]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
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 | |
624dffcb | 59 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 60 | * Enforced range limit on SEM_UNDO |
046c6884 | 61 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
62 | * Lockless wakeup |
63 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
64 | * |
65 | * support for audit of ipc object properties and permission changes | |
66 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
67 | * |
68 | * namespaces support | |
69 | * OpenVZ, SWsoft Inc. | |
70 | * Pavel Emelianov <xemul@openvz.org> | |
1da177e4 LT |
71 | */ |
72 | ||
1da177e4 LT |
73 | #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> | |
1da177e4 LT |
78 | #include <linux/security.h> |
79 | #include <linux/syscalls.h> | |
80 | #include <linux/audit.h> | |
c59ede7b | 81 | #include <linux/capability.h> |
19b4946c | 82 | #include <linux/seq_file.h> |
3e148c79 | 83 | #include <linux/rwsem.h> |
e3893534 | 84 | #include <linux/nsproxy.h> |
ae5e1b22 | 85 | #include <linux/ipc_namespace.h> |
5f921ae9 | 86 | |
1da177e4 LT |
87 | #include <asm/uaccess.h> |
88 | #include "util.h" | |
89 | ||
ed2ddbf8 | 90 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 91 | |
e3893534 | 92 | #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm) |
1b531f21 | 93 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 94 | |
7748dbfa | 95 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 96 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 97 | #ifdef CONFIG_PROC_FS |
19b4946c | 98 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
99 | #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 | ||
e3893534 KK |
113 | #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] | |
117 | ||
ed2ddbf8 | 118 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 119 | { |
e3893534 KK |
120 | ns->sc_semmsl = SEMMSL; |
121 | ns->sc_semmns = SEMMNS; | |
122 | ns->sc_semopm = SEMOPM; | |
123 | ns->sc_semmni = SEMMNI; | |
124 | ns->used_sems = 0; | |
ed2ddbf8 | 125 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
126 | } |
127 | ||
ae5e1b22 | 128 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
129 | void sem_exit_ns(struct ipc_namespace *ns) |
130 | { | |
01b8b07a | 131 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 132 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 133 | } |
ae5e1b22 | 134 | #endif |
1da177e4 LT |
135 | |
136 | void __init sem_init (void) | |
137 | { | |
ed2ddbf8 | 138 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
139 | ipc_init_proc_interface("sysvipc/sem", |
140 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 141 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
142 | } |
143 | ||
3e148c79 ND |
144 | /* |
145 | * sem_lock_(check_) routines are called in the paths where the rw_mutex | |
146 | * is not held. | |
147 | */ | |
023a5355 ND |
148 | static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id) |
149 | { | |
03f02c76 ND |
150 | struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id); |
151 | ||
b1ed88b4 PP |
152 | if (IS_ERR(ipcp)) |
153 | return (struct sem_array *)ipcp; | |
154 | ||
03f02c76 | 155 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
156 | } |
157 | ||
158 | static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns, | |
159 | int id) | |
160 | { | |
03f02c76 ND |
161 | struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id); |
162 | ||
b1ed88b4 PP |
163 | if (IS_ERR(ipcp)) |
164 | return (struct sem_array *)ipcp; | |
165 | ||
03f02c76 | 166 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
167 | } |
168 | ||
6ff37972 PP |
169 | static inline void sem_lock_and_putref(struct sem_array *sma) |
170 | { | |
171 | ipc_lock_by_ptr(&sma->sem_perm); | |
172 | ipc_rcu_putref(sma); | |
173 | } | |
174 | ||
175 | static inline void sem_getref_and_unlock(struct sem_array *sma) | |
176 | { | |
177 | ipc_rcu_getref(sma); | |
178 | ipc_unlock(&(sma)->sem_perm); | |
179 | } | |
180 | ||
181 | static inline void sem_putref(struct sem_array *sma) | |
182 | { | |
183 | ipc_lock_by_ptr(&sma->sem_perm); | |
184 | ipc_rcu_putref(sma); | |
185 | ipc_unlock(&(sma)->sem_perm); | |
186 | } | |
187 | ||
7ca7e564 ND |
188 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
189 | { | |
190 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
191 | } | |
192 | ||
1da177e4 LT |
193 | /* |
194 | * Lockless wakeup algorithm: | |
195 | * Without the check/retry algorithm a lockless wakeup is possible: | |
196 | * - queue.status is initialized to -EINTR before blocking. | |
197 | * - wakeup is performed by | |
198 | * * unlinking the queue entry from sma->sem_pending | |
199 | * * setting queue.status to IN_WAKEUP | |
200 | * This is the notification for the blocked thread that a | |
201 | * result value is imminent. | |
202 | * * call wake_up_process | |
203 | * * set queue.status to the final value. | |
204 | * - the previously blocked thread checks queue.status: | |
205 | * * if it's IN_WAKEUP, then it must wait until the value changes | |
206 | * * if it's not -EINTR, then the operation was completed by | |
207 | * update_queue. semtimedop can return queue.status without | |
5f921ae9 | 208 | * performing any operation on the sem array. |
1da177e4 LT |
209 | * * otherwise it must acquire the spinlock and check what's up. |
210 | * | |
211 | * The two-stage algorithm is necessary to protect against the following | |
212 | * races: | |
213 | * - if queue.status is set after wake_up_process, then the woken up idle | |
214 | * thread could race forward and try (and fail) to acquire sma->lock | |
215 | * before update_queue had a chance to set queue.status | |
216 | * - if queue.status is written before wake_up_process and if the | |
217 | * blocked process is woken up by a signal between writing | |
218 | * queue.status and the wake_up_process, then the woken up | |
219 | * process could return from semtimedop and die by calling | |
220 | * sys_exit before wake_up_process is called. Then wake_up_process | |
221 | * will oops, because the task structure is already invalid. | |
222 | * (yes, this happened on s390 with sysv msg). | |
223 | * | |
224 | */ | |
225 | #define IN_WAKEUP 1 | |
226 | ||
f4566f04 ND |
227 | /** |
228 | * newary - Create a new semaphore set | |
229 | * @ns: namespace | |
230 | * @params: ptr to the structure that contains key, semflg and nsems | |
231 | * | |
3e148c79 | 232 | * Called with sem_ids.rw_mutex held (as a writer) |
f4566f04 ND |
233 | */ |
234 | ||
7748dbfa | 235 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
236 | { |
237 | int id; | |
238 | int retval; | |
239 | struct sem_array *sma; | |
240 | int size; | |
7748dbfa ND |
241 | key_t key = params->key; |
242 | int nsems = params->u.nsems; | |
243 | int semflg = params->flg; | |
1da177e4 LT |
244 | |
245 | if (!nsems) | |
246 | return -EINVAL; | |
e3893534 | 247 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
248 | return -ENOSPC; |
249 | ||
250 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
251 | sma = ipc_rcu_alloc(size); | |
252 | if (!sma) { | |
253 | return -ENOMEM; | |
254 | } | |
255 | memset (sma, 0, size); | |
256 | ||
257 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
258 | sma->sem_perm.key = key; | |
259 | ||
260 | sma->sem_perm.security = NULL; | |
261 | retval = security_sem_alloc(sma); | |
262 | if (retval) { | |
263 | ipc_rcu_putref(sma); | |
264 | return retval; | |
265 | } | |
266 | ||
e3893534 | 267 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 268 | if (id < 0) { |
1da177e4 LT |
269 | security_sem_free(sma); |
270 | ipc_rcu_putref(sma); | |
283bb7fa | 271 | return id; |
1da177e4 | 272 | } |
e3893534 | 273 | ns->used_sems += nsems; |
1da177e4 LT |
274 | |
275 | sma->sem_base = (struct sem *) &sma[1]; | |
a1193f8e | 276 | INIT_LIST_HEAD(&sma->sem_pending); |
4daa28f6 | 277 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
278 | sma->sem_nsems = nsems; |
279 | sma->sem_ctime = get_seconds(); | |
280 | sem_unlock(sma); | |
281 | ||
7ca7e564 | 282 | return sma->sem_perm.id; |
1da177e4 LT |
283 | } |
284 | ||
7748dbfa | 285 | |
f4566f04 | 286 | /* |
3e148c79 | 287 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 288 | */ |
03f02c76 | 289 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 290 | { |
03f02c76 ND |
291 | struct sem_array *sma; |
292 | ||
293 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
294 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
295 | } |
296 | ||
f4566f04 | 297 | /* |
3e148c79 | 298 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 299 | */ |
03f02c76 ND |
300 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
301 | struct ipc_params *params) | |
7748dbfa | 302 | { |
03f02c76 ND |
303 | struct sem_array *sma; |
304 | ||
305 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
306 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
307 | return -EINVAL; |
308 | ||
309 | return 0; | |
310 | } | |
311 | ||
d5460c99 | 312 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 313 | { |
e3893534 | 314 | struct ipc_namespace *ns; |
7748dbfa ND |
315 | struct ipc_ops sem_ops; |
316 | struct ipc_params sem_params; | |
e3893534 KK |
317 | |
318 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 319 | |
e3893534 | 320 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 321 | return -EINVAL; |
7ca7e564 | 322 | |
7748dbfa ND |
323 | sem_ops.getnew = newary; |
324 | sem_ops.associate = sem_security; | |
325 | sem_ops.more_checks = sem_more_checks; | |
326 | ||
327 | sem_params.key = key; | |
328 | sem_params.flg = semflg; | |
329 | sem_params.u.nsems = nsems; | |
1da177e4 | 330 | |
7748dbfa | 331 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
332 | } |
333 | ||
1da177e4 LT |
334 | /* |
335 | * Determine whether a sequence of semaphore operations would succeed | |
336 | * all at once. Return 0 if yes, 1 if need to sleep, else return error code. | |
337 | */ | |
338 | ||
339 | static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, | |
340 | int nsops, struct sem_undo *un, int pid) | |
341 | { | |
342 | int result, sem_op; | |
343 | struct sembuf *sop; | |
344 | struct sem * curr; | |
345 | ||
346 | for (sop = sops; sop < sops + nsops; sop++) { | |
347 | curr = sma->sem_base + sop->sem_num; | |
348 | sem_op = sop->sem_op; | |
349 | result = curr->semval; | |
350 | ||
351 | if (!sem_op && result) | |
352 | goto would_block; | |
353 | ||
354 | result += sem_op; | |
355 | if (result < 0) | |
356 | goto would_block; | |
357 | if (result > SEMVMX) | |
358 | goto out_of_range; | |
359 | if (sop->sem_flg & SEM_UNDO) { | |
360 | int undo = un->semadj[sop->sem_num] - sem_op; | |
361 | /* | |
362 | * Exceeding the undo range is an error. | |
363 | */ | |
364 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
365 | goto out_of_range; | |
366 | } | |
367 | curr->semval = result; | |
368 | } | |
369 | ||
370 | sop--; | |
371 | while (sop >= sops) { | |
372 | sma->sem_base[sop->sem_num].sempid = pid; | |
373 | if (sop->sem_flg & SEM_UNDO) | |
374 | un->semadj[sop->sem_num] -= sop->sem_op; | |
375 | sop--; | |
376 | } | |
377 | ||
378 | sma->sem_otime = get_seconds(); | |
379 | return 0; | |
380 | ||
381 | out_of_range: | |
382 | result = -ERANGE; | |
383 | goto undo; | |
384 | ||
385 | would_block: | |
386 | if (sop->sem_flg & IPC_NOWAIT) | |
387 | result = -EAGAIN; | |
388 | else | |
389 | result = 1; | |
390 | ||
391 | undo: | |
392 | sop--; | |
393 | while (sop >= sops) { | |
394 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; | |
395 | sop--; | |
396 | } | |
397 | ||
398 | return result; | |
399 | } | |
400 | ||
d4212093 NP |
401 | /* |
402 | * Wake up a process waiting on the sem queue with a given error. | |
403 | * The queue is invalid (may not be accessed) after the function returns. | |
404 | */ | |
405 | static void wake_up_sem_queue(struct sem_queue *q, int error) | |
406 | { | |
407 | /* | |
408 | * Hold preempt off so that we don't get preempted and have the | |
409 | * wakee busy-wait until we're scheduled back on. We're holding | |
410 | * locks here so it may not strictly be needed, however if the | |
411 | * locks become preemptible then this prevents such a problem. | |
412 | */ | |
413 | preempt_disable(); | |
414 | q->status = IN_WAKEUP; | |
415 | wake_up_process(q->sleeper); | |
416 | /* hands-off: q can disappear immediately after writing q->status. */ | |
417 | smp_wmb(); | |
418 | q->status = error; | |
419 | preempt_enable(); | |
420 | } | |
421 | ||
1da177e4 LT |
422 | /* Go through the pending queue for the indicated semaphore |
423 | * looking for tasks that can be completed. | |
424 | */ | |
425 | static void update_queue (struct sem_array * sma) | |
426 | { | |
9cad200c NP |
427 | struct sem_queue *q, *tq; |
428 | ||
429 | again: | |
430 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { | |
431 | int error; | |
432 | int alter; | |
1da177e4 | 433 | |
1da177e4 LT |
434 | error = try_atomic_semop(sma, q->sops, q->nsops, |
435 | q->undo, q->pid); | |
436 | ||
437 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
438 | if (error > 0) |
439 | continue; | |
440 | ||
441 | list_del(&q->list); | |
442 | ||
443 | /* | |
444 | * The next operation that must be checked depends on the type | |
445 | * of the completed operation: | |
446 | * - if the operation modified the array, then restart from the | |
447 | * head of the queue and check for threads that might be | |
b6e90822 | 448 | * waiting for the new semaphore values. |
9cad200c NP |
449 | * - if the operation didn't modify the array, then just |
450 | * continue. | |
451 | */ | |
452 | alter = q->alter; | |
d4212093 | 453 | wake_up_sem_queue(q, error); |
b6e90822 | 454 | if (alter && !error) |
9cad200c | 455 | goto again; |
1da177e4 LT |
456 | } |
457 | } | |
458 | ||
459 | /* The following counts are associated to each semaphore: | |
460 | * semncnt number of tasks waiting on semval being nonzero | |
461 | * semzcnt number of tasks waiting on semval being zero | |
462 | * This model assumes that a task waits on exactly one semaphore. | |
463 | * Since semaphore operations are to be performed atomically, tasks actually | |
464 | * wait on a whole sequence of semaphores simultaneously. | |
465 | * The counts we return here are a rough approximation, but still | |
466 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
467 | */ | |
468 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
469 | { | |
470 | int semncnt; | |
471 | struct sem_queue * q; | |
472 | ||
473 | semncnt = 0; | |
a1193f8e | 474 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
475 | struct sembuf * sops = q->sops; |
476 | int nsops = q->nsops; | |
477 | int i; | |
478 | for (i = 0; i < nsops; i++) | |
479 | if (sops[i].sem_num == semnum | |
480 | && (sops[i].sem_op < 0) | |
481 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
482 | semncnt++; | |
483 | } | |
484 | return semncnt; | |
485 | } | |
a1193f8e | 486 | |
1da177e4 LT |
487 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
488 | { | |
489 | int semzcnt; | |
490 | struct sem_queue * q; | |
491 | ||
492 | semzcnt = 0; | |
a1193f8e | 493 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
494 | struct sembuf * sops = q->sops; |
495 | int nsops = q->nsops; | |
496 | int i; | |
497 | for (i = 0; i < nsops; i++) | |
498 | if (sops[i].sem_num == semnum | |
499 | && (sops[i].sem_op == 0) | |
500 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
501 | semzcnt++; | |
502 | } | |
503 | return semzcnt; | |
504 | } | |
505 | ||
6d97e234 | 506 | static void free_un(struct rcu_head *head) |
380af1b3 MS |
507 | { |
508 | struct sem_undo *un = container_of(head, struct sem_undo, rcu); | |
509 | kfree(un); | |
510 | } | |
511 | ||
3e148c79 ND |
512 | /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked |
513 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex | |
514 | * remains locked on exit. | |
1da177e4 | 515 | */ |
01b8b07a | 516 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 517 | { |
380af1b3 MS |
518 | struct sem_undo *un, *tu; |
519 | struct sem_queue *q, *tq; | |
01b8b07a | 520 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 | 521 | |
380af1b3 | 522 | /* Free the existing undo structures for this semaphore set. */ |
4daa28f6 | 523 | assert_spin_locked(&sma->sem_perm.lock); |
380af1b3 MS |
524 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
525 | list_del(&un->list_id); | |
526 | spin_lock(&un->ulp->lock); | |
1da177e4 | 527 | un->semid = -1; |
380af1b3 MS |
528 | list_del_rcu(&un->list_proc); |
529 | spin_unlock(&un->ulp->lock); | |
530 | call_rcu(&un->rcu, free_un); | |
531 | } | |
1da177e4 LT |
532 | |
533 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
380af1b3 | 534 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { |
a1193f8e MS |
535 | list_del(&q->list); |
536 | ||
d4212093 | 537 | wake_up_sem_queue(q, -EIDRM); |
1da177e4 LT |
538 | } |
539 | ||
7ca7e564 ND |
540 | /* Remove the semaphore set from the IDR */ |
541 | sem_rmid(ns, sma); | |
1da177e4 LT |
542 | sem_unlock(sma); |
543 | ||
e3893534 | 544 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
545 | security_sem_free(sma); |
546 | ipc_rcu_putref(sma); | |
547 | } | |
548 | ||
549 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
550 | { | |
551 | switch(version) { | |
552 | case IPC_64: | |
553 | return copy_to_user(buf, in, sizeof(*in)); | |
554 | case IPC_OLD: | |
555 | { | |
556 | struct semid_ds out; | |
557 | ||
558 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); | |
559 | ||
560 | out.sem_otime = in->sem_otime; | |
561 | out.sem_ctime = in->sem_ctime; | |
562 | out.sem_nsems = in->sem_nsems; | |
563 | ||
564 | return copy_to_user(buf, &out, sizeof(out)); | |
565 | } | |
566 | default: | |
567 | return -EINVAL; | |
568 | } | |
569 | } | |
570 | ||
4b9fcb0e PP |
571 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
572 | int cmd, int version, union semun arg) | |
1da177e4 LT |
573 | { |
574 | int err = -EINVAL; | |
575 | struct sem_array *sma; | |
576 | ||
577 | switch(cmd) { | |
578 | case IPC_INFO: | |
579 | case SEM_INFO: | |
580 | { | |
581 | struct seminfo seminfo; | |
582 | int max_id; | |
583 | ||
584 | err = security_sem_semctl(NULL, cmd); | |
585 | if (err) | |
586 | return err; | |
587 | ||
588 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
589 | seminfo.semmni = ns->sc_semmni; |
590 | seminfo.semmns = ns->sc_semmns; | |
591 | seminfo.semmsl = ns->sc_semmsl; | |
592 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
593 | seminfo.semvmx = SEMVMX; |
594 | seminfo.semmnu = SEMMNU; | |
595 | seminfo.semmap = SEMMAP; | |
596 | seminfo.semume = SEMUME; | |
3e148c79 | 597 | down_read(&sem_ids(ns).rw_mutex); |
1da177e4 | 598 | if (cmd == SEM_INFO) { |
e3893534 KK |
599 | seminfo.semusz = sem_ids(ns).in_use; |
600 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
601 | } else { |
602 | seminfo.semusz = SEMUSZ; | |
603 | seminfo.semaem = SEMAEM; | |
604 | } | |
7ca7e564 | 605 | max_id = ipc_get_maxid(&sem_ids(ns)); |
3e148c79 | 606 | up_read(&sem_ids(ns).rw_mutex); |
1da177e4 LT |
607 | if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) |
608 | return -EFAULT; | |
609 | return (max_id < 0) ? 0: max_id; | |
610 | } | |
4b9fcb0e | 611 | case IPC_STAT: |
1da177e4 LT |
612 | case SEM_STAT: |
613 | { | |
614 | struct semid64_ds tbuf; | |
615 | int id; | |
616 | ||
4b9fcb0e PP |
617 | if (cmd == SEM_STAT) { |
618 | sma = sem_lock(ns, semid); | |
619 | if (IS_ERR(sma)) | |
620 | return PTR_ERR(sma); | |
621 | id = sma->sem_perm.id; | |
622 | } else { | |
623 | sma = sem_lock_check(ns, semid); | |
624 | if (IS_ERR(sma)) | |
625 | return PTR_ERR(sma); | |
626 | id = 0; | |
627 | } | |
1da177e4 LT |
628 | |
629 | err = -EACCES; | |
630 | if (ipcperms (&sma->sem_perm, S_IRUGO)) | |
631 | goto out_unlock; | |
632 | ||
633 | err = security_sem_semctl(sma, cmd); | |
634 | if (err) | |
635 | goto out_unlock; | |
636 | ||
023a5355 ND |
637 | memset(&tbuf, 0, sizeof(tbuf)); |
638 | ||
1da177e4 LT |
639 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
640 | tbuf.sem_otime = sma->sem_otime; | |
641 | tbuf.sem_ctime = sma->sem_ctime; | |
642 | tbuf.sem_nsems = sma->sem_nsems; | |
643 | sem_unlock(sma); | |
644 | if (copy_semid_to_user (arg.buf, &tbuf, version)) | |
645 | return -EFAULT; | |
646 | return id; | |
647 | } | |
648 | default: | |
649 | return -EINVAL; | |
650 | } | |
651 | return err; | |
652 | out_unlock: | |
653 | sem_unlock(sma); | |
654 | return err; | |
655 | } | |
656 | ||
e3893534 KK |
657 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
658 | int cmd, int version, union semun arg) | |
1da177e4 LT |
659 | { |
660 | struct sem_array *sma; | |
661 | struct sem* curr; | |
662 | int err; | |
663 | ushort fast_sem_io[SEMMSL_FAST]; | |
664 | ushort* sem_io = fast_sem_io; | |
665 | int nsems; | |
666 | ||
023a5355 ND |
667 | sma = sem_lock_check(ns, semid); |
668 | if (IS_ERR(sma)) | |
669 | return PTR_ERR(sma); | |
1da177e4 LT |
670 | |
671 | nsems = sma->sem_nsems; | |
672 | ||
1da177e4 LT |
673 | err = -EACCES; |
674 | if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) | |
675 | goto out_unlock; | |
676 | ||
677 | err = security_sem_semctl(sma, cmd); | |
678 | if (err) | |
679 | goto out_unlock; | |
680 | ||
681 | err = -EACCES; | |
682 | switch (cmd) { | |
683 | case GETALL: | |
684 | { | |
685 | ushort __user *array = arg.array; | |
686 | int i; | |
687 | ||
688 | if(nsems > SEMMSL_FAST) { | |
6ff37972 | 689 | sem_getref_and_unlock(sma); |
1da177e4 LT |
690 | |
691 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
692 | if(sem_io == NULL) { | |
6ff37972 | 693 | sem_putref(sma); |
1da177e4 LT |
694 | return -ENOMEM; |
695 | } | |
696 | ||
6ff37972 | 697 | sem_lock_and_putref(sma); |
1da177e4 LT |
698 | if (sma->sem_perm.deleted) { |
699 | sem_unlock(sma); | |
700 | err = -EIDRM; | |
701 | goto out_free; | |
702 | } | |
703 | } | |
704 | ||
705 | for (i = 0; i < sma->sem_nsems; i++) | |
706 | sem_io[i] = sma->sem_base[i].semval; | |
707 | sem_unlock(sma); | |
708 | err = 0; | |
709 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
710 | err = -EFAULT; | |
711 | goto out_free; | |
712 | } | |
713 | case SETALL: | |
714 | { | |
715 | int i; | |
716 | struct sem_undo *un; | |
717 | ||
6ff37972 | 718 | sem_getref_and_unlock(sma); |
1da177e4 LT |
719 | |
720 | if(nsems > SEMMSL_FAST) { | |
721 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
722 | if(sem_io == NULL) { | |
6ff37972 | 723 | sem_putref(sma); |
1da177e4 LT |
724 | return -ENOMEM; |
725 | } | |
726 | } | |
727 | ||
728 | if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { | |
6ff37972 | 729 | sem_putref(sma); |
1da177e4 LT |
730 | err = -EFAULT; |
731 | goto out_free; | |
732 | } | |
733 | ||
734 | for (i = 0; i < nsems; i++) { | |
735 | if (sem_io[i] > SEMVMX) { | |
6ff37972 | 736 | sem_putref(sma); |
1da177e4 LT |
737 | err = -ERANGE; |
738 | goto out_free; | |
739 | } | |
740 | } | |
6ff37972 | 741 | sem_lock_and_putref(sma); |
1da177e4 LT |
742 | if (sma->sem_perm.deleted) { |
743 | sem_unlock(sma); | |
744 | err = -EIDRM; | |
745 | goto out_free; | |
746 | } | |
747 | ||
748 | for (i = 0; i < nsems; i++) | |
749 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 MS |
750 | |
751 | assert_spin_locked(&sma->sem_perm.lock); | |
752 | list_for_each_entry(un, &sma->list_id, list_id) { | |
1da177e4 LT |
753 | for (i = 0; i < nsems; i++) |
754 | un->semadj[i] = 0; | |
4daa28f6 | 755 | } |
1da177e4 LT |
756 | sma->sem_ctime = get_seconds(); |
757 | /* maybe some queued-up processes were waiting for this */ | |
758 | update_queue(sma); | |
759 | err = 0; | |
760 | goto out_unlock; | |
761 | } | |
1da177e4 LT |
762 | /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ |
763 | } | |
764 | err = -EINVAL; | |
765 | if(semnum < 0 || semnum >= nsems) | |
766 | goto out_unlock; | |
767 | ||
768 | curr = &sma->sem_base[semnum]; | |
769 | ||
770 | switch (cmd) { | |
771 | case GETVAL: | |
772 | err = curr->semval; | |
773 | goto out_unlock; | |
774 | case GETPID: | |
775 | err = curr->sempid; | |
776 | goto out_unlock; | |
777 | case GETNCNT: | |
778 | err = count_semncnt(sma,semnum); | |
779 | goto out_unlock; | |
780 | case GETZCNT: | |
781 | err = count_semzcnt(sma,semnum); | |
782 | goto out_unlock; | |
783 | case SETVAL: | |
784 | { | |
785 | int val = arg.val; | |
786 | struct sem_undo *un; | |
4daa28f6 | 787 | |
1da177e4 LT |
788 | err = -ERANGE; |
789 | if (val > SEMVMX || val < 0) | |
790 | goto out_unlock; | |
791 | ||
4daa28f6 MS |
792 | assert_spin_locked(&sma->sem_perm.lock); |
793 | list_for_each_entry(un, &sma->list_id, list_id) | |
1da177e4 | 794 | un->semadj[semnum] = 0; |
4daa28f6 | 795 | |
1da177e4 | 796 | curr->semval = val; |
b488893a | 797 | curr->sempid = task_tgid_vnr(current); |
1da177e4 LT |
798 | sma->sem_ctime = get_seconds(); |
799 | /* maybe some queued-up processes were waiting for this */ | |
800 | update_queue(sma); | |
801 | err = 0; | |
802 | goto out_unlock; | |
803 | } | |
804 | } | |
805 | out_unlock: | |
806 | sem_unlock(sma); | |
807 | out_free: | |
808 | if(sem_io != fast_sem_io) | |
809 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
810 | return err; | |
811 | } | |
812 | ||
016d7132 PP |
813 | static inline unsigned long |
814 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
815 | { |
816 | switch(version) { | |
817 | case IPC_64: | |
016d7132 | 818 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 819 | return -EFAULT; |
1da177e4 | 820 | return 0; |
1da177e4 LT |
821 | case IPC_OLD: |
822 | { | |
823 | struct semid_ds tbuf_old; | |
824 | ||
825 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
826 | return -EFAULT; | |
827 | ||
016d7132 PP |
828 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
829 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
830 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
831 | |
832 | return 0; | |
833 | } | |
834 | default: | |
835 | return -EINVAL; | |
836 | } | |
837 | } | |
838 | ||
522bb2a2 PP |
839 | /* |
840 | * This function handles some semctl commands which require the rw_mutex | |
841 | * to be held in write mode. | |
842 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
843 | */ | |
21a4826a PP |
844 | static int semctl_down(struct ipc_namespace *ns, int semid, |
845 | int cmd, int version, union semun arg) | |
1da177e4 LT |
846 | { |
847 | struct sem_array *sma; | |
848 | int err; | |
016d7132 | 849 | struct semid64_ds semid64; |
1da177e4 LT |
850 | struct kern_ipc_perm *ipcp; |
851 | ||
852 | if(cmd == IPC_SET) { | |
016d7132 | 853 | if (copy_semid_from_user(&semid64, arg.buf, version)) |
1da177e4 | 854 | return -EFAULT; |
1da177e4 | 855 | } |
073115d6 | 856 | |
a5f75e7f PP |
857 | ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0); |
858 | if (IS_ERR(ipcp)) | |
859 | return PTR_ERR(ipcp); | |
073115d6 | 860 | |
a5f75e7f | 861 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
862 | |
863 | err = security_sem_semctl(sma, cmd); | |
864 | if (err) | |
865 | goto out_unlock; | |
866 | ||
867 | switch(cmd){ | |
868 | case IPC_RMID: | |
01b8b07a | 869 | freeary(ns, ipcp); |
522bb2a2 | 870 | goto out_up; |
1da177e4 | 871 | case IPC_SET: |
8f4a3809 | 872 | ipc_update_perm(&semid64.sem_perm, ipcp); |
1da177e4 | 873 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
874 | break; |
875 | default: | |
1da177e4 | 876 | err = -EINVAL; |
1da177e4 | 877 | } |
1da177e4 LT |
878 | |
879 | out_unlock: | |
880 | sem_unlock(sma); | |
522bb2a2 PP |
881 | out_up: |
882 | up_write(&sem_ids(ns).rw_mutex); | |
1da177e4 LT |
883 | return err; |
884 | } | |
885 | ||
6673e0c3 | 886 | SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg) |
1da177e4 LT |
887 | { |
888 | int err = -EINVAL; | |
889 | int version; | |
e3893534 | 890 | struct ipc_namespace *ns; |
1da177e4 LT |
891 | |
892 | if (semid < 0) | |
893 | return -EINVAL; | |
894 | ||
895 | version = ipc_parse_version(&cmd); | |
e3893534 | 896 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
897 | |
898 | switch(cmd) { | |
899 | case IPC_INFO: | |
900 | case SEM_INFO: | |
4b9fcb0e | 901 | case IPC_STAT: |
1da177e4 | 902 | case SEM_STAT: |
4b9fcb0e | 903 | err = semctl_nolock(ns, semid, cmd, version, arg); |
1da177e4 LT |
904 | return err; |
905 | case GETALL: | |
906 | case GETVAL: | |
907 | case GETPID: | |
908 | case GETNCNT: | |
909 | case GETZCNT: | |
1da177e4 LT |
910 | case SETVAL: |
911 | case SETALL: | |
e3893534 | 912 | err = semctl_main(ns,semid,semnum,cmd,version,arg); |
1da177e4 LT |
913 | return err; |
914 | case IPC_RMID: | |
915 | case IPC_SET: | |
21a4826a | 916 | err = semctl_down(ns, semid, cmd, version, arg); |
1da177e4 LT |
917 | return err; |
918 | default: | |
919 | return -EINVAL; | |
920 | } | |
921 | } | |
6673e0c3 HC |
922 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
923 | asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg) | |
924 | { | |
925 | return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg); | |
926 | } | |
927 | SYSCALL_ALIAS(sys_semctl, SyS_semctl); | |
928 | #endif | |
1da177e4 | 929 | |
1da177e4 LT |
930 | /* If the task doesn't already have a undo_list, then allocate one |
931 | * here. We guarantee there is only one thread using this undo list, | |
932 | * and current is THE ONE | |
933 | * | |
934 | * If this allocation and assignment succeeds, but later | |
935 | * portions of this code fail, there is no need to free the sem_undo_list. | |
936 | * Just let it stay associated with the task, and it'll be freed later | |
937 | * at exit time. | |
938 | * | |
939 | * This can block, so callers must hold no locks. | |
940 | */ | |
941 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
942 | { | |
943 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
944 | |
945 | undo_list = current->sysvsem.undo_list; | |
946 | if (!undo_list) { | |
2453a306 | 947 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
948 | if (undo_list == NULL) |
949 | return -ENOMEM; | |
00a5dfdb | 950 | spin_lock_init(&undo_list->lock); |
1da177e4 | 951 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
952 | INIT_LIST_HEAD(&undo_list->list_proc); |
953 | ||
1da177e4 LT |
954 | current->sysvsem.undo_list = undo_list; |
955 | } | |
956 | *undo_listp = undo_list; | |
957 | return 0; | |
958 | } | |
959 | ||
bf17bb71 | 960 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 961 | { |
bf17bb71 | 962 | struct sem_undo *un; |
4daa28f6 | 963 | |
bf17bb71 NP |
964 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
965 | if (un->semid == semid) | |
966 | return un; | |
1da177e4 | 967 | } |
4daa28f6 | 968 | return NULL; |
1da177e4 LT |
969 | } |
970 | ||
bf17bb71 NP |
971 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
972 | { | |
973 | struct sem_undo *un; | |
974 | ||
975 | assert_spin_locked(&ulp->lock); | |
976 | ||
977 | un = __lookup_undo(ulp, semid); | |
978 | if (un) { | |
979 | list_del_rcu(&un->list_proc); | |
980 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
981 | } | |
982 | return un; | |
983 | } | |
984 | ||
4daa28f6 MS |
985 | /** |
986 | * find_alloc_undo - Lookup (and if not present create) undo array | |
987 | * @ns: namespace | |
988 | * @semid: semaphore array id | |
989 | * | |
990 | * The function looks up (and if not present creates) the undo structure. | |
991 | * The size of the undo structure depends on the size of the semaphore | |
992 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
993 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
994 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
995 | */ |
996 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
997 | { |
998 | struct sem_array *sma; | |
999 | struct sem_undo_list *ulp; | |
1000 | struct sem_undo *un, *new; | |
1001 | int nsems; | |
1002 | int error; | |
1003 | ||
1004 | error = get_undo_list(&ulp); | |
1005 | if (error) | |
1006 | return ERR_PTR(error); | |
1007 | ||
380af1b3 | 1008 | rcu_read_lock(); |
c530c6ac | 1009 | spin_lock(&ulp->lock); |
1da177e4 | 1010 | un = lookup_undo(ulp, semid); |
c530c6ac | 1011 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1012 | if (likely(un!=NULL)) |
1013 | goto out; | |
380af1b3 | 1014 | rcu_read_unlock(); |
1da177e4 LT |
1015 | |
1016 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1017 | /* step 1: figure out the size of the semaphore array */ |
023a5355 ND |
1018 | sma = sem_lock_check(ns, semid); |
1019 | if (IS_ERR(sma)) | |
1020 | return ERR_PTR(PTR_ERR(sma)); | |
1021 | ||
1da177e4 | 1022 | nsems = sma->sem_nsems; |
6ff37972 | 1023 | sem_getref_and_unlock(sma); |
1da177e4 | 1024 | |
4daa28f6 | 1025 | /* step 2: allocate new undo structure */ |
4668edc3 | 1026 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1027 | if (!new) { |
6ff37972 | 1028 | sem_putref(sma); |
1da177e4 LT |
1029 | return ERR_PTR(-ENOMEM); |
1030 | } | |
1da177e4 | 1031 | |
380af1b3 | 1032 | /* step 3: Acquire the lock on semaphore array */ |
6ff37972 | 1033 | sem_lock_and_putref(sma); |
1da177e4 LT |
1034 | if (sma->sem_perm.deleted) { |
1035 | sem_unlock(sma); | |
1da177e4 LT |
1036 | kfree(new); |
1037 | un = ERR_PTR(-EIDRM); | |
1038 | goto out; | |
1039 | } | |
380af1b3 MS |
1040 | spin_lock(&ulp->lock); |
1041 | ||
1042 | /* | |
1043 | * step 4: check for races: did someone else allocate the undo struct? | |
1044 | */ | |
1045 | un = lookup_undo(ulp, semid); | |
1046 | if (un) { | |
1047 | kfree(new); | |
1048 | goto success; | |
1049 | } | |
4daa28f6 MS |
1050 | /* step 5: initialize & link new undo structure */ |
1051 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1052 | new->ulp = ulp; |
4daa28f6 MS |
1053 | new->semid = semid; |
1054 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1055 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
4daa28f6 MS |
1056 | assert_spin_locked(&sma->sem_perm.lock); |
1057 | list_add(&new->list_id, &sma->list_id); | |
380af1b3 | 1058 | un = new; |
4daa28f6 | 1059 | |
380af1b3 | 1060 | success: |
c530c6ac | 1061 | spin_unlock(&ulp->lock); |
380af1b3 MS |
1062 | rcu_read_lock(); |
1063 | sem_unlock(sma); | |
1da177e4 LT |
1064 | out: |
1065 | return un; | |
1066 | } | |
1067 | ||
d5460c99 HC |
1068 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1069 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1070 | { |
1071 | int error = -EINVAL; | |
1072 | struct sem_array *sma; | |
1073 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1074 | struct sembuf* sops = fast_sops, *sop; | |
1075 | struct sem_undo *un; | |
b78755ab | 1076 | int undos = 0, alter = 0, max; |
1da177e4 LT |
1077 | struct sem_queue queue; |
1078 | unsigned long jiffies_left = 0; | |
e3893534 KK |
1079 | struct ipc_namespace *ns; |
1080 | ||
1081 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1082 | |
1083 | if (nsops < 1 || semid < 0) | |
1084 | return -EINVAL; | |
e3893534 | 1085 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1086 | return -E2BIG; |
1087 | if(nsops > SEMOPM_FAST) { | |
1088 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1089 | if(sops==NULL) | |
1090 | return -ENOMEM; | |
1091 | } | |
1092 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1093 | error=-EFAULT; | |
1094 | goto out_free; | |
1095 | } | |
1096 | if (timeout) { | |
1097 | struct timespec _timeout; | |
1098 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1099 | error = -EFAULT; | |
1100 | goto out_free; | |
1101 | } | |
1102 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1103 | _timeout.tv_nsec >= 1000000000L) { | |
1104 | error = -EINVAL; | |
1105 | goto out_free; | |
1106 | } | |
1107 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1108 | } | |
1109 | max = 0; | |
1110 | for (sop = sops; sop < sops + nsops; sop++) { | |
1111 | if (sop->sem_num >= max) | |
1112 | max = sop->sem_num; | |
1113 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1114 | undos = 1; |
1115 | if (sop->sem_op != 0) | |
1da177e4 LT |
1116 | alter = 1; |
1117 | } | |
1da177e4 | 1118 | |
1da177e4 | 1119 | if (undos) { |
4daa28f6 | 1120 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1121 | if (IS_ERR(un)) { |
1122 | error = PTR_ERR(un); | |
1123 | goto out_free; | |
1124 | } | |
1125 | } else | |
1126 | un = NULL; | |
1127 | ||
023a5355 ND |
1128 | sma = sem_lock_check(ns, semid); |
1129 | if (IS_ERR(sma)) { | |
380af1b3 MS |
1130 | if (un) |
1131 | rcu_read_unlock(); | |
023a5355 | 1132 | error = PTR_ERR(sma); |
1da177e4 | 1133 | goto out_free; |
023a5355 ND |
1134 | } |
1135 | ||
1da177e4 | 1136 | /* |
4daa28f6 | 1137 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1138 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1139 | * and now a new array with received the same id. Check and fail. |
380af1b3 MS |
1140 | * This case can be detected checking un->semid. The existance of |
1141 | * "un" itself is guaranteed by rcu. | |
1da177e4 | 1142 | */ |
4daa28f6 | 1143 | error = -EIDRM; |
380af1b3 MS |
1144 | if (un) { |
1145 | if (un->semid == -1) { | |
1146 | rcu_read_unlock(); | |
1147 | goto out_unlock_free; | |
1148 | } else { | |
1149 | /* | |
1150 | * rcu lock can be released, "un" cannot disappear: | |
1151 | * - sem_lock is acquired, thus IPC_RMID is | |
1152 | * impossible. | |
1153 | * - exit_sem is impossible, it always operates on | |
1154 | * current (or a dead task). | |
1155 | */ | |
1156 | ||
1157 | rcu_read_unlock(); | |
1158 | } | |
1159 | } | |
4daa28f6 | 1160 | |
1da177e4 LT |
1161 | error = -EFBIG; |
1162 | if (max >= sma->sem_nsems) | |
1163 | goto out_unlock_free; | |
1164 | ||
1165 | error = -EACCES; | |
1166 | if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) | |
1167 | goto out_unlock_free; | |
1168 | ||
1169 | error = security_sem_semop(sma, sops, nsops, alter); | |
1170 | if (error) | |
1171 | goto out_unlock_free; | |
1172 | ||
b488893a | 1173 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1174 | if (error <= 0) { |
1175 | if (alter && error == 0) | |
1176 | update_queue (sma); | |
1177 | goto out_unlock_free; | |
1178 | } | |
1179 | ||
1180 | /* We need to sleep on this operation, so we put the current | |
1181 | * task into the pending queue and go to sleep. | |
1182 | */ | |
1183 | ||
1da177e4 LT |
1184 | queue.sops = sops; |
1185 | queue.nsops = nsops; | |
1186 | queue.undo = un; | |
b488893a | 1187 | queue.pid = task_tgid_vnr(current); |
1da177e4 LT |
1188 | queue.alter = alter; |
1189 | if (alter) | |
a1193f8e | 1190 | list_add_tail(&queue.list, &sma->sem_pending); |
1da177e4 | 1191 | else |
a1193f8e | 1192 | list_add(&queue.list, &sma->sem_pending); |
1da177e4 LT |
1193 | |
1194 | queue.status = -EINTR; | |
1195 | queue.sleeper = current; | |
1196 | current->state = TASK_INTERRUPTIBLE; | |
1197 | sem_unlock(sma); | |
1198 | ||
1199 | if (timeout) | |
1200 | jiffies_left = schedule_timeout(jiffies_left); | |
1201 | else | |
1202 | schedule(); | |
1203 | ||
1204 | error = queue.status; | |
1205 | while(unlikely(error == IN_WAKEUP)) { | |
1206 | cpu_relax(); | |
1207 | error = queue.status; | |
1208 | } | |
1209 | ||
1210 | if (error != -EINTR) { | |
1211 | /* fast path: update_queue already obtained all requested | |
1212 | * resources */ | |
1213 | goto out_free; | |
1214 | } | |
1215 | ||
e3893534 | 1216 | sma = sem_lock(ns, semid); |
023a5355 | 1217 | if (IS_ERR(sma)) { |
1da177e4 LT |
1218 | error = -EIDRM; |
1219 | goto out_free; | |
1220 | } | |
1221 | ||
1222 | /* | |
1223 | * If queue.status != -EINTR we are woken up by another process | |
1224 | */ | |
1225 | error = queue.status; | |
1226 | if (error != -EINTR) { | |
1227 | goto out_unlock_free; | |
1228 | } | |
1229 | ||
1230 | /* | |
1231 | * If an interrupt occurred we have to clean up the queue | |
1232 | */ | |
1233 | if (timeout && jiffies_left == 0) | |
1234 | error = -EAGAIN; | |
a1193f8e | 1235 | list_del(&queue.list); |
1da177e4 LT |
1236 | |
1237 | out_unlock_free: | |
1238 | sem_unlock(sma); | |
1239 | out_free: | |
1240 | if(sops != fast_sops) | |
1241 | kfree(sops); | |
1242 | return error; | |
1243 | } | |
1244 | ||
d5460c99 HC |
1245 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
1246 | unsigned, nsops) | |
1da177e4 LT |
1247 | { |
1248 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1249 | } | |
1250 | ||
1251 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1252 | * parent and child tasks. | |
1da177e4 LT |
1253 | */ |
1254 | ||
1255 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1256 | { | |
1257 | struct sem_undo_list *undo_list; | |
1258 | int error; | |
1259 | ||
1260 | if (clone_flags & CLONE_SYSVSEM) { | |
1261 | error = get_undo_list(&undo_list); | |
1262 | if (error) | |
1263 | return error; | |
1da177e4 LT |
1264 | atomic_inc(&undo_list->refcnt); |
1265 | tsk->sysvsem.undo_list = undo_list; | |
1266 | } else | |
1267 | tsk->sysvsem.undo_list = NULL; | |
1268 | ||
1269 | return 0; | |
1270 | } | |
1271 | ||
1272 | /* | |
1273 | * add semadj values to semaphores, free undo structures. | |
1274 | * undo structures are not freed when semaphore arrays are destroyed | |
1275 | * so some of them may be out of date. | |
1276 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1277 | * set of adjustments that needs to be done should be done in an atomic | |
1278 | * manner or not. That is, if we are attempting to decrement the semval | |
1279 | * should we queue up and wait until we can do so legally? | |
1280 | * The original implementation attempted to do this (queue and wait). | |
1281 | * The current implementation does not do so. The POSIX standard | |
1282 | * and SVID should be consulted to determine what behavior is mandated. | |
1283 | */ | |
1284 | void exit_sem(struct task_struct *tsk) | |
1285 | { | |
4daa28f6 | 1286 | struct sem_undo_list *ulp; |
1da177e4 | 1287 | |
4daa28f6 MS |
1288 | ulp = tsk->sysvsem.undo_list; |
1289 | if (!ulp) | |
1da177e4 | 1290 | return; |
9edff4ab | 1291 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 1292 | |
4daa28f6 | 1293 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
1294 | return; |
1295 | ||
380af1b3 | 1296 | for (;;) { |
1da177e4 | 1297 | struct sem_array *sma; |
380af1b3 MS |
1298 | struct sem_undo *un; |
1299 | int semid; | |
4daa28f6 MS |
1300 | int i; |
1301 | ||
380af1b3 | 1302 | rcu_read_lock(); |
05725f7e JP |
1303 | un = list_entry_rcu(ulp->list_proc.next, |
1304 | struct sem_undo, list_proc); | |
380af1b3 MS |
1305 | if (&un->list_proc == &ulp->list_proc) |
1306 | semid = -1; | |
1307 | else | |
1308 | semid = un->semid; | |
1309 | rcu_read_unlock(); | |
4daa28f6 | 1310 | |
380af1b3 MS |
1311 | if (semid == -1) |
1312 | break; | |
1da177e4 | 1313 | |
380af1b3 | 1314 | sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid); |
1da177e4 | 1315 | |
380af1b3 MS |
1316 | /* exit_sem raced with IPC_RMID, nothing to do */ |
1317 | if (IS_ERR(sma)) | |
1318 | continue; | |
1da177e4 | 1319 | |
bf17bb71 | 1320 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
1321 | if (un == NULL) { |
1322 | /* exit_sem raced with IPC_RMID+semget() that created | |
1323 | * exactly the same semid. Nothing to do. | |
1324 | */ | |
1325 | sem_unlock(sma); | |
1326 | continue; | |
1327 | } | |
1328 | ||
1329 | /* remove un from the linked lists */ | |
4daa28f6 MS |
1330 | assert_spin_locked(&sma->sem_perm.lock); |
1331 | list_del(&un->list_id); | |
1332 | ||
380af1b3 MS |
1333 | spin_lock(&ulp->lock); |
1334 | list_del_rcu(&un->list_proc); | |
1335 | spin_unlock(&ulp->lock); | |
1336 | ||
4daa28f6 MS |
1337 | /* perform adjustments registered in un */ |
1338 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 1339 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
1340 | if (un->semadj[i]) { |
1341 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
1342 | /* |
1343 | * Range checks of the new semaphore value, | |
1344 | * not defined by sus: | |
1345 | * - Some unices ignore the undo entirely | |
1346 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
1347 | * - some cap the value (e.g. FreeBSD caps | |
1348 | * at 0, but doesn't enforce SEMVMX) | |
1349 | * | |
1350 | * Linux caps the semaphore value, both at 0 | |
1351 | * and at SEMVMX. | |
1352 | * | |
1353 | * Manfred <manfred@colorfullife.com> | |
1354 | */ | |
5f921ae9 IM |
1355 | if (semaphore->semval < 0) |
1356 | semaphore->semval = 0; | |
1357 | if (semaphore->semval > SEMVMX) | |
1358 | semaphore->semval = SEMVMX; | |
b488893a | 1359 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
1360 | } |
1361 | } | |
1362 | sma->sem_otime = get_seconds(); | |
1363 | /* maybe some queued-up processes were waiting for this */ | |
1364 | update_queue(sma); | |
1da177e4 | 1365 | sem_unlock(sma); |
380af1b3 MS |
1366 | |
1367 | call_rcu(&un->rcu, free_un); | |
1da177e4 | 1368 | } |
4daa28f6 | 1369 | kfree(ulp); |
1da177e4 LT |
1370 | } |
1371 | ||
1372 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1373 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1374 | { |
19b4946c MW |
1375 | struct sem_array *sma = it; |
1376 | ||
1377 | return seq_printf(s, | |
1378 | "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n", | |
1379 | sma->sem_perm.key, | |
7ca7e564 | 1380 | sma->sem_perm.id, |
19b4946c MW |
1381 | sma->sem_perm.mode, |
1382 | sma->sem_nsems, | |
1383 | sma->sem_perm.uid, | |
1384 | sma->sem_perm.gid, | |
1385 | sma->sem_perm.cuid, | |
1386 | sma->sem_perm.cgid, | |
1387 | sma->sem_otime, | |
1388 | sma->sem_ctime); | |
1da177e4 LT |
1389 | } |
1390 | #endif |