]>
Commit | Line | Data |
---|---|---|
1 | /* | |
2 | * linux/ipc/sem.c | |
3 | * Copyright (C) 1992 Krishna Balasubramanian | |
4 | * Copyright (C) 1995 Eric Schenk, Bruno Haible | |
5 | * | |
6 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> | |
7 | * | |
8 | * SMP-threaded, sysctl's added | |
9 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> | |
10 | * Enforced range limit on SEM_UNDO | |
11 | * (c) 2001 Red Hat Inc | |
12 | * Lockless wakeup | |
13 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
14 | * (c) 2016 Davidlohr Bueso <dave@stgolabs.net> | |
15 | * Further wakeup optimizations, documentation | |
16 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
17 | * | |
18 | * support for audit of ipc object properties and permission changes | |
19 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
20 | * | |
21 | * namespaces support | |
22 | * OpenVZ, SWsoft Inc. | |
23 | * Pavel Emelianov <xemul@openvz.org> | |
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. | |
51 | * - semncnt and semzcnt are calculated on demand in count_semcnt() | |
52 | * - the task that performs a successful semop() scans the list of all | |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
57 | * dropping all locks. (see wake_up_sem_queue_prepare()) | |
58 | * - 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 | * - 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. | |
70 | */ | |
71 | ||
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> | |
77 | #include <linux/security.h> | |
78 | #include <linux/syscalls.h> | |
79 | #include <linux/audit.h> | |
80 | #include <linux/capability.h> | |
81 | #include <linux/seq_file.h> | |
82 | #include <linux/rwsem.h> | |
83 | #include <linux/nsproxy.h> | |
84 | #include <linux/ipc_namespace.h> | |
85 | #include <linux/sched/wake_q.h> | |
86 | ||
87 | #include <linux/uaccess.h> | |
88 | #include "util.h" | |
89 | ||
90 | ||
91 | /* One queue for each sleeping process in the system. */ | |
92 | struct sem_queue { | |
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 */ | |
99 | struct sembuf *blocking; /* the operation that blocked */ | |
100 | int nsops; /* number of operations */ | |
101 | bool alter; /* does *sops alter the array? */ | |
102 | bool dupsop; /* sops on more than one sem_num */ | |
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 | ||
131 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) | |
132 | ||
133 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) | |
134 | ||
135 | static int newary(struct ipc_namespace *, struct ipc_params *); | |
136 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); | |
137 | #ifdef CONFIG_PROC_FS | |
138 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); | |
139 | #endif | |
140 | ||
141 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
142 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
143 | ||
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 | ||
151 | /* | |
152 | * Locking: | |
153 | * a) global sem_lock() for read/write | |
154 | * sem_undo.id_next, | |
155 | * sem_array.complex_count, | |
156 | * sem_array.pending{_alter,_const}, | |
157 | * sem_array.sem_undo | |
158 | * | |
159 | * b) global or semaphore sem_lock() for read/write: | |
160 | * sem_array.sems[i].pending_{const,alter}: | |
161 | * | |
162 | * c) special: | |
163 | * sem_undo_list.list_proc: | |
164 | * * undo_list->lock for write | |
165 | * * rcu for read | |
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. | |
181 | */ | |
182 | ||
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 | ||
188 | void sem_init_ns(struct ipc_namespace *ns) | |
189 | { | |
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; | |
195 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); | |
196 | } | |
197 | ||
198 | #ifdef CONFIG_IPC_NS | |
199 | void sem_exit_ns(struct ipc_namespace *ns) | |
200 | { | |
201 | free_ipcs(ns, &sem_ids(ns), freeary); | |
202 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); | |
203 | } | |
204 | #endif | |
205 | ||
206 | void __init sem_init(void) | |
207 | { | |
208 | sem_init_ns(&init_ipc_ns); | |
209 | ipc_init_proc_interface("sysvipc/sem", | |
210 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
211 | IPC_SEM_IDS, sysvipc_sem_proc_show); | |
212 | } | |
213 | ||
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; | |
235 | curr = &sma->sems[q->sops[0].sem_num]; | |
236 | ||
237 | list_add_tail(&q->list, &curr->pending_alter); | |
238 | } | |
239 | INIT_LIST_HEAD(&sma->pending_alter); | |
240 | } | |
241 | ||
242 | /** | |
243 | * merge_queues - merge single semop queues into global queue | |
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++) { | |
255 | struct sem *sem = &sma->sems[i]; | |
256 | ||
257 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
258 | } | |
259 | } | |
260 | ||
261 | static void sem_rcu_free(struct rcu_head *head) | |
262 | { | |
263 | struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu); | |
264 | struct sem_array *sma = container_of(p, struct sem_array, sem_perm); | |
265 | ||
266 | security_sem_free(sma); | |
267 | kvfree(sma); | |
268 | } | |
269 | ||
270 | /* | |
271 | * Enter the mode suitable for non-simple operations: | |
272 | * Caller must own sem_perm.lock. | |
273 | */ | |
274 | static void complexmode_enter(struct sem_array *sma) | |
275 | { | |
276 | int i; | |
277 | struct sem *sem; | |
278 | ||
279 | if (sma->use_global_lock > 0) { | |
280 | /* | |
281 | * We are already in global lock mode. | |
282 | * Nothing to do, just reset the | |
283 | * counter until we return to simple mode. | |
284 | */ | |
285 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; | |
286 | return; | |
287 | } | |
288 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; | |
289 | ||
290 | for (i = 0; i < sma->sem_nsems; i++) { | |
291 | sem = &sma->sems[i]; | |
292 | spin_lock(&sem->lock); | |
293 | spin_unlock(&sem->lock); | |
294 | } | |
295 | } | |
296 | ||
297 | /* | |
298 | * Try to leave the mode that disallows simple operations: | |
299 | * Caller must own sem_perm.lock. | |
300 | */ | |
301 | static void complexmode_tryleave(struct sem_array *sma) | |
302 | { | |
303 | if (sma->complex_count) { | |
304 | /* Complex ops are sleeping. | |
305 | * We must stay in complex mode | |
306 | */ | |
307 | return; | |
308 | } | |
309 | if (sma->use_global_lock == 1) { | |
310 | /* | |
311 | * Immediately after setting use_global_lock to 0, | |
312 | * a simple op can start. Thus: all memory writes | |
313 | * performed by the current operation must be visible | |
314 | * before we set use_global_lock to 0. | |
315 | */ | |
316 | smp_store_release(&sma->use_global_lock, 0); | |
317 | } else { | |
318 | sma->use_global_lock--; | |
319 | } | |
320 | } | |
321 | ||
322 | #define SEM_GLOBAL_LOCK (-1) | |
323 | /* | |
324 | * If the request contains only one semaphore operation, and there are | |
325 | * no complex transactions pending, lock only the semaphore involved. | |
326 | * Otherwise, lock the entire semaphore array, since we either have | |
327 | * multiple semaphores in our own semops, or we need to look at | |
328 | * semaphores from other pending complex operations. | |
329 | */ | |
330 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
331 | int nsops) | |
332 | { | |
333 | struct sem *sem; | |
334 | ||
335 | if (nsops != 1) { | |
336 | /* Complex operation - acquire a full lock */ | |
337 | ipc_lock_object(&sma->sem_perm); | |
338 | ||
339 | /* Prevent parallel simple ops */ | |
340 | complexmode_enter(sma); | |
341 | return SEM_GLOBAL_LOCK; | |
342 | } | |
343 | ||
344 | /* | |
345 | * Only one semaphore affected - try to optimize locking. | |
346 | * Optimized locking is possible if no complex operation | |
347 | * is either enqueued or processed right now. | |
348 | * | |
349 | * Both facts are tracked by use_global_mode. | |
350 | */ | |
351 | sem = &sma->sems[sops->sem_num]; | |
352 | ||
353 | /* | |
354 | * Initial check for use_global_lock. Just an optimization, | |
355 | * no locking, no memory barrier. | |
356 | */ | |
357 | if (!sma->use_global_lock) { | |
358 | /* | |
359 | * It appears that no complex operation is around. | |
360 | * Acquire the per-semaphore lock. | |
361 | */ | |
362 | spin_lock(&sem->lock); | |
363 | ||
364 | /* pairs with smp_store_release() */ | |
365 | if (!smp_load_acquire(&sma->use_global_lock)) { | |
366 | /* fast path successful! */ | |
367 | return sops->sem_num; | |
368 | } | |
369 | spin_unlock(&sem->lock); | |
370 | } | |
371 | ||
372 | /* slow path: acquire the full lock */ | |
373 | ipc_lock_object(&sma->sem_perm); | |
374 | ||
375 | if (sma->use_global_lock == 0) { | |
376 | /* | |
377 | * The use_global_lock mode ended while we waited for | |
378 | * sma->sem_perm.lock. Thus we must switch to locking | |
379 | * with sem->lock. | |
380 | * Unlike in the fast path, there is no need to recheck | |
381 | * sma->use_global_lock after we have acquired sem->lock: | |
382 | * We own sma->sem_perm.lock, thus use_global_lock cannot | |
383 | * change. | |
384 | */ | |
385 | spin_lock(&sem->lock); | |
386 | ||
387 | ipc_unlock_object(&sma->sem_perm); | |
388 | return sops->sem_num; | |
389 | } else { | |
390 | /* | |
391 | * Not a false alarm, thus continue to use the global lock | |
392 | * mode. No need for complexmode_enter(), this was done by | |
393 | * the caller that has set use_global_mode to non-zero. | |
394 | */ | |
395 | return SEM_GLOBAL_LOCK; | |
396 | } | |
397 | } | |
398 | ||
399 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
400 | { | |
401 | if (locknum == SEM_GLOBAL_LOCK) { | |
402 | unmerge_queues(sma); | |
403 | complexmode_tryleave(sma); | |
404 | ipc_unlock_object(&sma->sem_perm); | |
405 | } else { | |
406 | struct sem *sem = &sma->sems[locknum]; | |
407 | spin_unlock(&sem->lock); | |
408 | } | |
409 | } | |
410 | ||
411 | /* | |
412 | * sem_lock_(check_) routines are called in the paths where the rwsem | |
413 | * is not held. | |
414 | * | |
415 | * The caller holds the RCU read lock. | |
416 | */ | |
417 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) | |
418 | { | |
419 | struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); | |
420 | ||
421 | if (IS_ERR(ipcp)) | |
422 | return ERR_CAST(ipcp); | |
423 | ||
424 | return container_of(ipcp, struct sem_array, sem_perm); | |
425 | } | |
426 | ||
427 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, | |
428 | int id) | |
429 | { | |
430 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
431 | ||
432 | if (IS_ERR(ipcp)) | |
433 | return ERR_CAST(ipcp); | |
434 | ||
435 | return container_of(ipcp, struct sem_array, sem_perm); | |
436 | } | |
437 | ||
438 | static inline void sem_lock_and_putref(struct sem_array *sma) | |
439 | { | |
440 | sem_lock(sma, NULL, -1); | |
441 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
442 | } | |
443 | ||
444 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) | |
445 | { | |
446 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
447 | } | |
448 | ||
449 | static struct sem_array *sem_alloc(size_t nsems) | |
450 | { | |
451 | struct sem_array *sma; | |
452 | size_t size; | |
453 | ||
454 | if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0])) | |
455 | return NULL; | |
456 | ||
457 | size = sizeof(*sma) + nsems * sizeof(sma->sems[0]); | |
458 | sma = kvmalloc(size, GFP_KERNEL); | |
459 | if (unlikely(!sma)) | |
460 | return NULL; | |
461 | ||
462 | memset(sma, 0, size); | |
463 | ||
464 | return sma; | |
465 | } | |
466 | ||
467 | /** | |
468 | * newary - Create a new semaphore set | |
469 | * @ns: namespace | |
470 | * @params: ptr to the structure that contains key, semflg and nsems | |
471 | * | |
472 | * Called with sem_ids.rwsem held (as a writer) | |
473 | */ | |
474 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) | |
475 | { | |
476 | int retval; | |
477 | struct sem_array *sma; | |
478 | key_t key = params->key; | |
479 | int nsems = params->u.nsems; | |
480 | int semflg = params->flg; | |
481 | int i; | |
482 | ||
483 | if (!nsems) | |
484 | return -EINVAL; | |
485 | if (ns->used_sems + nsems > ns->sc_semmns) | |
486 | return -ENOSPC; | |
487 | ||
488 | sma = sem_alloc(nsems); | |
489 | if (!sma) | |
490 | return -ENOMEM; | |
491 | ||
492 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
493 | sma->sem_perm.key = key; | |
494 | ||
495 | sma->sem_perm.security = NULL; | |
496 | retval = security_sem_alloc(sma); | |
497 | if (retval) { | |
498 | kvfree(sma); | |
499 | return retval; | |
500 | } | |
501 | ||
502 | for (i = 0; i < nsems; i++) { | |
503 | INIT_LIST_HEAD(&sma->sems[i].pending_alter); | |
504 | INIT_LIST_HEAD(&sma->sems[i].pending_const); | |
505 | spin_lock_init(&sma->sems[i].lock); | |
506 | } | |
507 | ||
508 | sma->complex_count = 0; | |
509 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; | |
510 | INIT_LIST_HEAD(&sma->pending_alter); | |
511 | INIT_LIST_HEAD(&sma->pending_const); | |
512 | INIT_LIST_HEAD(&sma->list_id); | |
513 | sma->sem_nsems = nsems; | |
514 | sma->sem_ctime = get_seconds(); | |
515 | ||
516 | retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); | |
517 | if (retval < 0) { | |
518 | call_rcu(&sma->sem_perm.rcu, sem_rcu_free); | |
519 | return retval; | |
520 | } | |
521 | ns->used_sems += nsems; | |
522 | ||
523 | sem_unlock(sma, -1); | |
524 | rcu_read_unlock(); | |
525 | ||
526 | return sma->sem_perm.id; | |
527 | } | |
528 | ||
529 | ||
530 | /* | |
531 | * Called with sem_ids.rwsem and ipcp locked. | |
532 | */ | |
533 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) | |
534 | { | |
535 | struct sem_array *sma; | |
536 | ||
537 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
538 | return security_sem_associate(sma, semflg); | |
539 | } | |
540 | ||
541 | /* | |
542 | * Called with sem_ids.rwsem and ipcp locked. | |
543 | */ | |
544 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, | |
545 | struct ipc_params *params) | |
546 | { | |
547 | struct sem_array *sma; | |
548 | ||
549 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
550 | if (params->u.nsems > sma->sem_nsems) | |
551 | return -EINVAL; | |
552 | ||
553 | return 0; | |
554 | } | |
555 | ||
556 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) | |
557 | { | |
558 | struct ipc_namespace *ns; | |
559 | static const struct ipc_ops sem_ops = { | |
560 | .getnew = newary, | |
561 | .associate = sem_security, | |
562 | .more_checks = sem_more_checks, | |
563 | }; | |
564 | struct ipc_params sem_params; | |
565 | ||
566 | ns = current->nsproxy->ipc_ns; | |
567 | ||
568 | if (nsems < 0 || nsems > ns->sc_semmsl) | |
569 | return -EINVAL; | |
570 | ||
571 | sem_params.key = key; | |
572 | sem_params.flg = semflg; | |
573 | sem_params.u.nsems = nsems; | |
574 | ||
575 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); | |
576 | } | |
577 | ||
578 | /** | |
579 | * perform_atomic_semop[_slow] - Attempt to perform semaphore | |
580 | * operations on a given array. | |
581 | * @sma: semaphore array | |
582 | * @q: struct sem_queue that describes the operation | |
583 | * | |
584 | * Caller blocking are as follows, based the value | |
585 | * indicated by the semaphore operation (sem_op): | |
586 | * | |
587 | * (1) >0 never blocks. | |
588 | * (2) 0 (wait-for-zero operation): semval is non-zero. | |
589 | * (3) <0 attempting to decrement semval to a value smaller than zero. | |
590 | * | |
591 | * Returns 0 if the operation was possible. | |
592 | * Returns 1 if the operation is impossible, the caller must sleep. | |
593 | * Returns <0 for error codes. | |
594 | */ | |
595 | static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q) | |
596 | { | |
597 | int result, sem_op, nsops, pid; | |
598 | struct sembuf *sop; | |
599 | struct sem *curr; | |
600 | struct sembuf *sops; | |
601 | struct sem_undo *un; | |
602 | ||
603 | sops = q->sops; | |
604 | nsops = q->nsops; | |
605 | un = q->undo; | |
606 | ||
607 | for (sop = sops; sop < sops + nsops; sop++) { | |
608 | curr = &sma->sems[sop->sem_num]; | |
609 | sem_op = sop->sem_op; | |
610 | result = curr->semval; | |
611 | ||
612 | if (!sem_op && result) | |
613 | goto would_block; | |
614 | ||
615 | result += sem_op; | |
616 | if (result < 0) | |
617 | goto would_block; | |
618 | if (result > SEMVMX) | |
619 | goto out_of_range; | |
620 | ||
621 | if (sop->sem_flg & SEM_UNDO) { | |
622 | int undo = un->semadj[sop->sem_num] - sem_op; | |
623 | /* Exceeding the undo range is an error. */ | |
624 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
625 | goto out_of_range; | |
626 | un->semadj[sop->sem_num] = undo; | |
627 | } | |
628 | ||
629 | curr->semval = result; | |
630 | } | |
631 | ||
632 | sop--; | |
633 | pid = q->pid; | |
634 | while (sop >= sops) { | |
635 | sma->sems[sop->sem_num].sempid = pid; | |
636 | sop--; | |
637 | } | |
638 | ||
639 | return 0; | |
640 | ||
641 | out_of_range: | |
642 | result = -ERANGE; | |
643 | goto undo; | |
644 | ||
645 | would_block: | |
646 | q->blocking = sop; | |
647 | ||
648 | if (sop->sem_flg & IPC_NOWAIT) | |
649 | result = -EAGAIN; | |
650 | else | |
651 | result = 1; | |
652 | ||
653 | undo: | |
654 | sop--; | |
655 | while (sop >= sops) { | |
656 | sem_op = sop->sem_op; | |
657 | sma->sems[sop->sem_num].semval -= sem_op; | |
658 | if (sop->sem_flg & SEM_UNDO) | |
659 | un->semadj[sop->sem_num] += sem_op; | |
660 | sop--; | |
661 | } | |
662 | ||
663 | return result; | |
664 | } | |
665 | ||
666 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) | |
667 | { | |
668 | int result, sem_op, nsops; | |
669 | struct sembuf *sop; | |
670 | struct sem *curr; | |
671 | struct sembuf *sops; | |
672 | struct sem_undo *un; | |
673 | ||
674 | sops = q->sops; | |
675 | nsops = q->nsops; | |
676 | un = q->undo; | |
677 | ||
678 | if (unlikely(q->dupsop)) | |
679 | return perform_atomic_semop_slow(sma, q); | |
680 | ||
681 | /* | |
682 | * We scan the semaphore set twice, first to ensure that the entire | |
683 | * operation can succeed, therefore avoiding any pointless writes | |
684 | * to shared memory and having to undo such changes in order to block | |
685 | * until the operations can go through. | |
686 | */ | |
687 | for (sop = sops; sop < sops + nsops; sop++) { | |
688 | curr = &sma->sems[sop->sem_num]; | |
689 | sem_op = sop->sem_op; | |
690 | result = curr->semval; | |
691 | ||
692 | if (!sem_op && result) | |
693 | goto would_block; /* wait-for-zero */ | |
694 | ||
695 | result += sem_op; | |
696 | if (result < 0) | |
697 | goto would_block; | |
698 | ||
699 | if (result > SEMVMX) | |
700 | return -ERANGE; | |
701 | ||
702 | if (sop->sem_flg & SEM_UNDO) { | |
703 | int undo = un->semadj[sop->sem_num] - sem_op; | |
704 | ||
705 | /* Exceeding the undo range is an error. */ | |
706 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
707 | return -ERANGE; | |
708 | } | |
709 | } | |
710 | ||
711 | for (sop = sops; sop < sops + nsops; sop++) { | |
712 | curr = &sma->sems[sop->sem_num]; | |
713 | sem_op = sop->sem_op; | |
714 | result = curr->semval; | |
715 | ||
716 | if (sop->sem_flg & SEM_UNDO) { | |
717 | int undo = un->semadj[sop->sem_num] - sem_op; | |
718 | ||
719 | un->semadj[sop->sem_num] = undo; | |
720 | } | |
721 | curr->semval += sem_op; | |
722 | curr->sempid = q->pid; | |
723 | } | |
724 | ||
725 | return 0; | |
726 | ||
727 | would_block: | |
728 | q->blocking = sop; | |
729 | return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1; | |
730 | } | |
731 | ||
732 | static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error, | |
733 | struct wake_q_head *wake_q) | |
734 | { | |
735 | wake_q_add(wake_q, q->sleeper); | |
736 | /* | |
737 | * Rely on the above implicit barrier, such that we can | |
738 | * ensure that we hold reference to the task before setting | |
739 | * q->status. Otherwise we could race with do_exit if the | |
740 | * task is awoken by an external event before calling | |
741 | * wake_up_process(). | |
742 | */ | |
743 | WRITE_ONCE(q->status, error); | |
744 | } | |
745 | ||
746 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) | |
747 | { | |
748 | list_del(&q->list); | |
749 | if (q->nsops > 1) | |
750 | sma->complex_count--; | |
751 | } | |
752 | ||
753 | /** check_restart(sma, q) | |
754 | * @sma: semaphore array | |
755 | * @q: the operation that just completed | |
756 | * | |
757 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
758 | * waiting operations. Therefore this function checks if the restart is | |
759 | * really necessary. It is called after a previously waiting operation | |
760 | * modified the array. | |
761 | * Note that wait-for-zero operations are handled without restart. | |
762 | */ | |
763 | static inline int check_restart(struct sem_array *sma, struct sem_queue *q) | |
764 | { | |
765 | /* pending complex alter operations are too difficult to analyse */ | |
766 | if (!list_empty(&sma->pending_alter)) | |
767 | return 1; | |
768 | ||
769 | /* we were a sleeping complex operation. Too difficult */ | |
770 | if (q->nsops > 1) | |
771 | return 1; | |
772 | ||
773 | /* It is impossible that someone waits for the new value: | |
774 | * - complex operations always restart. | |
775 | * - wait-for-zero are handled seperately. | |
776 | * - q is a previously sleeping simple operation that | |
777 | * altered the array. It must be a decrement, because | |
778 | * simple increments never sleep. | |
779 | * - If there are older (higher priority) decrements | |
780 | * in the queue, then they have observed the original | |
781 | * semval value and couldn't proceed. The operation | |
782 | * decremented to value - thus they won't proceed either. | |
783 | */ | |
784 | return 0; | |
785 | } | |
786 | ||
787 | /** | |
788 | * wake_const_ops - wake up non-alter tasks | |
789 | * @sma: semaphore array. | |
790 | * @semnum: semaphore that was modified. | |
791 | * @wake_q: lockless wake-queue head. | |
792 | * | |
793 | * wake_const_ops must be called after a semaphore in a semaphore array | |
794 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
795 | * be called with semnum = -1, as well as with the number of each modified | |
796 | * semaphore. | |
797 | * The tasks that must be woken up are added to @wake_q. The return code | |
798 | * is stored in q->pid. | |
799 | * The function returns 1 if at least one operation was completed successfully. | |
800 | */ | |
801 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
802 | struct wake_q_head *wake_q) | |
803 | { | |
804 | struct sem_queue *q, *tmp; | |
805 | struct list_head *pending_list; | |
806 | int semop_completed = 0; | |
807 | ||
808 | if (semnum == -1) | |
809 | pending_list = &sma->pending_const; | |
810 | else | |
811 | pending_list = &sma->sems[semnum].pending_const; | |
812 | ||
813 | list_for_each_entry_safe(q, tmp, pending_list, list) { | |
814 | int error = perform_atomic_semop(sma, q); | |
815 | ||
816 | if (error > 0) | |
817 | continue; | |
818 | /* operation completed, remove from queue & wakeup */ | |
819 | unlink_queue(sma, q); | |
820 | ||
821 | wake_up_sem_queue_prepare(q, error, wake_q); | |
822 | if (error == 0) | |
823 | semop_completed = 1; | |
824 | } | |
825 | ||
826 | return semop_completed; | |
827 | } | |
828 | ||
829 | /** | |
830 | * do_smart_wakeup_zero - wakeup all wait for zero tasks | |
831 | * @sma: semaphore array | |
832 | * @sops: operations that were performed | |
833 | * @nsops: number of operations | |
834 | * @wake_q: lockless wake-queue head | |
835 | * | |
836 | * Checks all required queue for wait-for-zero operations, based | |
837 | * on the actual changes that were performed on the semaphore array. | |
838 | * The function returns 1 if at least one operation was completed successfully. | |
839 | */ | |
840 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
841 | int nsops, struct wake_q_head *wake_q) | |
842 | { | |
843 | int i; | |
844 | int semop_completed = 0; | |
845 | int got_zero = 0; | |
846 | ||
847 | /* first: the per-semaphore queues, if known */ | |
848 | if (sops) { | |
849 | for (i = 0; i < nsops; i++) { | |
850 | int num = sops[i].sem_num; | |
851 | ||
852 | if (sma->sems[num].semval == 0) { | |
853 | got_zero = 1; | |
854 | semop_completed |= wake_const_ops(sma, num, wake_q); | |
855 | } | |
856 | } | |
857 | } else { | |
858 | /* | |
859 | * No sops means modified semaphores not known. | |
860 | * Assume all were changed. | |
861 | */ | |
862 | for (i = 0; i < sma->sem_nsems; i++) { | |
863 | if (sma->sems[i].semval == 0) { | |
864 | got_zero = 1; | |
865 | semop_completed |= wake_const_ops(sma, i, wake_q); | |
866 | } | |
867 | } | |
868 | } | |
869 | /* | |
870 | * If one of the modified semaphores got 0, | |
871 | * then check the global queue, too. | |
872 | */ | |
873 | if (got_zero) | |
874 | semop_completed |= wake_const_ops(sma, -1, wake_q); | |
875 | ||
876 | return semop_completed; | |
877 | } | |
878 | ||
879 | ||
880 | /** | |
881 | * update_queue - look for tasks that can be completed. | |
882 | * @sma: semaphore array. | |
883 | * @semnum: semaphore that was modified. | |
884 | * @wake_q: lockless wake-queue head. | |
885 | * | |
886 | * update_queue must be called after a semaphore in a semaphore array | |
887 | * was modified. If multiple semaphores were modified, update_queue must | |
888 | * be called with semnum = -1, as well as with the number of each modified | |
889 | * semaphore. | |
890 | * The tasks that must be woken up are added to @wake_q. The return code | |
891 | * is stored in q->pid. | |
892 | * The function internally checks if const operations can now succeed. | |
893 | * | |
894 | * The function return 1 if at least one semop was completed successfully. | |
895 | */ | |
896 | static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q) | |
897 | { | |
898 | struct sem_queue *q, *tmp; | |
899 | struct list_head *pending_list; | |
900 | int semop_completed = 0; | |
901 | ||
902 | if (semnum == -1) | |
903 | pending_list = &sma->pending_alter; | |
904 | else | |
905 | pending_list = &sma->sems[semnum].pending_alter; | |
906 | ||
907 | again: | |
908 | list_for_each_entry_safe(q, tmp, pending_list, list) { | |
909 | int error, restart; | |
910 | ||
911 | /* If we are scanning the single sop, per-semaphore list of | |
912 | * one semaphore and that semaphore is 0, then it is not | |
913 | * necessary to scan further: simple increments | |
914 | * that affect only one entry succeed immediately and cannot | |
915 | * be in the per semaphore pending queue, and decrements | |
916 | * cannot be successful if the value is already 0. | |
917 | */ | |
918 | if (semnum != -1 && sma->sems[semnum].semval == 0) | |
919 | break; | |
920 | ||
921 | error = perform_atomic_semop(sma, q); | |
922 | ||
923 | /* Does q->sleeper still need to sleep? */ | |
924 | if (error > 0) | |
925 | continue; | |
926 | ||
927 | unlink_queue(sma, q); | |
928 | ||
929 | if (error) { | |
930 | restart = 0; | |
931 | } else { | |
932 | semop_completed = 1; | |
933 | do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q); | |
934 | restart = check_restart(sma, q); | |
935 | } | |
936 | ||
937 | wake_up_sem_queue_prepare(q, error, wake_q); | |
938 | if (restart) | |
939 | goto again; | |
940 | } | |
941 | return semop_completed; | |
942 | } | |
943 | ||
944 | /** | |
945 | * set_semotime - set sem_otime | |
946 | * @sma: semaphore array | |
947 | * @sops: operations that modified the array, may be NULL | |
948 | * | |
949 | * sem_otime is replicated to avoid cache line trashing. | |
950 | * This function sets one instance to the current time. | |
951 | */ | |
952 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
953 | { | |
954 | if (sops == NULL) { | |
955 | sma->sems[0].sem_otime = get_seconds(); | |
956 | } else { | |
957 | sma->sems[sops[0].sem_num].sem_otime = | |
958 | get_seconds(); | |
959 | } | |
960 | } | |
961 | ||
962 | /** | |
963 | * do_smart_update - optimized update_queue | |
964 | * @sma: semaphore array | |
965 | * @sops: operations that were performed | |
966 | * @nsops: number of operations | |
967 | * @otime: force setting otime | |
968 | * @wake_q: lockless wake-queue head | |
969 | * | |
970 | * do_smart_update() does the required calls to update_queue and wakeup_zero, | |
971 | * based on the actual changes that were performed on the semaphore array. | |
972 | * Note that the function does not do the actual wake-up: the caller is | |
973 | * responsible for calling wake_up_q(). | |
974 | * It is safe to perform this call after dropping all locks. | |
975 | */ | |
976 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, | |
977 | int otime, struct wake_q_head *wake_q) | |
978 | { | |
979 | int i; | |
980 | ||
981 | otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q); | |
982 | ||
983 | if (!list_empty(&sma->pending_alter)) { | |
984 | /* semaphore array uses the global queue - just process it. */ | |
985 | otime |= update_queue(sma, -1, wake_q); | |
986 | } else { | |
987 | if (!sops) { | |
988 | /* | |
989 | * No sops, thus the modified semaphores are not | |
990 | * known. Check all. | |
991 | */ | |
992 | for (i = 0; i < sma->sem_nsems; i++) | |
993 | otime |= update_queue(sma, i, wake_q); | |
994 | } else { | |
995 | /* | |
996 | * Check the semaphores that were increased: | |
997 | * - No complex ops, thus all sleeping ops are | |
998 | * decrease. | |
999 | * - if we decreased the value, then any sleeping | |
1000 | * semaphore ops wont be able to run: If the | |
1001 | * previous value was too small, then the new | |
1002 | * value will be too small, too. | |
1003 | */ | |
1004 | for (i = 0; i < nsops; i++) { | |
1005 | if (sops[i].sem_op > 0) { | |
1006 | otime |= update_queue(sma, | |
1007 | sops[i].sem_num, wake_q); | |
1008 | } | |
1009 | } | |
1010 | } | |
1011 | } | |
1012 | if (otime) | |
1013 | set_semotime(sma, sops); | |
1014 | } | |
1015 | ||
1016 | /* | |
1017 | * check_qop: Test if a queued operation sleeps on the semaphore semnum | |
1018 | */ | |
1019 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1020 | bool count_zero) | |
1021 | { | |
1022 | struct sembuf *sop = q->blocking; | |
1023 | ||
1024 | /* | |
1025 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1026 | * semaphores. This violates SUS, therefore it was changed to the | |
1027 | * standard compliant behavior. | |
1028 | * Give the administrators a chance to notice that an application | |
1029 | * might misbehave because it relies on the Linux behavior. | |
1030 | */ | |
1031 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1032 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1033 | current->comm, task_pid_nr(current)); | |
1034 | ||
1035 | if (sop->sem_num != semnum) | |
1036 | return 0; | |
1037 | ||
1038 | if (count_zero && sop->sem_op == 0) | |
1039 | return 1; | |
1040 | if (!count_zero && sop->sem_op < 0) | |
1041 | return 1; | |
1042 | ||
1043 | return 0; | |
1044 | } | |
1045 | ||
1046 | /* The following counts are associated to each semaphore: | |
1047 | * semncnt number of tasks waiting on semval being nonzero | |
1048 | * semzcnt number of tasks waiting on semval being zero | |
1049 | * | |
1050 | * Per definition, a task waits only on the semaphore of the first semop | |
1051 | * that cannot proceed, even if additional operation would block, too. | |
1052 | */ | |
1053 | static int count_semcnt(struct sem_array *sma, ushort semnum, | |
1054 | bool count_zero) | |
1055 | { | |
1056 | struct list_head *l; | |
1057 | struct sem_queue *q; | |
1058 | int semcnt; | |
1059 | ||
1060 | semcnt = 0; | |
1061 | /* First: check the simple operations. They are easy to evaluate */ | |
1062 | if (count_zero) | |
1063 | l = &sma->sems[semnum].pending_const; | |
1064 | else | |
1065 | l = &sma->sems[semnum].pending_alter; | |
1066 | ||
1067 | list_for_each_entry(q, l, list) { | |
1068 | /* all task on a per-semaphore list sleep on exactly | |
1069 | * that semaphore | |
1070 | */ | |
1071 | semcnt++; | |
1072 | } | |
1073 | ||
1074 | /* Then: check the complex operations. */ | |
1075 | list_for_each_entry(q, &sma->pending_alter, list) { | |
1076 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1077 | } | |
1078 | if (count_zero) { | |
1079 | list_for_each_entry(q, &sma->pending_const, list) { | |
1080 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1081 | } | |
1082 | } | |
1083 | return semcnt; | |
1084 | } | |
1085 | ||
1086 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked | |
1087 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
1088 | * remains locked on exit. | |
1089 | */ | |
1090 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) | |
1091 | { | |
1092 | struct sem_undo *un, *tu; | |
1093 | struct sem_queue *q, *tq; | |
1094 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); | |
1095 | int i; | |
1096 | DEFINE_WAKE_Q(wake_q); | |
1097 | ||
1098 | /* Free the existing undo structures for this semaphore set. */ | |
1099 | ipc_assert_locked_object(&sma->sem_perm); | |
1100 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { | |
1101 | list_del(&un->list_id); | |
1102 | spin_lock(&un->ulp->lock); | |
1103 | un->semid = -1; | |
1104 | list_del_rcu(&un->list_proc); | |
1105 | spin_unlock(&un->ulp->lock); | |
1106 | kfree_rcu(un, rcu); | |
1107 | } | |
1108 | ||
1109 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
1110 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { | |
1111 | unlink_queue(sma, q); | |
1112 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); | |
1113 | } | |
1114 | ||
1115 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
1116 | unlink_queue(sma, q); | |
1117 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); | |
1118 | } | |
1119 | for (i = 0; i < sma->sem_nsems; i++) { | |
1120 | struct sem *sem = &sma->sems[i]; | |
1121 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { | |
1122 | unlink_queue(sma, q); | |
1123 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); | |
1124 | } | |
1125 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
1126 | unlink_queue(sma, q); | |
1127 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | /* Remove the semaphore set from the IDR */ | |
1132 | sem_rmid(ns, sma); | |
1133 | sem_unlock(sma, -1); | |
1134 | rcu_read_unlock(); | |
1135 | ||
1136 | wake_up_q(&wake_q); | |
1137 | ns->used_sems -= sma->sem_nsems; | |
1138 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1139 | } | |
1140 | ||
1141 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1142 | { | |
1143 | switch (version) { | |
1144 | case IPC_64: | |
1145 | return copy_to_user(buf, in, sizeof(*in)); | |
1146 | case IPC_OLD: | |
1147 | { | |
1148 | struct semid_ds out; | |
1149 | ||
1150 | memset(&out, 0, sizeof(out)); | |
1151 | ||
1152 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); | |
1153 | ||
1154 | out.sem_otime = in->sem_otime; | |
1155 | out.sem_ctime = in->sem_ctime; | |
1156 | out.sem_nsems = in->sem_nsems; | |
1157 | ||
1158 | return copy_to_user(buf, &out, sizeof(out)); | |
1159 | } | |
1160 | default: | |
1161 | return -EINVAL; | |
1162 | } | |
1163 | } | |
1164 | ||
1165 | static time_t get_semotime(struct sem_array *sma) | |
1166 | { | |
1167 | int i; | |
1168 | time_t res; | |
1169 | ||
1170 | res = sma->sems[0].sem_otime; | |
1171 | for (i = 1; i < sma->sem_nsems; i++) { | |
1172 | time_t to = sma->sems[i].sem_otime; | |
1173 | ||
1174 | if (to > res) | |
1175 | res = to; | |
1176 | } | |
1177 | return res; | |
1178 | } | |
1179 | ||
1180 | static int semctl_nolock(struct ipc_namespace *ns, int semid, | |
1181 | int cmd, int version, void __user *p) | |
1182 | { | |
1183 | int err; | |
1184 | struct sem_array *sma; | |
1185 | ||
1186 | switch (cmd) { | |
1187 | case IPC_INFO: | |
1188 | case SEM_INFO: | |
1189 | { | |
1190 | struct seminfo seminfo; | |
1191 | int max_id; | |
1192 | ||
1193 | err = security_sem_semctl(NULL, cmd); | |
1194 | if (err) | |
1195 | return err; | |
1196 | ||
1197 | memset(&seminfo, 0, sizeof(seminfo)); | |
1198 | seminfo.semmni = ns->sc_semmni; | |
1199 | seminfo.semmns = ns->sc_semmns; | |
1200 | seminfo.semmsl = ns->sc_semmsl; | |
1201 | seminfo.semopm = ns->sc_semopm; | |
1202 | seminfo.semvmx = SEMVMX; | |
1203 | seminfo.semmnu = SEMMNU; | |
1204 | seminfo.semmap = SEMMAP; | |
1205 | seminfo.semume = SEMUME; | |
1206 | down_read(&sem_ids(ns).rwsem); | |
1207 | if (cmd == SEM_INFO) { | |
1208 | seminfo.semusz = sem_ids(ns).in_use; | |
1209 | seminfo.semaem = ns->used_sems; | |
1210 | } else { | |
1211 | seminfo.semusz = SEMUSZ; | |
1212 | seminfo.semaem = SEMAEM; | |
1213 | } | |
1214 | max_id = ipc_get_maxid(&sem_ids(ns)); | |
1215 | up_read(&sem_ids(ns).rwsem); | |
1216 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) | |
1217 | return -EFAULT; | |
1218 | return (max_id < 0) ? 0 : max_id; | |
1219 | } | |
1220 | case IPC_STAT: | |
1221 | case SEM_STAT: | |
1222 | { | |
1223 | struct semid64_ds tbuf; | |
1224 | int id = 0; | |
1225 | ||
1226 | memset(&tbuf, 0, sizeof(tbuf)); | |
1227 | ||
1228 | rcu_read_lock(); | |
1229 | if (cmd == SEM_STAT) { | |
1230 | sma = sem_obtain_object(ns, semid); | |
1231 | if (IS_ERR(sma)) { | |
1232 | err = PTR_ERR(sma); | |
1233 | goto out_unlock; | |
1234 | } | |
1235 | id = sma->sem_perm.id; | |
1236 | } else { | |
1237 | sma = sem_obtain_object_check(ns, semid); | |
1238 | if (IS_ERR(sma)) { | |
1239 | err = PTR_ERR(sma); | |
1240 | goto out_unlock; | |
1241 | } | |
1242 | } | |
1243 | ||
1244 | err = -EACCES; | |
1245 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) | |
1246 | goto out_unlock; | |
1247 | ||
1248 | err = security_sem_semctl(sma, cmd); | |
1249 | if (err) | |
1250 | goto out_unlock; | |
1251 | ||
1252 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); | |
1253 | tbuf.sem_otime = get_semotime(sma); | |
1254 | tbuf.sem_ctime = sma->sem_ctime; | |
1255 | tbuf.sem_nsems = sma->sem_nsems; | |
1256 | rcu_read_unlock(); | |
1257 | if (copy_semid_to_user(p, &tbuf, version)) | |
1258 | return -EFAULT; | |
1259 | return id; | |
1260 | } | |
1261 | default: | |
1262 | return -EINVAL; | |
1263 | } | |
1264 | out_unlock: | |
1265 | rcu_read_unlock(); | |
1266 | return err; | |
1267 | } | |
1268 | ||
1269 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, | |
1270 | unsigned long arg) | |
1271 | { | |
1272 | struct sem_undo *un; | |
1273 | struct sem_array *sma; | |
1274 | struct sem *curr; | |
1275 | int err, val; | |
1276 | DEFINE_WAKE_Q(wake_q); | |
1277 | ||
1278 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1279 | /* big-endian 64bit */ | |
1280 | val = arg >> 32; | |
1281 | #else | |
1282 | /* 32bit or little-endian 64bit */ | |
1283 | val = arg; | |
1284 | #endif | |
1285 | ||
1286 | if (val > SEMVMX || val < 0) | |
1287 | return -ERANGE; | |
1288 | ||
1289 | rcu_read_lock(); | |
1290 | sma = sem_obtain_object_check(ns, semid); | |
1291 | if (IS_ERR(sma)) { | |
1292 | rcu_read_unlock(); | |
1293 | return PTR_ERR(sma); | |
1294 | } | |
1295 | ||
1296 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1297 | rcu_read_unlock(); | |
1298 | return -EINVAL; | |
1299 | } | |
1300 | ||
1301 | ||
1302 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1303 | rcu_read_unlock(); | |
1304 | return -EACCES; | |
1305 | } | |
1306 | ||
1307 | err = security_sem_semctl(sma, SETVAL); | |
1308 | if (err) { | |
1309 | rcu_read_unlock(); | |
1310 | return -EACCES; | |
1311 | } | |
1312 | ||
1313 | sem_lock(sma, NULL, -1); | |
1314 | ||
1315 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1316 | sem_unlock(sma, -1); | |
1317 | rcu_read_unlock(); | |
1318 | return -EIDRM; | |
1319 | } | |
1320 | ||
1321 | curr = &sma->sems[semnum]; | |
1322 | ||
1323 | ipc_assert_locked_object(&sma->sem_perm); | |
1324 | list_for_each_entry(un, &sma->list_id, list_id) | |
1325 | un->semadj[semnum] = 0; | |
1326 | ||
1327 | curr->semval = val; | |
1328 | curr->sempid = task_tgid_vnr(current); | |
1329 | sma->sem_ctime = get_seconds(); | |
1330 | /* maybe some queued-up processes were waiting for this */ | |
1331 | do_smart_update(sma, NULL, 0, 0, &wake_q); | |
1332 | sem_unlock(sma, -1); | |
1333 | rcu_read_unlock(); | |
1334 | wake_up_q(&wake_q); | |
1335 | return 0; | |
1336 | } | |
1337 | ||
1338 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, | |
1339 | int cmd, void __user *p) | |
1340 | { | |
1341 | struct sem_array *sma; | |
1342 | struct sem *curr; | |
1343 | int err, nsems; | |
1344 | ushort fast_sem_io[SEMMSL_FAST]; | |
1345 | ushort *sem_io = fast_sem_io; | |
1346 | DEFINE_WAKE_Q(wake_q); | |
1347 | ||
1348 | rcu_read_lock(); | |
1349 | sma = sem_obtain_object_check(ns, semid); | |
1350 | if (IS_ERR(sma)) { | |
1351 | rcu_read_unlock(); | |
1352 | return PTR_ERR(sma); | |
1353 | } | |
1354 | ||
1355 | nsems = sma->sem_nsems; | |
1356 | ||
1357 | err = -EACCES; | |
1358 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) | |
1359 | goto out_rcu_wakeup; | |
1360 | ||
1361 | err = security_sem_semctl(sma, cmd); | |
1362 | if (err) | |
1363 | goto out_rcu_wakeup; | |
1364 | ||
1365 | err = -EACCES; | |
1366 | switch (cmd) { | |
1367 | case GETALL: | |
1368 | { | |
1369 | ushort __user *array = p; | |
1370 | int i; | |
1371 | ||
1372 | sem_lock(sma, NULL, -1); | |
1373 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1374 | err = -EIDRM; | |
1375 | goto out_unlock; | |
1376 | } | |
1377 | if (nsems > SEMMSL_FAST) { | |
1378 | if (!ipc_rcu_getref(&sma->sem_perm)) { | |
1379 | err = -EIDRM; | |
1380 | goto out_unlock; | |
1381 | } | |
1382 | sem_unlock(sma, -1); | |
1383 | rcu_read_unlock(); | |
1384 | sem_io = kvmalloc_array(nsems, sizeof(ushort), | |
1385 | GFP_KERNEL); | |
1386 | if (sem_io == NULL) { | |
1387 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1388 | return -ENOMEM; | |
1389 | } | |
1390 | ||
1391 | rcu_read_lock(); | |
1392 | sem_lock_and_putref(sma); | |
1393 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1394 | err = -EIDRM; | |
1395 | goto out_unlock; | |
1396 | } | |
1397 | } | |
1398 | for (i = 0; i < sma->sem_nsems; i++) | |
1399 | sem_io[i] = sma->sems[i].semval; | |
1400 | sem_unlock(sma, -1); | |
1401 | rcu_read_unlock(); | |
1402 | err = 0; | |
1403 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
1404 | err = -EFAULT; | |
1405 | goto out_free; | |
1406 | } | |
1407 | case SETALL: | |
1408 | { | |
1409 | int i; | |
1410 | struct sem_undo *un; | |
1411 | ||
1412 | if (!ipc_rcu_getref(&sma->sem_perm)) { | |
1413 | err = -EIDRM; | |
1414 | goto out_rcu_wakeup; | |
1415 | } | |
1416 | rcu_read_unlock(); | |
1417 | ||
1418 | if (nsems > SEMMSL_FAST) { | |
1419 | sem_io = kvmalloc_array(nsems, sizeof(ushort), | |
1420 | GFP_KERNEL); | |
1421 | if (sem_io == NULL) { | |
1422 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1423 | return -ENOMEM; | |
1424 | } | |
1425 | } | |
1426 | ||
1427 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { | |
1428 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1429 | err = -EFAULT; | |
1430 | goto out_free; | |
1431 | } | |
1432 | ||
1433 | for (i = 0; i < nsems; i++) { | |
1434 | if (sem_io[i] > SEMVMX) { | |
1435 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1436 | err = -ERANGE; | |
1437 | goto out_free; | |
1438 | } | |
1439 | } | |
1440 | rcu_read_lock(); | |
1441 | sem_lock_and_putref(sma); | |
1442 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1443 | err = -EIDRM; | |
1444 | goto out_unlock; | |
1445 | } | |
1446 | ||
1447 | for (i = 0; i < nsems; i++) { | |
1448 | sma->sems[i].semval = sem_io[i]; | |
1449 | sma->sems[i].sempid = task_tgid_vnr(current); | |
1450 | } | |
1451 | ||
1452 | ipc_assert_locked_object(&sma->sem_perm); | |
1453 | list_for_each_entry(un, &sma->list_id, list_id) { | |
1454 | for (i = 0; i < nsems; i++) | |
1455 | un->semadj[i] = 0; | |
1456 | } | |
1457 | sma->sem_ctime = get_seconds(); | |
1458 | /* maybe some queued-up processes were waiting for this */ | |
1459 | do_smart_update(sma, NULL, 0, 0, &wake_q); | |
1460 | err = 0; | |
1461 | goto out_unlock; | |
1462 | } | |
1463 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ | |
1464 | } | |
1465 | err = -EINVAL; | |
1466 | if (semnum < 0 || semnum >= nsems) | |
1467 | goto out_rcu_wakeup; | |
1468 | ||
1469 | sem_lock(sma, NULL, -1); | |
1470 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1471 | err = -EIDRM; | |
1472 | goto out_unlock; | |
1473 | } | |
1474 | curr = &sma->sems[semnum]; | |
1475 | ||
1476 | switch (cmd) { | |
1477 | case GETVAL: | |
1478 | err = curr->semval; | |
1479 | goto out_unlock; | |
1480 | case GETPID: | |
1481 | err = curr->sempid; | |
1482 | goto out_unlock; | |
1483 | case GETNCNT: | |
1484 | err = count_semcnt(sma, semnum, 0); | |
1485 | goto out_unlock; | |
1486 | case GETZCNT: | |
1487 | err = count_semcnt(sma, semnum, 1); | |
1488 | goto out_unlock; | |
1489 | } | |
1490 | ||
1491 | out_unlock: | |
1492 | sem_unlock(sma, -1); | |
1493 | out_rcu_wakeup: | |
1494 | rcu_read_unlock(); | |
1495 | wake_up_q(&wake_q); | |
1496 | out_free: | |
1497 | if (sem_io != fast_sem_io) | |
1498 | kvfree(sem_io); | |
1499 | return err; | |
1500 | } | |
1501 | ||
1502 | static inline unsigned long | |
1503 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1504 | { | |
1505 | switch (version) { | |
1506 | case IPC_64: | |
1507 | if (copy_from_user(out, buf, sizeof(*out))) | |
1508 | return -EFAULT; | |
1509 | return 0; | |
1510 | case IPC_OLD: | |
1511 | { | |
1512 | struct semid_ds tbuf_old; | |
1513 | ||
1514 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
1515 | return -EFAULT; | |
1516 | ||
1517 | out->sem_perm.uid = tbuf_old.sem_perm.uid; | |
1518 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1519 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1520 | ||
1521 | return 0; | |
1522 | } | |
1523 | default: | |
1524 | return -EINVAL; | |
1525 | } | |
1526 | } | |
1527 | ||
1528 | /* | |
1529 | * This function handles some semctl commands which require the rwsem | |
1530 | * to be held in write mode. | |
1531 | * NOTE: no locks must be held, the rwsem is taken inside this function. | |
1532 | */ | |
1533 | static int semctl_down(struct ipc_namespace *ns, int semid, | |
1534 | int cmd, int version, void __user *p) | |
1535 | { | |
1536 | struct sem_array *sma; | |
1537 | int err; | |
1538 | struct semid64_ds semid64; | |
1539 | struct kern_ipc_perm *ipcp; | |
1540 | ||
1541 | if (cmd == IPC_SET) { | |
1542 | if (copy_semid_from_user(&semid64, p, version)) | |
1543 | return -EFAULT; | |
1544 | } | |
1545 | ||
1546 | down_write(&sem_ids(ns).rwsem); | |
1547 | rcu_read_lock(); | |
1548 | ||
1549 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, | |
1550 | &semid64.sem_perm, 0); | |
1551 | if (IS_ERR(ipcp)) { | |
1552 | err = PTR_ERR(ipcp); | |
1553 | goto out_unlock1; | |
1554 | } | |
1555 | ||
1556 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
1557 | ||
1558 | err = security_sem_semctl(sma, cmd); | |
1559 | if (err) | |
1560 | goto out_unlock1; | |
1561 | ||
1562 | switch (cmd) { | |
1563 | case IPC_RMID: | |
1564 | sem_lock(sma, NULL, -1); | |
1565 | /* freeary unlocks the ipc object and rcu */ | |
1566 | freeary(ns, ipcp); | |
1567 | goto out_up; | |
1568 | case IPC_SET: | |
1569 | sem_lock(sma, NULL, -1); | |
1570 | err = ipc_update_perm(&semid64.sem_perm, ipcp); | |
1571 | if (err) | |
1572 | goto out_unlock0; | |
1573 | sma->sem_ctime = get_seconds(); | |
1574 | break; | |
1575 | default: | |
1576 | err = -EINVAL; | |
1577 | goto out_unlock1; | |
1578 | } | |
1579 | ||
1580 | out_unlock0: | |
1581 | sem_unlock(sma, -1); | |
1582 | out_unlock1: | |
1583 | rcu_read_unlock(); | |
1584 | out_up: | |
1585 | up_write(&sem_ids(ns).rwsem); | |
1586 | return err; | |
1587 | } | |
1588 | ||
1589 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) | |
1590 | { | |
1591 | int version; | |
1592 | struct ipc_namespace *ns; | |
1593 | void __user *p = (void __user *)arg; | |
1594 | ||
1595 | if (semid < 0) | |
1596 | return -EINVAL; | |
1597 | ||
1598 | version = ipc_parse_version(&cmd); | |
1599 | ns = current->nsproxy->ipc_ns; | |
1600 | ||
1601 | switch (cmd) { | |
1602 | case IPC_INFO: | |
1603 | case SEM_INFO: | |
1604 | case IPC_STAT: | |
1605 | case SEM_STAT: | |
1606 | return semctl_nolock(ns, semid, cmd, version, p); | |
1607 | case GETALL: | |
1608 | case GETVAL: | |
1609 | case GETPID: | |
1610 | case GETNCNT: | |
1611 | case GETZCNT: | |
1612 | case SETALL: | |
1613 | return semctl_main(ns, semid, semnum, cmd, p); | |
1614 | case SETVAL: | |
1615 | return semctl_setval(ns, semid, semnum, arg); | |
1616 | case IPC_RMID: | |
1617 | case IPC_SET: | |
1618 | return semctl_down(ns, semid, cmd, version, p); | |
1619 | default: | |
1620 | return -EINVAL; | |
1621 | } | |
1622 | } | |
1623 | ||
1624 | /* If the task doesn't already have a undo_list, then allocate one | |
1625 | * here. We guarantee there is only one thread using this undo list, | |
1626 | * and current is THE ONE | |
1627 | * | |
1628 | * If this allocation and assignment succeeds, but later | |
1629 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1630 | * Just let it stay associated with the task, and it'll be freed later | |
1631 | * at exit time. | |
1632 | * | |
1633 | * This can block, so callers must hold no locks. | |
1634 | */ | |
1635 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1636 | { | |
1637 | struct sem_undo_list *undo_list; | |
1638 | ||
1639 | undo_list = current->sysvsem.undo_list; | |
1640 | if (!undo_list) { | |
1641 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); | |
1642 | if (undo_list == NULL) | |
1643 | return -ENOMEM; | |
1644 | spin_lock_init(&undo_list->lock); | |
1645 | atomic_set(&undo_list->refcnt, 1); | |
1646 | INIT_LIST_HEAD(&undo_list->list_proc); | |
1647 | ||
1648 | current->sysvsem.undo_list = undo_list; | |
1649 | } | |
1650 | *undo_listp = undo_list; | |
1651 | return 0; | |
1652 | } | |
1653 | ||
1654 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) | |
1655 | { | |
1656 | struct sem_undo *un; | |
1657 | ||
1658 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { | |
1659 | if (un->semid == semid) | |
1660 | return un; | |
1661 | } | |
1662 | return NULL; | |
1663 | } | |
1664 | ||
1665 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) | |
1666 | { | |
1667 | struct sem_undo *un; | |
1668 | ||
1669 | assert_spin_locked(&ulp->lock); | |
1670 | ||
1671 | un = __lookup_undo(ulp, semid); | |
1672 | if (un) { | |
1673 | list_del_rcu(&un->list_proc); | |
1674 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1675 | } | |
1676 | return un; | |
1677 | } | |
1678 | ||
1679 | /** | |
1680 | * find_alloc_undo - lookup (and if not present create) undo array | |
1681 | * @ns: namespace | |
1682 | * @semid: semaphore array id | |
1683 | * | |
1684 | * The function looks up (and if not present creates) the undo structure. | |
1685 | * The size of the undo structure depends on the size of the semaphore | |
1686 | * array, thus the alloc path is not that straightforward. | |
1687 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function | |
1688 | * performs a rcu_read_lock(). | |
1689 | */ | |
1690 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1691 | { | |
1692 | struct sem_array *sma; | |
1693 | struct sem_undo_list *ulp; | |
1694 | struct sem_undo *un, *new; | |
1695 | int nsems, error; | |
1696 | ||
1697 | error = get_undo_list(&ulp); | |
1698 | if (error) | |
1699 | return ERR_PTR(error); | |
1700 | ||
1701 | rcu_read_lock(); | |
1702 | spin_lock(&ulp->lock); | |
1703 | un = lookup_undo(ulp, semid); | |
1704 | spin_unlock(&ulp->lock); | |
1705 | if (likely(un != NULL)) | |
1706 | goto out; | |
1707 | ||
1708 | /* no undo structure around - allocate one. */ | |
1709 | /* step 1: figure out the size of the semaphore array */ | |
1710 | sma = sem_obtain_object_check(ns, semid); | |
1711 | if (IS_ERR(sma)) { | |
1712 | rcu_read_unlock(); | |
1713 | return ERR_CAST(sma); | |
1714 | } | |
1715 | ||
1716 | nsems = sma->sem_nsems; | |
1717 | if (!ipc_rcu_getref(&sma->sem_perm)) { | |
1718 | rcu_read_unlock(); | |
1719 | un = ERR_PTR(-EIDRM); | |
1720 | goto out; | |
1721 | } | |
1722 | rcu_read_unlock(); | |
1723 | ||
1724 | /* step 2: allocate new undo structure */ | |
1725 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); | |
1726 | if (!new) { | |
1727 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); | |
1728 | return ERR_PTR(-ENOMEM); | |
1729 | } | |
1730 | ||
1731 | /* step 3: Acquire the lock on semaphore array */ | |
1732 | rcu_read_lock(); | |
1733 | sem_lock_and_putref(sma); | |
1734 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1735 | sem_unlock(sma, -1); | |
1736 | rcu_read_unlock(); | |
1737 | kfree(new); | |
1738 | un = ERR_PTR(-EIDRM); | |
1739 | goto out; | |
1740 | } | |
1741 | spin_lock(&ulp->lock); | |
1742 | ||
1743 | /* | |
1744 | * step 4: check for races: did someone else allocate the undo struct? | |
1745 | */ | |
1746 | un = lookup_undo(ulp, semid); | |
1747 | if (un) { | |
1748 | kfree(new); | |
1749 | goto success; | |
1750 | } | |
1751 | /* step 5: initialize & link new undo structure */ | |
1752 | new->semadj = (short *) &new[1]; | |
1753 | new->ulp = ulp; | |
1754 | new->semid = semid; | |
1755 | assert_spin_locked(&ulp->lock); | |
1756 | list_add_rcu(&new->list_proc, &ulp->list_proc); | |
1757 | ipc_assert_locked_object(&sma->sem_perm); | |
1758 | list_add(&new->list_id, &sma->list_id); | |
1759 | un = new; | |
1760 | ||
1761 | success: | |
1762 | spin_unlock(&ulp->lock); | |
1763 | sem_unlock(sma, -1); | |
1764 | out: | |
1765 | return un; | |
1766 | } | |
1767 | ||
1768 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, | |
1769 | unsigned, nsops, const struct timespec __user *, timeout) | |
1770 | { | |
1771 | int error = -EINVAL; | |
1772 | struct sem_array *sma; | |
1773 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1774 | struct sembuf *sops = fast_sops, *sop; | |
1775 | struct sem_undo *un; | |
1776 | int max, locknum; | |
1777 | bool undos = false, alter = false, dupsop = false; | |
1778 | struct sem_queue queue; | |
1779 | unsigned long dup = 0, jiffies_left = 0; | |
1780 | struct ipc_namespace *ns; | |
1781 | ||
1782 | ns = current->nsproxy->ipc_ns; | |
1783 | ||
1784 | if (nsops < 1 || semid < 0) | |
1785 | return -EINVAL; | |
1786 | if (nsops > ns->sc_semopm) | |
1787 | return -E2BIG; | |
1788 | if (nsops > SEMOPM_FAST) { | |
1789 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1790 | if (sops == NULL) | |
1791 | return -ENOMEM; | |
1792 | } | |
1793 | ||
1794 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { | |
1795 | error = -EFAULT; | |
1796 | goto out_free; | |
1797 | } | |
1798 | ||
1799 | if (timeout) { | |
1800 | struct timespec _timeout; | |
1801 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1802 | error = -EFAULT; | |
1803 | goto out_free; | |
1804 | } | |
1805 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1806 | _timeout.tv_nsec >= 1000000000L) { | |
1807 | error = -EINVAL; | |
1808 | goto out_free; | |
1809 | } | |
1810 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1811 | } | |
1812 | ||
1813 | max = 0; | |
1814 | for (sop = sops; sop < sops + nsops; sop++) { | |
1815 | unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG); | |
1816 | ||
1817 | if (sop->sem_num >= max) | |
1818 | max = sop->sem_num; | |
1819 | if (sop->sem_flg & SEM_UNDO) | |
1820 | undos = true; | |
1821 | if (dup & mask) { | |
1822 | /* | |
1823 | * There was a previous alter access that appears | |
1824 | * to have accessed the same semaphore, thus use | |
1825 | * the dupsop logic. "appears", because the detection | |
1826 | * can only check % BITS_PER_LONG. | |
1827 | */ | |
1828 | dupsop = true; | |
1829 | } | |
1830 | if (sop->sem_op != 0) { | |
1831 | alter = true; | |
1832 | dup |= mask; | |
1833 | } | |
1834 | } | |
1835 | ||
1836 | if (undos) { | |
1837 | /* On success, find_alloc_undo takes the rcu_read_lock */ | |
1838 | un = find_alloc_undo(ns, semid); | |
1839 | if (IS_ERR(un)) { | |
1840 | error = PTR_ERR(un); | |
1841 | goto out_free; | |
1842 | } | |
1843 | } else { | |
1844 | un = NULL; | |
1845 | rcu_read_lock(); | |
1846 | } | |
1847 | ||
1848 | sma = sem_obtain_object_check(ns, semid); | |
1849 | if (IS_ERR(sma)) { | |
1850 | rcu_read_unlock(); | |
1851 | error = PTR_ERR(sma); | |
1852 | goto out_free; | |
1853 | } | |
1854 | ||
1855 | error = -EFBIG; | |
1856 | if (max >= sma->sem_nsems) { | |
1857 | rcu_read_unlock(); | |
1858 | goto out_free; | |
1859 | } | |
1860 | ||
1861 | error = -EACCES; | |
1862 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) { | |
1863 | rcu_read_unlock(); | |
1864 | goto out_free; | |
1865 | } | |
1866 | ||
1867 | error = security_sem_semop(sma, sops, nsops, alter); | |
1868 | if (error) { | |
1869 | rcu_read_unlock(); | |
1870 | goto out_free; | |
1871 | } | |
1872 | ||
1873 | error = -EIDRM; | |
1874 | locknum = sem_lock(sma, sops, nsops); | |
1875 | /* | |
1876 | * We eventually might perform the following check in a lockless | |
1877 | * fashion, considering ipc_valid_object() locking constraints. | |
1878 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1879 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1880 | * check below. More details on the fine grained locking scheme | |
1881 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1882 | */ | |
1883 | if (!ipc_valid_object(&sma->sem_perm)) | |
1884 | goto out_unlock_free; | |
1885 | /* | |
1886 | * semid identifiers are not unique - find_alloc_undo may have | |
1887 | * allocated an undo structure, it was invalidated by an RMID | |
1888 | * and now a new array with received the same id. Check and fail. | |
1889 | * This case can be detected checking un->semid. The existence of | |
1890 | * "un" itself is guaranteed by rcu. | |
1891 | */ | |
1892 | if (un && un->semid == -1) | |
1893 | goto out_unlock_free; | |
1894 | ||
1895 | queue.sops = sops; | |
1896 | queue.nsops = nsops; | |
1897 | queue.undo = un; | |
1898 | queue.pid = task_tgid_vnr(current); | |
1899 | queue.alter = alter; | |
1900 | queue.dupsop = dupsop; | |
1901 | ||
1902 | error = perform_atomic_semop(sma, &queue); | |
1903 | if (error == 0) { /* non-blocking succesfull path */ | |
1904 | DEFINE_WAKE_Q(wake_q); | |
1905 | ||
1906 | /* | |
1907 | * If the operation was successful, then do | |
1908 | * the required updates. | |
1909 | */ | |
1910 | if (alter) | |
1911 | do_smart_update(sma, sops, nsops, 1, &wake_q); | |
1912 | else | |
1913 | set_semotime(sma, sops); | |
1914 | ||
1915 | sem_unlock(sma, locknum); | |
1916 | rcu_read_unlock(); | |
1917 | wake_up_q(&wake_q); | |
1918 | ||
1919 | goto out_free; | |
1920 | } | |
1921 | if (error < 0) /* non-blocking error path */ | |
1922 | goto out_unlock_free; | |
1923 | ||
1924 | /* | |
1925 | * We need to sleep on this operation, so we put the current | |
1926 | * task into the pending queue and go to sleep. | |
1927 | */ | |
1928 | if (nsops == 1) { | |
1929 | struct sem *curr; | |
1930 | curr = &sma->sems[sops->sem_num]; | |
1931 | ||
1932 | if (alter) { | |
1933 | if (sma->complex_count) { | |
1934 | list_add_tail(&queue.list, | |
1935 | &sma->pending_alter); | |
1936 | } else { | |
1937 | ||
1938 | list_add_tail(&queue.list, | |
1939 | &curr->pending_alter); | |
1940 | } | |
1941 | } else { | |
1942 | list_add_tail(&queue.list, &curr->pending_const); | |
1943 | } | |
1944 | } else { | |
1945 | if (!sma->complex_count) | |
1946 | merge_queues(sma); | |
1947 | ||
1948 | if (alter) | |
1949 | list_add_tail(&queue.list, &sma->pending_alter); | |
1950 | else | |
1951 | list_add_tail(&queue.list, &sma->pending_const); | |
1952 | ||
1953 | sma->complex_count++; | |
1954 | } | |
1955 | ||
1956 | do { | |
1957 | queue.status = -EINTR; | |
1958 | queue.sleeper = current; | |
1959 | ||
1960 | __set_current_state(TASK_INTERRUPTIBLE); | |
1961 | sem_unlock(sma, locknum); | |
1962 | rcu_read_unlock(); | |
1963 | ||
1964 | if (timeout) | |
1965 | jiffies_left = schedule_timeout(jiffies_left); | |
1966 | else | |
1967 | schedule(); | |
1968 | ||
1969 | /* | |
1970 | * fastpath: the semop has completed, either successfully or | |
1971 | * not, from the syscall pov, is quite irrelevant to us at this | |
1972 | * point; we're done. | |
1973 | * | |
1974 | * We _do_ care, nonetheless, about being awoken by a signal or | |
1975 | * spuriously. The queue.status is checked again in the | |
1976 | * slowpath (aka after taking sem_lock), such that we can detect | |
1977 | * scenarios where we were awakened externally, during the | |
1978 | * window between wake_q_add() and wake_up_q(). | |
1979 | */ | |
1980 | error = READ_ONCE(queue.status); | |
1981 | if (error != -EINTR) { | |
1982 | /* | |
1983 | * User space could assume that semop() is a memory | |
1984 | * barrier: Without the mb(), the cpu could | |
1985 | * speculatively read in userspace stale data that was | |
1986 | * overwritten by the previous owner of the semaphore. | |
1987 | */ | |
1988 | smp_mb(); | |
1989 | goto out_free; | |
1990 | } | |
1991 | ||
1992 | rcu_read_lock(); | |
1993 | locknum = sem_lock(sma, sops, nsops); | |
1994 | ||
1995 | if (!ipc_valid_object(&sma->sem_perm)) | |
1996 | goto out_unlock_free; | |
1997 | ||
1998 | error = READ_ONCE(queue.status); | |
1999 | ||
2000 | /* | |
2001 | * If queue.status != -EINTR we are woken up by another process. | |
2002 | * Leave without unlink_queue(), but with sem_unlock(). | |
2003 | */ | |
2004 | if (error != -EINTR) | |
2005 | goto out_unlock_free; | |
2006 | ||
2007 | /* | |
2008 | * If an interrupt occurred we have to clean up the queue. | |
2009 | */ | |
2010 | if (timeout && jiffies_left == 0) | |
2011 | error = -EAGAIN; | |
2012 | } while (error == -EINTR && !signal_pending(current)); /* spurious */ | |
2013 | ||
2014 | unlink_queue(sma, &queue); | |
2015 | ||
2016 | out_unlock_free: | |
2017 | sem_unlock(sma, locknum); | |
2018 | rcu_read_unlock(); | |
2019 | out_free: | |
2020 | if (sops != fast_sops) | |
2021 | kfree(sops); | |
2022 | return error; | |
2023 | } | |
2024 | ||
2025 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, | |
2026 | unsigned, nsops) | |
2027 | { | |
2028 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2029 | } | |
2030 | ||
2031 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2032 | * parent and child tasks. | |
2033 | */ | |
2034 | ||
2035 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2036 | { | |
2037 | struct sem_undo_list *undo_list; | |
2038 | int error; | |
2039 | ||
2040 | if (clone_flags & CLONE_SYSVSEM) { | |
2041 | error = get_undo_list(&undo_list); | |
2042 | if (error) | |
2043 | return error; | |
2044 | atomic_inc(&undo_list->refcnt); | |
2045 | tsk->sysvsem.undo_list = undo_list; | |
2046 | } else | |
2047 | tsk->sysvsem.undo_list = NULL; | |
2048 | ||
2049 | return 0; | |
2050 | } | |
2051 | ||
2052 | /* | |
2053 | * add semadj values to semaphores, free undo structures. | |
2054 | * undo structures are not freed when semaphore arrays are destroyed | |
2055 | * so some of them may be out of date. | |
2056 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2057 | * set of adjustments that needs to be done should be done in an atomic | |
2058 | * manner or not. That is, if we are attempting to decrement the semval | |
2059 | * should we queue up and wait until we can do so legally? | |
2060 | * The original implementation attempted to do this (queue and wait). | |
2061 | * The current implementation does not do so. The POSIX standard | |
2062 | * and SVID should be consulted to determine what behavior is mandated. | |
2063 | */ | |
2064 | void exit_sem(struct task_struct *tsk) | |
2065 | { | |
2066 | struct sem_undo_list *ulp; | |
2067 | ||
2068 | ulp = tsk->sysvsem.undo_list; | |
2069 | if (!ulp) | |
2070 | return; | |
2071 | tsk->sysvsem.undo_list = NULL; | |
2072 | ||
2073 | if (!atomic_dec_and_test(&ulp->refcnt)) | |
2074 | return; | |
2075 | ||
2076 | for (;;) { | |
2077 | struct sem_array *sma; | |
2078 | struct sem_undo *un; | |
2079 | int semid, i; | |
2080 | DEFINE_WAKE_Q(wake_q); | |
2081 | ||
2082 | cond_resched(); | |
2083 | ||
2084 | rcu_read_lock(); | |
2085 | un = list_entry_rcu(ulp->list_proc.next, | |
2086 | struct sem_undo, list_proc); | |
2087 | if (&un->list_proc == &ulp->list_proc) { | |
2088 | /* | |
2089 | * We must wait for freeary() before freeing this ulp, | |
2090 | * in case we raced with last sem_undo. There is a small | |
2091 | * possibility where we exit while freeary() didn't | |
2092 | * finish unlocking sem_undo_list. | |
2093 | */ | |
2094 | spin_unlock_wait(&ulp->lock); | |
2095 | rcu_read_unlock(); | |
2096 | break; | |
2097 | } | |
2098 | spin_lock(&ulp->lock); | |
2099 | semid = un->semid; | |
2100 | spin_unlock(&ulp->lock); | |
2101 | ||
2102 | /* exit_sem raced with IPC_RMID, nothing to do */ | |
2103 | if (semid == -1) { | |
2104 | rcu_read_unlock(); | |
2105 | continue; | |
2106 | } | |
2107 | ||
2108 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); | |
2109 | /* exit_sem raced with IPC_RMID, nothing to do */ | |
2110 | if (IS_ERR(sma)) { | |
2111 | rcu_read_unlock(); | |
2112 | continue; | |
2113 | } | |
2114 | ||
2115 | sem_lock(sma, NULL, -1); | |
2116 | /* exit_sem raced with IPC_RMID, nothing to do */ | |
2117 | if (!ipc_valid_object(&sma->sem_perm)) { | |
2118 | sem_unlock(sma, -1); | |
2119 | rcu_read_unlock(); | |
2120 | continue; | |
2121 | } | |
2122 | un = __lookup_undo(ulp, semid); | |
2123 | if (un == NULL) { | |
2124 | /* exit_sem raced with IPC_RMID+semget() that created | |
2125 | * exactly the same semid. Nothing to do. | |
2126 | */ | |
2127 | sem_unlock(sma, -1); | |
2128 | rcu_read_unlock(); | |
2129 | continue; | |
2130 | } | |
2131 | ||
2132 | /* remove un from the linked lists */ | |
2133 | ipc_assert_locked_object(&sma->sem_perm); | |
2134 | list_del(&un->list_id); | |
2135 | ||
2136 | /* we are the last process using this ulp, acquiring ulp->lock | |
2137 | * isn't required. Besides that, we are also protected against | |
2138 | * IPC_RMID as we hold sma->sem_perm lock now | |
2139 | */ | |
2140 | list_del_rcu(&un->list_proc); | |
2141 | ||
2142 | /* perform adjustments registered in un */ | |
2143 | for (i = 0; i < sma->sem_nsems; i++) { | |
2144 | struct sem *semaphore = &sma->sems[i]; | |
2145 | if (un->semadj[i]) { | |
2146 | semaphore->semval += un->semadj[i]; | |
2147 | /* | |
2148 | * Range checks of the new semaphore value, | |
2149 | * not defined by sus: | |
2150 | * - Some unices ignore the undo entirely | |
2151 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2152 | * - some cap the value (e.g. FreeBSD caps | |
2153 | * at 0, but doesn't enforce SEMVMX) | |
2154 | * | |
2155 | * Linux caps the semaphore value, both at 0 | |
2156 | * and at SEMVMX. | |
2157 | * | |
2158 | * Manfred <manfred@colorfullife.com> | |
2159 | */ | |
2160 | if (semaphore->semval < 0) | |
2161 | semaphore->semval = 0; | |
2162 | if (semaphore->semval > SEMVMX) | |
2163 | semaphore->semval = SEMVMX; | |
2164 | semaphore->sempid = task_tgid_vnr(current); | |
2165 | } | |
2166 | } | |
2167 | /* maybe some queued-up processes were waiting for this */ | |
2168 | do_smart_update(sma, NULL, 0, 1, &wake_q); | |
2169 | sem_unlock(sma, -1); | |
2170 | rcu_read_unlock(); | |
2171 | wake_up_q(&wake_q); | |
2172 | ||
2173 | kfree_rcu(un, rcu); | |
2174 | } | |
2175 | kfree(ulp); | |
2176 | } | |
2177 | ||
2178 | #ifdef CONFIG_PROC_FS | |
2179 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) | |
2180 | { | |
2181 | struct user_namespace *user_ns = seq_user_ns(s); | |
2182 | struct kern_ipc_perm *ipcp = it; | |
2183 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); | |
2184 | time_t sem_otime; | |
2185 | ||
2186 | /* | |
2187 | * The proc interface isn't aware of sem_lock(), it calls | |
2188 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
2189 | * In order to stay compatible with sem_lock(), we must | |
2190 | * enter / leave complex_mode. | |
2191 | */ | |
2192 | complexmode_enter(sma); | |
2193 | ||
2194 | sem_otime = get_semotime(sma); | |
2195 | ||
2196 | seq_printf(s, | |
2197 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", | |
2198 | sma->sem_perm.key, | |
2199 | sma->sem_perm.id, | |
2200 | sma->sem_perm.mode, | |
2201 | sma->sem_nsems, | |
2202 | from_kuid_munged(user_ns, sma->sem_perm.uid), | |
2203 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2204 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2205 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
2206 | sem_otime, | |
2207 | sma->sem_ctime); | |
2208 | ||
2209 | complexmode_tryleave(sma); | |
2210 | ||
2211 | return 0; | |
2212 | } | |
2213 | #endif |