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> | |
9ae949fa | 14 | * (c) 2016 Davidlohr Bueso <dave@stgolabs.net> |
c5cf6359 MS |
15 | * Further wakeup optimizations, documentation |
16 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
17 | * |
18 | * support for audit of ipc object properties and permission changes | |
19 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
20 | * |
21 | * namespaces support | |
22 | * OpenVZ, SWsoft Inc. | |
23 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
24 | * |
25 | * Implementation notes: (May 2010) | |
26 | * This file implements System V semaphores. | |
27 | * | |
28 | * User space visible behavior: | |
29 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
30 | * protection) | |
31 | * - multiple semaphore operations that alter the same semaphore in | |
32 | * one semop() are handled. | |
33 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
34 | * SETALL calls. | |
35 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
36 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
37 | * - namespace are supported. | |
38 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
39 | * to /proc/sys/kernel/sem. | |
40 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
41 | * | |
42 | * Internals: | |
43 | * - scalability: | |
44 | * - all global variables are read-mostly. | |
45 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
46 | * - most operations do write operations (actually: spin_lock calls) to | |
47 | * the per-semaphore array structure. | |
48 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
49 | * If multiple semaphores in one array are used, then cache line | |
50 | * trashing on the semaphore array spinlock will limit the scaling. | |
2f2ed41d | 51 | * - semncnt and semzcnt are calculated on demand in count_semcnt() |
c5cf6359 MS |
52 | * - the task that performs a successful semop() scans the list of all |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
9ae949fa | 57 | * dropping all locks. (see wake_up_sem_queue_prepare()) |
c5cf6359 MS |
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). | |
c5cf6359 MS |
62 | * - UNDO values are stored in an array (one per process and per |
63 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
64 | * modes for the UNDO variables are supported (per process, per thread) | |
65 | * (see copy_semundo, CLONE_SYSVSEM) | |
66 | * - There are two lists of the pending operations: a per-array list | |
67 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
68 | * ordering without always scanning all pending operations. | |
69 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
70 | */ |
71 | ||
1da177e4 LT |
72 | #include <linux/slab.h> |
73 | #include <linux/spinlock.h> | |
74 | #include <linux/init.h> | |
75 | #include <linux/proc_fs.h> | |
76 | #include <linux/time.h> | |
1da177e4 LT |
77 | #include <linux/security.h> |
78 | #include <linux/syscalls.h> | |
79 | #include <linux/audit.h> | |
c59ede7b | 80 | #include <linux/capability.h> |
19b4946c | 81 | #include <linux/seq_file.h> |
3e148c79 | 82 | #include <linux/rwsem.h> |
e3893534 | 83 | #include <linux/nsproxy.h> |
ae5e1b22 | 84 | #include <linux/ipc_namespace.h> |
84f001e1 | 85 | #include <linux/sched/wake_q.h> |
5f921ae9 | 86 | |
7153e402 | 87 | #include <linux/uaccess.h> |
1da177e4 LT |
88 | #include "util.h" |
89 | ||
e57940d7 MS |
90 | |
91 | /* One queue for each sleeping process in the system. */ | |
92 | struct sem_queue { | |
e57940d7 MS |
93 | struct list_head list; /* queue of pending operations */ |
94 | struct task_struct *sleeper; /* this process */ | |
95 | struct sem_undo *undo; /* undo structure */ | |
96 | int pid; /* process id of requesting process */ | |
97 | int status; /* completion status of operation */ | |
98 | struct sembuf *sops; /* array of pending operations */ | |
ed247b7c | 99 | struct sembuf *blocking; /* the operation that blocked */ |
e57940d7 | 100 | int nsops; /* number of operations */ |
4ce33ec2 DB |
101 | bool alter; /* does *sops alter the array? */ |
102 | bool dupsop; /* sops on more than one sem_num */ | |
e57940d7 MS |
103 | }; |
104 | ||
105 | /* Each task has a list of undo requests. They are executed automatically | |
106 | * when the process exits. | |
107 | */ | |
108 | struct sem_undo { | |
109 | struct list_head list_proc; /* per-process list: * | |
110 | * all undos from one process | |
111 | * rcu protected */ | |
112 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
113 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
114 | struct list_head list_id; /* per semaphore array list: | |
115 | * all undos for one array */ | |
116 | int semid; /* semaphore set identifier */ | |
117 | short *semadj; /* array of adjustments */ | |
118 | /* one per semaphore */ | |
119 | }; | |
120 | ||
121 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
122 | * that may be shared among all a CLONE_SYSVSEM task group. | |
123 | */ | |
124 | struct sem_undo_list { | |
125 | atomic_t refcnt; | |
126 | spinlock_t lock; | |
127 | struct list_head list_proc; | |
128 | }; | |
129 | ||
130 | ||
ed2ddbf8 | 131 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 132 | |
1b531f21 | 133 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 134 | |
7748dbfa | 135 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 136 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 137 | #ifdef CONFIG_PROC_FS |
19b4946c | 138 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
139 | #endif |
140 | ||
141 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
142 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
143 | ||
9de5ab8a MS |
144 | /* |
145 | * Switching from the mode suitable for simple ops | |
146 | * to the mode for complex ops is costly. Therefore: | |
147 | * use some hysteresis | |
148 | */ | |
149 | #define USE_GLOBAL_LOCK_HYSTERESIS 10 | |
150 | ||
1da177e4 | 151 | /* |
758a6ba3 | 152 | * Locking: |
5864a2fd | 153 | * a) global sem_lock() for read/write |
1da177e4 | 154 | * sem_undo.id_next, |
758a6ba3 | 155 | * sem_array.complex_count, |
5864a2fd MS |
156 | * sem_array.pending{_alter,_const}, |
157 | * sem_array.sem_undo | |
46c0a8ca | 158 | * |
5864a2fd | 159 | * b) global or semaphore sem_lock() for read/write: |
1a233956 | 160 | * sem_array.sems[i].pending_{const,alter}: |
5864a2fd MS |
161 | * |
162 | * c) special: | |
163 | * sem_undo_list.list_proc: | |
164 | * * undo_list->lock for write | |
165 | * * rcu for read | |
9de5ab8a MS |
166 | * use_global_lock: |
167 | * * global sem_lock() for write | |
168 | * * either local or global sem_lock() for read. | |
169 | * | |
170 | * Memory ordering: | |
171 | * Most ordering is enforced by using spin_lock() and spin_unlock(). | |
172 | * The special case is use_global_lock: | |
173 | * Setting it from non-zero to 0 is a RELEASE, this is ensured by | |
174 | * using smp_store_release(). | |
175 | * Testing if it is non-zero is an ACQUIRE, this is ensured by using | |
176 | * smp_load_acquire(). | |
177 | * Setting it from 0 to non-zero must be ordered with regards to | |
178 | * this smp_load_acquire(), this is guaranteed because the smp_load_acquire() | |
179 | * is inside a spin_lock() and after a write from 0 to non-zero a | |
180 | * spin_lock()+spin_unlock() is done. | |
1da177e4 LT |
181 | */ |
182 | ||
e3893534 KK |
183 | #define sc_semmsl sem_ctls[0] |
184 | #define sc_semmns sem_ctls[1] | |
185 | #define sc_semopm sem_ctls[2] | |
186 | #define sc_semmni sem_ctls[3] | |
187 | ||
ed2ddbf8 | 188 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 189 | { |
e3893534 KK |
190 | ns->sc_semmsl = SEMMSL; |
191 | ns->sc_semmns = SEMMNS; | |
192 | ns->sc_semopm = SEMOPM; | |
193 | ns->sc_semmni = SEMMNI; | |
194 | ns->used_sems = 0; | |
ed2ddbf8 | 195 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
196 | } |
197 | ||
ae5e1b22 | 198 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
199 | void sem_exit_ns(struct ipc_namespace *ns) |
200 | { | |
01b8b07a | 201 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 202 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 203 | } |
ae5e1b22 | 204 | #endif |
1da177e4 | 205 | |
239521f3 | 206 | void __init sem_init(void) |
1da177e4 | 207 | { |
ed2ddbf8 | 208 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
209 | ipc_init_proc_interface("sysvipc/sem", |
210 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 211 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
212 | } |
213 | ||
f269f40a MS |
214 | /** |
215 | * unmerge_queues - unmerge queues, if possible. | |
216 | * @sma: semaphore array | |
217 | * | |
218 | * The function unmerges the wait queues if complex_count is 0. | |
219 | * It must be called prior to dropping the global semaphore array lock. | |
220 | */ | |
221 | static void unmerge_queues(struct sem_array *sma) | |
222 | { | |
223 | struct sem_queue *q, *tq; | |
224 | ||
225 | /* complex operations still around? */ | |
226 | if (sma->complex_count) | |
227 | return; | |
228 | /* | |
229 | * We will switch back to simple mode. | |
230 | * Move all pending operation back into the per-semaphore | |
231 | * queues. | |
232 | */ | |
233 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
234 | struct sem *curr; | |
1a233956 | 235 | curr = &sma->sems[q->sops[0].sem_num]; |
f269f40a MS |
236 | |
237 | list_add_tail(&q->list, &curr->pending_alter); | |
238 | } | |
239 | INIT_LIST_HEAD(&sma->pending_alter); | |
240 | } | |
241 | ||
242 | /** | |
8001c858 | 243 | * merge_queues - merge single semop queues into global queue |
f269f40a MS |
244 | * @sma: semaphore array |
245 | * | |
246 | * This function merges all per-semaphore queues into the global queue. | |
247 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
248 | * operations when a multi-semop operation must sleep. | |
249 | * Only the alter operations must be moved, the const operations can stay. | |
250 | */ | |
251 | static void merge_queues(struct sem_array *sma) | |
252 | { | |
253 | int i; | |
254 | for (i = 0; i < sma->sem_nsems; i++) { | |
1a233956 | 255 | struct sem *sem = &sma->sems[i]; |
f269f40a MS |
256 | |
257 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
258 | } | |
259 | } | |
260 | ||
1b4654ef KC |
261 | static void __sem_free(struct sem_array *sma) |
262 | { | |
263 | kvfree(sma); | |
264 | } | |
265 | ||
53dad6d3 DB |
266 | static void sem_rcu_free(struct rcu_head *head) |
267 | { | |
dba4cdd3 MS |
268 | struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu); |
269 | struct sem_array *sma = container_of(p, struct sem_array, sem_perm); | |
53dad6d3 DB |
270 | |
271 | security_sem_free(sma); | |
1b4654ef | 272 | __sem_free(sma); |
53dad6d3 DB |
273 | } |
274 | ||
5e9d5275 | 275 | /* |
5864a2fd | 276 | * Enter the mode suitable for non-simple operations: |
5e9d5275 | 277 | * Caller must own sem_perm.lock. |
5e9d5275 | 278 | */ |
5864a2fd | 279 | static void complexmode_enter(struct sem_array *sma) |
5e9d5275 MS |
280 | { |
281 | int i; | |
282 | struct sem *sem; | |
283 | ||
9de5ab8a MS |
284 | if (sma->use_global_lock > 0) { |
285 | /* | |
286 | * We are already in global lock mode. | |
287 | * Nothing to do, just reset the | |
288 | * counter until we return to simple mode. | |
289 | */ | |
290 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; | |
6d07b68c MS |
291 | return; |
292 | } | |
9de5ab8a | 293 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; |
5864a2fd | 294 | |
5e9d5275 | 295 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 296 | sem = &sma->sems[i]; |
27d7be18 MS |
297 | spin_lock(&sem->lock); |
298 | spin_unlock(&sem->lock); | |
5e9d5275 | 299 | } |
5864a2fd MS |
300 | } |
301 | ||
302 | /* | |
303 | * Try to leave the mode that disallows simple operations: | |
304 | * Caller must own sem_perm.lock. | |
305 | */ | |
306 | static void complexmode_tryleave(struct sem_array *sma) | |
307 | { | |
308 | if (sma->complex_count) { | |
309 | /* Complex ops are sleeping. | |
310 | * We must stay in complex mode | |
311 | */ | |
312 | return; | |
313 | } | |
9de5ab8a MS |
314 | if (sma->use_global_lock == 1) { |
315 | /* | |
316 | * Immediately after setting use_global_lock to 0, | |
317 | * a simple op can start. Thus: all memory writes | |
318 | * performed by the current operation must be visible | |
319 | * before we set use_global_lock to 0. | |
320 | */ | |
321 | smp_store_release(&sma->use_global_lock, 0); | |
322 | } else { | |
323 | sma->use_global_lock--; | |
324 | } | |
5e9d5275 MS |
325 | } |
326 | ||
5864a2fd | 327 | #define SEM_GLOBAL_LOCK (-1) |
6062a8dc RR |
328 | /* |
329 | * If the request contains only one semaphore operation, and there are | |
330 | * no complex transactions pending, lock only the semaphore involved. | |
331 | * Otherwise, lock the entire semaphore array, since we either have | |
332 | * multiple semaphores in our own semops, or we need to look at | |
333 | * semaphores from other pending complex operations. | |
6062a8dc RR |
334 | */ |
335 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
336 | int nsops) | |
337 | { | |
5e9d5275 | 338 | struct sem *sem; |
6062a8dc | 339 | |
5e9d5275 MS |
340 | if (nsops != 1) { |
341 | /* Complex operation - acquire a full lock */ | |
342 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 343 | |
5864a2fd MS |
344 | /* Prevent parallel simple ops */ |
345 | complexmode_enter(sma); | |
346 | return SEM_GLOBAL_LOCK; | |
5e9d5275 MS |
347 | } |
348 | ||
349 | /* | |
350 | * Only one semaphore affected - try to optimize locking. | |
5864a2fd MS |
351 | * Optimized locking is possible if no complex operation |
352 | * is either enqueued or processed right now. | |
353 | * | |
9de5ab8a | 354 | * Both facts are tracked by use_global_mode. |
5e9d5275 | 355 | */ |
1a233956 | 356 | sem = &sma->sems[sops->sem_num]; |
6062a8dc | 357 | |
5864a2fd | 358 | /* |
9de5ab8a | 359 | * Initial check for use_global_lock. Just an optimization, |
5864a2fd MS |
360 | * no locking, no memory barrier. |
361 | */ | |
9de5ab8a | 362 | if (!sma->use_global_lock) { |
6062a8dc | 363 | /* |
5e9d5275 MS |
364 | * It appears that no complex operation is around. |
365 | * Acquire the per-semaphore lock. | |
6062a8dc | 366 | */ |
5e9d5275 MS |
367 | spin_lock(&sem->lock); |
368 | ||
9de5ab8a MS |
369 | /* pairs with smp_store_release() */ |
370 | if (!smp_load_acquire(&sma->use_global_lock)) { | |
5864a2fd MS |
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 | |
9de5ab8a MS |
380 | if (sma->use_global_lock == 0) { |
381 | /* | |
382 | * The use_global_lock mode ended while we waited for | |
383 | * sma->sem_perm.lock. Thus we must switch to locking | |
384 | * with sem->lock. | |
385 | * Unlike in the fast path, there is no need to recheck | |
386 | * sma->use_global_lock after we have acquired sem->lock: | |
387 | * We own sma->sem_perm.lock, thus use_global_lock cannot | |
388 | * change. | |
5e9d5275 MS |
389 | */ |
390 | spin_lock(&sem->lock); | |
9de5ab8a | 391 | |
5e9d5275 MS |
392 | ipc_unlock_object(&sma->sem_perm); |
393 | return sops->sem_num; | |
6062a8dc | 394 | } else { |
9de5ab8a MS |
395 | /* |
396 | * Not a false alarm, thus continue to use the global lock | |
397 | * mode. No need for complexmode_enter(), this was done by | |
398 | * the caller that has set use_global_mode to non-zero. | |
6062a8dc | 399 | */ |
5864a2fd | 400 | return SEM_GLOBAL_LOCK; |
6062a8dc | 401 | } |
6062a8dc RR |
402 | } |
403 | ||
404 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
405 | { | |
5864a2fd | 406 | if (locknum == SEM_GLOBAL_LOCK) { |
f269f40a | 407 | unmerge_queues(sma); |
5864a2fd | 408 | complexmode_tryleave(sma); |
cf9d5d78 | 409 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc | 410 | } else { |
1a233956 | 411 | struct sem *sem = &sma->sems[locknum]; |
6062a8dc RR |
412 | spin_unlock(&sem->lock); |
413 | } | |
6062a8dc RR |
414 | } |
415 | ||
3e148c79 | 416 | /* |
d9a605e4 | 417 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 418 | * is not held. |
321310ce LT |
419 | * |
420 | * The caller holds the RCU read lock. | |
3e148c79 | 421 | */ |
16df3674 DB |
422 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
423 | { | |
55b7ae50 | 424 | struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
16df3674 DB |
425 | |
426 | if (IS_ERR(ipcp)) | |
427 | return ERR_CAST(ipcp); | |
428 | ||
429 | return container_of(ipcp, struct sem_array, sem_perm); | |
430 | } | |
431 | ||
16df3674 DB |
432 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
433 | int id) | |
434 | { | |
435 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
436 | ||
437 | if (IS_ERR(ipcp)) | |
438 | return ERR_CAST(ipcp); | |
b1ed88b4 | 439 | |
03f02c76 | 440 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
441 | } |
442 | ||
6ff37972 PP |
443 | static inline void sem_lock_and_putref(struct sem_array *sma) |
444 | { | |
6062a8dc | 445 | sem_lock(sma, NULL, -1); |
dba4cdd3 | 446 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
6ff37972 PP |
447 | } |
448 | ||
7ca7e564 ND |
449 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
450 | { | |
451 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
452 | } | |
453 | ||
f4566f04 ND |
454 | /** |
455 | * newary - Create a new semaphore set | |
456 | * @ns: namespace | |
457 | * @params: ptr to the structure that contains key, semflg and nsems | |
458 | * | |
d9a605e4 | 459 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 | 460 | */ |
7748dbfa | 461 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
462 | { |
463 | int id; | |
464 | int retval; | |
465 | struct sem_array *sma; | |
466 | int size; | |
7748dbfa ND |
467 | key_t key = params->key; |
468 | int nsems = params->u.nsems; | |
469 | int semflg = params->flg; | |
b97e820f | 470 | int i; |
1da177e4 LT |
471 | |
472 | if (!nsems) | |
473 | return -EINVAL; | |
e3893534 | 474 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
475 | return -ENOSPC; |
476 | ||
dba4cdd3 MS |
477 | BUILD_BUG_ON(offsetof(struct sem_array, sem_perm) != 0); |
478 | ||
1a233956 | 479 | size = sizeof(*sma) + nsems * sizeof(sma->sems[0]); |
dba4cdd3 | 480 | sma = container_of(ipc_rcu_alloc(size), struct sem_array, sem_perm); |
3ab08fe2 | 481 | if (!sma) |
1da177e4 | 482 | return -ENOMEM; |
3ab08fe2 | 483 | |
1da177e4 LT |
484 | sma->sem_perm.mode = (semflg & S_IRWXUGO); |
485 | sma->sem_perm.key = key; | |
486 | ||
487 | sma->sem_perm.security = NULL; | |
488 | retval = security_sem_alloc(sma); | |
489 | if (retval) { | |
1b4654ef | 490 | __sem_free(sma); |
1da177e4 LT |
491 | return retval; |
492 | } | |
493 | ||
6062a8dc | 494 | for (i = 0; i < nsems; i++) { |
1a233956 MS |
495 | INIT_LIST_HEAD(&sma->sems[i].pending_alter); |
496 | INIT_LIST_HEAD(&sma->sems[i].pending_const); | |
497 | spin_lock_init(&sma->sems[i].lock); | |
6062a8dc | 498 | } |
b97e820f MS |
499 | |
500 | sma->complex_count = 0; | |
9de5ab8a | 501 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; |
1a82e9e1 MS |
502 | INIT_LIST_HEAD(&sma->pending_alter); |
503 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 504 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
505 | sma->sem_nsems = nsems; |
506 | sma->sem_ctime = get_seconds(); | |
e8577d1f MS |
507 | |
508 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); | |
509 | if (id < 0) { | |
dba4cdd3 | 510 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
e8577d1f MS |
511 | return id; |
512 | } | |
513 | ns->used_sems += nsems; | |
514 | ||
6062a8dc | 515 | sem_unlock(sma, -1); |
6d49dab8 | 516 | rcu_read_unlock(); |
1da177e4 | 517 | |
7ca7e564 | 518 | return sma->sem_perm.id; |
1da177e4 LT |
519 | } |
520 | ||
7748dbfa | 521 | |
f4566f04 | 522 | /* |
d9a605e4 | 523 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 524 | */ |
03f02c76 | 525 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 526 | { |
03f02c76 ND |
527 | struct sem_array *sma; |
528 | ||
529 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
530 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
531 | } |
532 | ||
f4566f04 | 533 | /* |
d9a605e4 | 534 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 535 | */ |
03f02c76 ND |
536 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
537 | struct ipc_params *params) | |
7748dbfa | 538 | { |
03f02c76 ND |
539 | struct sem_array *sma; |
540 | ||
541 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
542 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
543 | return -EINVAL; |
544 | ||
545 | return 0; | |
546 | } | |
547 | ||
d5460c99 | 548 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 549 | { |
e3893534 | 550 | struct ipc_namespace *ns; |
eb66ec44 MK |
551 | static const struct ipc_ops sem_ops = { |
552 | .getnew = newary, | |
553 | .associate = sem_security, | |
554 | .more_checks = sem_more_checks, | |
555 | }; | |
7748dbfa | 556 | struct ipc_params sem_params; |
e3893534 KK |
557 | |
558 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 559 | |
e3893534 | 560 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 561 | return -EINVAL; |
7ca7e564 | 562 | |
7748dbfa ND |
563 | sem_params.key = key; |
564 | sem_params.flg = semflg; | |
565 | sem_params.u.nsems = nsems; | |
1da177e4 | 566 | |
7748dbfa | 567 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
568 | } |
569 | ||
78f5009c | 570 | /** |
4ce33ec2 DB |
571 | * perform_atomic_semop[_slow] - Attempt to perform semaphore |
572 | * operations on a given array. | |
758a6ba3 | 573 | * @sma: semaphore array |
d198cd6d | 574 | * @q: struct sem_queue that describes the operation |
758a6ba3 | 575 | * |
4ce33ec2 DB |
576 | * Caller blocking are as follows, based the value |
577 | * indicated by the semaphore operation (sem_op): | |
578 | * | |
579 | * (1) >0 never blocks. | |
580 | * (2) 0 (wait-for-zero operation): semval is non-zero. | |
581 | * (3) <0 attempting to decrement semval to a value smaller than zero. | |
582 | * | |
758a6ba3 MS |
583 | * Returns 0 if the operation was possible. |
584 | * Returns 1 if the operation is impossible, the caller must sleep. | |
4ce33ec2 | 585 | * Returns <0 for error codes. |
1da177e4 | 586 | */ |
4ce33ec2 | 587 | static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q) |
1da177e4 | 588 | { |
d198cd6d | 589 | int result, sem_op, nsops, pid; |
1da177e4 | 590 | struct sembuf *sop; |
239521f3 | 591 | struct sem *curr; |
d198cd6d MS |
592 | struct sembuf *sops; |
593 | struct sem_undo *un; | |
594 | ||
595 | sops = q->sops; | |
596 | nsops = q->nsops; | |
597 | un = q->undo; | |
1da177e4 LT |
598 | |
599 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 600 | curr = &sma->sems[sop->sem_num]; |
1da177e4 LT |
601 | sem_op = sop->sem_op; |
602 | result = curr->semval; | |
78f5009c | 603 | |
1da177e4 LT |
604 | if (!sem_op && result) |
605 | goto would_block; | |
606 | ||
607 | result += sem_op; | |
608 | if (result < 0) | |
609 | goto would_block; | |
610 | if (result > SEMVMX) | |
611 | goto out_of_range; | |
78f5009c | 612 | |
1da177e4 LT |
613 | if (sop->sem_flg & SEM_UNDO) { |
614 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 615 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
616 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
617 | goto out_of_range; | |
78f5009c | 618 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 619 | } |
78f5009c | 620 | |
1da177e4 LT |
621 | curr->semval = result; |
622 | } | |
623 | ||
624 | sop--; | |
d198cd6d | 625 | pid = q->pid; |
1da177e4 | 626 | while (sop >= sops) { |
1a233956 | 627 | sma->sems[sop->sem_num].sempid = pid; |
1da177e4 LT |
628 | sop--; |
629 | } | |
78f5009c | 630 | |
1da177e4 LT |
631 | return 0; |
632 | ||
633 | out_of_range: | |
634 | result = -ERANGE; | |
635 | goto undo; | |
636 | ||
637 | would_block: | |
ed247b7c MS |
638 | q->blocking = sop; |
639 | ||
1da177e4 LT |
640 | if (sop->sem_flg & IPC_NOWAIT) |
641 | result = -EAGAIN; | |
642 | else | |
643 | result = 1; | |
644 | ||
645 | undo: | |
646 | sop--; | |
647 | while (sop >= sops) { | |
78f5009c | 648 | sem_op = sop->sem_op; |
1a233956 | 649 | sma->sems[sop->sem_num].semval -= sem_op; |
78f5009c PM |
650 | if (sop->sem_flg & SEM_UNDO) |
651 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
652 | sop--; |
653 | } | |
654 | ||
655 | return result; | |
656 | } | |
657 | ||
4ce33ec2 DB |
658 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) |
659 | { | |
660 | int result, sem_op, nsops; | |
661 | struct sembuf *sop; | |
662 | struct sem *curr; | |
663 | struct sembuf *sops; | |
664 | struct sem_undo *un; | |
665 | ||
666 | sops = q->sops; | |
667 | nsops = q->nsops; | |
668 | un = q->undo; | |
669 | ||
670 | if (unlikely(q->dupsop)) | |
671 | return perform_atomic_semop_slow(sma, q); | |
672 | ||
673 | /* | |
674 | * We scan the semaphore set twice, first to ensure that the entire | |
675 | * operation can succeed, therefore avoiding any pointless writes | |
676 | * to shared memory and having to undo such changes in order to block | |
677 | * until the operations can go through. | |
678 | */ | |
679 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 680 | curr = &sma->sems[sop->sem_num]; |
4ce33ec2 DB |
681 | sem_op = sop->sem_op; |
682 | result = curr->semval; | |
683 | ||
684 | if (!sem_op && result) | |
685 | goto would_block; /* wait-for-zero */ | |
686 | ||
687 | result += sem_op; | |
688 | if (result < 0) | |
689 | goto would_block; | |
690 | ||
691 | if (result > SEMVMX) | |
692 | return -ERANGE; | |
693 | ||
694 | if (sop->sem_flg & SEM_UNDO) { | |
695 | int undo = un->semadj[sop->sem_num] - sem_op; | |
696 | ||
697 | /* Exceeding the undo range is an error. */ | |
698 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
699 | return -ERANGE; | |
700 | } | |
701 | } | |
702 | ||
703 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 704 | curr = &sma->sems[sop->sem_num]; |
4ce33ec2 DB |
705 | sem_op = sop->sem_op; |
706 | result = curr->semval; | |
707 | ||
708 | if (sop->sem_flg & SEM_UNDO) { | |
709 | int undo = un->semadj[sop->sem_num] - sem_op; | |
710 | ||
711 | un->semadj[sop->sem_num] = undo; | |
712 | } | |
713 | curr->semval += sem_op; | |
714 | curr->sempid = q->pid; | |
715 | } | |
716 | ||
717 | return 0; | |
718 | ||
719 | would_block: | |
720 | q->blocking = sop; | |
721 | return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1; | |
722 | } | |
723 | ||
9ae949fa DB |
724 | static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error, |
725 | struct wake_q_head *wake_q) | |
0a2b9d4c | 726 | { |
9ae949fa DB |
727 | wake_q_add(wake_q, q->sleeper); |
728 | /* | |
729 | * Rely on the above implicit barrier, such that we can | |
730 | * ensure that we hold reference to the task before setting | |
731 | * q->status. Otherwise we could race with do_exit if the | |
732 | * task is awoken by an external event before calling | |
733 | * wake_up_process(). | |
734 | */ | |
735 | WRITE_ONCE(q->status, error); | |
d4212093 NP |
736 | } |
737 | ||
b97e820f MS |
738 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
739 | { | |
740 | list_del(&q->list); | |
9f1bc2c9 | 741 | if (q->nsops > 1) |
b97e820f MS |
742 | sma->complex_count--; |
743 | } | |
744 | ||
fd5db422 MS |
745 | /** check_restart(sma, q) |
746 | * @sma: semaphore array | |
747 | * @q: the operation that just completed | |
748 | * | |
749 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
750 | * waiting operations. Therefore this function checks if the restart is | |
751 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
752 | * modified the array. |
753 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 | 754 | */ |
4663d3e8 | 755 | static inline int check_restart(struct sem_array *sma, struct sem_queue *q) |
fd5db422 | 756 | { |
1a82e9e1 MS |
757 | /* pending complex alter operations are too difficult to analyse */ |
758 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
759 | return 1; |
760 | ||
761 | /* we were a sleeping complex operation. Too difficult */ | |
762 | if (q->nsops > 1) | |
763 | return 1; | |
764 | ||
1a82e9e1 MS |
765 | /* It is impossible that someone waits for the new value: |
766 | * - complex operations always restart. | |
767 | * - wait-for-zero are handled seperately. | |
768 | * - q is a previously sleeping simple operation that | |
769 | * altered the array. It must be a decrement, because | |
770 | * simple increments never sleep. | |
771 | * - If there are older (higher priority) decrements | |
772 | * in the queue, then they have observed the original | |
773 | * semval value and couldn't proceed. The operation | |
774 | * decremented to value - thus they won't proceed either. | |
775 | */ | |
776 | return 0; | |
777 | } | |
fd5db422 | 778 | |
1a82e9e1 | 779 | /** |
8001c858 | 780 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
781 | * @sma: semaphore array. |
782 | * @semnum: semaphore that was modified. | |
9ae949fa | 783 | * @wake_q: lockless wake-queue head. |
1a82e9e1 MS |
784 | * |
785 | * wake_const_ops must be called after a semaphore in a semaphore array | |
786 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
787 | * be called with semnum = -1, as well as with the number of each modified | |
788 | * semaphore. | |
9ae949fa | 789 | * The tasks that must be woken up are added to @wake_q. The return code |
1a82e9e1 MS |
790 | * is stored in q->pid. |
791 | * The function returns 1 if at least one operation was completed successfully. | |
792 | */ | |
793 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
9ae949fa | 794 | struct wake_q_head *wake_q) |
1a82e9e1 | 795 | { |
f150f02c | 796 | struct sem_queue *q, *tmp; |
1a82e9e1 MS |
797 | struct list_head *pending_list; |
798 | int semop_completed = 0; | |
799 | ||
800 | if (semnum == -1) | |
801 | pending_list = &sma->pending_const; | |
802 | else | |
1a233956 | 803 | pending_list = &sma->sems[semnum].pending_const; |
fd5db422 | 804 | |
f150f02c DB |
805 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
806 | int error = perform_atomic_semop(sma, q); | |
1a82e9e1 | 807 | |
f150f02c DB |
808 | if (error > 0) |
809 | continue; | |
810 | /* operation completed, remove from queue & wakeup */ | |
811 | unlink_queue(sma, q); | |
1a82e9e1 | 812 | |
f150f02c DB |
813 | wake_up_sem_queue_prepare(q, error, wake_q); |
814 | if (error == 0) | |
815 | semop_completed = 1; | |
1a82e9e1 | 816 | } |
f150f02c | 817 | |
1a82e9e1 MS |
818 | return semop_completed; |
819 | } | |
820 | ||
821 | /** | |
8001c858 | 822 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
823 | * @sma: semaphore array |
824 | * @sops: operations that were performed | |
825 | * @nsops: number of operations | |
9ae949fa | 826 | * @wake_q: lockless wake-queue head |
1a82e9e1 | 827 | * |
8001c858 DB |
828 | * Checks all required queue for wait-for-zero operations, based |
829 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
830 | * The function returns 1 if at least one operation was completed successfully. |
831 | */ | |
832 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
9ae949fa | 833 | int nsops, struct wake_q_head *wake_q) |
1a82e9e1 MS |
834 | { |
835 | int i; | |
836 | int semop_completed = 0; | |
837 | int got_zero = 0; | |
838 | ||
839 | /* first: the per-semaphore queues, if known */ | |
840 | if (sops) { | |
841 | for (i = 0; i < nsops; i++) { | |
842 | int num = sops[i].sem_num; | |
843 | ||
1a233956 | 844 | if (sma->sems[num].semval == 0) { |
1a82e9e1 | 845 | got_zero = 1; |
9ae949fa | 846 | semop_completed |= wake_const_ops(sma, num, wake_q); |
1a82e9e1 MS |
847 | } |
848 | } | |
849 | } else { | |
850 | /* | |
851 | * No sops means modified semaphores not known. | |
852 | * Assume all were changed. | |
fd5db422 | 853 | */ |
1a82e9e1 | 854 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 855 | if (sma->sems[i].semval == 0) { |
1a82e9e1 | 856 | got_zero = 1; |
9ae949fa | 857 | semop_completed |= wake_const_ops(sma, i, wake_q); |
1a82e9e1 MS |
858 | } |
859 | } | |
fd5db422 MS |
860 | } |
861 | /* | |
1a82e9e1 MS |
862 | * If one of the modified semaphores got 0, |
863 | * then check the global queue, too. | |
fd5db422 | 864 | */ |
1a82e9e1 | 865 | if (got_zero) |
9ae949fa | 866 | semop_completed |= wake_const_ops(sma, -1, wake_q); |
fd5db422 | 867 | |
1a82e9e1 | 868 | return semop_completed; |
fd5db422 MS |
869 | } |
870 | ||
636c6be8 MS |
871 | |
872 | /** | |
8001c858 | 873 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
874 | * @sma: semaphore array. |
875 | * @semnum: semaphore that was modified. | |
9ae949fa | 876 | * @wake_q: lockless wake-queue head. |
636c6be8 MS |
877 | * |
878 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
879 | * was modified. If multiple semaphores were modified, update_queue must |
880 | * be called with semnum = -1, as well as with the number of each modified | |
881 | * semaphore. | |
9ae949fa | 882 | * The tasks that must be woken up are added to @wake_q. The return code |
0a2b9d4c | 883 | * is stored in q->pid. |
1a82e9e1 MS |
884 | * The function internally checks if const operations can now succeed. |
885 | * | |
0a2b9d4c | 886 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 887 | */ |
9ae949fa | 888 | static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q) |
1da177e4 | 889 | { |
f150f02c | 890 | struct sem_queue *q, *tmp; |
636c6be8 | 891 | struct list_head *pending_list; |
0a2b9d4c | 892 | int semop_completed = 0; |
636c6be8 | 893 | |
9f1bc2c9 | 894 | if (semnum == -1) |
1a82e9e1 | 895 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 896 | else |
1a233956 | 897 | pending_list = &sma->sems[semnum].pending_alter; |
9cad200c NP |
898 | |
899 | again: | |
f150f02c | 900 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
fd5db422 | 901 | int error, restart; |
636c6be8 | 902 | |
d987f8b2 MS |
903 | /* If we are scanning the single sop, per-semaphore list of |
904 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 905 | * necessary to scan further: simple increments |
d987f8b2 MS |
906 | * that affect only one entry succeed immediately and cannot |
907 | * be in the per semaphore pending queue, and decrements | |
908 | * cannot be successful if the value is already 0. | |
909 | */ | |
1a233956 | 910 | if (semnum != -1 && sma->sems[semnum].semval == 0) |
d987f8b2 MS |
911 | break; |
912 | ||
d198cd6d | 913 | error = perform_atomic_semop(sma, q); |
1da177e4 LT |
914 | |
915 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
916 | if (error > 0) |
917 | continue; | |
918 | ||
b97e820f | 919 | unlink_queue(sma, q); |
9cad200c | 920 | |
0a2b9d4c | 921 | if (error) { |
fd5db422 | 922 | restart = 0; |
0a2b9d4c MS |
923 | } else { |
924 | semop_completed = 1; | |
9ae949fa | 925 | do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q); |
fd5db422 | 926 | restart = check_restart(sma, q); |
0a2b9d4c | 927 | } |
fd5db422 | 928 | |
9ae949fa | 929 | wake_up_sem_queue_prepare(q, error, wake_q); |
fd5db422 | 930 | if (restart) |
9cad200c | 931 | goto again; |
1da177e4 | 932 | } |
0a2b9d4c | 933 | return semop_completed; |
1da177e4 LT |
934 | } |
935 | ||
0e8c6656 | 936 | /** |
8001c858 | 937 | * set_semotime - set sem_otime |
0e8c6656 MS |
938 | * @sma: semaphore array |
939 | * @sops: operations that modified the array, may be NULL | |
940 | * | |
941 | * sem_otime is replicated to avoid cache line trashing. | |
942 | * This function sets one instance to the current time. | |
943 | */ | |
944 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
945 | { | |
946 | if (sops == NULL) { | |
1a233956 | 947 | sma->sems[0].sem_otime = get_seconds(); |
0e8c6656 | 948 | } else { |
1a233956 | 949 | sma->sems[sops[0].sem_num].sem_otime = |
0e8c6656 MS |
950 | get_seconds(); |
951 | } | |
952 | } | |
953 | ||
0a2b9d4c | 954 | /** |
8001c858 | 955 | * do_smart_update - optimized update_queue |
fd5db422 MS |
956 | * @sma: semaphore array |
957 | * @sops: operations that were performed | |
958 | * @nsops: number of operations | |
0a2b9d4c | 959 | * @otime: force setting otime |
9ae949fa | 960 | * @wake_q: lockless wake-queue head |
fd5db422 | 961 | * |
1a82e9e1 MS |
962 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
963 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c | 964 | * Note that the function does not do the actual wake-up: the caller is |
9ae949fa | 965 | * responsible for calling wake_up_q(). |
0a2b9d4c | 966 | * It is safe to perform this call after dropping all locks. |
fd5db422 | 967 | */ |
0a2b9d4c | 968 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
9ae949fa | 969 | int otime, struct wake_q_head *wake_q) |
fd5db422 MS |
970 | { |
971 | int i; | |
972 | ||
9ae949fa | 973 | otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q); |
1a82e9e1 | 974 | |
f269f40a MS |
975 | if (!list_empty(&sma->pending_alter)) { |
976 | /* semaphore array uses the global queue - just process it. */ | |
9ae949fa | 977 | otime |= update_queue(sma, -1, wake_q); |
f269f40a MS |
978 | } else { |
979 | if (!sops) { | |
980 | /* | |
981 | * No sops, thus the modified semaphores are not | |
982 | * known. Check all. | |
983 | */ | |
984 | for (i = 0; i < sma->sem_nsems; i++) | |
9ae949fa | 985 | otime |= update_queue(sma, i, wake_q); |
f269f40a MS |
986 | } else { |
987 | /* | |
988 | * Check the semaphores that were increased: | |
989 | * - No complex ops, thus all sleeping ops are | |
990 | * decrease. | |
991 | * - if we decreased the value, then any sleeping | |
992 | * semaphore ops wont be able to run: If the | |
993 | * previous value was too small, then the new | |
994 | * value will be too small, too. | |
995 | */ | |
996 | for (i = 0; i < nsops; i++) { | |
997 | if (sops[i].sem_op > 0) { | |
998 | otime |= update_queue(sma, | |
9ae949fa | 999 | sops[i].sem_num, wake_q); |
f269f40a | 1000 | } |
ab465df9 | 1001 | } |
9f1bc2c9 | 1002 | } |
fd5db422 | 1003 | } |
0e8c6656 MS |
1004 | if (otime) |
1005 | set_semotime(sma, sops); | |
fd5db422 MS |
1006 | } |
1007 | ||
2f2ed41d | 1008 | /* |
b220c57a | 1009 | * check_qop: Test if a queued operation sleeps on the semaphore semnum |
2f2ed41d MS |
1010 | */ |
1011 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1012 | bool count_zero) | |
1013 | { | |
b220c57a | 1014 | struct sembuf *sop = q->blocking; |
2f2ed41d | 1015 | |
9b44ee2e MS |
1016 | /* |
1017 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1018 | * semaphores. This violates SUS, therefore it was changed to the | |
1019 | * standard compliant behavior. | |
1020 | * Give the administrators a chance to notice that an application | |
1021 | * might misbehave because it relies on the Linux behavior. | |
1022 | */ | |
1023 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1024 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1025 | current->comm, task_pid_nr(current)); | |
1026 | ||
b220c57a MS |
1027 | if (sop->sem_num != semnum) |
1028 | return 0; | |
2f2ed41d | 1029 | |
b220c57a MS |
1030 | if (count_zero && sop->sem_op == 0) |
1031 | return 1; | |
1032 | if (!count_zero && sop->sem_op < 0) | |
1033 | return 1; | |
1034 | ||
1035 | return 0; | |
2f2ed41d MS |
1036 | } |
1037 | ||
1da177e4 LT |
1038 | /* The following counts are associated to each semaphore: |
1039 | * semncnt number of tasks waiting on semval being nonzero | |
1040 | * semzcnt number of tasks waiting on semval being zero | |
b220c57a MS |
1041 | * |
1042 | * Per definition, a task waits only on the semaphore of the first semop | |
1043 | * that cannot proceed, even if additional operation would block, too. | |
1da177e4 | 1044 | */ |
2f2ed41d MS |
1045 | static int count_semcnt(struct sem_array *sma, ushort semnum, |
1046 | bool count_zero) | |
1da177e4 | 1047 | { |
2f2ed41d | 1048 | struct list_head *l; |
239521f3 | 1049 | struct sem_queue *q; |
2f2ed41d | 1050 | int semcnt; |
1da177e4 | 1051 | |
2f2ed41d MS |
1052 | semcnt = 0; |
1053 | /* First: check the simple operations. They are easy to evaluate */ | |
1054 | if (count_zero) | |
1a233956 | 1055 | l = &sma->sems[semnum].pending_const; |
2f2ed41d | 1056 | else |
1a233956 | 1057 | l = &sma->sems[semnum].pending_alter; |
1da177e4 | 1058 | |
2f2ed41d MS |
1059 | list_for_each_entry(q, l, list) { |
1060 | /* all task on a per-semaphore list sleep on exactly | |
1061 | * that semaphore | |
1062 | */ | |
1063 | semcnt++; | |
ebc2e5e6 RR |
1064 | } |
1065 | ||
2f2ed41d | 1066 | /* Then: check the complex operations. */ |
1994862d | 1067 | list_for_each_entry(q, &sma->pending_alter, list) { |
2f2ed41d MS |
1068 | semcnt += check_qop(sma, semnum, q, count_zero); |
1069 | } | |
1070 | if (count_zero) { | |
1071 | list_for_each_entry(q, &sma->pending_const, list) { | |
1072 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1073 | } | |
1994862d | 1074 | } |
2f2ed41d | 1075 | return semcnt; |
1da177e4 LT |
1076 | } |
1077 | ||
d9a605e4 DB |
1078 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1079 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1080 | * remains locked on exit. |
1da177e4 | 1081 | */ |
01b8b07a | 1082 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1083 | { |
380af1b3 MS |
1084 | struct sem_undo *un, *tu; |
1085 | struct sem_queue *q, *tq; | |
01b8b07a | 1086 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
9f1bc2c9 | 1087 | int i; |
9ae949fa | 1088 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1089 | |
380af1b3 | 1090 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1091 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1092 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1093 | list_del(&un->list_id); | |
1094 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1095 | un->semid = -1; |
380af1b3 MS |
1096 | list_del_rcu(&un->list_proc); |
1097 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1098 | kfree_rcu(un, rcu); |
380af1b3 | 1099 | } |
1da177e4 LT |
1100 | |
1101 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
1a82e9e1 MS |
1102 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1103 | unlink_queue(sma, q); | |
9ae949fa | 1104 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1105 | } |
1106 | ||
1107 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1108 | unlink_queue(sma, q); |
9ae949fa | 1109 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1da177e4 | 1110 | } |
9f1bc2c9 | 1111 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 1112 | struct sem *sem = &sma->sems[i]; |
1a82e9e1 MS |
1113 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1114 | unlink_queue(sma, q); | |
9ae949fa | 1115 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1116 | } |
1117 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 | 1118 | unlink_queue(sma, q); |
9ae949fa | 1119 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
9f1bc2c9 RR |
1120 | } |
1121 | } | |
1da177e4 | 1122 | |
7ca7e564 ND |
1123 | /* Remove the semaphore set from the IDR */ |
1124 | sem_rmid(ns, sma); | |
6062a8dc | 1125 | sem_unlock(sma, -1); |
6d49dab8 | 1126 | rcu_read_unlock(); |
1da177e4 | 1127 | |
9ae949fa | 1128 | wake_up_q(&wake_q); |
e3893534 | 1129 | ns->used_sems -= sma->sem_nsems; |
dba4cdd3 | 1130 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1131 | } |
1132 | ||
1133 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1134 | { | |
239521f3 | 1135 | switch (version) { |
1da177e4 LT |
1136 | case IPC_64: |
1137 | return copy_to_user(buf, in, sizeof(*in)); | |
1138 | case IPC_OLD: | |
1139 | { | |
1140 | struct semid_ds out; | |
1141 | ||
982f7c2b DR |
1142 | memset(&out, 0, sizeof(out)); |
1143 | ||
1da177e4 LT |
1144 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1145 | ||
1146 | out.sem_otime = in->sem_otime; | |
1147 | out.sem_ctime = in->sem_ctime; | |
1148 | out.sem_nsems = in->sem_nsems; | |
1149 | ||
1150 | return copy_to_user(buf, &out, sizeof(out)); | |
1151 | } | |
1152 | default: | |
1153 | return -EINVAL; | |
1154 | } | |
1155 | } | |
1156 | ||
d12e1e50 MS |
1157 | static time_t get_semotime(struct sem_array *sma) |
1158 | { | |
1159 | int i; | |
1160 | time_t res; | |
1161 | ||
1a233956 | 1162 | res = sma->sems[0].sem_otime; |
d12e1e50 | 1163 | for (i = 1; i < sma->sem_nsems; i++) { |
1a233956 | 1164 | time_t to = sma->sems[i].sem_otime; |
d12e1e50 MS |
1165 | |
1166 | if (to > res) | |
1167 | res = to; | |
1168 | } | |
1169 | return res; | |
1170 | } | |
1171 | ||
4b9fcb0e | 1172 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1173 | int cmd, int version, void __user *p) |
1da177e4 | 1174 | { |
e5cc9c7b | 1175 | int err; |
1da177e4 LT |
1176 | struct sem_array *sma; |
1177 | ||
239521f3 | 1178 | switch (cmd) { |
1da177e4 LT |
1179 | case IPC_INFO: |
1180 | case SEM_INFO: | |
1181 | { | |
1182 | struct seminfo seminfo; | |
1183 | int max_id; | |
1184 | ||
1185 | err = security_sem_semctl(NULL, cmd); | |
1186 | if (err) | |
1187 | return err; | |
46c0a8ca | 1188 | |
239521f3 | 1189 | memset(&seminfo, 0, sizeof(seminfo)); |
e3893534 KK |
1190 | seminfo.semmni = ns->sc_semmni; |
1191 | seminfo.semmns = ns->sc_semmns; | |
1192 | seminfo.semmsl = ns->sc_semmsl; | |
1193 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1194 | seminfo.semvmx = SEMVMX; |
1195 | seminfo.semmnu = SEMMNU; | |
1196 | seminfo.semmap = SEMMAP; | |
1197 | seminfo.semume = SEMUME; | |
d9a605e4 | 1198 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1199 | if (cmd == SEM_INFO) { |
e3893534 KK |
1200 | seminfo.semusz = sem_ids(ns).in_use; |
1201 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1202 | } else { |
1203 | seminfo.semusz = SEMUSZ; | |
1204 | seminfo.semaem = SEMAEM; | |
1205 | } | |
7ca7e564 | 1206 | max_id = ipc_get_maxid(&sem_ids(ns)); |
d9a605e4 | 1207 | up_read(&sem_ids(ns).rwsem); |
46c0a8ca | 1208 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 | 1209 | return -EFAULT; |
239521f3 | 1210 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 1211 | } |
4b9fcb0e | 1212 | case IPC_STAT: |
1da177e4 LT |
1213 | case SEM_STAT: |
1214 | { | |
1215 | struct semid64_ds tbuf; | |
16df3674 DB |
1216 | int id = 0; |
1217 | ||
1218 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1219 | |
941b0304 | 1220 | rcu_read_lock(); |
4b9fcb0e | 1221 | if (cmd == SEM_STAT) { |
16df3674 DB |
1222 | sma = sem_obtain_object(ns, semid); |
1223 | if (IS_ERR(sma)) { | |
1224 | err = PTR_ERR(sma); | |
1225 | goto out_unlock; | |
1226 | } | |
4b9fcb0e PP |
1227 | id = sma->sem_perm.id; |
1228 | } else { | |
16df3674 DB |
1229 | sma = sem_obtain_object_check(ns, semid); |
1230 | if (IS_ERR(sma)) { | |
1231 | err = PTR_ERR(sma); | |
1232 | goto out_unlock; | |
1233 | } | |
4b9fcb0e | 1234 | } |
1da177e4 LT |
1235 | |
1236 | err = -EACCES; | |
b0e77598 | 1237 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1238 | goto out_unlock; |
1239 | ||
1240 | err = security_sem_semctl(sma, cmd); | |
1241 | if (err) | |
1242 | goto out_unlock; | |
1243 | ||
1da177e4 | 1244 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
d12e1e50 MS |
1245 | tbuf.sem_otime = get_semotime(sma); |
1246 | tbuf.sem_ctime = sma->sem_ctime; | |
1247 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1248 | rcu_read_unlock(); |
e1fd1f49 | 1249 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1250 | return -EFAULT; |
1251 | return id; | |
1252 | } | |
1253 | default: | |
1254 | return -EINVAL; | |
1255 | } | |
1da177e4 | 1256 | out_unlock: |
16df3674 | 1257 | rcu_read_unlock(); |
1da177e4 LT |
1258 | return err; |
1259 | } | |
1260 | ||
e1fd1f49 AV |
1261 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1262 | unsigned long arg) | |
1263 | { | |
1264 | struct sem_undo *un; | |
1265 | struct sem_array *sma; | |
239521f3 | 1266 | struct sem *curr; |
9ae949fa DB |
1267 | int err, val; |
1268 | DEFINE_WAKE_Q(wake_q); | |
1269 | ||
e1fd1f49 AV |
1270 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) |
1271 | /* big-endian 64bit */ | |
1272 | val = arg >> 32; | |
1273 | #else | |
1274 | /* 32bit or little-endian 64bit */ | |
1275 | val = arg; | |
1276 | #endif | |
1277 | ||
6062a8dc RR |
1278 | if (val > SEMVMX || val < 0) |
1279 | return -ERANGE; | |
e1fd1f49 | 1280 | |
6062a8dc RR |
1281 | rcu_read_lock(); |
1282 | sma = sem_obtain_object_check(ns, semid); | |
1283 | if (IS_ERR(sma)) { | |
1284 | rcu_read_unlock(); | |
1285 | return PTR_ERR(sma); | |
1286 | } | |
1287 | ||
1288 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1289 | rcu_read_unlock(); | |
1290 | return -EINVAL; | |
1291 | } | |
1292 | ||
1293 | ||
1294 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1295 | rcu_read_unlock(); | |
1296 | return -EACCES; | |
1297 | } | |
e1fd1f49 AV |
1298 | |
1299 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1300 | if (err) { |
1301 | rcu_read_unlock(); | |
1302 | return -EACCES; | |
1303 | } | |
e1fd1f49 | 1304 | |
6062a8dc | 1305 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1306 | |
0f3d2b01 | 1307 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1308 | sem_unlock(sma, -1); |
1309 | rcu_read_unlock(); | |
1310 | return -EIDRM; | |
1311 | } | |
1312 | ||
1a233956 | 1313 | curr = &sma->sems[semnum]; |
e1fd1f49 | 1314 | |
cf9d5d78 | 1315 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1316 | list_for_each_entry(un, &sma->list_id, list_id) |
1317 | un->semadj[semnum] = 0; | |
1318 | ||
1319 | curr->semval = val; | |
1320 | curr->sempid = task_tgid_vnr(current); | |
1321 | sma->sem_ctime = get_seconds(); | |
1322 | /* maybe some queued-up processes were waiting for this */ | |
9ae949fa | 1323 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
6062a8dc | 1324 | sem_unlock(sma, -1); |
6d49dab8 | 1325 | rcu_read_unlock(); |
9ae949fa | 1326 | wake_up_q(&wake_q); |
6062a8dc | 1327 | return 0; |
e1fd1f49 AV |
1328 | } |
1329 | ||
e3893534 | 1330 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1331 | int cmd, void __user *p) |
1da177e4 LT |
1332 | { |
1333 | struct sem_array *sma; | |
239521f3 | 1334 | struct sem *curr; |
16df3674 | 1335 | int err, nsems; |
1da177e4 | 1336 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1337 | ushort *sem_io = fast_sem_io; |
9ae949fa | 1338 | DEFINE_WAKE_Q(wake_q); |
16df3674 DB |
1339 | |
1340 | rcu_read_lock(); | |
1341 | sma = sem_obtain_object_check(ns, semid); | |
1342 | if (IS_ERR(sma)) { | |
1343 | rcu_read_unlock(); | |
023a5355 | 1344 | return PTR_ERR(sma); |
16df3674 | 1345 | } |
1da177e4 LT |
1346 | |
1347 | nsems = sma->sem_nsems; | |
1348 | ||
1da177e4 | 1349 | err = -EACCES; |
c728b9c8 LT |
1350 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1351 | goto out_rcu_wakeup; | |
1da177e4 LT |
1352 | |
1353 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1354 | if (err) |
1355 | goto out_rcu_wakeup; | |
1da177e4 LT |
1356 | |
1357 | err = -EACCES; | |
1358 | switch (cmd) { | |
1359 | case GETALL: | |
1360 | { | |
e1fd1f49 | 1361 | ushort __user *array = p; |
1da177e4 LT |
1362 | int i; |
1363 | ||
ce857229 | 1364 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1365 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1366 | err = -EIDRM; |
1367 | goto out_unlock; | |
1368 | } | |
239521f3 | 1369 | if (nsems > SEMMSL_FAST) { |
dba4cdd3 | 1370 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
ce857229 | 1371 | err = -EIDRM; |
6e224f94 | 1372 | goto out_unlock; |
ce857229 AV |
1373 | } |
1374 | sem_unlock(sma, -1); | |
6d49dab8 | 1375 | rcu_read_unlock(); |
f8dbe8d2 KC |
1376 | sem_io = kvmalloc_array(nsems, sizeof(ushort), |
1377 | GFP_KERNEL); | |
239521f3 | 1378 | if (sem_io == NULL) { |
dba4cdd3 | 1379 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1380 | return -ENOMEM; |
1381 | } | |
1382 | ||
4091fd94 | 1383 | rcu_read_lock(); |
6ff37972 | 1384 | sem_lock_and_putref(sma); |
0f3d2b01 | 1385 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1386 | err = -EIDRM; |
6e224f94 | 1387 | goto out_unlock; |
1da177e4 | 1388 | } |
ce857229 | 1389 | } |
1da177e4 | 1390 | for (i = 0; i < sma->sem_nsems; i++) |
1a233956 | 1391 | sem_io[i] = sma->sems[i].semval; |
6062a8dc | 1392 | sem_unlock(sma, -1); |
6d49dab8 | 1393 | rcu_read_unlock(); |
1da177e4 | 1394 | err = 0; |
239521f3 | 1395 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1396 | err = -EFAULT; |
1397 | goto out_free; | |
1398 | } | |
1399 | case SETALL: | |
1400 | { | |
1401 | int i; | |
1402 | struct sem_undo *un; | |
1403 | ||
dba4cdd3 | 1404 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
6e224f94 MS |
1405 | err = -EIDRM; |
1406 | goto out_rcu_wakeup; | |
6062a8dc | 1407 | } |
16df3674 | 1408 | rcu_read_unlock(); |
1da177e4 | 1409 | |
239521f3 | 1410 | if (nsems > SEMMSL_FAST) { |
f8dbe8d2 KC |
1411 | sem_io = kvmalloc_array(nsems, sizeof(ushort), |
1412 | GFP_KERNEL); | |
239521f3 | 1413 | if (sem_io == NULL) { |
dba4cdd3 | 1414 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1415 | return -ENOMEM; |
1416 | } | |
1417 | } | |
1418 | ||
239521f3 | 1419 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
dba4cdd3 | 1420 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1421 | err = -EFAULT; |
1422 | goto out_free; | |
1423 | } | |
1424 | ||
1425 | for (i = 0; i < nsems; i++) { | |
1426 | if (sem_io[i] > SEMVMX) { | |
dba4cdd3 | 1427 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1428 | err = -ERANGE; |
1429 | goto out_free; | |
1430 | } | |
1431 | } | |
4091fd94 | 1432 | rcu_read_lock(); |
6ff37972 | 1433 | sem_lock_and_putref(sma); |
0f3d2b01 | 1434 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1435 | err = -EIDRM; |
6e224f94 | 1436 | goto out_unlock; |
1da177e4 LT |
1437 | } |
1438 | ||
a5f4db87 | 1439 | for (i = 0; i < nsems; i++) { |
1a233956 MS |
1440 | sma->sems[i].semval = sem_io[i]; |
1441 | sma->sems[i].sempid = task_tgid_vnr(current); | |
a5f4db87 | 1442 | } |
4daa28f6 | 1443 | |
cf9d5d78 | 1444 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1445 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1446 | for (i = 0; i < nsems; i++) |
1447 | un->semadj[i] = 0; | |
4daa28f6 | 1448 | } |
1da177e4 LT |
1449 | sma->sem_ctime = get_seconds(); |
1450 | /* maybe some queued-up processes were waiting for this */ | |
9ae949fa | 1451 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
1da177e4 LT |
1452 | err = 0; |
1453 | goto out_unlock; | |
1454 | } | |
e1fd1f49 | 1455 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1456 | } |
1457 | err = -EINVAL; | |
c728b9c8 LT |
1458 | if (semnum < 0 || semnum >= nsems) |
1459 | goto out_rcu_wakeup; | |
1da177e4 | 1460 | |
6062a8dc | 1461 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1462 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1463 | err = -EIDRM; |
1464 | goto out_unlock; | |
1465 | } | |
1a233956 | 1466 | curr = &sma->sems[semnum]; |
1da177e4 LT |
1467 | |
1468 | switch (cmd) { | |
1469 | case GETVAL: | |
1470 | err = curr->semval; | |
1471 | goto out_unlock; | |
1472 | case GETPID: | |
1473 | err = curr->sempid; | |
1474 | goto out_unlock; | |
1475 | case GETNCNT: | |
2f2ed41d | 1476 | err = count_semcnt(sma, semnum, 0); |
1da177e4 LT |
1477 | goto out_unlock; |
1478 | case GETZCNT: | |
2f2ed41d | 1479 | err = count_semcnt(sma, semnum, 1); |
1da177e4 | 1480 | goto out_unlock; |
1da177e4 | 1481 | } |
16df3674 | 1482 | |
1da177e4 | 1483 | out_unlock: |
6062a8dc | 1484 | sem_unlock(sma, -1); |
c728b9c8 | 1485 | out_rcu_wakeup: |
6d49dab8 | 1486 | rcu_read_unlock(); |
9ae949fa | 1487 | wake_up_q(&wake_q); |
1da177e4 | 1488 | out_free: |
239521f3 | 1489 | if (sem_io != fast_sem_io) |
f8dbe8d2 | 1490 | kvfree(sem_io); |
1da177e4 LT |
1491 | return err; |
1492 | } | |
1493 | ||
016d7132 PP |
1494 | static inline unsigned long |
1495 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1496 | { |
239521f3 | 1497 | switch (version) { |
1da177e4 | 1498 | case IPC_64: |
016d7132 | 1499 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1500 | return -EFAULT; |
1da177e4 | 1501 | return 0; |
1da177e4 LT |
1502 | case IPC_OLD: |
1503 | { | |
1504 | struct semid_ds tbuf_old; | |
1505 | ||
239521f3 | 1506 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1507 | return -EFAULT; |
1508 | ||
016d7132 PP |
1509 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1510 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1511 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1512 | |
1513 | return 0; | |
1514 | } | |
1515 | default: | |
1516 | return -EINVAL; | |
1517 | } | |
1518 | } | |
1519 | ||
522bb2a2 | 1520 | /* |
d9a605e4 | 1521 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1522 | * to be held in write mode. |
d9a605e4 | 1523 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1524 | */ |
21a4826a | 1525 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1526 | int cmd, int version, void __user *p) |
1da177e4 LT |
1527 | { |
1528 | struct sem_array *sma; | |
1529 | int err; | |
016d7132 | 1530 | struct semid64_ds semid64; |
1da177e4 LT |
1531 | struct kern_ipc_perm *ipcp; |
1532 | ||
239521f3 | 1533 | if (cmd == IPC_SET) { |
e1fd1f49 | 1534 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1535 | return -EFAULT; |
1da177e4 | 1536 | } |
073115d6 | 1537 | |
d9a605e4 | 1538 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1539 | rcu_read_lock(); |
1540 | ||
16df3674 DB |
1541 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1542 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1543 | if (IS_ERR(ipcp)) { |
1544 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1545 | goto out_unlock1; |
1546 | } | |
073115d6 | 1547 | |
a5f75e7f | 1548 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1549 | |
1550 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1551 | if (err) |
1552 | goto out_unlock1; | |
1da177e4 | 1553 | |
7b4cc5d8 | 1554 | switch (cmd) { |
1da177e4 | 1555 | case IPC_RMID: |
6062a8dc | 1556 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1557 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1558 | freeary(ns, ipcp); |
522bb2a2 | 1559 | goto out_up; |
1da177e4 | 1560 | case IPC_SET: |
6062a8dc | 1561 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1562 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1563 | if (err) | |
7b4cc5d8 | 1564 | goto out_unlock0; |
1da177e4 | 1565 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1566 | break; |
1567 | default: | |
1da177e4 | 1568 | err = -EINVAL; |
7b4cc5d8 | 1569 | goto out_unlock1; |
1da177e4 | 1570 | } |
1da177e4 | 1571 | |
7b4cc5d8 | 1572 | out_unlock0: |
6062a8dc | 1573 | sem_unlock(sma, -1); |
7b4cc5d8 | 1574 | out_unlock1: |
6d49dab8 | 1575 | rcu_read_unlock(); |
522bb2a2 | 1576 | out_up: |
d9a605e4 | 1577 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1578 | return err; |
1579 | } | |
1580 | ||
e1fd1f49 | 1581 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1582 | { |
1da177e4 | 1583 | int version; |
e3893534 | 1584 | struct ipc_namespace *ns; |
e1fd1f49 | 1585 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1586 | |
1587 | if (semid < 0) | |
1588 | return -EINVAL; | |
1589 | ||
1590 | version = ipc_parse_version(&cmd); | |
e3893534 | 1591 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1592 | |
239521f3 | 1593 | switch (cmd) { |
1da177e4 LT |
1594 | case IPC_INFO: |
1595 | case SEM_INFO: | |
4b9fcb0e | 1596 | case IPC_STAT: |
1da177e4 | 1597 | case SEM_STAT: |
e1fd1f49 | 1598 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1599 | case GETALL: |
1600 | case GETVAL: | |
1601 | case GETPID: | |
1602 | case GETNCNT: | |
1603 | case GETZCNT: | |
1da177e4 | 1604 | case SETALL: |
e1fd1f49 AV |
1605 | return semctl_main(ns, semid, semnum, cmd, p); |
1606 | case SETVAL: | |
1607 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1608 | case IPC_RMID: |
1609 | case IPC_SET: | |
e1fd1f49 | 1610 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1611 | default: |
1612 | return -EINVAL; | |
1613 | } | |
1614 | } | |
1615 | ||
1da177e4 LT |
1616 | /* If the task doesn't already have a undo_list, then allocate one |
1617 | * here. We guarantee there is only one thread using this undo list, | |
1618 | * and current is THE ONE | |
1619 | * | |
1620 | * If this allocation and assignment succeeds, but later | |
1621 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1622 | * Just let it stay associated with the task, and it'll be freed later | |
1623 | * at exit time. | |
1624 | * | |
1625 | * This can block, so callers must hold no locks. | |
1626 | */ | |
1627 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1628 | { | |
1629 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1630 | |
1631 | undo_list = current->sysvsem.undo_list; | |
1632 | if (!undo_list) { | |
2453a306 | 1633 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1634 | if (undo_list == NULL) |
1635 | return -ENOMEM; | |
00a5dfdb | 1636 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1637 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1638 | INIT_LIST_HEAD(&undo_list->list_proc); |
1639 | ||
1da177e4 LT |
1640 | current->sysvsem.undo_list = undo_list; |
1641 | } | |
1642 | *undo_listp = undo_list; | |
1643 | return 0; | |
1644 | } | |
1645 | ||
bf17bb71 | 1646 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1647 | { |
bf17bb71 | 1648 | struct sem_undo *un; |
4daa28f6 | 1649 | |
bf17bb71 NP |
1650 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1651 | if (un->semid == semid) | |
1652 | return un; | |
1da177e4 | 1653 | } |
4daa28f6 | 1654 | return NULL; |
1da177e4 LT |
1655 | } |
1656 | ||
bf17bb71 NP |
1657 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1658 | { | |
1659 | struct sem_undo *un; | |
1660 | ||
239521f3 | 1661 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1662 | |
1663 | un = __lookup_undo(ulp, semid); | |
1664 | if (un) { | |
1665 | list_del_rcu(&un->list_proc); | |
1666 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1667 | } | |
1668 | return un; | |
1669 | } | |
1670 | ||
4daa28f6 | 1671 | /** |
8001c858 | 1672 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1673 | * @ns: namespace |
1674 | * @semid: semaphore array id | |
1675 | * | |
1676 | * The function looks up (and if not present creates) the undo structure. | |
1677 | * The size of the undo structure depends on the size of the semaphore | |
1678 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1679 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1680 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1681 | */ |
1682 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1683 | { |
1684 | struct sem_array *sma; | |
1685 | struct sem_undo_list *ulp; | |
1686 | struct sem_undo *un, *new; | |
6062a8dc | 1687 | int nsems, error; |
1da177e4 LT |
1688 | |
1689 | error = get_undo_list(&ulp); | |
1690 | if (error) | |
1691 | return ERR_PTR(error); | |
1692 | ||
380af1b3 | 1693 | rcu_read_lock(); |
c530c6ac | 1694 | spin_lock(&ulp->lock); |
1da177e4 | 1695 | un = lookup_undo(ulp, semid); |
c530c6ac | 1696 | spin_unlock(&ulp->lock); |
239521f3 | 1697 | if (likely(un != NULL)) |
1da177e4 LT |
1698 | goto out; |
1699 | ||
1700 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1701 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1702 | sma = sem_obtain_object_check(ns, semid); |
1703 | if (IS_ERR(sma)) { | |
1704 | rcu_read_unlock(); | |
4de85cd6 | 1705 | return ERR_CAST(sma); |
16df3674 | 1706 | } |
023a5355 | 1707 | |
1da177e4 | 1708 | nsems = sma->sem_nsems; |
dba4cdd3 | 1709 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
6062a8dc RR |
1710 | rcu_read_unlock(); |
1711 | un = ERR_PTR(-EIDRM); | |
1712 | goto out; | |
1713 | } | |
16df3674 | 1714 | rcu_read_unlock(); |
1da177e4 | 1715 | |
4daa28f6 | 1716 | /* step 2: allocate new undo structure */ |
4668edc3 | 1717 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1718 | if (!new) { |
dba4cdd3 | 1719 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1720 | return ERR_PTR(-ENOMEM); |
1721 | } | |
1da177e4 | 1722 | |
380af1b3 | 1723 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1724 | rcu_read_lock(); |
6ff37972 | 1725 | sem_lock_and_putref(sma); |
0f3d2b01 | 1726 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1727 | sem_unlock(sma, -1); |
6d49dab8 | 1728 | rcu_read_unlock(); |
1da177e4 LT |
1729 | kfree(new); |
1730 | un = ERR_PTR(-EIDRM); | |
1731 | goto out; | |
1732 | } | |
380af1b3 MS |
1733 | spin_lock(&ulp->lock); |
1734 | ||
1735 | /* | |
1736 | * step 4: check for races: did someone else allocate the undo struct? | |
1737 | */ | |
1738 | un = lookup_undo(ulp, semid); | |
1739 | if (un) { | |
1740 | kfree(new); | |
1741 | goto success; | |
1742 | } | |
4daa28f6 MS |
1743 | /* step 5: initialize & link new undo structure */ |
1744 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1745 | new->ulp = ulp; |
4daa28f6 MS |
1746 | new->semid = semid; |
1747 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1748 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1749 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1750 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1751 | un = new; |
4daa28f6 | 1752 | |
380af1b3 | 1753 | success: |
c530c6ac | 1754 | spin_unlock(&ulp->lock); |
6062a8dc | 1755 | sem_unlock(sma, -1); |
1da177e4 LT |
1756 | out: |
1757 | return un; | |
1758 | } | |
1759 | ||
d5460c99 HC |
1760 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1761 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1762 | { |
1763 | int error = -EINVAL; | |
1764 | struct sem_array *sma; | |
1765 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1766 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1767 | struct sem_undo *un; |
4ce33ec2 DB |
1768 | int max, locknum; |
1769 | bool undos = false, alter = false, dupsop = false; | |
1da177e4 | 1770 | struct sem_queue queue; |
4ce33ec2 | 1771 | unsigned long dup = 0, jiffies_left = 0; |
e3893534 KK |
1772 | struct ipc_namespace *ns; |
1773 | ||
1774 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1775 | |
1776 | if (nsops < 1 || semid < 0) | |
1777 | return -EINVAL; | |
e3893534 | 1778 | if (nsops > ns->sc_semopm) |
1da177e4 | 1779 | return -E2BIG; |
239521f3 MS |
1780 | if (nsops > SEMOPM_FAST) { |
1781 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1782 | if (sops == NULL) | |
1da177e4 LT |
1783 | return -ENOMEM; |
1784 | } | |
4ce33ec2 | 1785 | |
239521f3 MS |
1786 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1787 | error = -EFAULT; | |
1da177e4 LT |
1788 | goto out_free; |
1789 | } | |
4ce33ec2 | 1790 | |
1da177e4 LT |
1791 | if (timeout) { |
1792 | struct timespec _timeout; | |
1793 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1794 | error = -EFAULT; | |
1795 | goto out_free; | |
1796 | } | |
1797 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1798 | _timeout.tv_nsec >= 1000000000L) { | |
1799 | error = -EINVAL; | |
1800 | goto out_free; | |
1801 | } | |
1802 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1803 | } | |
4ce33ec2 | 1804 | |
1da177e4 LT |
1805 | max = 0; |
1806 | for (sop = sops; sop < sops + nsops; sop++) { | |
4ce33ec2 DB |
1807 | unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG); |
1808 | ||
1da177e4 LT |
1809 | if (sop->sem_num >= max) |
1810 | max = sop->sem_num; | |
1811 | if (sop->sem_flg & SEM_UNDO) | |
4ce33ec2 DB |
1812 | undos = true; |
1813 | if (dup & mask) { | |
1814 | /* | |
1815 | * There was a previous alter access that appears | |
1816 | * to have accessed the same semaphore, thus use | |
1817 | * the dupsop logic. "appears", because the detection | |
1818 | * can only check % BITS_PER_LONG. | |
1819 | */ | |
1820 | dupsop = true; | |
1821 | } | |
1822 | if (sop->sem_op != 0) { | |
1823 | alter = true; | |
1824 | dup |= mask; | |
1825 | } | |
1da177e4 | 1826 | } |
1da177e4 | 1827 | |
1da177e4 | 1828 | if (undos) { |
6062a8dc | 1829 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1830 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1831 | if (IS_ERR(un)) { |
1832 | error = PTR_ERR(un); | |
1833 | goto out_free; | |
1834 | } | |
6062a8dc | 1835 | } else { |
1da177e4 | 1836 | un = NULL; |
6062a8dc RR |
1837 | rcu_read_lock(); |
1838 | } | |
1da177e4 | 1839 | |
16df3674 | 1840 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1841 | if (IS_ERR(sma)) { |
6062a8dc | 1842 | rcu_read_unlock(); |
023a5355 | 1843 | error = PTR_ERR(sma); |
1da177e4 | 1844 | goto out_free; |
023a5355 ND |
1845 | } |
1846 | ||
16df3674 | 1847 | error = -EFBIG; |
248e7357 DB |
1848 | if (max >= sma->sem_nsems) { |
1849 | rcu_read_unlock(); | |
1850 | goto out_free; | |
1851 | } | |
16df3674 DB |
1852 | |
1853 | error = -EACCES; | |
248e7357 DB |
1854 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) { |
1855 | rcu_read_unlock(); | |
1856 | goto out_free; | |
1857 | } | |
16df3674 DB |
1858 | |
1859 | error = security_sem_semop(sma, sops, nsops, alter); | |
248e7357 DB |
1860 | if (error) { |
1861 | rcu_read_unlock(); | |
1862 | goto out_free; | |
1863 | } | |
16df3674 | 1864 | |
6e224f94 MS |
1865 | error = -EIDRM; |
1866 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
1867 | /* |
1868 | * We eventually might perform the following check in a lockless | |
1869 | * fashion, considering ipc_valid_object() locking constraints. | |
1870 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1871 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1872 | * check below. More details on the fine grained locking scheme | |
1873 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1874 | */ | |
1875 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 1876 | goto out_unlock_free; |
1da177e4 | 1877 | /* |
4daa28f6 | 1878 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1879 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1880 | * and now a new array with received the same id. Check and fail. |
25985edc | 1881 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1882 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1883 | */ |
6062a8dc RR |
1884 | if (un && un->semid == -1) |
1885 | goto out_unlock_free; | |
4daa28f6 | 1886 | |
d198cd6d MS |
1887 | queue.sops = sops; |
1888 | queue.nsops = nsops; | |
1889 | queue.undo = un; | |
1890 | queue.pid = task_tgid_vnr(current); | |
1891 | queue.alter = alter; | |
4ce33ec2 | 1892 | queue.dupsop = dupsop; |
d198cd6d MS |
1893 | |
1894 | error = perform_atomic_semop(sma, &queue); | |
9ae949fa DB |
1895 | if (error == 0) { /* non-blocking succesfull path */ |
1896 | DEFINE_WAKE_Q(wake_q); | |
1897 | ||
1898 | /* | |
1899 | * If the operation was successful, then do | |
0e8c6656 MS |
1900 | * the required updates. |
1901 | */ | |
1902 | if (alter) | |
9ae949fa | 1903 | do_smart_update(sma, sops, nsops, 1, &wake_q); |
0e8c6656 MS |
1904 | else |
1905 | set_semotime(sma, sops); | |
9ae949fa DB |
1906 | |
1907 | sem_unlock(sma, locknum); | |
1908 | rcu_read_unlock(); | |
1909 | wake_up_q(&wake_q); | |
1910 | ||
1911 | goto out_free; | |
1da177e4 | 1912 | } |
9ae949fa | 1913 | if (error < 0) /* non-blocking error path */ |
0e8c6656 | 1914 | goto out_unlock_free; |
1da177e4 | 1915 | |
9ae949fa DB |
1916 | /* |
1917 | * We need to sleep on this operation, so we put the current | |
1da177e4 LT |
1918 | * task into the pending queue and go to sleep. |
1919 | */ | |
b97e820f MS |
1920 | if (nsops == 1) { |
1921 | struct sem *curr; | |
1a233956 | 1922 | curr = &sma->sems[sops->sem_num]; |
b97e820f | 1923 | |
f269f40a MS |
1924 | if (alter) { |
1925 | if (sma->complex_count) { | |
1926 | list_add_tail(&queue.list, | |
1927 | &sma->pending_alter); | |
1928 | } else { | |
1929 | ||
1930 | list_add_tail(&queue.list, | |
1931 | &curr->pending_alter); | |
1932 | } | |
1933 | } else { | |
1a82e9e1 | 1934 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 1935 | } |
b97e820f | 1936 | } else { |
f269f40a MS |
1937 | if (!sma->complex_count) |
1938 | merge_queues(sma); | |
1939 | ||
9f1bc2c9 | 1940 | if (alter) |
1a82e9e1 | 1941 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1942 | else |
1a82e9e1 MS |
1943 | list_add_tail(&queue.list, &sma->pending_const); |
1944 | ||
b97e820f MS |
1945 | sma->complex_count++; |
1946 | } | |
1947 | ||
b5fa01a2 DB |
1948 | do { |
1949 | queue.status = -EINTR; | |
1950 | queue.sleeper = current; | |
0b0577f6 | 1951 | |
b5fa01a2 DB |
1952 | __set_current_state(TASK_INTERRUPTIBLE); |
1953 | sem_unlock(sma, locknum); | |
1954 | rcu_read_unlock(); | |
1da177e4 | 1955 | |
b5fa01a2 DB |
1956 | if (timeout) |
1957 | jiffies_left = schedule_timeout(jiffies_left); | |
1958 | else | |
1959 | schedule(); | |
1da177e4 | 1960 | |
9ae949fa | 1961 | /* |
b5fa01a2 DB |
1962 | * fastpath: the semop has completed, either successfully or |
1963 | * not, from the syscall pov, is quite irrelevant to us at this | |
1964 | * point; we're done. | |
1965 | * | |
1966 | * We _do_ care, nonetheless, about being awoken by a signal or | |
1967 | * spuriously. The queue.status is checked again in the | |
1968 | * slowpath (aka after taking sem_lock), such that we can detect | |
1969 | * scenarios where we were awakened externally, during the | |
1970 | * window between wake_q_add() and wake_up_q(). | |
c61284e9 | 1971 | */ |
b5fa01a2 DB |
1972 | error = READ_ONCE(queue.status); |
1973 | if (error != -EINTR) { | |
1974 | /* | |
1975 | * User space could assume that semop() is a memory | |
1976 | * barrier: Without the mb(), the cpu could | |
1977 | * speculatively read in userspace stale data that was | |
1978 | * overwritten by the previous owner of the semaphore. | |
1979 | */ | |
1980 | smp_mb(); | |
1981 | goto out_free; | |
1982 | } | |
d694ad62 | 1983 | |
b5fa01a2 | 1984 | rcu_read_lock(); |
c626bc46 | 1985 | locknum = sem_lock(sma, sops, nsops); |
1da177e4 | 1986 | |
370b262c DB |
1987 | if (!ipc_valid_object(&sma->sem_perm)) |
1988 | goto out_unlock_free; | |
1989 | ||
1990 | error = READ_ONCE(queue.status); | |
1da177e4 | 1991 | |
b5fa01a2 DB |
1992 | /* |
1993 | * If queue.status != -EINTR we are woken up by another process. | |
1994 | * Leave without unlink_queue(), but with sem_unlock(). | |
1995 | */ | |
1996 | if (error != -EINTR) | |
1997 | goto out_unlock_free; | |
0b0577f6 | 1998 | |
b5fa01a2 DB |
1999 | /* |
2000 | * If an interrupt occurred we have to clean up the queue. | |
2001 | */ | |
2002 | if (timeout && jiffies_left == 0) | |
2003 | error = -EAGAIN; | |
2004 | } while (error == -EINTR && !signal_pending(current)); /* spurious */ | |
0b0577f6 | 2005 | |
b97e820f | 2006 | unlink_queue(sma, &queue); |
1da177e4 LT |
2007 | |
2008 | out_unlock_free: | |
6062a8dc | 2009 | sem_unlock(sma, locknum); |
6d49dab8 | 2010 | rcu_read_unlock(); |
1da177e4 | 2011 | out_free: |
239521f3 | 2012 | if (sops != fast_sops) |
1da177e4 LT |
2013 | kfree(sops); |
2014 | return error; | |
2015 | } | |
2016 | ||
d5460c99 HC |
2017 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2018 | unsigned, nsops) | |
1da177e4 LT |
2019 | { |
2020 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2021 | } | |
2022 | ||
2023 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2024 | * parent and child tasks. | |
1da177e4 LT |
2025 | */ |
2026 | ||
2027 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2028 | { | |
2029 | struct sem_undo_list *undo_list; | |
2030 | int error; | |
2031 | ||
2032 | if (clone_flags & CLONE_SYSVSEM) { | |
2033 | error = get_undo_list(&undo_list); | |
2034 | if (error) | |
2035 | return error; | |
1da177e4 LT |
2036 | atomic_inc(&undo_list->refcnt); |
2037 | tsk->sysvsem.undo_list = undo_list; | |
46c0a8ca | 2038 | } else |
1da177e4 LT |
2039 | tsk->sysvsem.undo_list = NULL; |
2040 | ||
2041 | return 0; | |
2042 | } | |
2043 | ||
2044 | /* | |
2045 | * add semadj values to semaphores, free undo structures. | |
2046 | * undo structures are not freed when semaphore arrays are destroyed | |
2047 | * so some of them may be out of date. | |
2048 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2049 | * set of adjustments that needs to be done should be done in an atomic | |
2050 | * manner or not. That is, if we are attempting to decrement the semval | |
2051 | * should we queue up and wait until we can do so legally? | |
2052 | * The original implementation attempted to do this (queue and wait). | |
2053 | * The current implementation does not do so. The POSIX standard | |
2054 | * and SVID should be consulted to determine what behavior is mandated. | |
2055 | */ | |
2056 | void exit_sem(struct task_struct *tsk) | |
2057 | { | |
4daa28f6 | 2058 | struct sem_undo_list *ulp; |
1da177e4 | 2059 | |
4daa28f6 MS |
2060 | ulp = tsk->sysvsem.undo_list; |
2061 | if (!ulp) | |
1da177e4 | 2062 | return; |
9edff4ab | 2063 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2064 | |
4daa28f6 | 2065 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2066 | return; |
2067 | ||
380af1b3 | 2068 | for (;;) { |
1da177e4 | 2069 | struct sem_array *sma; |
380af1b3 | 2070 | struct sem_undo *un; |
6062a8dc | 2071 | int semid, i; |
9ae949fa | 2072 | DEFINE_WAKE_Q(wake_q); |
4daa28f6 | 2073 | |
2a1613a5 NB |
2074 | cond_resched(); |
2075 | ||
380af1b3 | 2076 | rcu_read_lock(); |
05725f7e JP |
2077 | un = list_entry_rcu(ulp->list_proc.next, |
2078 | struct sem_undo, list_proc); | |
602b8593 HK |
2079 | if (&un->list_proc == &ulp->list_proc) { |
2080 | /* | |
2081 | * We must wait for freeary() before freeing this ulp, | |
2082 | * in case we raced with last sem_undo. There is a small | |
2083 | * possibility where we exit while freeary() didn't | |
2084 | * finish unlocking sem_undo_list. | |
2085 | */ | |
2086 | spin_unlock_wait(&ulp->lock); | |
2087 | rcu_read_unlock(); | |
2088 | break; | |
2089 | } | |
2090 | spin_lock(&ulp->lock); | |
2091 | semid = un->semid; | |
2092 | spin_unlock(&ulp->lock); | |
4daa28f6 | 2093 | |
602b8593 | 2094 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2095 | if (semid == -1) { |
2096 | rcu_read_unlock(); | |
602b8593 | 2097 | continue; |
6062a8dc | 2098 | } |
1da177e4 | 2099 | |
602b8593 | 2100 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); |
380af1b3 | 2101 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2102 | if (IS_ERR(sma)) { |
2103 | rcu_read_unlock(); | |
380af1b3 | 2104 | continue; |
6062a8dc | 2105 | } |
1da177e4 | 2106 | |
6062a8dc | 2107 | sem_lock(sma, NULL, -1); |
6e224f94 | 2108 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2109 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2110 | sem_unlock(sma, -1); |
2111 | rcu_read_unlock(); | |
2112 | continue; | |
2113 | } | |
bf17bb71 | 2114 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2115 | if (un == NULL) { |
2116 | /* exit_sem raced with IPC_RMID+semget() that created | |
2117 | * exactly the same semid. Nothing to do. | |
2118 | */ | |
6062a8dc | 2119 | sem_unlock(sma, -1); |
6d49dab8 | 2120 | rcu_read_unlock(); |
380af1b3 MS |
2121 | continue; |
2122 | } | |
2123 | ||
2124 | /* remove un from the linked lists */ | |
cf9d5d78 | 2125 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2126 | list_del(&un->list_id); |
2127 | ||
a9795584 HK |
2128 | /* we are the last process using this ulp, acquiring ulp->lock |
2129 | * isn't required. Besides that, we are also protected against | |
2130 | * IPC_RMID as we hold sma->sem_perm lock now | |
2131 | */ | |
380af1b3 | 2132 | list_del_rcu(&un->list_proc); |
380af1b3 | 2133 | |
4daa28f6 MS |
2134 | /* perform adjustments registered in un */ |
2135 | for (i = 0; i < sma->sem_nsems; i++) { | |
1a233956 | 2136 | struct sem *semaphore = &sma->sems[i]; |
4daa28f6 MS |
2137 | if (un->semadj[i]) { |
2138 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2139 | /* |
2140 | * Range checks of the new semaphore value, | |
2141 | * not defined by sus: | |
2142 | * - Some unices ignore the undo entirely | |
2143 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2144 | * - some cap the value (e.g. FreeBSD caps | |
2145 | * at 0, but doesn't enforce SEMVMX) | |
2146 | * | |
2147 | * Linux caps the semaphore value, both at 0 | |
2148 | * and at SEMVMX. | |
2149 | * | |
239521f3 | 2150 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2151 | */ |
5f921ae9 IM |
2152 | if (semaphore->semval < 0) |
2153 | semaphore->semval = 0; | |
2154 | if (semaphore->semval > SEMVMX) | |
2155 | semaphore->semval = SEMVMX; | |
b488893a | 2156 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2157 | } |
2158 | } | |
1da177e4 | 2159 | /* maybe some queued-up processes were waiting for this */ |
9ae949fa | 2160 | do_smart_update(sma, NULL, 0, 1, &wake_q); |
6062a8dc | 2161 | sem_unlock(sma, -1); |
6d49dab8 | 2162 | rcu_read_unlock(); |
9ae949fa | 2163 | wake_up_q(&wake_q); |
380af1b3 | 2164 | |
693a8b6e | 2165 | kfree_rcu(un, rcu); |
1da177e4 | 2166 | } |
4daa28f6 | 2167 | kfree(ulp); |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2171 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2172 | { |
1efdb69b | 2173 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2174 | struct sem_array *sma = it; |
d12e1e50 MS |
2175 | time_t sem_otime; |
2176 | ||
d8c63376 MS |
2177 | /* |
2178 | * The proc interface isn't aware of sem_lock(), it calls | |
2179 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
5864a2fd MS |
2180 | * In order to stay compatible with sem_lock(), we must |
2181 | * enter / leave complex_mode. | |
d8c63376 | 2182 | */ |
5864a2fd | 2183 | complexmode_enter(sma); |
d8c63376 | 2184 | |
d12e1e50 | 2185 | sem_otime = get_semotime(sma); |
19b4946c | 2186 | |
7f032d6e JP |
2187 | seq_printf(s, |
2188 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", | |
2189 | sma->sem_perm.key, | |
2190 | sma->sem_perm.id, | |
2191 | sma->sem_perm.mode, | |
2192 | sma->sem_nsems, | |
2193 | from_kuid_munged(user_ns, sma->sem_perm.uid), | |
2194 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2195 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2196 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
2197 | sem_otime, | |
2198 | sma->sem_ctime); | |
2199 | ||
5864a2fd MS |
2200 | complexmode_tryleave(sma); |
2201 | ||
7f032d6e | 2202 | return 0; |
1da177e4 LT |
2203 | } |
2204 | #endif |