blob: 77162eddcbfc3d8bf5873aa385f66d0bb7564c2e [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/ipc/sem.c
3 * Copyright (C) 1992 Krishna Balasubramanian
4 * Copyright (C) 1995 Eric Schenk, Bruno Haible
5 *
6 * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995):
7 * This code underwent a massive rewrite in order to solve some problems
8 * with the original code. In particular the original code failed to
9 * wake up processes that were waiting for semval to go to 0 if the
10 * value went to 0 and was then incremented rapidly enough. In solving
11 * this problem I have also modified the implementation so that it
12 * processes pending operations in a FIFO manner, thus give a guarantee
13 * that processes waiting for a lock on the semaphore won't starve
14 * unless another locking process fails to unlock.
15 * In addition the following two changes in behavior have been introduced:
16 * - The original implementation of semop returned the value
17 * last semaphore element examined on success. This does not
18 * match the manual page specifications, and effectively
19 * allows the user to read the semaphore even if they do not
20 * have read permissions. The implementation now returns 0
21 * on success as stated in the manual page.
22 * - There is some confusion over whether the set of undo adjustments
23 * to be performed at exit should be done in an atomic manner.
24 * That is, if we are attempting to decrement the semval should we queue
25 * up and wait until we can do so legally?
26 * The original implementation attempted to do this.
27 * The current implementation does not do so. This is because I don't
28 * think it is the right thing (TM) to do, and because I couldn't
29 * see a clean way to get the old behavior with the new design.
30 * The POSIX standard and SVID should be consulted to determine
31 * what behavior is mandated.
32 *
33 * Further notes on refinement (Christoph Rohland, December 1998):
34 * - The POSIX standard says, that the undo adjustments simply should
35 * redo. So the current implementation is o.K.
36 * - The previous code had two flaws:
37 * 1) It actively gave the semaphore to the next waiting process
38 * sleeping on the semaphore. Since this process did not have the
39 * cpu this led to many unnecessary context switches and bad
40 * performance. Now we only check which process should be able to
41 * get the semaphore and if this process wants to reduce some
42 * semaphore value we simply wake it up without doing the
43 * operation. So it has to try to get it later. Thus e.g. the
44 * running process may reacquire the semaphore during the current
45 * time slice. If it only waits for zero or increases the semaphore,
46 * we do the operation in advance and wake it up.
47 * 2) It did not wake up all zero waiting processes. We try to do
48 * better but only get the semops right which only wait for zero or
49 * increase. If there are decrement operations in the operations
50 * array we do the same as before.
51 *
52 * With the incarnation of O(1) scheduler, it becomes unnecessary to perform
53 * check/retry algorithm for waking up blocked processes as the new scheduler
54 * is better at handling thread switch than the old one.
55 *
56 * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com>
57 *
58 * SMP-threaded, sysctl's added
Christian Kujau624dffc2006-01-15 02:43:54 +010059 * (c) 1999 Manfred Spraul <manfred@colorfullife.com>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 * Enforced range limit on SEM_UNDO
61 * (c) 2001 Red Hat Inc <alan@redhat.com>
62 * Lockless wakeup
63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com>
Steve Grubb073115d2006-04-02 17:07:33 -040064 *
65 * support for audit of ipc object properties and permission changes
66 * Dustin Kirkland <dustin.kirkland@us.ibm.com>
Kirill Korotaeve3893532006-10-02 02:18:22 -070067 *
68 * namespaces support
69 * OpenVZ, SWsoft Inc.
70 * Pavel Emelianov <xemul@openvz.org>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071 */
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#include <linux/slab.h>
74#include <linux/spinlock.h>
75#include <linux/init.h>
76#include <linux/proc_fs.h>
77#include <linux/time.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070078#include <linux/security.h>
79#include <linux/syscalls.h>
80#include <linux/audit.h>
Randy.Dunlapc59ede72006-01-11 12:17:46 -080081#include <linux/capability.h>
Mike Waychison19b49462005-09-06 15:17:10 -070082#include <linux/seq_file.h>
Nadia Derbey3e148c72007-10-18 23:40:54 -070083#include <linux/rwsem.h>
Kirill Korotaeve3893532006-10-02 02:18:22 -070084#include <linux/nsproxy.h>
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -080085#include <linux/ipc_namespace.h>
Ingo Molnar5f921ae2006-03-26 01:37:17 -080086
Linus Torvalds1da177e2005-04-16 15:20:36 -070087#include <asm/uaccess.h>
88#include "util.h"
89
Pierre Peiffered2ddbf2008-02-08 04:18:57 -080090#define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS])
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Kirill Korotaeve3893532006-10-02 02:18:22 -070092#define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm)
Nadia Derbey1b531f22007-10-18 23:40:55 -070093#define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -070094
Nadia Derbey7748dbf2007-10-18 23:40:49 -070095static int newary(struct ipc_namespace *, struct ipc_params *);
Pierre Peiffer01b8b072008-02-08 04:18:57 -080096static void freeary(struct ipc_namespace *, struct kern_ipc_perm *);
Linus Torvalds1da177e2005-04-16 15:20:36 -070097#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -070098static int sysvipc_sem_proc_show(struct seq_file *s, void *it);
Linus Torvalds1da177e2005-04-16 15:20:36 -070099#endif
100
101#define SEMMSL_FAST 256 /* 512 bytes on stack */
102#define SEMOPM_FAST 64 /* ~ 372 bytes on stack */
103
104/*
105 * linked list protection:
106 * sem_undo.id_next,
107 * sem_array.sem_pending{,last},
108 * sem_array.sem_undo: sem_lock() for read/write
109 * sem_undo.proc_next: only "current" is allowed to read/write that field.
110 *
111 */
112
Kirill Korotaeve3893532006-10-02 02:18:22 -0700113#define sc_semmsl sem_ctls[0]
114#define sc_semmns sem_ctls[1]
115#define sc_semopm sem_ctls[2]
116#define sc_semmni sem_ctls[3]
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800118void sem_init_ns(struct ipc_namespace *ns)
Kirill Korotaeve3893532006-10-02 02:18:22 -0700119{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700120 ns->sc_semmsl = SEMMSL;
121 ns->sc_semmns = SEMMNS;
122 ns->sc_semopm = SEMOPM;
123 ns->sc_semmni = SEMMNI;
124 ns->used_sems = 0;
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800125 ipc_init_ids(&ns->ids[IPC_SEM_IDS]);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700126}
127
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800128#ifdef CONFIG_IPC_NS
Kirill Korotaeve3893532006-10-02 02:18:22 -0700129void sem_exit_ns(struct ipc_namespace *ns)
130{
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800131 free_ipcs(ns, &sem_ids(ns), freeary);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700132}
Pavel Emelyanovae5e1b22008-02-08 04:18:22 -0800133#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134
135void __init sem_init (void)
136{
Pierre Peiffered2ddbf2008-02-08 04:18:57 -0800137 sem_init_ns(&init_ipc_ns);
Mike Waychison19b49462005-09-06 15:17:10 -0700138 ipc_init_proc_interface("sysvipc/sem",
139 " key semid perms nsems uid gid cuid cgid otime ctime\n",
Kirill Korotaeve3893532006-10-02 02:18:22 -0700140 IPC_SEM_IDS, sysvipc_sem_proc_show);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
Nadia Derbey3e148c72007-10-18 23:40:54 -0700143/*
144 * This routine is called in the paths where the rw_mutex is held to protect
145 * access to the idr tree.
146 */
147static inline struct sem_array *sem_lock_check_down(struct ipc_namespace *ns,
148 int id)
149{
150 struct kern_ipc_perm *ipcp = ipc_lock_check_down(&sem_ids(ns), id);
151
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800152 if (IS_ERR(ipcp))
153 return (struct sem_array *)ipcp;
154
Nadia Derbey3e148c72007-10-18 23:40:54 -0700155 return container_of(ipcp, struct sem_array, sem_perm);
156}
157
158/*
159 * sem_lock_(check_) routines are called in the paths where the rw_mutex
160 * is not held.
161 */
Nadia Derbey023a5352007-10-18 23:40:51 -0700162static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id)
163{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700164 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id);
165
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800166 if (IS_ERR(ipcp))
167 return (struct sem_array *)ipcp;
168
Nadia Derbey03f02c72007-10-18 23:40:51 -0700169 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700170}
171
172static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns,
173 int id)
174{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700175 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id);
176
Pierre Peifferb1ed88b2008-02-06 01:36:23 -0800177 if (IS_ERR(ipcp))
178 return (struct sem_array *)ipcp;
179
Nadia Derbey03f02c72007-10-18 23:40:51 -0700180 return container_of(ipcp, struct sem_array, sem_perm);
Nadia Derbey023a5352007-10-18 23:40:51 -0700181}
182
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700183static inline void sem_lock_and_putref(struct sem_array *sma)
184{
185 ipc_lock_by_ptr(&sma->sem_perm);
186 ipc_rcu_putref(sma);
187}
188
189static inline void sem_getref_and_unlock(struct sem_array *sma)
190{
191 ipc_rcu_getref(sma);
192 ipc_unlock(&(sma)->sem_perm);
193}
194
195static inline void sem_putref(struct sem_array *sma)
196{
197 ipc_lock_by_ptr(&sma->sem_perm);
198 ipc_rcu_putref(sma);
199 ipc_unlock(&(sma)->sem_perm);
200}
201
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700202static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s)
203{
204 ipc_rmid(&sem_ids(ns), &s->sem_perm);
205}
206
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207/*
208 * Lockless wakeup algorithm:
209 * Without the check/retry algorithm a lockless wakeup is possible:
210 * - queue.status is initialized to -EINTR before blocking.
211 * - wakeup is performed by
212 * * unlinking the queue entry from sma->sem_pending
213 * * setting queue.status to IN_WAKEUP
214 * This is the notification for the blocked thread that a
215 * result value is imminent.
216 * * call wake_up_process
217 * * set queue.status to the final value.
218 * - the previously blocked thread checks queue.status:
219 * * if it's IN_WAKEUP, then it must wait until the value changes
220 * * if it's not -EINTR, then the operation was completed by
221 * update_queue. semtimedop can return queue.status without
Ingo Molnar5f921ae2006-03-26 01:37:17 -0800222 * performing any operation on the sem array.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223 * * otherwise it must acquire the spinlock and check what's up.
224 *
225 * The two-stage algorithm is necessary to protect against the following
226 * races:
227 * - if queue.status is set after wake_up_process, then the woken up idle
228 * thread could race forward and try (and fail) to acquire sma->lock
229 * before update_queue had a chance to set queue.status
230 * - if queue.status is written before wake_up_process and if the
231 * blocked process is woken up by a signal between writing
232 * queue.status and the wake_up_process, then the woken up
233 * process could return from semtimedop and die by calling
234 * sys_exit before wake_up_process is called. Then wake_up_process
235 * will oops, because the task structure is already invalid.
236 * (yes, this happened on s390 with sysv msg).
237 *
238 */
239#define IN_WAKEUP 1
240
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700241/**
242 * newary - Create a new semaphore set
243 * @ns: namespace
244 * @params: ptr to the structure that contains key, semflg and nsems
245 *
Nadia Derbey3e148c72007-10-18 23:40:54 -0700246 * Called with sem_ids.rw_mutex held (as a writer)
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700247 */
248
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700249static int newary(struct ipc_namespace *ns, struct ipc_params *params)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250{
251 int id;
252 int retval;
253 struct sem_array *sma;
254 int size;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700255 key_t key = params->key;
256 int nsems = params->u.nsems;
257 int semflg = params->flg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258
259 if (!nsems)
260 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700261 if (ns->used_sems + nsems > ns->sc_semmns)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return -ENOSPC;
263
264 size = sizeof (*sma) + nsems * sizeof (struct sem);
265 sma = ipc_rcu_alloc(size);
266 if (!sma) {
267 return -ENOMEM;
268 }
269 memset (sma, 0, size);
270
271 sma->sem_perm.mode = (semflg & S_IRWXUGO);
272 sma->sem_perm.key = key;
273
274 sma->sem_perm.security = NULL;
275 retval = security_sem_alloc(sma);
276 if (retval) {
277 ipc_rcu_putref(sma);
278 return retval;
279 }
280
Kirill Korotaeve3893532006-10-02 02:18:22 -0700281 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700282 if (id < 0) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 security_sem_free(sma);
284 ipc_rcu_putref(sma);
Pierre Peiffer283bb7f2007-10-18 23:40:57 -0700285 return id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286 }
Kirill Korotaeve3893532006-10-02 02:18:22 -0700287 ns->used_sems += nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700288
289 sma->sem_base = (struct sem *) &sma[1];
290 /* sma->sem_pending = NULL; */
291 sma->sem_pending_last = &sma->sem_pending;
292 /* sma->undo = NULL; */
293 sma->sem_nsems = nsems;
294 sma->sem_ctime = get_seconds();
295 sem_unlock(sma);
296
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700297 return sma->sem_perm.id;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298}
299
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700300
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700301/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700302 * Called with sem_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700303 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700304static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700306 struct sem_array *sma;
307
308 sma = container_of(ipcp, struct sem_array, sem_perm);
309 return security_sem_associate(sma, semflg);
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700310}
311
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700312/*
Nadia Derbey3e148c72007-10-18 23:40:54 -0700313 * Called with sem_ids.rw_mutex and ipcp locked.
Nadia Derbeyf4566f02007-10-18 23:40:53 -0700314 */
Nadia Derbey03f02c72007-10-18 23:40:51 -0700315static inline int sem_more_checks(struct kern_ipc_perm *ipcp,
316 struct ipc_params *params)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700317{
Nadia Derbey03f02c72007-10-18 23:40:51 -0700318 struct sem_array *sma;
319
320 sma = container_of(ipcp, struct sem_array, sem_perm);
321 if (params->u.nsems > sma->sem_nsems)
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700322 return -EINVAL;
323
324 return 0;
325}
326
327asmlinkage long sys_semget(key_t key, int nsems, int semflg)
328{
Kirill Korotaeve3893532006-10-02 02:18:22 -0700329 struct ipc_namespace *ns;
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700330 struct ipc_ops sem_ops;
331 struct ipc_params sem_params;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Kirill Korotaeve3893532006-10-02 02:18:22 -0700333 ns = current->nsproxy->ipc_ns;
334
335 if (nsems < 0 || nsems > ns->sc_semmsl)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336 return -EINVAL;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700337
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700338 sem_ops.getnew = newary;
339 sem_ops.associate = sem_security;
340 sem_ops.more_checks = sem_more_checks;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700341
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700342 sem_params.key = key;
343 sem_params.flg = semflg;
344 sem_params.u.nsems = nsems;
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700345
Nadia Derbey7748dbf2007-10-18 23:40:49 -0700346 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700347}
348
349/* Manage the doubly linked list sma->sem_pending as a FIFO:
350 * insert new queue elements at the tail sma->sem_pending_last.
351 */
352static inline void append_to_queue (struct sem_array * sma,
353 struct sem_queue * q)
354{
355 *(q->prev = sma->sem_pending_last) = q;
356 *(sma->sem_pending_last = &q->next) = NULL;
357}
358
359static inline void prepend_to_queue (struct sem_array * sma,
360 struct sem_queue * q)
361{
362 q->next = sma->sem_pending;
363 *(q->prev = &sma->sem_pending) = q;
364 if (q->next)
365 q->next->prev = &q->next;
366 else /* sma->sem_pending_last == &sma->sem_pending */
367 sma->sem_pending_last = &q->next;
368}
369
370static inline void remove_from_queue (struct sem_array * sma,
371 struct sem_queue * q)
372{
373 *(q->prev) = q->next;
374 if (q->next)
375 q->next->prev = q->prev;
376 else /* sma->sem_pending_last == &q->next */
377 sma->sem_pending_last = q->prev;
378 q->prev = NULL; /* mark as removed */
379}
380
381/*
382 * Determine whether a sequence of semaphore operations would succeed
383 * all at once. Return 0 if yes, 1 if need to sleep, else return error code.
384 */
385
386static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops,
387 int nsops, struct sem_undo *un, int pid)
388{
389 int result, sem_op;
390 struct sembuf *sop;
391 struct sem * curr;
392
393 for (sop = sops; sop < sops + nsops; sop++) {
394 curr = sma->sem_base + sop->sem_num;
395 sem_op = sop->sem_op;
396 result = curr->semval;
397
398 if (!sem_op && result)
399 goto would_block;
400
401 result += sem_op;
402 if (result < 0)
403 goto would_block;
404 if (result > SEMVMX)
405 goto out_of_range;
406 if (sop->sem_flg & SEM_UNDO) {
407 int undo = un->semadj[sop->sem_num] - sem_op;
408 /*
409 * Exceeding the undo range is an error.
410 */
411 if (undo < (-SEMAEM - 1) || undo > SEMAEM)
412 goto out_of_range;
413 }
414 curr->semval = result;
415 }
416
417 sop--;
418 while (sop >= sops) {
419 sma->sem_base[sop->sem_num].sempid = pid;
420 if (sop->sem_flg & SEM_UNDO)
421 un->semadj[sop->sem_num] -= sop->sem_op;
422 sop--;
423 }
424
425 sma->sem_otime = get_seconds();
426 return 0;
427
428out_of_range:
429 result = -ERANGE;
430 goto undo;
431
432would_block:
433 if (sop->sem_flg & IPC_NOWAIT)
434 result = -EAGAIN;
435 else
436 result = 1;
437
438undo:
439 sop--;
440 while (sop >= sops) {
441 sma->sem_base[sop->sem_num].semval -= sop->sem_op;
442 sop--;
443 }
444
445 return result;
446}
447
448/* Go through the pending queue for the indicated semaphore
449 * looking for tasks that can be completed.
450 */
451static void update_queue (struct sem_array * sma)
452{
453 int error;
454 struct sem_queue * q;
455
456 q = sma->sem_pending;
457 while(q) {
458 error = try_atomic_semop(sma, q->sops, q->nsops,
459 q->undo, q->pid);
460
461 /* Does q->sleeper still need to sleep? */
462 if (error <= 0) {
463 struct sem_queue *n;
464 remove_from_queue(sma,q);
465 q->status = IN_WAKEUP;
466 /*
467 * Continue scanning. The next operation
468 * that must be checked depends on the type of the
469 * completed operation:
470 * - if the operation modified the array, then
471 * restart from the head of the queue and
472 * check for threads that might be waiting
473 * for semaphore values to become 0.
474 * - if the operation didn't modify the array,
475 * then just continue.
476 */
477 if (q->alter)
478 n = sma->sem_pending;
479 else
480 n = q->next;
481 wake_up_process(q->sleeper);
482 /* hands-off: q will disappear immediately after
483 * writing q->status.
484 */
Linus Torvalds1224b372005-12-24 12:19:38 -0800485 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486 q->status = error;
487 q = n;
488 } else {
489 q = q->next;
490 }
491 }
492}
493
494/* The following counts are associated to each semaphore:
495 * semncnt number of tasks waiting on semval being nonzero
496 * semzcnt number of tasks waiting on semval being zero
497 * This model assumes that a task waits on exactly one semaphore.
498 * Since semaphore operations are to be performed atomically, tasks actually
499 * wait on a whole sequence of semaphores simultaneously.
500 * The counts we return here are a rough approximation, but still
501 * warrant that semncnt+semzcnt>0 if the task is on the pending queue.
502 */
503static int count_semncnt (struct sem_array * sma, ushort semnum)
504{
505 int semncnt;
506 struct sem_queue * q;
507
508 semncnt = 0;
509 for (q = sma->sem_pending; q; q = q->next) {
510 struct sembuf * sops = q->sops;
511 int nsops = q->nsops;
512 int i;
513 for (i = 0; i < nsops; i++)
514 if (sops[i].sem_num == semnum
515 && (sops[i].sem_op < 0)
516 && !(sops[i].sem_flg & IPC_NOWAIT))
517 semncnt++;
518 }
519 return semncnt;
520}
521static int count_semzcnt (struct sem_array * sma, ushort semnum)
522{
523 int semzcnt;
524 struct sem_queue * q;
525
526 semzcnt = 0;
527 for (q = sma->sem_pending; q; q = q->next) {
528 struct sembuf * sops = q->sops;
529 int nsops = q->nsops;
530 int i;
531 for (i = 0; i < nsops; i++)
532 if (sops[i].sem_num == semnum
533 && (sops[i].sem_op == 0)
534 && !(sops[i].sem_flg & IPC_NOWAIT))
535 semzcnt++;
536 }
537 return semzcnt;
538}
539
Nadia Derbey3e148c72007-10-18 23:40:54 -0700540/* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked
541 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex
542 * remains locked on exit.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800544static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545{
546 struct sem_undo *un;
547 struct sem_queue *q;
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800548 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549
550 /* Invalidate the existing undo structures for this semaphore set.
551 * (They will be freed without any further action in exit_sem()
552 * or during the next semop.)
553 */
554 for (un = sma->undo; un; un = un->id_next)
555 un->semid = -1;
556
557 /* Wake up all pending processes and let them fail with EIDRM. */
558 q = sma->sem_pending;
559 while(q) {
560 struct sem_queue *n;
561 /* lazy remove_from_queue: we are killing the whole queue */
562 q->prev = NULL;
563 n = q->next;
564 q->status = IN_WAKEUP;
565 wake_up_process(q->sleeper); /* doesn't sleep */
Manfred Spraul6003a932005-12-23 23:57:41 +0100566 smp_wmb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 q->status = -EIDRM; /* hands-off q */
568 q = n;
569 }
570
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700571 /* Remove the semaphore set from the IDR */
572 sem_rmid(ns, sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 sem_unlock(sma);
574
Kirill Korotaeve3893532006-10-02 02:18:22 -0700575 ns->used_sems -= sma->sem_nsems;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 security_sem_free(sma);
577 ipc_rcu_putref(sma);
578}
579
580static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version)
581{
582 switch(version) {
583 case IPC_64:
584 return copy_to_user(buf, in, sizeof(*in));
585 case IPC_OLD:
586 {
587 struct semid_ds out;
588
589 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm);
590
591 out.sem_otime = in->sem_otime;
592 out.sem_ctime = in->sem_ctime;
593 out.sem_nsems = in->sem_nsems;
594
595 return copy_to_user(buf, &out, sizeof(out));
596 }
597 default:
598 return -EINVAL;
599 }
600}
601
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -0800602static int semctl_nolock(struct ipc_namespace *ns, int semid,
603 int cmd, int version, union semun arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
605 int err = -EINVAL;
606 struct sem_array *sma;
607
608 switch(cmd) {
609 case IPC_INFO:
610 case SEM_INFO:
611 {
612 struct seminfo seminfo;
613 int max_id;
614
615 err = security_sem_semctl(NULL, cmd);
616 if (err)
617 return err;
618
619 memset(&seminfo,0,sizeof(seminfo));
Kirill Korotaeve3893532006-10-02 02:18:22 -0700620 seminfo.semmni = ns->sc_semmni;
621 seminfo.semmns = ns->sc_semmns;
622 seminfo.semmsl = ns->sc_semmsl;
623 seminfo.semopm = ns->sc_semopm;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 seminfo.semvmx = SEMVMX;
625 seminfo.semmnu = SEMMNU;
626 seminfo.semmap = SEMMAP;
627 seminfo.semume = SEMUME;
Nadia Derbey3e148c72007-10-18 23:40:54 -0700628 down_read(&sem_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 if (cmd == SEM_INFO) {
Kirill Korotaeve3893532006-10-02 02:18:22 -0700630 seminfo.semusz = sem_ids(ns).in_use;
631 seminfo.semaem = ns->used_sems;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 } else {
633 seminfo.semusz = SEMUSZ;
634 seminfo.semaem = SEMAEM;
635 }
Nadia Derbey7ca7e562007-10-18 23:40:48 -0700636 max_id = ipc_get_maxid(&sem_ids(ns));
Nadia Derbey3e148c72007-10-18 23:40:54 -0700637 up_read(&sem_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo)))
639 return -EFAULT;
640 return (max_id < 0) ? 0: max_id;
641 }
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -0800642 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 case SEM_STAT:
644 {
645 struct semid64_ds tbuf;
646 int id;
647
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -0800648 if (cmd == SEM_STAT) {
649 sma = sem_lock(ns, semid);
650 if (IS_ERR(sma))
651 return PTR_ERR(sma);
652 id = sma->sem_perm.id;
653 } else {
654 sma = sem_lock_check(ns, semid);
655 if (IS_ERR(sma))
656 return PTR_ERR(sma);
657 id = 0;
658 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659
660 err = -EACCES;
661 if (ipcperms (&sma->sem_perm, S_IRUGO))
662 goto out_unlock;
663
664 err = security_sem_semctl(sma, cmd);
665 if (err)
666 goto out_unlock;
667
Nadia Derbey023a5352007-10-18 23:40:51 -0700668 memset(&tbuf, 0, sizeof(tbuf));
669
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm);
671 tbuf.sem_otime = sma->sem_otime;
672 tbuf.sem_ctime = sma->sem_ctime;
673 tbuf.sem_nsems = sma->sem_nsems;
674 sem_unlock(sma);
675 if (copy_semid_to_user (arg.buf, &tbuf, version))
676 return -EFAULT;
677 return id;
678 }
679 default:
680 return -EINVAL;
681 }
682 return err;
683out_unlock:
684 sem_unlock(sma);
685 return err;
686}
687
Kirill Korotaeve3893532006-10-02 02:18:22 -0700688static int semctl_main(struct ipc_namespace *ns, int semid, int semnum,
689 int cmd, int version, union semun arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690{
691 struct sem_array *sma;
692 struct sem* curr;
693 int err;
694 ushort fast_sem_io[SEMMSL_FAST];
695 ushort* sem_io = fast_sem_io;
696 int nsems;
697
Nadia Derbey023a5352007-10-18 23:40:51 -0700698 sma = sem_lock_check(ns, semid);
699 if (IS_ERR(sma))
700 return PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
702 nsems = sma->sem_nsems;
703
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704 err = -EACCES;
705 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO))
706 goto out_unlock;
707
708 err = security_sem_semctl(sma, cmd);
709 if (err)
710 goto out_unlock;
711
712 err = -EACCES;
713 switch (cmd) {
714 case GETALL:
715 {
716 ushort __user *array = arg.array;
717 int i;
718
719 if(nsems > SEMMSL_FAST) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700720 sem_getref_and_unlock(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721
722 sem_io = ipc_alloc(sizeof(ushort)*nsems);
723 if(sem_io == NULL) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700724 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725 return -ENOMEM;
726 }
727
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700728 sem_lock_and_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 if (sma->sem_perm.deleted) {
730 sem_unlock(sma);
731 err = -EIDRM;
732 goto out_free;
733 }
734 }
735
736 for (i = 0; i < sma->sem_nsems; i++)
737 sem_io[i] = sma->sem_base[i].semval;
738 sem_unlock(sma);
739 err = 0;
740 if(copy_to_user(array, sem_io, nsems*sizeof(ushort)))
741 err = -EFAULT;
742 goto out_free;
743 }
744 case SETALL:
745 {
746 int i;
747 struct sem_undo *un;
748
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700749 sem_getref_and_unlock(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750
751 if(nsems > SEMMSL_FAST) {
752 sem_io = ipc_alloc(sizeof(ushort)*nsems);
753 if(sem_io == NULL) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700754 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 return -ENOMEM;
756 }
757 }
758
759 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700760 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 err = -EFAULT;
762 goto out_free;
763 }
764
765 for (i = 0; i < nsems; i++) {
766 if (sem_io[i] > SEMVMX) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700767 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 err = -ERANGE;
769 goto out_free;
770 }
771 }
Pierre Peiffer6ff37972008-04-29 01:00:46 -0700772 sem_lock_and_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 if (sma->sem_perm.deleted) {
774 sem_unlock(sma);
775 err = -EIDRM;
776 goto out_free;
777 }
778
779 for (i = 0; i < nsems; i++)
780 sma->sem_base[i].semval = sem_io[i];
781 for (un = sma->undo; un; un = un->id_next)
782 for (i = 0; i < nsems; i++)
783 un->semadj[i] = 0;
784 sma->sem_ctime = get_seconds();
785 /* maybe some queued-up processes were waiting for this */
786 update_queue(sma);
787 err = 0;
788 goto out_unlock;
789 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */
791 }
792 err = -EINVAL;
793 if(semnum < 0 || semnum >= nsems)
794 goto out_unlock;
795
796 curr = &sma->sem_base[semnum];
797
798 switch (cmd) {
799 case GETVAL:
800 err = curr->semval;
801 goto out_unlock;
802 case GETPID:
803 err = curr->sempid;
804 goto out_unlock;
805 case GETNCNT:
806 err = count_semncnt(sma,semnum);
807 goto out_unlock;
808 case GETZCNT:
809 err = count_semzcnt(sma,semnum);
810 goto out_unlock;
811 case SETVAL:
812 {
813 int val = arg.val;
814 struct sem_undo *un;
815 err = -ERANGE;
816 if (val > SEMVMX || val < 0)
817 goto out_unlock;
818
819 for (un = sma->undo; un; un = un->id_next)
820 un->semadj[semnum] = 0;
821 curr->semval = val;
Pavel Emelyanovb4888932007-10-18 23:40:14 -0700822 curr->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 sma->sem_ctime = get_seconds();
824 /* maybe some queued-up processes were waiting for this */
825 update_queue(sma);
826 err = 0;
827 goto out_unlock;
828 }
829 }
830out_unlock:
831 sem_unlock(sma);
832out_free:
833 if(sem_io != fast_sem_io)
834 ipc_free(sem_io, sizeof(ushort)*nsems);
835 return err;
836}
837
838struct sem_setbuf {
839 uid_t uid;
840 gid_t gid;
841 mode_t mode;
842};
843
844static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version)
845{
846 switch(version) {
847 case IPC_64:
848 {
849 struct semid64_ds tbuf;
850
851 if(copy_from_user(&tbuf, buf, sizeof(tbuf)))
852 return -EFAULT;
853
854 out->uid = tbuf.sem_perm.uid;
855 out->gid = tbuf.sem_perm.gid;
856 out->mode = tbuf.sem_perm.mode;
857
858 return 0;
859 }
860 case IPC_OLD:
861 {
862 struct semid_ds tbuf_old;
863
864 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old)))
865 return -EFAULT;
866
867 out->uid = tbuf_old.sem_perm.uid;
868 out->gid = tbuf_old.sem_perm.gid;
869 out->mode = tbuf_old.sem_perm.mode;
870
871 return 0;
872 }
873 default:
874 return -EINVAL;
875 }
876}
877
Pierre Peiffer522bb2a2008-04-29 01:00:49 -0700878/*
879 * This function handles some semctl commands which require the rw_mutex
880 * to be held in write mode.
881 * NOTE: no locks must be held, the rw_mutex is taken inside this function.
882 */
Kirill Korotaeve3893532006-10-02 02:18:22 -0700883static int semctl_down(struct ipc_namespace *ns, int semid, int semnum,
884 int cmd, int version, union semun arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885{
886 struct sem_array *sma;
887 int err;
Jeff Garzik8e1c0912007-07-17 05:40:59 -0400888 struct sem_setbuf uninitialized_var(setbuf);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 struct kern_ipc_perm *ipcp;
890
891 if(cmd == IPC_SET) {
892 if(copy_semid_from_user (&setbuf, arg.buf, version))
893 return -EFAULT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 }
Pierre Peiffer522bb2a2008-04-29 01:00:49 -0700895 down_write(&sem_ids(ns).rw_mutex);
Nadia Derbey3e148c72007-10-18 23:40:54 -0700896 sma = sem_lock_check_down(ns, semid);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -0700897 if (IS_ERR(sma)) {
898 err = PTR_ERR(sma);
899 goto out_up;
900 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700901
Linus Torvalds1da177e2005-04-16 15:20:36 -0700902 ipcp = &sma->sem_perm;
Steve Grubb073115d2006-04-02 17:07:33 -0400903
904 err = audit_ipc_obj(ipcp);
905 if (err)
906 goto out_unlock;
907
Linda Knippersac032212006-05-16 22:03:48 -0400908 if (cmd == IPC_SET) {
909 err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode);
910 if (err)
911 goto out_unlock;
912 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913 if (current->euid != ipcp->cuid &&
914 current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) {
915 err=-EPERM;
916 goto out_unlock;
917 }
918
919 err = security_sem_semctl(sma, cmd);
920 if (err)
921 goto out_unlock;
922
923 switch(cmd){
924 case IPC_RMID:
Pierre Peiffer01b8b072008-02-08 04:18:57 -0800925 freeary(ns, ipcp);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -0700926 goto out_up;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700927 case IPC_SET:
928 ipcp->uid = setbuf.uid;
929 ipcp->gid = setbuf.gid;
930 ipcp->mode = (ipcp->mode & ~S_IRWXUGO)
931 | (setbuf.mode & S_IRWXUGO);
932 sma->sem_ctime = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 break;
934 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935 err = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700937
938out_unlock:
939 sem_unlock(sma);
Pierre Peiffer522bb2a2008-04-29 01:00:49 -0700940out_up:
941 up_write(&sem_ids(ns).rw_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942 return err;
943}
944
945asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg)
946{
947 int err = -EINVAL;
948 int version;
Kirill Korotaeve3893532006-10-02 02:18:22 -0700949 struct ipc_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950
951 if (semid < 0)
952 return -EINVAL;
953
954 version = ipc_parse_version(&cmd);
Kirill Korotaeve3893532006-10-02 02:18:22 -0700955 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956
957 switch(cmd) {
958 case IPC_INFO:
959 case SEM_INFO:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -0800960 case IPC_STAT:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 case SEM_STAT:
Pierre Peiffer4b9fcb02008-02-08 04:18:56 -0800962 err = semctl_nolock(ns, semid, cmd, version, arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963 return err;
964 case GETALL:
965 case GETVAL:
966 case GETPID:
967 case GETNCNT:
968 case GETZCNT:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 case SETVAL:
970 case SETALL:
Kirill Korotaeve3893532006-10-02 02:18:22 -0700971 err = semctl_main(ns,semid,semnum,cmd,version,arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700972 return err;
973 case IPC_RMID:
974 case IPC_SET:
Kirill Korotaeve3893532006-10-02 02:18:22 -0700975 err = semctl_down(ns,semid,semnum,cmd,version,arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 return err;
977 default:
978 return -EINVAL;
979 }
980}
981
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982/* If the task doesn't already have a undo_list, then allocate one
983 * here. We guarantee there is only one thread using this undo list,
984 * and current is THE ONE
985 *
986 * If this allocation and assignment succeeds, but later
987 * portions of this code fail, there is no need to free the sem_undo_list.
988 * Just let it stay associated with the task, and it'll be freed later
989 * at exit time.
990 *
991 * This can block, so callers must hold no locks.
992 */
993static inline int get_undo_list(struct sem_undo_list **undo_listp)
994{
995 struct sem_undo_list *undo_list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996
997 undo_list = current->sysvsem.undo_list;
998 if (!undo_list) {
Matt Helsley2453a302006-10-02 02:18:25 -0700999 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (undo_list == NULL)
1001 return -ENOMEM;
Ingo Molnar00a5dfd2005-08-05 23:05:27 +02001002 spin_lock_init(&undo_list->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 atomic_set(&undo_list->refcnt, 1);
1004 current->sysvsem.undo_list = undo_list;
1005 }
1006 *undo_listp = undo_list;
1007 return 0;
1008}
1009
1010static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid)
1011{
1012 struct sem_undo **last, *un;
1013
1014 last = &ulp->proc_list;
1015 un = *last;
1016 while(un != NULL) {
1017 if(un->semid==semid)
1018 break;
1019 if(un->semid==-1) {
1020 *last=un->proc_next;
1021 kfree(un);
1022 } else {
1023 last=&un->proc_next;
1024 }
1025 un=*last;
1026 }
1027 return un;
1028}
1029
Kirill Korotaeve3893532006-10-02 02:18:22 -07001030static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 struct sem_array *sma;
1033 struct sem_undo_list *ulp;
1034 struct sem_undo *un, *new;
1035 int nsems;
1036 int error;
1037
1038 error = get_undo_list(&ulp);
1039 if (error)
1040 return ERR_PTR(error);
1041
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001042 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 un = lookup_undo(ulp, semid);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001044 spin_unlock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 if (likely(un!=NULL))
1046 goto out;
1047
1048 /* no undo structure around - allocate one. */
Nadia Derbey023a5352007-10-18 23:40:51 -07001049 sma = sem_lock_check(ns, semid);
1050 if (IS_ERR(sma))
1051 return ERR_PTR(PTR_ERR(sma));
1052
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 nsems = sma->sem_nsems;
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001054 sem_getref_and_unlock(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055
Burman Yan4668edc2006-12-06 20:38:51 -08001056 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057 if (!new) {
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001058 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return ERR_PTR(-ENOMEM);
1060 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061 new->semadj = (short *) &new[1];
1062 new->semid = semid;
1063
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001064 spin_lock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 un = lookup_undo(ulp, semid);
1066 if (un) {
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001067 spin_unlock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001068 kfree(new);
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001069 sem_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 goto out;
1071 }
Pierre Peiffer6ff37972008-04-29 01:00:46 -07001072 sem_lock_and_putref(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073 if (sma->sem_perm.deleted) {
1074 sem_unlock(sma);
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001075 spin_unlock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 kfree(new);
1077 un = ERR_PTR(-EIDRM);
1078 goto out;
1079 }
1080 new->proc_next = ulp->proc_list;
1081 ulp->proc_list = new;
1082 new->id_next = sma->undo;
1083 sma->undo = new;
1084 sem_unlock(sma);
1085 un = new;
Pierre Peifferc530c6a2007-10-18 23:40:55 -07001086 spin_unlock(&ulp->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087out:
1088 return un;
1089}
1090
1091asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops,
1092 unsigned nsops, const struct timespec __user *timeout)
1093{
1094 int error = -EINVAL;
1095 struct sem_array *sma;
1096 struct sembuf fast_sops[SEMOPM_FAST];
1097 struct sembuf* sops = fast_sops, *sop;
1098 struct sem_undo *un;
Manfred Spraulb78755a2005-06-23 00:10:06 -07001099 int undos = 0, alter = 0, max;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100 struct sem_queue queue;
1101 unsigned long jiffies_left = 0;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001102 struct ipc_namespace *ns;
1103
1104 ns = current->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105
1106 if (nsops < 1 || semid < 0)
1107 return -EINVAL;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001108 if (nsops > ns->sc_semopm)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 return -E2BIG;
1110 if(nsops > SEMOPM_FAST) {
1111 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL);
1112 if(sops==NULL)
1113 return -ENOMEM;
1114 }
1115 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) {
1116 error=-EFAULT;
1117 goto out_free;
1118 }
1119 if (timeout) {
1120 struct timespec _timeout;
1121 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) {
1122 error = -EFAULT;
1123 goto out_free;
1124 }
1125 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 ||
1126 _timeout.tv_nsec >= 1000000000L) {
1127 error = -EINVAL;
1128 goto out_free;
1129 }
1130 jiffies_left = timespec_to_jiffies(&_timeout);
1131 }
1132 max = 0;
1133 for (sop = sops; sop < sops + nsops; sop++) {
1134 if (sop->sem_num >= max)
1135 max = sop->sem_num;
1136 if (sop->sem_flg & SEM_UNDO)
Manfred Spraulb78755a2005-06-23 00:10:06 -07001137 undos = 1;
1138 if (sop->sem_op != 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 alter = 1;
1140 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
1142retry_undos:
1143 if (undos) {
Kirill Korotaeve3893532006-10-02 02:18:22 -07001144 un = find_undo(ns, semid);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001145 if (IS_ERR(un)) {
1146 error = PTR_ERR(un);
1147 goto out_free;
1148 }
1149 } else
1150 un = NULL;
1151
Nadia Derbey023a5352007-10-18 23:40:51 -07001152 sma = sem_lock_check(ns, semid);
1153 if (IS_ERR(sma)) {
1154 error = PTR_ERR(sma);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 goto out_free;
Nadia Derbey023a5352007-10-18 23:40:51 -07001156 }
1157
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158 /*
Nadia Derbey023a5352007-10-18 23:40:51 -07001159 * semid identifiers are not unique - find_undo may have
Linus Torvalds1da177e2005-04-16 15:20:36 -07001160 * allocated an undo structure, it was invalidated by an RMID
1161 * and now a new array with received the same id. Check and retry.
1162 */
1163 if (un && un->semid == -1) {
1164 sem_unlock(sma);
1165 goto retry_undos;
1166 }
1167 error = -EFBIG;
1168 if (max >= sma->sem_nsems)
1169 goto out_unlock_free;
1170
1171 error = -EACCES;
1172 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO))
1173 goto out_unlock_free;
1174
1175 error = security_sem_semop(sma, sops, nsops, alter);
1176 if (error)
1177 goto out_unlock_free;
1178
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001179 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 if (error <= 0) {
1181 if (alter && error == 0)
1182 update_queue (sma);
1183 goto out_unlock_free;
1184 }
1185
1186 /* We need to sleep on this operation, so we put the current
1187 * task into the pending queue and go to sleep.
1188 */
1189
1190 queue.sma = sma;
1191 queue.sops = sops;
1192 queue.nsops = nsops;
1193 queue.undo = un;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001194 queue.pid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 queue.id = semid;
1196 queue.alter = alter;
1197 if (alter)
1198 append_to_queue(sma ,&queue);
1199 else
1200 prepend_to_queue(sma ,&queue);
1201
1202 queue.status = -EINTR;
1203 queue.sleeper = current;
1204 current->state = TASK_INTERRUPTIBLE;
1205 sem_unlock(sma);
1206
1207 if (timeout)
1208 jiffies_left = schedule_timeout(jiffies_left);
1209 else
1210 schedule();
1211
1212 error = queue.status;
1213 while(unlikely(error == IN_WAKEUP)) {
1214 cpu_relax();
1215 error = queue.status;
1216 }
1217
1218 if (error != -EINTR) {
1219 /* fast path: update_queue already obtained all requested
1220 * resources */
1221 goto out_free;
1222 }
1223
Kirill Korotaeve3893532006-10-02 02:18:22 -07001224 sma = sem_lock(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001225 if (IS_ERR(sma)) {
Eric Sesterhenn27315c92006-03-26 18:28:38 +02001226 BUG_ON(queue.prev != NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 error = -EIDRM;
1228 goto out_free;
1229 }
1230
1231 /*
1232 * If queue.status != -EINTR we are woken up by another process
1233 */
1234 error = queue.status;
1235 if (error != -EINTR) {
1236 goto out_unlock_free;
1237 }
1238
1239 /*
1240 * If an interrupt occurred we have to clean up the queue
1241 */
1242 if (timeout && jiffies_left == 0)
1243 error = -EAGAIN;
1244 remove_from_queue(sma,&queue);
1245 goto out_unlock_free;
1246
1247out_unlock_free:
1248 sem_unlock(sma);
1249out_free:
1250 if(sops != fast_sops)
1251 kfree(sops);
1252 return error;
1253}
1254
1255asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops)
1256{
1257 return sys_semtimedop(semid, tsops, nsops, NULL);
1258}
1259
1260/* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between
1261 * parent and child tasks.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001262 */
1263
1264int copy_semundo(unsigned long clone_flags, struct task_struct *tsk)
1265{
1266 struct sem_undo_list *undo_list;
1267 int error;
1268
1269 if (clone_flags & CLONE_SYSVSEM) {
1270 error = get_undo_list(&undo_list);
1271 if (error)
1272 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 atomic_inc(&undo_list->refcnt);
1274 tsk->sysvsem.undo_list = undo_list;
1275 } else
1276 tsk->sysvsem.undo_list = NULL;
1277
1278 return 0;
1279}
1280
1281/*
1282 * add semadj values to semaphores, free undo structures.
1283 * undo structures are not freed when semaphore arrays are destroyed
1284 * so some of them may be out of date.
1285 * IMPLEMENTATION NOTE: There is some confusion over whether the
1286 * set of adjustments that needs to be done should be done in an atomic
1287 * manner or not. That is, if we are attempting to decrement the semval
1288 * should we queue up and wait until we can do so legally?
1289 * The original implementation attempted to do this (queue and wait).
1290 * The current implementation does not do so. The POSIX standard
1291 * and SVID should be consulted to determine what behavior is mandated.
1292 */
1293void exit_sem(struct task_struct *tsk)
1294{
1295 struct sem_undo_list *undo_list;
1296 struct sem_undo *u, **up;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001297 struct ipc_namespace *ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298
1299 undo_list = tsk->sysvsem.undo_list;
1300 if (!undo_list)
1301 return;
1302
1303 if (!atomic_dec_and_test(&undo_list->refcnt))
1304 return;
1305
Kirill Korotaeve3893532006-10-02 02:18:22 -07001306 ns = tsk->nsproxy->ipc_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 /* There's no need to hold the semundo list lock, as current
1308 * is the last task exiting for this undo list.
1309 */
1310 for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) {
1311 struct sem_array *sma;
1312 int nsems, i;
1313 struct sem_undo *un, **unp;
1314 int semid;
1315
1316 semid = u->semid;
1317
1318 if(semid == -1)
1319 continue;
Kirill Korotaeve3893532006-10-02 02:18:22 -07001320 sma = sem_lock(ns, semid);
Nadia Derbey023a5352007-10-18 23:40:51 -07001321 if (IS_ERR(sma))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001322 continue;
1323
1324 if (u->semid == -1)
1325 goto next_entry;
1326
Nadia Derbey1b531f22007-10-18 23:40:55 -07001327 BUG_ON(sem_checkid(sma, u->semid));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
1329 /* remove u from the sma->undo list */
1330 for (unp = &sma->undo; (un = *unp); unp = &un->id_next) {
1331 if (u == un)
1332 goto found;
1333 }
1334 printk ("exit_sem undo list error id=%d\n", u->semid);
1335 goto next_entry;
1336found:
1337 *unp = un->id_next;
1338 /* perform adjustments registered in u */
1339 nsems = sma->sem_nsems;
1340 for (i = 0; i < nsems; i++) {
Ingo Molnar5f921ae2006-03-26 01:37:17 -08001341 struct sem * semaphore = &sma->sem_base[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 if (u->semadj[i]) {
Ingo Molnar5f921ae2006-03-26 01:37:17 -08001343 semaphore->semval += u->semadj[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 /*
1345 * Range checks of the new semaphore value,
1346 * not defined by sus:
1347 * - Some unices ignore the undo entirely
1348 * (e.g. HP UX 11i 11.22, Tru64 V5.1)
1349 * - some cap the value (e.g. FreeBSD caps
1350 * at 0, but doesn't enforce SEMVMX)
1351 *
1352 * Linux caps the semaphore value, both at 0
1353 * and at SEMVMX.
1354 *
1355 * Manfred <manfred@colorfullife.com>
1356 */
Ingo Molnar5f921ae2006-03-26 01:37:17 -08001357 if (semaphore->semval < 0)
1358 semaphore->semval = 0;
1359 if (semaphore->semval > SEMVMX)
1360 semaphore->semval = SEMVMX;
Pavel Emelyanovb4888932007-10-18 23:40:14 -07001361 semaphore->sempid = task_tgid_vnr(current);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 }
1363 }
1364 sma->sem_otime = get_seconds();
1365 /* maybe some queued-up processes were waiting for this */
1366 update_queue(sma);
1367next_entry:
1368 sem_unlock(sma);
1369 }
1370 kfree(undo_list);
1371}
1372
1373#ifdef CONFIG_PROC_FS
Mike Waychison19b49462005-09-06 15:17:10 -07001374static int sysvipc_sem_proc_show(struct seq_file *s, void *it)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
Mike Waychison19b49462005-09-06 15:17:10 -07001376 struct sem_array *sma = it;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377
Mike Waychison19b49462005-09-06 15:17:10 -07001378 return seq_printf(s,
1379 "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n",
1380 sma->sem_perm.key,
Nadia Derbey7ca7e562007-10-18 23:40:48 -07001381 sma->sem_perm.id,
Mike Waychison19b49462005-09-06 15:17:10 -07001382 sma->sem_perm.mode,
1383 sma->sem_nsems,
1384 sma->sem_perm.uid,
1385 sma->sem_perm.gid,
1386 sma->sem_perm.cuid,
1387 sma->sem_perm.cgid,
1388 sma->sem_otime,
1389 sma->sem_ctime);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391#endif