]>
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; | |
b97e820f | 244 | int i; |
1da177e4 LT |
245 | |
246 | if (!nsems) | |
247 | return -EINVAL; | |
e3893534 | 248 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
249 | return -ENOSPC; |
250 | ||
251 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
252 | sma = ipc_rcu_alloc(size); | |
253 | if (!sma) { | |
254 | return -ENOMEM; | |
255 | } | |
256 | memset (sma, 0, size); | |
257 | ||
258 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
259 | sma->sem_perm.key = key; | |
260 | ||
261 | sma->sem_perm.security = NULL; | |
262 | retval = security_sem_alloc(sma); | |
263 | if (retval) { | |
264 | ipc_rcu_putref(sma); | |
265 | return retval; | |
266 | } | |
267 | ||
e3893534 | 268 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 269 | if (id < 0) { |
1da177e4 LT |
270 | security_sem_free(sma); |
271 | ipc_rcu_putref(sma); | |
283bb7fa | 272 | return id; |
1da177e4 | 273 | } |
e3893534 | 274 | ns->used_sems += nsems; |
1da177e4 LT |
275 | |
276 | sma->sem_base = (struct sem *) &sma[1]; | |
b97e820f MS |
277 | |
278 | for (i = 0; i < nsems; i++) | |
279 | INIT_LIST_HEAD(&sma->sem_base[i].sem_pending); | |
280 | ||
281 | sma->complex_count = 0; | |
a1193f8e | 282 | INIT_LIST_HEAD(&sma->sem_pending); |
4daa28f6 | 283 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
284 | sma->sem_nsems = nsems; |
285 | sma->sem_ctime = get_seconds(); | |
286 | sem_unlock(sma); | |
287 | ||
7ca7e564 | 288 | return sma->sem_perm.id; |
1da177e4 LT |
289 | } |
290 | ||
7748dbfa | 291 | |
f4566f04 | 292 | /* |
3e148c79 | 293 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 294 | */ |
03f02c76 | 295 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 296 | { |
03f02c76 ND |
297 | struct sem_array *sma; |
298 | ||
299 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
300 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
301 | } |
302 | ||
f4566f04 | 303 | /* |
3e148c79 | 304 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 305 | */ |
03f02c76 ND |
306 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
307 | struct ipc_params *params) | |
7748dbfa | 308 | { |
03f02c76 ND |
309 | struct sem_array *sma; |
310 | ||
311 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
312 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
313 | return -EINVAL; |
314 | ||
315 | return 0; | |
316 | } | |
317 | ||
d5460c99 | 318 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 319 | { |
e3893534 | 320 | struct ipc_namespace *ns; |
7748dbfa ND |
321 | struct ipc_ops sem_ops; |
322 | struct ipc_params sem_params; | |
e3893534 KK |
323 | |
324 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 325 | |
e3893534 | 326 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 327 | return -EINVAL; |
7ca7e564 | 328 | |
7748dbfa ND |
329 | sem_ops.getnew = newary; |
330 | sem_ops.associate = sem_security; | |
331 | sem_ops.more_checks = sem_more_checks; | |
332 | ||
333 | sem_params.key = key; | |
334 | sem_params.flg = semflg; | |
335 | sem_params.u.nsems = nsems; | |
1da177e4 | 336 | |
7748dbfa | 337 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
338 | } |
339 | ||
1da177e4 LT |
340 | /* |
341 | * Determine whether a sequence of semaphore operations would succeed | |
342 | * all at once. Return 0 if yes, 1 if need to sleep, else return error code. | |
343 | */ | |
344 | ||
345 | static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, | |
346 | int nsops, struct sem_undo *un, int pid) | |
347 | { | |
348 | int result, sem_op; | |
349 | struct sembuf *sop; | |
350 | struct sem * curr; | |
351 | ||
352 | for (sop = sops; sop < sops + nsops; sop++) { | |
353 | curr = sma->sem_base + sop->sem_num; | |
354 | sem_op = sop->sem_op; | |
355 | result = curr->semval; | |
356 | ||
357 | if (!sem_op && result) | |
358 | goto would_block; | |
359 | ||
360 | result += sem_op; | |
361 | if (result < 0) | |
362 | goto would_block; | |
363 | if (result > SEMVMX) | |
364 | goto out_of_range; | |
365 | if (sop->sem_flg & SEM_UNDO) { | |
366 | int undo = un->semadj[sop->sem_num] - sem_op; | |
367 | /* | |
368 | * Exceeding the undo range is an error. | |
369 | */ | |
370 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
371 | goto out_of_range; | |
372 | } | |
373 | curr->semval = result; | |
374 | } | |
375 | ||
376 | sop--; | |
377 | while (sop >= sops) { | |
378 | sma->sem_base[sop->sem_num].sempid = pid; | |
379 | if (sop->sem_flg & SEM_UNDO) | |
380 | un->semadj[sop->sem_num] -= sop->sem_op; | |
381 | sop--; | |
382 | } | |
383 | ||
384 | sma->sem_otime = get_seconds(); | |
385 | return 0; | |
386 | ||
387 | out_of_range: | |
388 | result = -ERANGE; | |
389 | goto undo; | |
390 | ||
391 | would_block: | |
392 | if (sop->sem_flg & IPC_NOWAIT) | |
393 | result = -EAGAIN; | |
394 | else | |
395 | result = 1; | |
396 | ||
397 | undo: | |
398 | sop--; | |
399 | while (sop >= sops) { | |
400 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; | |
401 | sop--; | |
402 | } | |
403 | ||
404 | return result; | |
405 | } | |
406 | ||
d4212093 NP |
407 | /* |
408 | * Wake up a process waiting on the sem queue with a given error. | |
409 | * The queue is invalid (may not be accessed) after the function returns. | |
410 | */ | |
411 | static void wake_up_sem_queue(struct sem_queue *q, int error) | |
412 | { | |
413 | /* | |
414 | * Hold preempt off so that we don't get preempted and have the | |
415 | * wakee busy-wait until we're scheduled back on. We're holding | |
416 | * locks here so it may not strictly be needed, however if the | |
417 | * locks become preemptible then this prevents such a problem. | |
418 | */ | |
419 | preempt_disable(); | |
420 | q->status = IN_WAKEUP; | |
421 | wake_up_process(q->sleeper); | |
422 | /* hands-off: q can disappear immediately after writing q->status. */ | |
423 | smp_wmb(); | |
424 | q->status = error; | |
425 | preempt_enable(); | |
426 | } | |
427 | ||
b97e820f MS |
428 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
429 | { | |
430 | list_del(&q->list); | |
431 | if (q->nsops == 1) | |
432 | list_del(&q->simple_list); | |
433 | else | |
434 | sma->complex_count--; | |
435 | } | |
436 | ||
1da177e4 LT |
437 | /* Go through the pending queue for the indicated semaphore |
438 | * looking for tasks that can be completed. | |
439 | */ | |
440 | static void update_queue (struct sem_array * sma) | |
441 | { | |
9cad200c NP |
442 | struct sem_queue *q, *tq; |
443 | ||
444 | again: | |
445 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { | |
446 | int error; | |
447 | int alter; | |
1da177e4 | 448 | |
1da177e4 LT |
449 | error = try_atomic_semop(sma, q->sops, q->nsops, |
450 | q->undo, q->pid); | |
451 | ||
452 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
453 | if (error > 0) |
454 | continue; | |
455 | ||
b97e820f | 456 | unlink_queue(sma, q); |
9cad200c NP |
457 | |
458 | /* | |
459 | * The next operation that must be checked depends on the type | |
460 | * of the completed operation: | |
461 | * - if the operation modified the array, then restart from the | |
462 | * head of the queue and check for threads that might be | |
b6e90822 | 463 | * waiting for the new semaphore values. |
9cad200c NP |
464 | * - if the operation didn't modify the array, then just |
465 | * continue. | |
466 | */ | |
467 | alter = q->alter; | |
d4212093 | 468 | wake_up_sem_queue(q, error); |
b6e90822 | 469 | if (alter && !error) |
9cad200c | 470 | goto again; |
1da177e4 LT |
471 | } |
472 | } | |
473 | ||
474 | /* The following counts are associated to each semaphore: | |
475 | * semncnt number of tasks waiting on semval being nonzero | |
476 | * semzcnt number of tasks waiting on semval being zero | |
477 | * This model assumes that a task waits on exactly one semaphore. | |
478 | * Since semaphore operations are to be performed atomically, tasks actually | |
479 | * wait on a whole sequence of semaphores simultaneously. | |
480 | * The counts we return here are a rough approximation, but still | |
481 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
482 | */ | |
483 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
484 | { | |
485 | int semncnt; | |
486 | struct sem_queue * q; | |
487 | ||
488 | semncnt = 0; | |
a1193f8e | 489 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
490 | struct sembuf * sops = q->sops; |
491 | int nsops = q->nsops; | |
492 | int i; | |
493 | for (i = 0; i < nsops; i++) | |
494 | if (sops[i].sem_num == semnum | |
495 | && (sops[i].sem_op < 0) | |
496 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
497 | semncnt++; | |
498 | } | |
499 | return semncnt; | |
500 | } | |
a1193f8e | 501 | |
1da177e4 LT |
502 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
503 | { | |
504 | int semzcnt; | |
505 | struct sem_queue * q; | |
506 | ||
507 | semzcnt = 0; | |
a1193f8e | 508 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
509 | struct sembuf * sops = q->sops; |
510 | int nsops = q->nsops; | |
511 | int i; | |
512 | for (i = 0; i < nsops; i++) | |
513 | if (sops[i].sem_num == semnum | |
514 | && (sops[i].sem_op == 0) | |
515 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
516 | semzcnt++; | |
517 | } | |
518 | return semzcnt; | |
519 | } | |
520 | ||
6d97e234 | 521 | static void free_un(struct rcu_head *head) |
380af1b3 MS |
522 | { |
523 | struct sem_undo *un = container_of(head, struct sem_undo, rcu); | |
524 | kfree(un); | |
525 | } | |
526 | ||
3e148c79 ND |
527 | /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked |
528 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex | |
529 | * remains locked on exit. | |
1da177e4 | 530 | */ |
01b8b07a | 531 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 532 | { |
380af1b3 MS |
533 | struct sem_undo *un, *tu; |
534 | struct sem_queue *q, *tq; | |
01b8b07a | 535 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 | 536 | |
380af1b3 | 537 | /* Free the existing undo structures for this semaphore set. */ |
4daa28f6 | 538 | assert_spin_locked(&sma->sem_perm.lock); |
380af1b3 MS |
539 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
540 | list_del(&un->list_id); | |
541 | spin_lock(&un->ulp->lock); | |
1da177e4 | 542 | un->semid = -1; |
380af1b3 MS |
543 | list_del_rcu(&un->list_proc); |
544 | spin_unlock(&un->ulp->lock); | |
545 | call_rcu(&un->rcu, free_un); | |
546 | } | |
1da177e4 LT |
547 | |
548 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
380af1b3 | 549 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { |
b97e820f | 550 | unlink_queue(sma, q); |
d4212093 | 551 | wake_up_sem_queue(q, -EIDRM); |
1da177e4 LT |
552 | } |
553 | ||
7ca7e564 ND |
554 | /* Remove the semaphore set from the IDR */ |
555 | sem_rmid(ns, sma); | |
1da177e4 LT |
556 | sem_unlock(sma); |
557 | ||
e3893534 | 558 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
559 | security_sem_free(sma); |
560 | ipc_rcu_putref(sma); | |
561 | } | |
562 | ||
563 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
564 | { | |
565 | switch(version) { | |
566 | case IPC_64: | |
567 | return copy_to_user(buf, in, sizeof(*in)); | |
568 | case IPC_OLD: | |
569 | { | |
570 | struct semid_ds out; | |
571 | ||
572 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); | |
573 | ||
574 | out.sem_otime = in->sem_otime; | |
575 | out.sem_ctime = in->sem_ctime; | |
576 | out.sem_nsems = in->sem_nsems; | |
577 | ||
578 | return copy_to_user(buf, &out, sizeof(out)); | |
579 | } | |
580 | default: | |
581 | return -EINVAL; | |
582 | } | |
583 | } | |
584 | ||
4b9fcb0e PP |
585 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
586 | int cmd, int version, union semun arg) | |
1da177e4 LT |
587 | { |
588 | int err = -EINVAL; | |
589 | struct sem_array *sma; | |
590 | ||
591 | switch(cmd) { | |
592 | case IPC_INFO: | |
593 | case SEM_INFO: | |
594 | { | |
595 | struct seminfo seminfo; | |
596 | int max_id; | |
597 | ||
598 | err = security_sem_semctl(NULL, cmd); | |
599 | if (err) | |
600 | return err; | |
601 | ||
602 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
603 | seminfo.semmni = ns->sc_semmni; |
604 | seminfo.semmns = ns->sc_semmns; | |
605 | seminfo.semmsl = ns->sc_semmsl; | |
606 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
607 | seminfo.semvmx = SEMVMX; |
608 | seminfo.semmnu = SEMMNU; | |
609 | seminfo.semmap = SEMMAP; | |
610 | seminfo.semume = SEMUME; | |
3e148c79 | 611 | down_read(&sem_ids(ns).rw_mutex); |
1da177e4 | 612 | if (cmd == SEM_INFO) { |
e3893534 KK |
613 | seminfo.semusz = sem_ids(ns).in_use; |
614 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
615 | } else { |
616 | seminfo.semusz = SEMUSZ; | |
617 | seminfo.semaem = SEMAEM; | |
618 | } | |
7ca7e564 | 619 | max_id = ipc_get_maxid(&sem_ids(ns)); |
3e148c79 | 620 | up_read(&sem_ids(ns).rw_mutex); |
1da177e4 LT |
621 | if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) |
622 | return -EFAULT; | |
623 | return (max_id < 0) ? 0: max_id; | |
624 | } | |
4b9fcb0e | 625 | case IPC_STAT: |
1da177e4 LT |
626 | case SEM_STAT: |
627 | { | |
628 | struct semid64_ds tbuf; | |
629 | int id; | |
630 | ||
4b9fcb0e PP |
631 | if (cmd == SEM_STAT) { |
632 | sma = sem_lock(ns, semid); | |
633 | if (IS_ERR(sma)) | |
634 | return PTR_ERR(sma); | |
635 | id = sma->sem_perm.id; | |
636 | } else { | |
637 | sma = sem_lock_check(ns, semid); | |
638 | if (IS_ERR(sma)) | |
639 | return PTR_ERR(sma); | |
640 | id = 0; | |
641 | } | |
1da177e4 LT |
642 | |
643 | err = -EACCES; | |
644 | if (ipcperms (&sma->sem_perm, S_IRUGO)) | |
645 | goto out_unlock; | |
646 | ||
647 | err = security_sem_semctl(sma, cmd); | |
648 | if (err) | |
649 | goto out_unlock; | |
650 | ||
023a5355 ND |
651 | memset(&tbuf, 0, sizeof(tbuf)); |
652 | ||
1da177e4 LT |
653 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
654 | tbuf.sem_otime = sma->sem_otime; | |
655 | tbuf.sem_ctime = sma->sem_ctime; | |
656 | tbuf.sem_nsems = sma->sem_nsems; | |
657 | sem_unlock(sma); | |
658 | if (copy_semid_to_user (arg.buf, &tbuf, version)) | |
659 | return -EFAULT; | |
660 | return id; | |
661 | } | |
662 | default: | |
663 | return -EINVAL; | |
664 | } | |
665 | return err; | |
666 | out_unlock: | |
667 | sem_unlock(sma); | |
668 | return err; | |
669 | } | |
670 | ||
e3893534 KK |
671 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
672 | int cmd, int version, union semun arg) | |
1da177e4 LT |
673 | { |
674 | struct sem_array *sma; | |
675 | struct sem* curr; | |
676 | int err; | |
677 | ushort fast_sem_io[SEMMSL_FAST]; | |
678 | ushort* sem_io = fast_sem_io; | |
679 | int nsems; | |
680 | ||
023a5355 ND |
681 | sma = sem_lock_check(ns, semid); |
682 | if (IS_ERR(sma)) | |
683 | return PTR_ERR(sma); | |
1da177e4 LT |
684 | |
685 | nsems = sma->sem_nsems; | |
686 | ||
1da177e4 LT |
687 | err = -EACCES; |
688 | if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) | |
689 | goto out_unlock; | |
690 | ||
691 | err = security_sem_semctl(sma, cmd); | |
692 | if (err) | |
693 | goto out_unlock; | |
694 | ||
695 | err = -EACCES; | |
696 | switch (cmd) { | |
697 | case GETALL: | |
698 | { | |
699 | ushort __user *array = arg.array; | |
700 | int i; | |
701 | ||
702 | if(nsems > SEMMSL_FAST) { | |
6ff37972 | 703 | sem_getref_and_unlock(sma); |
1da177e4 LT |
704 | |
705 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
706 | if(sem_io == NULL) { | |
6ff37972 | 707 | sem_putref(sma); |
1da177e4 LT |
708 | return -ENOMEM; |
709 | } | |
710 | ||
6ff37972 | 711 | sem_lock_and_putref(sma); |
1da177e4 LT |
712 | if (sma->sem_perm.deleted) { |
713 | sem_unlock(sma); | |
714 | err = -EIDRM; | |
715 | goto out_free; | |
716 | } | |
717 | } | |
718 | ||
719 | for (i = 0; i < sma->sem_nsems; i++) | |
720 | sem_io[i] = sma->sem_base[i].semval; | |
721 | sem_unlock(sma); | |
722 | err = 0; | |
723 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
724 | err = -EFAULT; | |
725 | goto out_free; | |
726 | } | |
727 | case SETALL: | |
728 | { | |
729 | int i; | |
730 | struct sem_undo *un; | |
731 | ||
6ff37972 | 732 | sem_getref_and_unlock(sma); |
1da177e4 LT |
733 | |
734 | if(nsems > SEMMSL_FAST) { | |
735 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
736 | if(sem_io == NULL) { | |
6ff37972 | 737 | sem_putref(sma); |
1da177e4 LT |
738 | return -ENOMEM; |
739 | } | |
740 | } | |
741 | ||
742 | if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { | |
6ff37972 | 743 | sem_putref(sma); |
1da177e4 LT |
744 | err = -EFAULT; |
745 | goto out_free; | |
746 | } | |
747 | ||
748 | for (i = 0; i < nsems; i++) { | |
749 | if (sem_io[i] > SEMVMX) { | |
6ff37972 | 750 | sem_putref(sma); |
1da177e4 LT |
751 | err = -ERANGE; |
752 | goto out_free; | |
753 | } | |
754 | } | |
6ff37972 | 755 | sem_lock_and_putref(sma); |
1da177e4 LT |
756 | if (sma->sem_perm.deleted) { |
757 | sem_unlock(sma); | |
758 | err = -EIDRM; | |
759 | goto out_free; | |
760 | } | |
761 | ||
762 | for (i = 0; i < nsems; i++) | |
763 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 MS |
764 | |
765 | assert_spin_locked(&sma->sem_perm.lock); | |
766 | list_for_each_entry(un, &sma->list_id, list_id) { | |
1da177e4 LT |
767 | for (i = 0; i < nsems; i++) |
768 | un->semadj[i] = 0; | |
4daa28f6 | 769 | } |
1da177e4 LT |
770 | sma->sem_ctime = get_seconds(); |
771 | /* maybe some queued-up processes were waiting for this */ | |
772 | update_queue(sma); | |
773 | err = 0; | |
774 | goto out_unlock; | |
775 | } | |
1da177e4 LT |
776 | /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ |
777 | } | |
778 | err = -EINVAL; | |
779 | if(semnum < 0 || semnum >= nsems) | |
780 | goto out_unlock; | |
781 | ||
782 | curr = &sma->sem_base[semnum]; | |
783 | ||
784 | switch (cmd) { | |
785 | case GETVAL: | |
786 | err = curr->semval; | |
787 | goto out_unlock; | |
788 | case GETPID: | |
789 | err = curr->sempid; | |
790 | goto out_unlock; | |
791 | case GETNCNT: | |
792 | err = count_semncnt(sma,semnum); | |
793 | goto out_unlock; | |
794 | case GETZCNT: | |
795 | err = count_semzcnt(sma,semnum); | |
796 | goto out_unlock; | |
797 | case SETVAL: | |
798 | { | |
799 | int val = arg.val; | |
800 | struct sem_undo *un; | |
4daa28f6 | 801 | |
1da177e4 LT |
802 | err = -ERANGE; |
803 | if (val > SEMVMX || val < 0) | |
804 | goto out_unlock; | |
805 | ||
4daa28f6 MS |
806 | assert_spin_locked(&sma->sem_perm.lock); |
807 | list_for_each_entry(un, &sma->list_id, list_id) | |
1da177e4 | 808 | un->semadj[semnum] = 0; |
4daa28f6 | 809 | |
1da177e4 | 810 | curr->semval = val; |
b488893a | 811 | curr->sempid = task_tgid_vnr(current); |
1da177e4 LT |
812 | sma->sem_ctime = get_seconds(); |
813 | /* maybe some queued-up processes were waiting for this */ | |
814 | update_queue(sma); | |
815 | err = 0; | |
816 | goto out_unlock; | |
817 | } | |
818 | } | |
819 | out_unlock: | |
820 | sem_unlock(sma); | |
821 | out_free: | |
822 | if(sem_io != fast_sem_io) | |
823 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
824 | return err; | |
825 | } | |
826 | ||
016d7132 PP |
827 | static inline unsigned long |
828 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
829 | { |
830 | switch(version) { | |
831 | case IPC_64: | |
016d7132 | 832 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 833 | return -EFAULT; |
1da177e4 | 834 | return 0; |
1da177e4 LT |
835 | case IPC_OLD: |
836 | { | |
837 | struct semid_ds tbuf_old; | |
838 | ||
839 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
840 | return -EFAULT; | |
841 | ||
016d7132 PP |
842 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
843 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
844 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
845 | |
846 | return 0; | |
847 | } | |
848 | default: | |
849 | return -EINVAL; | |
850 | } | |
851 | } | |
852 | ||
522bb2a2 PP |
853 | /* |
854 | * This function handles some semctl commands which require the rw_mutex | |
855 | * to be held in write mode. | |
856 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
857 | */ | |
21a4826a PP |
858 | static int semctl_down(struct ipc_namespace *ns, int semid, |
859 | int cmd, int version, union semun arg) | |
1da177e4 LT |
860 | { |
861 | struct sem_array *sma; | |
862 | int err; | |
016d7132 | 863 | struct semid64_ds semid64; |
1da177e4 LT |
864 | struct kern_ipc_perm *ipcp; |
865 | ||
866 | if(cmd == IPC_SET) { | |
016d7132 | 867 | if (copy_semid_from_user(&semid64, arg.buf, version)) |
1da177e4 | 868 | return -EFAULT; |
1da177e4 | 869 | } |
073115d6 | 870 | |
a5f75e7f PP |
871 | ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0); |
872 | if (IS_ERR(ipcp)) | |
873 | return PTR_ERR(ipcp); | |
073115d6 | 874 | |
a5f75e7f | 875 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
876 | |
877 | err = security_sem_semctl(sma, cmd); | |
878 | if (err) | |
879 | goto out_unlock; | |
880 | ||
881 | switch(cmd){ | |
882 | case IPC_RMID: | |
01b8b07a | 883 | freeary(ns, ipcp); |
522bb2a2 | 884 | goto out_up; |
1da177e4 | 885 | case IPC_SET: |
8f4a3809 | 886 | ipc_update_perm(&semid64.sem_perm, ipcp); |
1da177e4 | 887 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
888 | break; |
889 | default: | |
1da177e4 | 890 | err = -EINVAL; |
1da177e4 | 891 | } |
1da177e4 LT |
892 | |
893 | out_unlock: | |
894 | sem_unlock(sma); | |
522bb2a2 PP |
895 | out_up: |
896 | up_write(&sem_ids(ns).rw_mutex); | |
1da177e4 LT |
897 | return err; |
898 | } | |
899 | ||
6673e0c3 | 900 | SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg) |
1da177e4 LT |
901 | { |
902 | int err = -EINVAL; | |
903 | int version; | |
e3893534 | 904 | struct ipc_namespace *ns; |
1da177e4 LT |
905 | |
906 | if (semid < 0) | |
907 | return -EINVAL; | |
908 | ||
909 | version = ipc_parse_version(&cmd); | |
e3893534 | 910 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
911 | |
912 | switch(cmd) { | |
913 | case IPC_INFO: | |
914 | case SEM_INFO: | |
4b9fcb0e | 915 | case IPC_STAT: |
1da177e4 | 916 | case SEM_STAT: |
4b9fcb0e | 917 | err = semctl_nolock(ns, semid, cmd, version, arg); |
1da177e4 LT |
918 | return err; |
919 | case GETALL: | |
920 | case GETVAL: | |
921 | case GETPID: | |
922 | case GETNCNT: | |
923 | case GETZCNT: | |
1da177e4 LT |
924 | case SETVAL: |
925 | case SETALL: | |
e3893534 | 926 | err = semctl_main(ns,semid,semnum,cmd,version,arg); |
1da177e4 LT |
927 | return err; |
928 | case IPC_RMID: | |
929 | case IPC_SET: | |
21a4826a | 930 | err = semctl_down(ns, semid, cmd, version, arg); |
1da177e4 LT |
931 | return err; |
932 | default: | |
933 | return -EINVAL; | |
934 | } | |
935 | } | |
6673e0c3 HC |
936 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
937 | asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg) | |
938 | { | |
939 | return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg); | |
940 | } | |
941 | SYSCALL_ALIAS(sys_semctl, SyS_semctl); | |
942 | #endif | |
1da177e4 | 943 | |
1da177e4 LT |
944 | /* If the task doesn't already have a undo_list, then allocate one |
945 | * here. We guarantee there is only one thread using this undo list, | |
946 | * and current is THE ONE | |
947 | * | |
948 | * If this allocation and assignment succeeds, but later | |
949 | * portions of this code fail, there is no need to free the sem_undo_list. | |
950 | * Just let it stay associated with the task, and it'll be freed later | |
951 | * at exit time. | |
952 | * | |
953 | * This can block, so callers must hold no locks. | |
954 | */ | |
955 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
956 | { | |
957 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
958 | |
959 | undo_list = current->sysvsem.undo_list; | |
960 | if (!undo_list) { | |
2453a306 | 961 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
962 | if (undo_list == NULL) |
963 | return -ENOMEM; | |
00a5dfdb | 964 | spin_lock_init(&undo_list->lock); |
1da177e4 | 965 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
966 | INIT_LIST_HEAD(&undo_list->list_proc); |
967 | ||
1da177e4 LT |
968 | current->sysvsem.undo_list = undo_list; |
969 | } | |
970 | *undo_listp = undo_list; | |
971 | return 0; | |
972 | } | |
973 | ||
bf17bb71 | 974 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 975 | { |
bf17bb71 | 976 | struct sem_undo *un; |
4daa28f6 | 977 | |
bf17bb71 NP |
978 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
979 | if (un->semid == semid) | |
980 | return un; | |
1da177e4 | 981 | } |
4daa28f6 | 982 | return NULL; |
1da177e4 LT |
983 | } |
984 | ||
bf17bb71 NP |
985 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
986 | { | |
987 | struct sem_undo *un; | |
988 | ||
989 | assert_spin_locked(&ulp->lock); | |
990 | ||
991 | un = __lookup_undo(ulp, semid); | |
992 | if (un) { | |
993 | list_del_rcu(&un->list_proc); | |
994 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
995 | } | |
996 | return un; | |
997 | } | |
998 | ||
4daa28f6 MS |
999 | /** |
1000 | * find_alloc_undo - Lookup (and if not present create) undo array | |
1001 | * @ns: namespace | |
1002 | * @semid: semaphore array id | |
1003 | * | |
1004 | * The function looks up (and if not present creates) the undo structure. | |
1005 | * The size of the undo structure depends on the size of the semaphore | |
1006 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1007 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1008 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1009 | */ |
1010 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1011 | { |
1012 | struct sem_array *sma; | |
1013 | struct sem_undo_list *ulp; | |
1014 | struct sem_undo *un, *new; | |
1015 | int nsems; | |
1016 | int error; | |
1017 | ||
1018 | error = get_undo_list(&ulp); | |
1019 | if (error) | |
1020 | return ERR_PTR(error); | |
1021 | ||
380af1b3 | 1022 | rcu_read_lock(); |
c530c6ac | 1023 | spin_lock(&ulp->lock); |
1da177e4 | 1024 | un = lookup_undo(ulp, semid); |
c530c6ac | 1025 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1026 | if (likely(un!=NULL)) |
1027 | goto out; | |
380af1b3 | 1028 | rcu_read_unlock(); |
1da177e4 LT |
1029 | |
1030 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1031 | /* step 1: figure out the size of the semaphore array */ |
023a5355 ND |
1032 | sma = sem_lock_check(ns, semid); |
1033 | if (IS_ERR(sma)) | |
1034 | return ERR_PTR(PTR_ERR(sma)); | |
1035 | ||
1da177e4 | 1036 | nsems = sma->sem_nsems; |
6ff37972 | 1037 | sem_getref_and_unlock(sma); |
1da177e4 | 1038 | |
4daa28f6 | 1039 | /* step 2: allocate new undo structure */ |
4668edc3 | 1040 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1041 | if (!new) { |
6ff37972 | 1042 | sem_putref(sma); |
1da177e4 LT |
1043 | return ERR_PTR(-ENOMEM); |
1044 | } | |
1da177e4 | 1045 | |
380af1b3 | 1046 | /* step 3: Acquire the lock on semaphore array */ |
6ff37972 | 1047 | sem_lock_and_putref(sma); |
1da177e4 LT |
1048 | if (sma->sem_perm.deleted) { |
1049 | sem_unlock(sma); | |
1da177e4 LT |
1050 | kfree(new); |
1051 | un = ERR_PTR(-EIDRM); | |
1052 | goto out; | |
1053 | } | |
380af1b3 MS |
1054 | spin_lock(&ulp->lock); |
1055 | ||
1056 | /* | |
1057 | * step 4: check for races: did someone else allocate the undo struct? | |
1058 | */ | |
1059 | un = lookup_undo(ulp, semid); | |
1060 | if (un) { | |
1061 | kfree(new); | |
1062 | goto success; | |
1063 | } | |
4daa28f6 MS |
1064 | /* step 5: initialize & link new undo structure */ |
1065 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1066 | new->ulp = ulp; |
4daa28f6 MS |
1067 | new->semid = semid; |
1068 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1069 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
4daa28f6 MS |
1070 | assert_spin_locked(&sma->sem_perm.lock); |
1071 | list_add(&new->list_id, &sma->list_id); | |
380af1b3 | 1072 | un = new; |
4daa28f6 | 1073 | |
380af1b3 | 1074 | success: |
c530c6ac | 1075 | spin_unlock(&ulp->lock); |
380af1b3 MS |
1076 | rcu_read_lock(); |
1077 | sem_unlock(sma); | |
1da177e4 LT |
1078 | out: |
1079 | return un; | |
1080 | } | |
1081 | ||
d5460c99 HC |
1082 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1083 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1084 | { |
1085 | int error = -EINVAL; | |
1086 | struct sem_array *sma; | |
1087 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1088 | struct sembuf* sops = fast_sops, *sop; | |
1089 | struct sem_undo *un; | |
b78755ab | 1090 | int undos = 0, alter = 0, max; |
1da177e4 LT |
1091 | struct sem_queue queue; |
1092 | unsigned long jiffies_left = 0; | |
e3893534 KK |
1093 | struct ipc_namespace *ns; |
1094 | ||
1095 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1096 | |
1097 | if (nsops < 1 || semid < 0) | |
1098 | return -EINVAL; | |
e3893534 | 1099 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1100 | return -E2BIG; |
1101 | if(nsops > SEMOPM_FAST) { | |
1102 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1103 | if(sops==NULL) | |
1104 | return -ENOMEM; | |
1105 | } | |
1106 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1107 | error=-EFAULT; | |
1108 | goto out_free; | |
1109 | } | |
1110 | if (timeout) { | |
1111 | struct timespec _timeout; | |
1112 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1113 | error = -EFAULT; | |
1114 | goto out_free; | |
1115 | } | |
1116 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1117 | _timeout.tv_nsec >= 1000000000L) { | |
1118 | error = -EINVAL; | |
1119 | goto out_free; | |
1120 | } | |
1121 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1122 | } | |
1123 | max = 0; | |
1124 | for (sop = sops; sop < sops + nsops; sop++) { | |
1125 | if (sop->sem_num >= max) | |
1126 | max = sop->sem_num; | |
1127 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1128 | undos = 1; |
1129 | if (sop->sem_op != 0) | |
1da177e4 LT |
1130 | alter = 1; |
1131 | } | |
1da177e4 | 1132 | |
1da177e4 | 1133 | if (undos) { |
4daa28f6 | 1134 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1135 | if (IS_ERR(un)) { |
1136 | error = PTR_ERR(un); | |
1137 | goto out_free; | |
1138 | } | |
1139 | } else | |
1140 | un = NULL; | |
1141 | ||
023a5355 ND |
1142 | sma = sem_lock_check(ns, semid); |
1143 | if (IS_ERR(sma)) { | |
380af1b3 MS |
1144 | if (un) |
1145 | rcu_read_unlock(); | |
023a5355 | 1146 | error = PTR_ERR(sma); |
1da177e4 | 1147 | goto out_free; |
023a5355 ND |
1148 | } |
1149 | ||
1da177e4 | 1150 | /* |
4daa28f6 | 1151 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1152 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1153 | * and now a new array with received the same id. Check and fail. |
380af1b3 MS |
1154 | * This case can be detected checking un->semid. The existance of |
1155 | * "un" itself is guaranteed by rcu. | |
1da177e4 | 1156 | */ |
4daa28f6 | 1157 | error = -EIDRM; |
380af1b3 MS |
1158 | if (un) { |
1159 | if (un->semid == -1) { | |
1160 | rcu_read_unlock(); | |
1161 | goto out_unlock_free; | |
1162 | } else { | |
1163 | /* | |
1164 | * rcu lock can be released, "un" cannot disappear: | |
1165 | * - sem_lock is acquired, thus IPC_RMID is | |
1166 | * impossible. | |
1167 | * - exit_sem is impossible, it always operates on | |
1168 | * current (or a dead task). | |
1169 | */ | |
1170 | ||
1171 | rcu_read_unlock(); | |
1172 | } | |
1173 | } | |
4daa28f6 | 1174 | |
1da177e4 LT |
1175 | error = -EFBIG; |
1176 | if (max >= sma->sem_nsems) | |
1177 | goto out_unlock_free; | |
1178 | ||
1179 | error = -EACCES; | |
1180 | if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) | |
1181 | goto out_unlock_free; | |
1182 | ||
1183 | error = security_sem_semop(sma, sops, nsops, alter); | |
1184 | if (error) | |
1185 | goto out_unlock_free; | |
1186 | ||
b488893a | 1187 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1188 | if (error <= 0) { |
1189 | if (alter && error == 0) | |
1190 | update_queue (sma); | |
1191 | goto out_unlock_free; | |
1192 | } | |
1193 | ||
1194 | /* We need to sleep on this operation, so we put the current | |
1195 | * task into the pending queue and go to sleep. | |
1196 | */ | |
1197 | ||
1da177e4 LT |
1198 | queue.sops = sops; |
1199 | queue.nsops = nsops; | |
1200 | queue.undo = un; | |
b488893a | 1201 | queue.pid = task_tgid_vnr(current); |
1da177e4 LT |
1202 | queue.alter = alter; |
1203 | if (alter) | |
a1193f8e | 1204 | list_add_tail(&queue.list, &sma->sem_pending); |
1da177e4 | 1205 | else |
a1193f8e | 1206 | list_add(&queue.list, &sma->sem_pending); |
1da177e4 | 1207 | |
b97e820f MS |
1208 | if (nsops == 1) { |
1209 | struct sem *curr; | |
1210 | curr = &sma->sem_base[sops->sem_num]; | |
1211 | ||
1212 | if (alter) | |
1213 | list_add_tail(&queue.simple_list, &curr->sem_pending); | |
1214 | else | |
1215 | list_add(&queue.simple_list, &curr->sem_pending); | |
1216 | } else { | |
1217 | INIT_LIST_HEAD(&queue.simple_list); | |
1218 | sma->complex_count++; | |
1219 | } | |
1220 | ||
1da177e4 LT |
1221 | queue.status = -EINTR; |
1222 | queue.sleeper = current; | |
1223 | current->state = TASK_INTERRUPTIBLE; | |
1224 | sem_unlock(sma); | |
1225 | ||
1226 | if (timeout) | |
1227 | jiffies_left = schedule_timeout(jiffies_left); | |
1228 | else | |
1229 | schedule(); | |
1230 | ||
1231 | error = queue.status; | |
1232 | while(unlikely(error == IN_WAKEUP)) { | |
1233 | cpu_relax(); | |
1234 | error = queue.status; | |
1235 | } | |
1236 | ||
1237 | if (error != -EINTR) { | |
1238 | /* fast path: update_queue already obtained all requested | |
1239 | * resources */ | |
1240 | goto out_free; | |
1241 | } | |
1242 | ||
e3893534 | 1243 | sma = sem_lock(ns, semid); |
023a5355 | 1244 | if (IS_ERR(sma)) { |
1da177e4 LT |
1245 | error = -EIDRM; |
1246 | goto out_free; | |
1247 | } | |
1248 | ||
1249 | /* | |
1250 | * If queue.status != -EINTR we are woken up by another process | |
1251 | */ | |
1252 | error = queue.status; | |
1253 | if (error != -EINTR) { | |
1254 | goto out_unlock_free; | |
1255 | } | |
1256 | ||
1257 | /* | |
1258 | * If an interrupt occurred we have to clean up the queue | |
1259 | */ | |
1260 | if (timeout && jiffies_left == 0) | |
1261 | error = -EAGAIN; | |
b97e820f | 1262 | unlink_queue(sma, &queue); |
1da177e4 LT |
1263 | |
1264 | out_unlock_free: | |
1265 | sem_unlock(sma); | |
1266 | out_free: | |
1267 | if(sops != fast_sops) | |
1268 | kfree(sops); | |
1269 | return error; | |
1270 | } | |
1271 | ||
d5460c99 HC |
1272 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
1273 | unsigned, nsops) | |
1da177e4 LT |
1274 | { |
1275 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1276 | } | |
1277 | ||
1278 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1279 | * parent and child tasks. | |
1da177e4 LT |
1280 | */ |
1281 | ||
1282 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1283 | { | |
1284 | struct sem_undo_list *undo_list; | |
1285 | int error; | |
1286 | ||
1287 | if (clone_flags & CLONE_SYSVSEM) { | |
1288 | error = get_undo_list(&undo_list); | |
1289 | if (error) | |
1290 | return error; | |
1da177e4 LT |
1291 | atomic_inc(&undo_list->refcnt); |
1292 | tsk->sysvsem.undo_list = undo_list; | |
1293 | } else | |
1294 | tsk->sysvsem.undo_list = NULL; | |
1295 | ||
1296 | return 0; | |
1297 | } | |
1298 | ||
1299 | /* | |
1300 | * add semadj values to semaphores, free undo structures. | |
1301 | * undo structures are not freed when semaphore arrays are destroyed | |
1302 | * so some of them may be out of date. | |
1303 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1304 | * set of adjustments that needs to be done should be done in an atomic | |
1305 | * manner or not. That is, if we are attempting to decrement the semval | |
1306 | * should we queue up and wait until we can do so legally? | |
1307 | * The original implementation attempted to do this (queue and wait). | |
1308 | * The current implementation does not do so. The POSIX standard | |
1309 | * and SVID should be consulted to determine what behavior is mandated. | |
1310 | */ | |
1311 | void exit_sem(struct task_struct *tsk) | |
1312 | { | |
4daa28f6 | 1313 | struct sem_undo_list *ulp; |
1da177e4 | 1314 | |
4daa28f6 MS |
1315 | ulp = tsk->sysvsem.undo_list; |
1316 | if (!ulp) | |
1da177e4 | 1317 | return; |
9edff4ab | 1318 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 1319 | |
4daa28f6 | 1320 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
1321 | return; |
1322 | ||
380af1b3 | 1323 | for (;;) { |
1da177e4 | 1324 | struct sem_array *sma; |
380af1b3 MS |
1325 | struct sem_undo *un; |
1326 | int semid; | |
4daa28f6 MS |
1327 | int i; |
1328 | ||
380af1b3 | 1329 | rcu_read_lock(); |
05725f7e JP |
1330 | un = list_entry_rcu(ulp->list_proc.next, |
1331 | struct sem_undo, list_proc); | |
380af1b3 MS |
1332 | if (&un->list_proc == &ulp->list_proc) |
1333 | semid = -1; | |
1334 | else | |
1335 | semid = un->semid; | |
1336 | rcu_read_unlock(); | |
4daa28f6 | 1337 | |
380af1b3 MS |
1338 | if (semid == -1) |
1339 | break; | |
1da177e4 | 1340 | |
380af1b3 | 1341 | sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid); |
1da177e4 | 1342 | |
380af1b3 MS |
1343 | /* exit_sem raced with IPC_RMID, nothing to do */ |
1344 | if (IS_ERR(sma)) | |
1345 | continue; | |
1da177e4 | 1346 | |
bf17bb71 | 1347 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
1348 | if (un == NULL) { |
1349 | /* exit_sem raced with IPC_RMID+semget() that created | |
1350 | * exactly the same semid. Nothing to do. | |
1351 | */ | |
1352 | sem_unlock(sma); | |
1353 | continue; | |
1354 | } | |
1355 | ||
1356 | /* remove un from the linked lists */ | |
4daa28f6 MS |
1357 | assert_spin_locked(&sma->sem_perm.lock); |
1358 | list_del(&un->list_id); | |
1359 | ||
380af1b3 MS |
1360 | spin_lock(&ulp->lock); |
1361 | list_del_rcu(&un->list_proc); | |
1362 | spin_unlock(&ulp->lock); | |
1363 | ||
4daa28f6 MS |
1364 | /* perform adjustments registered in un */ |
1365 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 1366 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
1367 | if (un->semadj[i]) { |
1368 | semaphore->semval += un->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); | |
1da177e4 | 1392 | sem_unlock(sma); |
380af1b3 MS |
1393 | |
1394 | call_rcu(&un->rcu, free_un); | |
1da177e4 | 1395 | } |
4daa28f6 | 1396 | kfree(ulp); |
1da177e4 LT |
1397 | } |
1398 | ||
1399 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1400 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1401 | { |
19b4946c MW |
1402 | struct sem_array *sma = it; |
1403 | ||
1404 | return seq_printf(s, | |
b97e820f | 1405 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 1406 | sma->sem_perm.key, |
7ca7e564 | 1407 | sma->sem_perm.id, |
19b4946c MW |
1408 | sma->sem_perm.mode, |
1409 | sma->sem_nsems, | |
1410 | sma->sem_perm.uid, | |
1411 | sma->sem_perm.gid, | |
1412 | sma->sem_perm.cuid, | |
1413 | sma->sem_perm.cgid, | |
1414 | sma->sem_otime, | |
1415 | sma->sem_ctime); | |
1da177e4 LT |
1416 | } |
1417 | #endif |