(&_timeout);
1329 }
1330 max = 0;
1331 for (sop = sops; sop < sops + nsops; sop++) {
1332 if (sop->sem_num >= max)
1333 max = sop->sem_num;
1334 if (sop->sem_flg & SEM_UNDO)
b78755ab
MS
1335 undos = 1;
1336 if (sop->sem_op != 0)
1da177e4
LT
1337 alter = 1;
1338 }
1da177e4 1339
1da177e4 1340 if (undos) {
4daa28f6 1341 un = find_alloc_undo(ns, semid);
1da177e4
LT
1342 if (IS_ERR(un)) {
1343 error = PTR_ERR(un);
1344 goto out_free;
1345 }
1346 } else
1347 un = NULL;
1348
0a2b9d4c
MS
1349 INIT_LIST_HEAD(&tasks);
1350
023a5355
ND
1351 sma = sem_lock_check(ns, semid);
1352 if (IS_ERR(sma)) {
380af1b3
MS
1353 if (un)
1354 rcu_read_unlock();
023a5355 1355 error = PTR_ERR(sma);
1da177e4 1356 goto out_free;
023a5355
ND
1357 }
1358
1da177e4 1359 /*
4daa28f6 1360 * semid identifiers are not unique - find_alloc_undo may have
1da177e4 1361 * allocated an undo structure, it was invalidated by an RMID
4daa28f6 1362 * and now a new array with received the same id. Check and fail.
380af1b3
MS
1363 * This case can be detected checking un->semid. The existance of
1364 * "un" itself is guaranteed by rcu.
1da177e4 1365 */
4daa28f6 1366 error = -EIDRM;
380af1b3
MS
1367 if (un) {
1368 if (un->semid == -1) {
1369 rcu_read_unlock();
1370 goto out_unlock_free;
1371 } else {
1372 /*
1373 * rcu lock can be released, "un" cannot disappear:
1374 * - sem_lock is acquired, thus IPC_RMID is
1375 * impossible.
1376 * - exit_sem is impossible, it always operates on
1377 * current (or a dead task).
1378 */
1379
1380 rcu_read_unlock();
1381 }
1382 }
4daa28f6 1383
1da177e4
LT
1384 error = -EFBIG;
1385 if (max >= sma->sem_nsems)
1386 goto out_unlock_free;
1387
1388 error = -EACCES;
1389 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1390 goto out_unlock_free;
1391
1392 error = security_sem_semop(sma, sops, nsops, alter);
1393 if (error)
1394 goto out_unlock_free;
1395
b488893a 1396 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
1da177e4
LT
1397 if (error <= 0) {
1398 if (alter && error == 0)
0a2b9d4c 1399 do_smart_update(sma, sops, nsops, 1, &tasks);
636c6be8 1400
1da177e4
LT
1401 goto out_unlock_free;
1402 }
1403
1404 /* We need to sleep on this operation, so we put the current
1405 * task into the pending queue and go to sleep.
1406 */
1407
1da177e4
LT
1408 queue.sops = sops;
1409 queue.nsops = nsops;
1410 queue.undo = un;
b488893a 1411 queue.pid = task_tgid_vnr(current);
1da177e4
LT
1412 queue.alter = alter;
1413 if (alter)
a1193f8e 1414 list_add_tail(&queue.list, &sma->sem_pending);
1da177e4 1415 else
a1193f8e 1416 list_add(&queue.list, &sma->sem_pending);
1da177e4 1417
b97e820f
MS
1418 if (nsops == 1) {
1419 struct sem *curr;
1420 curr = &sma->sem_base[sops->sem_num];
1421
1422 if (alter)
1423 list_add_tail(&queue.simple_list, &curr->sem_pending);
1424 else
1425 list_add(&queue.simple_list, &curr->sem_pending);
1426 } else {
1427 INIT_LIST_HEAD(&queue.simple_list);
1428 sma->complex_count++;
1429 }
1430
1da177e4
LT
1431 queue.status = -EINTR;
1432 queue.sleeper = current;
1433 current->state = TASK_INTERRUPTIBLE;
1434 sem_unlock(sma);
1435
1436 if (timeout)
1437 jiffies_left = schedule_timeout(jiffies_left);
1438 else
1439 schedule();
1440
c61284e9 1441 error = get_queue_result(&queue);
1da177e4
LT
1442
1443 if (error != -EINTR) {
1444 /* fast path: update_queue already obtained all requested
c61284e9
MS
1445 * resources.
1446 * Perform a smp_mb(): User space could assume that semop()
1447 * is a memory barrier: Without the mb(), the cpu could
1448 * speculatively read in user space stale data that was
1449 * overwritten by the previous owner of the semaphore.
1450 */
1451 smp_mb();
1452
1da177e4
LT
1453 goto out_free;
1454 }
1455
e3893534 1456 sma = sem_lock(ns, semid);
023a5355 1457 if (IS_ERR(sma)) {
1da177e4
LT
1458 error = -EIDRM;
1459 goto out_free;
1460 }
1461
c61284e9
MS
1462 error = get_queue_result(&queue);
1463
1da177e4
LT
1464 /*
1465 * If queue.status != -EINTR we are woken up by another process
1466 */
c61284e9 1467
1da177e4
LT
1468 if (error != -EINTR) {
1469 goto out_unlock_free;
1470 }
1471
1472 /*
1473 * If an interrupt occurred we have to clean up the queue
1474 */
1475 if (timeout && jiffies_left == 0)
1476 error = -EAGAIN;
b97e820f 1477 unlink_queue(sma, &queue);
1da177e4
LT
1478
1479out_unlock_free:
1480 sem_unlock(sma);
0a2b9d4c
MS
1481
1482 wake_up_sem_queue_do(&tasks);
1da177e4
LT
1483out_free:
1484 if(sops != fast_sops)
1485 kfree(sops);
1486 return error;
1487}
1488
d5460c99
HC
1489SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops,
1490 unsigned, nsops)
1da177e4
LT
1491{
1492 return sys_semtimedop(semid, tsops, nsops, NULL);
1493}
1494
1495/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1496 * parent and child tasks.
1da177e4
LT
1497 */
1498
1499int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1500{
1501 struct sem_undo_list *undo_list;
1502 int error;
1503
1504 if (clone_flags & CLONE_SYSVSEM) {
1505 error = get_undo_list(&undo_list);
1506 if (error)
1507 return error;
1da177e4
LT
1508 atomic_inc(&undo_list->refcnt);
1509 tsk->sysvsem.undo_list = undo_list;
1510 } else
1511 tsk->sysvsem.undo_list = NULL;
1512
1513 return 0;
1514}
1515
1516/*
1517 * add semadj values to semaphores, free undo structures.
1518 * undo structures are not freed when semaphore arrays are destroyed
1519 * so some of them may be out of date.
1520 * IMPLEMENTATION NOTE: There is some confusion over whether the
1521 * set of adjustments that needs to be done should be done in an atomic
1522 * manner or not. That is, if we are attempting to decrement the semval
1523 * should we queue up and wait until we can do so legally?
1524 * The original implementation attempted to do this (queue and wait).
1525 * The current implementation does not do so. The POSIX standard
1526 * and SVID should be consulted to determine what behavior is mandated.
1527 */
1528void exit_sem(struct task_struct *tsk)
1529{
4daa28f6 1530 struct sem_undo_list *ulp;
1da177e4 1531
4daa28f6
MS
1532 ulp = tsk->sysvsem.undo_list;
1533 if (!ulp)
1da177e4 1534 return;
9edff4ab 1535 tsk->sysvsem.undo_list = NULL;
1da177e4 1536
4daa28f6 1537 if (!atomic_dec_and_test(&ulp->refcnt))
1da177e4
LT
1538 return;
1539
380af1b3 1540 for (;;) {
1da177e4 1541 struct sem_array *sma;
380af1b3 1542 struct sem_undo *un;
0a2b9d4c 1543 struct list_head tasks;
380af1b3 1544 int semid;
4daa28f6
MS
1545 int i;
1546
380af1b3 1547 rcu_read_lock();
05725f7e
JP
1548 un = list_entry_rcu(ulp->list_proc.next,
1549 struct sem_undo, list_proc);
380af1b3
MS
1550 if (&un->list_proc == &ulp->list_proc)
1551 semid = -1;
1552 else
1553 semid = un->semid;
1554 rcu_read_unlock();
4daa28f6 1555
380af1b3
MS
1556 if (semid == -1)
1557 break;
1da177e4 1558
380af1b3 1559 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid);
1da177e4 1560
380af1b3
MS
1561 /* exit_sem raced with IPC_RMID, nothing to do */
1562 if (IS_ERR(sma))
1563 continue;
1da177e4 1564
bf17bb71 1565 un = __lookup_undo(ulp, semid);
380af1b3
MS
1566 if (un == NULL) {
1567 /* exit_sem raced with IPC_RMID+semget() that created
1568 * exactly the same semid. Nothing to do.
1569 */
1570 sem_unlock(sma);
1571 continue;
1572 }
1573
1574 /* remove un from the linked lists */
4daa28f6
MS
1575 assert_spin_locked(&sma->sem_perm.lock);
1576 list_del(&un->list_id);
1577
380af1b3
MS
1578 spin_lock(&ulp->lock);
1579 list_del_rcu(&un->list_proc);
1580 spin_unlock(&ulp->lock);
1581
4daa28f6
MS
1582 /* perform adjustments registered in un */
1583 for (i = 0; i < sma->sem_nsems; i++) {
5f921ae9 1584 struct sem * semaphore = &sma->sem_base[i];
4daa28f6
MS
1585 if (un->semadj[i]) {
1586 semaphore->semval += un->semadj[i];
1da177e4
LT
1587 /*
1588 * Range checks of the new semaphore value,
1589 * not defined by sus:
1590 * - Some unices ignore the undo entirely
1591 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1592 * - some cap the value (e.g. FreeBSD caps
1593 * at 0, but doesn't enforce SEMVMX)
1594 *
1595 * Linux caps the semaphore value, both at 0
1596 * and at SEMVMX.
1597 *
1598 * Manfred <manfred@colorfullife.com>
1599 */
5f921ae9
IM
1600 if (semaphore->semval < 0)
1601 semaphore->semval = 0;
1602 if (semaphore->semval > SEMVMX)
1603 semaphore->semval = SEMVMX;
b488893a 1604 semaphore->sempid = task_tgid_vnr(current);
1da177e4
LT
1605 }
1606 }
1da177e4 1607 /* maybe some queued-up processes were waiting for this */
0a2b9d4c
MS
1608 INIT_LIST_HEAD(&tasks);
1609 do_smart_update(sma, NULL, 0, 1, &tasks);
1da177e4 1610 sem_unlock(sma);
0a2b9d4c 1611 wake_up_sem_queue_do(&tasks);
380af1b3
MS
1612
1613 call_rcu(&un->rcu, free_un);
1da177e4 1614 }
4daa28f6 1615 kfree(ulp);
1da177e4
LT
1616}
1617
1618#ifdef CONFIG_PROC_FS
19b4946c 1619static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
1da177e4 1620{
19b4946c
MW
1621 struct sem_array *sma = it;
1622
1623 return seq_printf(s,
b97e820f 1624 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n",
19b4946c 1625 sma->sem_perm.key,
7ca7e564 1626 sma->sem_perm.id,
19b4946c
MW
1627 sma->sem_perm.mode,
1628 sma->sem_nsems,
1629 sma->sem_perm.uid,
1630 sma->sem_perm.gid,
1631 sma->sem_perm.cuid,
1632 sma->sem_perm.cgid,
1633 sma->sem_otime,
1634 sma->sem_ctime);
1da177e4
LT
1635}
1636#endif