]>
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. | |
2f2ed41d | 50 | * - semncnt and semzcnt are calculated on demand in count_semcnt() |
c5cf6359 MS |
51 | * - the task that performs a successful semop() scans the list of all |
52 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
53 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
54 | * (see update_queue()) | |
55 | * - To improve the scalability, the actual wake-up calls are performed after | |
56 | * dropping all locks. (see wake_up_sem_queue_prepare(), | |
57 | * wake_up_sem_queue_do()) | |
58 | * - All work is done by the waker, the woken up task does not have to do | |
59 | * anything - not even acquiring a lock or dropping a refcount. | |
60 | * - A woken up task may not even touch the semaphore array anymore, it may | |
61 | * have been destroyed already by a semctl(RMID). | |
62 | * - The synchronizations between wake-ups due to a timeout/signal and a | |
63 | * wake-up due to a completed semaphore operation is achieved by using an | |
64 | * intermediate state (IN_WAKEUP). | |
65 | * - UNDO values are stored in an array (one per process and per | |
66 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
67 | * modes for the UNDO variables are supported (per process, per thread) | |
68 | * (see copy_semundo, CLONE_SYSVSEM) | |
69 | * - There are two lists of the pending operations: a per-array list | |
70 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
71 | * ordering without always scanning all pending operations. | |
72 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
73 | */ |
74 | ||
1da177e4 LT |
75 | #include <linux/slab.h> |
76 | #include <linux/spinlock.h> | |
77 | #include <linux/init.h> | |
78 | #include <linux/proc_fs.h> | |
79 | #include <linux/time.h> | |
1da177e4 LT |
80 | #include <linux/security.h> |
81 | #include <linux/syscalls.h> | |
82 | #include <linux/audit.h> | |
c59ede7b | 83 | #include <linux/capability.h> |
19b4946c | 84 | #include <linux/seq_file.h> |
3e148c79 | 85 | #include <linux/rwsem.h> |
e3893534 | 86 | #include <linux/nsproxy.h> |
ae5e1b22 | 87 | #include <linux/ipc_namespace.h> |
5f921ae9 | 88 | |
7153e402 | 89 | #include <linux/uaccess.h> |
1da177e4 LT |
90 | #include "util.h" |
91 | ||
e57940d7 MS |
92 | /* One semaphore structure for each semaphore in the system. */ |
93 | struct sem { | |
94 | int semval; /* current value */ | |
95 | int sempid; /* pid of last operation */ | |
6062a8dc | 96 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
1a82e9e1 MS |
97 | struct list_head pending_alter; /* pending single-sop operations */ |
98 | /* that alter the semaphore */ | |
99 | struct list_head pending_const; /* pending single-sop operations */ | |
100 | /* that do not alter the semaphore*/ | |
d12e1e50 | 101 | time_t sem_otime; /* candidate for sem_otime */ |
f5c936c0 | 102 | } ____cacheline_aligned_in_smp; |
e57940d7 MS |
103 | |
104 | /* One queue for each sleeping process in the system. */ | |
105 | struct sem_queue { | |
e57940d7 MS |
106 | struct list_head list; /* queue of pending operations */ |
107 | struct task_struct *sleeper; /* this process */ | |
108 | struct sem_undo *undo; /* undo structure */ | |
109 | int pid; /* process id of requesting process */ | |
110 | int status; /* completion status of operation */ | |
111 | struct sembuf *sops; /* array of pending operations */ | |
ed247b7c | 112 | struct sembuf *blocking; /* the operation that blocked */ |
e57940d7 MS |
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 | ||
1da177e4 | 510 | sma->sem_base = (struct sem *) &sma[1]; |
b97e820f | 511 | |
6062a8dc | 512 | for (i = 0; i < nsems; i++) { |
1a82e9e1 MS |
513 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
514 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
515 | spin_lock_init(&sma->sem_base[i].lock); |
516 | } | |
b97e820f MS |
517 | |
518 | sma->complex_count = 0; | |
1a82e9e1 MS |
519 | INIT_LIST_HEAD(&sma->pending_alter); |
520 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 521 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
522 | sma->sem_nsems = nsems; |
523 | sma->sem_ctime = get_seconds(); | |
e8577d1f MS |
524 | |
525 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); | |
526 | if (id < 0) { | |
527 | ipc_rcu_putref(sma, sem_rcu_free); | |
528 | return id; | |
529 | } | |
530 | ns->used_sems += nsems; | |
531 | ||
6062a8dc | 532 | sem_unlock(sma, -1); |
6d49dab8 | 533 | rcu_read_unlock(); |
1da177e4 | 534 | |
7ca7e564 | 535 | return sma->sem_perm.id; |
1da177e4 LT |
536 | } |
537 | ||
7748dbfa | 538 | |
f4566f04 | 539 | /* |
d9a605e4 | 540 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 541 | */ |
03f02c76 | 542 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 543 | { |
03f02c76 ND |
544 | struct sem_array *sma; |
545 | ||
546 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
547 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
548 | } |
549 | ||
f4566f04 | 550 | /* |
d9a605e4 | 551 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 552 | */ |
03f02c76 ND |
553 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
554 | struct ipc_params *params) | |
7748dbfa | 555 | { |
03f02c76 ND |
556 | struct sem_array *sma; |
557 | ||
558 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
559 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
560 | return -EINVAL; |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
d5460c99 | 565 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 566 | { |
e3893534 | 567 | struct ipc_namespace *ns; |
eb66ec44 MK |
568 | static const struct ipc_ops sem_ops = { |
569 | .getnew = newary, | |
570 | .associate = sem_security, | |
571 | .more_checks = sem_more_checks, | |
572 | }; | |
7748dbfa | 573 | struct ipc_params sem_params; |
e3893534 KK |
574 | |
575 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 576 | |
e3893534 | 577 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 578 | return -EINVAL; |
7ca7e564 | 579 | |
7748dbfa ND |
580 | sem_params.key = key; |
581 | sem_params.flg = semflg; | |
582 | sem_params.u.nsems = nsems; | |
1da177e4 | 583 | |
7748dbfa | 584 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
585 | } |
586 | ||
78f5009c PM |
587 | /** |
588 | * perform_atomic_semop - Perform (if possible) a semaphore operation | |
758a6ba3 | 589 | * @sma: semaphore array |
d198cd6d | 590 | * @q: struct sem_queue that describes the operation |
758a6ba3 MS |
591 | * |
592 | * Returns 0 if the operation was possible. | |
593 | * Returns 1 if the operation is impossible, the caller must sleep. | |
594 | * Negative values are error codes. | |
1da177e4 | 595 | */ |
d198cd6d | 596 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) |
1da177e4 | 597 | { |
d198cd6d | 598 | int result, sem_op, nsops, pid; |
1da177e4 | 599 | struct sembuf *sop; |
239521f3 | 600 | struct sem *curr; |
d198cd6d MS |
601 | struct sembuf *sops; |
602 | struct sem_undo *un; | |
603 | ||
604 | sops = q->sops; | |
605 | nsops = q->nsops; | |
606 | un = q->undo; | |
1da177e4 LT |
607 | |
608 | for (sop = sops; sop < sops + nsops; sop++) { | |
609 | curr = sma->sem_base + sop->sem_num; | |
610 | sem_op = sop->sem_op; | |
611 | result = curr->semval; | |
78f5009c | 612 | |
1da177e4 LT |
613 | if (!sem_op && result) |
614 | goto would_block; | |
615 | ||
616 | result += sem_op; | |
617 | if (result < 0) | |
618 | goto would_block; | |
619 | if (result > SEMVMX) | |
620 | goto out_of_range; | |
78f5009c | 621 | |
1da177e4 LT |
622 | if (sop->sem_flg & SEM_UNDO) { |
623 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 624 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
625 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
626 | goto out_of_range; | |
78f5009c | 627 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 628 | } |
78f5009c | 629 | |
1da177e4 LT |
630 | curr->semval = result; |
631 | } | |
632 | ||
633 | sop--; | |
d198cd6d | 634 | pid = q->pid; |
1da177e4 LT |
635 | while (sop >= sops) { |
636 | sma->sem_base[sop->sem_num].sempid = pid; | |
1da177e4 LT |
637 | sop--; |
638 | } | |
78f5009c | 639 | |
1da177e4 LT |
640 | return 0; |
641 | ||
642 | out_of_range: | |
643 | result = -ERANGE; | |
644 | goto undo; | |
645 | ||
646 | would_block: | |
ed247b7c MS |
647 | q->blocking = sop; |
648 | ||
1da177e4 LT |
649 | if (sop->sem_flg & IPC_NOWAIT) |
650 | result = -EAGAIN; | |
651 | else | |
652 | result = 1; | |
653 | ||
654 | undo: | |
655 | sop--; | |
656 | while (sop >= sops) { | |
78f5009c PM |
657 | sem_op = sop->sem_op; |
658 | sma->sem_base[sop->sem_num].semval -= sem_op; | |
659 | if (sop->sem_flg & SEM_UNDO) | |
660 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
661 | sop--; |
662 | } | |
663 | ||
664 | return result; | |
665 | } | |
666 | ||
0a2b9d4c MS |
667 | /** wake_up_sem_queue_prepare(q, error): Prepare wake-up |
668 | * @q: queue entry that must be signaled | |
669 | * @error: Error value for the signal | |
670 | * | |
671 | * Prepare the wake-up of the queue entry q. | |
d4212093 | 672 | */ |
0a2b9d4c MS |
673 | static void wake_up_sem_queue_prepare(struct list_head *pt, |
674 | struct sem_queue *q, int error) | |
d4212093 | 675 | { |
0a2b9d4c MS |
676 | if (list_empty(pt)) { |
677 | /* | |
678 | * Hold preempt off so that we don't get preempted and have the | |
679 | * wakee busy-wait until we're scheduled back on. | |
680 | */ | |
681 | preempt_disable(); | |
682 | } | |
d4212093 | 683 | q->status = IN_WAKEUP; |
0a2b9d4c MS |
684 | q->pid = error; |
685 | ||
9f1bc2c9 | 686 | list_add_tail(&q->list, pt); |
0a2b9d4c MS |
687 | } |
688 | ||
689 | /** | |
8001c858 | 690 | * wake_up_sem_queue_do - do the actual wake-up |
0a2b9d4c MS |
691 | * @pt: list of tasks to be woken up |
692 | * | |
693 | * Do the actual wake-up. | |
694 | * The function is called without any locks held, thus the semaphore array | |
695 | * could be destroyed already and the tasks can disappear as soon as the | |
696 | * status is set to the actual return code. | |
697 | */ | |
698 | static void wake_up_sem_queue_do(struct list_head *pt) | |
699 | { | |
700 | struct sem_queue *q, *t; | |
701 | int did_something; | |
702 | ||
703 | did_something = !list_empty(pt); | |
9f1bc2c9 | 704 | list_for_each_entry_safe(q, t, pt, list) { |
0a2b9d4c MS |
705 | wake_up_process(q->sleeper); |
706 | /* q can disappear immediately after writing q->status. */ | |
707 | smp_wmb(); | |
708 | q->status = q->pid; | |
709 | } | |
710 | if (did_something) | |
711 | preempt_enable(); | |
d4212093 NP |
712 | } |
713 | ||
b97e820f MS |
714 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
715 | { | |
716 | list_del(&q->list); | |
9f1bc2c9 | 717 | if (q->nsops > 1) |
b97e820f MS |
718 | sma->complex_count--; |
719 | } | |
720 | ||
fd5db422 MS |
721 | /** check_restart(sma, q) |
722 | * @sma: semaphore array | |
723 | * @q: the operation that just completed | |
724 | * | |
725 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
726 | * waiting operations. Therefore this function checks if the restart is | |
727 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
728 | * modified the array. |
729 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 MS |
730 | */ |
731 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
732 | { | |
1a82e9e1 MS |
733 | /* pending complex alter operations are too difficult to analyse */ |
734 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
735 | return 1; |
736 | ||
737 | /* we were a sleeping complex operation. Too difficult */ | |
738 | if (q->nsops > 1) | |
739 | return 1; | |
740 | ||
1a82e9e1 MS |
741 | /* It is impossible that someone waits for the new value: |
742 | * - complex operations always restart. | |
743 | * - wait-for-zero are handled seperately. | |
744 | * - q is a previously sleeping simple operation that | |
745 | * altered the array. It must be a decrement, because | |
746 | * simple increments never sleep. | |
747 | * - If there are older (higher priority) decrements | |
748 | * in the queue, then they have observed the original | |
749 | * semval value and couldn't proceed. The operation | |
750 | * decremented to value - thus they won't proceed either. | |
751 | */ | |
752 | return 0; | |
753 | } | |
fd5db422 | 754 | |
1a82e9e1 | 755 | /** |
8001c858 | 756 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
757 | * @sma: semaphore array. |
758 | * @semnum: semaphore that was modified. | |
759 | * @pt: list head for the tasks that must be woken up. | |
760 | * | |
761 | * wake_const_ops must be called after a semaphore in a semaphore array | |
762 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
763 | * be called with semnum = -1, as well as with the number of each modified | |
764 | * semaphore. | |
765 | * The tasks that must be woken up are added to @pt. The return code | |
766 | * is stored in q->pid. | |
767 | * The function returns 1 if at least one operation was completed successfully. | |
768 | */ | |
769 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
770 | struct list_head *pt) | |
771 | { | |
772 | struct sem_queue *q; | |
773 | struct list_head *walk; | |
774 | struct list_head *pending_list; | |
775 | int semop_completed = 0; | |
776 | ||
777 | if (semnum == -1) | |
778 | pending_list = &sma->pending_const; | |
779 | else | |
780 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 781 | |
1a82e9e1 MS |
782 | walk = pending_list->next; |
783 | while (walk != pending_list) { | |
784 | int error; | |
785 | ||
786 | q = container_of(walk, struct sem_queue, list); | |
787 | walk = walk->next; | |
788 | ||
d198cd6d | 789 | error = perform_atomic_semop(sma, q); |
1a82e9e1 MS |
790 | |
791 | if (error <= 0) { | |
792 | /* operation completed, remove from queue & wakeup */ | |
793 | ||
794 | unlink_queue(sma, q); | |
795 | ||
796 | wake_up_sem_queue_prepare(pt, q, error); | |
797 | if (error == 0) | |
798 | semop_completed = 1; | |
799 | } | |
800 | } | |
801 | return semop_completed; | |
802 | } | |
803 | ||
804 | /** | |
8001c858 | 805 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
806 | * @sma: semaphore array |
807 | * @sops: operations that were performed | |
808 | * @nsops: number of operations | |
809 | * @pt: list head of the tasks that must be woken up. | |
810 | * | |
8001c858 DB |
811 | * Checks all required queue for wait-for-zero operations, based |
812 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
813 | * The function returns 1 if at least one operation was completed successfully. |
814 | */ | |
815 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
816 | int nsops, struct list_head *pt) | |
817 | { | |
818 | int i; | |
819 | int semop_completed = 0; | |
820 | int got_zero = 0; | |
821 | ||
822 | /* first: the per-semaphore queues, if known */ | |
823 | if (sops) { | |
824 | for (i = 0; i < nsops; i++) { | |
825 | int num = sops[i].sem_num; | |
826 | ||
827 | if (sma->sem_base[num].semval == 0) { | |
828 | got_zero = 1; | |
829 | semop_completed |= wake_const_ops(sma, num, pt); | |
830 | } | |
831 | } | |
832 | } else { | |
833 | /* | |
834 | * No sops means modified semaphores not known. | |
835 | * Assume all were changed. | |
fd5db422 | 836 | */ |
1a82e9e1 MS |
837 | for (i = 0; i < sma->sem_nsems; i++) { |
838 | if (sma->sem_base[i].semval == 0) { | |
839 | got_zero = 1; | |
840 | semop_completed |= wake_const_ops(sma, i, pt); | |
841 | } | |
842 | } | |
fd5db422 MS |
843 | } |
844 | /* | |
1a82e9e1 MS |
845 | * If one of the modified semaphores got 0, |
846 | * then check the global queue, too. | |
fd5db422 | 847 | */ |
1a82e9e1 MS |
848 | if (got_zero) |
849 | semop_completed |= wake_const_ops(sma, -1, pt); | |
fd5db422 | 850 | |
1a82e9e1 | 851 | return semop_completed; |
fd5db422 MS |
852 | } |
853 | ||
636c6be8 MS |
854 | |
855 | /** | |
8001c858 | 856 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
857 | * @sma: semaphore array. |
858 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 859 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
860 | * |
861 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
862 | * was modified. If multiple semaphores were modified, update_queue must |
863 | * be called with semnum = -1, as well as with the number of each modified | |
864 | * semaphore. | |
0a2b9d4c MS |
865 | * The tasks that must be woken up are added to @pt. The return code |
866 | * is stored in q->pid. | |
1a82e9e1 MS |
867 | * The function internally checks if const operations can now succeed. |
868 | * | |
0a2b9d4c | 869 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 870 | */ |
0a2b9d4c | 871 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 872 | { |
636c6be8 MS |
873 | struct sem_queue *q; |
874 | struct list_head *walk; | |
875 | struct list_head *pending_list; | |
0a2b9d4c | 876 | int semop_completed = 0; |
636c6be8 | 877 | |
9f1bc2c9 | 878 | if (semnum == -1) |
1a82e9e1 | 879 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 880 | else |
1a82e9e1 | 881 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
882 | |
883 | again: | |
636c6be8 MS |
884 | walk = pending_list->next; |
885 | while (walk != pending_list) { | |
fd5db422 | 886 | int error, restart; |
636c6be8 | 887 | |
9f1bc2c9 | 888 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 889 | walk = walk->next; |
1da177e4 | 890 | |
d987f8b2 MS |
891 | /* If we are scanning the single sop, per-semaphore list of |
892 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 893 | * necessary to scan further: simple increments |
d987f8b2 MS |
894 | * that affect only one entry succeed immediately and cannot |
895 | * be in the per semaphore pending queue, and decrements | |
896 | * cannot be successful if the value is already 0. | |
897 | */ | |
1a82e9e1 | 898 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
899 | break; |
900 | ||
d198cd6d | 901 | error = perform_atomic_semop(sma, q); |
1da177e4 LT |
902 | |
903 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
904 | if (error > 0) |
905 | continue; | |
906 | ||
b97e820f | 907 | unlink_queue(sma, q); |
9cad200c | 908 | |
0a2b9d4c | 909 | if (error) { |
fd5db422 | 910 | restart = 0; |
0a2b9d4c MS |
911 | } else { |
912 | semop_completed = 1; | |
1a82e9e1 | 913 | do_smart_wakeup_zero(sma, q->sops, q->nsops, pt); |
fd5db422 | 914 | restart = check_restart(sma, q); |
0a2b9d4c | 915 | } |
fd5db422 | 916 | |
0a2b9d4c | 917 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 918 | if (restart) |
9cad200c | 919 | goto again; |
1da177e4 | 920 | } |
0a2b9d4c | 921 | return semop_completed; |
1da177e4 LT |
922 | } |
923 | ||
0e8c6656 | 924 | /** |
8001c858 | 925 | * set_semotime - set sem_otime |
0e8c6656 MS |
926 | * @sma: semaphore array |
927 | * @sops: operations that modified the array, may be NULL | |
928 | * | |
929 | * sem_otime is replicated to avoid cache line trashing. | |
930 | * This function sets one instance to the current time. | |
931 | */ | |
932 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
933 | { | |
934 | if (sops == NULL) { | |
935 | sma->sem_base[0].sem_otime = get_seconds(); | |
936 | } else { | |
937 | sma->sem_base[sops[0].sem_num].sem_otime = | |
938 | get_seconds(); | |
939 | } | |
940 | } | |
941 | ||
0a2b9d4c | 942 | /** |
8001c858 | 943 | * do_smart_update - optimized update_queue |
fd5db422 MS |
944 | * @sma: semaphore array |
945 | * @sops: operations that were performed | |
946 | * @nsops: number of operations | |
0a2b9d4c MS |
947 | * @otime: force setting otime |
948 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 | 949 | * |
1a82e9e1 MS |
950 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
951 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
952 | * Note that the function does not do the actual wake-up: the caller is |
953 | * responsible for calling wake_up_sem_queue_do(@pt). | |
954 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 955 | */ |
0a2b9d4c MS |
956 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
957 | int otime, struct list_head *pt) | |
fd5db422 MS |
958 | { |
959 | int i; | |
960 | ||
1a82e9e1 MS |
961 | otime |= do_smart_wakeup_zero(sma, sops, nsops, pt); |
962 | ||
f269f40a MS |
963 | if (!list_empty(&sma->pending_alter)) { |
964 | /* semaphore array uses the global queue - just process it. */ | |
965 | otime |= update_queue(sma, -1, pt); | |
966 | } else { | |
967 | if (!sops) { | |
968 | /* | |
969 | * No sops, thus the modified semaphores are not | |
970 | * known. Check all. | |
971 | */ | |
972 | for (i = 0; i < sma->sem_nsems; i++) | |
973 | otime |= update_queue(sma, i, pt); | |
974 | } else { | |
975 | /* | |
976 | * Check the semaphores that were increased: | |
977 | * - No complex ops, thus all sleeping ops are | |
978 | * decrease. | |
979 | * - if we decreased the value, then any sleeping | |
980 | * semaphore ops wont be able to run: If the | |
981 | * previous value was too small, then the new | |
982 | * value will be too small, too. | |
983 | */ | |
984 | for (i = 0; i < nsops; i++) { | |
985 | if (sops[i].sem_op > 0) { | |
986 | otime |= update_queue(sma, | |
987 | sops[i].sem_num, pt); | |
988 | } | |
ab465df9 | 989 | } |
9f1bc2c9 | 990 | } |
fd5db422 | 991 | } |
0e8c6656 MS |
992 | if (otime) |
993 | set_semotime(sma, sops); | |
fd5db422 MS |
994 | } |
995 | ||
2f2ed41d | 996 | /* |
b220c57a | 997 | * check_qop: Test if a queued operation sleeps on the semaphore semnum |
2f2ed41d MS |
998 | */ |
999 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1000 | bool count_zero) | |
1001 | { | |
b220c57a | 1002 | struct sembuf *sop = q->blocking; |
2f2ed41d | 1003 | |
9b44ee2e MS |
1004 | /* |
1005 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1006 | * semaphores. This violates SUS, therefore it was changed to the | |
1007 | * standard compliant behavior. | |
1008 | * Give the administrators a chance to notice that an application | |
1009 | * might misbehave because it relies on the Linux behavior. | |
1010 | */ | |
1011 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1012 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1013 | current->comm, task_pid_nr(current)); | |
1014 | ||
b220c57a MS |
1015 | if (sop->sem_num != semnum) |
1016 | return 0; | |
2f2ed41d | 1017 | |
b220c57a MS |
1018 | if (count_zero && sop->sem_op == 0) |
1019 | return 1; | |
1020 | if (!count_zero && sop->sem_op < 0) | |
1021 | return 1; | |
1022 | ||
1023 | return 0; | |
2f2ed41d MS |
1024 | } |
1025 | ||
1da177e4 LT |
1026 | /* The following counts are associated to each semaphore: |
1027 | * semncnt number of tasks waiting on semval being nonzero | |
1028 | * semzcnt number of tasks waiting on semval being zero | |
b220c57a MS |
1029 | * |
1030 | * Per definition, a task waits only on the semaphore of the first semop | |
1031 | * that cannot proceed, even if additional operation would block, too. | |
1da177e4 | 1032 | */ |
2f2ed41d MS |
1033 | static int count_semcnt(struct sem_array *sma, ushort semnum, |
1034 | bool count_zero) | |
1da177e4 | 1035 | { |
2f2ed41d | 1036 | struct list_head *l; |
239521f3 | 1037 | struct sem_queue *q; |
2f2ed41d | 1038 | int semcnt; |
1da177e4 | 1039 | |
2f2ed41d MS |
1040 | semcnt = 0; |
1041 | /* First: check the simple operations. They are easy to evaluate */ | |
1042 | if (count_zero) | |
1043 | l = &sma->sem_base[semnum].pending_const; | |
1044 | else | |
1045 | l = &sma->sem_base[semnum].pending_alter; | |
1da177e4 | 1046 | |
2f2ed41d MS |
1047 | list_for_each_entry(q, l, list) { |
1048 | /* all task on a per-semaphore list sleep on exactly | |
1049 | * that semaphore | |
1050 | */ | |
1051 | semcnt++; | |
ebc2e5e6 RR |
1052 | } |
1053 | ||
2f2ed41d | 1054 | /* Then: check the complex operations. */ |
1994862d | 1055 | list_for_each_entry(q, &sma->pending_alter, list) { |
2f2ed41d MS |
1056 | semcnt += check_qop(sma, semnum, q, count_zero); |
1057 | } | |
1058 | if (count_zero) { | |
1059 | list_for_each_entry(q, &sma->pending_const, list) { | |
1060 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1061 | } | |
1994862d | 1062 | } |
2f2ed41d | 1063 | return semcnt; |
1da177e4 LT |
1064 | } |
1065 | ||
d9a605e4 DB |
1066 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1067 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1068 | * remains locked on exit. |
1da177e4 | 1069 | */ |
01b8b07a | 1070 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1071 | { |
380af1b3 MS |
1072 | struct sem_undo *un, *tu; |
1073 | struct sem_queue *q, *tq; | |
01b8b07a | 1074 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 1075 | struct list_head tasks; |
9f1bc2c9 | 1076 | int i; |
1da177e4 | 1077 | |
380af1b3 | 1078 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1079 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1080 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1081 | list_del(&un->list_id); | |
1082 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1083 | un->semid = -1; |
380af1b3 MS |
1084 | list_del_rcu(&un->list_proc); |
1085 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1086 | kfree_rcu(un, rcu); |
380af1b3 | 1087 | } |
1da177e4 LT |
1088 | |
1089 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 1090 | INIT_LIST_HEAD(&tasks); |
1a82e9e1 MS |
1091 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1092 | unlink_queue(sma, q); | |
1093 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1094 | } | |
1095 | ||
1096 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1097 | unlink_queue(sma, q); |
0a2b9d4c | 1098 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 1099 | } |
9f1bc2c9 RR |
1100 | for (i = 0; i < sma->sem_nsems; i++) { |
1101 | struct sem *sem = sma->sem_base + i; | |
1a82e9e1 MS |
1102 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1103 | unlink_queue(sma, q); | |
1104 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1105 | } | |
1106 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 RR |
1107 | unlink_queue(sma, q); |
1108 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1109 | } | |
1110 | } | |
1da177e4 | 1111 | |
7ca7e564 ND |
1112 | /* Remove the semaphore set from the IDR */ |
1113 | sem_rmid(ns, sma); | |
6062a8dc | 1114 | sem_unlock(sma, -1); |
6d49dab8 | 1115 | rcu_read_unlock(); |
1da177e4 | 1116 | |
0a2b9d4c | 1117 | wake_up_sem_queue_do(&tasks); |
e3893534 | 1118 | ns->used_sems -= sma->sem_nsems; |
53dad6d3 | 1119 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1123 | { | |
239521f3 | 1124 | switch (version) { |
1da177e4 LT |
1125 | case IPC_64: |
1126 | return copy_to_user(buf, in, sizeof(*in)); | |
1127 | case IPC_OLD: | |
1128 | { | |
1129 | struct semid_ds out; | |
1130 | ||
982f7c2b DR |
1131 | memset(&out, 0, sizeof(out)); |
1132 | ||
1da177e4 LT |
1133 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1134 | ||
1135 | out.sem_otime = in->sem_otime; | |
1136 | out.sem_ctime = in->sem_ctime; | |
1137 | out.sem_nsems = in->sem_nsems; | |
1138 | ||
1139 | return copy_to_user(buf, &out, sizeof(out)); | |
1140 | } | |
1141 | default: | |
1142 | return -EINVAL; | |
1143 | } | |
1144 | } | |
1145 | ||
d12e1e50 MS |
1146 | static time_t get_semotime(struct sem_array *sma) |
1147 | { | |
1148 | int i; | |
1149 | time_t res; | |
1150 | ||
1151 | res = sma->sem_base[0].sem_otime; | |
1152 | for (i = 1; i < sma->sem_nsems; i++) { | |
1153 | time_t to = sma->sem_base[i].sem_otime; | |
1154 | ||
1155 | if (to > res) | |
1156 | res = to; | |
1157 | } | |
1158 | return res; | |
1159 | } | |
1160 | ||
4b9fcb0e | 1161 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1162 | int cmd, int version, void __user *p) |
1da177e4 | 1163 | { |
e5cc9c7b | 1164 | int err; |
1da177e4 LT |
1165 | struct sem_array *sma; |
1166 | ||
239521f3 | 1167 | switch (cmd) { |
1da177e4 LT |
1168 | case IPC_INFO: |
1169 | case SEM_INFO: | |
1170 | { | |
1171 | struct seminfo seminfo; | |
1172 | int max_id; | |
1173 | ||
1174 | err = security_sem_semctl(NULL, cmd); | |
1175 | if (err) | |
1176 | return err; | |
46c0a8ca | 1177 | |
239521f3 | 1178 | memset(&seminfo, 0, sizeof(seminfo)); |
e3893534 KK |
1179 | seminfo.semmni = ns->sc_semmni; |
1180 | seminfo.semmns = ns->sc_semmns; | |
1181 | seminfo.semmsl = ns->sc_semmsl; | |
1182 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1183 | seminfo.semvmx = SEMVMX; |
1184 | seminfo.semmnu = SEMMNU; | |
1185 | seminfo.semmap = SEMMAP; | |
1186 | seminfo.semume = SEMUME; | |
d9a605e4 | 1187 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1188 | if (cmd == SEM_INFO) { |
e3893534 KK |
1189 | seminfo.semusz = sem_ids(ns).in_use; |
1190 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1191 | } else { |
1192 | seminfo.semusz = SEMUSZ; | |
1193 | seminfo.semaem = SEMAEM; | |
1194 | } | |
7ca7e564 | 1195 | max_id = ipc_get_maxid(&sem_ids(ns)); |
d9a605e4 | 1196 | up_read(&sem_ids(ns).rwsem); |
46c0a8ca | 1197 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 | 1198 | return -EFAULT; |
239521f3 | 1199 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 1200 | } |
4b9fcb0e | 1201 | case IPC_STAT: |
1da177e4 LT |
1202 | case SEM_STAT: |
1203 | { | |
1204 | struct semid64_ds tbuf; | |
16df3674 DB |
1205 | int id = 0; |
1206 | ||
1207 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1208 | |
941b0304 | 1209 | rcu_read_lock(); |
4b9fcb0e | 1210 | if (cmd == SEM_STAT) { |
16df3674 DB |
1211 | sma = sem_obtain_object(ns, semid); |
1212 | if (IS_ERR(sma)) { | |
1213 | err = PTR_ERR(sma); | |
1214 | goto out_unlock; | |
1215 | } | |
4b9fcb0e PP |
1216 | id = sma->sem_perm.id; |
1217 | } else { | |
16df3674 DB |
1218 | sma = sem_obtain_object_check(ns, semid); |
1219 | if (IS_ERR(sma)) { | |
1220 | err = PTR_ERR(sma); | |
1221 | goto out_unlock; | |
1222 | } | |
4b9fcb0e | 1223 | } |
1da177e4 LT |
1224 | |
1225 | err = -EACCES; | |
b0e77598 | 1226 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1227 | goto out_unlock; |
1228 | ||
1229 | err = security_sem_semctl(sma, cmd); | |
1230 | if (err) | |
1231 | goto out_unlock; | |
1232 | ||
1da177e4 | 1233 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
d12e1e50 MS |
1234 | tbuf.sem_otime = get_semotime(sma); |
1235 | tbuf.sem_ctime = sma->sem_ctime; | |
1236 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1237 | rcu_read_unlock(); |
e1fd1f49 | 1238 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1239 | return -EFAULT; |
1240 | return id; | |
1241 | } | |
1242 | default: | |
1243 | return -EINVAL; | |
1244 | } | |
1da177e4 | 1245 | out_unlock: |
16df3674 | 1246 | rcu_read_unlock(); |
1da177e4 LT |
1247 | return err; |
1248 | } | |
1249 | ||
e1fd1f49 AV |
1250 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1251 | unsigned long arg) | |
1252 | { | |
1253 | struct sem_undo *un; | |
1254 | struct sem_array *sma; | |
239521f3 | 1255 | struct sem *curr; |
e1fd1f49 | 1256 | int err; |
e1fd1f49 AV |
1257 | struct list_head tasks; |
1258 | int val; | |
1259 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1260 | /* big-endian 64bit */ | |
1261 | val = arg >> 32; | |
1262 | #else | |
1263 | /* 32bit or little-endian 64bit */ | |
1264 | val = arg; | |
1265 | #endif | |
1266 | ||
6062a8dc RR |
1267 | if (val > SEMVMX || val < 0) |
1268 | return -ERANGE; | |
e1fd1f49 AV |
1269 | |
1270 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1271 | |
6062a8dc RR |
1272 | rcu_read_lock(); |
1273 | sma = sem_obtain_object_check(ns, semid); | |
1274 | if (IS_ERR(sma)) { | |
1275 | rcu_read_unlock(); | |
1276 | return PTR_ERR(sma); | |
1277 | } | |
1278 | ||
1279 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1280 | rcu_read_unlock(); | |
1281 | return -EINVAL; | |
1282 | } | |
1283 | ||
1284 | ||
1285 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1286 | rcu_read_unlock(); | |
1287 | return -EACCES; | |
1288 | } | |
e1fd1f49 AV |
1289 | |
1290 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1291 | if (err) { |
1292 | rcu_read_unlock(); | |
1293 | return -EACCES; | |
1294 | } | |
e1fd1f49 | 1295 | |
6062a8dc | 1296 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1297 | |
0f3d2b01 | 1298 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1299 | sem_unlock(sma, -1); |
1300 | rcu_read_unlock(); | |
1301 | return -EIDRM; | |
1302 | } | |
1303 | ||
e1fd1f49 AV |
1304 | curr = &sma->sem_base[semnum]; |
1305 | ||
cf9d5d78 | 1306 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1307 | list_for_each_entry(un, &sma->list_id, list_id) |
1308 | un->semadj[semnum] = 0; | |
1309 | ||
1310 | curr->semval = val; | |
1311 | curr->sempid = task_tgid_vnr(current); | |
1312 | sma->sem_ctime = get_seconds(); | |
1313 | /* maybe some queued-up processes were waiting for this */ | |
1314 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1315 | sem_unlock(sma, -1); |
6d49dab8 | 1316 | rcu_read_unlock(); |
e1fd1f49 | 1317 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1318 | return 0; |
e1fd1f49 AV |
1319 | } |
1320 | ||
e3893534 | 1321 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1322 | int cmd, void __user *p) |
1da177e4 LT |
1323 | { |
1324 | struct sem_array *sma; | |
239521f3 | 1325 | struct sem *curr; |
16df3674 | 1326 | int err, nsems; |
1da177e4 | 1327 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1328 | ushort *sem_io = fast_sem_io; |
0a2b9d4c | 1329 | struct list_head tasks; |
1da177e4 | 1330 | |
16df3674 DB |
1331 | INIT_LIST_HEAD(&tasks); |
1332 | ||
1333 | rcu_read_lock(); | |
1334 | sma = sem_obtain_object_check(ns, semid); | |
1335 | if (IS_ERR(sma)) { | |
1336 | rcu_read_unlock(); | |
023a5355 | 1337 | return PTR_ERR(sma); |
16df3674 | 1338 | } |
1da177e4 LT |
1339 | |
1340 | nsems = sma->sem_nsems; | |
1341 | ||
1da177e4 | 1342 | err = -EACCES; |
c728b9c8 LT |
1343 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1344 | goto out_rcu_wakeup; | |
1da177e4 LT |
1345 | |
1346 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1347 | if (err) |
1348 | goto out_rcu_wakeup; | |
1da177e4 LT |
1349 | |
1350 | err = -EACCES; | |
1351 | switch (cmd) { | |
1352 | case GETALL: | |
1353 | { | |
e1fd1f49 | 1354 | ushort __user *array = p; |
1da177e4 LT |
1355 | int i; |
1356 | ||
ce857229 | 1357 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1358 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1359 | err = -EIDRM; |
1360 | goto out_unlock; | |
1361 | } | |
239521f3 | 1362 | if (nsems > SEMMSL_FAST) { |
ce857229 | 1363 | if (!ipc_rcu_getref(sma)) { |
ce857229 | 1364 | err = -EIDRM; |
6e224f94 | 1365 | goto out_unlock; |
ce857229 AV |
1366 | } |
1367 | sem_unlock(sma, -1); | |
6d49dab8 | 1368 | rcu_read_unlock(); |
1da177e4 | 1369 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1370 | if (sem_io == NULL) { |
53dad6d3 | 1371 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1372 | return -ENOMEM; |
1373 | } | |
1374 | ||
4091fd94 | 1375 | rcu_read_lock(); |
6ff37972 | 1376 | sem_lock_and_putref(sma); |
0f3d2b01 | 1377 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1378 | err = -EIDRM; |
6e224f94 | 1379 | goto out_unlock; |
1da177e4 | 1380 | } |
ce857229 | 1381 | } |
1da177e4 LT |
1382 | for (i = 0; i < sma->sem_nsems; i++) |
1383 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1384 | sem_unlock(sma, -1); |
6d49dab8 | 1385 | rcu_read_unlock(); |
1da177e4 | 1386 | err = 0; |
239521f3 | 1387 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1388 | err = -EFAULT; |
1389 | goto out_free; | |
1390 | } | |
1391 | case SETALL: | |
1392 | { | |
1393 | int i; | |
1394 | struct sem_undo *un; | |
1395 | ||
6062a8dc | 1396 | if (!ipc_rcu_getref(sma)) { |
6e224f94 MS |
1397 | err = -EIDRM; |
1398 | goto out_rcu_wakeup; | |
6062a8dc | 1399 | } |
16df3674 | 1400 | rcu_read_unlock(); |
1da177e4 | 1401 | |
239521f3 | 1402 | if (nsems > SEMMSL_FAST) { |
1da177e4 | 1403 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1404 | if (sem_io == NULL) { |
53dad6d3 | 1405 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1406 | return -ENOMEM; |
1407 | } | |
1408 | } | |
1409 | ||
239521f3 | 1410 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
53dad6d3 | 1411 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1412 | err = -EFAULT; |
1413 | goto out_free; | |
1414 | } | |
1415 | ||
1416 | for (i = 0; i < nsems; i++) { | |
1417 | if (sem_io[i] > SEMVMX) { | |
53dad6d3 | 1418 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1419 | err = -ERANGE; |
1420 | goto out_free; | |
1421 | } | |
1422 | } | |
4091fd94 | 1423 | rcu_read_lock(); |
6ff37972 | 1424 | sem_lock_and_putref(sma); |
0f3d2b01 | 1425 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1426 | err = -EIDRM; |
6e224f94 | 1427 | goto out_unlock; |
1da177e4 LT |
1428 | } |
1429 | ||
1430 | for (i = 0; i < nsems; i++) | |
1431 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 | 1432 | |
cf9d5d78 | 1433 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1434 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1435 | for (i = 0; i < nsems; i++) |
1436 | un->semadj[i] = 0; | |
4daa28f6 | 1437 | } |
1da177e4 LT |
1438 | sma->sem_ctime = get_seconds(); |
1439 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1440 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1441 | err = 0; |
1442 | goto out_unlock; | |
1443 | } | |
e1fd1f49 | 1444 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1445 | } |
1446 | err = -EINVAL; | |
c728b9c8 LT |
1447 | if (semnum < 0 || semnum >= nsems) |
1448 | goto out_rcu_wakeup; | |
1da177e4 | 1449 | |
6062a8dc | 1450 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1451 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1452 | err = -EIDRM; |
1453 | goto out_unlock; | |
1454 | } | |
1da177e4 LT |
1455 | curr = &sma->sem_base[semnum]; |
1456 | ||
1457 | switch (cmd) { | |
1458 | case GETVAL: | |
1459 | err = curr->semval; | |
1460 | goto out_unlock; | |
1461 | case GETPID: | |
1462 | err = curr->sempid; | |
1463 | goto out_unlock; | |
1464 | case GETNCNT: | |
2f2ed41d | 1465 | err = count_semcnt(sma, semnum, 0); |
1da177e4 LT |
1466 | goto out_unlock; |
1467 | case GETZCNT: | |
2f2ed41d | 1468 | err = count_semcnt(sma, semnum, 1); |
1da177e4 | 1469 | goto out_unlock; |
1da177e4 | 1470 | } |
16df3674 | 1471 | |
1da177e4 | 1472 | out_unlock: |
6062a8dc | 1473 | sem_unlock(sma, -1); |
c728b9c8 | 1474 | out_rcu_wakeup: |
6d49dab8 | 1475 | rcu_read_unlock(); |
0a2b9d4c | 1476 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 1477 | out_free: |
239521f3 | 1478 | if (sem_io != fast_sem_io) |
1da177e4 LT |
1479 | ipc_free(sem_io, sizeof(ushort)*nsems); |
1480 | return err; | |
1481 | } | |
1482 | ||
016d7132 PP |
1483 | static inline unsigned long |
1484 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1485 | { |
239521f3 | 1486 | switch (version) { |
1da177e4 | 1487 | case IPC_64: |
016d7132 | 1488 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1489 | return -EFAULT; |
1da177e4 | 1490 | return 0; |
1da177e4 LT |
1491 | case IPC_OLD: |
1492 | { | |
1493 | struct semid_ds tbuf_old; | |
1494 | ||
239521f3 | 1495 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1496 | return -EFAULT; |
1497 | ||
016d7132 PP |
1498 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1499 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1500 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1501 | |
1502 | return 0; | |
1503 | } | |
1504 | default: | |
1505 | return -EINVAL; | |
1506 | } | |
1507 | } | |
1508 | ||
522bb2a2 | 1509 | /* |
d9a605e4 | 1510 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1511 | * to be held in write mode. |
d9a605e4 | 1512 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1513 | */ |
21a4826a | 1514 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1515 | int cmd, int version, void __user *p) |
1da177e4 LT |
1516 | { |
1517 | struct sem_array *sma; | |
1518 | int err; | |
016d7132 | 1519 | struct semid64_ds semid64; |
1da177e4 LT |
1520 | struct kern_ipc_perm *ipcp; |
1521 | ||
239521f3 | 1522 | if (cmd == IPC_SET) { |
e1fd1f49 | 1523 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1524 | return -EFAULT; |
1da177e4 | 1525 | } |
073115d6 | 1526 | |
d9a605e4 | 1527 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1528 | rcu_read_lock(); |
1529 | ||
16df3674 DB |
1530 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1531 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1532 | if (IS_ERR(ipcp)) { |
1533 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1534 | goto out_unlock1; |
1535 | } | |
073115d6 | 1536 | |
a5f75e7f | 1537 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1538 | |
1539 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1540 | if (err) |
1541 | goto out_unlock1; | |
1da177e4 | 1542 | |
7b4cc5d8 | 1543 | switch (cmd) { |
1da177e4 | 1544 | case IPC_RMID: |
6062a8dc | 1545 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1546 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1547 | freeary(ns, ipcp); |
522bb2a2 | 1548 | goto out_up; |
1da177e4 | 1549 | case IPC_SET: |
6062a8dc | 1550 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1551 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1552 | if (err) | |
7b4cc5d8 | 1553 | goto out_unlock0; |
1da177e4 | 1554 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1555 | break; |
1556 | default: | |
1da177e4 | 1557 | err = -EINVAL; |
7b4cc5d8 | 1558 | goto out_unlock1; |
1da177e4 | 1559 | } |
1da177e4 | 1560 | |
7b4cc5d8 | 1561 | out_unlock0: |
6062a8dc | 1562 | sem_unlock(sma, -1); |
7b4cc5d8 | 1563 | out_unlock1: |
6d49dab8 | 1564 | rcu_read_unlock(); |
522bb2a2 | 1565 | out_up: |
d9a605e4 | 1566 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1567 | return err; |
1568 | } | |
1569 | ||
e1fd1f49 | 1570 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1571 | { |
1da177e4 | 1572 | int version; |
e3893534 | 1573 | struct ipc_namespace *ns; |
e1fd1f49 | 1574 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1575 | |
1576 | if (semid < 0) | |
1577 | return -EINVAL; | |
1578 | ||
1579 | version = ipc_parse_version(&cmd); | |
e3893534 | 1580 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1581 | |
239521f3 | 1582 | switch (cmd) { |
1da177e4 LT |
1583 | case IPC_INFO: |
1584 | case SEM_INFO: | |
4b9fcb0e | 1585 | case IPC_STAT: |
1da177e4 | 1586 | case SEM_STAT: |
e1fd1f49 | 1587 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1588 | case GETALL: |
1589 | case GETVAL: | |
1590 | case GETPID: | |
1591 | case GETNCNT: | |
1592 | case GETZCNT: | |
1da177e4 | 1593 | case SETALL: |
e1fd1f49 AV |
1594 | return semctl_main(ns, semid, semnum, cmd, p); |
1595 | case SETVAL: | |
1596 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1597 | case IPC_RMID: |
1598 | case IPC_SET: | |
e1fd1f49 | 1599 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1600 | default: |
1601 | return -EINVAL; | |
1602 | } | |
1603 | } | |
1604 | ||
1da177e4 LT |
1605 | /* If the task doesn't already have a undo_list, then allocate one |
1606 | * here. We guarantee there is only one thread using this undo list, | |
1607 | * and current is THE ONE | |
1608 | * | |
1609 | * If this allocation and assignment succeeds, but later | |
1610 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1611 | * Just let it stay associated with the task, and it'll be freed later | |
1612 | * at exit time. | |
1613 | * | |
1614 | * This can block, so callers must hold no locks. | |
1615 | */ | |
1616 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1617 | { | |
1618 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1619 | |
1620 | undo_list = current->sysvsem.undo_list; | |
1621 | if (!undo_list) { | |
2453a306 | 1622 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1623 | if (undo_list == NULL) |
1624 | return -ENOMEM; | |
00a5dfdb | 1625 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1626 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1627 | INIT_LIST_HEAD(&undo_list->list_proc); |
1628 | ||
1da177e4 LT |
1629 | current->sysvsem.undo_list = undo_list; |
1630 | } | |
1631 | *undo_listp = undo_list; | |
1632 | return 0; | |
1633 | } | |
1634 | ||
bf17bb71 | 1635 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1636 | { |
bf17bb71 | 1637 | struct sem_undo *un; |
4daa28f6 | 1638 | |
bf17bb71 NP |
1639 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1640 | if (un->semid == semid) | |
1641 | return un; | |
1da177e4 | 1642 | } |
4daa28f6 | 1643 | return NULL; |
1da177e4 LT |
1644 | } |
1645 | ||
bf17bb71 NP |
1646 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1647 | { | |
1648 | struct sem_undo *un; | |
1649 | ||
239521f3 | 1650 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1651 | |
1652 | un = __lookup_undo(ulp, semid); | |
1653 | if (un) { | |
1654 | list_del_rcu(&un->list_proc); | |
1655 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1656 | } | |
1657 | return un; | |
1658 | } | |
1659 | ||
4daa28f6 | 1660 | /** |
8001c858 | 1661 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1662 | * @ns: namespace |
1663 | * @semid: semaphore array id | |
1664 | * | |
1665 | * The function looks up (and if not present creates) the undo structure. | |
1666 | * The size of the undo structure depends on the size of the semaphore | |
1667 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1668 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1669 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1670 | */ |
1671 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1672 | { |
1673 | struct sem_array *sma; | |
1674 | struct sem_undo_list *ulp; | |
1675 | struct sem_undo *un, *new; | |
6062a8dc | 1676 | int nsems, error; |
1da177e4 LT |
1677 | |
1678 | error = get_undo_list(&ulp); | |
1679 | if (error) | |
1680 | return ERR_PTR(error); | |
1681 | ||
380af1b3 | 1682 | rcu_read_lock(); |
c530c6ac | 1683 | spin_lock(&ulp->lock); |
1da177e4 | 1684 | un = lookup_undo(ulp, semid); |
c530c6ac | 1685 | spin_unlock(&ulp->lock); |
239521f3 | 1686 | if (likely(un != NULL)) |
1da177e4 LT |
1687 | goto out; |
1688 | ||
1689 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1690 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1691 | sma = sem_obtain_object_check(ns, semid); |
1692 | if (IS_ERR(sma)) { | |
1693 | rcu_read_unlock(); | |
4de85cd6 | 1694 | return ERR_CAST(sma); |
16df3674 | 1695 | } |
023a5355 | 1696 | |
1da177e4 | 1697 | nsems = sma->sem_nsems; |
6062a8dc RR |
1698 | if (!ipc_rcu_getref(sma)) { |
1699 | rcu_read_unlock(); | |
1700 | un = ERR_PTR(-EIDRM); | |
1701 | goto out; | |
1702 | } | |
16df3674 | 1703 | rcu_read_unlock(); |
1da177e4 | 1704 | |
4daa28f6 | 1705 | /* step 2: allocate new undo structure */ |
4668edc3 | 1706 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1707 | if (!new) { |
53dad6d3 | 1708 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
1709 | return ERR_PTR(-ENOMEM); |
1710 | } | |
1da177e4 | 1711 | |
380af1b3 | 1712 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1713 | rcu_read_lock(); |
6ff37972 | 1714 | sem_lock_and_putref(sma); |
0f3d2b01 | 1715 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1716 | sem_unlock(sma, -1); |
6d49dab8 | 1717 | rcu_read_unlock(); |
1da177e4 LT |
1718 | kfree(new); |
1719 | un = ERR_PTR(-EIDRM); | |
1720 | goto out; | |
1721 | } | |
380af1b3 MS |
1722 | spin_lock(&ulp->lock); |
1723 | ||
1724 | /* | |
1725 | * step 4: check for races: did someone else allocate the undo struct? | |
1726 | */ | |
1727 | un = lookup_undo(ulp, semid); | |
1728 | if (un) { | |
1729 | kfree(new); | |
1730 | goto success; | |
1731 | } | |
4daa28f6 MS |
1732 | /* step 5: initialize & link new undo structure */ |
1733 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1734 | new->ulp = ulp; |
4daa28f6 MS |
1735 | new->semid = semid; |
1736 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1737 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1738 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1739 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1740 | un = new; |
4daa28f6 | 1741 | |
380af1b3 | 1742 | success: |
c530c6ac | 1743 | spin_unlock(&ulp->lock); |
6062a8dc | 1744 | sem_unlock(sma, -1); |
1da177e4 LT |
1745 | out: |
1746 | return un; | |
1747 | } | |
1748 | ||
c61284e9 MS |
1749 | |
1750 | /** | |
8001c858 | 1751 | * get_queue_result - retrieve the result code from sem_queue |
c61284e9 MS |
1752 | * @q: Pointer to queue structure |
1753 | * | |
1754 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1755 | * q->status, then we must loop until the value is replaced with the final | |
1756 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1757 | * signal) and in parallel the task is woken up by another task because it got | |
1758 | * the requested semaphores. | |
1759 | * | |
1760 | * The function can be called with or without holding the semaphore spinlock. | |
1761 | */ | |
1762 | static int get_queue_result(struct sem_queue *q) | |
1763 | { | |
1764 | int error; | |
1765 | ||
1766 | error = q->status; | |
1767 | while (unlikely(error == IN_WAKEUP)) { | |
1768 | cpu_relax(); | |
1769 | error = q->status; | |
1770 | } | |
1771 | ||
1772 | return error; | |
1773 | } | |
1774 | ||
d5460c99 HC |
1775 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1776 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1777 | { |
1778 | int error = -EINVAL; | |
1779 | struct sem_array *sma; | |
1780 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1781 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1782 | struct sem_undo *un; |
6062a8dc | 1783 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1784 | struct sem_queue queue; |
1785 | unsigned long jiffies_left = 0; | |
e3893534 | 1786 | struct ipc_namespace *ns; |
0a2b9d4c | 1787 | struct list_head tasks; |
e3893534 KK |
1788 | |
1789 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1790 | |
1791 | if (nsops < 1 || semid < 0) | |
1792 | return -EINVAL; | |
e3893534 | 1793 | if (nsops > ns->sc_semopm) |
1da177e4 | 1794 | return -E2BIG; |
239521f3 MS |
1795 | if (nsops > SEMOPM_FAST) { |
1796 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1797 | if (sops == NULL) | |
1da177e4 LT |
1798 | return -ENOMEM; |
1799 | } | |
239521f3 MS |
1800 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1801 | error = -EFAULT; | |
1da177e4 LT |
1802 | goto out_free; |
1803 | } | |
1804 | if (timeout) { | |
1805 | struct timespec _timeout; | |
1806 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1807 | error = -EFAULT; | |
1808 | goto out_free; | |
1809 | } | |
1810 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1811 | _timeout.tv_nsec >= 1000000000L) { | |
1812 | error = -EINVAL; | |
1813 | goto out_free; | |
1814 | } | |
1815 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1816 | } | |
1817 | max = 0; | |
1818 | for (sop = sops; sop < sops + nsops; sop++) { | |
1819 | if (sop->sem_num >= max) | |
1820 | max = sop->sem_num; | |
1821 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1822 | undos = 1; |
1823 | if (sop->sem_op != 0) | |
1da177e4 LT |
1824 | alter = 1; |
1825 | } | |
1da177e4 | 1826 | |
6062a8dc RR |
1827 | INIT_LIST_HEAD(&tasks); |
1828 | ||
1da177e4 | 1829 | if (undos) { |
6062a8dc | 1830 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1831 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1832 | if (IS_ERR(un)) { |
1833 | error = PTR_ERR(un); | |
1834 | goto out_free; | |
1835 | } | |
6062a8dc | 1836 | } else { |
1da177e4 | 1837 | un = NULL; |
6062a8dc RR |
1838 | rcu_read_lock(); |
1839 | } | |
1da177e4 | 1840 | |
16df3674 | 1841 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1842 | if (IS_ERR(sma)) { |
6062a8dc | 1843 | rcu_read_unlock(); |
023a5355 | 1844 | error = PTR_ERR(sma); |
1da177e4 | 1845 | goto out_free; |
023a5355 ND |
1846 | } |
1847 | ||
16df3674 | 1848 | error = -EFBIG; |
c728b9c8 LT |
1849 | if (max >= sma->sem_nsems) |
1850 | goto out_rcu_wakeup; | |
16df3674 DB |
1851 | |
1852 | error = -EACCES; | |
c728b9c8 LT |
1853 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
1854 | goto out_rcu_wakeup; | |
16df3674 DB |
1855 | |
1856 | error = security_sem_semop(sma, sops, nsops, alter); | |
c728b9c8 LT |
1857 | if (error) |
1858 | goto out_rcu_wakeup; | |
16df3674 | 1859 | |
6e224f94 MS |
1860 | error = -EIDRM; |
1861 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
1862 | /* |
1863 | * We eventually might perform the following check in a lockless | |
1864 | * fashion, considering ipc_valid_object() locking constraints. | |
1865 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1866 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1867 | * check below. More details on the fine grained locking scheme | |
1868 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1869 | */ | |
1870 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 1871 | goto out_unlock_free; |
1da177e4 | 1872 | /* |
4daa28f6 | 1873 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1874 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1875 | * and now a new array with received the same id. Check and fail. |
25985edc | 1876 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1877 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1878 | */ |
6062a8dc RR |
1879 | if (un && un->semid == -1) |
1880 | goto out_unlock_free; | |
4daa28f6 | 1881 | |
d198cd6d MS |
1882 | queue.sops = sops; |
1883 | queue.nsops = nsops; | |
1884 | queue.undo = un; | |
1885 | queue.pid = task_tgid_vnr(current); | |
1886 | queue.alter = alter; | |
1887 | ||
1888 | error = perform_atomic_semop(sma, &queue); | |
0e8c6656 MS |
1889 | if (error == 0) { |
1890 | /* If the operation was successful, then do | |
1891 | * the required updates. | |
1892 | */ | |
1893 | if (alter) | |
0a2b9d4c | 1894 | do_smart_update(sma, sops, nsops, 1, &tasks); |
0e8c6656 MS |
1895 | else |
1896 | set_semotime(sma, sops); | |
1da177e4 | 1897 | } |
0e8c6656 MS |
1898 | if (error <= 0) |
1899 | goto out_unlock_free; | |
1da177e4 LT |
1900 | |
1901 | /* We need to sleep on this operation, so we put the current | |
1902 | * task into the pending queue and go to sleep. | |
1903 | */ | |
46c0a8ca | 1904 | |
b97e820f MS |
1905 | if (nsops == 1) { |
1906 | struct sem *curr; | |
1907 | curr = &sma->sem_base[sops->sem_num]; | |
1908 | ||
f269f40a MS |
1909 | if (alter) { |
1910 | if (sma->complex_count) { | |
1911 | list_add_tail(&queue.list, | |
1912 | &sma->pending_alter); | |
1913 | } else { | |
1914 | ||
1915 | list_add_tail(&queue.list, | |
1916 | &curr->pending_alter); | |
1917 | } | |
1918 | } else { | |
1a82e9e1 | 1919 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 1920 | } |
b97e820f | 1921 | } else { |
f269f40a MS |
1922 | if (!sma->complex_count) |
1923 | merge_queues(sma); | |
1924 | ||
9f1bc2c9 | 1925 | if (alter) |
1a82e9e1 | 1926 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1927 | else |
1a82e9e1 MS |
1928 | list_add_tail(&queue.list, &sma->pending_const); |
1929 | ||
b97e820f MS |
1930 | sma->complex_count++; |
1931 | } | |
1932 | ||
1da177e4 LT |
1933 | queue.status = -EINTR; |
1934 | queue.sleeper = current; | |
0b0577f6 MS |
1935 | |
1936 | sleep_again: | |
1da177e4 | 1937 | current->state = TASK_INTERRUPTIBLE; |
6062a8dc | 1938 | sem_unlock(sma, locknum); |
6d49dab8 | 1939 | rcu_read_unlock(); |
1da177e4 LT |
1940 | |
1941 | if (timeout) | |
1942 | jiffies_left = schedule_timeout(jiffies_left); | |
1943 | else | |
1944 | schedule(); | |
1945 | ||
c61284e9 | 1946 | error = get_queue_result(&queue); |
1da177e4 LT |
1947 | |
1948 | if (error != -EINTR) { | |
1949 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1950 | * resources. |
1951 | * Perform a smp_mb(): User space could assume that semop() | |
1952 | * is a memory barrier: Without the mb(), the cpu could | |
1953 | * speculatively read in user space stale data that was | |
1954 | * overwritten by the previous owner of the semaphore. | |
1955 | */ | |
1956 | smp_mb(); | |
1957 | ||
1da177e4 LT |
1958 | goto out_free; |
1959 | } | |
1960 | ||
321310ce | 1961 | rcu_read_lock(); |
6062a8dc | 1962 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1963 | |
1964 | /* | |
1965 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
1966 | */ | |
1967 | error = get_queue_result(&queue); | |
1968 | ||
1969 | /* | |
1970 | * Array removed? If yes, leave without sem_unlock(). | |
1971 | */ | |
023a5355 | 1972 | if (IS_ERR(sma)) { |
321310ce | 1973 | rcu_read_unlock(); |
1da177e4 LT |
1974 | goto out_free; |
1975 | } | |
1976 | ||
c61284e9 | 1977 | |
1da177e4 | 1978 | /* |
d694ad62 MS |
1979 | * If queue.status != -EINTR we are woken up by another process. |
1980 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 1981 | */ |
3ab08fe2 | 1982 | if (error != -EINTR) |
1da177e4 | 1983 | goto out_unlock_free; |
1da177e4 LT |
1984 | |
1985 | /* | |
1986 | * If an interrupt occurred we have to clean up the queue | |
1987 | */ | |
1988 | if (timeout && jiffies_left == 0) | |
1989 | error = -EAGAIN; | |
0b0577f6 MS |
1990 | |
1991 | /* | |
1992 | * If the wakeup was spurious, just retry | |
1993 | */ | |
1994 | if (error == -EINTR && !signal_pending(current)) | |
1995 | goto sleep_again; | |
1996 | ||
b97e820f | 1997 | unlink_queue(sma, &queue); |
1da177e4 LT |
1998 | |
1999 | out_unlock_free: | |
6062a8dc | 2000 | sem_unlock(sma, locknum); |
c728b9c8 | 2001 | out_rcu_wakeup: |
6d49dab8 | 2002 | rcu_read_unlock(); |
0a2b9d4c | 2003 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 2004 | out_free: |
239521f3 | 2005 | if (sops != fast_sops) |
1da177e4 LT |
2006 | kfree(sops); |
2007 | return error; | |
2008 | } | |
2009 | ||
d5460c99 HC |
2010 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2011 | unsigned, nsops) | |
1da177e4 LT |
2012 | { |
2013 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2014 | } | |
2015 | ||
2016 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2017 | * parent and child tasks. | |
1da177e4 LT |
2018 | */ |
2019 | ||
2020 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2021 | { | |
2022 | struct sem_undo_list *undo_list; | |
2023 | int error; | |
2024 | ||
2025 | if (clone_flags & CLONE_SYSVSEM) { | |
2026 | error = get_undo_list(&undo_list); | |
2027 | if (error) | |
2028 | return error; | |
1da177e4 LT |
2029 | atomic_inc(&undo_list->refcnt); |
2030 | tsk->sysvsem.undo_list = undo_list; | |
46c0a8ca | 2031 | } else |
1da177e4 LT |
2032 | tsk->sysvsem.undo_list = NULL; |
2033 | ||
2034 | return 0; | |
2035 | } | |
2036 | ||
2037 | /* | |
2038 | * add semadj values to semaphores, free undo structures. | |
2039 | * undo structures are not freed when semaphore arrays are destroyed | |
2040 | * so some of them may be out of date. | |
2041 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2042 | * set of adjustments that needs to be done should be done in an atomic | |
2043 | * manner or not. That is, if we are attempting to decrement the semval | |
2044 | * should we queue up and wait until we can do so legally? | |
2045 | * The original implementation attempted to do this (queue and wait). | |
2046 | * The current implementation does not do so. The POSIX standard | |
2047 | * and SVID should be consulted to determine what behavior is mandated. | |
2048 | */ | |
2049 | void exit_sem(struct task_struct *tsk) | |
2050 | { | |
4daa28f6 | 2051 | struct sem_undo_list *ulp; |
1da177e4 | 2052 | |
4daa28f6 MS |
2053 | ulp = tsk->sysvsem.undo_list; |
2054 | if (!ulp) | |
1da177e4 | 2055 | return; |
9edff4ab | 2056 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2057 | |
4daa28f6 | 2058 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2059 | return; |
2060 | ||
380af1b3 | 2061 | for (;;) { |
1da177e4 | 2062 | struct sem_array *sma; |
380af1b3 | 2063 | struct sem_undo *un; |
0a2b9d4c | 2064 | struct list_head tasks; |
6062a8dc | 2065 | int semid, i; |
4daa28f6 | 2066 | |
380af1b3 | 2067 | rcu_read_lock(); |
05725f7e JP |
2068 | un = list_entry_rcu(ulp->list_proc.next, |
2069 | struct sem_undo, list_proc); | |
380af1b3 MS |
2070 | if (&un->list_proc == &ulp->list_proc) |
2071 | semid = -1; | |
2072 | else | |
2073 | semid = un->semid; | |
4daa28f6 | 2074 | |
6062a8dc RR |
2075 | if (semid == -1) { |
2076 | rcu_read_unlock(); | |
380af1b3 | 2077 | break; |
6062a8dc | 2078 | } |
1da177e4 | 2079 | |
6062a8dc | 2080 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid); |
380af1b3 | 2081 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2082 | if (IS_ERR(sma)) { |
2083 | rcu_read_unlock(); | |
380af1b3 | 2084 | continue; |
6062a8dc | 2085 | } |
1da177e4 | 2086 | |
6062a8dc | 2087 | sem_lock(sma, NULL, -1); |
6e224f94 | 2088 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2089 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2090 | sem_unlock(sma, -1); |
2091 | rcu_read_unlock(); | |
2092 | continue; | |
2093 | } | |
bf17bb71 | 2094 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2095 | if (un == NULL) { |
2096 | /* exit_sem raced with IPC_RMID+semget() that created | |
2097 | * exactly the same semid. Nothing to do. | |
2098 | */ | |
6062a8dc | 2099 | sem_unlock(sma, -1); |
6d49dab8 | 2100 | rcu_read_unlock(); |
380af1b3 MS |
2101 | continue; |
2102 | } | |
2103 | ||
2104 | /* remove un from the linked lists */ | |
cf9d5d78 | 2105 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2106 | list_del(&un->list_id); |
2107 | ||
380af1b3 MS |
2108 | spin_lock(&ulp->lock); |
2109 | list_del_rcu(&un->list_proc); | |
2110 | spin_unlock(&ulp->lock); | |
2111 | ||
4daa28f6 MS |
2112 | /* perform adjustments registered in un */ |
2113 | for (i = 0; i < sma->sem_nsems; i++) { | |
239521f3 | 2114 | struct sem *semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
2115 | if (un->semadj[i]) { |
2116 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2117 | /* |
2118 | * Range checks of the new semaphore value, | |
2119 | * not defined by sus: | |
2120 | * - Some unices ignore the undo entirely | |
2121 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2122 | * - some cap the value (e.g. FreeBSD caps | |
2123 | * at 0, but doesn't enforce SEMVMX) | |
2124 | * | |
2125 | * Linux caps the semaphore value, both at 0 | |
2126 | * and at SEMVMX. | |
2127 | * | |
239521f3 | 2128 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2129 | */ |
5f921ae9 IM |
2130 | if (semaphore->semval < 0) |
2131 | semaphore->semval = 0; | |
2132 | if (semaphore->semval > SEMVMX) | |
2133 | semaphore->semval = SEMVMX; | |
b488893a | 2134 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2135 | } |
2136 | } | |
1da177e4 | 2137 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
2138 | INIT_LIST_HEAD(&tasks); |
2139 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 2140 | sem_unlock(sma, -1); |
6d49dab8 | 2141 | rcu_read_unlock(); |
0a2b9d4c | 2142 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 2143 | |
693a8b6e | 2144 | kfree_rcu(un, rcu); |
1da177e4 | 2145 | } |
4daa28f6 | 2146 | kfree(ulp); |
1da177e4 LT |
2147 | } |
2148 | ||
2149 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2150 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2151 | { |
1efdb69b | 2152 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2153 | struct sem_array *sma = it; |
d12e1e50 MS |
2154 | time_t sem_otime; |
2155 | ||
d8c63376 MS |
2156 | /* |
2157 | * The proc interface isn't aware of sem_lock(), it calls | |
2158 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
2159 | * In order to stay compatible with sem_lock(), we must wait until | |
2160 | * all simple semop() calls have left their critical regions. | |
2161 | */ | |
2162 | sem_wait_array(sma); | |
2163 | ||
d12e1e50 | 2164 | sem_otime = get_semotime(sma); |
19b4946c MW |
2165 | |
2166 | return seq_printf(s, | |
b97e820f | 2167 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 2168 | sma->sem_perm.key, |
7ca7e564 | 2169 | sma->sem_perm.id, |
19b4946c MW |
2170 | sma->sem_perm.mode, |
2171 | sma->sem_nsems, | |
1efdb69b EB |
2172 | from_kuid_munged(user_ns, sma->sem_perm.uid), |
2173 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2174 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2175 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
d12e1e50 | 2176 | sem_otime, |
19b4946c | 2177 | sma->sem_ctime); |
1da177e4 LT |
2178 | } |
2179 | #endif |