]>
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 | ||
fd5db422 MS |
437 | /** check_restart(sma, q) |
438 | * @sma: semaphore array | |
439 | * @q: the operation that just completed | |
440 | * | |
441 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
442 | * waiting operations. Therefore this function checks if the restart is | |
443 | * really necessary. It is called after a previously waiting operation | |
444 | * was completed. | |
445 | */ | |
446 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
447 | { | |
448 | struct sem *curr; | |
449 | struct sem_queue *h; | |
450 | ||
451 | /* if the operation didn't modify the array, then no restart */ | |
452 | if (q->alter == 0) | |
453 | return 0; | |
454 | ||
455 | /* pending complex operations are too difficult to analyse */ | |
456 | if (sma->complex_count) | |
457 | return 1; | |
458 | ||
459 | /* we were a sleeping complex operation. Too difficult */ | |
460 | if (q->nsops > 1) | |
461 | return 1; | |
462 | ||
463 | curr = sma->sem_base + q->sops[0].sem_num; | |
464 | ||
465 | /* No-one waits on this queue */ | |
466 | if (list_empty(&curr->sem_pending)) | |
467 | return 0; | |
468 | ||
469 | /* the new semaphore value */ | |
470 | if (curr->semval) { | |
471 | /* It is impossible that someone waits for the new value: | |
472 | * - q is a previously sleeping simple operation that | |
473 | * altered the array. It must be a decrement, because | |
474 | * simple increments never sleep. | |
475 | * - The value is not 0, thus wait-for-zero won't proceed. | |
476 | * - If there are older (higher priority) decrements | |
477 | * in the queue, then they have observed the original | |
478 | * semval value and couldn't proceed. The operation | |
479 | * decremented to value - thus they won't proceed either. | |
480 | */ | |
481 | BUG_ON(q->sops[0].sem_op >= 0); | |
482 | return 0; | |
483 | } | |
484 | /* | |
485 | * semval is 0. Check if there are wait-for-zero semops. | |
486 | * They must be the first entries in the per-semaphore simple queue | |
487 | */ | |
488 | h = list_first_entry(&curr->sem_pending, struct sem_queue, simple_list); | |
489 | BUG_ON(h->nsops != 1); | |
490 | BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num); | |
491 | ||
492 | /* Yes, there is a wait-for-zero semop. Restart */ | |
493 | if (h->sops[0].sem_op == 0) | |
494 | return 1; | |
495 | ||
496 | /* Again - no-one is waiting for the new value. */ | |
497 | return 0; | |
498 | } | |
499 | ||
636c6be8 MS |
500 | |
501 | /** | |
502 | * update_queue(sma, semnum): Look for tasks that can be completed. | |
503 | * @sma: semaphore array. | |
504 | * @semnum: semaphore that was modified. | |
505 | * | |
506 | * update_queue must be called after a semaphore in a semaphore array | |
507 | * was modified. If multiple semaphore were modified, then @semnum | |
508 | * must be set to -1. | |
1da177e4 | 509 | */ |
636c6be8 | 510 | static void update_queue(struct sem_array *sma, int semnum) |
1da177e4 | 511 | { |
636c6be8 MS |
512 | struct sem_queue *q; |
513 | struct list_head *walk; | |
514 | struct list_head *pending_list; | |
515 | int offset; | |
516 | ||
517 | /* if there are complex operations around, then knowing the semaphore | |
518 | * that was modified doesn't help us. Assume that multiple semaphores | |
519 | * were modified. | |
520 | */ | |
521 | if (sma->complex_count) | |
522 | semnum = -1; | |
523 | ||
524 | if (semnum == -1) { | |
525 | pending_list = &sma->sem_pending; | |
526 | offset = offsetof(struct sem_queue, list); | |
527 | } else { | |
528 | pending_list = &sma->sem_base[semnum].sem_pending; | |
529 | offset = offsetof(struct sem_queue, simple_list); | |
530 | } | |
9cad200c NP |
531 | |
532 | again: | |
636c6be8 MS |
533 | walk = pending_list->next; |
534 | while (walk != pending_list) { | |
fd5db422 | 535 | int error, restart; |
636c6be8 MS |
536 | |
537 | q = (struct sem_queue *)((char *)walk - offset); | |
538 | walk = walk->next; | |
1da177e4 | 539 | |
d987f8b2 MS |
540 | /* If we are scanning the single sop, per-semaphore list of |
541 | * one semaphore and that semaphore is 0, then it is not | |
542 | * necessary to scan the "alter" entries: simple increments | |
543 | * that affect only one entry succeed immediately and cannot | |
544 | * be in the per semaphore pending queue, and decrements | |
545 | * cannot be successful if the value is already 0. | |
546 | */ | |
547 | if (semnum != -1 && sma->sem_base[semnum].semval == 0 && | |
548 | q->alter) | |
549 | break; | |
550 | ||
1da177e4 LT |
551 | error = try_atomic_semop(sma, q->sops, q->nsops, |
552 | q->undo, q->pid); | |
553 | ||
554 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
555 | if (error > 0) |
556 | continue; | |
557 | ||
b97e820f | 558 | unlink_queue(sma, q); |
9cad200c | 559 | |
fd5db422 MS |
560 | if (error) |
561 | restart = 0; | |
562 | else | |
563 | restart = check_restart(sma, q); | |
564 | ||
d4212093 | 565 | wake_up_sem_queue(q, error); |
fd5db422 | 566 | if (restart) |
9cad200c | 567 | goto again; |
1da177e4 LT |
568 | } |
569 | } | |
570 | ||
fd5db422 MS |
571 | /** do_smart_update(sma, sops, nsops): Optimized update_queue |
572 | * @sma: semaphore array | |
573 | * @sops: operations that were performed | |
574 | * @nsops: number of operations | |
575 | * | |
576 | * do_smart_update() does the required called to update_queue, based on the | |
577 | * actual changes that were performed on the semaphore array. | |
578 | */ | |
579 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops) | |
580 | { | |
581 | int i; | |
582 | ||
583 | if (sma->complex_count || sops == NULL) { | |
584 | update_queue(sma, -1); | |
585 | return; | |
586 | } | |
587 | ||
588 | for (i = 0; i < nsops; i++) { | |
589 | if (sops[i].sem_op > 0 || | |
590 | (sops[i].sem_op < 0 && | |
591 | sma->sem_base[sops[i].sem_num].semval == 0)) | |
592 | update_queue(sma, sops[i].sem_num); | |
593 | } | |
594 | } | |
595 | ||
596 | ||
1da177e4 LT |
597 | /* The following counts are associated to each semaphore: |
598 | * semncnt number of tasks waiting on semval being nonzero | |
599 | * semzcnt number of tasks waiting on semval being zero | |
600 | * This model assumes that a task waits on exactly one semaphore. | |
601 | * Since semaphore operations are to be performed atomically, tasks actually | |
602 | * wait on a whole sequence of semaphores simultaneously. | |
603 | * The counts we return here are a rough approximation, but still | |
604 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
605 | */ | |
606 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
607 | { | |
608 | int semncnt; | |
609 | struct sem_queue * q; | |
610 | ||
611 | semncnt = 0; | |
a1193f8e | 612 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
613 | struct sembuf * sops = q->sops; |
614 | int nsops = q->nsops; | |
615 | int i; | |
616 | for (i = 0; i < nsops; i++) | |
617 | if (sops[i].sem_num == semnum | |
618 | && (sops[i].sem_op < 0) | |
619 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
620 | semncnt++; | |
621 | } | |
622 | return semncnt; | |
623 | } | |
a1193f8e | 624 | |
1da177e4 LT |
625 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
626 | { | |
627 | int semzcnt; | |
628 | struct sem_queue * q; | |
629 | ||
630 | semzcnt = 0; | |
a1193f8e | 631 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
632 | struct sembuf * sops = q->sops; |
633 | int nsops = q->nsops; | |
634 | int i; | |
635 | for (i = 0; i < nsops; i++) | |
636 | if (sops[i].sem_num == semnum | |
637 | && (sops[i].sem_op == 0) | |
638 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
639 | semzcnt++; | |
640 | } | |
641 | return semzcnt; | |
642 | } | |
643 | ||
6d97e234 | 644 | static void free_un(struct rcu_head *head) |
380af1b3 MS |
645 | { |
646 | struct sem_undo *un = container_of(head, struct sem_undo, rcu); | |
647 | kfree(un); | |
648 | } | |
649 | ||
3e148c79 ND |
650 | /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked |
651 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex | |
652 | * remains locked on exit. | |
1da177e4 | 653 | */ |
01b8b07a | 654 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 655 | { |
380af1b3 MS |
656 | struct sem_undo *un, *tu; |
657 | struct sem_queue *q, *tq; | |
01b8b07a | 658 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 | 659 | |
380af1b3 | 660 | /* Free the existing undo structures for this semaphore set. */ |
4daa28f6 | 661 | assert_spin_locked(&sma->sem_perm.lock); |
380af1b3 MS |
662 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
663 | list_del(&un->list_id); | |
664 | spin_lock(&un->ulp->lock); | |
1da177e4 | 665 | un->semid = -1; |
380af1b3 MS |
666 | list_del_rcu(&un->list_proc); |
667 | spin_unlock(&un->ulp->lock); | |
668 | call_rcu(&un->rcu, free_un); | |
669 | } | |
1da177e4 LT |
670 | |
671 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
380af1b3 | 672 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { |
b97e820f | 673 | unlink_queue(sma, q); |
d4212093 | 674 | wake_up_sem_queue(q, -EIDRM); |
1da177e4 LT |
675 | } |
676 | ||
7ca7e564 ND |
677 | /* Remove the semaphore set from the IDR */ |
678 | sem_rmid(ns, sma); | |
1da177e4 LT |
679 | sem_unlock(sma); |
680 | ||
e3893534 | 681 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
682 | security_sem_free(sma); |
683 | ipc_rcu_putref(sma); | |
684 | } | |
685 | ||
686 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
687 | { | |
688 | switch(version) { | |
689 | case IPC_64: | |
690 | return copy_to_user(buf, in, sizeof(*in)); | |
691 | case IPC_OLD: | |
692 | { | |
693 | struct semid_ds out; | |
694 | ||
695 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); | |
696 | ||
697 | out.sem_otime = in->sem_otime; | |
698 | out.sem_ctime = in->sem_ctime; | |
699 | out.sem_nsems = in->sem_nsems; | |
700 | ||
701 | return copy_to_user(buf, &out, sizeof(out)); | |
702 | } | |
703 | default: | |
704 | return -EINVAL; | |
705 | } | |
706 | } | |
707 | ||
4b9fcb0e PP |
708 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
709 | int cmd, int version, union semun arg) | |
1da177e4 | 710 | { |
e5cc9c7b | 711 | int err; |
1da177e4 LT |
712 | struct sem_array *sma; |
713 | ||
714 | switch(cmd) { | |
715 | case IPC_INFO: | |
716 | case SEM_INFO: | |
717 | { | |
718 | struct seminfo seminfo; | |
719 | int max_id; | |
720 | ||
721 | err = security_sem_semctl(NULL, cmd); | |
722 | if (err) | |
723 | return err; | |
724 | ||
725 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
726 | seminfo.semmni = ns->sc_semmni; |
727 | seminfo.semmns = ns->sc_semmns; | |
728 | seminfo.semmsl = ns->sc_semmsl; | |
729 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
730 | seminfo.semvmx = SEMVMX; |
731 | seminfo.semmnu = SEMMNU; | |
732 | seminfo.semmap = SEMMAP; | |
733 | seminfo.semume = SEMUME; | |
3e148c79 | 734 | down_read(&sem_ids(ns).rw_mutex); |
1da177e4 | 735 | if (cmd == SEM_INFO) { |
e3893534 KK |
736 | seminfo.semusz = sem_ids(ns).in_use; |
737 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
738 | } else { |
739 | seminfo.semusz = SEMUSZ; | |
740 | seminfo.semaem = SEMAEM; | |
741 | } | |
7ca7e564 | 742 | max_id = ipc_get_maxid(&sem_ids(ns)); |
3e148c79 | 743 | up_read(&sem_ids(ns).rw_mutex); |
1da177e4 LT |
744 | if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) |
745 | return -EFAULT; | |
746 | return (max_id < 0) ? 0: max_id; | |
747 | } | |
4b9fcb0e | 748 | case IPC_STAT: |
1da177e4 LT |
749 | case SEM_STAT: |
750 | { | |
751 | struct semid64_ds tbuf; | |
752 | int id; | |
753 | ||
4b9fcb0e PP |
754 | if (cmd == SEM_STAT) { |
755 | sma = sem_lock(ns, semid); | |
756 | if (IS_ERR(sma)) | |
757 | return PTR_ERR(sma); | |
758 | id = sma->sem_perm.id; | |
759 | } else { | |
760 | sma = sem_lock_check(ns, semid); | |
761 | if (IS_ERR(sma)) | |
762 | return PTR_ERR(sma); | |
763 | id = 0; | |
764 | } | |
1da177e4 LT |
765 | |
766 | err = -EACCES; | |
767 | if (ipcperms (&sma->sem_perm, S_IRUGO)) | |
768 | goto out_unlock; | |
769 | ||
770 | err = security_sem_semctl(sma, cmd); | |
771 | if (err) | |
772 | goto out_unlock; | |
773 | ||
023a5355 ND |
774 | memset(&tbuf, 0, sizeof(tbuf)); |
775 | ||
1da177e4 LT |
776 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
777 | tbuf.sem_otime = sma->sem_otime; | |
778 | tbuf.sem_ctime = sma->sem_ctime; | |
779 | tbuf.sem_nsems = sma->sem_nsems; | |
780 | sem_unlock(sma); | |
781 | if (copy_semid_to_user (arg.buf, &tbuf, version)) | |
782 | return -EFAULT; | |
783 | return id; | |
784 | } | |
785 | default: | |
786 | return -EINVAL; | |
787 | } | |
1da177e4 LT |
788 | out_unlock: |
789 | sem_unlock(sma); | |
790 | return err; | |
791 | } | |
792 | ||
e3893534 KK |
793 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
794 | int cmd, int version, union semun arg) | |
1da177e4 LT |
795 | { |
796 | struct sem_array *sma; | |
797 | struct sem* curr; | |
798 | int err; | |
799 | ushort fast_sem_io[SEMMSL_FAST]; | |
800 | ushort* sem_io = fast_sem_io; | |
801 | int nsems; | |
802 | ||
023a5355 ND |
803 | sma = sem_lock_check(ns, semid); |
804 | if (IS_ERR(sma)) | |
805 | return PTR_ERR(sma); | |
1da177e4 LT |
806 | |
807 | nsems = sma->sem_nsems; | |
808 | ||
1da177e4 LT |
809 | err = -EACCES; |
810 | if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) | |
811 | goto out_unlock; | |
812 | ||
813 | err = security_sem_semctl(sma, cmd); | |
814 | if (err) | |
815 | goto out_unlock; | |
816 | ||
817 | err = -EACCES; | |
818 | switch (cmd) { | |
819 | case GETALL: | |
820 | { | |
821 | ushort __user *array = arg.array; | |
822 | int i; | |
823 | ||
824 | if(nsems > SEMMSL_FAST) { | |
6ff37972 | 825 | sem_getref_and_unlock(sma); |
1da177e4 LT |
826 | |
827 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
828 | if(sem_io == NULL) { | |
6ff37972 | 829 | sem_putref(sma); |
1da177e4 LT |
830 | return -ENOMEM; |
831 | } | |
832 | ||
6ff37972 | 833 | sem_lock_and_putref(sma); |
1da177e4 LT |
834 | if (sma->sem_perm.deleted) { |
835 | sem_unlock(sma); | |
836 | err = -EIDRM; | |
837 | goto out_free; | |
838 | } | |
839 | } | |
840 | ||
841 | for (i = 0; i < sma->sem_nsems; i++) | |
842 | sem_io[i] = sma->sem_base[i].semval; | |
843 | sem_unlock(sma); | |
844 | err = 0; | |
845 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
846 | err = -EFAULT; | |
847 | goto out_free; | |
848 | } | |
849 | case SETALL: | |
850 | { | |
851 | int i; | |
852 | struct sem_undo *un; | |
853 | ||
6ff37972 | 854 | sem_getref_and_unlock(sma); |
1da177e4 LT |
855 | |
856 | if(nsems > SEMMSL_FAST) { | |
857 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
858 | if(sem_io == NULL) { | |
6ff37972 | 859 | sem_putref(sma); |
1da177e4 LT |
860 | return -ENOMEM; |
861 | } | |
862 | } | |
863 | ||
864 | if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { | |
6ff37972 | 865 | sem_putref(sma); |
1da177e4 LT |
866 | err = -EFAULT; |
867 | goto out_free; | |
868 | } | |
869 | ||
870 | for (i = 0; i < nsems; i++) { | |
871 | if (sem_io[i] > SEMVMX) { | |
6ff37972 | 872 | sem_putref(sma); |
1da177e4 LT |
873 | err = -ERANGE; |
874 | goto out_free; | |
875 | } | |
876 | } | |
6ff37972 | 877 | sem_lock_and_putref(sma); |
1da177e4 LT |
878 | if (sma->sem_perm.deleted) { |
879 | sem_unlock(sma); | |
880 | err = -EIDRM; | |
881 | goto out_free; | |
882 | } | |
883 | ||
884 | for (i = 0; i < nsems; i++) | |
885 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 MS |
886 | |
887 | assert_spin_locked(&sma->sem_perm.lock); | |
888 | list_for_each_entry(un, &sma->list_id, list_id) { | |
1da177e4 LT |
889 | for (i = 0; i < nsems; i++) |
890 | un->semadj[i] = 0; | |
4daa28f6 | 891 | } |
1da177e4 LT |
892 | sma->sem_ctime = get_seconds(); |
893 | /* maybe some queued-up processes were waiting for this */ | |
636c6be8 | 894 | update_queue(sma, -1); |
1da177e4 LT |
895 | err = 0; |
896 | goto out_unlock; | |
897 | } | |
1da177e4 LT |
898 | /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ |
899 | } | |
900 | err = -EINVAL; | |
901 | if(semnum < 0 || semnum >= nsems) | |
902 | goto out_unlock; | |
903 | ||
904 | curr = &sma->sem_base[semnum]; | |
905 | ||
906 | switch (cmd) { | |
907 | case GETVAL: | |
908 | err = curr->semval; | |
909 | goto out_unlock; | |
910 | case GETPID: | |
911 | err = curr->sempid; | |
912 | goto out_unlock; | |
913 | case GETNCNT: | |
914 | err = count_semncnt(sma,semnum); | |
915 | goto out_unlock; | |
916 | case GETZCNT: | |
917 | err = count_semzcnt(sma,semnum); | |
918 | goto out_unlock; | |
919 | case SETVAL: | |
920 | { | |
921 | int val = arg.val; | |
922 | struct sem_undo *un; | |
4daa28f6 | 923 | |
1da177e4 LT |
924 | err = -ERANGE; |
925 | if (val > SEMVMX || val < 0) | |
926 | goto out_unlock; | |
927 | ||
4daa28f6 MS |
928 | assert_spin_locked(&sma->sem_perm.lock); |
929 | list_for_each_entry(un, &sma->list_id, list_id) | |
1da177e4 | 930 | un->semadj[semnum] = 0; |
4daa28f6 | 931 | |
1da177e4 | 932 | curr->semval = val; |
b488893a | 933 | curr->sempid = task_tgid_vnr(current); |
1da177e4 LT |
934 | sma->sem_ctime = get_seconds(); |
935 | /* maybe some queued-up processes were waiting for this */ | |
636c6be8 | 936 | update_queue(sma, semnum); |
1da177e4 LT |
937 | err = 0; |
938 | goto out_unlock; | |
939 | } | |
940 | } | |
941 | out_unlock: | |
942 | sem_unlock(sma); | |
943 | out_free: | |
944 | if(sem_io != fast_sem_io) | |
945 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
946 | return err; | |
947 | } | |
948 | ||
016d7132 PP |
949 | static inline unsigned long |
950 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
951 | { |
952 | switch(version) { | |
953 | case IPC_64: | |
016d7132 | 954 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 955 | return -EFAULT; |
1da177e4 | 956 | return 0; |
1da177e4 LT |
957 | case IPC_OLD: |
958 | { | |
959 | struct semid_ds tbuf_old; | |
960 | ||
961 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
962 | return -EFAULT; | |
963 | ||
016d7132 PP |
964 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
965 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
966 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
967 | |
968 | return 0; | |
969 | } | |
970 | default: | |
971 | return -EINVAL; | |
972 | } | |
973 | } | |
974 | ||
522bb2a2 PP |
975 | /* |
976 | * This function handles some semctl commands which require the rw_mutex | |
977 | * to be held in write mode. | |
978 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
979 | */ | |
21a4826a PP |
980 | static int semctl_down(struct ipc_namespace *ns, int semid, |
981 | int cmd, int version, union semun arg) | |
1da177e4 LT |
982 | { |
983 | struct sem_array *sma; | |
984 | int err; | |
016d7132 | 985 | struct semid64_ds semid64; |
1da177e4 LT |
986 | struct kern_ipc_perm *ipcp; |
987 | ||
988 | if(cmd == IPC_SET) { | |
016d7132 | 989 | if (copy_semid_from_user(&semid64, arg.buf, version)) |
1da177e4 | 990 | return -EFAULT; |
1da177e4 | 991 | } |
073115d6 | 992 | |
a5f75e7f PP |
993 | ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0); |
994 | if (IS_ERR(ipcp)) | |
995 | return PTR_ERR(ipcp); | |
073115d6 | 996 | |
a5f75e7f | 997 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
998 | |
999 | err = security_sem_semctl(sma, cmd); | |
1000 | if (err) | |
1001 | goto out_unlock; | |
1002 | ||
1003 | switch(cmd){ | |
1004 | case IPC_RMID: | |
01b8b07a | 1005 | freeary(ns, ipcp); |
522bb2a2 | 1006 | goto out_up; |
1da177e4 | 1007 | case IPC_SET: |
8f4a3809 | 1008 | ipc_update_perm(&semid64.sem_perm, ipcp); |
1da177e4 | 1009 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1010 | break; |
1011 | default: | |
1da177e4 | 1012 | err = -EINVAL; |
1da177e4 | 1013 | } |
1da177e4 LT |
1014 | |
1015 | out_unlock: | |
1016 | sem_unlock(sma); | |
522bb2a2 PP |
1017 | out_up: |
1018 | up_write(&sem_ids(ns).rw_mutex); | |
1da177e4 LT |
1019 | return err; |
1020 | } | |
1021 | ||
6673e0c3 | 1022 | SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg) |
1da177e4 LT |
1023 | { |
1024 | int err = -EINVAL; | |
1025 | int version; | |
e3893534 | 1026 | struct ipc_namespace *ns; |
1da177e4 LT |
1027 | |
1028 | if (semid < 0) | |
1029 | return -EINVAL; | |
1030 | ||
1031 | version = ipc_parse_version(&cmd); | |
e3893534 | 1032 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
1033 | |
1034 | switch(cmd) { | |
1035 | case IPC_INFO: | |
1036 | case SEM_INFO: | |
4b9fcb0e | 1037 | case IPC_STAT: |
1da177e4 | 1038 | case SEM_STAT: |
4b9fcb0e | 1039 | err = semctl_nolock(ns, semid, cmd, version, arg); |
1da177e4 LT |
1040 | return err; |
1041 | case GETALL: | |
1042 | case GETVAL: | |
1043 | case GETPID: | |
1044 | case GETNCNT: | |
1045 | case GETZCNT: | |
1da177e4 LT |
1046 | case SETVAL: |
1047 | case SETALL: | |
e3893534 | 1048 | err = semctl_main(ns,semid,semnum,cmd,version,arg); |
1da177e4 LT |
1049 | return err; |
1050 | case IPC_RMID: | |
1051 | case IPC_SET: | |
21a4826a | 1052 | err = semctl_down(ns, semid, cmd, version, arg); |
1da177e4 LT |
1053 | return err; |
1054 | default: | |
1055 | return -EINVAL; | |
1056 | } | |
1057 | } | |
6673e0c3 HC |
1058 | #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS |
1059 | asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg) | |
1060 | { | |
1061 | return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg); | |
1062 | } | |
1063 | SYSCALL_ALIAS(sys_semctl, SyS_semctl); | |
1064 | #endif | |
1da177e4 | 1065 | |
1da177e4 LT |
1066 | /* If the task doesn't already have a undo_list, then allocate one |
1067 | * here. We guarantee there is only one thread using this undo list, | |
1068 | * and current is THE ONE | |
1069 | * | |
1070 | * If this allocation and assignment succeeds, but later | |
1071 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1072 | * Just let it stay associated with the task, and it'll be freed later | |
1073 | * at exit time. | |
1074 | * | |
1075 | * This can block, so callers must hold no locks. | |
1076 | */ | |
1077 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1078 | { | |
1079 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1080 | |
1081 | undo_list = current->sysvsem.undo_list; | |
1082 | if (!undo_list) { | |
2453a306 | 1083 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1084 | if (undo_list == NULL) |
1085 | return -ENOMEM; | |
00a5dfdb | 1086 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1087 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1088 | INIT_LIST_HEAD(&undo_list->list_proc); |
1089 | ||
1da177e4 LT |
1090 | current->sysvsem.undo_list = undo_list; |
1091 | } | |
1092 | *undo_listp = undo_list; | |
1093 | return 0; | |
1094 | } | |
1095 | ||
bf17bb71 | 1096 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1097 | { |
bf17bb71 | 1098 | struct sem_undo *un; |
4daa28f6 | 1099 | |
bf17bb71 NP |
1100 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1101 | if (un->semid == semid) | |
1102 | return un; | |
1da177e4 | 1103 | } |
4daa28f6 | 1104 | return NULL; |
1da177e4 LT |
1105 | } |
1106 | ||
bf17bb71 NP |
1107 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1108 | { | |
1109 | struct sem_undo *un; | |
1110 | ||
1111 | assert_spin_locked(&ulp->lock); | |
1112 | ||
1113 | un = __lookup_undo(ulp, semid); | |
1114 | if (un) { | |
1115 | list_del_rcu(&un->list_proc); | |
1116 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1117 | } | |
1118 | return un; | |
1119 | } | |
1120 | ||
4daa28f6 MS |
1121 | /** |
1122 | * find_alloc_undo - Lookup (and if not present create) undo array | |
1123 | * @ns: namespace | |
1124 | * @semid: semaphore array id | |
1125 | * | |
1126 | * The function looks up (and if not present creates) the undo structure. | |
1127 | * The size of the undo structure depends on the size of the semaphore | |
1128 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1129 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1130 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1131 | */ |
1132 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1133 | { |
1134 | struct sem_array *sma; | |
1135 | struct sem_undo_list *ulp; | |
1136 | struct sem_undo *un, *new; | |
1137 | int nsems; | |
1138 | int error; | |
1139 | ||
1140 | error = get_undo_list(&ulp); | |
1141 | if (error) | |
1142 | return ERR_PTR(error); | |
1143 | ||
380af1b3 | 1144 | rcu_read_lock(); |
c530c6ac | 1145 | spin_lock(&ulp->lock); |
1da177e4 | 1146 | un = lookup_undo(ulp, semid); |
c530c6ac | 1147 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1148 | if (likely(un!=NULL)) |
1149 | goto out; | |
380af1b3 | 1150 | rcu_read_unlock(); |
1da177e4 LT |
1151 | |
1152 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1153 | /* step 1: figure out the size of the semaphore array */ |
023a5355 ND |
1154 | sma = sem_lock_check(ns, semid); |
1155 | if (IS_ERR(sma)) | |
1156 | return ERR_PTR(PTR_ERR(sma)); | |
1157 | ||
1da177e4 | 1158 | nsems = sma->sem_nsems; |
6ff37972 | 1159 | sem_getref_and_unlock(sma); |
1da177e4 | 1160 | |
4daa28f6 | 1161 | /* step 2: allocate new undo structure */ |
4668edc3 | 1162 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1163 | if (!new) { |
6ff37972 | 1164 | sem_putref(sma); |
1da177e4 LT |
1165 | return ERR_PTR(-ENOMEM); |
1166 | } | |
1da177e4 | 1167 | |
380af1b3 | 1168 | /* step 3: Acquire the lock on semaphore array */ |
6ff37972 | 1169 | sem_lock_and_putref(sma); |
1da177e4 LT |
1170 | if (sma->sem_perm.deleted) { |
1171 | sem_unlock(sma); | |
1da177e4 LT |
1172 | kfree(new); |
1173 | un = ERR_PTR(-EIDRM); | |
1174 | goto out; | |
1175 | } | |
380af1b3 MS |
1176 | spin_lock(&ulp->lock); |
1177 | ||
1178 | /* | |
1179 | * step 4: check for races: did someone else allocate the undo struct? | |
1180 | */ | |
1181 | un = lookup_undo(ulp, semid); | |
1182 | if (un) { | |
1183 | kfree(new); | |
1184 | goto success; | |
1185 | } | |
4daa28f6 MS |
1186 | /* step 5: initialize & link new undo structure */ |
1187 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1188 | new->ulp = ulp; |
4daa28f6 MS |
1189 | new->semid = semid; |
1190 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1191 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
4daa28f6 MS |
1192 | assert_spin_locked(&sma->sem_perm.lock); |
1193 | list_add(&new->list_id, &sma->list_id); | |
380af1b3 | 1194 | un = new; |
4daa28f6 | 1195 | |
380af1b3 | 1196 | success: |
c530c6ac | 1197 | spin_unlock(&ulp->lock); |
380af1b3 MS |
1198 | rcu_read_lock(); |
1199 | sem_unlock(sma); | |
1da177e4 LT |
1200 | out: |
1201 | return un; | |
1202 | } | |
1203 | ||
d5460c99 HC |
1204 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1205 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1206 | { |
1207 | int error = -EINVAL; | |
1208 | struct sem_array *sma; | |
1209 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1210 | struct sembuf* sops = fast_sops, *sop; | |
1211 | struct sem_undo *un; | |
b78755ab | 1212 | int undos = 0, alter = 0, max; |
1da177e4 LT |
1213 | struct sem_queue queue; |
1214 | unsigned long jiffies_left = 0; | |
e3893534 KK |
1215 | struct ipc_namespace *ns; |
1216 | ||
1217 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1218 | |
1219 | if (nsops < 1 || semid < 0) | |
1220 | return -EINVAL; | |
e3893534 | 1221 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1222 | return -E2BIG; |
1223 | if(nsops > SEMOPM_FAST) { | |
1224 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1225 | if(sops==NULL) | |
1226 | return -ENOMEM; | |
1227 | } | |
1228 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1229 | error=-EFAULT; | |
1230 | goto out_free; | |
1231 | } | |
1232 | if (timeout) { | |
1233 | struct timespec _timeout; | |
1234 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1235 | error = -EFAULT; | |
1236 | goto out_free; | |
1237 | } | |
1238 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1239 | _timeout.tv_nsec >= 1000000000L) { | |
1240 | error = -EINVAL; | |
1241 | goto out_free; | |
1242 | } | |
1243 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1244 | } | |
1245 | max = 0; | |
1246 | for (sop = sops; sop < sops + nsops; sop++) { | |
1247 | if (sop->sem_num >= max) | |
1248 | max = sop->sem_num; | |
1249 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1250 | undos = 1; |
1251 | if (sop->sem_op != 0) | |
1da177e4 LT |
1252 | alter = 1; |
1253 | } | |
1da177e4 | 1254 | |
1da177e4 | 1255 | if (undos) { |
4daa28f6 | 1256 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1257 | if (IS_ERR(un)) { |
1258 | error = PTR_ERR(un); | |
1259 | goto out_free; | |
1260 | } | |
1261 | } else | |
1262 | un = NULL; | |
1263 | ||
023a5355 ND |
1264 | sma = sem_lock_check(ns, semid); |
1265 | if (IS_ERR(sma)) { | |
380af1b3 MS |
1266 | if (un) |
1267 | rcu_read_unlock(); | |
023a5355 | 1268 | error = PTR_ERR(sma); |
1da177e4 | 1269 | goto out_free; |
023a5355 ND |
1270 | } |
1271 | ||
1da177e4 | 1272 | /* |
4daa28f6 | 1273 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1274 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1275 | * and now a new array with received the same id. Check and fail. |
380af1b3 MS |
1276 | * This case can be detected checking un->semid. The existance of |
1277 | * "un" itself is guaranteed by rcu. | |
1da177e4 | 1278 | */ |
4daa28f6 | 1279 | error = -EIDRM; |
380af1b3 MS |
1280 | if (un) { |
1281 | if (un->semid == -1) { | |
1282 | rcu_read_unlock(); | |
1283 | goto out_unlock_free; | |
1284 | } else { | |
1285 | /* | |
1286 | * rcu lock can be released, "un" cannot disappear: | |
1287 | * - sem_lock is acquired, thus IPC_RMID is | |
1288 | * impossible. | |
1289 | * - exit_sem is impossible, it always operates on | |
1290 | * current (or a dead task). | |
1291 | */ | |
1292 | ||
1293 | rcu_read_unlock(); | |
1294 | } | |
1295 | } | |
4daa28f6 | 1296 | |
1da177e4 LT |
1297 | error = -EFBIG; |
1298 | if (max >= sma->sem_nsems) | |
1299 | goto out_unlock_free; | |
1300 | ||
1301 | error = -EACCES; | |
1302 | if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) | |
1303 | goto out_unlock_free; | |
1304 | ||
1305 | error = security_sem_semop(sma, sops, nsops, alter); | |
1306 | if (error) | |
1307 | goto out_unlock_free; | |
1308 | ||
b488893a | 1309 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1310 | if (error <= 0) { |
1311 | if (alter && error == 0) | |
fd5db422 | 1312 | do_smart_update(sma, sops, nsops); |
636c6be8 | 1313 | |
1da177e4 LT |
1314 | goto out_unlock_free; |
1315 | } | |
1316 | ||
1317 | /* We need to sleep on this operation, so we put the current | |
1318 | * task into the pending queue and go to sleep. | |
1319 | */ | |
1320 | ||
1da177e4 LT |
1321 | queue.sops = sops; |
1322 | queue.nsops = nsops; | |
1323 | queue.undo = un; | |
b488893a | 1324 | queue.pid = task_tgid_vnr(current); |
1da177e4 LT |
1325 | queue.alter = alter; |
1326 | if (alter) | |
a1193f8e | 1327 | list_add_tail(&queue.list, &sma->sem_pending); |
1da177e4 | 1328 | else |
a1193f8e | 1329 | list_add(&queue.list, &sma->sem_pending); |
1da177e4 | 1330 | |
b97e820f MS |
1331 | if (nsops == 1) { |
1332 | struct sem *curr; | |
1333 | curr = &sma->sem_base[sops->sem_num]; | |
1334 | ||
1335 | if (alter) | |
1336 | list_add_tail(&queue.simple_list, &curr->sem_pending); | |
1337 | else | |
1338 | list_add(&queue.simple_list, &curr->sem_pending); | |
1339 | } else { | |
1340 | INIT_LIST_HEAD(&queue.simple_list); | |
1341 | sma->complex_count++; | |
1342 | } | |
1343 | ||
1da177e4 LT |
1344 | queue.status = -EINTR; |
1345 | queue.sleeper = current; | |
1346 | current->state = TASK_INTERRUPTIBLE; | |
1347 | sem_unlock(sma); | |
1348 | ||
1349 | if (timeout) | |
1350 | jiffies_left = schedule_timeout(jiffies_left); | |
1351 | else | |
1352 | schedule(); | |
1353 | ||
1354 | error = queue.status; | |
1355 | while(unlikely(error == IN_WAKEUP)) { | |
1356 | cpu_relax(); | |
1357 | error = queue.status; | |
1358 | } | |
1359 | ||
1360 | if (error != -EINTR) { | |
1361 | /* fast path: update_queue already obtained all requested | |
1362 | * resources */ | |
1363 | goto out_free; | |
1364 | } | |
1365 | ||
e3893534 | 1366 | sma = sem_lock(ns, semid); |
023a5355 | 1367 | if (IS_ERR(sma)) { |
1da177e4 LT |
1368 | error = -EIDRM; |
1369 | goto out_free; | |
1370 | } | |
1371 | ||
1372 | /* | |
1373 | * If queue.status != -EINTR we are woken up by another process | |
1374 | */ | |
1375 | error = queue.status; | |
1376 | if (error != -EINTR) { | |
1377 | goto out_unlock_free; | |
1378 | } | |
1379 | ||
1380 | /* | |
1381 | * If an interrupt occurred we have to clean up the queue | |
1382 | */ | |
1383 | if (timeout && jiffies_left == 0) | |
1384 | error = -EAGAIN; | |
b97e820f | 1385 | unlink_queue(sma, &queue); |
1da177e4 LT |
1386 | |
1387 | out_unlock_free: | |
1388 | sem_unlock(sma); | |
1389 | out_free: | |
1390 | if(sops != fast_sops) | |
1391 | kfree(sops); | |
1392 | return error; | |
1393 | } | |
1394 | ||
d5460c99 HC |
1395 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
1396 | unsigned, nsops) | |
1da177e4 LT |
1397 | { |
1398 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1399 | } | |
1400 | ||
1401 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1402 | * parent and child tasks. | |
1da177e4 LT |
1403 | */ |
1404 | ||
1405 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1406 | { | |
1407 | struct sem_undo_list *undo_list; | |
1408 | int error; | |
1409 | ||
1410 | if (clone_flags & CLONE_SYSVSEM) { | |
1411 | error = get_undo_list(&undo_list); | |
1412 | if (error) | |
1413 | return error; | |
1da177e4 LT |
1414 | atomic_inc(&undo_list->refcnt); |
1415 | tsk->sysvsem.undo_list = undo_list; | |
1416 | } else | |
1417 | tsk->sysvsem.undo_list = NULL; | |
1418 | ||
1419 | return 0; | |
1420 | } | |
1421 | ||
1422 | /* | |
1423 | * add semadj values to semaphores, free undo structures. | |
1424 | * undo structures are not freed when semaphore arrays are destroyed | |
1425 | * so some of them may be out of date. | |
1426 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1427 | * set of adjustments that needs to be done should be done in an atomic | |
1428 | * manner or not. That is, if we are attempting to decrement the semval | |
1429 | * should we queue up and wait until we can do so legally? | |
1430 | * The original implementation attempted to do this (queue and wait). | |
1431 | * The current implementation does not do so. The POSIX standard | |
1432 | * and SVID should be consulted to determine what behavior is mandated. | |
1433 | */ | |
1434 | void exit_sem(struct task_struct *tsk) | |
1435 | { | |
4daa28f6 | 1436 | struct sem_undo_list *ulp; |
1da177e4 | 1437 | |
4daa28f6 MS |
1438 | ulp = tsk->sysvsem.undo_list; |
1439 | if (!ulp) | |
1da177e4 | 1440 | return; |
9edff4ab | 1441 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 1442 | |
4daa28f6 | 1443 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
1444 | return; |
1445 | ||
380af1b3 | 1446 | for (;;) { |
1da177e4 | 1447 | struct sem_array *sma; |
380af1b3 MS |
1448 | struct sem_undo *un; |
1449 | int semid; | |
4daa28f6 MS |
1450 | int i; |
1451 | ||
380af1b3 | 1452 | rcu_read_lock(); |
05725f7e JP |
1453 | un = list_entry_rcu(ulp->list_proc.next, |
1454 | struct sem_undo, list_proc); | |
380af1b3 MS |
1455 | if (&un->list_proc == &ulp->list_proc) |
1456 | semid = -1; | |
1457 | else | |
1458 | semid = un->semid; | |
1459 | rcu_read_unlock(); | |
4daa28f6 | 1460 | |
380af1b3 MS |
1461 | if (semid == -1) |
1462 | break; | |
1da177e4 | 1463 | |
380af1b3 | 1464 | sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid); |
1da177e4 | 1465 | |
380af1b3 MS |
1466 | /* exit_sem raced with IPC_RMID, nothing to do */ |
1467 | if (IS_ERR(sma)) | |
1468 | continue; | |
1da177e4 | 1469 | |
bf17bb71 | 1470 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
1471 | if (un == NULL) { |
1472 | /* exit_sem raced with IPC_RMID+semget() that created | |
1473 | * exactly the same semid. Nothing to do. | |
1474 | */ | |
1475 | sem_unlock(sma); | |
1476 | continue; | |
1477 | } | |
1478 | ||
1479 | /* remove un from the linked lists */ | |
4daa28f6 MS |
1480 | assert_spin_locked(&sma->sem_perm.lock); |
1481 | list_del(&un->list_id); | |
1482 | ||
380af1b3 MS |
1483 | spin_lock(&ulp->lock); |
1484 | list_del_rcu(&un->list_proc); | |
1485 | spin_unlock(&ulp->lock); | |
1486 | ||
4daa28f6 MS |
1487 | /* perform adjustments registered in un */ |
1488 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 1489 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
1490 | if (un->semadj[i]) { |
1491 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
1492 | /* |
1493 | * Range checks of the new semaphore value, | |
1494 | * not defined by sus: | |
1495 | * - Some unices ignore the undo entirely | |
1496 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
1497 | * - some cap the value (e.g. FreeBSD caps | |
1498 | * at 0, but doesn't enforce SEMVMX) | |
1499 | * | |
1500 | * Linux caps the semaphore value, both at 0 | |
1501 | * and at SEMVMX. | |
1502 | * | |
1503 | * Manfred <manfred@colorfullife.com> | |
1504 | */ | |
5f921ae9 IM |
1505 | if (semaphore->semval < 0) |
1506 | semaphore->semval = 0; | |
1507 | if (semaphore->semval > SEMVMX) | |
1508 | semaphore->semval = SEMVMX; | |
b488893a | 1509 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
1510 | } |
1511 | } | |
1512 | sma->sem_otime = get_seconds(); | |
1513 | /* maybe some queued-up processes were waiting for this */ | |
636c6be8 | 1514 | update_queue(sma, -1); |
1da177e4 | 1515 | sem_unlock(sma); |
380af1b3 MS |
1516 | |
1517 | call_rcu(&un->rcu, free_un); | |
1da177e4 | 1518 | } |
4daa28f6 | 1519 | kfree(ulp); |
1da177e4 LT |
1520 | } |
1521 | ||
1522 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1523 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1524 | { |
19b4946c MW |
1525 | struct sem_array *sma = it; |
1526 | ||
1527 | return seq_printf(s, | |
b97e820f | 1528 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 1529 | sma->sem_perm.key, |
7ca7e564 | 1530 | sma->sem_perm.id, |
19b4946c MW |
1531 | sma->sem_perm.mode, |
1532 | sma->sem_nsems, | |
1533 | sma->sem_perm.uid, | |
1534 | sma->sem_perm.gid, | |
1535 | sma->sem_perm.cuid, | |
1536 | sma->sem_perm.cgid, | |
1537 | sma->sem_otime, | |
1538 | sma->sem_ctime); | |
1da177e4 LT |
1539 | } |
1540 | #endif |