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: |
f6031d95 | 158 | * a) global sem_lock() for read/write |
1da177e4 | 159 | * sem_undo.id_next, |
758a6ba3 | 160 | * sem_array.complex_count, |
f6031d95 MS |
161 | * sem_array.complex_mode |
162 | * sem_array.pending{_alter,_const}, | |
163 | * sem_array.sem_undo | |
46c0a8ca | 164 | * |
f6031d95 | 165 | * b) global or semaphore sem_lock() for read/write: |
758a6ba3 | 166 | * sem_array.sem_base[i].pending_{const,alter}: |
f6031d95 MS |
167 | * sem_array.complex_mode (for read) |
168 | * | |
169 | * c) special: | |
170 | * sem_undo_list.list_proc: | |
171 | * * undo_list->lock for write | |
172 | * * rcu for read | |
1da177e4 LT |
173 | */ |
174 | ||
e3893534 KK |
175 | #define sc_semmsl sem_ctls[0] |
176 | #define sc_semmns sem_ctls[1] | |
177 | #define sc_semopm sem_ctls[2] | |
178 | #define sc_semmni sem_ctls[3] | |
179 | ||
ed2ddbf8 | 180 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 181 | { |
e3893534 KK |
182 | ns->sc_semmsl = SEMMSL; |
183 | ns->sc_semmns = SEMMNS; | |
184 | ns->sc_semopm = SEMOPM; | |
185 | ns->sc_semmni = SEMMNI; | |
186 | ns->used_sems = 0; | |
ed2ddbf8 | 187 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
188 | } |
189 | ||
ae5e1b22 | 190 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
191 | void sem_exit_ns(struct ipc_namespace *ns) |
192 | { | |
01b8b07a | 193 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 194 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 195 | } |
ae5e1b22 | 196 | #endif |
1da177e4 | 197 | |
239521f3 | 198 | void __init sem_init(void) |
1da177e4 | 199 | { |
ed2ddbf8 | 200 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
201 | ipc_init_proc_interface("sysvipc/sem", |
202 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 203 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
204 | } |
205 | ||
f269f40a MS |
206 | /** |
207 | * unmerge_queues - unmerge queues, if possible. | |
208 | * @sma: semaphore array | |
209 | * | |
210 | * The function unmerges the wait queues if complex_count is 0. | |
211 | * It must be called prior to dropping the global semaphore array lock. | |
212 | */ | |
213 | static void unmerge_queues(struct sem_array *sma) | |
214 | { | |
215 | struct sem_queue *q, *tq; | |
216 | ||
217 | /* complex operations still around? */ | |
218 | if (sma->complex_count) | |
219 | return; | |
220 | /* | |
221 | * We will switch back to simple mode. | |
222 | * Move all pending operation back into the per-semaphore | |
223 | * queues. | |
224 | */ | |
225 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
226 | struct sem *curr; | |
227 | curr = &sma->sem_base[q->sops[0].sem_num]; | |
228 | ||
229 | list_add_tail(&q->list, &curr->pending_alter); | |
230 | } | |
231 | INIT_LIST_HEAD(&sma->pending_alter); | |
232 | } | |
233 | ||
234 | /** | |
8001c858 | 235 | * merge_queues - merge single semop queues into global queue |
f269f40a MS |
236 | * @sma: semaphore array |
237 | * | |
238 | * This function merges all per-semaphore queues into the global queue. | |
239 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
240 | * operations when a multi-semop operation must sleep. | |
241 | * Only the alter operations must be moved, the const operations can stay. | |
242 | */ | |
243 | static void merge_queues(struct sem_array *sma) | |
244 | { | |
245 | int i; | |
246 | for (i = 0; i < sma->sem_nsems; i++) { | |
247 | struct sem *sem = sma->sem_base + i; | |
248 | ||
249 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
250 | } | |
251 | } | |
252 | ||
53dad6d3 DB |
253 | static void sem_rcu_free(struct rcu_head *head) |
254 | { | |
255 | struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu); | |
256 | struct sem_array *sma = ipc_rcu_to_struct(p); | |
257 | ||
258 | security_sem_free(sma); | |
259 | ipc_rcu_free(head); | |
260 | } | |
261 | ||
3ed1f8a9 MS |
262 | /* |
263 | * spin_unlock_wait() and !spin_is_locked() are not memory barriers, they | |
264 | * are only control barriers. | |
265 | * The code must pair with spin_unlock(&sem->lock) or | |
266 | * spin_unlock(&sem_perm.lock), thus just the control barrier is insufficient. | |
267 | * | |
268 | * smp_rmb() is sufficient, as writes cannot pass the control barrier. | |
269 | */ | |
270 | #define ipc_smp_acquire__after_spin_is_unlocked() smp_rmb() | |
271 | ||
5e9d5275 | 272 | /* |
f6031d95 | 273 | * Enter the mode suitable for non-simple operations: |
5e9d5275 | 274 | * Caller must own sem_perm.lock. |
5e9d5275 | 275 | */ |
f6031d95 | 276 | static void complexmode_enter(struct sem_array *sma) |
5e9d5275 MS |
277 | { |
278 | int i; | |
279 | struct sem *sem; | |
280 | ||
f6031d95 MS |
281 | if (sma->complex_mode) { |
282 | /* We are already in complex_mode. Nothing to do */ | |
6d07b68c MS |
283 | return; |
284 | } | |
285 | ||
f6031d95 MS |
286 | /* We need a full barrier after seting complex_mode: |
287 | * The write to complex_mode must be visible | |
288 | * before we read the first sem->lock spinlock state. | |
289 | */ | |
290 | smp_store_mb(sma->complex_mode, true); | |
291 | ||
5e9d5275 MS |
292 | for (i = 0; i < sma->sem_nsems; i++) { |
293 | sem = sma->sem_base + i; | |
294 | spin_unlock_wait(&sem->lock); | |
295 | } | |
3ed1f8a9 | 296 | ipc_smp_acquire__after_spin_is_unlocked(); |
5e9d5275 MS |
297 | } |
298 | ||
f6031d95 MS |
299 | /* |
300 | * Try to leave the mode that disallows simple operations: | |
301 | * Caller must own sem_perm.lock. | |
302 | */ | |
303 | static void complexmode_tryleave(struct sem_array *sma) | |
304 | { | |
305 | if (sma->complex_count) { | |
306 | /* Complex ops are sleeping. | |
307 | * We must stay in complex mode | |
308 | */ | |
309 | return; | |
310 | } | |
311 | /* | |
312 | * Immediately after setting complex_mode to false, | |
313 | * a simple op can start. Thus: all memory writes | |
314 | * performed by the current operation must be visible | |
315 | * before we set complex_mode to false. | |
316 | */ | |
317 | smp_store_release(&sma->complex_mode, false); | |
318 | } | |
319 | ||
320 | #define SEM_GLOBAL_LOCK (-1) | |
6062a8dc RR |
321 | /* |
322 | * If the request contains only one semaphore operation, and there are | |
323 | * no complex transactions pending, lock only the semaphore involved. | |
324 | * Otherwise, lock the entire semaphore array, since we either have | |
325 | * multiple semaphores in our own semops, or we need to look at | |
326 | * semaphores from other pending complex operations. | |
6062a8dc RR |
327 | */ |
328 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
329 | int nsops) | |
330 | { | |
5e9d5275 | 331 | struct sem *sem; |
6062a8dc | 332 | |
5e9d5275 MS |
333 | if (nsops != 1) { |
334 | /* Complex operation - acquire a full lock */ | |
335 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 336 | |
f6031d95 MS |
337 | /* Prevent parallel simple ops */ |
338 | complexmode_enter(sma); | |
339 | return SEM_GLOBAL_LOCK; | |
5e9d5275 MS |
340 | } |
341 | ||
342 | /* | |
343 | * Only one semaphore affected - try to optimize locking. | |
f6031d95 MS |
344 | * Optimized locking is possible if no complex operation |
345 | * is either enqueued or processed right now. | |
346 | * | |
347 | * Both facts are tracked by complex_mode. | |
5e9d5275 MS |
348 | */ |
349 | sem = sma->sem_base + sops->sem_num; | |
6062a8dc | 350 | |
f6031d95 MS |
351 | /* |
352 | * Initial check for complex_mode. Just an optimization, | |
353 | * no locking, no memory barrier. | |
354 | */ | |
355 | if (!sma->complex_mode) { | |
6062a8dc | 356 | /* |
5e9d5275 MS |
357 | * It appears that no complex operation is around. |
358 | * Acquire the per-semaphore lock. | |
6062a8dc | 359 | */ |
5e9d5275 MS |
360 | spin_lock(&sem->lock); |
361 | ||
f6031d95 MS |
362 | /* |
363 | * See 51d7d5205d33 | |
364 | * ("powerpc: Add smp_mb() to arch_spin_is_locked()"): | |
365 | * A full barrier is required: the write of sem->lock | |
366 | * must be visible before the read is executed | |
367 | */ | |
368 | smp_mb(); | |
5e9d5275 | 369 | |
f6031d95 MS |
370 | if (!smp_load_acquire(&sma->complex_mode)) { |
371 | /* fast path successful! */ | |
372 | return sops->sem_num; | |
6062a8dc | 373 | } |
5e9d5275 MS |
374 | spin_unlock(&sem->lock); |
375 | } | |
376 | ||
377 | /* slow path: acquire the full lock */ | |
378 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 379 | |
5e9d5275 MS |
380 | if (sma->complex_count == 0) { |
381 | /* False alarm: | |
382 | * There is no complex operation, thus we can switch | |
383 | * back to the fast path. | |
384 | */ | |
385 | spin_lock(&sem->lock); | |
386 | ipc_unlock_object(&sma->sem_perm); | |
387 | return sops->sem_num; | |
6062a8dc | 388 | } else { |
5e9d5275 MS |
389 | /* Not a false alarm, thus complete the sequence for a |
390 | * full lock. | |
6062a8dc | 391 | */ |
f6031d95 MS |
392 | complexmode_enter(sma); |
393 | return SEM_GLOBAL_LOCK; | |
6062a8dc | 394 | } |
6062a8dc RR |
395 | } |
396 | ||
397 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
398 | { | |
f6031d95 | 399 | if (locknum == SEM_GLOBAL_LOCK) { |
f269f40a | 400 | unmerge_queues(sma); |
f6031d95 | 401 | complexmode_tryleave(sma); |
cf9d5d78 | 402 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc RR |
403 | } else { |
404 | struct sem *sem = sma->sem_base + locknum; | |
405 | spin_unlock(&sem->lock); | |
406 | } | |
6062a8dc RR |
407 | } |
408 | ||
3e148c79 | 409 | /* |
d9a605e4 | 410 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 411 | * is not held. |
321310ce LT |
412 | * |
413 | * The caller holds the RCU read lock. | |
3e148c79 | 414 | */ |
6062a8dc RR |
415 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
416 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 417 | { |
c460b662 RR |
418 | struct kern_ipc_perm *ipcp; |
419 | struct sem_array *sma; | |
03f02c76 | 420 | |
55b7ae50 | 421 | ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
321310ce LT |
422 | if (IS_ERR(ipcp)) |
423 | return ERR_CAST(ipcp); | |
b1ed88b4 | 424 | |
6062a8dc RR |
425 | sma = container_of(ipcp, struct sem_array, sem_perm); |
426 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
427 | |
428 | /* ipc_rmid() may have already freed the ID while sem_lock | |
429 | * was spinning: verify that the structure is still valid | |
430 | */ | |
72a8ff2f | 431 | if (ipc_valid_object(ipcp)) |
c460b662 RR |
432 | return container_of(ipcp, struct sem_array, sem_perm); |
433 | ||
6062a8dc | 434 | sem_unlock(sma, *locknum); |
321310ce | 435 | return ERR_PTR(-EINVAL); |
023a5355 ND |
436 | } |
437 | ||
16df3674 DB |
438 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
439 | { | |
55b7ae50 | 440 | struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
16df3674 DB |
441 | |
442 | if (IS_ERR(ipcp)) | |
443 | return ERR_CAST(ipcp); | |
444 | ||
445 | return container_of(ipcp, struct sem_array, sem_perm); | |
446 | } | |
447 | ||
16df3674 DB |
448 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
449 | int id) | |
450 | { | |
451 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
452 | ||
453 | if (IS_ERR(ipcp)) | |
454 | return ERR_CAST(ipcp); | |
b1ed88b4 | 455 | |
03f02c76 | 456 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
457 | } |
458 | ||
6ff37972 PP |
459 | static inline void sem_lock_and_putref(struct sem_array *sma) |
460 | { | |
6062a8dc | 461 | sem_lock(sma, NULL, -1); |
62659f0b | 462 | ipc_rcu_putref(sma, sem_rcu_free); |
6ff37972 PP |
463 | } |
464 | ||
7ca7e564 ND |
465 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
466 | { | |
467 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
468 | } | |
469 | ||
1da177e4 LT |
470 | /* |
471 | * Lockless wakeup algorithm: | |
472 | * Without the check/retry algorithm a lockless wakeup is possible: | |
473 | * - queue.status is initialized to -EINTR before blocking. | |
474 | * - wakeup is performed by | |
1a82e9e1 | 475 | * * unlinking the queue entry from the pending list |
1da177e4 LT |
476 | * * setting queue.status to IN_WAKEUP |
477 | * This is the notification for the blocked thread that a | |
478 | * result value is imminent. | |
479 | * * call wake_up_process | |
480 | * * set queue.status to the final value. | |
481 | * - the previously blocked thread checks queue.status: | |
239521f3 MS |
482 | * * if it's IN_WAKEUP, then it must wait until the value changes |
483 | * * if it's not -EINTR, then the operation was completed by | |
484 | * update_queue. semtimedop can return queue.status without | |
485 | * performing any operation on the sem array. | |
486 | * * otherwise it must acquire the spinlock and check what's up. | |
1da177e4 LT |
487 | * |
488 | * The two-stage algorithm is necessary to protect against the following | |
489 | * races: | |
490 | * - if queue.status is set after wake_up_process, then the woken up idle | |
491 | * thread could race forward and try (and fail) to acquire sma->lock | |
492 | * before update_queue had a chance to set queue.status | |
493 | * - if queue.status is written before wake_up_process and if the | |
494 | * blocked process is woken up by a signal between writing | |
495 | * queue.status and the wake_up_process, then the woken up | |
496 | * process could return from semtimedop and die by calling | |
497 | * sys_exit before wake_up_process is called. Then wake_up_process | |
498 | * will oops, because the task structure is already invalid. | |
499 | * (yes, this happened on s390 with sysv msg). | |
500 | * | |
501 | */ | |
502 | #define IN_WAKEUP 1 | |
503 | ||
f4566f04 ND |
504 | /** |
505 | * newary - Create a new semaphore set | |
506 | * @ns: namespace | |
507 | * @params: ptr to the structure that contains key, semflg and nsems | |
508 | * | |
d9a605e4 | 509 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 | 510 | */ |
7748dbfa | 511 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
512 | { |
513 | int id; | |
514 | int retval; | |
515 | struct sem_array *sma; | |
516 | int size; | |
7748dbfa ND |
517 | key_t key = params->key; |
518 | int nsems = params->u.nsems; | |
519 | int semflg = params->flg; | |
b97e820f | 520 | int i; |
1da177e4 LT |
521 | |
522 | if (!nsems) | |
523 | return -EINVAL; | |
e3893534 | 524 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
525 | return -ENOSPC; |
526 | ||
239521f3 | 527 | size = sizeof(*sma) + nsems * sizeof(struct sem); |
1da177e4 | 528 | sma = ipc_rcu_alloc(size); |
3ab08fe2 | 529 | if (!sma) |
1da177e4 | 530 | return -ENOMEM; |
3ab08fe2 | 531 | |
239521f3 | 532 | memset(sma, 0, size); |
1da177e4 LT |
533 | |
534 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
535 | sma->sem_perm.key = key; | |
536 | ||
537 | sma->sem_perm.security = NULL; | |
538 | retval = security_sem_alloc(sma); | |
539 | if (retval) { | |
53dad6d3 | 540 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
541 | return retval; |
542 | } | |
543 | ||
1da177e4 | 544 | sma->sem_base = (struct sem *) &sma[1]; |
b97e820f | 545 | |
6062a8dc | 546 | for (i = 0; i < nsems; i++) { |
1a82e9e1 MS |
547 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
548 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
549 | spin_lock_init(&sma->sem_base[i].lock); |
550 | } | |
b97e820f MS |
551 | |
552 | sma->complex_count = 0; | |
f6031d95 | 553 | sma->complex_mode = true; /* dropped by sem_unlock below */ |
1a82e9e1 MS |
554 | INIT_LIST_HEAD(&sma->pending_alter); |
555 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 556 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
557 | sma->sem_nsems = nsems; |
558 | sma->sem_ctime = get_seconds(); | |
e8577d1f MS |
559 | |
560 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); | |
561 | if (id < 0) { | |
562 | ipc_rcu_putref(sma, sem_rcu_free); | |
563 | return id; | |
564 | } | |
565 | ns->used_sems += nsems; | |
566 | ||
6062a8dc | 567 | sem_unlock(sma, -1); |
6d49dab8 | 568 | rcu_read_unlock(); |
1da177e4 | 569 | |
7ca7e564 | 570 | return sma->sem_perm.id; |
1da177e4 LT |
571 | } |
572 | ||
7748dbfa | 573 | |
f4566f04 | 574 | /* |
d9a605e4 | 575 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 576 | */ |
03f02c76 | 577 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 578 | { |
03f02c76 ND |
579 | struct sem_array *sma; |
580 | ||
581 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
582 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
583 | } |
584 | ||
f4566f04 | 585 | /* |
d9a605e4 | 586 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 587 | */ |
03f02c76 ND |
588 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
589 | struct ipc_params *params) | |
7748dbfa | 590 | { |
03f02c76 ND |
591 | struct sem_array *sma; |
592 | ||
593 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
594 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
595 | return -EINVAL; |
596 | ||
597 | return 0; | |
598 | } | |
599 | ||
d5460c99 | 600 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 601 | { |
e3893534 | 602 | struct ipc_namespace *ns; |
eb66ec44 MK |
603 | static const struct ipc_ops sem_ops = { |
604 | .getnew = newary, | |
605 | .associate = sem_security, | |
606 | .more_checks = sem_more_checks, | |
607 | }; | |
7748dbfa | 608 | struct ipc_params sem_params; |
e3893534 KK |
609 | |
610 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 611 | |
e3893534 | 612 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 613 | return -EINVAL; |
7ca7e564 | 614 | |
7748dbfa ND |
615 | sem_params.key = key; |
616 | sem_params.flg = semflg; | |
617 | sem_params.u.nsems = nsems; | |
1da177e4 | 618 | |
7748dbfa | 619 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
620 | } |
621 | ||
78f5009c PM |
622 | /** |
623 | * perform_atomic_semop - Perform (if possible) a semaphore operation | |
758a6ba3 | 624 | * @sma: semaphore array |
d198cd6d | 625 | * @q: struct sem_queue that describes the operation |
758a6ba3 MS |
626 | * |
627 | * Returns 0 if the operation was possible. | |
628 | * Returns 1 if the operation is impossible, the caller must sleep. | |
629 | * Negative values are error codes. | |
1da177e4 | 630 | */ |
d198cd6d | 631 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) |
1da177e4 | 632 | { |
d198cd6d | 633 | int result, sem_op, nsops, pid; |
1da177e4 | 634 | struct sembuf *sop; |
239521f3 | 635 | struct sem *curr; |
d198cd6d MS |
636 | struct sembuf *sops; |
637 | struct sem_undo *un; | |
638 | ||
639 | sops = q->sops; | |
640 | nsops = q->nsops; | |
641 | un = q->undo; | |
1da177e4 LT |
642 | |
643 | for (sop = sops; sop < sops + nsops; sop++) { | |
644 | curr = sma->sem_base + sop->sem_num; | |
645 | sem_op = sop->sem_op; | |
646 | result = curr->semval; | |
78f5009c | 647 | |
1da177e4 LT |
648 | if (!sem_op && result) |
649 | goto would_block; | |
650 | ||
651 | result += sem_op; | |
652 | if (result < 0) | |
653 | goto would_block; | |
654 | if (result > SEMVMX) | |
655 | goto out_of_range; | |
78f5009c | 656 | |
1da177e4 LT |
657 | if (sop->sem_flg & SEM_UNDO) { |
658 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 659 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
660 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
661 | goto out_of_range; | |
78f5009c | 662 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 663 | } |
78f5009c | 664 | |
1da177e4 LT |
665 | curr->semval = result; |
666 | } | |
667 | ||
668 | sop--; | |
d198cd6d | 669 | pid = q->pid; |
1da177e4 LT |
670 | while (sop >= sops) { |
671 | sma->sem_base[sop->sem_num].sempid = pid; | |
1da177e4 LT |
672 | sop--; |
673 | } | |
78f5009c | 674 | |
1da177e4 LT |
675 | return 0; |
676 | ||
677 | out_of_range: | |
678 | result = -ERANGE; | |
679 | goto undo; | |
680 | ||
681 | would_block: | |
ed247b7c MS |
682 | q->blocking = sop; |
683 | ||
1da177e4 LT |
684 | if (sop->sem_flg & IPC_NOWAIT) |
685 | result = -EAGAIN; | |
686 | else | |
687 | result = 1; | |
688 | ||
689 | undo: | |
690 | sop--; | |
691 | while (sop >= sops) { | |
78f5009c PM |
692 | sem_op = sop->sem_op; |
693 | sma->sem_base[sop->sem_num].semval -= sem_op; | |
694 | if (sop->sem_flg & SEM_UNDO) | |
695 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
696 | sop--; |
697 | } | |
698 | ||
699 | return result; | |
700 | } | |
701 | ||
0a2b9d4c MS |
702 | /** wake_up_sem_queue_prepare(q, error): Prepare wake-up |
703 | * @q: queue entry that must be signaled | |
704 | * @error: Error value for the signal | |
705 | * | |
706 | * Prepare the wake-up of the queue entry q. | |
d4212093 | 707 | */ |
0a2b9d4c MS |
708 | static void wake_up_sem_queue_prepare(struct list_head *pt, |
709 | struct sem_queue *q, int error) | |
d4212093 | 710 | { |
0a2b9d4c MS |
711 | if (list_empty(pt)) { |
712 | /* | |
713 | * Hold preempt off so that we don't get preempted and have the | |
714 | * wakee busy-wait until we're scheduled back on. | |
715 | */ | |
716 | preempt_disable(); | |
717 | } | |
d4212093 | 718 | q->status = IN_WAKEUP; |
0a2b9d4c MS |
719 | q->pid = error; |
720 | ||
9f1bc2c9 | 721 | list_add_tail(&q->list, pt); |
0a2b9d4c MS |
722 | } |
723 | ||
724 | /** | |
8001c858 | 725 | * wake_up_sem_queue_do - do the actual wake-up |
0a2b9d4c MS |
726 | * @pt: list of tasks to be woken up |
727 | * | |
728 | * Do the actual wake-up. | |
729 | * The function is called without any locks held, thus the semaphore array | |
730 | * could be destroyed already and the tasks can disappear as soon as the | |
731 | * status is set to the actual return code. | |
732 | */ | |
733 | static void wake_up_sem_queue_do(struct list_head *pt) | |
734 | { | |
735 | struct sem_queue *q, *t; | |
736 | int did_something; | |
737 | ||
738 | did_something = !list_empty(pt); | |
9f1bc2c9 | 739 | list_for_each_entry_safe(q, t, pt, list) { |
0a2b9d4c MS |
740 | wake_up_process(q->sleeper); |
741 | /* q can disappear immediately after writing q->status. */ | |
742 | smp_wmb(); | |
743 | q->status = q->pid; | |
744 | } | |
745 | if (did_something) | |
746 | preempt_enable(); | |
d4212093 NP |
747 | } |
748 | ||
b97e820f MS |
749 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
750 | { | |
751 | list_del(&q->list); | |
9f1bc2c9 | 752 | if (q->nsops > 1) |
b97e820f MS |
753 | sma->complex_count--; |
754 | } | |
755 | ||
fd5db422 MS |
756 | /** check_restart(sma, q) |
757 | * @sma: semaphore array | |
758 | * @q: the operation that just completed | |
759 | * | |
760 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
761 | * waiting operations. Therefore this function checks if the restart is | |
762 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
763 | * modified the array. |
764 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 MS |
765 | */ |
766 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
767 | { | |
1a82e9e1 MS |
768 | /* pending complex alter operations are too difficult to analyse */ |
769 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
770 | return 1; |
771 | ||
772 | /* we were a sleeping complex operation. Too difficult */ | |
773 | if (q->nsops > 1) | |
774 | return 1; | |
775 | ||
1a82e9e1 MS |
776 | /* It is impossible that someone waits for the new value: |
777 | * - complex operations always restart. | |
778 | * - wait-for-zero are handled seperately. | |
779 | * - q is a previously sleeping simple operation that | |
780 | * altered the array. It must be a decrement, because | |
781 | * simple increments never sleep. | |
782 | * - If there are older (higher priority) decrements | |
783 | * in the queue, then they have observed the original | |
784 | * semval value and couldn't proceed. The operation | |
785 | * decremented to value - thus they won't proceed either. | |
786 | */ | |
787 | return 0; | |
788 | } | |
fd5db422 | 789 | |
1a82e9e1 | 790 | /** |
8001c858 | 791 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
792 | * @sma: semaphore array. |
793 | * @semnum: semaphore that was modified. | |
794 | * @pt: list head for the tasks that must be woken up. | |
795 | * | |
796 | * wake_const_ops must be called after a semaphore in a semaphore array | |
797 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
798 | * be called with semnum = -1, as well as with the number of each modified | |
799 | * semaphore. | |
800 | * The tasks that must be woken up are added to @pt. The return code | |
801 | * is stored in q->pid. | |
802 | * The function returns 1 if at least one operation was completed successfully. | |
803 | */ | |
804 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
805 | struct list_head *pt) | |
806 | { | |
807 | struct sem_queue *q; | |
808 | struct list_head *walk; | |
809 | struct list_head *pending_list; | |
810 | int semop_completed = 0; | |
811 | ||
812 | if (semnum == -1) | |
813 | pending_list = &sma->pending_const; | |
814 | else | |
815 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 816 | |
1a82e9e1 MS |
817 | walk = pending_list->next; |
818 | while (walk != pending_list) { | |
819 | int error; | |
820 | ||
821 | q = container_of(walk, struct sem_queue, list); | |
822 | walk = walk->next; | |
823 | ||
d198cd6d | 824 | error = perform_atomic_semop(sma, q); |
1a82e9e1 MS |
825 | |
826 | if (error <= 0) { | |
827 | /* operation completed, remove from queue & wakeup */ | |
828 | ||
829 | unlink_queue(sma, q); | |
830 | ||
831 | wake_up_sem_queue_prepare(pt, q, error); | |
832 | if (error == 0) | |
833 | semop_completed = 1; | |
834 | } | |
835 | } | |
836 | return semop_completed; | |
837 | } | |
838 | ||
839 | /** | |
8001c858 | 840 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
841 | * @sma: semaphore array |
842 | * @sops: operations that were performed | |
843 | * @nsops: number of operations | |
844 | * @pt: list head of the tasks that must be woken up. | |
845 | * | |
8001c858 DB |
846 | * Checks all required queue for wait-for-zero operations, based |
847 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
848 | * The function returns 1 if at least one operation was completed successfully. |
849 | */ | |
850 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
851 | int nsops, struct list_head *pt) | |
852 | { | |
853 | int i; | |
854 | int semop_completed = 0; | |
855 | int got_zero = 0; | |
856 | ||
857 | /* first: the per-semaphore queues, if known */ | |
858 | if (sops) { | |
859 | for (i = 0; i < nsops; i++) { | |
860 | int num = sops[i].sem_num; | |
861 | ||
862 | if (sma->sem_base[num].semval == 0) { | |
863 | got_zero = 1; | |
864 | semop_completed |= wake_const_ops(sma, num, pt); | |
865 | } | |
866 | } | |
867 | } else { | |
868 | /* | |
869 | * No sops means modified semaphores not known. | |
870 | * Assume all were changed. | |
fd5db422 | 871 | */ |
1a82e9e1 MS |
872 | for (i = 0; i < sma->sem_nsems; i++) { |
873 | if (sma->sem_base[i].semval == 0) { | |
874 | got_zero = 1; | |
875 | semop_completed |= wake_const_ops(sma, i, pt); | |
876 | } | |
877 | } | |
fd5db422 MS |
878 | } |
879 | /* | |
1a82e9e1 MS |
880 | * If one of the modified semaphores got 0, |
881 | * then check the global queue, too. | |
fd5db422 | 882 | */ |
1a82e9e1 MS |
883 | if (got_zero) |
884 | semop_completed |= wake_const_ops(sma, -1, pt); | |
fd5db422 | 885 | |
1a82e9e1 | 886 | return semop_completed; |
fd5db422 MS |
887 | } |
888 | ||
636c6be8 MS |
889 | |
890 | /** | |
8001c858 | 891 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
892 | * @sma: semaphore array. |
893 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 894 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
895 | * |
896 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
897 | * was modified. If multiple semaphores were modified, update_queue must |
898 | * be called with semnum = -1, as well as with the number of each modified | |
899 | * semaphore. | |
0a2b9d4c MS |
900 | * The tasks that must be woken up are added to @pt. The return code |
901 | * is stored in q->pid. | |
1a82e9e1 MS |
902 | * The function internally checks if const operations can now succeed. |
903 | * | |
0a2b9d4c | 904 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 905 | */ |
0a2b9d4c | 906 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 907 | { |
636c6be8 MS |
908 | struct sem_queue *q; |
909 | struct list_head *walk; | |
910 | struct list_head *pending_list; | |
0a2b9d4c | 911 | int semop_completed = 0; |
636c6be8 | 912 | |
9f1bc2c9 | 913 | if (semnum == -1) |
1a82e9e1 | 914 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 915 | else |
1a82e9e1 | 916 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
917 | |
918 | again: | |
636c6be8 MS |
919 | walk = pending_list->next; |
920 | while (walk != pending_list) { | |
fd5db422 | 921 | int error, restart; |
636c6be8 | 922 | |
9f1bc2c9 | 923 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 924 | walk = walk->next; |
1da177e4 | 925 | |
d987f8b2 MS |
926 | /* If we are scanning the single sop, per-semaphore list of |
927 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 928 | * necessary to scan further: simple increments |
d987f8b2 MS |
929 | * that affect only one entry succeed immediately and cannot |
930 | * be in the per semaphore pending queue, and decrements | |
931 | * cannot be successful if the value is already 0. | |
932 | */ | |
1a82e9e1 | 933 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
934 | break; |
935 | ||
d198cd6d | 936 | error = perform_atomic_semop(sma, q); |
1da177e4 LT |
937 | |
938 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
939 | if (error > 0) |
940 | continue; | |
941 | ||
b97e820f | 942 | unlink_queue(sma, q); |
9cad200c | 943 | |
0a2b9d4c | 944 | if (error) { |
fd5db422 | 945 | restart = 0; |
0a2b9d4c MS |
946 | } else { |
947 | semop_completed = 1; | |
1a82e9e1 | 948 | do_smart_wakeup_zero(sma, q->sops, q->nsops, pt); |
fd5db422 | 949 | restart = check_restart(sma, q); |
0a2b9d4c | 950 | } |
fd5db422 | 951 | |
0a2b9d4c | 952 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 953 | if (restart) |
9cad200c | 954 | goto again; |
1da177e4 | 955 | } |
0a2b9d4c | 956 | return semop_completed; |
1da177e4 LT |
957 | } |
958 | ||
0e8c6656 | 959 | /** |
8001c858 | 960 | * set_semotime - set sem_otime |
0e8c6656 MS |
961 | * @sma: semaphore array |
962 | * @sops: operations that modified the array, may be NULL | |
963 | * | |
964 | * sem_otime is replicated to avoid cache line trashing. | |
965 | * This function sets one instance to the current time. | |
966 | */ | |
967 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
968 | { | |
969 | if (sops == NULL) { | |
970 | sma->sem_base[0].sem_otime = get_seconds(); | |
971 | } else { | |
972 | sma->sem_base[sops[0].sem_num].sem_otime = | |
973 | get_seconds(); | |
974 | } | |
975 | } | |
976 | ||
0a2b9d4c | 977 | /** |
8001c858 | 978 | * do_smart_update - optimized update_queue |
fd5db422 MS |
979 | * @sma: semaphore array |
980 | * @sops: operations that were performed | |
981 | * @nsops: number of operations | |
0a2b9d4c MS |
982 | * @otime: force setting otime |
983 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 | 984 | * |
1a82e9e1 MS |
985 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
986 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
987 | * Note that the function does not do the actual wake-up: the caller is |
988 | * responsible for calling wake_up_sem_queue_do(@pt). | |
989 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 990 | */ |
0a2b9d4c MS |
991 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
992 | int otime, struct list_head *pt) | |
fd5db422 MS |
993 | { |
994 | int i; | |
995 | ||
1a82e9e1 MS |
996 | otime |= do_smart_wakeup_zero(sma, sops, nsops, pt); |
997 | ||
f269f40a MS |
998 | if (!list_empty(&sma->pending_alter)) { |
999 | /* semaphore array uses the global queue - just process it. */ | |
1000 | otime |= update_queue(sma, -1, pt); | |
1001 | } else { | |
1002 | if (!sops) { | |
1003 | /* | |
1004 | * No sops, thus the modified semaphores are not | |
1005 | * known. Check all. | |
1006 | */ | |
1007 | for (i = 0; i < sma->sem_nsems; i++) | |
1008 | otime |= update_queue(sma, i, pt); | |
1009 | } else { | |
1010 | /* | |
1011 | * Check the semaphores that were increased: | |
1012 | * - No complex ops, thus all sleeping ops are | |
1013 | * decrease. | |
1014 | * - if we decreased the value, then any sleeping | |
1015 | * semaphore ops wont be able to run: If the | |
1016 | * previous value was too small, then the new | |
1017 | * value will be too small, too. | |
1018 | */ | |
1019 | for (i = 0; i < nsops; i++) { | |
1020 | if (sops[i].sem_op > 0) { | |
1021 | otime |= update_queue(sma, | |
1022 | sops[i].sem_num, pt); | |
1023 | } | |
ab465df9 | 1024 | } |
9f1bc2c9 | 1025 | } |
fd5db422 | 1026 | } |
0e8c6656 MS |
1027 | if (otime) |
1028 | set_semotime(sma, sops); | |
fd5db422 MS |
1029 | } |
1030 | ||
2f2ed41d | 1031 | /* |
b220c57a | 1032 | * check_qop: Test if a queued operation sleeps on the semaphore semnum |
2f2ed41d MS |
1033 | */ |
1034 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1035 | bool count_zero) | |
1036 | { | |
b220c57a | 1037 | struct sembuf *sop = q->blocking; |
2f2ed41d | 1038 | |
9b44ee2e MS |
1039 | /* |
1040 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1041 | * semaphores. This violates SUS, therefore it was changed to the | |
1042 | * standard compliant behavior. | |
1043 | * Give the administrators a chance to notice that an application | |
1044 | * might misbehave because it relies on the Linux behavior. | |
1045 | */ | |
1046 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1047 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1048 | current->comm, task_pid_nr(current)); | |
1049 | ||
b220c57a MS |
1050 | if (sop->sem_num != semnum) |
1051 | return 0; | |
2f2ed41d | 1052 | |
b220c57a MS |
1053 | if (count_zero && sop->sem_op == 0) |
1054 | return 1; | |
1055 | if (!count_zero && sop->sem_op < 0) | |
1056 | return 1; | |
1057 | ||
1058 | return 0; | |
2f2ed41d MS |
1059 | } |
1060 | ||
1da177e4 LT |
1061 | /* The following counts are associated to each semaphore: |
1062 | * semncnt number of tasks waiting on semval being nonzero | |
1063 | * semzcnt number of tasks waiting on semval being zero | |
b220c57a MS |
1064 | * |
1065 | * Per definition, a task waits only on the semaphore of the first semop | |
1066 | * that cannot proceed, even if additional operation would block, too. | |
1da177e4 | 1067 | */ |
2f2ed41d MS |
1068 | static int count_semcnt(struct sem_array *sma, ushort semnum, |
1069 | bool count_zero) | |
1da177e4 | 1070 | { |
2f2ed41d | 1071 | struct list_head *l; |
239521f3 | 1072 | struct sem_queue *q; |
2f2ed41d | 1073 | int semcnt; |
1da177e4 | 1074 | |
2f2ed41d MS |
1075 | semcnt = 0; |
1076 | /* First: check the simple operations. They are easy to evaluate */ | |
1077 | if (count_zero) | |
1078 | l = &sma->sem_base[semnum].pending_const; | |
1079 | else | |
1080 | l = &sma->sem_base[semnum].pending_alter; | |
1da177e4 | 1081 | |
2f2ed41d MS |
1082 | list_for_each_entry(q, l, list) { |
1083 | /* all task on a per-semaphore list sleep on exactly | |
1084 | * that semaphore | |
1085 | */ | |
1086 | semcnt++; | |
ebc2e5e6 RR |
1087 | } |
1088 | ||
2f2ed41d | 1089 | /* Then: check the complex operations. */ |
1994862d | 1090 | list_for_each_entry(q, &sma->pending_alter, list) { |
2f2ed41d MS |
1091 | semcnt += check_qop(sma, semnum, q, count_zero); |
1092 | } | |
1093 | if (count_zero) { | |
1094 | list_for_each_entry(q, &sma->pending_const, list) { | |
1095 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1096 | } | |
1994862d | 1097 | } |
2f2ed41d | 1098 | return semcnt; |
1da177e4 LT |
1099 | } |
1100 | ||
d9a605e4 DB |
1101 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1102 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1103 | * remains locked on exit. |
1da177e4 | 1104 | */ |
01b8b07a | 1105 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1106 | { |
380af1b3 MS |
1107 | struct sem_undo *un, *tu; |
1108 | struct sem_queue *q, *tq; | |
01b8b07a | 1109 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 1110 | struct list_head tasks; |
9f1bc2c9 | 1111 | int i; |
1da177e4 | 1112 | |
380af1b3 | 1113 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1114 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1115 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1116 | list_del(&un->list_id); | |
1117 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1118 | un->semid = -1; |
380af1b3 MS |
1119 | list_del_rcu(&un->list_proc); |
1120 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1121 | kfree_rcu(un, rcu); |
380af1b3 | 1122 | } |
1da177e4 LT |
1123 | |
1124 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 1125 | INIT_LIST_HEAD(&tasks); |
1a82e9e1 MS |
1126 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1127 | unlink_queue(sma, q); | |
1128 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1129 | } | |
1130 | ||
1131 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1132 | unlink_queue(sma, q); |
0a2b9d4c | 1133 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 1134 | } |
9f1bc2c9 RR |
1135 | for (i = 0; i < sma->sem_nsems; i++) { |
1136 | struct sem *sem = sma->sem_base + i; | |
1a82e9e1 MS |
1137 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1138 | unlink_queue(sma, q); | |
1139 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1140 | } | |
1141 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 RR |
1142 | unlink_queue(sma, q); |
1143 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
1144 | } | |
1145 | } | |
1da177e4 | 1146 | |
7ca7e564 ND |
1147 | /* Remove the semaphore set from the IDR */ |
1148 | sem_rmid(ns, sma); | |
6062a8dc | 1149 | sem_unlock(sma, -1); |
6d49dab8 | 1150 | rcu_read_unlock(); |
1da177e4 | 1151 | |
0a2b9d4c | 1152 | wake_up_sem_queue_do(&tasks); |
e3893534 | 1153 | ns->used_sems -= sma->sem_nsems; |
53dad6d3 | 1154 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1155 | } |
1156 | ||
1157 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1158 | { | |
239521f3 | 1159 | switch (version) { |
1da177e4 LT |
1160 | case IPC_64: |
1161 | return copy_to_user(buf, in, sizeof(*in)); | |
1162 | case IPC_OLD: | |
1163 | { | |
1164 | struct semid_ds out; | |
1165 | ||
982f7c2b DR |
1166 | memset(&out, 0, sizeof(out)); |
1167 | ||
1da177e4 LT |
1168 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1169 | ||
1170 | out.sem_otime = in->sem_otime; | |
1171 | out.sem_ctime = in->sem_ctime; | |
1172 | out.sem_nsems = in->sem_nsems; | |
1173 | ||
1174 | return copy_to_user(buf, &out, sizeof(out)); | |
1175 | } | |
1176 | default: | |
1177 | return -EINVAL; | |
1178 | } | |
1179 | } | |
1180 | ||
d12e1e50 MS |
1181 | static time_t get_semotime(struct sem_array *sma) |
1182 | { | |
1183 | int i; | |
1184 | time_t res; | |
1185 | ||
1186 | res = sma->sem_base[0].sem_otime; | |
1187 | for (i = 1; i < sma->sem_nsems; i++) { | |
1188 | time_t to = sma->sem_base[i].sem_otime; | |
1189 | ||
1190 | if (to > res) | |
1191 | res = to; | |
1192 | } | |
1193 | return res; | |
1194 | } | |
1195 | ||
4b9fcb0e | 1196 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1197 | int cmd, int version, void __user *p) |
1da177e4 | 1198 | { |
e5cc9c7b | 1199 | int err; |
1da177e4 LT |
1200 | struct sem_array *sma; |
1201 | ||
239521f3 | 1202 | switch (cmd) { |
1da177e4 LT |
1203 | case IPC_INFO: |
1204 | case SEM_INFO: | |
1205 | { | |
1206 | struct seminfo seminfo; | |
1207 | int max_id; | |
1208 | ||
1209 | err = security_sem_semctl(NULL, cmd); | |
1210 | if (err) | |
1211 | return err; | |
46c0a8ca | 1212 | |
239521f3 | 1213 | memset(&seminfo, 0, sizeof(seminfo)); |
e3893534 KK |
1214 | seminfo.semmni = ns->sc_semmni; |
1215 | seminfo.semmns = ns->sc_semmns; | |
1216 | seminfo.semmsl = ns->sc_semmsl; | |
1217 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1218 | seminfo.semvmx = SEMVMX; |
1219 | seminfo.semmnu = SEMMNU; | |
1220 | seminfo.semmap = SEMMAP; | |
1221 | seminfo.semume = SEMUME; | |
d9a605e4 | 1222 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1223 | if (cmd == SEM_INFO) { |
e3893534 KK |
1224 | seminfo.semusz = sem_ids(ns).in_use; |
1225 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1226 | } else { |
1227 | seminfo.semusz = SEMUSZ; | |
1228 | seminfo.semaem = SEMAEM; | |
1229 | } | |
7ca7e564 | 1230 | max_id = ipc_get_maxid(&sem_ids(ns)); |
d9a605e4 | 1231 | up_read(&sem_ids(ns).rwsem); |
46c0a8ca | 1232 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 | 1233 | return -EFAULT; |
239521f3 | 1234 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 1235 | } |
4b9fcb0e | 1236 | case IPC_STAT: |
1da177e4 LT |
1237 | case SEM_STAT: |
1238 | { | |
1239 | struct semid64_ds tbuf; | |
16df3674 DB |
1240 | int id = 0; |
1241 | ||
1242 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1243 | |
941b0304 | 1244 | rcu_read_lock(); |
4b9fcb0e | 1245 | if (cmd == SEM_STAT) { |
16df3674 DB |
1246 | sma = sem_obtain_object(ns, semid); |
1247 | if (IS_ERR(sma)) { | |
1248 | err = PTR_ERR(sma); | |
1249 | goto out_unlock; | |
1250 | } | |
4b9fcb0e PP |
1251 | id = sma->sem_perm.id; |
1252 | } else { | |
16df3674 DB |
1253 | sma = sem_obtain_object_check(ns, semid); |
1254 | if (IS_ERR(sma)) { | |
1255 | err = PTR_ERR(sma); | |
1256 | goto out_unlock; | |
1257 | } | |
4b9fcb0e | 1258 | } |
1da177e4 LT |
1259 | |
1260 | err = -EACCES; | |
b0e77598 | 1261 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1262 | goto out_unlock; |
1263 | ||
1264 | err = security_sem_semctl(sma, cmd); | |
1265 | if (err) | |
1266 | goto out_unlock; | |
1267 | ||
1da177e4 | 1268 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
d12e1e50 MS |
1269 | tbuf.sem_otime = get_semotime(sma); |
1270 | tbuf.sem_ctime = sma->sem_ctime; | |
1271 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1272 | rcu_read_unlock(); |
e1fd1f49 | 1273 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1274 | return -EFAULT; |
1275 | return id; | |
1276 | } | |
1277 | default: | |
1278 | return -EINVAL; | |
1279 | } | |
1da177e4 | 1280 | out_unlock: |
16df3674 | 1281 | rcu_read_unlock(); |
1da177e4 LT |
1282 | return err; |
1283 | } | |
1284 | ||
e1fd1f49 AV |
1285 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1286 | unsigned long arg) | |
1287 | { | |
1288 | struct sem_undo *un; | |
1289 | struct sem_array *sma; | |
239521f3 | 1290 | struct sem *curr; |
e1fd1f49 | 1291 | int err; |
e1fd1f49 AV |
1292 | struct list_head tasks; |
1293 | int val; | |
1294 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1295 | /* big-endian 64bit */ | |
1296 | val = arg >> 32; | |
1297 | #else | |
1298 | /* 32bit or little-endian 64bit */ | |
1299 | val = arg; | |
1300 | #endif | |
1301 | ||
6062a8dc RR |
1302 | if (val > SEMVMX || val < 0) |
1303 | return -ERANGE; | |
e1fd1f49 AV |
1304 | |
1305 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1306 | |
6062a8dc RR |
1307 | rcu_read_lock(); |
1308 | sma = sem_obtain_object_check(ns, semid); | |
1309 | if (IS_ERR(sma)) { | |
1310 | rcu_read_unlock(); | |
1311 | return PTR_ERR(sma); | |
1312 | } | |
1313 | ||
1314 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1315 | rcu_read_unlock(); | |
1316 | return -EINVAL; | |
1317 | } | |
1318 | ||
1319 | ||
1320 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1321 | rcu_read_unlock(); | |
1322 | return -EACCES; | |
1323 | } | |
e1fd1f49 AV |
1324 | |
1325 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1326 | if (err) { |
1327 | rcu_read_unlock(); | |
1328 | return -EACCES; | |
1329 | } | |
e1fd1f49 | 1330 | |
6062a8dc | 1331 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1332 | |
0f3d2b01 | 1333 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1334 | sem_unlock(sma, -1); |
1335 | rcu_read_unlock(); | |
1336 | return -EIDRM; | |
1337 | } | |
1338 | ||
e1fd1f49 AV |
1339 | curr = &sma->sem_base[semnum]; |
1340 | ||
cf9d5d78 | 1341 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1342 | list_for_each_entry(un, &sma->list_id, list_id) |
1343 | un->semadj[semnum] = 0; | |
1344 | ||
1345 | curr->semval = val; | |
1346 | curr->sempid = task_tgid_vnr(current); | |
1347 | sma->sem_ctime = get_seconds(); | |
1348 | /* maybe some queued-up processes were waiting for this */ | |
1349 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1350 | sem_unlock(sma, -1); |
6d49dab8 | 1351 | rcu_read_unlock(); |
e1fd1f49 | 1352 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1353 | return 0; |
e1fd1f49 AV |
1354 | } |
1355 | ||
e3893534 | 1356 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1357 | int cmd, void __user *p) |
1da177e4 LT |
1358 | { |
1359 | struct sem_array *sma; | |
239521f3 | 1360 | struct sem *curr; |
16df3674 | 1361 | int err, nsems; |
1da177e4 | 1362 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1363 | ushort *sem_io = fast_sem_io; |
0a2b9d4c | 1364 | struct list_head tasks; |
1da177e4 | 1365 | |
16df3674 DB |
1366 | INIT_LIST_HEAD(&tasks); |
1367 | ||
1368 | rcu_read_lock(); | |
1369 | sma = sem_obtain_object_check(ns, semid); | |
1370 | if (IS_ERR(sma)) { | |
1371 | rcu_read_unlock(); | |
023a5355 | 1372 | return PTR_ERR(sma); |
16df3674 | 1373 | } |
1da177e4 LT |
1374 | |
1375 | nsems = sma->sem_nsems; | |
1376 | ||
1da177e4 | 1377 | err = -EACCES; |
c728b9c8 LT |
1378 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1379 | goto out_rcu_wakeup; | |
1da177e4 LT |
1380 | |
1381 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1382 | if (err) |
1383 | goto out_rcu_wakeup; | |
1da177e4 LT |
1384 | |
1385 | err = -EACCES; | |
1386 | switch (cmd) { | |
1387 | case GETALL: | |
1388 | { | |
e1fd1f49 | 1389 | ushort __user *array = p; |
1da177e4 LT |
1390 | int i; |
1391 | ||
ce857229 | 1392 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1393 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1394 | err = -EIDRM; |
1395 | goto out_unlock; | |
1396 | } | |
239521f3 | 1397 | if (nsems > SEMMSL_FAST) { |
ce857229 | 1398 | if (!ipc_rcu_getref(sma)) { |
ce857229 | 1399 | err = -EIDRM; |
6e224f94 | 1400 | goto out_unlock; |
ce857229 AV |
1401 | } |
1402 | sem_unlock(sma, -1); | |
6d49dab8 | 1403 | rcu_read_unlock(); |
1da177e4 | 1404 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1405 | if (sem_io == NULL) { |
62659f0b | 1406 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1407 | return -ENOMEM; |
1408 | } | |
1409 | ||
4091fd94 | 1410 | rcu_read_lock(); |
6ff37972 | 1411 | sem_lock_and_putref(sma); |
0f3d2b01 | 1412 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1413 | err = -EIDRM; |
6e224f94 | 1414 | goto out_unlock; |
1da177e4 | 1415 | } |
ce857229 | 1416 | } |
1da177e4 LT |
1417 | for (i = 0; i < sma->sem_nsems; i++) |
1418 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1419 | sem_unlock(sma, -1); |
6d49dab8 | 1420 | rcu_read_unlock(); |
1da177e4 | 1421 | err = 0; |
239521f3 | 1422 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1423 | err = -EFAULT; |
1424 | goto out_free; | |
1425 | } | |
1426 | case SETALL: | |
1427 | { | |
1428 | int i; | |
1429 | struct sem_undo *un; | |
1430 | ||
6062a8dc | 1431 | if (!ipc_rcu_getref(sma)) { |
6e224f94 MS |
1432 | err = -EIDRM; |
1433 | goto out_rcu_wakeup; | |
6062a8dc | 1434 | } |
16df3674 | 1435 | rcu_read_unlock(); |
1da177e4 | 1436 | |
239521f3 | 1437 | if (nsems > SEMMSL_FAST) { |
1da177e4 | 1438 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1439 | if (sem_io == NULL) { |
62659f0b | 1440 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1441 | return -ENOMEM; |
1442 | } | |
1443 | } | |
1444 | ||
239521f3 | 1445 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
62659f0b | 1446 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1447 | err = -EFAULT; |
1448 | goto out_free; | |
1449 | } | |
1450 | ||
1451 | for (i = 0; i < nsems; i++) { | |
1452 | if (sem_io[i] > SEMVMX) { | |
62659f0b | 1453 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1454 | err = -ERANGE; |
1455 | goto out_free; | |
1456 | } | |
1457 | } | |
4091fd94 | 1458 | rcu_read_lock(); |
6ff37972 | 1459 | sem_lock_and_putref(sma); |
0f3d2b01 | 1460 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1461 | err = -EIDRM; |
6e224f94 | 1462 | goto out_unlock; |
1da177e4 LT |
1463 | } |
1464 | ||
1465 | for (i = 0; i < nsems; i++) | |
1466 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 | 1467 | |
cf9d5d78 | 1468 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1469 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1470 | for (i = 0; i < nsems; i++) |
1471 | un->semadj[i] = 0; | |
4daa28f6 | 1472 | } |
1da177e4 LT |
1473 | sma->sem_ctime = get_seconds(); |
1474 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1475 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1476 | err = 0; |
1477 | goto out_unlock; | |
1478 | } | |
e1fd1f49 | 1479 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1480 | } |
1481 | err = -EINVAL; | |
c728b9c8 LT |
1482 | if (semnum < 0 || semnum >= nsems) |
1483 | goto out_rcu_wakeup; | |
1da177e4 | 1484 | |
6062a8dc | 1485 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1486 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1487 | err = -EIDRM; |
1488 | goto out_unlock; | |
1489 | } | |
1da177e4 LT |
1490 | curr = &sma->sem_base[semnum]; |
1491 | ||
1492 | switch (cmd) { | |
1493 | case GETVAL: | |
1494 | err = curr->semval; | |
1495 | goto out_unlock; | |
1496 | case GETPID: | |
1497 | err = curr->sempid; | |
1498 | goto out_unlock; | |
1499 | case GETNCNT: | |
2f2ed41d | 1500 | err = count_semcnt(sma, semnum, 0); |
1da177e4 LT |
1501 | goto out_unlock; |
1502 | case GETZCNT: | |
2f2ed41d | 1503 | err = count_semcnt(sma, semnum, 1); |
1da177e4 | 1504 | goto out_unlock; |
1da177e4 | 1505 | } |
16df3674 | 1506 | |
1da177e4 | 1507 | out_unlock: |
6062a8dc | 1508 | sem_unlock(sma, -1); |
c728b9c8 | 1509 | out_rcu_wakeup: |
6d49dab8 | 1510 | rcu_read_unlock(); |
0a2b9d4c | 1511 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 1512 | out_free: |
239521f3 | 1513 | if (sem_io != fast_sem_io) |
1da177e4 LT |
1514 | ipc_free(sem_io, sizeof(ushort)*nsems); |
1515 | return err; | |
1516 | } | |
1517 | ||
016d7132 PP |
1518 | static inline unsigned long |
1519 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1520 | { |
239521f3 | 1521 | switch (version) { |
1da177e4 | 1522 | case IPC_64: |
016d7132 | 1523 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1524 | return -EFAULT; |
1da177e4 | 1525 | return 0; |
1da177e4 LT |
1526 | case IPC_OLD: |
1527 | { | |
1528 | struct semid_ds tbuf_old; | |
1529 | ||
239521f3 | 1530 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1531 | return -EFAULT; |
1532 | ||
016d7132 PP |
1533 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1534 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1535 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1536 | |
1537 | return 0; | |
1538 | } | |
1539 | default: | |
1540 | return -EINVAL; | |
1541 | } | |
1542 | } | |
1543 | ||
522bb2a2 | 1544 | /* |
d9a605e4 | 1545 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1546 | * to be held in write mode. |
d9a605e4 | 1547 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1548 | */ |
21a4826a | 1549 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1550 | int cmd, int version, void __user *p) |
1da177e4 LT |
1551 | { |
1552 | struct sem_array *sma; | |
1553 | int err; | |
016d7132 | 1554 | struct semid64_ds semid64; |
1da177e4 LT |
1555 | struct kern_ipc_perm *ipcp; |
1556 | ||
239521f3 | 1557 | if (cmd == IPC_SET) { |
e1fd1f49 | 1558 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1559 | return -EFAULT; |
1da177e4 | 1560 | } |
073115d6 | 1561 | |
d9a605e4 | 1562 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1563 | rcu_read_lock(); |
1564 | ||
16df3674 DB |
1565 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1566 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1567 | if (IS_ERR(ipcp)) { |
1568 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1569 | goto out_unlock1; |
1570 | } | |
073115d6 | 1571 | |
a5f75e7f | 1572 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1573 | |
1574 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1575 | if (err) |
1576 | goto out_unlock1; | |
1da177e4 | 1577 | |
7b4cc5d8 | 1578 | switch (cmd) { |
1da177e4 | 1579 | case IPC_RMID: |
6062a8dc | 1580 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1581 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1582 | freeary(ns, ipcp); |
522bb2a2 | 1583 | goto out_up; |
1da177e4 | 1584 | case IPC_SET: |
6062a8dc | 1585 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1586 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1587 | if (err) | |
7b4cc5d8 | 1588 | goto out_unlock0; |
1da177e4 | 1589 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1590 | break; |
1591 | default: | |
1da177e4 | 1592 | err = -EINVAL; |
7b4cc5d8 | 1593 | goto out_unlock1; |
1da177e4 | 1594 | } |
1da177e4 | 1595 | |
7b4cc5d8 | 1596 | out_unlock0: |
6062a8dc | 1597 | sem_unlock(sma, -1); |
7b4cc5d8 | 1598 | out_unlock1: |
6d49dab8 | 1599 | rcu_read_unlock(); |
522bb2a2 | 1600 | out_up: |
d9a605e4 | 1601 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1602 | return err; |
1603 | } | |
1604 | ||
e1fd1f49 | 1605 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1606 | { |
1da177e4 | 1607 | int version; |
e3893534 | 1608 | struct ipc_namespace *ns; |
e1fd1f49 | 1609 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1610 | |
1611 | if (semid < 0) | |
1612 | return -EINVAL; | |
1613 | ||
1614 | version = ipc_parse_version(&cmd); | |
e3893534 | 1615 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1616 | |
239521f3 | 1617 | switch (cmd) { |
1da177e4 LT |
1618 | case IPC_INFO: |
1619 | case SEM_INFO: | |
4b9fcb0e | 1620 | case IPC_STAT: |
1da177e4 | 1621 | case SEM_STAT: |
e1fd1f49 | 1622 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1623 | case GETALL: |
1624 | case GETVAL: | |
1625 | case GETPID: | |
1626 | case GETNCNT: | |
1627 | case GETZCNT: | |
1da177e4 | 1628 | case SETALL: |
e1fd1f49 AV |
1629 | return semctl_main(ns, semid, semnum, cmd, p); |
1630 | case SETVAL: | |
1631 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1632 | case IPC_RMID: |
1633 | case IPC_SET: | |
e1fd1f49 | 1634 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1635 | default: |
1636 | return -EINVAL; | |
1637 | } | |
1638 | } | |
1639 | ||
1da177e4 LT |
1640 | /* If the task doesn't already have a undo_list, then allocate one |
1641 | * here. We guarantee there is only one thread using this undo list, | |
1642 | * and current is THE ONE | |
1643 | * | |
1644 | * If this allocation and assignment succeeds, but later | |
1645 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1646 | * Just let it stay associated with the task, and it'll be freed later | |
1647 | * at exit time. | |
1648 | * | |
1649 | * This can block, so callers must hold no locks. | |
1650 | */ | |
1651 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1652 | { | |
1653 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1654 | |
1655 | undo_list = current->sysvsem.undo_list; | |
1656 | if (!undo_list) { | |
2453a306 | 1657 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1658 | if (undo_list == NULL) |
1659 | return -ENOMEM; | |
00a5dfdb | 1660 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1661 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1662 | INIT_LIST_HEAD(&undo_list->list_proc); |
1663 | ||
1da177e4 LT |
1664 | current->sysvsem.undo_list = undo_list; |
1665 | } | |
1666 | *undo_listp = undo_list; | |
1667 | return 0; | |
1668 | } | |
1669 | ||
bf17bb71 | 1670 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1671 | { |
bf17bb71 | 1672 | struct sem_undo *un; |
4daa28f6 | 1673 | |
bf17bb71 NP |
1674 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1675 | if (un->semid == semid) | |
1676 | return un; | |
1da177e4 | 1677 | } |
4daa28f6 | 1678 | return NULL; |
1da177e4 LT |
1679 | } |
1680 | ||
bf17bb71 NP |
1681 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1682 | { | |
1683 | struct sem_undo *un; | |
1684 | ||
239521f3 | 1685 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1686 | |
1687 | un = __lookup_undo(ulp, semid); | |
1688 | if (un) { | |
1689 | list_del_rcu(&un->list_proc); | |
1690 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1691 | } | |
1692 | return un; | |
1693 | } | |
1694 | ||
4daa28f6 | 1695 | /** |
8001c858 | 1696 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1697 | * @ns: namespace |
1698 | * @semid: semaphore array id | |
1699 | * | |
1700 | * The function looks up (and if not present creates) the undo structure. | |
1701 | * The size of the undo structure depends on the size of the semaphore | |
1702 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1703 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1704 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1705 | */ |
1706 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1707 | { |
1708 | struct sem_array *sma; | |
1709 | struct sem_undo_list *ulp; | |
1710 | struct sem_undo *un, *new; | |
6062a8dc | 1711 | int nsems, error; |
1da177e4 LT |
1712 | |
1713 | error = get_undo_list(&ulp); | |
1714 | if (error) | |
1715 | return ERR_PTR(error); | |
1716 | ||
380af1b3 | 1717 | rcu_read_lock(); |
c530c6ac | 1718 | spin_lock(&ulp->lock); |
1da177e4 | 1719 | un = lookup_undo(ulp, semid); |
c530c6ac | 1720 | spin_unlock(&ulp->lock); |
239521f3 | 1721 | if (likely(un != NULL)) |
1da177e4 LT |
1722 | goto out; |
1723 | ||
1724 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1725 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1726 | sma = sem_obtain_object_check(ns, semid); |
1727 | if (IS_ERR(sma)) { | |
1728 | rcu_read_unlock(); | |
4de85cd6 | 1729 | return ERR_CAST(sma); |
16df3674 | 1730 | } |
023a5355 | 1731 | |
1da177e4 | 1732 | nsems = sma->sem_nsems; |
6062a8dc RR |
1733 | if (!ipc_rcu_getref(sma)) { |
1734 | rcu_read_unlock(); | |
1735 | un = ERR_PTR(-EIDRM); | |
1736 | goto out; | |
1737 | } | |
16df3674 | 1738 | rcu_read_unlock(); |
1da177e4 | 1739 | |
4daa28f6 | 1740 | /* step 2: allocate new undo structure */ |
4668edc3 | 1741 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1742 | if (!new) { |
62659f0b | 1743 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1744 | return ERR_PTR(-ENOMEM); |
1745 | } | |
1da177e4 | 1746 | |
380af1b3 | 1747 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1748 | rcu_read_lock(); |
6ff37972 | 1749 | sem_lock_and_putref(sma); |
0f3d2b01 | 1750 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1751 | sem_unlock(sma, -1); |
6d49dab8 | 1752 | rcu_read_unlock(); |
1da177e4 LT |
1753 | kfree(new); |
1754 | un = ERR_PTR(-EIDRM); | |
1755 | goto out; | |
1756 | } | |
380af1b3 MS |
1757 | spin_lock(&ulp->lock); |
1758 | ||
1759 | /* | |
1760 | * step 4: check for races: did someone else allocate the undo struct? | |
1761 | */ | |
1762 | un = lookup_undo(ulp, semid); | |
1763 | if (un) { | |
1764 | kfree(new); | |
1765 | goto success; | |
1766 | } | |
4daa28f6 MS |
1767 | /* step 5: initialize & link new undo structure */ |
1768 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1769 | new->ulp = ulp; |
4daa28f6 MS |
1770 | new->semid = semid; |
1771 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1772 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1773 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1774 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1775 | un = new; |
4daa28f6 | 1776 | |
380af1b3 | 1777 | success: |
c530c6ac | 1778 | spin_unlock(&ulp->lock); |
6062a8dc | 1779 | sem_unlock(sma, -1); |
1da177e4 LT |
1780 | out: |
1781 | return un; | |
1782 | } | |
1783 | ||
c61284e9 MS |
1784 | |
1785 | /** | |
8001c858 | 1786 | * get_queue_result - retrieve the result code from sem_queue |
c61284e9 MS |
1787 | * @q: Pointer to queue structure |
1788 | * | |
1789 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1790 | * q->status, then we must loop until the value is replaced with the final | |
1791 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1792 | * signal) and in parallel the task is woken up by another task because it got | |
1793 | * the requested semaphores. | |
1794 | * | |
1795 | * The function can be called with or without holding the semaphore spinlock. | |
1796 | */ | |
1797 | static int get_queue_result(struct sem_queue *q) | |
1798 | { | |
1799 | int error; | |
1800 | ||
1801 | error = q->status; | |
1802 | while (unlikely(error == IN_WAKEUP)) { | |
1803 | cpu_relax(); | |
1804 | error = q->status; | |
1805 | } | |
1806 | ||
1807 | return error; | |
1808 | } | |
1809 | ||
d5460c99 HC |
1810 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1811 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1812 | { |
1813 | int error = -EINVAL; | |
1814 | struct sem_array *sma; | |
1815 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1816 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1817 | struct sem_undo *un; |
6062a8dc | 1818 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1819 | struct sem_queue queue; |
1820 | unsigned long jiffies_left = 0; | |
e3893534 | 1821 | struct ipc_namespace *ns; |
0a2b9d4c | 1822 | struct list_head tasks; |
e3893534 KK |
1823 | |
1824 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1825 | |
1826 | if (nsops < 1 || semid < 0) | |
1827 | return -EINVAL; | |
e3893534 | 1828 | if (nsops > ns->sc_semopm) |
1da177e4 | 1829 | return -E2BIG; |
239521f3 MS |
1830 | if (nsops > SEMOPM_FAST) { |
1831 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1832 | if (sops == NULL) | |
1da177e4 LT |
1833 | return -ENOMEM; |
1834 | } | |
239521f3 MS |
1835 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1836 | error = -EFAULT; | |
1da177e4 LT |
1837 | goto out_free; |
1838 | } | |
1839 | if (timeout) { | |
1840 | struct timespec _timeout; | |
1841 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1842 | error = -EFAULT; | |
1843 | goto out_free; | |
1844 | } | |
1845 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1846 | _timeout.tv_nsec >= 1000000000L) { | |
1847 | error = -EINVAL; | |
1848 | goto out_free; | |
1849 | } | |
1850 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1851 | } | |
1852 | max = 0; | |
1853 | for (sop = sops; sop < sops + nsops; sop++) { | |
1854 | if (sop->sem_num >= max) | |
1855 | max = sop->sem_num; | |
1856 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1857 | undos = 1; |
1858 | if (sop->sem_op != 0) | |
1da177e4 LT |
1859 | alter = 1; |
1860 | } | |
1da177e4 | 1861 | |
6062a8dc RR |
1862 | INIT_LIST_HEAD(&tasks); |
1863 | ||
1da177e4 | 1864 | if (undos) { |
6062a8dc | 1865 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1866 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1867 | if (IS_ERR(un)) { |
1868 | error = PTR_ERR(un); | |
1869 | goto out_free; | |
1870 | } | |
6062a8dc | 1871 | } else { |
1da177e4 | 1872 | un = NULL; |
6062a8dc RR |
1873 | rcu_read_lock(); |
1874 | } | |
1da177e4 | 1875 | |
16df3674 | 1876 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1877 | if (IS_ERR(sma)) { |
6062a8dc | 1878 | rcu_read_unlock(); |
023a5355 | 1879 | error = PTR_ERR(sma); |
1da177e4 | 1880 | goto out_free; |
023a5355 ND |
1881 | } |
1882 | ||
16df3674 | 1883 | error = -EFBIG; |
c728b9c8 LT |
1884 | if (max >= sma->sem_nsems) |
1885 | goto out_rcu_wakeup; | |
16df3674 DB |
1886 | |
1887 | error = -EACCES; | |
c728b9c8 LT |
1888 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
1889 | goto out_rcu_wakeup; | |
16df3674 DB |
1890 | |
1891 | error = security_sem_semop(sma, sops, nsops, alter); | |
c728b9c8 LT |
1892 | if (error) |
1893 | goto out_rcu_wakeup; | |
16df3674 | 1894 | |
6e224f94 MS |
1895 | error = -EIDRM; |
1896 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
1897 | /* |
1898 | * We eventually might perform the following check in a lockless | |
1899 | * fashion, considering ipc_valid_object() locking constraints. | |
1900 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1901 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1902 | * check below. More details on the fine grained locking scheme | |
1903 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1904 | */ | |
1905 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 1906 | goto out_unlock_free; |
1da177e4 | 1907 | /* |
4daa28f6 | 1908 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1909 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1910 | * and now a new array with received the same id. Check and fail. |
25985edc | 1911 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1912 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1913 | */ |
6062a8dc RR |
1914 | if (un && un->semid == -1) |
1915 | goto out_unlock_free; | |
4daa28f6 | 1916 | |
d198cd6d MS |
1917 | queue.sops = sops; |
1918 | queue.nsops = nsops; | |
1919 | queue.undo = un; | |
1920 | queue.pid = task_tgid_vnr(current); | |
1921 | queue.alter = alter; | |
1922 | ||
1923 | error = perform_atomic_semop(sma, &queue); | |
0e8c6656 MS |
1924 | if (error == 0) { |
1925 | /* If the operation was successful, then do | |
1926 | * the required updates. | |
1927 | */ | |
1928 | if (alter) | |
0a2b9d4c | 1929 | do_smart_update(sma, sops, nsops, 1, &tasks); |
0e8c6656 MS |
1930 | else |
1931 | set_semotime(sma, sops); | |
1da177e4 | 1932 | } |
0e8c6656 MS |
1933 | if (error <= 0) |
1934 | goto out_unlock_free; | |
1da177e4 LT |
1935 | |
1936 | /* We need to sleep on this operation, so we put the current | |
1937 | * task into the pending queue and go to sleep. | |
1938 | */ | |
46c0a8ca | 1939 | |
b97e820f MS |
1940 | if (nsops == 1) { |
1941 | struct sem *curr; | |
1942 | curr = &sma->sem_base[sops->sem_num]; | |
1943 | ||
f269f40a MS |
1944 | if (alter) { |
1945 | if (sma->complex_count) { | |
1946 | list_add_tail(&queue.list, | |
1947 | &sma->pending_alter); | |
1948 | } else { | |
1949 | ||
1950 | list_add_tail(&queue.list, | |
1951 | &curr->pending_alter); | |
1952 | } | |
1953 | } else { | |
1a82e9e1 | 1954 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 1955 | } |
b97e820f | 1956 | } else { |
f269f40a MS |
1957 | if (!sma->complex_count) |
1958 | merge_queues(sma); | |
1959 | ||
9f1bc2c9 | 1960 | if (alter) |
1a82e9e1 | 1961 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1962 | else |
1a82e9e1 MS |
1963 | list_add_tail(&queue.list, &sma->pending_const); |
1964 | ||
b97e820f MS |
1965 | sma->complex_count++; |
1966 | } | |
1967 | ||
1da177e4 LT |
1968 | queue.status = -EINTR; |
1969 | queue.sleeper = current; | |
0b0577f6 MS |
1970 | |
1971 | sleep_again: | |
52644c9a | 1972 | __set_current_state(TASK_INTERRUPTIBLE); |
6062a8dc | 1973 | sem_unlock(sma, locknum); |
6d49dab8 | 1974 | rcu_read_unlock(); |
1da177e4 LT |
1975 | |
1976 | if (timeout) | |
1977 | jiffies_left = schedule_timeout(jiffies_left); | |
1978 | else | |
1979 | schedule(); | |
1980 | ||
c61284e9 | 1981 | error = get_queue_result(&queue); |
1da177e4 LT |
1982 | |
1983 | if (error != -EINTR) { | |
1984 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1985 | * resources. |
1986 | * Perform a smp_mb(): User space could assume that semop() | |
1987 | * is a memory barrier: Without the mb(), the cpu could | |
1988 | * speculatively read in user space stale data that was | |
1989 | * overwritten by the previous owner of the semaphore. | |
1990 | */ | |
1991 | smp_mb(); | |
1992 | ||
1da177e4 LT |
1993 | goto out_free; |
1994 | } | |
1995 | ||
321310ce | 1996 | rcu_read_lock(); |
6062a8dc | 1997 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1998 | |
1999 | /* | |
2000 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
2001 | */ | |
2002 | error = get_queue_result(&queue); | |
2003 | ||
2004 | /* | |
2005 | * Array removed? If yes, leave without sem_unlock(). | |
2006 | */ | |
023a5355 | 2007 | if (IS_ERR(sma)) { |
321310ce | 2008 | rcu_read_unlock(); |
1da177e4 LT |
2009 | goto out_free; |
2010 | } | |
2011 | ||
c61284e9 | 2012 | |
1da177e4 | 2013 | /* |
d694ad62 MS |
2014 | * If queue.status != -EINTR we are woken up by another process. |
2015 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 2016 | */ |
3ab08fe2 | 2017 | if (error != -EINTR) |
1da177e4 | 2018 | goto out_unlock_free; |
1da177e4 LT |
2019 | |
2020 | /* | |
2021 | * If an interrupt occurred we have to clean up the queue | |
2022 | */ | |
2023 | if (timeout && jiffies_left == 0) | |
2024 | error = -EAGAIN; | |
0b0577f6 MS |
2025 | |
2026 | /* | |
2027 | * If the wakeup was spurious, just retry | |
2028 | */ | |
2029 | if (error == -EINTR && !signal_pending(current)) | |
2030 | goto sleep_again; | |
2031 | ||
b97e820f | 2032 | unlink_queue(sma, &queue); |
1da177e4 LT |
2033 | |
2034 | out_unlock_free: | |
6062a8dc | 2035 | sem_unlock(sma, locknum); |
c728b9c8 | 2036 | out_rcu_wakeup: |
6d49dab8 | 2037 | rcu_read_unlock(); |
0a2b9d4c | 2038 | wake_up_sem_queue_do(&tasks); |
1da177e4 | 2039 | out_free: |
239521f3 | 2040 | if (sops != fast_sops) |
1da177e4 LT |
2041 | kfree(sops); |
2042 | return error; | |
2043 | } | |
2044 | ||
d5460c99 HC |
2045 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2046 | unsigned, nsops) | |
1da177e4 LT |
2047 | { |
2048 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2049 | } | |
2050 | ||
2051 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2052 | * parent and child tasks. | |
1da177e4 LT |
2053 | */ |
2054 | ||
2055 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2056 | { | |
2057 | struct sem_undo_list *undo_list; | |
2058 | int error; | |
2059 | ||
2060 | if (clone_flags & CLONE_SYSVSEM) { | |
2061 | error = get_undo_list(&undo_list); | |
2062 | if (error) | |
2063 | return error; | |
1da177e4 LT |
2064 | atomic_inc(&undo_list->refcnt); |
2065 | tsk->sysvsem.undo_list = undo_list; | |
46c0a8ca | 2066 | } else |
1da177e4 LT |
2067 | tsk->sysvsem.undo_list = NULL; |
2068 | ||
2069 | return 0; | |
2070 | } | |
2071 | ||
2072 | /* | |
2073 | * add semadj values to semaphores, free undo structures. | |
2074 | * undo structures are not freed when semaphore arrays are destroyed | |
2075 | * so some of them may be out of date. | |
2076 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2077 | * set of adjustments that needs to be done should be done in an atomic | |
2078 | * manner or not. That is, if we are attempting to decrement the semval | |
2079 | * should we queue up and wait until we can do so legally? | |
2080 | * The original implementation attempted to do this (queue and wait). | |
2081 | * The current implementation does not do so. The POSIX standard | |
2082 | * and SVID should be consulted to determine what behavior is mandated. | |
2083 | */ | |
2084 | void exit_sem(struct task_struct *tsk) | |
2085 | { | |
4daa28f6 | 2086 | struct sem_undo_list *ulp; |
1da177e4 | 2087 | |
4daa28f6 MS |
2088 | ulp = tsk->sysvsem.undo_list; |
2089 | if (!ulp) | |
1da177e4 | 2090 | return; |
9edff4ab | 2091 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2092 | |
4daa28f6 | 2093 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2094 | return; |
2095 | ||
380af1b3 | 2096 | for (;;) { |
1da177e4 | 2097 | struct sem_array *sma; |
380af1b3 | 2098 | struct sem_undo *un; |
0a2b9d4c | 2099 | struct list_head tasks; |
6062a8dc | 2100 | int semid, i; |
4daa28f6 | 2101 | |
380af1b3 | 2102 | rcu_read_lock(); |
05725f7e JP |
2103 | un = list_entry_rcu(ulp->list_proc.next, |
2104 | struct sem_undo, list_proc); | |
602b8593 HK |
2105 | if (&un->list_proc == &ulp->list_proc) { |
2106 | /* | |
2107 | * We must wait for freeary() before freeing this ulp, | |
2108 | * in case we raced with last sem_undo. There is a small | |
2109 | * possibility where we exit while freeary() didn't | |
2110 | * finish unlocking sem_undo_list. | |
2111 | */ | |
2112 | spin_unlock_wait(&ulp->lock); | |
2113 | rcu_read_unlock(); | |
2114 | break; | |
2115 | } | |
2116 | spin_lock(&ulp->lock); | |
2117 | semid = un->semid; | |
2118 | spin_unlock(&ulp->lock); | |
4daa28f6 | 2119 | |
602b8593 | 2120 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2121 | if (semid == -1) { |
2122 | rcu_read_unlock(); | |
602b8593 | 2123 | continue; |
6062a8dc | 2124 | } |
1da177e4 | 2125 | |
602b8593 | 2126 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); |
380af1b3 | 2127 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2128 | if (IS_ERR(sma)) { |
2129 | rcu_read_unlock(); | |
380af1b3 | 2130 | continue; |
6062a8dc | 2131 | } |
1da177e4 | 2132 | |
6062a8dc | 2133 | sem_lock(sma, NULL, -1); |
6e224f94 | 2134 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2135 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2136 | sem_unlock(sma, -1); |
2137 | rcu_read_unlock(); | |
2138 | continue; | |
2139 | } | |
bf17bb71 | 2140 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2141 | if (un == NULL) { |
2142 | /* exit_sem raced with IPC_RMID+semget() that created | |
2143 | * exactly the same semid. Nothing to do. | |
2144 | */ | |
6062a8dc | 2145 | sem_unlock(sma, -1); |
6d49dab8 | 2146 | rcu_read_unlock(); |
380af1b3 MS |
2147 | continue; |
2148 | } | |
2149 | ||
2150 | /* remove un from the linked lists */ | |
cf9d5d78 | 2151 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2152 | list_del(&un->list_id); |
2153 | ||
a9795584 HK |
2154 | /* we are the last process using this ulp, acquiring ulp->lock |
2155 | * isn't required. Besides that, we are also protected against | |
2156 | * IPC_RMID as we hold sma->sem_perm lock now | |
2157 | */ | |
380af1b3 | 2158 | list_del_rcu(&un->list_proc); |
380af1b3 | 2159 | |
4daa28f6 MS |
2160 | /* perform adjustments registered in un */ |
2161 | for (i = 0; i < sma->sem_nsems; i++) { | |
239521f3 | 2162 | struct sem *semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
2163 | if (un->semadj[i]) { |
2164 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2165 | /* |
2166 | * Range checks of the new semaphore value, | |
2167 | * not defined by sus: | |
2168 | * - Some unices ignore the undo entirely | |
2169 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2170 | * - some cap the value (e.g. FreeBSD caps | |
2171 | * at 0, but doesn't enforce SEMVMX) | |
2172 | * | |
2173 | * Linux caps the semaphore value, both at 0 | |
2174 | * and at SEMVMX. | |
2175 | * | |
239521f3 | 2176 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2177 | */ |
5f921ae9 IM |
2178 | if (semaphore->semval < 0) |
2179 | semaphore->semval = 0; | |
2180 | if (semaphore->semval > SEMVMX) | |
2181 | semaphore->semval = SEMVMX; | |
b488893a | 2182 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2183 | } |
2184 | } | |
1da177e4 | 2185 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
2186 | INIT_LIST_HEAD(&tasks); |
2187 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 2188 | sem_unlock(sma, -1); |
6d49dab8 | 2189 | rcu_read_unlock(); |
0a2b9d4c | 2190 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 2191 | |
693a8b6e | 2192 | kfree_rcu(un, rcu); |
1da177e4 | 2193 | } |
4daa28f6 | 2194 | kfree(ulp); |
1da177e4 LT |
2195 | } |
2196 | ||
2197 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2198 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2199 | { |
1efdb69b | 2200 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2201 | struct sem_array *sma = it; |
d12e1e50 MS |
2202 | time_t sem_otime; |
2203 | ||
d8c63376 MS |
2204 | /* |
2205 | * The proc interface isn't aware of sem_lock(), it calls | |
2206 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
f6031d95 MS |
2207 | * In order to stay compatible with sem_lock(), we must |
2208 | * enter / leave complex_mode. | |
d8c63376 | 2209 | */ |
f6031d95 | 2210 | complexmode_enter(sma); |
d8c63376 | 2211 | |
d12e1e50 | 2212 | sem_otime = get_semotime(sma); |
19b4946c | 2213 | |
7f032d6e JP |
2214 | seq_printf(s, |
2215 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", | |
2216 | sma->sem_perm.key, | |
2217 | sma->sem_perm.id, | |
2218 | sma->sem_perm.mode, | |
2219 | sma->sem_nsems, | |
2220 | from_kuid_munged(user_ns, sma->sem_perm.uid), | |
2221 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2222 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2223 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
2224 | sem_otime, | |
2225 | sma->sem_ctime); | |
2226 | ||
f6031d95 MS |
2227 | complexmode_tryleave(sma); |
2228 | ||
7f032d6e | 2229 | return 0; |
1da177e4 LT |
2230 | } |
2231 | #endif |