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 LT |
60 | * Enforced range limit on SEM_UNDO |
61 | * (c) 2001 Red Hat Inc <alan@redhat.com> | |
62 | * Lockless wakeup | |
63 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
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> |
5f921ae9 | 83 | #include <linux/mutex.h> |
e3893534 | 84 | #include <linux/nsproxy.h> |
5f921ae9 | 85 | |
1da177e4 LT |
86 | #include <asm/uaccess.h> |
87 | #include "util.h" | |
88 | ||
e3893534 KK |
89 | #define sem_ids(ns) (*((ns)->ids[IPC_SEM_IDS])) |
90 | ||
91 | #define sem_lock(ns, id) ((struct sem_array*)ipc_lock(&sem_ids(ns), id)) | |
92 | #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm) | |
e3893534 KK |
93 | #define sem_checkid(ns, sma, semid) \ |
94 | ipc_checkid(&sem_ids(ns),&sma->sem_perm,semid) | |
95 | #define sem_buildid(ns, id, seq) \ | |
96 | ipc_buildid(&sem_ids(ns), id, seq) | |
1da177e4 | 97 | |
e3893534 | 98 | static struct ipc_ids init_sem_ids; |
1da177e4 | 99 | |
7748dbfa | 100 | static int newary(struct ipc_namespace *, struct ipc_params *); |
7ca7e564 | 101 | static void freeary(struct ipc_namespace *, struct sem_array *); |
1da177e4 | 102 | #ifdef CONFIG_PROC_FS |
19b4946c | 103 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
104 | #endif |
105 | ||
106 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
107 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
108 | ||
109 | /* | |
110 | * linked list protection: | |
111 | * sem_undo.id_next, | |
112 | * sem_array.sem_pending{,last}, | |
113 | * sem_array.sem_undo: sem_lock() for read/write | |
114 | * sem_undo.proc_next: only "current" is allowed to read/write that field. | |
115 | * | |
116 | */ | |
117 | ||
e3893534 KK |
118 | #define sc_semmsl sem_ctls[0] |
119 | #define sc_semmns sem_ctls[1] | |
120 | #define sc_semopm sem_ctls[2] | |
121 | #define sc_semmni sem_ctls[3] | |
122 | ||
7d69a1f4 | 123 | static void __sem_init_ns(struct ipc_namespace *ns, struct ipc_ids *ids) |
e3893534 KK |
124 | { |
125 | ns->ids[IPC_SEM_IDS] = ids; | |
126 | ns->sc_semmsl = SEMMSL; | |
127 | ns->sc_semmns = SEMMNS; | |
128 | ns->sc_semopm = SEMOPM; | |
129 | ns->sc_semmni = SEMMNI; | |
130 | ns->used_sems = 0; | |
7ca7e564 | 131 | ipc_init_ids(ids); |
e3893534 KK |
132 | } |
133 | ||
e3893534 KK |
134 | int sem_init_ns(struct ipc_namespace *ns) |
135 | { | |
136 | struct ipc_ids *ids; | |
137 | ||
138 | ids = kmalloc(sizeof(struct ipc_ids), GFP_KERNEL); | |
139 | if (ids == NULL) | |
140 | return -ENOMEM; | |
141 | ||
142 | __sem_init_ns(ns, ids); | |
143 | return 0; | |
144 | } | |
145 | ||
146 | void sem_exit_ns(struct ipc_namespace *ns) | |
147 | { | |
e3893534 | 148 | struct sem_array *sma; |
7ca7e564 ND |
149 | int next_id; |
150 | int total, in_use; | |
e3893534 KK |
151 | |
152 | mutex_lock(&sem_ids(ns).mutex); | |
7ca7e564 ND |
153 | |
154 | in_use = sem_ids(ns).in_use; | |
155 | ||
156 | for (total = 0, next_id = 0; total < in_use; next_id++) { | |
157 | sma = idr_find(&sem_ids(ns).ipcs_idr, next_id); | |
e3893534 KK |
158 | if (sma == NULL) |
159 | continue; | |
7ca7e564 ND |
160 | ipc_lock_by_ptr(&sma->sem_perm); |
161 | freeary(ns, sma); | |
162 | total++; | |
e3893534 KK |
163 | } |
164 | mutex_unlock(&sem_ids(ns).mutex); | |
1da177e4 | 165 | |
e3893534 KK |
166 | kfree(ns->ids[IPC_SEM_IDS]); |
167 | ns->ids[IPC_SEM_IDS] = NULL; | |
168 | } | |
1da177e4 LT |
169 | |
170 | void __init sem_init (void) | |
171 | { | |
e3893534 | 172 | __sem_init_ns(&init_ipc_ns, &init_sem_ids); |
19b4946c MW |
173 | ipc_init_proc_interface("sysvipc/sem", |
174 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 175 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
176 | } |
177 | ||
7ca7e564 ND |
178 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
179 | { | |
180 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
181 | } | |
182 | ||
1da177e4 LT |
183 | /* |
184 | * Lockless wakeup algorithm: | |
185 | * Without the check/retry algorithm a lockless wakeup is possible: | |
186 | * - queue.status is initialized to -EINTR before blocking. | |
187 | * - wakeup is performed by | |
188 | * * unlinking the queue entry from sma->sem_pending | |
189 | * * setting queue.status to IN_WAKEUP | |
190 | * This is the notification for the blocked thread that a | |
191 | * result value is imminent. | |
192 | * * call wake_up_process | |
193 | * * set queue.status to the final value. | |
194 | * - the previously blocked thread checks queue.status: | |
195 | * * if it's IN_WAKEUP, then it must wait until the value changes | |
196 | * * if it's not -EINTR, then the operation was completed by | |
197 | * update_queue. semtimedop can return queue.status without | |
5f921ae9 | 198 | * performing any operation on the sem array. |
1da177e4 LT |
199 | * * otherwise it must acquire the spinlock and check what's up. |
200 | * | |
201 | * The two-stage algorithm is necessary to protect against the following | |
202 | * races: | |
203 | * - if queue.status is set after wake_up_process, then the woken up idle | |
204 | * thread could race forward and try (and fail) to acquire sma->lock | |
205 | * before update_queue had a chance to set queue.status | |
206 | * - if queue.status is written before wake_up_process and if the | |
207 | * blocked process is woken up by a signal between writing | |
208 | * queue.status and the wake_up_process, then the woken up | |
209 | * process could return from semtimedop and die by calling | |
210 | * sys_exit before wake_up_process is called. Then wake_up_process | |
211 | * will oops, because the task structure is already invalid. | |
212 | * (yes, this happened on s390 with sysv msg). | |
213 | * | |
214 | */ | |
215 | #define IN_WAKEUP 1 | |
216 | ||
7748dbfa | 217 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
218 | { |
219 | int id; | |
220 | int retval; | |
221 | struct sem_array *sma; | |
222 | int size; | |
7748dbfa ND |
223 | key_t key = params->key; |
224 | int nsems = params->u.nsems; | |
225 | int semflg = params->flg; | |
1da177e4 LT |
226 | |
227 | if (!nsems) | |
228 | return -EINVAL; | |
e3893534 | 229 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
230 | return -ENOSPC; |
231 | ||
232 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
233 | sma = ipc_rcu_alloc(size); | |
234 | if (!sma) { | |
235 | return -ENOMEM; | |
236 | } | |
237 | memset (sma, 0, size); | |
238 | ||
239 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
240 | sma->sem_perm.key = key; | |
241 | ||
242 | sma->sem_perm.security = NULL; | |
243 | retval = security_sem_alloc(sma); | |
244 | if (retval) { | |
245 | ipc_rcu_putref(sma); | |
246 | return retval; | |
247 | } | |
248 | ||
e3893534 | 249 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
1da177e4 LT |
250 | if(id == -1) { |
251 | security_sem_free(sma); | |
252 | ipc_rcu_putref(sma); | |
253 | return -ENOSPC; | |
254 | } | |
e3893534 | 255 | ns->used_sems += nsems; |
1da177e4 | 256 | |
7ca7e564 | 257 | sma->sem_perm.id = sem_buildid(ns, id, sma->sem_perm.seq); |
1da177e4 LT |
258 | sma->sem_base = (struct sem *) &sma[1]; |
259 | /* sma->sem_pending = NULL; */ | |
260 | sma->sem_pending_last = &sma->sem_pending; | |
261 | /* sma->undo = NULL; */ | |
262 | sma->sem_nsems = nsems; | |
263 | sma->sem_ctime = get_seconds(); | |
264 | sem_unlock(sma); | |
265 | ||
7ca7e564 | 266 | return sma->sem_perm.id; |
1da177e4 LT |
267 | } |
268 | ||
7748dbfa ND |
269 | |
270 | static inline int sem_security(void *sma, int semflg) | |
271 | { | |
272 | return security_sem_associate((struct sem_array *) sma, semflg); | |
273 | } | |
274 | ||
275 | static inline int sem_more_checks(void *sma, struct ipc_params *params) | |
276 | { | |
277 | if (params->u.nsems > ((struct sem_array *)sma)->sem_nsems) | |
278 | return -EINVAL; | |
279 | ||
280 | return 0; | |
281 | } | |
282 | ||
283 | asmlinkage long sys_semget(key_t key, int nsems, int semflg) | |
1da177e4 | 284 | { |
e3893534 | 285 | struct ipc_namespace *ns; |
7748dbfa ND |
286 | struct ipc_ops sem_ops; |
287 | struct ipc_params sem_params; | |
e3893534 KK |
288 | |
289 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 290 | |
e3893534 | 291 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 292 | return -EINVAL; |
7ca7e564 | 293 | |
7748dbfa ND |
294 | sem_ops.getnew = newary; |
295 | sem_ops.associate = sem_security; | |
296 | sem_ops.more_checks = sem_more_checks; | |
297 | ||
298 | sem_params.key = key; | |
299 | sem_params.flg = semflg; | |
300 | sem_params.u.nsems = nsems; | |
1da177e4 | 301 | |
7748dbfa | 302 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
303 | } |
304 | ||
305 | /* Manage the doubly linked list sma->sem_pending as a FIFO: | |
306 | * insert new queue elements at the tail sma->sem_pending_last. | |
307 | */ | |
308 | static inline void append_to_queue (struct sem_array * sma, | |
309 | struct sem_queue * q) | |
310 | { | |
311 | *(q->prev = sma->sem_pending_last) = q; | |
312 | *(sma->sem_pending_last = &q->next) = NULL; | |
313 | } | |
314 | ||
315 | static inline void prepend_to_queue (struct sem_array * sma, | |
316 | struct sem_queue * q) | |
317 | { | |
318 | q->next = sma->sem_pending; | |
319 | *(q->prev = &sma->sem_pending) = q; | |
320 | if (q->next) | |
321 | q->next->prev = &q->next; | |
322 | else /* sma->sem_pending_last == &sma->sem_pending */ | |
323 | sma->sem_pending_last = &q->next; | |
324 | } | |
325 | ||
326 | static inline void remove_from_queue (struct sem_array * sma, | |
327 | struct sem_queue * q) | |
328 | { | |
329 | *(q->prev) = q->next; | |
330 | if (q->next) | |
331 | q->next->prev = q->prev; | |
332 | else /* sma->sem_pending_last == &q->next */ | |
333 | sma->sem_pending_last = q->prev; | |
334 | q->prev = NULL; /* mark as removed */ | |
335 | } | |
336 | ||
337 | /* | |
338 | * Determine whether a sequence of semaphore operations would succeed | |
339 | * all at once. Return 0 if yes, 1 if need to sleep, else return error code. | |
340 | */ | |
341 | ||
342 | static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, | |
343 | int nsops, struct sem_undo *un, int pid) | |
344 | { | |
345 | int result, sem_op; | |
346 | struct sembuf *sop; | |
347 | struct sem * curr; | |
348 | ||
349 | for (sop = sops; sop < sops + nsops; sop++) { | |
350 | curr = sma->sem_base + sop->sem_num; | |
351 | sem_op = sop->sem_op; | |
352 | result = curr->semval; | |
353 | ||
354 | if (!sem_op && result) | |
355 | goto would_block; | |
356 | ||
357 | result += sem_op; | |
358 | if (result < 0) | |
359 | goto would_block; | |
360 | if (result > SEMVMX) | |
361 | goto out_of_range; | |
362 | if (sop->sem_flg & SEM_UNDO) { | |
363 | int undo = un->semadj[sop->sem_num] - sem_op; | |
364 | /* | |
365 | * Exceeding the undo range is an error. | |
366 | */ | |
367 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
368 | goto out_of_range; | |
369 | } | |
370 | curr->semval = result; | |
371 | } | |
372 | ||
373 | sop--; | |
374 | while (sop >= sops) { | |
375 | sma->sem_base[sop->sem_num].sempid = pid; | |
376 | if (sop->sem_flg & SEM_UNDO) | |
377 | un->semadj[sop->sem_num] -= sop->sem_op; | |
378 | sop--; | |
379 | } | |
380 | ||
381 | sma->sem_otime = get_seconds(); | |
382 | return 0; | |
383 | ||
384 | out_of_range: | |
385 | result = -ERANGE; | |
386 | goto undo; | |
387 | ||
388 | would_block: | |
389 | if (sop->sem_flg & IPC_NOWAIT) | |
390 | result = -EAGAIN; | |
391 | else | |
392 | result = 1; | |
393 | ||
394 | undo: | |
395 | sop--; | |
396 | while (sop >= sops) { | |
397 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; | |
398 | sop--; | |
399 | } | |
400 | ||
401 | return result; | |
402 | } | |
403 | ||
404 | /* Go through the pending queue for the indicated semaphore | |
405 | * looking for tasks that can be completed. | |
406 | */ | |
407 | static void update_queue (struct sem_array * sma) | |
408 | { | |
409 | int error; | |
410 | struct sem_queue * q; | |
411 | ||
412 | q = sma->sem_pending; | |
413 | while(q) { | |
414 | error = try_atomic_semop(sma, q->sops, q->nsops, | |
415 | q->undo, q->pid); | |
416 | ||
417 | /* Does q->sleeper still need to sleep? */ | |
418 | if (error <= 0) { | |
419 | struct sem_queue *n; | |
420 | remove_from_queue(sma,q); | |
421 | q->status = IN_WAKEUP; | |
422 | /* | |
423 | * Continue scanning. The next operation | |
424 | * that must be checked depends on the type of the | |
425 | * completed operation: | |
426 | * - if the operation modified the array, then | |
427 | * restart from the head of the queue and | |
428 | * check for threads that might be waiting | |
429 | * for semaphore values to become 0. | |
430 | * - if the operation didn't modify the array, | |
431 | * then just continue. | |
432 | */ | |
433 | if (q->alter) | |
434 | n = sma->sem_pending; | |
435 | else | |
436 | n = q->next; | |
437 | wake_up_process(q->sleeper); | |
438 | /* hands-off: q will disappear immediately after | |
439 | * writing q->status. | |
440 | */ | |
1224b375 | 441 | smp_wmb(); |
1da177e4 LT |
442 | q->status = error; |
443 | q = n; | |
444 | } else { | |
445 | q = q->next; | |
446 | } | |
447 | } | |
448 | } | |
449 | ||
450 | /* The following counts are associated to each semaphore: | |
451 | * semncnt number of tasks waiting on semval being nonzero | |
452 | * semzcnt number of tasks waiting on semval being zero | |
453 | * This model assumes that a task waits on exactly one semaphore. | |
454 | * Since semaphore operations are to be performed atomically, tasks actually | |
455 | * wait on a whole sequence of semaphores simultaneously. | |
456 | * The counts we return here are a rough approximation, but still | |
457 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
458 | */ | |
459 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
460 | { | |
461 | int semncnt; | |
462 | struct sem_queue * q; | |
463 | ||
464 | semncnt = 0; | |
465 | for (q = sma->sem_pending; q; q = q->next) { | |
466 | struct sembuf * sops = q->sops; | |
467 | int nsops = q->nsops; | |
468 | int i; | |
469 | for (i = 0; i < nsops; i++) | |
470 | if (sops[i].sem_num == semnum | |
471 | && (sops[i].sem_op < 0) | |
472 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
473 | semncnt++; | |
474 | } | |
475 | return semncnt; | |
476 | } | |
477 | static int count_semzcnt (struct sem_array * sma, ushort semnum) | |
478 | { | |
479 | int semzcnt; | |
480 | struct sem_queue * q; | |
481 | ||
482 | semzcnt = 0; | |
483 | for (q = sma->sem_pending; q; q = q->next) { | |
484 | struct sembuf * sops = q->sops; | |
485 | int nsops = q->nsops; | |
486 | int i; | |
487 | for (i = 0; i < nsops; i++) | |
488 | if (sops[i].sem_num == semnum | |
489 | && (sops[i].sem_op == 0) | |
490 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
491 | semzcnt++; | |
492 | } | |
493 | return semzcnt; | |
494 | } | |
495 | ||
5f921ae9 IM |
496 | /* Free a semaphore set. freeary() is called with sem_ids.mutex locked and |
497 | * the spinlock for this semaphore set hold. sem_ids.mutex remains locked | |
1da177e4 LT |
498 | * on exit. |
499 | */ | |
7ca7e564 | 500 | static void freeary(struct ipc_namespace *ns, struct sem_array *sma) |
1da177e4 LT |
501 | { |
502 | struct sem_undo *un; | |
503 | struct sem_queue *q; | |
1da177e4 LT |
504 | |
505 | /* Invalidate the existing undo structures for this semaphore set. | |
506 | * (They will be freed without any further action in exit_sem() | |
507 | * or during the next semop.) | |
508 | */ | |
509 | for (un = sma->undo; un; un = un->id_next) | |
510 | un->semid = -1; | |
511 | ||
512 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
513 | q = sma->sem_pending; | |
514 | while(q) { | |
515 | struct sem_queue *n; | |
516 | /* lazy remove_from_queue: we are killing the whole queue */ | |
517 | q->prev = NULL; | |
518 | n = q->next; | |
519 | q->status = IN_WAKEUP; | |
520 | wake_up_process(q->sleeper); /* doesn't sleep */ | |
6003a93e | 521 | smp_wmb(); |
1da177e4 LT |
522 | q->status = -EIDRM; /* hands-off q */ |
523 | q = n; | |
524 | } | |
525 | ||
7ca7e564 ND |
526 | /* Remove the semaphore set from the IDR */ |
527 | sem_rmid(ns, sma); | |
1da177e4 LT |
528 | sem_unlock(sma); |
529 | ||
e3893534 | 530 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
531 | security_sem_free(sma); |
532 | ipc_rcu_putref(sma); | |
533 | } | |
534 | ||
535 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
536 | { | |
537 | switch(version) { | |
538 | case IPC_64: | |
539 | return copy_to_user(buf, in, sizeof(*in)); | |
540 | case IPC_OLD: | |
541 | { | |
542 | struct semid_ds out; | |
543 | ||
544 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); | |
545 | ||
546 | out.sem_otime = in->sem_otime; | |
547 | out.sem_ctime = in->sem_ctime; | |
548 | out.sem_nsems = in->sem_nsems; | |
549 | ||
550 | return copy_to_user(buf, &out, sizeof(out)); | |
551 | } | |
552 | default: | |
553 | return -EINVAL; | |
554 | } | |
555 | } | |
556 | ||
e3893534 KK |
557 | static int semctl_nolock(struct ipc_namespace *ns, int semid, int semnum, |
558 | int cmd, int version, union semun arg) | |
1da177e4 LT |
559 | { |
560 | int err = -EINVAL; | |
561 | struct sem_array *sma; | |
562 | ||
563 | switch(cmd) { | |
564 | case IPC_INFO: | |
565 | case SEM_INFO: | |
566 | { | |
567 | struct seminfo seminfo; | |
568 | int max_id; | |
569 | ||
570 | err = security_sem_semctl(NULL, cmd); | |
571 | if (err) | |
572 | return err; | |
573 | ||
574 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
575 | seminfo.semmni = ns->sc_semmni; |
576 | seminfo.semmns = ns->sc_semmns; | |
577 | seminfo.semmsl = ns->sc_semmsl; | |
578 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
579 | seminfo.semvmx = SEMVMX; |
580 | seminfo.semmnu = SEMMNU; | |
581 | seminfo.semmap = SEMMAP; | |
582 | seminfo.semume = SEMUME; | |
e3893534 | 583 | mutex_lock(&sem_ids(ns).mutex); |
1da177e4 | 584 | if (cmd == SEM_INFO) { |
e3893534 KK |
585 | seminfo.semusz = sem_ids(ns).in_use; |
586 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
587 | } else { |
588 | seminfo.semusz = SEMUSZ; | |
589 | seminfo.semaem = SEMAEM; | |
590 | } | |
7ca7e564 | 591 | max_id = ipc_get_maxid(&sem_ids(ns)); |
e3893534 | 592 | mutex_unlock(&sem_ids(ns).mutex); |
1da177e4 LT |
593 | if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) |
594 | return -EFAULT; | |
595 | return (max_id < 0) ? 0: max_id; | |
596 | } | |
597 | case SEM_STAT: | |
598 | { | |
599 | struct semid64_ds tbuf; | |
600 | int id; | |
601 | ||
1da177e4 LT |
602 | memset(&tbuf,0,sizeof(tbuf)); |
603 | ||
e3893534 | 604 | sma = sem_lock(ns, semid); |
1da177e4 LT |
605 | if(sma == NULL) |
606 | return -EINVAL; | |
607 | ||
608 | err = -EACCES; | |
609 | if (ipcperms (&sma->sem_perm, S_IRUGO)) | |
610 | goto out_unlock; | |
611 | ||
612 | err = security_sem_semctl(sma, cmd); | |
613 | if (err) | |
614 | goto out_unlock; | |
615 | ||
7ca7e564 | 616 | id = sma->sem_perm.id; |
1da177e4 LT |
617 | |
618 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); | |
619 | tbuf.sem_otime = sma->sem_otime; | |
620 | tbuf.sem_ctime = sma->sem_ctime; | |
621 | tbuf.sem_nsems = sma->sem_nsems; | |
622 | sem_unlock(sma); | |
623 | if (copy_semid_to_user (arg.buf, &tbuf, version)) | |
624 | return -EFAULT; | |
625 | return id; | |
626 | } | |
627 | default: | |
628 | return -EINVAL; | |
629 | } | |
630 | return err; | |
631 | out_unlock: | |
632 | sem_unlock(sma); | |
633 | return err; | |
634 | } | |
635 | ||
e3893534 KK |
636 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
637 | int cmd, int version, union semun arg) | |
1da177e4 LT |
638 | { |
639 | struct sem_array *sma; | |
640 | struct sem* curr; | |
641 | int err; | |
642 | ushort fast_sem_io[SEMMSL_FAST]; | |
643 | ushort* sem_io = fast_sem_io; | |
644 | int nsems; | |
645 | ||
e3893534 | 646 | sma = sem_lock(ns, semid); |
1da177e4 LT |
647 | if(sma==NULL) |
648 | return -EINVAL; | |
649 | ||
650 | nsems = sma->sem_nsems; | |
651 | ||
652 | err=-EIDRM; | |
e3893534 | 653 | if (sem_checkid(ns,sma,semid)) |
1da177e4 LT |
654 | goto out_unlock; |
655 | ||
656 | err = -EACCES; | |
657 | if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) | |
658 | goto out_unlock; | |
659 | ||
660 | err = security_sem_semctl(sma, cmd); | |
661 | if (err) | |
662 | goto out_unlock; | |
663 | ||
664 | err = -EACCES; | |
665 | switch (cmd) { | |
666 | case GETALL: | |
667 | { | |
668 | ushort __user *array = arg.array; | |
669 | int i; | |
670 | ||
671 | if(nsems > SEMMSL_FAST) { | |
672 | ipc_rcu_getref(sma); | |
673 | sem_unlock(sma); | |
674 | ||
675 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
676 | if(sem_io == NULL) { | |
677 | ipc_lock_by_ptr(&sma->sem_perm); | |
678 | ipc_rcu_putref(sma); | |
679 | sem_unlock(sma); | |
680 | return -ENOMEM; | |
681 | } | |
682 | ||
683 | ipc_lock_by_ptr(&sma->sem_perm); | |
684 | ipc_rcu_putref(sma); | |
685 | if (sma->sem_perm.deleted) { | |
686 | sem_unlock(sma); | |
687 | err = -EIDRM; | |
688 | goto out_free; | |
689 | } | |
690 | } | |
691 | ||
692 | for (i = 0; i < sma->sem_nsems; i++) | |
693 | sem_io[i] = sma->sem_base[i].semval; | |
694 | sem_unlock(sma); | |
695 | err = 0; | |
696 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
697 | err = -EFAULT; | |
698 | goto out_free; | |
699 | } | |
700 | case SETALL: | |
701 | { | |
702 | int i; | |
703 | struct sem_undo *un; | |
704 | ||
705 | ipc_rcu_getref(sma); | |
706 | sem_unlock(sma); | |
707 | ||
708 | if(nsems > SEMMSL_FAST) { | |
709 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
710 | if(sem_io == NULL) { | |
711 | ipc_lock_by_ptr(&sma->sem_perm); | |
712 | ipc_rcu_putref(sma); | |
713 | sem_unlock(sma); | |
714 | return -ENOMEM; | |
715 | } | |
716 | } | |
717 | ||
718 | if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { | |
719 | ipc_lock_by_ptr(&sma->sem_perm); | |
720 | ipc_rcu_putref(sma); | |
721 | sem_unlock(sma); | |
722 | err = -EFAULT; | |
723 | goto out_free; | |
724 | } | |
725 | ||
726 | for (i = 0; i < nsems; i++) { | |
727 | if (sem_io[i] > SEMVMX) { | |
728 | ipc_lock_by_ptr(&sma->sem_perm); | |
729 | ipc_rcu_putref(sma); | |
730 | sem_unlock(sma); | |
731 | err = -ERANGE; | |
732 | goto out_free; | |
733 | } | |
734 | } | |
735 | ipc_lock_by_ptr(&sma->sem_perm); | |
736 | ipc_rcu_putref(sma); | |
737 | if (sma->sem_perm.deleted) { | |
738 | sem_unlock(sma); | |
739 | err = -EIDRM; | |
740 | goto out_free; | |
741 | } | |
742 | ||
743 | for (i = 0; i < nsems; i++) | |
744 | sma->sem_base[i].semval = sem_io[i]; | |
745 | for (un = sma->undo; un; un = un->id_next) | |
746 | for (i = 0; i < nsems; i++) | |
747 | un->semadj[i] = 0; | |
748 | sma->sem_ctime = get_seconds(); | |
749 | /* maybe some queued-up processes were waiting for this */ | |
750 | update_queue(sma); | |
751 | err = 0; | |
752 | goto out_unlock; | |
753 | } | |
754 | case IPC_STAT: | |
755 | { | |
756 | struct semid64_ds tbuf; | |
757 | memset(&tbuf,0,sizeof(tbuf)); | |
758 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); | |
759 | tbuf.sem_otime = sma->sem_otime; | |
760 | tbuf.sem_ctime = sma->sem_ctime; | |
761 | tbuf.sem_nsems = sma->sem_nsems; | |
762 | sem_unlock(sma); | |
763 | if (copy_semid_to_user (arg.buf, &tbuf, version)) | |
764 | return -EFAULT; | |
765 | return 0; | |
766 | } | |
767 | /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ | |
768 | } | |
769 | err = -EINVAL; | |
770 | if(semnum < 0 || semnum >= nsems) | |
771 | goto out_unlock; | |
772 | ||
773 | curr = &sma->sem_base[semnum]; | |
774 | ||
775 | switch (cmd) { | |
776 | case GETVAL: | |
777 | err = curr->semval; | |
778 | goto out_unlock; | |
779 | case GETPID: | |
780 | err = curr->sempid; | |
781 | goto out_unlock; | |
782 | case GETNCNT: | |
783 | err = count_semncnt(sma,semnum); | |
784 | goto out_unlock; | |
785 | case GETZCNT: | |
786 | err = count_semzcnt(sma,semnum); | |
787 | goto out_unlock; | |
788 | case SETVAL: | |
789 | { | |
790 | int val = arg.val; | |
791 | struct sem_undo *un; | |
792 | err = -ERANGE; | |
793 | if (val > SEMVMX || val < 0) | |
794 | goto out_unlock; | |
795 | ||
796 | for (un = sma->undo; un; un = un->id_next) | |
797 | un->semadj[semnum] = 0; | |
798 | curr->semval = val; | |
b488893a | 799 | curr->sempid = task_tgid_vnr(current); |
1da177e4 LT |
800 | sma->sem_ctime = get_seconds(); |
801 | /* maybe some queued-up processes were waiting for this */ | |
802 | update_queue(sma); | |
803 | err = 0; | |
804 | goto out_unlock; | |
805 | } | |
806 | } | |
807 | out_unlock: | |
808 | sem_unlock(sma); | |
809 | out_free: | |
810 | if(sem_io != fast_sem_io) | |
811 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
812 | return err; | |
813 | } | |
814 | ||
815 | struct sem_setbuf { | |
816 | uid_t uid; | |
817 | gid_t gid; | |
818 | mode_t mode; | |
819 | }; | |
820 | ||
821 | static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version) | |
822 | { | |
823 | switch(version) { | |
824 | case IPC_64: | |
825 | { | |
826 | struct semid64_ds tbuf; | |
827 | ||
828 | if(copy_from_user(&tbuf, buf, sizeof(tbuf))) | |
829 | return -EFAULT; | |
830 | ||
831 | out->uid = tbuf.sem_perm.uid; | |
832 | out->gid = tbuf.sem_perm.gid; | |
833 | out->mode = tbuf.sem_perm.mode; | |
834 | ||
835 | return 0; | |
836 | } | |
837 | case IPC_OLD: | |
838 | { | |
839 | struct semid_ds tbuf_old; | |
840 | ||
841 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
842 | return -EFAULT; | |
843 | ||
844 | out->uid = tbuf_old.sem_perm.uid; | |
845 | out->gid = tbuf_old.sem_perm.gid; | |
846 | out->mode = tbuf_old.sem_perm.mode; | |
847 | ||
848 | return 0; | |
849 | } | |
850 | default: | |
851 | return -EINVAL; | |
852 | } | |
853 | } | |
854 | ||
e3893534 KK |
855 | static int semctl_down(struct ipc_namespace *ns, int semid, int semnum, |
856 | int cmd, int version, union semun arg) | |
1da177e4 LT |
857 | { |
858 | struct sem_array *sma; | |
859 | int err; | |
8e1c091c | 860 | struct sem_setbuf uninitialized_var(setbuf); |
1da177e4 LT |
861 | struct kern_ipc_perm *ipcp; |
862 | ||
863 | if(cmd == IPC_SET) { | |
864 | if(copy_semid_from_user (&setbuf, arg.buf, version)) | |
865 | return -EFAULT; | |
1da177e4 | 866 | } |
e3893534 | 867 | sma = sem_lock(ns, semid); |
1da177e4 LT |
868 | if(sma==NULL) |
869 | return -EINVAL; | |
870 | ||
e3893534 | 871 | if (sem_checkid(ns,sma,semid)) { |
1da177e4 LT |
872 | err=-EIDRM; |
873 | goto out_unlock; | |
874 | } | |
875 | ipcp = &sma->sem_perm; | |
073115d6 SG |
876 | |
877 | err = audit_ipc_obj(ipcp); | |
878 | if (err) | |
879 | goto out_unlock; | |
880 | ||
ac03221a LK |
881 | if (cmd == IPC_SET) { |
882 | err = audit_ipc_set_perm(0, setbuf.uid, setbuf.gid, setbuf.mode); | |
883 | if (err) | |
884 | goto out_unlock; | |
885 | } | |
1da177e4 LT |
886 | if (current->euid != ipcp->cuid && |
887 | current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) { | |
888 | err=-EPERM; | |
889 | goto out_unlock; | |
890 | } | |
891 | ||
892 | err = security_sem_semctl(sma, cmd); | |
893 | if (err) | |
894 | goto out_unlock; | |
895 | ||
896 | switch(cmd){ | |
897 | case IPC_RMID: | |
7ca7e564 | 898 | freeary(ns, sma); |
1da177e4 LT |
899 | err = 0; |
900 | break; | |
901 | case IPC_SET: | |
902 | ipcp->uid = setbuf.uid; | |
903 | ipcp->gid = setbuf.gid; | |
904 | ipcp->mode = (ipcp->mode & ~S_IRWXUGO) | |
905 | | (setbuf.mode & S_IRWXUGO); | |
906 | sma->sem_ctime = get_seconds(); | |
907 | sem_unlock(sma); | |
908 | err = 0; | |
909 | break; | |
910 | default: | |
911 | sem_unlock(sma); | |
912 | err = -EINVAL; | |
913 | break; | |
914 | } | |
915 | return err; | |
916 | ||
917 | out_unlock: | |
918 | sem_unlock(sma); | |
919 | return err; | |
920 | } | |
921 | ||
922 | asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg) | |
923 | { | |
924 | int err = -EINVAL; | |
925 | int version; | |
e3893534 | 926 | struct ipc_namespace *ns; |
1da177e4 LT |
927 | |
928 | if (semid < 0) | |
929 | return -EINVAL; | |
930 | ||
931 | version = ipc_parse_version(&cmd); | |
e3893534 | 932 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
933 | |
934 | switch(cmd) { | |
935 | case IPC_INFO: | |
936 | case SEM_INFO: | |
937 | case SEM_STAT: | |
e3893534 | 938 | err = semctl_nolock(ns,semid,semnum,cmd,version,arg); |
1da177e4 LT |
939 | return err; |
940 | case GETALL: | |
941 | case GETVAL: | |
942 | case GETPID: | |
943 | case GETNCNT: | |
944 | case GETZCNT: | |
945 | case IPC_STAT: | |
946 | case SETVAL: | |
947 | case SETALL: | |
e3893534 | 948 | err = semctl_main(ns,semid,semnum,cmd,version,arg); |
1da177e4 LT |
949 | return err; |
950 | case IPC_RMID: | |
951 | case IPC_SET: | |
e3893534 KK |
952 | mutex_lock(&sem_ids(ns).mutex); |
953 | err = semctl_down(ns,semid,semnum,cmd,version,arg); | |
954 | mutex_unlock(&sem_ids(ns).mutex); | |
1da177e4 LT |
955 | return err; |
956 | default: | |
957 | return -EINVAL; | |
958 | } | |
959 | } | |
960 | ||
961 | static inline void lock_semundo(void) | |
962 | { | |
963 | struct sem_undo_list *undo_list; | |
964 | ||
965 | undo_list = current->sysvsem.undo_list; | |
00a5dfdb | 966 | if (undo_list) |
1da177e4 LT |
967 | spin_lock(&undo_list->lock); |
968 | } | |
969 | ||
970 | /* This code has an interaction with copy_semundo(). | |
971 | * Consider; two tasks are sharing the undo_list. task1 | |
972 | * acquires the undo_list lock in lock_semundo(). If task2 now | |
973 | * exits before task1 releases the lock (by calling | |
974 | * unlock_semundo()), then task1 will never call spin_unlock(). | |
975 | * This leave the sem_undo_list in a locked state. If task1 now creats task3 | |
976 | * and once again shares the sem_undo_list, the sem_undo_list will still be | |
977 | * locked, and future SEM_UNDO operations will deadlock. This case is | |
978 | * dealt with in copy_semundo() by having it reinitialize the spin lock when | |
979 | * the refcnt goes from 1 to 2. | |
980 | */ | |
981 | static inline void unlock_semundo(void) | |
982 | { | |
983 | struct sem_undo_list *undo_list; | |
984 | ||
985 | undo_list = current->sysvsem.undo_list; | |
00a5dfdb | 986 | if (undo_list) |
1da177e4 LT |
987 | spin_unlock(&undo_list->lock); |
988 | } | |
989 | ||
990 | ||
991 | /* If the task doesn't already have a undo_list, then allocate one | |
992 | * here. We guarantee there is only one thread using this undo list, | |
993 | * and current is THE ONE | |
994 | * | |
995 | * If this allocation and assignment succeeds, but later | |
996 | * portions of this code fail, there is no need to free the sem_undo_list. | |
997 | * Just let it stay associated with the task, and it'll be freed later | |
998 | * at exit time. | |
999 | * | |
1000 | * This can block, so callers must hold no locks. | |
1001 | */ | |
1002 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1003 | { | |
1004 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1005 | |
1006 | undo_list = current->sysvsem.undo_list; | |
1007 | if (!undo_list) { | |
2453a306 | 1008 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1009 | if (undo_list == NULL) |
1010 | return -ENOMEM; | |
00a5dfdb | 1011 | spin_lock_init(&undo_list->lock); |
1da177e4 LT |
1012 | atomic_set(&undo_list->refcnt, 1); |
1013 | current->sysvsem.undo_list = undo_list; | |
1014 | } | |
1015 | *undo_listp = undo_list; | |
1016 | return 0; | |
1017 | } | |
1018 | ||
1019 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) | |
1020 | { | |
1021 | struct sem_undo **last, *un; | |
1022 | ||
1023 | last = &ulp->proc_list; | |
1024 | un = *last; | |
1025 | while(un != NULL) { | |
1026 | if(un->semid==semid) | |
1027 | break; | |
1028 | if(un->semid==-1) { | |
1029 | *last=un->proc_next; | |
1030 | kfree(un); | |
1031 | } else { | |
1032 | last=&un->proc_next; | |
1033 | } | |
1034 | un=*last; | |
1035 | } | |
1036 | return un; | |
1037 | } | |
1038 | ||
e3893534 | 1039 | static struct sem_undo *find_undo(struct ipc_namespace *ns, int semid) |
1da177e4 LT |
1040 | { |
1041 | struct sem_array *sma; | |
1042 | struct sem_undo_list *ulp; | |
1043 | struct sem_undo *un, *new; | |
1044 | int nsems; | |
1045 | int error; | |
1046 | ||
1047 | error = get_undo_list(&ulp); | |
1048 | if (error) | |
1049 | return ERR_PTR(error); | |
1050 | ||
1051 | lock_semundo(); | |
1052 | un = lookup_undo(ulp, semid); | |
1053 | unlock_semundo(); | |
1054 | if (likely(un!=NULL)) | |
1055 | goto out; | |
1056 | ||
1057 | /* no undo structure around - allocate one. */ | |
e3893534 | 1058 | sma = sem_lock(ns, semid); |
1da177e4 LT |
1059 | un = ERR_PTR(-EINVAL); |
1060 | if(sma==NULL) | |
1061 | goto out; | |
1062 | un = ERR_PTR(-EIDRM); | |
e3893534 | 1063 | if (sem_checkid(ns,sma,semid)) { |
1da177e4 LT |
1064 | sem_unlock(sma); |
1065 | goto out; | |
1066 | } | |
1067 | nsems = sma->sem_nsems; | |
1068 | ipc_rcu_getref(sma); | |
1069 | sem_unlock(sma); | |
1070 | ||
4668edc3 | 1071 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 LT |
1072 | if (!new) { |
1073 | ipc_lock_by_ptr(&sma->sem_perm); | |
1074 | ipc_rcu_putref(sma); | |
1075 | sem_unlock(sma); | |
1076 | return ERR_PTR(-ENOMEM); | |
1077 | } | |
1da177e4 LT |
1078 | new->semadj = (short *) &new[1]; |
1079 | new->semid = semid; | |
1080 | ||
1081 | lock_semundo(); | |
1082 | un = lookup_undo(ulp, semid); | |
1083 | if (un) { | |
1084 | unlock_semundo(); | |
1085 | kfree(new); | |
1086 | ipc_lock_by_ptr(&sma->sem_perm); | |
1087 | ipc_rcu_putref(sma); | |
1088 | sem_unlock(sma); | |
1089 | goto out; | |
1090 | } | |
1091 | ipc_lock_by_ptr(&sma->sem_perm); | |
1092 | ipc_rcu_putref(sma); | |
1093 | if (sma->sem_perm.deleted) { | |
1094 | sem_unlock(sma); | |
1095 | unlock_semundo(); | |
1096 | kfree(new); | |
1097 | un = ERR_PTR(-EIDRM); | |
1098 | goto out; | |
1099 | } | |
1100 | new->proc_next = ulp->proc_list; | |
1101 | ulp->proc_list = new; | |
1102 | new->id_next = sma->undo; | |
1103 | sma->undo = new; | |
1104 | sem_unlock(sma); | |
1105 | un = new; | |
1106 | unlock_semundo(); | |
1107 | out: | |
1108 | return un; | |
1109 | } | |
1110 | ||
1111 | asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops, | |
1112 | unsigned nsops, const struct timespec __user *timeout) | |
1113 | { | |
1114 | int error = -EINVAL; | |
1115 | struct sem_array *sma; | |
1116 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1117 | struct sembuf* sops = fast_sops, *sop; | |
1118 | struct sem_undo *un; | |
b78755ab | 1119 | int undos = 0, alter = 0, max; |
1da177e4 LT |
1120 | struct sem_queue queue; |
1121 | unsigned long jiffies_left = 0; | |
e3893534 KK |
1122 | struct ipc_namespace *ns; |
1123 | ||
1124 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1125 | |
1126 | if (nsops < 1 || semid < 0) | |
1127 | return -EINVAL; | |
e3893534 | 1128 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1129 | return -E2BIG; |
1130 | if(nsops > SEMOPM_FAST) { | |
1131 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1132 | if(sops==NULL) | |
1133 | return -ENOMEM; | |
1134 | } | |
1135 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1136 | error=-EFAULT; | |
1137 | goto out_free; | |
1138 | } | |
1139 | if (timeout) { | |
1140 | struct timespec _timeout; | |
1141 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1142 | error = -EFAULT; | |
1143 | goto out_free; | |
1144 | } | |
1145 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1146 | _timeout.tv_nsec >= 1000000000L) { | |
1147 | error = -EINVAL; | |
1148 | goto out_free; | |
1149 | } | |
1150 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1151 | } | |
1152 | max = 0; | |
1153 | for (sop = sops; sop < sops + nsops; sop++) { | |
1154 | if (sop->sem_num >= max) | |
1155 | max = sop->sem_num; | |
1156 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1157 | undos = 1; |
1158 | if (sop->sem_op != 0) | |
1da177e4 LT |
1159 | alter = 1; |
1160 | } | |
1da177e4 LT |
1161 | |
1162 | retry_undos: | |
1163 | if (undos) { | |
e3893534 | 1164 | un = find_undo(ns, semid); |
1da177e4 LT |
1165 | if (IS_ERR(un)) { |
1166 | error = PTR_ERR(un); | |
1167 | goto out_free; | |
1168 | } | |
1169 | } else | |
1170 | un = NULL; | |
1171 | ||
e3893534 | 1172 | sma = sem_lock(ns, semid); |
1da177e4 LT |
1173 | error=-EINVAL; |
1174 | if(sma==NULL) | |
1175 | goto out_free; | |
1176 | error = -EIDRM; | |
e3893534 | 1177 | if (sem_checkid(ns,sma,semid)) |
1da177e4 LT |
1178 | goto out_unlock_free; |
1179 | /* | |
1180 | * semid identifies are not unique - find_undo may have | |
1181 | * allocated an undo structure, it was invalidated by an RMID | |
1182 | * and now a new array with received the same id. Check and retry. | |
1183 | */ | |
1184 | if (un && un->semid == -1) { | |
1185 | sem_unlock(sma); | |
1186 | goto retry_undos; | |
1187 | } | |
1188 | error = -EFBIG; | |
1189 | if (max >= sma->sem_nsems) | |
1190 | goto out_unlock_free; | |
1191 | ||
1192 | error = -EACCES; | |
1193 | if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) | |
1194 | goto out_unlock_free; | |
1195 | ||
1196 | error = security_sem_semop(sma, sops, nsops, alter); | |
1197 | if (error) | |
1198 | goto out_unlock_free; | |
1199 | ||
b488893a | 1200 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1201 | if (error <= 0) { |
1202 | if (alter && error == 0) | |
1203 | update_queue (sma); | |
1204 | goto out_unlock_free; | |
1205 | } | |
1206 | ||
1207 | /* We need to sleep on this operation, so we put the current | |
1208 | * task into the pending queue and go to sleep. | |
1209 | */ | |
1210 | ||
1211 | queue.sma = sma; | |
1212 | queue.sops = sops; | |
1213 | queue.nsops = nsops; | |
1214 | queue.undo = un; | |
b488893a | 1215 | queue.pid = task_tgid_vnr(current); |
1da177e4 LT |
1216 | queue.id = semid; |
1217 | queue.alter = alter; | |
1218 | if (alter) | |
1219 | append_to_queue(sma ,&queue); | |
1220 | else | |
1221 | prepend_to_queue(sma ,&queue); | |
1222 | ||
1223 | queue.status = -EINTR; | |
1224 | queue.sleeper = current; | |
1225 | current->state = TASK_INTERRUPTIBLE; | |
1226 | sem_unlock(sma); | |
1227 | ||
1228 | if (timeout) | |
1229 | jiffies_left = schedule_timeout(jiffies_left); | |
1230 | else | |
1231 | schedule(); | |
1232 | ||
1233 | error = queue.status; | |
1234 | while(unlikely(error == IN_WAKEUP)) { | |
1235 | cpu_relax(); | |
1236 | error = queue.status; | |
1237 | } | |
1238 | ||
1239 | if (error != -EINTR) { | |
1240 | /* fast path: update_queue already obtained all requested | |
1241 | * resources */ | |
1242 | goto out_free; | |
1243 | } | |
1244 | ||
e3893534 | 1245 | sma = sem_lock(ns, semid); |
1da177e4 | 1246 | if(sma==NULL) { |
27315c96 | 1247 | BUG_ON(queue.prev != NULL); |
1da177e4 LT |
1248 | error = -EIDRM; |
1249 | goto out_free; | |
1250 | } | |
1251 | ||
1252 | /* | |
1253 | * If queue.status != -EINTR we are woken up by another process | |
1254 | */ | |
1255 | error = queue.status; | |
1256 | if (error != -EINTR) { | |
1257 | goto out_unlock_free; | |
1258 | } | |
1259 | ||
1260 | /* | |
1261 | * If an interrupt occurred we have to clean up the queue | |
1262 | */ | |
1263 | if (timeout && jiffies_left == 0) | |
1264 | error = -EAGAIN; | |
1265 | remove_from_queue(sma,&queue); | |
1266 | goto out_unlock_free; | |
1267 | ||
1268 | out_unlock_free: | |
1269 | sem_unlock(sma); | |
1270 | out_free: | |
1271 | if(sops != fast_sops) | |
1272 | kfree(sops); | |
1273 | return error; | |
1274 | } | |
1275 | ||
1276 | asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops) | |
1277 | { | |
1278 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1279 | } | |
1280 | ||
1281 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1282 | * parent and child tasks. | |
1283 | * | |
1284 | * See the notes above unlock_semundo() regarding the spin_lock_init() | |
1285 | * in this code. Initialize the undo_list->lock here instead of get_undo_list() | |
1286 | * because of the reasoning in the comment above unlock_semundo. | |
1287 | */ | |
1288 | ||
1289 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1290 | { | |
1291 | struct sem_undo_list *undo_list; | |
1292 | int error; | |
1293 | ||
1294 | if (clone_flags & CLONE_SYSVSEM) { | |
1295 | error = get_undo_list(&undo_list); | |
1296 | if (error) | |
1297 | return error; | |
1da177e4 LT |
1298 | atomic_inc(&undo_list->refcnt); |
1299 | tsk->sysvsem.undo_list = undo_list; | |
1300 | } else | |
1301 | tsk->sysvsem.undo_list = NULL; | |
1302 | ||
1303 | return 0; | |
1304 | } | |
1305 | ||
1306 | /* | |
1307 | * add semadj values to semaphores, free undo structures. | |
1308 | * undo structures are not freed when semaphore arrays are destroyed | |
1309 | * so some of them may be out of date. | |
1310 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1311 | * set of adjustments that needs to be done should be done in an atomic | |
1312 | * manner or not. That is, if we are attempting to decrement the semval | |
1313 | * should we queue up and wait until we can do so legally? | |
1314 | * The original implementation attempted to do this (queue and wait). | |
1315 | * The current implementation does not do so. The POSIX standard | |
1316 | * and SVID should be consulted to determine what behavior is mandated. | |
1317 | */ | |
1318 | void exit_sem(struct task_struct *tsk) | |
1319 | { | |
1320 | struct sem_undo_list *undo_list; | |
1321 | struct sem_undo *u, **up; | |
e3893534 | 1322 | struct ipc_namespace *ns; |
1da177e4 LT |
1323 | |
1324 | undo_list = tsk->sysvsem.undo_list; | |
1325 | if (!undo_list) | |
1326 | return; | |
1327 | ||
1328 | if (!atomic_dec_and_test(&undo_list->refcnt)) | |
1329 | return; | |
1330 | ||
e3893534 | 1331 | ns = tsk->nsproxy->ipc_ns; |
1da177e4 LT |
1332 | /* There's no need to hold the semundo list lock, as current |
1333 | * is the last task exiting for this undo list. | |
1334 | */ | |
1335 | for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) { | |
1336 | struct sem_array *sma; | |
1337 | int nsems, i; | |
1338 | struct sem_undo *un, **unp; | |
1339 | int semid; | |
1340 | ||
1341 | semid = u->semid; | |
1342 | ||
1343 | if(semid == -1) | |
1344 | continue; | |
e3893534 | 1345 | sma = sem_lock(ns, semid); |
1da177e4 LT |
1346 | if (sma == NULL) |
1347 | continue; | |
1348 | ||
1349 | if (u->semid == -1) | |
1350 | goto next_entry; | |
1351 | ||
e3893534 | 1352 | BUG_ON(sem_checkid(ns,sma,u->semid)); |
1da177e4 LT |
1353 | |
1354 | /* remove u from the sma->undo list */ | |
1355 | for (unp = &sma->undo; (un = *unp); unp = &un->id_next) { | |
1356 | if (u == un) | |
1357 | goto found; | |
1358 | } | |
1359 | printk ("exit_sem undo list error id=%d\n", u->semid); | |
1360 | goto next_entry; | |
1361 | found: | |
1362 | *unp = un->id_next; | |
1363 | /* perform adjustments registered in u */ | |
1364 | nsems = sma->sem_nsems; | |
1365 | for (i = 0; i < nsems; i++) { | |
5f921ae9 | 1366 | struct sem * semaphore = &sma->sem_base[i]; |
1da177e4 | 1367 | if (u->semadj[i]) { |
5f921ae9 | 1368 | semaphore->semval += u->semadj[i]; |
1da177e4 LT |
1369 | /* |
1370 | * Range checks of the new semaphore value, | |
1371 | * not defined by sus: | |
1372 | * - Some unices ignore the undo entirely | |
1373 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
1374 | * - some cap the value (e.g. FreeBSD caps | |
1375 | * at 0, but doesn't enforce SEMVMX) | |
1376 | * | |
1377 | * Linux caps the semaphore value, both at 0 | |
1378 | * and at SEMVMX. | |
1379 | * | |
1380 | * Manfred <manfred@colorfullife.com> | |
1381 | */ | |
5f921ae9 IM |
1382 | if (semaphore->semval < 0) |
1383 | semaphore->semval = 0; | |
1384 | if (semaphore->semval > SEMVMX) | |
1385 | semaphore->semval = SEMVMX; | |
b488893a | 1386 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
1387 | } |
1388 | } | |
1389 | sma->sem_otime = get_seconds(); | |
1390 | /* maybe some queued-up processes were waiting for this */ | |
1391 | update_queue(sma); | |
1392 | next_entry: | |
1393 | sem_unlock(sma); | |
1394 | } | |
1395 | kfree(undo_list); | |
1396 | } | |
1397 | ||
1398 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1399 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1400 | { |
19b4946c MW |
1401 | struct sem_array *sma = it; |
1402 | ||
1403 | return seq_printf(s, | |
1404 | "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n", | |
1405 | sma->sem_perm.key, | |
7ca7e564 | 1406 | sma->sem_perm.id, |
19b4946c MW |
1407 | sma->sem_perm.mode, |
1408 | sma->sem_nsems, | |
1409 | sma->sem_perm.uid, | |
1410 | sma->sem_perm.gid, | |
1411 | sma->sem_perm.cuid, | |
1412 | sma->sem_perm.cgid, | |
1413 | sma->sem_otime, | |
1414 | sma->sem_ctime); | |
1da177e4 LT |
1415 | } |
1416 | #endif |