]>
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 | * | |
1da177e4 LT |
6 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
7 | * | |
8 | * SMP-threaded, sysctl's added | |
624dffcb | 9 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 10 | * Enforced range limit on SEM_UNDO |
046c6884 | 11 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
12 | * Lockless wakeup |
13 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
c5cf6359 MS |
14 | * Further wakeup optimizations, documentation |
15 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
16 | * |
17 | * support for audit of ipc object properties and permission changes | |
18 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
19 | * |
20 | * namespaces support | |
21 | * OpenVZ, SWsoft Inc. | |
22 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
23 | * |
24 | * Implementation notes: (May 2010) | |
25 | * This file implements System V semaphores. | |
26 | * | |
27 | * User space visible behavior: | |
28 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
29 | * protection) | |
30 | * - multiple semaphore operations that alter the same semaphore in | |
31 | * one semop() are handled. | |
32 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
33 | * SETALL calls. | |
34 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
35 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
36 | * - namespace are supported. | |
37 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
38 | * to /proc/sys/kernel/sem. | |
39 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
40 | * | |
41 | * Internals: | |
42 | * - scalability: | |
43 | * - all global variables are read-mostly. | |
44 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
45 | * - most operations do write operations (actually: spin_lock calls) to | |
46 | * the per-semaphore array structure. | |
47 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
48 | * If multiple semaphores in one array are used, then cache line | |
49 | * trashing on the semaphore array spinlock will limit the scaling. | |
50 | * - semncnt and semzcnt are calculated on demand in count_semncnt() and | |
51 | * count_semzcnt() | |
52 | * - the task that performs a successful semop() scans the list of all | |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
57 | * dropping all locks. (see wake_up_sem_queue_prepare(), | |
58 | * wake_up_sem_queue_do()) | |
59 | * - All work is done by the waker, the woken up task does not have to do | |
60 | * anything - not even acquiring a lock or dropping a refcount. | |
61 | * - A woken up task may not even touch the semaphore array anymore, it may | |
62 | * have been destroyed already by a semctl(RMID). | |
63 | * - The synchronizations between wake-ups due to a timeout/signal and a | |
64 | * wake-up due to a completed semaphore operation is achieved by using an | |
65 | * intermediate state (IN_WAKEUP). | |
66 | * - UNDO values are stored in an array (one per process and per | |
67 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
68 | * modes for the UNDO variables are supported (per process, per thread) | |
69 | * (see copy_semundo, CLONE_SYSVSEM) | |
70 | * - There are two lists of the pending operations: a per-array list | |
71 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
72 | * ordering without always scanning all pending operations. | |
73 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
74 | */ |
75 | ||
1da177e4 LT |
76 | #include <linux/slab.h> |
77 | #include <linux/spinlock.h> | |
78 | #include <linux/init.h> | |
79 | #include <linux/proc_fs.h> | |
80 | #include <linux/time.h> | |
1da177e4 LT |
81 | #include <linux/security.h> |
82 | #include <linux/syscalls.h> | |
83 | #include <linux/audit.h> | |
c59ede7b | 84 | #include <linux/capability.h> |
19b4946c | 85 | #include <linux/seq_file.h> |
3e148c79 | 86 | #include <linux/rwsem.h> |
e3893534 | 87 | #include <linux/nsproxy.h> |
ae5e1b22 | 88 | #include <linux/ipc_namespace.h> |
5f921ae9 | 89 | |
7153e402 | 90 | #include <linux/uaccess.h> |
1da177e4 LT |
91 | #include "util.h" |
92 | ||
e57940d7 MS |
93 | /* One semaphore structure for each semaphore in the system. */ |
94 | struct sem { | |
95 | int semval; /* current value */ | |
96 | int sempid; /* pid of last operation */ | |
6062a8dc | 97 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
1a82e9e1 MS |
98 | struct list_head pending_alter; /* pending single-sop operations */ |
99 | /* that alter the semaphore */ | |
100 | struct list_head pending_const; /* pending single-sop operations */ | |
101 | /* that do not alter the semaphore*/ | |
d12e1e50 | 102 | time_t sem_otime; /* candidate for sem_otime */ |
f5c936c0 | 103 | } ____cacheline_aligned_in_smp; |
e57940d7 MS |
104 | |
105 | /* One queue for each sleeping process in the system. */ | |
106 | struct sem_queue { | |
e57940d7 MS |
107 | struct list_head list; /* queue of pending operations */ |
108 | struct task_struct *sleeper; /* this process */ | |
109 | struct sem_undo *undo; /* undo structure */ | |
110 | int pid; /* process id of requesting process */ | |
111 | int status; /* completion status of operation */ | |
112 | struct sembuf *sops; /* array of pending operations */ | |
113 | int nsops; /* number of operations */ | |
114 | int alter; /* does *sops alter the array? */ | |
115 | }; | |
116 | ||
117 | /* Each task has a list of undo requests. They are executed automatically | |
118 | * when the process exits. | |
119 | */ | |
120 | struct sem_undo { | |
121 | struct list_head list_proc; /* per-process list: * | |
122 | * all undos from one process | |
123 | * rcu protected */ | |
124 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
125 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
126 | struct list_head list_id; /* per semaphore array list: | |
127 | * all undos for one array */ | |
128 | int semid; /* semaphore set identifier */ | |
129 | short *semadj; /* array of adjustments */ | |
130 | /* one per semaphore */ | |
131 | }; | |
132 | ||
133 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
134 | * that may be shared among all a CLONE_SYSVSEM task group. | |
135 | */ | |
136 | struct sem_undo_list { | |
137 | atomic_t refcnt; | |
138 | spinlock_t lock; | |
139 | struct list_head list_proc; | |
140 | }; | |
141 | ||
142 | ||
ed2ddbf8 | 143 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 144 | |
1b531f21 | 145 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 146 | |
7748dbfa | 147 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 148 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 149 | #ifdef CONFIG_PROC_FS |
19b4946c | 150 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
151 | #endif |
152 | ||
153 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
154 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
155 | ||
156 | /* | |
758a6ba3 | 157 | * Locking: |
1da177e4 | 158 | * sem_undo.id_next, |
758a6ba3 | 159 | * sem_array.complex_count, |
1a82e9e1 | 160 | * sem_array.pending{_alter,_cont}, |
758a6ba3 | 161 | * sem_array.sem_undo: global sem_lock() for read/write |
1da177e4 | 162 | * sem_undo.proc_next: only "current" is allowed to read/write that field. |
46c0a8ca | 163 | * |
758a6ba3 MS |
164 | * sem_array.sem_base[i].pending_{const,alter}: |
165 | * global or semaphore sem_lock() for read/write | |
1da177e4 LT |
166 | */ |
167 | ||
e3893534 KK |
168 | #define sc_semmsl sem_ctls[0] |
169 | #define sc_semmns sem_ctls[1] | |
170 | #define sc_semopm sem_ctls[2] | |
171 | #define sc_semmni sem_ctls[3] | |
172 | ||
ed2ddbf8 | 173 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 174 | { |
e3893534 KK |
175 | ns->sc_semmsl = SEMMSL; |
176 | ns->sc_semmns = SEMMNS; | |
177 | ns->sc_semopm = SEMOPM; | |
178 | ns->sc_semmni = SEMMNI; | |
179 | ns->used_sems = 0; | |
ed2ddbf8 | 180 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
181 | } |
182 | ||
ae5e1b22 | 183 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
184 | void sem_exit_ns(struct ipc_namespace *ns) |
185 | { | |
01b8b07a | 186 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 187 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 188 | } |
ae5e1b22 | 189 | #endif |
1da177e4 | 190 | |
239521f3 | 191 | void __init sem_init(void) |
1da177e4 | 192 | { |
ed2ddbf8 | 193 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
194 | ipc_init_proc_interface("sysvipc/sem", |
195 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 196 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
197 | } |
198 | ||
f269f40a MS |
199 | /** |
200 | * unmerge_queues - unmerge queues, if possible. | |
201 | * @sma: semaphore array | |
202 | * | |
203 | * The function unmerges the wait queues if complex_count is 0. | |
204 | * It must be called prior to dropping the global semaphore array lock. | |
205 | */ | |
206 | static void unmerge_queues(struct sem_array *sma) | |
207 | { | |
208 | struct sem_queue *q, *tq; | |
209 | ||
210 | /* complex operations still around? */ | |
211 | if (sma->complex_count) | |
212 | return; | |
213 | /* | |
214 | * We will switch back to simple mode. | |
215 | * Move all pending operation back into the per-semaphore | |
216 | * queues. | |
217 | */ | |
218 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
219 | struct sem *curr; | |
220 | curr = &sma->sem_base[q->sops[0].sem_num]; | |
221 | ||
222 | list_add_tail(&q->list, &curr->pending_alter); | |
223 | } | |
224 | INIT_LIST_HEAD(&sma->pending_alter); | |
225 | } | |
226 | ||
227 | /** | |
8001c858 | 228 | * merge_queues - merge single semop queues into global queue |
f269f40a MS |
229 | * @sma: semaphore array |
230 | * | |
231 | * This function merges all per-semaphore queues into the global queue. | |
232 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
233 | * operations when a multi-semop operation must sleep. | |
234 | * Only the alter operations must be moved, the const operations can stay. | |
235 | */ | |
236 | static void merge_queues(struct sem_array *sma) | |
237 | { | |
238 | int i; | |
239 | for (i = 0; i < sma->sem_nsems; i++) { | |
240 | struct sem *sem = sma->sem_base + i; | |
241 | ||
242 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
243 | } | |
244 | } | |
245 | ||
53dad6d3 DB |
246 | static void sem_rcu_free(struct rcu_head *head) |
247 | { | |
248 | struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu); | |
249 | struct sem_array *sma = ipc_rcu_to_struct(p); | |
250 | ||
251 | security_sem_free(sma); | |
252 | ipc_rcu_free(head); | |
253 | } | |
254 | ||
5e9d5275 MS |
255 | /* |
256 | * Wait until all currently ongoing simple ops have completed. | |
257 | * Caller must own sem_perm.lock. | |
258 | * New simple ops cannot start, because simple ops first check | |
259 | * that sem_perm.lock is free. | |
6d07b68c | 260 | * that a) sem_perm.lock is free and b) complex_count is 0. |
5e9d5275 MS |
261 | */ |
262 | static void sem_wait_array(struct sem_array *sma) | |
263 | { | |
264 | int i; | |
265 | struct sem *sem; | |
266 | ||
6d07b68c MS |
267 | if (sma->complex_count) { |
268 | /* The thread that increased sma->complex_count waited on | |
269 | * all sem->lock locks. Thus we don't need to wait again. | |
270 | */ | |
271 | return; | |
272 | } | |
273 | ||
5e9d5275 MS |
274 | for (i = 0; i < sma->sem_nsems; i++) { |
275 | sem = sma->sem_base + i; | |
276 | spin_unlock_wait(&sem->lock); | |
277 | } | |
278 | } | |
279 | ||
6062a8dc RR |
280 | /* |
281 | * If the request contains only one semaphore operation, and there are | |
282 | * no complex transactions pending, lock only the semaphore involved. | |
283 | * Otherwise, lock the entire semaphore array, since we either have | |
284 | * multiple semaphores in our own semops, or we need to look at | |
285 | * semaphores from other pending complex operations. | |
6062a8dc RR |
286 | */ |
287 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
288 | int nsops) | |
289 | { | |
5e9d5275 | 290 | struct sem *sem; |
6062a8dc | 291 | |
5e9d5275 MS |
292 | if (nsops != 1) { |
293 | /* Complex operation - acquire a full lock */ | |
294 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 295 | |
5e9d5275 MS |
296 | /* And wait until all simple ops that are processed |
297 | * right now have dropped their locks. | |
6062a8dc | 298 | */ |
5e9d5275 MS |
299 | sem_wait_array(sma); |
300 | return -1; | |
301 | } | |
302 | ||
303 | /* | |
304 | * Only one semaphore affected - try to optimize locking. | |
305 | * The rules are: | |
306 | * - optimized locking is possible if no complex operation | |
307 | * is either enqueued or processed right now. | |
308 | * - The test for enqueued complex ops is simple: | |
309 | * sma->complex_count != 0 | |
310 | * - Testing for complex ops that are processed right now is | |
311 | * a bit more difficult. Complex ops acquire the full lock | |
312 | * and first wait that the running simple ops have completed. | |
313 | * (see above) | |
314 | * Thus: If we own a simple lock and the global lock is free | |
315 | * and complex_count is now 0, then it will stay 0 and | |
316 | * thus just locking sem->lock is sufficient. | |
317 | */ | |
318 | sem = sma->sem_base + sops->sem_num; | |
6062a8dc | 319 | |
5e9d5275 | 320 | if (sma->complex_count == 0) { |
6062a8dc | 321 | /* |
5e9d5275 MS |
322 | * It appears that no complex operation is around. |
323 | * Acquire the per-semaphore lock. | |
6062a8dc | 324 | */ |
5e9d5275 MS |
325 | spin_lock(&sem->lock); |
326 | ||
327 | /* Then check that the global lock is free */ | |
328 | if (!spin_is_locked(&sma->sem_perm.lock)) { | |
329 | /* spin_is_locked() is not a memory barrier */ | |
330 | smp_mb(); | |
331 | ||
332 | /* Now repeat the test of complex_count: | |
333 | * It can't change anymore until we drop sem->lock. | |
334 | * Thus: if is now 0, then it will stay 0. | |
335 | */ | |
336 | if (sma->complex_count == 0) { | |
337 | /* fast path successful! */ | |
338 | return sops->sem_num; | |
339 | } | |
6062a8dc | 340 | } |
5e9d5275 MS |
341 | spin_unlock(&sem->lock); |
342 | } | |
343 | ||
344 | /* slow path: acquire the full lock */ | |
345 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 346 | |
5e9d5275 MS |
347 | if (sma->complex_count == 0) { |
348 | /* False alarm: | |
349 | * There is no complex operation, thus we can switch | |
350 | * back to the fast path. | |
351 | */ | |
352 | spin_lock(&sem->lock); | |
353 | ipc_unlock_object(&sma->sem_perm); | |
354 | return sops->sem_num; | |
6062a8dc | 355 | } else { |
5e9d5275 MS |
356 | /* Not a false alarm, thus complete the sequence for a |
357 | * full lock. | |
6062a8dc | 358 | */ |
5e9d5275 MS |
359 | sem_wait_array(sma); |
360 | return -1; | |
6062a8dc | 361 | } |
6062a8dc RR |
362 | } |
363 | ||
364 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
365 | { | |
366 | if (locknum == -1) { | |
f269f40a | 367 | unmerge_queues(sma); |
cf9d5d78 | 368 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc RR |
369 | } else { |
370 | struct sem *sem = sma->sem_base + locknum; | |
371 | spin_unlock(&sem->lock); | |
372 | } | |
6062a8dc RR |
373 | } |
374 | ||
3e148c79 | 375 | /* |
d9a605e4 | 376 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 377 | * is not held. |
321310ce LT |
378 | * |
379 | * The caller holds the RCU read lock. | |
3e148c79 | 380 | */ |
6062a8dc RR |
381 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
382 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 383 | { |
c460b662 RR |
384 | struct kern_ipc_perm *ipcp; |
385 | struct sem_array *sma; | |
03f02c76 | 386 | |
c460b662 | 387 | ipcp = ipc_obtain_object(&sem_ids(ns), id); |
321310ce LT |
388 | if (IS_ERR(ipcp)) |
389 | return ERR_CAST(ipcp); | |
b1ed88b4 | 390 | |
6062a8dc RR |
391 | sma = container_of(ipcp, struct sem_array, sem_perm); |
392 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
393 | |
394 | /* ipc_rmid() may have already freed the ID while sem_lock | |
395 | * was spinning: verify that the structure is still valid | |
396 | */ | |
72a8ff2f | 397 | if (ipc_valid_object(ipcp)) |
c460b662 RR |
398 | return container_of(ipcp, struct sem_array, sem_perm); |
399 | ||
6062a8dc | 400 | sem_unlock(sma, *locknum); |
321310ce | 401 | return ERR_PTR(-EINVAL); |
023a5355 ND |
402 | } |
403 | ||
16df3674 DB |
404 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
405 | { | |
406 | struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id); | |
407 | ||
408 | if (IS_ERR(ipcp)) | |
409 | return ERR_CAST(ipcp); | |
410 | ||
411 | return container_of(ipcp, struct sem_array, sem_perm); | |
412 | } | |
413 | ||
16df3674 DB |
414 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
415 | int id) | |
416 | { | |
417 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
418 | ||
419 | if (IS_ERR(ipcp)) | |
420 | return ERR_CAST(ipcp); | |
b1ed88b4 | 421 | |
03f02c76 | 422 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
423 | } |
424 | ||
6ff37972 PP |
425 | static inline void sem_lock_and_putref(struct sem_array *sma) |
426 | { | |
6062a8dc | 427 | sem_lock(sma, NULL, -1); |
53dad6d3 | 428 | ipc_rcu_putref(sma, ipc_rcu_free); |
6ff37972 PP |
429 | } |
430 | ||
7ca7e564 ND |
431 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
432 | { | |
433 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
434 | } | |
435 | ||
1da177e4 LT |
436 | /* |
437 | * Lockless wakeup algorithm: | |
438 | * Without the check/retry algorithm a lockless wakeup is possible: | |
439 | * - queue.status is initialized to -EINTR before blocking. | |
440 | * - wakeup is performed by | |
1a82e9e1 | 441 | * * unlinking the queue entry from the pending list |
1da177e4 LT |
442 | * * setting queue.status to IN_WAKEUP |
443 | * This is the notification for the blocked thread that a | |
444 | * result value is imminent. | |
445 | * * call wake_up_process | |
446 | * * set queue.status to the final value. | |
447 | * - the previously blocked thread checks queue.status: | |
239521f3 MS |
448 | * * if it's IN_WAKEUP, then it must wait until the value changes |
449 | * * if it's not -EINTR, then the operation was completed by | |
450 | * update_queue. semtimedop can return queue.status without | |
451 | * performing any operation on the sem array. | |
452 | * * otherwise it must acquire the spinlock and check what's up. | |
1da177e4 LT |
453 | * |
454 | * The two-stage algorithm is necessary to protect against the following | |
455 | * races: | |
456 | * - if queue.status is set after wake_up_process, then the woken up idle | |
457 | * thread could race forward and try (and fail) to acquire sma->lock | |
458 | * before update_queue had a chance to set queue.status | |
459 | * - if queue.status is written before wake_up_process and if the | |
460 | * blocked process is woken up by a signal between writing | |
461 | * queue.status and the wake_up_process, then the woken up | |
462 | * process could return from semtimedop and die by calling | |
463 | * sys_exit before wake_up_process is called. Then wake_up_process | |
464 | * will oops, because the task structure is already invalid. | |
465 | * (yes, this happened on s390 with sysv msg). | |
466 | * | |
467 | */ | |
468 | #define IN_WAKEUP 1 | |
469 | ||
f4566f04 ND |
470 | /** |
471 | * newary - Create a new semaphore set | |
472 | * @ns: namespace | |
473 | * @params: ptr to the structure that contains key, semflg and nsems | |
474 | * | |
d9a605e4 | 475 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 | 476 | */ |
7748dbfa | 477 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
478 | { |
479 | int id; | |
480 | int retval; | |
481 | struct sem_array *sma; | |
482 | int size; | |
7748dbfa ND |
483 | key_t key = params->key; |
484 | int nsems = params->u.nsems; | |
485 | int semflg = params->flg; | |
b97e820f | 486 | int i; |
1da177e4 LT |
487 | |
488 | if (!nsems) | |
489 | return -EINVAL; | |
e3893534 | 490 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
491 | return -ENOSPC; |
492 | ||
239521f3 | 493 | size = sizeof(*sma) + nsems * sizeof(struct sem); |
1da177e4 | 494 | sma = ipc_rcu_alloc(size); |
3ab08fe2 | 495 | if (!sma) |
1da177e4 | 496 | return -ENOMEM; |
3ab08fe2 | 497 | |
239521f3 | 498 | memset(sma, 0, size); |
1da177e4 LT |
499 | |
500 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
501 | sma->sem_perm.key = key; | |
502 | ||
503 | sma->sem_perm.security = NULL; | |
504 | retval = security_sem_alloc(sma); | |
505 | if (retval) { | |
53dad6d3 | 506 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
507 | return retval; |
508 | } | |
509 | ||
e3893534 | 510 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 511 | if (id < 0) { |
53dad6d3 | 512 | ipc_rcu_putref(sma, sem_rcu_free); |
283bb7fa | 513 | return id; |
1da177e4 | 514 | } |
e3893534 | 515 | ns->used_sems += nsems; |
1da177e4 LT |
516 | |
517 | sma->sem_base = (struct sem *) &sma[1]; | |
b97e820f | 518 | |
6062a8dc | 519 | for (i = 0; i < nsems; i++) { |
1a82e9e1 MS |
520 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
521 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
522 | spin_lock_init(&sma->sem_base[i].lock); |
523 | } | |
b97e820f MS |
524 | |
525 | sma->complex_count = 0; | |
1a82e9e1 MS |
526 | INIT_LIST_HEAD(&sma->pending_alter); |
527 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 528 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
529 | sma->sem_nsems = nsems; |
530 | sma->sem_ctime = get_seconds(); | |
6062a8dc | 531 | sem_unlock(sma, -1); |
6d49dab8 | 532 | rcu_read_unlock(); |
1da177e4 | 533 | |
7ca7e564 | 534 | return sma->sem_perm.id; |
1da177e4 LT |
535 | } |
536 | ||
7748dbfa | 537 | |
f4566f04 | 538 | /* |
d9a605e4 | 539 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 540 | */ |
03f02c76 | 541 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 542 | { |
03f02c76 ND |
543 | struct sem_array *sma; |
544 | ||
545 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
546 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
547 | } |
548 | ||
f4566f04 | 549 | /* |
d9a605e4 | 550 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 551 | */ |
03f02c76 ND |
552 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
553 | struct ipc_params *params) | |
7748dbfa | 554 | { |
03f02c76 ND |
555 | struct sem_array *sma; |
556 | ||
557 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
558 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
559 | return -EINVAL; |
560 | ||
561 | return 0; | |
562 | } | |
563 | ||
d5460c99 | 564 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 565 | { |
e3893534 | 566 | struct ipc_namespace *ns; |
eb66ec44 MK |
567 | static const struct ipc_ops sem_ops = { |
568 | .getnew = newary, | |
569 | .associate = sem_security, | |
570 | .more_checks = sem_more_checks, | |
571 | }; | |
7748dbfa | 572 | struct ipc_params sem_params; |
e3893534 KK |
573 | |
574 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 575 | |
e3893534 | 576 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 577 | return -EINVAL; |
7ca7e564 | 578 | |
7748dbfa ND |
579 | sem_params.key = key; |
580 | sem_params.flg = semflg; | |
581 | sem_params.u.nsems = nsems; | |
1da177e4 | 582 | |
7748dbfa | 583 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
584 | } |
585 | ||
78f5009c PM |
586 | /** |
587 | * perform_atomic_semop - Perform (if possible) a semaphore operation | |
758a6ba3 MS |
588 | * @sma: semaphore array |
589 | * @sops: array with operations that should be checked | |
78f5009c | 590 | * @nsops: number of operations |
758a6ba3 MS |
591 | * @un: undo array |
592 | * @pid: pid that did the change | |
593 | * | |
594 | * Returns 0 if the operation was possible. | |
595 | * Returns 1 if the operation is impossible, the caller must sleep. | |
596 | * Negative values are error codes. | |
1da177e4 | 597 | */ |
758a6ba3 | 598 | static int perform_atomic_semop(struct sem_array *sma, struct sembuf *sops, |
1da177e4 LT |
599 | int nsops, struct sem_undo *un, int pid) |
600 | { | |
601 | int result, sem_op; | |
602 | struct sembuf *sop; | |
239521f3 | 603 | struct sem *curr; |
1da177e4 LT |
604 | |
605 | for (sop = sops; sop < sops + nsops; sop++) { | |
606 | curr = sma->sem_base + sop->sem_num; | |
607 | sem_op = sop->sem_op; | |
608 | result = curr->semval; | |
78f5009c | 609 | |
1da177e4 LT |
610 | if (!sem_op && result) |
611 | goto would_block; | |
612 | ||
613 | result += sem_op; | |
614 | if (result < 0) | |
615 | goto would_block; | |
616 | if (result > SEMVMX) | |
617 | goto out_of_range; | |
78f5009c | 618 | |
1da177e4 LT |
619 | if (sop->sem_flg & SEM_UNDO) { |
620 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 621 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
622 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
623 | goto out_of_range; | |
78f5009c | 624 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 625 | } |
78f5009c | 626 | |
1da177e4 LT |
627 | curr->semval = result; |
628 | } | |
629 | ||
630 | sop--; | |
631 | while (sop >= sops) { | |
632 | sma->sem_base[sop->sem_num].sempid = pid; | |
1da177e4 LT |
633 | sop--; |
634 | } | |
78f5009c | 635 | |
1da177e4 LT |
636 | return 0; |
637 | ||
638 | out_of_range: | |
639 | result = -ERANGE; | |
640 | goto undo; | |
641 | ||
642 | would_block: | |
643 | if (sop->sem_flg & IPC_NOWAIT) | |
644 | result = -EAGAIN; | |
645 | else | |
646 | result = 1; | |
647 | ||
648 | undo: | |
649 | sop--; | |
650 | while (sop >= sops) { | |
78f5009c PM |
651 | sem_op = sop->sem_op; |
652 | sma->sem_base[sop->sem_num].semval -= sem_op; | |
653 | if (sop->sem_flg & SEM_UNDO) | |
654 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
655 | sop--; |
656 | } | |
657 | ||
658 | return result; | |
659 | } | |
660 | ||
0a2b9d4c MS |
661 | /** wake_up_sem_queue_prepare(q, error): Prepare wake-up |
662 | * @q: queue entry that must be signaled | |
663 | * @error: Error value for the signal | |
664 | * | |
665 | * Prepare the wake-up of the queue entry q. | |
d4212093 | 666 | */ |
0a2b9d4c MS |
667 | static void wake_up_sem_queue_prepare(struct list_head *pt, |
668 | struct sem_queue *q, int error) | |
d4212093 | 669 | { |
0a2b9d4c MS |
670 | if (list_empty(pt)) { |
671 | /* | |
672 | * Hold preempt off so that we don't get preempted and have the | |
673 | * wakee busy-wait until we're scheduled back on. | |
674 | */ | |
675 | preempt_disable(); | |
676 | } | |
d4212093 | 677 | q->status = IN_WAKEUP; |
0a2b9d4c MS |
678 | q->pid = error; |
679 | ||
9f1bc2c9 | 680 | list_add_tail(&q->list, pt); |
0a2b9d4c MS |
681 | } |
682 | ||
683 | /** | |
8001c858 | 684 | * wake_up_sem_queue_do - do the actual wake-up |
0a2b9d4c MS |
685 | * @pt: list of tasks to be woken up |
686 | * | |
687 | * Do the actual wake-up. | |
688 | * The function is called without any locks held, thus the semaphore array | |
689 | * could be destroyed already and the tasks can disappear as soon as the | |
690 | * status is set to the actual return code. | |
691 | */ | |
692 | static void wake_up_sem_queue_do(struct list_head *pt) | |
693 | { | |
694 | struct sem_queue *q, *t; | |
695 | int did_something; | |
696 | ||
697 | did_something = !list_empty(pt); | |
9f1bc2c9 | 698 | list_for_each_entry_safe(q, t, pt, list) { |
0a2b9d4c MS |
699 | wake_up_process(q->sleeper); |
700 | /* q can disappear immediately after writing q->status. */ | |
701 | smp_wmb(); | |
702 | q->status = q->pid; | |
703 | } | |
704 | if (did_something) | |
705 | preempt_enable(); | |
d4212093 NP |
706 | } |
707 | ||
b97e820f MS |
708 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
709 | { | |
710 | list_del(&q->list); | |
9f1bc2c9 | 711 | if (q->nsops > 1) |
b97e820f MS |
712 | sma->complex_count--; |
713 | } | |
714 | ||
fd5db422 MS |
715 | /** check_restart(sma, q) |
716 | * @sma: semaphore array | |
717 | * @q: the operation that just completed | |
718 | * | |
719 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
720 | * waiting operations. Therefore this function checks if the restart is | |
721 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
722 | * modified the array. |
723 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 MS |
724 | */ |
725 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
726 | { | |
1a82e9e1 MS |
727 | /* pending complex alter operations are too difficult to analyse */ |
728 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
729 | return 1; |
730 | ||
731 | /* we were a sleeping complex operation. Too difficult */ | |
732 | if (q->nsops > 1) | |
733 | return 1; | |
734 | ||
1a82e9e1 MS |
735 | /* It is impossible that someone waits for the new value: |
736 | * - complex operations always restart. | |
737 | * - wait-for-zero are handled seperately. | |
738 | * - q is a previously sleeping simple operation that | |
739 | * altered the array. It must be a decrement, because | |
740 | * simple increments never sleep. | |
741 | * - If there are older (higher priority) decrements | |
742 | * in the queue, then they have observed the original | |
743 | * semval value and couldn't proceed. The operation | |
744 | * decremented to value - thus they won't proceed either. | |
745 | */ | |
746 | return 0; | |
747 | } | |
fd5db422 | 748 | |
1a82e9e1 | 749 | /** |
8001c858 | 750 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
751 | * @sma: semaphore array. |
752 | * @semnum: semaphore that was modified. | |
753 | * @pt: list head for the tasks that must be woken up. | |
754 | * | |
755 | * wake_const_ops must be called after a semaphore in a semaphore array | |
756 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
757 | * be called with semnum = -1, as well as with the number of each modified | |
758 | * semaphore. | |
759 | * The tasks that must be woken up are added to @pt. The return code | |
760 | * is stored in q->pid. | |
761 | * The function returns 1 if at least one operation was completed successfully. | |
762 | */ | |
763 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
764 | struct list_head *pt) | |
765 | { | |
766 | struct sem_queue *q; | |
767 | struct list_head *walk; | |
768 | struct list_head *pending_list; | |
769 | int semop_completed = 0; | |
770 | ||
771 | if (semnum == -1) | |
772 | pending_list = &sma->pending_const; | |
773 | else | |
774 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 775 | |
1a82e9e1 MS |
776 | walk = pending_list->next; |
777 | while (walk != pending_list) { | |
778 | int error; | |
779 | ||
780 | q = container_of(walk, struct sem_queue, list); | |
781 | walk = walk->next; | |
782 | ||
758a6ba3 MS |
783 | error = perform_atomic_semop(sma, q->sops, q->nsops, |
784 | q->undo, q->pid); | |
1a82e9e1 MS |
785 | |
786 | if (error <= 0) { | |
787 | /* operation completed, remove from queue & wakeup */ | |
788 | ||
789 | unlink_queue(sma, q); | |
790 | ||
791 | wake_up_sem_queue_prepare(pt, q, error); | |
792 | if (error == 0) | |
793 | semop_completed = 1; | |
794 | } | |
795 | } | |
796 | return semop_completed; | |
797 | } | |
798 | ||
799 | /** | |
8001c858 | 800 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
801 | * @sma: semaphore array |
802 | * @sops: operations that were performed | |
803 | * @nsops: number of operations | |
804 | * @pt: list head of the tasks that must be woken up. | |
805 | * | |
8001c858 DB |
806 | * Checks all required queue for wait-for-zero operations, based |
807 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
808 | * The function returns 1 if at least one operation was completed successfully. |
809 | */ | |
810 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
811 | int nsops, struct list_head *pt) | |
812 | { | |
813 | int i; | |
814 | int semop_completed = 0; | |
815 | int got_zero = 0; | |
816 | ||
817 | /* first: the per-semaphore queues, if known */ | |
818 | if (sops) { | |
819 | for (i = 0; i < nsops; i++) { | |
820 | int num = sops[i].sem_num; | |
821 | ||
822 | if (sma->sem_base[num].semval == 0) { | |
823 | got_zero = 1; | |
824 | semop_completed |= wake_const_ops(sma, num, pt); | |
825 | } | |
826 | } | |
827 | } else { | |
828 | /* | |
829 | * No sops means modified semaphores not known. | |
830 | * Assume all were changed. | |
fd5db422 | 831 | */ |
1a82e9e1 MS |
832 | for (i = 0; i < sma->sem_nsems; i++) { |
833 | if (sma->sem_base[i].semval == 0) { | |
834 | got_zero = 1; | |
835 | semop_completed |= wake_const_ops(sma, i, pt); | |
836 | } | |
837 | } | |
fd5db422 MS |
838 | } |
839 | /* | |
1a82e9e1 MS |
840 | * If one of the modified semaphores got 0, |
841 | * then check the global queue, too. | |
fd5db422 | 842 | */ |
1a82e9e1 MS |
843 | if (got_zero) |
844 | semop_completed |= wake_const_ops(sma, -1, pt); | |
fd5db422 | 845 | |
1a82e9e1 | 846 | return semop_completed; |
fd5db422 MS |
847 | } |
848 | ||
636c6be8 MS |
849 | |
850 | /** | |
8001c858 | 851 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
852 | * @sma: semaphore array. |
853 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 854 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
855 | * |
856 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
857 | * was modified. If multiple semaphores were modified, update_queue must |
858 | * be called with semnum = -1, as well as with the number of each modified | |
859 | * semaphore. | |
0a2b9d4c MS |
860 | * The tasks that must be woken up are added to @pt. The return code |
861 | * is stored in q->pid. | |
1a82e9e1 MS |
862 | * The function internally checks if const operations can now succeed. |
863 | * | |
0a2b9d4c | 864 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 865 | */ |
0a2b9d4c | 866 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 867 | { |
636c6be8 MS |
868 | struct sem_queue *q; |
869 | struct list_head *walk; | |
870 | struct list_head *pending_list; | |
0a2b9d4c | 871 | int semop_completed = 0; |
636c6be8 | 872 | |
9f1bc2c9 | 873 | if (semnum == -1) |
1a82e9e1 | 874 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 875 | else |
1a82e9e1 | 876 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
877 | |
878 | again: | |
636c6be8 MS |
879 | walk = pending_list->next; |
880 | while (walk != pending_list) { | |
fd5db422 | 881 | int error, restart; |
636c6be8 | 882 | |
9f1bc2c9 | 883 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 884 | walk = walk->next; |
1da177e4 | 885 | |
d987f8b2 MS |
886 | /* If we are scanning the single sop, per-semaphore list of |
887 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 888 | * necessary to scan further: simple increments |
d987f8b2 MS |
889 | * that affect only one entry succeed immediately and cannot |
890 | * be in the per semaphore pending queue, and decrements | |
891 | * cannot be successful if the value is already 0. | |
892 | */ | |
1a82e9e1 | 893 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
894 | break; |
895 | ||
758a6ba3 | 896 | error = perform_atomic_semop(sma, q->sops, q->nsops, |
1da177e4 LT |
897 | q->undo, q->pid); |
898 | ||
899 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
900 | if (error > 0) |
901 | continue; | |
902 | ||
b97e820f | 903 | unlink_queue(sma, q); |
9cad200c | 904 | |
0a2b9d4c | 905 | if (error) { |
fd5db422 | 906 | restart = 0; |
0a2b9d4c MS |
907 | } else { |
908 | semop_completed = 1; | |
1a82e9e1 | 909 | do_smart_wakeup_zero(sma, q->sops, q->nsops, pt); |
fd5db422 | 910 | restart = check_restart(sma, q); |
0a2b9d4c | 911 | } |
fd5db422 | 912 | |
0a2b9d4c | 913 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 914 | if (restart) |
9cad200c | 915 | goto again; |
1da177e4 | 916 | } |
0a2b9d4c | 917 | return semop_completed; |
1da177e4 LT |
918 | } |
919 | ||
0e8c6656 | 920 | /** |
8001c858 | 921 | * set_semotime - set sem_otime |
0e8c6656 MS |
922 | * @sma: semaphore array |
923 | * @sops: operations that modified the array, may be NULL | |
924 | * | |
925 | * sem_otime is replicated to avoid cache line trashing. | |
926 | * This function sets one instance to the current time. | |
927 | */ | |
928 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
929 | { | |
930 | if (sops == NULL) { | |
931 | sma->sem_base[0].sem_otime = get_seconds(); | |
932 | } else { | |
933 | sma->sem_base[sops[0].sem_num].sem_otime = | |
934 | get_seconds(); | |
935 | } | |
936 | } | |
937 | ||
0a2b9d4c | 938 | /** |
8001c858 | 939 | * do_smart_update - optimized update_queue |
fd5db422 MS |
940 | * @sma: semaphore array |
941 | * @sops: operations that were performed | |
942 | * @nsops: number of operations | |
0a2b9d4c MS |
943 | * @otime: force setting otime |
944 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 | 945 | * |
1a82e9e1 MS |
946 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
947 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
948 | * Note that the function does not do the actual wake-up: the caller is |
949 | * responsible for calling wake_up_sem_queue_do(@pt). | |
950 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 951 | */ |
0a2b9d4c MS |
952 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
953 | int otime, struct list_head *pt) | |
fd5db422 MS |
954 | { |
955 | int i; | |
956 | ||
1a82e9e1 MS |
957 | otime |= do_smart_wakeup_zero(sma, sops, nsops, pt); |
958 | ||
f269f40a MS |
959 | if (!list_empty(&sma->pending_alter)) { |
960 | /* semaphore array uses the global queue - just process it. */ | |
961 | otime |= update_queue(sma, -1, pt); | |
962 | } else { | |
963 | if (!sops) { | |
964 | /* | |
965 | * No sops, thus the modified semaphores are not | |
966 | * known. Check all. | |
967 | */ | |
968 | for (i = 0; i < sma->sem_nsems; i++) | |
969 | otime |= update_queue(sma, i, pt); | |
970 | } else { | |
971 | /* | |
972 | * Check the semaphores that were increased: | |
973 | * - No complex ops, thus all sleeping ops are | |
974 | * decrease. | |
975 | * - if we decreased the value, then any sleeping | |
976 | * semaphore ops wont be able to run: If the | |
977 | * previous value was too small, then the new | |
978 | * value will be too small, too. | |
979 | */ | |
980 | for (i = 0; i < nsops; i++) { | |
981 | if (sops[i].sem_op > 0) { | |
982 | otime |= update_queue(sma, | |
983 | sops[i].sem_num, pt); | |
984 | } | |
ab465df9 | 985 | } |
9f1bc2c9 | 986 | } |
fd5db422 | 987 | } |
0e8c6656 MS |
988 | if (otime) |
989 | set_semotime(sma, sops); | |
fd5db422 MS |
990 | } |
991 | ||
1da177e4 LT |
992 | /* The following counts are associated to each semaphore: |
993 | * semncnt number of tasks waiting on semval being nonzero | |
994 | * semzcnt number of tasks waiting on semval being zero | |
995 | * This model assumes that a task waits on exactly one semaphore. | |
996 | * Since semaphore operations are to be performed atomically, tasks actually | |
997 | * wait on a whole sequence of semaphores simultaneously. | |
998 | * The counts we return here are a rough approximation, but still | |
999 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
1000 | */ | |
239521f3 | 1001 | static int count_semncnt(struct sem_array *sma, ushort semnum) |
1da177e4 LT |
1002 | { |
1003 | int semncnt; | |
239521f3 | 1004 | struct sem_queue *q; |
1da177e4 LT |
1005 | |
1006 | semncnt = 0; | |
1a82e9e1 | 1007 | list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) { |
239521f3 | 1008 | struct sembuf *sops = q->sops; |
de2657f9 RR |
1009 | BUG_ON(sops->sem_num != semnum); |
1010 | if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
1011 | semncnt++; | |
1012 | } | |
1013 | ||
1a82e9e1 | 1014 | list_for_each_entry(q, &sma->pending_alter, list) { |
239521f3 | 1015 | struct sembuf *sops = q->sops; |
1da177e4 LT |
1016 | int nsops = q->nsops; |
1017 | int i; | |
1018 | for (i = 0; i < nsops; i++) | |
1019 | if (sops[i].sem_num == semnum | |
1020 | && (sops[i].sem_op < 0) | |
1021 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
1022 | semncnt++; | |
1023 | } | |
1024 | return semncnt; | |
1025 | } | |
a1193f8e | 1026 | |
239521f3 | 1027 | static int count_semzcnt(struct sem_array *sma, ushort semnum) |
1da177e4 LT |
1028 | { |
1029 | int semzcnt; | |
239521f3 | 1030 | struct sem_queue *q; |
1da177e4 LT |
1031 | |
1032 | semzcnt = 0; | |
1a82e9e1 | 1033 | list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) { |
239521f3 | 1034 | struct sembuf *sops = q->sops; |
ebc2e5e6 RR |
1035 | BUG_ON(sops->sem_num != semnum); |
1036 | if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
1037 | semzcnt++; | |
1038 | } | |
1039 | ||
1a82e9e1 | 1040 | list_for_each_entry(q, &sma->pending_const, list) { |
239521f3 | 1041 | struct sembuf *sops = q->sops; |
1da177e4 LT |
1042 | int nsops = q->nsops; |
1043 | int i; | |
1044 | for (i = 0; i < nsops; i++) | |
1045 | if (sops[i].sem_num == semnum | |
1046 | && (sops[i].sem_op == 0) | |
1047 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
1048 | semzcnt++; | |
1049 | } | |
1994862d MS |
1050 | list_for_each_entry(q, &sma->pending_alter, list) { |
1051 | struct sembuf *sops = q->sops; | |
1052 | int nsops = q->nsops; | |
1053 | int i; | |
1054 | for (i = 0; i < nsops; i++) | |
1055 | if (sops[i].sem_num == semnum | |
1056 | && (sops[i].sem_op == 0) | |
1057 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
1058 | semzcnt++; | |
1059 | } | |
1da177e4 LT |
1060 | return semzcnt; |
1061 | } | |
1062 | ||
d9a605e4 DB |
1063 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1064 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1065 | * remains locked on exit. |
1da177e4 | 1066 | */ |
01b8b07a | 1067 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1068 | { |
380af1b3 MS |
1069 | struct sem_undo *un, *tu; |
1070 | struct sem_queue *q, *tq; | |
01b8b07a | 1071 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 1072 | struct list_head tasks; |
9f1bc2c9 | 1073 | int i; |
1da177e4 | 1074 | |
380af1b3 | 1075 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1076 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1077 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1078 | list_del(&un->list_id); | |
1079 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1080 | un->semid = -1; |
380af1b3 MS |
1081 | list_del_rcu(&un->list_proc); |
1082 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1083 | kfree_rcu(un, rcu); |
380af1b3 | 1084 | } |
1da177e4 LT |
1085 | |
1086 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 1087 | INIT_LIST_HEAD(&tasks); |
1a82e9e1 MS |
1088 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1089 | unlink_queue(sma, q); | |
1090 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1091 | } | |
1092 | ||
1093 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1094 | unlink_queue(sma, q); |
0a2b9d4c | 1095 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 1096 | } |
9f1bc2c9 RR |
1097 | for (i = 0; i < sma->sem_nsems; i++) { |
1098 | struct sem *sem = sma->sem_base + i; | |
1a82e9e1 MS |
1099 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1100 | unlink_queue(sma, q); | |
1101 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1102 | } | |
1103 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 RR |
1104 | unlink_queue(sma, q); |
1105 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1106 | } | |
1107 | } | |
1da177e4 | 1108 | |
7ca7e564 ND |
1109 | /* Remove the semaphore set from the IDR */ |
1110 | sem_rmid(ns, sma); | |
6062a8dc | 1111 | sem_unlock(sma, -1); |
6d49dab8 | 1112 | rcu_read_unlock(); |
1da177e4 | 1113 | |
0a2b9d4c | 1114 | wake_up_sem_queue_do(&tasks); |
e3893534 | 1115 | ns->used_sems -= sma->sem_nsems; |
53dad6d3 | 1116 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1117 | } |
1118 | ||
1119 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1120 | { | |
239521f3 | 1121 | switch (version) { |
1da177e4 LT |
1122 | case IPC_64: |
1123 | return copy_to_user(buf, in, sizeof(*in)); | |
1124 | case IPC_OLD: | |
1125 | { | |
1126 | struct semid_ds out; | |
1127 | ||
982f7c2b DR |
1128 | memset(&out, 0, sizeof(out)); |
1129 | ||
1da177e4 LT |
1130 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1131 | ||
1132 | out.sem_otime = in->sem_otime; | |
1133 | out.sem_ctime = in->sem_ctime; | |
1134 | out.sem_nsems = in->sem_nsems; | |
1135 | ||
1136 | return copy_to_user(buf, &out, sizeof(out)); | |
1137 | } | |
1138 | default: | |
1139 | return -EINVAL; | |
1140 | } | |
1141 | } | |
1142 | ||
d12e1e50 MS |
1143 | static time_t get_semotime(struct sem_array *sma) |
1144 | { | |
1145 | int i; | |
1146 | time_t res; | |
1147 | ||
1148 | res = sma->sem_base[0].sem_otime; | |
1149 | for (i = 1; i < sma->sem_nsems; i++) { | |
1150 | time_t to = sma->sem_base[i].sem_otime; | |
1151 | ||
1152 | if (to > res) | |
1153 | res = to; | |
1154 | } | |
1155 | return res; | |
1156 | } | |
1157 | ||
4b9fcb0e | 1158 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1159 | int cmd, int version, void __user *p) |
1da177e4 | 1160 | { |
e5cc9c7b | 1161 | int err; |
1da177e4 LT |
1162 | struct sem_array *sma; |
1163 | ||
239521f3 | 1164 | switch (cmd) { |
1da177e4 LT |
1165 | case IPC_INFO: |
1166 | case SEM_INFO: | |
1167 | { | |
1168 | struct seminfo seminfo; | |
1169 | int max_id; | |
1170 | ||
1171 | err = security_sem_semctl(NULL, cmd); | |
1172 | if (err) | |
1173 | return err; | |
46c0a8ca | 1174 | |
239521f3 | 1175 | memset(&seminfo, 0, sizeof(seminfo)); |
e3893534 KK |
1176 | seminfo.semmni = ns->sc_semmni; |
1177 | seminfo.semmns = ns->sc_semmns; | |
1178 | seminfo.semmsl = ns->sc_semmsl; | |
1179 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1180 | seminfo.semvmx = SEMVMX; |
1181 | seminfo.semmnu = SEMMNU; | |
1182 | seminfo.semmap = SEMMAP; | |
1183 | seminfo.semume = SEMUME; | |
d9a605e4 | 1184 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1185 | if (cmd == SEM_INFO) { |
e3893534 KK |
1186 | seminfo.semusz = sem_ids(ns).in_use; |
1187 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1188 | } else { |
1189 | seminfo.semusz = SEMUSZ; | |
1190 | seminfo.semaem = SEMAEM; | |
1191 | } | |
7ca7e564 | 1192 | max_id = ipc_get_maxid(&sem_ids(ns)); |
d9a605e4 | 1193 | up_read(&sem_ids(ns).rwsem); |
46c0a8ca | 1194 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 | 1195 | return -EFAULT; |
239521f3 | 1196 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 1197 | } |
4b9fcb0e | 1198 | case IPC_STAT: |
1da177e4 LT |
1199 | case SEM_STAT: |
1200 | { | |
1201 | struct semid64_ds tbuf; | |
16df3674 DB |
1202 | int id = 0; |
1203 | ||
1204 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1205 | |
941b0304 | 1206 | rcu_read_lock(); |
4b9fcb0e | 1207 | if (cmd == SEM_STAT) { |
16df3674 DB |
1208 | sma = sem_obtain_object(ns, semid); |
1209 | if (IS_ERR(sma)) { | |
1210 | err = PTR_ERR(sma); | |
1211 | goto out_unlock; | |
1212 | } | |
4b9fcb0e PP |
1213 | id = sma->sem_perm.id; |
1214 | } else { | |
16df3674 DB |
1215 | sma = sem_obtain_object_check(ns, semid); |
1216 | if (IS_ERR(sma)) { | |
1217 | err = PTR_ERR(sma); | |
1218 | goto out_unlock; | |
1219 | } | |
4b9fcb0e | 1220 | } |
1da177e4 LT |
1221 | |
1222 | err = -EACCES; | |
b0e77598 | 1223 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1224 | goto out_unlock; |
1225 | ||
1226 | err = security_sem_semctl(sma, cmd); | |
1227 | if (err) | |
1228 | goto out_unlock; | |
1229 | ||
1da177e4 | 1230 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
d12e1e50 MS |
1231 | tbuf.sem_otime = get_semotime(sma); |
1232 | tbuf.sem_ctime = sma->sem_ctime; | |
1233 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1234 | rcu_read_unlock(); |
e1fd1f49 | 1235 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1236 | return -EFAULT; |
1237 | return id; | |
1238 | } | |
1239 | default: | |
1240 | return -EINVAL; | |
1241 | } | |
1da177e4 | 1242 | out_unlock: |
16df3674 | 1243 | rcu_read_unlock(); |
1da177e4 LT |
1244 | return err; |
1245 | } | |
1246 | ||
e1fd1f49 AV |
1247 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1248 | unsigned long arg) | |
1249 | { | |
1250 | struct sem_undo *un; | |
1251 | struct sem_array *sma; | |
239521f3 | 1252 | struct sem *curr; |
e1fd1f49 | 1253 | int err; |
e1fd1f49 AV |
1254 | struct list_head tasks; |
1255 | int val; | |
1256 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1257 | /* big-endian 64bit */ | |
1258 | val = arg >> 32; | |
1259 | #else | |
1260 | /* 32bit or little-endian 64bit */ | |
1261 | val = arg; | |
1262 | #endif | |
1263 | ||
6062a8dc RR |
1264 | if (val > SEMVMX || val < 0) |
1265 | return -ERANGE; | |
e1fd1f49 AV |
1266 | |
1267 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1268 | |
6062a8dc RR |
1269 | rcu_read_lock(); |
1270 | sma = sem_obtain_object_check(ns, semid); | |
1271 | if (IS_ERR(sma)) { | |
1272 | rcu_read_unlock(); | |
1273 | return PTR_ERR(sma); | |
1274 | } | |
1275 | ||
1276 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1277 | rcu_read_unlock(); | |
1278 | return -EINVAL; | |
1279 | } | |
1280 | ||
1281 | ||
1282 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1283 | rcu_read_unlock(); | |
1284 | return -EACCES; | |
1285 | } | |
e1fd1f49 AV |
1286 | |
1287 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1288 | if (err) { |
1289 | rcu_read_unlock(); | |
1290 | return -EACCES; | |
1291 | } | |
e1fd1f49 | 1292 | |
6062a8dc | 1293 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1294 | |
0f3d2b01 | 1295 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1296 | sem_unlock(sma, -1); |
1297 | rcu_read_unlock(); | |
1298 | return -EIDRM; | |
1299 | } | |
1300 | ||
e1fd1f49 AV |
1301 | curr = &sma->sem_base[semnum]; |
1302 | ||
cf9d5d78 | 1303 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1304 | list_for_each_entry(un, &sma->list_id, list_id) |
1305 | un->semadj[semnum] = 0; | |
1306 | ||
1307 | curr->semval = val; | |
1308 | curr->sempid = task_tgid_vnr(current); | |
1309 | sma->sem_ctime = get_seconds(); | |
1310 | /* maybe some queued-up processes were waiting for this */ | |
1311 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1312 | sem_unlock(sma, -1); |
6d49dab8 | 1313 | rcu_read_unlock(); |
e1fd1f49 | 1314 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1315 | return 0; |
e1fd1f49 AV |
1316 | } |
1317 | ||
e3893534 | 1318 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1319 | int cmd, void __user *p) |
1da177e4 LT |
1320 | { |
1321 | struct sem_array *sma; | |
239521f3 | 1322 | struct sem *curr; |
16df3674 | 1323 | int err, nsems; |
1da177e4 | 1324 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1325 | ushort *sem_io = fast_sem_io; |
0a2b9d4c | 1326 | struct list_head tasks; |
1da177e4 | 1327 | |
16df3674 DB |
1328 | INIT_LIST_HEAD(&tasks); |
1329 | ||
1330 | rcu_read_lock(); | |
1331 | sma = sem_obtain_object_check(ns, semid); | |
1332 | if (IS_ERR(sma)) { | |
1333 | rcu_read_unlock(); | |
023a5355 | 1334 | return PTR_ERR(sma); |
16df3674 | 1335 | } |
1da177e4 LT |
1336 | |
1337 | nsems = sma->sem_nsems; | |
1338 | ||
1da177e4 | 1339 | err = -EACCES; |
c728b9c8 LT |
1340 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1341 | goto out_rcu_wakeup; | |
1da177e4 LT |
1342 | |
1343 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1344 | if (err) |
1345 | goto out_rcu_wakeup; | |
1da177e4 LT |
1346 | |
1347 | err = -EACCES; | |
1348 | switch (cmd) { | |
1349 | case GETALL: | |
1350 | { | |
e1fd1f49 | 1351 | ushort __user *array = p; |
1da177e4 LT |
1352 | int i; |
1353 | ||
ce857229 | 1354 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1355 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1356 | err = -EIDRM; |
1357 | goto out_unlock; | |
1358 | } | |
239521f3 | 1359 | if (nsems > SEMMSL_FAST) { |
ce857229 | 1360 | if (!ipc_rcu_getref(sma)) { |
ce857229 | 1361 | err = -EIDRM; |
6e224f94 | 1362 | goto out_unlock; |
ce857229 AV |
1363 | } |
1364 | sem_unlock(sma, -1); | |
6d49dab8 | 1365 | rcu_read_unlock(); |
1da177e4 | 1366 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1367 | if (sem_io == NULL) { |
53dad6d3 | 1368 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1369 | return -ENOMEM; |
1370 | } | |
1371 | ||
4091fd94 | 1372 | rcu_read_lock(); |
6ff37972 | 1373 | sem_lock_and_putref(sma); |
0f3d2b01 | 1374 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1375 | err = -EIDRM; |
6e224f94 | 1376 | goto out_unlock; |
1da177e4 | 1377 | } |
ce857229 | 1378 | } |
1da177e4 LT |
1379 | for (i = 0; i < sma->sem_nsems; i++) |
1380 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1381 | sem_unlock(sma, -1); |
6d49dab8 | 1382 | rcu_read_unlock(); |
1da177e4 | 1383 | err = 0; |
239521f3 | 1384 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1385 | err = -EFAULT; |
1386 | goto out_free; | |
1387 | } | |
1388 | case SETALL: | |
1389 | { | |
1390 | int i; | |
1391 | struct sem_undo *un; | |
1392 | ||
6062a8dc | 1393 | if (!ipc_rcu_getref(sma)) { |
6e224f94 MS |
1394 | err = -EIDRM; |
1395 | goto out_rcu_wakeup; | |
6062a8dc | 1396 | } |
16df3674 | 1397 | rcu_read_unlock(); |
1da177e4 | 1398 | |
239521f3 | 1399 | if (nsems > SEMMSL_FAST) { |
1da177e4 | 1400 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1401 | if (sem_io == NULL) { |
53dad6d3 | 1402 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1403 | return -ENOMEM; |
1404 | } | |
1405 | } | |
1406 | ||
239521f3 | 1407 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
53dad6d3 | 1408 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1409 | err = -EFAULT; |
1410 | goto out_free; | |
1411 | } | |
1412 | ||
1413 | for (i = 0; i < nsems; i++) { | |
1414 | if (sem_io[i] > SEMVMX) { | |
53dad6d3 | 1415 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1416 | err = -ERANGE; |
1417 | goto out_free; | |
1418 | } | |
1419 | } | |
4091fd94 | 1420 | rcu_read_lock(); |
6ff37972 | 1421 | sem_lock_and_putref(sma); |
0f3d2b01 | 1422 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1423 | err = -EIDRM; |
6e224f94 | 1424 | goto out_unlock; |
1da177e4 LT |
1425 | } |
1426 | ||
1427 | for (i = 0; i < nsems; i++) | |
1428 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 | 1429 | |
cf9d5d78 | 1430 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1431 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1432 | for (i = 0; i < nsems; i++) |
1433 | un->semadj[i] = 0; | |
4daa28f6 | 1434 | } |
1da177e4 LT |
1435 | sma->sem_ctime = get_seconds(); |
1436 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1437 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1438 | err = 0; |
1439 | goto out_unlock; | |
1440 | } | |
e1fd1f49 | 1441 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1442 | } |
1443 | err = -EINVAL; | |
c728b9c8 LT |
1444 | if (semnum < 0 || semnum >= nsems) |
1445 | goto out_rcu_wakeup; | |
1da177e4 | 1446 | |
6062a8dc | 1447 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1448 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1449 | err = -EIDRM; |
1450 | goto out_unlock; | |
1451 | } | |
1da177e4 LT |
1452 | curr = &sma->sem_base[semnum]; |
1453 | ||
1454 | switch (cmd) { | |
1455 | case GETVAL: | |
1456 | err = curr->semval; | |
1457 | goto out_unlock; | |
1458 | case GETPID: | |
1459 | err = curr->sempid; | |
1460 | goto out_unlock; | |
1461 | case GETNCNT: | |
239521f3 | 1462 | err = count_semncnt(sma, semnum); |
1da177e4 LT |
1463 | goto out_unlock; |
1464 | case GETZCNT: | |
239521f3 | 1465 | err = count_semzcnt(sma, semnum); |
1da177e4 | 1466 | goto out_unlock; |
1da177e4 | 1467 | } |
16df3674 | 1468 | |
1da177e4 | 1469 | out_unlock: |
6062a8dc | 1470 | sem_unlock(sma, -1); |
c728b9c8 | 1471 | out_rcu_wakeup: |
6d49dab8 | 1472 | rcu_read_unlock(); |
0a2b9d4c | 1473 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 1474 | out_free: |
239521f3 | 1475 | if (sem_io != fast_sem_io) |
1da177e4 LT |
1476 | ipc_free(sem_io, sizeof(ushort)*nsems); |
1477 | return err; | |
1478 | } | |
1479 | ||
016d7132 PP |
1480 | static inline unsigned long |
1481 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1482 | { |
239521f3 | 1483 | switch (version) { |
1da177e4 | 1484 | case IPC_64: |
016d7132 | 1485 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1486 | return -EFAULT; |
1da177e4 | 1487 | return 0; |
1da177e4 LT |
1488 | case IPC_OLD: |
1489 | { | |
1490 | struct semid_ds tbuf_old; | |
1491 | ||
239521f3 | 1492 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1493 | return -EFAULT; |
1494 | ||
016d7132 PP |
1495 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1496 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1497 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1498 | |
1499 | return 0; | |
1500 | } | |
1501 | default: | |
1502 | return -EINVAL; | |
1503 | } | |
1504 | } | |
1505 | ||
522bb2a2 | 1506 | /* |
d9a605e4 | 1507 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1508 | * to be held in write mode. |
d9a605e4 | 1509 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1510 | */ |
21a4826a | 1511 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1512 | int cmd, int version, void __user *p) |
1da177e4 LT |
1513 | { |
1514 | struct sem_array *sma; | |
1515 | int err; | |
016d7132 | 1516 | struct semid64_ds semid64; |
1da177e4 LT |
1517 | struct kern_ipc_perm *ipcp; |
1518 | ||
239521f3 | 1519 | if (cmd == IPC_SET) { |
e1fd1f49 | 1520 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1521 | return -EFAULT; |
1da177e4 | 1522 | } |
073115d6 | 1523 | |
d9a605e4 | 1524 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1525 | rcu_read_lock(); |
1526 | ||
16df3674 DB |
1527 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1528 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1529 | if (IS_ERR(ipcp)) { |
1530 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1531 | goto out_unlock1; |
1532 | } | |
073115d6 | 1533 | |
a5f75e7f | 1534 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1535 | |
1536 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1537 | if (err) |
1538 | goto out_unlock1; | |
1da177e4 | 1539 | |
7b4cc5d8 | 1540 | switch (cmd) { |
1da177e4 | 1541 | case IPC_RMID: |
6062a8dc | 1542 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1543 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1544 | freeary(ns, ipcp); |
522bb2a2 | 1545 | goto out_up; |
1da177e4 | 1546 | case IPC_SET: |
6062a8dc | 1547 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1548 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1549 | if (err) | |
7b4cc5d8 | 1550 | goto out_unlock0; |
1da177e4 | 1551 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1552 | break; |
1553 | default: | |
1da177e4 | 1554 | err = -EINVAL; |
7b4cc5d8 | 1555 | goto out_unlock1; |
1da177e4 | 1556 | } |
1da177e4 | 1557 | |
7b4cc5d8 | 1558 | out_unlock0: |
6062a8dc | 1559 | sem_unlock(sma, -1); |
7b4cc5d8 | 1560 | out_unlock1: |
6d49dab8 | 1561 | rcu_read_unlock(); |
522bb2a2 | 1562 | out_up: |
d9a605e4 | 1563 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1564 | return err; |
1565 | } | |
1566 | ||
e1fd1f49 | 1567 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1568 | { |
1da177e4 | 1569 | int version; |
e3893534 | 1570 | struct ipc_namespace *ns; |
e1fd1f49 | 1571 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1572 | |
1573 | if (semid < 0) | |
1574 | return -EINVAL; | |
1575 | ||
1576 | version = ipc_parse_version(&cmd); | |
e3893534 | 1577 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1578 | |
239521f3 | 1579 | switch (cmd) { |
1da177e4 LT |
1580 | case IPC_INFO: |
1581 | case SEM_INFO: | |
4b9fcb0e | 1582 | case IPC_STAT: |
1da177e4 | 1583 | case SEM_STAT: |
e1fd1f49 | 1584 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1585 | case GETALL: |
1586 | case GETVAL: | |
1587 | case GETPID: | |
1588 | case GETNCNT: | |
1589 | case GETZCNT: | |
1da177e4 | 1590 | case SETALL: |
e1fd1f49 AV |
1591 | return semctl_main(ns, semid, semnum, cmd, p); |
1592 | case SETVAL: | |
1593 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1594 | case IPC_RMID: |
1595 | case IPC_SET: | |
e1fd1f49 | 1596 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1597 | default: |
1598 | return -EINVAL; | |
1599 | } | |
1600 | } | |
1601 | ||
1da177e4 LT |
1602 | /* If the task doesn't already have a undo_list, then allocate one |
1603 | * here. We guarantee there is only one thread using this undo list, | |
1604 | * and current is THE ONE | |
1605 | * | |
1606 | * If this allocation and assignment succeeds, but later | |
1607 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1608 | * Just let it stay associated with the task, and it'll be freed later | |
1609 | * at exit time. | |
1610 | * | |
1611 | * This can block, so callers must hold no locks. | |
1612 | */ | |
1613 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1614 | { | |
1615 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1616 | |
1617 | undo_list = current->sysvsem.undo_list; | |
1618 | if (!undo_list) { | |
2453a306 | 1619 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1620 | if (undo_list == NULL) |
1621 | return -ENOMEM; | |
00a5dfdb | 1622 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1623 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1624 | INIT_LIST_HEAD(&undo_list->list_proc); |
1625 | ||
1da177e4 LT |
1626 | current->sysvsem.undo_list = undo_list; |
1627 | } | |
1628 | *undo_listp = undo_list; | |
1629 | return 0; | |
1630 | } | |
1631 | ||
bf17bb71 | 1632 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1633 | { |
bf17bb71 | 1634 | struct sem_undo *un; |
4daa28f6 | 1635 | |
bf17bb71 NP |
1636 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1637 | if (un->semid == semid) | |
1638 | return un; | |
1da177e4 | 1639 | } |
4daa28f6 | 1640 | return NULL; |
1da177e4 LT |
1641 | } |
1642 | ||
bf17bb71 NP |
1643 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1644 | { | |
1645 | struct sem_undo *un; | |
1646 | ||
239521f3 | 1647 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1648 | |
1649 | un = __lookup_undo(ulp, semid); | |
1650 | if (un) { | |
1651 | list_del_rcu(&un->list_proc); | |
1652 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1653 | } | |
1654 | return un; | |
1655 | } | |
1656 | ||
4daa28f6 | 1657 | /** |
8001c858 | 1658 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1659 | * @ns: namespace |
1660 | * @semid: semaphore array id | |
1661 | * | |
1662 | * The function looks up (and if not present creates) the undo structure. | |
1663 | * The size of the undo structure depends on the size of the semaphore | |
1664 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1665 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1666 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1667 | */ |
1668 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1669 | { |
1670 | struct sem_array *sma; | |
1671 | struct sem_undo_list *ulp; | |
1672 | struct sem_undo *un, *new; | |
6062a8dc | 1673 | int nsems, error; |
1da177e4 LT |
1674 | |
1675 | error = get_undo_list(&ulp); | |
1676 | if (error) | |
1677 | return ERR_PTR(error); | |
1678 | ||
380af1b3 | 1679 | rcu_read_lock(); |
c530c6ac | 1680 | spin_lock(&ulp->lock); |
1da177e4 | 1681 | un = lookup_undo(ulp, semid); |
c530c6ac | 1682 | spin_unlock(&ulp->lock); |
239521f3 | 1683 | if (likely(un != NULL)) |
1da177e4 LT |
1684 | goto out; |
1685 | ||
1686 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1687 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1688 | sma = sem_obtain_object_check(ns, semid); |
1689 | if (IS_ERR(sma)) { | |
1690 | rcu_read_unlock(); | |
4de85cd6 | 1691 | return ERR_CAST(sma); |
16df3674 | 1692 | } |
023a5355 | 1693 | |
1da177e4 | 1694 | nsems = sma->sem_nsems; |
6062a8dc RR |
1695 | if (!ipc_rcu_getref(sma)) { |
1696 | rcu_read_unlock(); | |
1697 | un = ERR_PTR(-EIDRM); | |
1698 | goto out; | |
1699 | } | |
16df3674 | 1700 | rcu_read_unlock(); |
1da177e4 | 1701 | |
4daa28f6 | 1702 | /* step 2: allocate new undo structure */ |
4668edc3 | 1703 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1704 | if (!new) { |
53dad6d3 | 1705 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1706 | return ERR_PTR(-ENOMEM); |
1707 | } | |
1da177e4 | 1708 | |
380af1b3 | 1709 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1710 | rcu_read_lock(); |
6ff37972 | 1711 | sem_lock_and_putref(sma); |
0f3d2b01 | 1712 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1713 | sem_unlock(sma, -1); |
6d49dab8 | 1714 | rcu_read_unlock(); |
1da177e4 LT |
1715 | kfree(new); |
1716 | un = ERR_PTR(-EIDRM); | |
1717 | goto out; | |
1718 | } | |
380af1b3 MS |
1719 | spin_lock(&ulp->lock); |
1720 | ||
1721 | /* | |
1722 | * step 4: check for races: did someone else allocate the undo struct? | |
1723 | */ | |
1724 | un = lookup_undo(ulp, semid); | |
1725 | if (un) { | |
1726 | kfree(new); | |
1727 | goto success; | |
1728 | } | |
4daa28f6 MS |
1729 | /* step 5: initialize & link new undo structure */ |
1730 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1731 | new->ulp = ulp; |
4daa28f6 MS |
1732 | new->semid = semid; |
1733 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1734 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1735 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1736 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1737 | un = new; |
4daa28f6 | 1738 | |
380af1b3 | 1739 | success: |
c530c6ac | 1740 | spin_unlock(&ulp->lock); |
6062a8dc | 1741 | sem_unlock(sma, -1); |
1da177e4 LT |
1742 | out: |
1743 | return un; | |
1744 | } | |
1745 | ||
c61284e9 MS |
1746 | |
1747 | /** | |
8001c858 | 1748 | * get_queue_result - retrieve the result code from sem_queue |
c61284e9 MS |
1749 | * @q: Pointer to queue structure |
1750 | * | |
1751 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1752 | * q->status, then we must loop until the value is replaced with the final | |
1753 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1754 | * signal) and in parallel the task is woken up by another task because it got | |
1755 | * the requested semaphores. | |
1756 | * | |
1757 | * The function can be called with or without holding the semaphore spinlock. | |
1758 | */ | |
1759 | static int get_queue_result(struct sem_queue *q) | |
1760 | { | |
1761 | int error; | |
1762 | ||
1763 | error = q->status; | |
1764 | while (unlikely(error == IN_WAKEUP)) { | |
1765 | cpu_relax(); | |
1766 | error = q->status; | |
1767 | } | |
1768 | ||
1769 | return error; | |
1770 | } | |
1771 | ||
d5460c99 HC |
1772 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1773 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1774 | { |
1775 | int error = -EINVAL; | |
1776 | struct sem_array *sma; | |
1777 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1778 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1779 | struct sem_undo *un; |
6062a8dc | 1780 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1781 | struct sem_queue queue; |
1782 | unsigned long jiffies_left = 0; | |
e3893534 | 1783 | struct ipc_namespace *ns; |
0a2b9d4c | 1784 | struct list_head tasks; |
e3893534 KK |
1785 | |
1786 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1787 | |
1788 | if (nsops < 1 || semid < 0) | |
1789 | return -EINVAL; | |
e3893534 | 1790 | if (nsops > ns->sc_semopm) |
1da177e4 | 1791 | return -E2BIG; |
239521f3 MS |
1792 | if (nsops > SEMOPM_FAST) { |
1793 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1794 | if (sops == NULL) | |
1da177e4 LT |
1795 | return -ENOMEM; |
1796 | } | |
239521f3 MS |
1797 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1798 | error = -EFAULT; | |
1da177e4 LT |
1799 | goto out_free; |
1800 | } | |
1801 | if (timeout) { | |
1802 | struct timespec _timeout; | |
1803 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1804 | error = -EFAULT; | |
1805 | goto out_free; | |
1806 | } | |
1807 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1808 | _timeout.tv_nsec >= 1000000000L) { | |
1809 | error = -EINVAL; | |
1810 | goto out_free; | |
1811 | } | |
1812 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1813 | } | |
1814 | max = 0; | |
1815 | for (sop = sops; sop < sops + nsops; sop++) { | |
1816 | if (sop->sem_num >= max) | |
1817 | max = sop->sem_num; | |
1818 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1819 | undos = 1; |
1820 | if (sop->sem_op != 0) | |
1da177e4 LT |
1821 | alter = 1; |
1822 | } | |
1da177e4 | 1823 | |
6062a8dc RR |
1824 | INIT_LIST_HEAD(&tasks); |
1825 | ||
1da177e4 | 1826 | if (undos) { |
6062a8dc | 1827 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1828 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1829 | if (IS_ERR(un)) { |
1830 | error = PTR_ERR(un); | |
1831 | goto out_free; | |
1832 | } | |
6062a8dc | 1833 | } else { |
1da177e4 | 1834 | un = NULL; |
6062a8dc RR |
1835 | rcu_read_lock(); |
1836 | } | |
1da177e4 | 1837 | |
16df3674 | 1838 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1839 | if (IS_ERR(sma)) { |
6062a8dc | 1840 | rcu_read_unlock(); |
023a5355 | 1841 | error = PTR_ERR(sma); |
1da177e4 | 1842 | goto out_free; |
023a5355 ND |
1843 | } |
1844 | ||
16df3674 | 1845 | error = -EFBIG; |
c728b9c8 LT |
1846 | if (max >= sma->sem_nsems) |
1847 | goto out_rcu_wakeup; | |
16df3674 DB |
1848 | |
1849 | error = -EACCES; | |
c728b9c8 LT |
1850 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
1851 | goto out_rcu_wakeup; | |
16df3674 DB |
1852 | |
1853 | error = security_sem_semop(sma, sops, nsops, alter); | |
c728b9c8 LT |
1854 | if (error) |
1855 | goto out_rcu_wakeup; | |
16df3674 | 1856 | |
6e224f94 MS |
1857 | error = -EIDRM; |
1858 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
1859 | /* |
1860 | * We eventually might perform the following check in a lockless | |
1861 | * fashion, considering ipc_valid_object() locking constraints. | |
1862 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1863 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1864 | * check below. More details on the fine grained locking scheme | |
1865 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1866 | */ | |
1867 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 1868 | goto out_unlock_free; |
1da177e4 | 1869 | /* |
4daa28f6 | 1870 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1871 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1872 | * and now a new array with received the same id. Check and fail. |
25985edc | 1873 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1874 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1875 | */ |
6062a8dc RR |
1876 | if (un && un->semid == -1) |
1877 | goto out_unlock_free; | |
4daa28f6 | 1878 | |
758a6ba3 MS |
1879 | error = perform_atomic_semop(sma, sops, nsops, un, |
1880 | task_tgid_vnr(current)); | |
0e8c6656 MS |
1881 | if (error == 0) { |
1882 | /* If the operation was successful, then do | |
1883 | * the required updates. | |
1884 | */ | |
1885 | if (alter) | |
0a2b9d4c | 1886 | do_smart_update(sma, sops, nsops, 1, &tasks); |
0e8c6656 MS |
1887 | else |
1888 | set_semotime(sma, sops); | |
1da177e4 | 1889 | } |
0e8c6656 MS |
1890 | if (error <= 0) |
1891 | goto out_unlock_free; | |
1da177e4 LT |
1892 | |
1893 | /* We need to sleep on this operation, so we put the current | |
1894 | * task into the pending queue and go to sleep. | |
1895 | */ | |
46c0a8ca | 1896 | |
1da177e4 LT |
1897 | queue.sops = sops; |
1898 | queue.nsops = nsops; | |
1899 | queue.undo = un; | |
b488893a | 1900 | queue.pid = task_tgid_vnr(current); |
1da177e4 | 1901 | queue.alter = alter; |
1da177e4 | 1902 | |
b97e820f MS |
1903 | if (nsops == 1) { |
1904 | struct sem *curr; | |
1905 | curr = &sma->sem_base[sops->sem_num]; | |
1906 | ||
f269f40a MS |
1907 | if (alter) { |
1908 | if (sma->complex_count) { | |
1909 | list_add_tail(&queue.list, | |
1910 | &sma->pending_alter); | |
1911 | } else { | |
1912 | ||
1913 | list_add_tail(&queue.list, | |
1914 | &curr->pending_alter); | |
1915 | } | |
1916 | } else { | |
1a82e9e1 | 1917 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 1918 | } |
b97e820f | 1919 | } else { |
f269f40a MS |
1920 | if (!sma->complex_count) |
1921 | merge_queues(sma); | |
1922 | ||
9f1bc2c9 | 1923 | if (alter) |
1a82e9e1 | 1924 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1925 | else |
1a82e9e1 MS |
1926 | list_add_tail(&queue.list, &sma->pending_const); |
1927 | ||
b97e820f MS |
1928 | sma->complex_count++; |
1929 | } | |
1930 | ||
1da177e4 LT |
1931 | queue.status = -EINTR; |
1932 | queue.sleeper = current; | |
0b0577f6 MS |
1933 | |
1934 | sleep_again: | |
1da177e4 | 1935 | current->state = TASK_INTERRUPTIBLE; |
6062a8dc | 1936 | sem_unlock(sma, locknum); |
6d49dab8 | 1937 | rcu_read_unlock(); |
1da177e4 LT |
1938 | |
1939 | if (timeout) | |
1940 | jiffies_left = schedule_timeout(jiffies_left); | |
1941 | else | |
1942 | schedule(); | |
1943 | ||
c61284e9 | 1944 | error = get_queue_result(&queue); |
1da177e4 LT |
1945 | |
1946 | if (error != -EINTR) { | |
1947 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1948 | * resources. |
1949 | * Perform a smp_mb(): User space could assume that semop() | |
1950 | * is a memory barrier: Without the mb(), the cpu could | |
1951 | * speculatively read in user space stale data that was | |
1952 | * overwritten by the previous owner of the semaphore. | |
1953 | */ | |
1954 | smp_mb(); | |
1955 | ||
1da177e4 LT |
1956 | goto out_free; |
1957 | } | |
1958 | ||
321310ce | 1959 | rcu_read_lock(); |
6062a8dc | 1960 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1961 | |
1962 | /* | |
1963 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
1964 | */ | |
1965 | error = get_queue_result(&queue); | |
1966 | ||
1967 | /* | |
1968 | * Array removed? If yes, leave without sem_unlock(). | |
1969 | */ | |
023a5355 | 1970 | if (IS_ERR(sma)) { |
321310ce | 1971 | rcu_read_unlock(); |
1da177e4 LT |
1972 | goto out_free; |
1973 | } | |
1974 | ||
c61284e9 | 1975 | |
1da177e4 | 1976 | /* |
d694ad62 MS |
1977 | * If queue.status != -EINTR we are woken up by another process. |
1978 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 1979 | */ |
3ab08fe2 | 1980 | if (error != -EINTR) |
1da177e4 | 1981 | goto out_unlock_free; |
1da177e4 LT |
1982 | |
1983 | /* | |
1984 | * If an interrupt occurred we have to clean up the queue | |
1985 | */ | |
1986 | if (timeout && jiffies_left == 0) | |
1987 | error = -EAGAIN; | |
0b0577f6 MS |
1988 | |
1989 | /* | |
1990 | * If the wakeup was spurious, just retry | |
1991 | */ | |
1992 | if (error == -EINTR && !signal_pending(current)) | |
1993 | goto sleep_again; | |
1994 | ||
b97e820f | 1995 | unlink_queue(sma, &queue); |
1da177e4 LT |
1996 | |
1997 | out_unlock_free: | |
6062a8dc | 1998 | sem_unlock(sma, locknum); |
c728b9c8 | 1999 | out_rcu_wakeup: |
6d49dab8 | 2000 | rcu_read_unlock(); |
0a2b9d4c | 2001 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 2002 | out_free: |
239521f3 | 2003 | if (sops != fast_sops) |
1da177e4 LT |
2004 | kfree(sops); |
2005 | return error; | |
2006 | } | |
2007 | ||
d5460c99 HC |
2008 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2009 | unsigned, nsops) | |
1da177e4 LT |
2010 | { |
2011 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2012 | } | |
2013 | ||
2014 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2015 | * parent and child tasks. | |
1da177e4 LT |
2016 | */ |
2017 | ||
2018 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2019 | { | |
2020 | struct sem_undo_list *undo_list; | |
2021 | int error; | |
2022 | ||
2023 | if (clone_flags & CLONE_SYSVSEM) { | |
2024 | error = get_undo_list(&undo_list); | |
2025 | if (error) | |
2026 | return error; | |
1da177e4 LT |
2027 | atomic_inc(&undo_list->refcnt); |
2028 | tsk->sysvsem.undo_list = undo_list; | |
46c0a8ca | 2029 | } else |
1da177e4 LT |
2030 | tsk->sysvsem.undo_list = NULL; |
2031 | ||
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | /* | |
2036 | * add semadj values to semaphores, free undo structures. | |
2037 | * undo structures are not freed when semaphore arrays are destroyed | |
2038 | * so some of them may be out of date. | |
2039 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2040 | * set of adjustments that needs to be done should be done in an atomic | |
2041 | * manner or not. That is, if we are attempting to decrement the semval | |
2042 | * should we queue up and wait until we can do so legally? | |
2043 | * The original implementation attempted to do this (queue and wait). | |
2044 | * The current implementation does not do so. The POSIX standard | |
2045 | * and SVID should be consulted to determine what behavior is mandated. | |
2046 | */ | |
2047 | void exit_sem(struct task_struct *tsk) | |
2048 | { | |
4daa28f6 | 2049 | struct sem_undo_list *ulp; |
1da177e4 | 2050 | |
4daa28f6 MS |
2051 | ulp = tsk->sysvsem.undo_list; |
2052 | if (!ulp) | |
1da177e4 | 2053 | return; |
9edff4ab | 2054 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2055 | |
4daa28f6 | 2056 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2057 | return; |
2058 | ||
380af1b3 | 2059 | for (;;) { |
1da177e4 | 2060 | struct sem_array *sma; |
380af1b3 | 2061 | struct sem_undo *un; |
0a2b9d4c | 2062 | struct list_head tasks; |
6062a8dc | 2063 | int semid, i; |
4daa28f6 | 2064 | |
380af1b3 | 2065 | rcu_read_lock(); |
05725f7e JP |
2066 | un = list_entry_rcu(ulp->list_proc.next, |
2067 | struct sem_undo, list_proc); | |
380af1b3 MS |
2068 | if (&un->list_proc == &ulp->list_proc) |
2069 | semid = -1; | |
2070 | else | |
2071 | semid = un->semid; | |
4daa28f6 | 2072 | |
6062a8dc RR |
2073 | if (semid == -1) { |
2074 | rcu_read_unlock(); | |
380af1b3 | 2075 | break; |
6062a8dc | 2076 | } |
1da177e4 | 2077 | |
6062a8dc | 2078 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid); |
380af1b3 | 2079 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2080 | if (IS_ERR(sma)) { |
2081 | rcu_read_unlock(); | |
380af1b3 | 2082 | continue; |
6062a8dc | 2083 | } |
1da177e4 | 2084 | |
6062a8dc | 2085 | sem_lock(sma, NULL, -1); |
6e224f94 | 2086 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2087 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2088 | sem_unlock(sma, -1); |
2089 | rcu_read_unlock(); | |
2090 | continue; | |
2091 | } | |
bf17bb71 | 2092 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2093 | if (un == NULL) { |
2094 | /* exit_sem raced with IPC_RMID+semget() that created | |
2095 | * exactly the same semid. Nothing to do. | |
2096 | */ | |
6062a8dc | 2097 | sem_unlock(sma, -1); |
6d49dab8 | 2098 | rcu_read_unlock(); |
380af1b3 MS |
2099 | continue; |
2100 | } | |
2101 | ||
2102 | /* remove un from the linked lists */ | |
cf9d5d78 | 2103 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2104 | list_del(&un->list_id); |
2105 | ||
380af1b3 MS |
2106 | spin_lock(&ulp->lock); |
2107 | list_del_rcu(&un->list_proc); | |
2108 | spin_unlock(&ulp->lock); | |
2109 | ||
4daa28f6 MS |
2110 | /* perform adjustments registered in un */ |
2111 | for (i = 0; i < sma->sem_nsems; i++) { | |
239521f3 | 2112 | struct sem *semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
2113 | if (un->semadj[i]) { |
2114 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2115 | /* |
2116 | * Range checks of the new semaphore value, | |
2117 | * not defined by sus: | |
2118 | * - Some unices ignore the undo entirely | |
2119 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2120 | * - some cap the value (e.g. FreeBSD caps | |
2121 | * at 0, but doesn't enforce SEMVMX) | |
2122 | * | |
2123 | * Linux caps the semaphore value, both at 0 | |
2124 | * and at SEMVMX. | |
2125 | * | |
239521f3 | 2126 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2127 | */ |
5f921ae9 IM |
2128 | if (semaphore->semval < 0) |
2129 | semaphore->semval = 0; | |
2130 | if (semaphore->semval > SEMVMX) | |
2131 | semaphore->semval = SEMVMX; | |
b488893a | 2132 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2133 | } |
2134 | } | |
1da177e4 | 2135 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
2136 | INIT_LIST_HEAD(&tasks); |
2137 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 2138 | sem_unlock(sma, -1); |
6d49dab8 | 2139 | rcu_read_unlock(); |
0a2b9d4c | 2140 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 2141 | |
693a8b6e | 2142 | kfree_rcu(un, rcu); |
1da177e4 | 2143 | } |
4daa28f6 | 2144 | kfree(ulp); |
1da177e4 LT |
2145 | } |
2146 | ||
2147 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2148 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2149 | { |
1efdb69b | 2150 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2151 | struct sem_array *sma = it; |
d12e1e50 MS |
2152 | time_t sem_otime; |
2153 | ||
d8c63376 MS |
2154 | /* |
2155 | * The proc interface isn't aware of sem_lock(), it calls | |
2156 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
2157 | * In order to stay compatible with sem_lock(), we must wait until | |
2158 | * all simple semop() calls have left their critical regions. | |
2159 | */ | |
2160 | sem_wait_array(sma); | |
2161 | ||
d12e1e50 | 2162 | sem_otime = get_semotime(sma); |
19b4946c MW |
2163 | |
2164 | return seq_printf(s, | |
b97e820f | 2165 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 2166 | sma->sem_perm.key, |
7ca7e564 | 2167 | sma->sem_perm.id, |
19b4946c MW |
2168 | sma->sem_perm.mode, |
2169 | sma->sem_nsems, | |
1efdb69b EB |
2170 | from_kuid_munged(user_ns, sma->sem_perm.uid), |
2171 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2172 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2173 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
d12e1e50 | 2174 | sem_otime, |
19b4946c | 2175 | sma->sem_ctime); |
1da177e4 LT |
2176 | } |
2177 | #endif |