]>
Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * linux/ipc/sem.c | |
4 | * Copyright (C) 1992 Krishna Balasubramanian | |
5 | * Copyright (C) 1995 Eric Schenk, Bruno Haible | |
6 | * | |
1da177e4 LT |
7 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
8 | * | |
9 | * SMP-threaded, sysctl's added | |
624dffcb | 10 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 11 | * Enforced range limit on SEM_UNDO |
046c6884 | 12 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
13 | * Lockless wakeup |
14 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
9ae949fa | 15 | * (c) 2016 Davidlohr Bueso <dave@stgolabs.net> |
c5cf6359 MS |
16 | * Further wakeup optimizations, documentation |
17 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
18 | * |
19 | * support for audit of ipc object properties and permission changes | |
20 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
21 | * |
22 | * namespaces support | |
23 | * OpenVZ, SWsoft Inc. | |
24 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
25 | * |
26 | * Implementation notes: (May 2010) | |
27 | * This file implements System V semaphores. | |
28 | * | |
29 | * User space visible behavior: | |
30 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
31 | * protection) | |
32 | * - multiple semaphore operations that alter the same semaphore in | |
33 | * one semop() are handled. | |
34 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
35 | * SETALL calls. | |
36 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
37 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
38 | * - namespace are supported. | |
39 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
40 | * to /proc/sys/kernel/sem. | |
41 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
42 | * | |
43 | * Internals: | |
44 | * - scalability: | |
45 | * - all global variables are read-mostly. | |
46 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
47 | * - most operations do write operations (actually: spin_lock calls) to | |
48 | * the per-semaphore array structure. | |
49 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
50 | * If multiple semaphores in one array are used, then cache line | |
51 | * trashing on the semaphore array spinlock will limit the scaling. | |
2f2ed41d | 52 | * - semncnt and semzcnt are calculated on demand in count_semcnt() |
c5cf6359 MS |
53 | * - the task that performs a successful semop() scans the list of all |
54 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
55 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
56 | * (see update_queue()) | |
57 | * - To improve the scalability, the actual wake-up calls are performed after | |
9ae949fa | 58 | * dropping all locks. (see wake_up_sem_queue_prepare()) |
c5cf6359 MS |
59 | * - All work is done by the waker, the woken up task does not have to do |
60 | * anything - not even acquiring a lock or dropping a refcount. | |
61 | * - A woken up task may not even touch the semaphore array anymore, it may | |
62 | * have been destroyed already by a semctl(RMID). | |
c5cf6359 MS |
63 | * - UNDO values are stored in an array (one per process and per |
64 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
65 | * modes for the UNDO variables are supported (per process, per thread) | |
66 | * (see copy_semundo, CLONE_SYSVSEM) | |
67 | * - There are two lists of the pending operations: a per-array list | |
68 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
69 | * ordering without always scanning all pending operations. | |
70 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
71 | */ |
72 | ||
1da177e4 LT |
73 | #include <linux/slab.h> |
74 | #include <linux/spinlock.h> | |
75 | #include <linux/init.h> | |
76 | #include <linux/proc_fs.h> | |
77 | #include <linux/time.h> | |
1da177e4 LT |
78 | #include <linux/security.h> |
79 | #include <linux/syscalls.h> | |
80 | #include <linux/audit.h> | |
c59ede7b | 81 | #include <linux/capability.h> |
19b4946c | 82 | #include <linux/seq_file.h> |
3e148c79 | 83 | #include <linux/rwsem.h> |
e3893534 | 84 | #include <linux/nsproxy.h> |
ae5e1b22 | 85 | #include <linux/ipc_namespace.h> |
84f001e1 | 86 | #include <linux/sched/wake_q.h> |
5f921ae9 | 87 | |
7153e402 | 88 | #include <linux/uaccess.h> |
1da177e4 LT |
89 | #include "util.h" |
90 | ||
1a5c1349 EB |
91 | /* One semaphore structure for each semaphore in the system. */ |
92 | struct sem { | |
93 | int semval; /* current value */ | |
94 | /* | |
95 | * PID of the process that last modified the semaphore. For | |
96 | * Linux, specifically these are: | |
97 | * - semop | |
98 | * - semctl, via SETVAL and SETALL. | |
99 | * - at task exit when performing undo adjustments (see exit_sem). | |
100 | */ | |
51d6f263 | 101 | struct pid *sempid; |
1a5c1349 EB |
102 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
103 | struct list_head pending_alter; /* pending single-sop operations */ | |
104 | /* that alter the semaphore */ | |
105 | struct list_head pending_const; /* pending single-sop operations */ | |
106 | /* that do not alter the semaphore*/ | |
107 | time_t sem_otime; /* candidate for sem_otime */ | |
108 | } ____cacheline_aligned_in_smp; | |
109 | ||
110 | /* One sem_array data structure for each set of semaphores in the system. */ | |
111 | struct sem_array { | |
112 | struct kern_ipc_perm sem_perm; /* permissions .. see ipc.h */ | |
113 | time64_t sem_ctime; /* create/last semctl() time */ | |
114 | struct list_head pending_alter; /* pending operations */ | |
115 | /* that alter the array */ | |
116 | struct list_head pending_const; /* pending complex operations */ | |
117 | /* that do not alter semvals */ | |
118 | struct list_head list_id; /* undo requests on this array */ | |
119 | int sem_nsems; /* no. of semaphores in array */ | |
120 | int complex_count; /* pending complex operations */ | |
121 | unsigned int use_global_lock;/* >0: global lock required */ | |
122 | ||
123 | struct sem sems[]; | |
124 | } __randomize_layout; | |
e57940d7 MS |
125 | |
126 | /* One queue for each sleeping process in the system. */ | |
127 | struct sem_queue { | |
e57940d7 MS |
128 | struct list_head list; /* queue of pending operations */ |
129 | struct task_struct *sleeper; /* this process */ | |
130 | struct sem_undo *undo; /* undo structure */ | |
51d6f263 | 131 | struct pid *pid; /* process id of requesting process */ |
e57940d7 MS |
132 | int status; /* completion status of operation */ |
133 | struct sembuf *sops; /* array of pending operations */ | |
ed247b7c | 134 | struct sembuf *blocking; /* the operation that blocked */ |
e57940d7 | 135 | int nsops; /* number of operations */ |
4ce33ec2 DB |
136 | bool alter; /* does *sops alter the array? */ |
137 | bool dupsop; /* sops on more than one sem_num */ | |
e57940d7 MS |
138 | }; |
139 | ||
140 | /* Each task has a list of undo requests. They are executed automatically | |
141 | * when the process exits. | |
142 | */ | |
143 | struct sem_undo { | |
144 | struct list_head list_proc; /* per-process list: * | |
145 | * all undos from one process | |
146 | * rcu protected */ | |
147 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
148 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
149 | struct list_head list_id; /* per semaphore array list: | |
150 | * all undos for one array */ | |
151 | int semid; /* semaphore set identifier */ | |
152 | short *semadj; /* array of adjustments */ | |
153 | /* one per semaphore */ | |
154 | }; | |
155 | ||
156 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
157 | * that may be shared among all a CLONE_SYSVSEM task group. | |
158 | */ | |
159 | struct sem_undo_list { | |
f74370b8 | 160 | refcount_t refcnt; |
e57940d7 MS |
161 | spinlock_t lock; |
162 | struct list_head list_proc; | |
163 | }; | |
164 | ||
165 | ||
ed2ddbf8 | 166 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 167 | |
7748dbfa | 168 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 169 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 170 | #ifdef CONFIG_PROC_FS |
19b4946c | 171 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
172 | #endif |
173 | ||
174 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
175 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
176 | ||
9de5ab8a MS |
177 | /* |
178 | * Switching from the mode suitable for simple ops | |
179 | * to the mode for complex ops is costly. Therefore: | |
180 | * use some hysteresis | |
181 | */ | |
182 | #define USE_GLOBAL_LOCK_HYSTERESIS 10 | |
183 | ||
1da177e4 | 184 | /* |
758a6ba3 | 185 | * Locking: |
5864a2fd | 186 | * a) global sem_lock() for read/write |
1da177e4 | 187 | * sem_undo.id_next, |
758a6ba3 | 188 | * sem_array.complex_count, |
5864a2fd MS |
189 | * sem_array.pending{_alter,_const}, |
190 | * sem_array.sem_undo | |
46c0a8ca | 191 | * |
5864a2fd | 192 | * b) global or semaphore sem_lock() for read/write: |
1a233956 | 193 | * sem_array.sems[i].pending_{const,alter}: |
5864a2fd MS |
194 | * |
195 | * c) special: | |
196 | * sem_undo_list.list_proc: | |
197 | * * undo_list->lock for write | |
198 | * * rcu for read | |
9de5ab8a MS |
199 | * use_global_lock: |
200 | * * global sem_lock() for write | |
201 | * * either local or global sem_lock() for read. | |
202 | * | |
203 | * Memory ordering: | |
204 | * Most ordering is enforced by using spin_lock() and spin_unlock(). | |
205 | * The special case is use_global_lock: | |
206 | * Setting it from non-zero to 0 is a RELEASE, this is ensured by | |
207 | * using smp_store_release(). | |
208 | * Testing if it is non-zero is an ACQUIRE, this is ensured by using | |
209 | * smp_load_acquire(). | |
210 | * Setting it from 0 to non-zero must be ordered with regards to | |
211 | * this smp_load_acquire(), this is guaranteed because the smp_load_acquire() | |
212 | * is inside a spin_lock() and after a write from 0 to non-zero a | |
213 | * spin_lock()+spin_unlock() is done. | |
1da177e4 LT |
214 | */ |
215 | ||
e3893534 KK |
216 | #define sc_semmsl sem_ctls[0] |
217 | #define sc_semmns sem_ctls[1] | |
218 | #define sc_semopm sem_ctls[2] | |
219 | #define sc_semmni sem_ctls[3] | |
220 | ||
0cfb6aee | 221 | int sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 222 | { |
e3893534 KK |
223 | ns->sc_semmsl = SEMMSL; |
224 | ns->sc_semmns = SEMMNS; | |
225 | ns->sc_semopm = SEMOPM; | |
226 | ns->sc_semmni = SEMMNI; | |
227 | ns->used_sems = 0; | |
0cfb6aee | 228 | return ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
229 | } |
230 | ||
ae5e1b22 | 231 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
232 | void sem_exit_ns(struct ipc_namespace *ns) |
233 | { | |
01b8b07a | 234 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 235 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
0cfb6aee | 236 | rhashtable_destroy(&ns->ids[IPC_SEM_IDS].key_ht); |
e3893534 | 237 | } |
ae5e1b22 | 238 | #endif |
1da177e4 | 239 | |
0cfb6aee | 240 | int __init sem_init(void) |
1da177e4 | 241 | { |
0cfb6aee GK |
242 | const int err = sem_init_ns(&init_ipc_ns); |
243 | ||
19b4946c MW |
244 | ipc_init_proc_interface("sysvipc/sem", |
245 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 246 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
0cfb6aee | 247 | return err; |
1da177e4 LT |
248 | } |
249 | ||
f269f40a MS |
250 | /** |
251 | * unmerge_queues - unmerge queues, if possible. | |
252 | * @sma: semaphore array | |
253 | * | |
254 | * The function unmerges the wait queues if complex_count is 0. | |
255 | * It must be called prior to dropping the global semaphore array lock. | |
256 | */ | |
257 | static void unmerge_queues(struct sem_array *sma) | |
258 | { | |
259 | struct sem_queue *q, *tq; | |
260 | ||
261 | /* complex operations still around? */ | |
262 | if (sma->complex_count) | |
263 | return; | |
264 | /* | |
265 | * We will switch back to simple mode. | |
266 | * Move all pending operation back into the per-semaphore | |
267 | * queues. | |
268 | */ | |
269 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
270 | struct sem *curr; | |
1a233956 | 271 | curr = &sma->sems[q->sops[0].sem_num]; |
f269f40a MS |
272 | |
273 | list_add_tail(&q->list, &curr->pending_alter); | |
274 | } | |
275 | INIT_LIST_HEAD(&sma->pending_alter); | |
276 | } | |
277 | ||
278 | /** | |
8001c858 | 279 | * merge_queues - merge single semop queues into global queue |
f269f40a MS |
280 | * @sma: semaphore array |
281 | * | |
282 | * This function merges all per-semaphore queues into the global queue. | |
283 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
284 | * operations when a multi-semop operation must sleep. | |
285 | * Only the alter operations must be moved, the const operations can stay. | |
286 | */ | |
287 | static void merge_queues(struct sem_array *sma) | |
288 | { | |
289 | int i; | |
290 | for (i = 0; i < sma->sem_nsems; i++) { | |
1a233956 | 291 | struct sem *sem = &sma->sems[i]; |
f269f40a MS |
292 | |
293 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
294 | } | |
295 | } | |
296 | ||
53dad6d3 DB |
297 | static void sem_rcu_free(struct rcu_head *head) |
298 | { | |
dba4cdd3 MS |
299 | struct kern_ipc_perm *p = container_of(head, struct kern_ipc_perm, rcu); |
300 | struct sem_array *sma = container_of(p, struct sem_array, sem_perm); | |
53dad6d3 | 301 | |
aefad959 | 302 | security_sem_free(&sma->sem_perm); |
e2029dfe | 303 | kvfree(sma); |
53dad6d3 DB |
304 | } |
305 | ||
5e9d5275 | 306 | /* |
5864a2fd | 307 | * Enter the mode suitable for non-simple operations: |
5e9d5275 | 308 | * Caller must own sem_perm.lock. |
5e9d5275 | 309 | */ |
5864a2fd | 310 | static void complexmode_enter(struct sem_array *sma) |
5e9d5275 MS |
311 | { |
312 | int i; | |
313 | struct sem *sem; | |
314 | ||
9de5ab8a MS |
315 | if (sma->use_global_lock > 0) { |
316 | /* | |
317 | * We are already in global lock mode. | |
318 | * Nothing to do, just reset the | |
319 | * counter until we return to simple mode. | |
320 | */ | |
321 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; | |
6d07b68c MS |
322 | return; |
323 | } | |
9de5ab8a | 324 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; |
5864a2fd | 325 | |
5e9d5275 | 326 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 327 | sem = &sma->sems[i]; |
27d7be18 MS |
328 | spin_lock(&sem->lock); |
329 | spin_unlock(&sem->lock); | |
5e9d5275 | 330 | } |
5864a2fd MS |
331 | } |
332 | ||
333 | /* | |
334 | * Try to leave the mode that disallows simple operations: | |
335 | * Caller must own sem_perm.lock. | |
336 | */ | |
337 | static void complexmode_tryleave(struct sem_array *sma) | |
338 | { | |
339 | if (sma->complex_count) { | |
340 | /* Complex ops are sleeping. | |
341 | * We must stay in complex mode | |
342 | */ | |
343 | return; | |
344 | } | |
9de5ab8a MS |
345 | if (sma->use_global_lock == 1) { |
346 | /* | |
347 | * Immediately after setting use_global_lock to 0, | |
348 | * a simple op can start. Thus: all memory writes | |
349 | * performed by the current operation must be visible | |
350 | * before we set use_global_lock to 0. | |
351 | */ | |
352 | smp_store_release(&sma->use_global_lock, 0); | |
353 | } else { | |
354 | sma->use_global_lock--; | |
355 | } | |
5e9d5275 MS |
356 | } |
357 | ||
5864a2fd | 358 | #define SEM_GLOBAL_LOCK (-1) |
6062a8dc RR |
359 | /* |
360 | * If the request contains only one semaphore operation, and there are | |
361 | * no complex transactions pending, lock only the semaphore involved. | |
362 | * Otherwise, lock the entire semaphore array, since we either have | |
363 | * multiple semaphores in our own semops, or we need to look at | |
364 | * semaphores from other pending complex operations. | |
6062a8dc RR |
365 | */ |
366 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
367 | int nsops) | |
368 | { | |
5e9d5275 | 369 | struct sem *sem; |
6062a8dc | 370 | |
5e9d5275 MS |
371 | if (nsops != 1) { |
372 | /* Complex operation - acquire a full lock */ | |
373 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 374 | |
5864a2fd MS |
375 | /* Prevent parallel simple ops */ |
376 | complexmode_enter(sma); | |
377 | return SEM_GLOBAL_LOCK; | |
5e9d5275 MS |
378 | } |
379 | ||
380 | /* | |
381 | * Only one semaphore affected - try to optimize locking. | |
5864a2fd MS |
382 | * Optimized locking is possible if no complex operation |
383 | * is either enqueued or processed right now. | |
384 | * | |
9de5ab8a | 385 | * Both facts are tracked by use_global_mode. |
5e9d5275 | 386 | */ |
1a233956 | 387 | sem = &sma->sems[sops->sem_num]; |
6062a8dc | 388 | |
5864a2fd | 389 | /* |
9de5ab8a | 390 | * Initial check for use_global_lock. Just an optimization, |
5864a2fd MS |
391 | * no locking, no memory barrier. |
392 | */ | |
9de5ab8a | 393 | if (!sma->use_global_lock) { |
6062a8dc | 394 | /* |
5e9d5275 MS |
395 | * It appears that no complex operation is around. |
396 | * Acquire the per-semaphore lock. | |
6062a8dc | 397 | */ |
5e9d5275 MS |
398 | spin_lock(&sem->lock); |
399 | ||
9de5ab8a MS |
400 | /* pairs with smp_store_release() */ |
401 | if (!smp_load_acquire(&sma->use_global_lock)) { | |
5864a2fd MS |
402 | /* fast path successful! */ |
403 | return sops->sem_num; | |
6062a8dc | 404 | } |
5e9d5275 MS |
405 | spin_unlock(&sem->lock); |
406 | } | |
407 | ||
408 | /* slow path: acquire the full lock */ | |
409 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 410 | |
9de5ab8a MS |
411 | if (sma->use_global_lock == 0) { |
412 | /* | |
413 | * The use_global_lock mode ended while we waited for | |
414 | * sma->sem_perm.lock. Thus we must switch to locking | |
415 | * with sem->lock. | |
416 | * Unlike in the fast path, there is no need to recheck | |
417 | * sma->use_global_lock after we have acquired sem->lock: | |
418 | * We own sma->sem_perm.lock, thus use_global_lock cannot | |
419 | * change. | |
5e9d5275 MS |
420 | */ |
421 | spin_lock(&sem->lock); | |
9de5ab8a | 422 | |
5e9d5275 MS |
423 | ipc_unlock_object(&sma->sem_perm); |
424 | return sops->sem_num; | |
6062a8dc | 425 | } else { |
9de5ab8a MS |
426 | /* |
427 | * Not a false alarm, thus continue to use the global lock | |
428 | * mode. No need for complexmode_enter(), this was done by | |
429 | * the caller that has set use_global_mode to non-zero. | |
6062a8dc | 430 | */ |
5864a2fd | 431 | return SEM_GLOBAL_LOCK; |
6062a8dc | 432 | } |
6062a8dc RR |
433 | } |
434 | ||
435 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
436 | { | |
5864a2fd | 437 | if (locknum == SEM_GLOBAL_LOCK) { |
f269f40a | 438 | unmerge_queues(sma); |
5864a2fd | 439 | complexmode_tryleave(sma); |
cf9d5d78 | 440 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc | 441 | } else { |
1a233956 | 442 | struct sem *sem = &sma->sems[locknum]; |
6062a8dc RR |
443 | spin_unlock(&sem->lock); |
444 | } | |
6062a8dc RR |
445 | } |
446 | ||
3e148c79 | 447 | /* |
d9a605e4 | 448 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 449 | * is not held. |
321310ce LT |
450 | * |
451 | * The caller holds the RCU read lock. | |
3e148c79 | 452 | */ |
16df3674 DB |
453 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
454 | { | |
55b7ae50 | 455 | struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
16df3674 DB |
456 | |
457 | if (IS_ERR(ipcp)) | |
458 | return ERR_CAST(ipcp); | |
459 | ||
460 | return container_of(ipcp, struct sem_array, sem_perm); | |
461 | } | |
462 | ||
16df3674 DB |
463 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
464 | int id) | |
465 | { | |
466 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
467 | ||
468 | if (IS_ERR(ipcp)) | |
469 | return ERR_CAST(ipcp); | |
b1ed88b4 | 470 | |
03f02c76 | 471 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
472 | } |
473 | ||
6ff37972 PP |
474 | static inline void sem_lock_and_putref(struct sem_array *sma) |
475 | { | |
6062a8dc | 476 | sem_lock(sma, NULL, -1); |
dba4cdd3 | 477 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
6ff37972 PP |
478 | } |
479 | ||
7ca7e564 ND |
480 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
481 | { | |
482 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
483 | } | |
484 | ||
101ede01 KC |
485 | static struct sem_array *sem_alloc(size_t nsems) |
486 | { | |
487 | struct sem_array *sma; | |
488 | size_t size; | |
489 | ||
490 | if (nsems > (INT_MAX - sizeof(*sma)) / sizeof(sma->sems[0])) | |
491 | return NULL; | |
492 | ||
493 | size = sizeof(*sma) + nsems * sizeof(sma->sems[0]); | |
494 | sma = kvmalloc(size, GFP_KERNEL); | |
495 | if (unlikely(!sma)) | |
496 | return NULL; | |
497 | ||
498 | memset(sma, 0, size); | |
101ede01 KC |
499 | |
500 | return sma; | |
501 | } | |
502 | ||
f4566f04 ND |
503 | /** |
504 | * newary - Create a new semaphore set | |
505 | * @ns: namespace | |
506 | * @params: ptr to the structure that contains key, semflg and nsems | |
507 | * | |
d9a605e4 | 508 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 | 509 | */ |
7748dbfa | 510 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 | 511 | { |
1da177e4 LT |
512 | int retval; |
513 | struct sem_array *sma; | |
7748dbfa ND |
514 | key_t key = params->key; |
515 | int nsems = params->u.nsems; | |
516 | int semflg = params->flg; | |
b97e820f | 517 | int i; |
1da177e4 LT |
518 | |
519 | if (!nsems) | |
520 | return -EINVAL; | |
e3893534 | 521 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
522 | return -ENOSPC; |
523 | ||
101ede01 | 524 | sma = sem_alloc(nsems); |
3ab08fe2 | 525 | if (!sma) |
1da177e4 | 526 | return -ENOMEM; |
3ab08fe2 | 527 | |
1da177e4 LT |
528 | sma->sem_perm.mode = (semflg & S_IRWXUGO); |
529 | sma->sem_perm.key = key; | |
530 | ||
531 | sma->sem_perm.security = NULL; | |
aefad959 | 532 | retval = security_sem_alloc(&sma->sem_perm); |
1da177e4 | 533 | if (retval) { |
e2029dfe | 534 | kvfree(sma); |
1da177e4 LT |
535 | return retval; |
536 | } | |
537 | ||
6062a8dc | 538 | for (i = 0; i < nsems; i++) { |
1a233956 MS |
539 | INIT_LIST_HEAD(&sma->sems[i].pending_alter); |
540 | INIT_LIST_HEAD(&sma->sems[i].pending_const); | |
541 | spin_lock_init(&sma->sems[i].lock); | |
6062a8dc | 542 | } |
b97e820f MS |
543 | |
544 | sma->complex_count = 0; | |
9de5ab8a | 545 | sma->use_global_lock = USE_GLOBAL_LOCK_HYSTERESIS; |
1a82e9e1 MS |
546 | INIT_LIST_HEAD(&sma->pending_alter); |
547 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 548 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 | 549 | sma->sem_nsems = nsems; |
e54d02b2 | 550 | sma->sem_ctime = ktime_get_real_seconds(); |
e8577d1f | 551 | |
39c96a1b | 552 | /* ipc_addid() locks sma upon success. */ |
2ec55f80 MS |
553 | retval = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
554 | if (retval < 0) { | |
555 | call_rcu(&sma->sem_perm.rcu, sem_rcu_free); | |
556 | return retval; | |
e8577d1f MS |
557 | } |
558 | ns->used_sems += nsems; | |
559 | ||
6062a8dc | 560 | sem_unlock(sma, -1); |
6d49dab8 | 561 | rcu_read_unlock(); |
1da177e4 | 562 | |
7ca7e564 | 563 | return sma->sem_perm.id; |
1da177e4 LT |
564 | } |
565 | ||
7748dbfa | 566 | |
f4566f04 | 567 | /* |
d9a605e4 | 568 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 569 | */ |
03f02c76 | 570 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 571 | { |
aefad959 | 572 | return security_sem_associate(ipcp, semflg); |
7748dbfa ND |
573 | } |
574 | ||
f4566f04 | 575 | /* |
d9a605e4 | 576 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 577 | */ |
03f02c76 ND |
578 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
579 | struct ipc_params *params) | |
7748dbfa | 580 | { |
03f02c76 ND |
581 | struct sem_array *sma; |
582 | ||
583 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
584 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
585 | return -EINVAL; |
586 | ||
587 | return 0; | |
588 | } | |
589 | ||
d5460c99 | 590 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 591 | { |
e3893534 | 592 | struct ipc_namespace *ns; |
eb66ec44 MK |
593 | static const struct ipc_ops sem_ops = { |
594 | .getnew = newary, | |
595 | .associate = sem_security, | |
596 | .more_checks = sem_more_checks, | |
597 | }; | |
7748dbfa | 598 | struct ipc_params sem_params; |
e3893534 KK |
599 | |
600 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 601 | |
e3893534 | 602 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 603 | return -EINVAL; |
7ca7e564 | 604 | |
7748dbfa ND |
605 | sem_params.key = key; |
606 | sem_params.flg = semflg; | |
607 | sem_params.u.nsems = nsems; | |
1da177e4 | 608 | |
7748dbfa | 609 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
610 | } |
611 | ||
78f5009c | 612 | /** |
4ce33ec2 DB |
613 | * perform_atomic_semop[_slow] - Attempt to perform semaphore |
614 | * operations on a given array. | |
758a6ba3 | 615 | * @sma: semaphore array |
d198cd6d | 616 | * @q: struct sem_queue that describes the operation |
758a6ba3 | 617 | * |
4ce33ec2 DB |
618 | * Caller blocking are as follows, based the value |
619 | * indicated by the semaphore operation (sem_op): | |
620 | * | |
621 | * (1) >0 never blocks. | |
622 | * (2) 0 (wait-for-zero operation): semval is non-zero. | |
623 | * (3) <0 attempting to decrement semval to a value smaller than zero. | |
624 | * | |
758a6ba3 MS |
625 | * Returns 0 if the operation was possible. |
626 | * Returns 1 if the operation is impossible, the caller must sleep. | |
4ce33ec2 | 627 | * Returns <0 for error codes. |
1da177e4 | 628 | */ |
4ce33ec2 | 629 | static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q) |
1da177e4 | 630 | { |
51d6f263 EB |
631 | int result, sem_op, nsops; |
632 | struct pid *pid; | |
1da177e4 | 633 | struct sembuf *sop; |
239521f3 | 634 | struct sem *curr; |
d198cd6d MS |
635 | struct sembuf *sops; |
636 | struct sem_undo *un; | |
637 | ||
638 | sops = q->sops; | |
639 | nsops = q->nsops; | |
640 | un = q->undo; | |
1da177e4 LT |
641 | |
642 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 643 | curr = &sma->sems[sop->sem_num]; |
1da177e4 LT |
644 | sem_op = sop->sem_op; |
645 | result = curr->semval; | |
78f5009c | 646 | |
1da177e4 LT |
647 | if (!sem_op && result) |
648 | goto would_block; | |
649 | ||
650 | result += sem_op; | |
651 | if (result < 0) | |
652 | goto would_block; | |
653 | if (result > SEMVMX) | |
654 | goto out_of_range; | |
78f5009c | 655 | |
1da177e4 LT |
656 | if (sop->sem_flg & SEM_UNDO) { |
657 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 658 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
659 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
660 | goto out_of_range; | |
78f5009c | 661 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 662 | } |
78f5009c | 663 | |
1da177e4 LT |
664 | curr->semval = result; |
665 | } | |
666 | ||
667 | sop--; | |
d198cd6d | 668 | pid = q->pid; |
1da177e4 | 669 | while (sop >= sops) { |
51d6f263 | 670 | ipc_update_pid(&sma->sems[sop->sem_num].sempid, pid); |
1da177e4 LT |
671 | sop--; |
672 | } | |
78f5009c | 673 | |
1da177e4 LT |
674 | return 0; |
675 | ||
676 | out_of_range: | |
677 | result = -ERANGE; | |
678 | goto undo; | |
679 | ||
680 | would_block: | |
ed247b7c MS |
681 | q->blocking = sop; |
682 | ||
1da177e4 LT |
683 | if (sop->sem_flg & IPC_NOWAIT) |
684 | result = -EAGAIN; | |
685 | else | |
686 | result = 1; | |
687 | ||
688 | undo: | |
689 | sop--; | |
690 | while (sop >= sops) { | |
78f5009c | 691 | sem_op = sop->sem_op; |
1a233956 | 692 | sma->sems[sop->sem_num].semval -= sem_op; |
78f5009c PM |
693 | if (sop->sem_flg & SEM_UNDO) |
694 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
695 | sop--; |
696 | } | |
697 | ||
698 | return result; | |
699 | } | |
700 | ||
4ce33ec2 DB |
701 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) |
702 | { | |
703 | int result, sem_op, nsops; | |
704 | struct sembuf *sop; | |
705 | struct sem *curr; | |
706 | struct sembuf *sops; | |
707 | struct sem_undo *un; | |
708 | ||
709 | sops = q->sops; | |
710 | nsops = q->nsops; | |
711 | un = q->undo; | |
712 | ||
713 | if (unlikely(q->dupsop)) | |
714 | return perform_atomic_semop_slow(sma, q); | |
715 | ||
716 | /* | |
717 | * We scan the semaphore set twice, first to ensure that the entire | |
718 | * operation can succeed, therefore avoiding any pointless writes | |
719 | * to shared memory and having to undo such changes in order to block | |
720 | * until the operations can go through. | |
721 | */ | |
722 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 723 | curr = &sma->sems[sop->sem_num]; |
4ce33ec2 DB |
724 | sem_op = sop->sem_op; |
725 | result = curr->semval; | |
726 | ||
727 | if (!sem_op && result) | |
728 | goto would_block; /* wait-for-zero */ | |
729 | ||
730 | result += sem_op; | |
731 | if (result < 0) | |
732 | goto would_block; | |
733 | ||
734 | if (result > SEMVMX) | |
735 | return -ERANGE; | |
736 | ||
737 | if (sop->sem_flg & SEM_UNDO) { | |
738 | int undo = un->semadj[sop->sem_num] - sem_op; | |
739 | ||
740 | /* Exceeding the undo range is an error. */ | |
741 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
742 | return -ERANGE; | |
743 | } | |
744 | } | |
745 | ||
746 | for (sop = sops; sop < sops + nsops; sop++) { | |
1a233956 | 747 | curr = &sma->sems[sop->sem_num]; |
4ce33ec2 DB |
748 | sem_op = sop->sem_op; |
749 | result = curr->semval; | |
750 | ||
751 | if (sop->sem_flg & SEM_UNDO) { | |
752 | int undo = un->semadj[sop->sem_num] - sem_op; | |
753 | ||
754 | un->semadj[sop->sem_num] = undo; | |
755 | } | |
756 | curr->semval += sem_op; | |
51d6f263 | 757 | ipc_update_pid(&curr->sempid, q->pid); |
4ce33ec2 DB |
758 | } |
759 | ||
760 | return 0; | |
761 | ||
762 | would_block: | |
763 | q->blocking = sop; | |
764 | return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1; | |
765 | } | |
766 | ||
9ae949fa DB |
767 | static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error, |
768 | struct wake_q_head *wake_q) | |
0a2b9d4c | 769 | { |
9ae949fa DB |
770 | wake_q_add(wake_q, q->sleeper); |
771 | /* | |
772 | * Rely on the above implicit barrier, such that we can | |
773 | * ensure that we hold reference to the task before setting | |
774 | * q->status. Otherwise we could race with do_exit if the | |
775 | * task is awoken by an external event before calling | |
776 | * wake_up_process(). | |
777 | */ | |
778 | WRITE_ONCE(q->status, error); | |
d4212093 NP |
779 | } |
780 | ||
b97e820f MS |
781 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
782 | { | |
783 | list_del(&q->list); | |
9f1bc2c9 | 784 | if (q->nsops > 1) |
b97e820f MS |
785 | sma->complex_count--; |
786 | } | |
787 | ||
fd5db422 MS |
788 | /** check_restart(sma, q) |
789 | * @sma: semaphore array | |
790 | * @q: the operation that just completed | |
791 | * | |
792 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
793 | * waiting operations. Therefore this function checks if the restart is | |
794 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
795 | * modified the array. |
796 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 | 797 | */ |
4663d3e8 | 798 | static inline int check_restart(struct sem_array *sma, struct sem_queue *q) |
fd5db422 | 799 | { |
1a82e9e1 MS |
800 | /* pending complex alter operations are too difficult to analyse */ |
801 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
802 | return 1; |
803 | ||
804 | /* we were a sleeping complex operation. Too difficult */ | |
805 | if (q->nsops > 1) | |
806 | return 1; | |
807 | ||
1a82e9e1 MS |
808 | /* It is impossible that someone waits for the new value: |
809 | * - complex operations always restart. | |
810 | * - wait-for-zero are handled seperately. | |
811 | * - q is a previously sleeping simple operation that | |
812 | * altered the array. It must be a decrement, because | |
813 | * simple increments never sleep. | |
814 | * - If there are older (higher priority) decrements | |
815 | * in the queue, then they have observed the original | |
816 | * semval value and couldn't proceed. The operation | |
817 | * decremented to value - thus they won't proceed either. | |
818 | */ | |
819 | return 0; | |
820 | } | |
fd5db422 | 821 | |
1a82e9e1 | 822 | /** |
8001c858 | 823 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
824 | * @sma: semaphore array. |
825 | * @semnum: semaphore that was modified. | |
9ae949fa | 826 | * @wake_q: lockless wake-queue head. |
1a82e9e1 MS |
827 | * |
828 | * wake_const_ops must be called after a semaphore in a semaphore array | |
829 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
830 | * be called with semnum = -1, as well as with the number of each modified | |
831 | * semaphore. | |
9ae949fa | 832 | * The tasks that must be woken up are added to @wake_q. The return code |
1a82e9e1 MS |
833 | * is stored in q->pid. |
834 | * The function returns 1 if at least one operation was completed successfully. | |
835 | */ | |
836 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
9ae949fa | 837 | struct wake_q_head *wake_q) |
1a82e9e1 | 838 | { |
f150f02c | 839 | struct sem_queue *q, *tmp; |
1a82e9e1 MS |
840 | struct list_head *pending_list; |
841 | int semop_completed = 0; | |
842 | ||
843 | if (semnum == -1) | |
844 | pending_list = &sma->pending_const; | |
845 | else | |
1a233956 | 846 | pending_list = &sma->sems[semnum].pending_const; |
fd5db422 | 847 | |
f150f02c DB |
848 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
849 | int error = perform_atomic_semop(sma, q); | |
1a82e9e1 | 850 | |
f150f02c DB |
851 | if (error > 0) |
852 | continue; | |
853 | /* operation completed, remove from queue & wakeup */ | |
854 | unlink_queue(sma, q); | |
1a82e9e1 | 855 | |
f150f02c DB |
856 | wake_up_sem_queue_prepare(q, error, wake_q); |
857 | if (error == 0) | |
858 | semop_completed = 1; | |
1a82e9e1 | 859 | } |
f150f02c | 860 | |
1a82e9e1 MS |
861 | return semop_completed; |
862 | } | |
863 | ||
864 | /** | |
8001c858 | 865 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
866 | * @sma: semaphore array |
867 | * @sops: operations that were performed | |
868 | * @nsops: number of operations | |
9ae949fa | 869 | * @wake_q: lockless wake-queue head |
1a82e9e1 | 870 | * |
8001c858 DB |
871 | * Checks all required queue for wait-for-zero operations, based |
872 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
873 | * The function returns 1 if at least one operation was completed successfully. |
874 | */ | |
875 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
9ae949fa | 876 | int nsops, struct wake_q_head *wake_q) |
1a82e9e1 MS |
877 | { |
878 | int i; | |
879 | int semop_completed = 0; | |
880 | int got_zero = 0; | |
881 | ||
882 | /* first: the per-semaphore queues, if known */ | |
883 | if (sops) { | |
884 | for (i = 0; i < nsops; i++) { | |
885 | int num = sops[i].sem_num; | |
886 | ||
1a233956 | 887 | if (sma->sems[num].semval == 0) { |
1a82e9e1 | 888 | got_zero = 1; |
9ae949fa | 889 | semop_completed |= wake_const_ops(sma, num, wake_q); |
1a82e9e1 MS |
890 | } |
891 | } | |
892 | } else { | |
893 | /* | |
894 | * No sops means modified semaphores not known. | |
895 | * Assume all were changed. | |
fd5db422 | 896 | */ |
1a82e9e1 | 897 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 898 | if (sma->sems[i].semval == 0) { |
1a82e9e1 | 899 | got_zero = 1; |
9ae949fa | 900 | semop_completed |= wake_const_ops(sma, i, wake_q); |
1a82e9e1 MS |
901 | } |
902 | } | |
fd5db422 MS |
903 | } |
904 | /* | |
1a82e9e1 MS |
905 | * If one of the modified semaphores got 0, |
906 | * then check the global queue, too. | |
fd5db422 | 907 | */ |
1a82e9e1 | 908 | if (got_zero) |
9ae949fa | 909 | semop_completed |= wake_const_ops(sma, -1, wake_q); |
fd5db422 | 910 | |
1a82e9e1 | 911 | return semop_completed; |
fd5db422 MS |
912 | } |
913 | ||
636c6be8 MS |
914 | |
915 | /** | |
8001c858 | 916 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
917 | * @sma: semaphore array. |
918 | * @semnum: semaphore that was modified. | |
9ae949fa | 919 | * @wake_q: lockless wake-queue head. |
636c6be8 MS |
920 | * |
921 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
922 | * was modified. If multiple semaphores were modified, update_queue must |
923 | * be called with semnum = -1, as well as with the number of each modified | |
924 | * semaphore. | |
9ae949fa | 925 | * The tasks that must be woken up are added to @wake_q. The return code |
0a2b9d4c | 926 | * is stored in q->pid. |
1a82e9e1 MS |
927 | * The function internally checks if const operations can now succeed. |
928 | * | |
0a2b9d4c | 929 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 930 | */ |
9ae949fa | 931 | static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q) |
1da177e4 | 932 | { |
f150f02c | 933 | struct sem_queue *q, *tmp; |
636c6be8 | 934 | struct list_head *pending_list; |
0a2b9d4c | 935 | int semop_completed = 0; |
636c6be8 | 936 | |
9f1bc2c9 | 937 | if (semnum == -1) |
1a82e9e1 | 938 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 939 | else |
1a233956 | 940 | pending_list = &sma->sems[semnum].pending_alter; |
9cad200c NP |
941 | |
942 | again: | |
f150f02c | 943 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
fd5db422 | 944 | int error, restart; |
636c6be8 | 945 | |
d987f8b2 MS |
946 | /* If we are scanning the single sop, per-semaphore list of |
947 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 948 | * necessary to scan further: simple increments |
d987f8b2 MS |
949 | * that affect only one entry succeed immediately and cannot |
950 | * be in the per semaphore pending queue, and decrements | |
951 | * cannot be successful if the value is already 0. | |
952 | */ | |
1a233956 | 953 | if (semnum != -1 && sma->sems[semnum].semval == 0) |
d987f8b2 MS |
954 | break; |
955 | ||
d198cd6d | 956 | error = perform_atomic_semop(sma, q); |
1da177e4 LT |
957 | |
958 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
959 | if (error > 0) |
960 | continue; | |
961 | ||
b97e820f | 962 | unlink_queue(sma, q); |
9cad200c | 963 | |
0a2b9d4c | 964 | if (error) { |
fd5db422 | 965 | restart = 0; |
0a2b9d4c MS |
966 | } else { |
967 | semop_completed = 1; | |
9ae949fa | 968 | do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q); |
fd5db422 | 969 | restart = check_restart(sma, q); |
0a2b9d4c | 970 | } |
fd5db422 | 971 | |
9ae949fa | 972 | wake_up_sem_queue_prepare(q, error, wake_q); |
fd5db422 | 973 | if (restart) |
9cad200c | 974 | goto again; |
1da177e4 | 975 | } |
0a2b9d4c | 976 | return semop_completed; |
1da177e4 LT |
977 | } |
978 | ||
0e8c6656 | 979 | /** |
8001c858 | 980 | * set_semotime - set sem_otime |
0e8c6656 MS |
981 | * @sma: semaphore array |
982 | * @sops: operations that modified the array, may be NULL | |
983 | * | |
984 | * sem_otime is replicated to avoid cache line trashing. | |
985 | * This function sets one instance to the current time. | |
986 | */ | |
987 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
988 | { | |
989 | if (sops == NULL) { | |
1a233956 | 990 | sma->sems[0].sem_otime = get_seconds(); |
0e8c6656 | 991 | } else { |
1a233956 | 992 | sma->sems[sops[0].sem_num].sem_otime = |
0e8c6656 MS |
993 | get_seconds(); |
994 | } | |
995 | } | |
996 | ||
0a2b9d4c | 997 | /** |
8001c858 | 998 | * do_smart_update - optimized update_queue |
fd5db422 MS |
999 | * @sma: semaphore array |
1000 | * @sops: operations that were performed | |
1001 | * @nsops: number of operations | |
0a2b9d4c | 1002 | * @otime: force setting otime |
9ae949fa | 1003 | * @wake_q: lockless wake-queue head |
fd5db422 | 1004 | * |
1a82e9e1 MS |
1005 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
1006 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c | 1007 | * Note that the function does not do the actual wake-up: the caller is |
9ae949fa | 1008 | * responsible for calling wake_up_q(). |
0a2b9d4c | 1009 | * It is safe to perform this call after dropping all locks. |
fd5db422 | 1010 | */ |
0a2b9d4c | 1011 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
9ae949fa | 1012 | int otime, struct wake_q_head *wake_q) |
fd5db422 MS |
1013 | { |
1014 | int i; | |
1015 | ||
9ae949fa | 1016 | otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q); |
1a82e9e1 | 1017 | |
f269f40a MS |
1018 | if (!list_empty(&sma->pending_alter)) { |
1019 | /* semaphore array uses the global queue - just process it. */ | |
9ae949fa | 1020 | otime |= update_queue(sma, -1, wake_q); |
f269f40a MS |
1021 | } else { |
1022 | if (!sops) { | |
1023 | /* | |
1024 | * No sops, thus the modified semaphores are not | |
1025 | * known. Check all. | |
1026 | */ | |
1027 | for (i = 0; i < sma->sem_nsems; i++) | |
9ae949fa | 1028 | otime |= update_queue(sma, i, wake_q); |
f269f40a MS |
1029 | } else { |
1030 | /* | |
1031 | * Check the semaphores that were increased: | |
1032 | * - No complex ops, thus all sleeping ops are | |
1033 | * decrease. | |
1034 | * - if we decreased the value, then any sleeping | |
1035 | * semaphore ops wont be able to run: If the | |
1036 | * previous value was too small, then the new | |
1037 | * value will be too small, too. | |
1038 | */ | |
1039 | for (i = 0; i < nsops; i++) { | |
1040 | if (sops[i].sem_op > 0) { | |
1041 | otime |= update_queue(sma, | |
9ae949fa | 1042 | sops[i].sem_num, wake_q); |
f269f40a | 1043 | } |
ab465df9 | 1044 | } |
9f1bc2c9 | 1045 | } |
fd5db422 | 1046 | } |
0e8c6656 MS |
1047 | if (otime) |
1048 | set_semotime(sma, sops); | |
fd5db422 MS |
1049 | } |
1050 | ||
2f2ed41d | 1051 | /* |
b220c57a | 1052 | * check_qop: Test if a queued operation sleeps on the semaphore semnum |
2f2ed41d MS |
1053 | */ |
1054 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1055 | bool count_zero) | |
1056 | { | |
b220c57a | 1057 | struct sembuf *sop = q->blocking; |
2f2ed41d | 1058 | |
9b44ee2e MS |
1059 | /* |
1060 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1061 | * semaphores. This violates SUS, therefore it was changed to the | |
1062 | * standard compliant behavior. | |
1063 | * Give the administrators a chance to notice that an application | |
1064 | * might misbehave because it relies on the Linux behavior. | |
1065 | */ | |
1066 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1067 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1068 | current->comm, task_pid_nr(current)); | |
1069 | ||
b220c57a MS |
1070 | if (sop->sem_num != semnum) |
1071 | return 0; | |
2f2ed41d | 1072 | |
b220c57a MS |
1073 | if (count_zero && sop->sem_op == 0) |
1074 | return 1; | |
1075 | if (!count_zero && sop->sem_op < 0) | |
1076 | return 1; | |
1077 | ||
1078 | return 0; | |
2f2ed41d MS |
1079 | } |
1080 | ||
1da177e4 LT |
1081 | /* The following counts are associated to each semaphore: |
1082 | * semncnt number of tasks waiting on semval being nonzero | |
1083 | * semzcnt number of tasks waiting on semval being zero | |
b220c57a MS |
1084 | * |
1085 | * Per definition, a task waits only on the semaphore of the first semop | |
1086 | * that cannot proceed, even if additional operation would block, too. | |
1da177e4 | 1087 | */ |
2f2ed41d MS |
1088 | static int count_semcnt(struct sem_array *sma, ushort semnum, |
1089 | bool count_zero) | |
1da177e4 | 1090 | { |
2f2ed41d | 1091 | struct list_head *l; |
239521f3 | 1092 | struct sem_queue *q; |
2f2ed41d | 1093 | int semcnt; |
1da177e4 | 1094 | |
2f2ed41d MS |
1095 | semcnt = 0; |
1096 | /* First: check the simple operations. They are easy to evaluate */ | |
1097 | if (count_zero) | |
1a233956 | 1098 | l = &sma->sems[semnum].pending_const; |
2f2ed41d | 1099 | else |
1a233956 | 1100 | l = &sma->sems[semnum].pending_alter; |
1da177e4 | 1101 | |
2f2ed41d MS |
1102 | list_for_each_entry(q, l, list) { |
1103 | /* all task on a per-semaphore list sleep on exactly | |
1104 | * that semaphore | |
1105 | */ | |
1106 | semcnt++; | |
ebc2e5e6 RR |
1107 | } |
1108 | ||
2f2ed41d | 1109 | /* Then: check the complex operations. */ |
1994862d | 1110 | list_for_each_entry(q, &sma->pending_alter, list) { |
2f2ed41d MS |
1111 | semcnt += check_qop(sma, semnum, q, count_zero); |
1112 | } | |
1113 | if (count_zero) { | |
1114 | list_for_each_entry(q, &sma->pending_const, list) { | |
1115 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1116 | } | |
1994862d | 1117 | } |
2f2ed41d | 1118 | return semcnt; |
1da177e4 LT |
1119 | } |
1120 | ||
d9a605e4 DB |
1121 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1122 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1123 | * remains locked on exit. |
1da177e4 | 1124 | */ |
01b8b07a | 1125 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1126 | { |
380af1b3 MS |
1127 | struct sem_undo *un, *tu; |
1128 | struct sem_queue *q, *tq; | |
01b8b07a | 1129 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
9f1bc2c9 | 1130 | int i; |
9ae949fa | 1131 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1132 | |
380af1b3 | 1133 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1134 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1135 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1136 | list_del(&un->list_id); | |
1137 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1138 | un->semid = -1; |
380af1b3 MS |
1139 | list_del_rcu(&un->list_proc); |
1140 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1141 | kfree_rcu(un, rcu); |
380af1b3 | 1142 | } |
1da177e4 LT |
1143 | |
1144 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
1a82e9e1 MS |
1145 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1146 | unlink_queue(sma, q); | |
9ae949fa | 1147 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1148 | } |
1149 | ||
1150 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1151 | unlink_queue(sma, q); |
9ae949fa | 1152 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1da177e4 | 1153 | } |
9f1bc2c9 | 1154 | for (i = 0; i < sma->sem_nsems; i++) { |
1a233956 | 1155 | struct sem *sem = &sma->sems[i]; |
1a82e9e1 MS |
1156 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1157 | unlink_queue(sma, q); | |
9ae949fa | 1158 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1159 | } |
1160 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 | 1161 | unlink_queue(sma, q); |
9ae949fa | 1162 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
9f1bc2c9 | 1163 | } |
51d6f263 | 1164 | ipc_update_pid(&sem->sempid, NULL); |
9f1bc2c9 | 1165 | } |
1da177e4 | 1166 | |
7ca7e564 ND |
1167 | /* Remove the semaphore set from the IDR */ |
1168 | sem_rmid(ns, sma); | |
6062a8dc | 1169 | sem_unlock(sma, -1); |
6d49dab8 | 1170 | rcu_read_unlock(); |
1da177e4 | 1171 | |
9ae949fa | 1172 | wake_up_q(&wake_q); |
e3893534 | 1173 | ns->used_sems -= sma->sem_nsems; |
dba4cdd3 | 1174 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1175 | } |
1176 | ||
1177 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1178 | { | |
239521f3 | 1179 | switch (version) { |
1da177e4 LT |
1180 | case IPC_64: |
1181 | return copy_to_user(buf, in, sizeof(*in)); | |
1182 | case IPC_OLD: | |
1183 | { | |
1184 | struct semid_ds out; | |
1185 | ||
982f7c2b DR |
1186 | memset(&out, 0, sizeof(out)); |
1187 | ||
1da177e4 LT |
1188 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1189 | ||
1190 | out.sem_otime = in->sem_otime; | |
1191 | out.sem_ctime = in->sem_ctime; | |
1192 | out.sem_nsems = in->sem_nsems; | |
1193 | ||
1194 | return copy_to_user(buf, &out, sizeof(out)); | |
1195 | } | |
1196 | default: | |
1197 | return -EINVAL; | |
1198 | } | |
1199 | } | |
1200 | ||
e54d02b2 | 1201 | static time64_t get_semotime(struct sem_array *sma) |
d12e1e50 MS |
1202 | { |
1203 | int i; | |
e54d02b2 | 1204 | time64_t res; |
d12e1e50 | 1205 | |
1a233956 | 1206 | res = sma->sems[0].sem_otime; |
d12e1e50 | 1207 | for (i = 1; i < sma->sem_nsems; i++) { |
e54d02b2 | 1208 | time64_t to = sma->sems[i].sem_otime; |
d12e1e50 MS |
1209 | |
1210 | if (to > res) | |
1211 | res = to; | |
1212 | } | |
1213 | return res; | |
1214 | } | |
1215 | ||
45a4a64a AV |
1216 | static int semctl_stat(struct ipc_namespace *ns, int semid, |
1217 | int cmd, struct semid64_ds *semid64) | |
1da177e4 | 1218 | { |
1da177e4 | 1219 | struct sem_array *sma; |
45a4a64a AV |
1220 | int id = 0; |
1221 | int err; | |
1da177e4 | 1222 | |
45a4a64a | 1223 | memset(semid64, 0, sizeof(*semid64)); |
46c0a8ca | 1224 | |
45a4a64a AV |
1225 | rcu_read_lock(); |
1226 | if (cmd == SEM_STAT) { | |
1227 | sma = sem_obtain_object(ns, semid); | |
1228 | if (IS_ERR(sma)) { | |
1229 | err = PTR_ERR(sma); | |
1230 | goto out_unlock; | |
1231 | } | |
1232 | id = sma->sem_perm.id; | |
1233 | } else { | |
1234 | sma = sem_obtain_object_check(ns, semid); | |
1235 | if (IS_ERR(sma)) { | |
1236 | err = PTR_ERR(sma); | |
1237 | goto out_unlock; | |
1da177e4 | 1238 | } |
1da177e4 | 1239 | } |
1da177e4 | 1240 | |
45a4a64a AV |
1241 | err = -EACCES; |
1242 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) | |
1243 | goto out_unlock; | |
1da177e4 | 1244 | |
aefad959 | 1245 | err = security_sem_semctl(&sma->sem_perm, cmd); |
45a4a64a AV |
1246 | if (err) |
1247 | goto out_unlock; | |
1da177e4 | 1248 | |
87ad4b0d PM |
1249 | ipc_lock_object(&sma->sem_perm); |
1250 | ||
1251 | if (!ipc_valid_object(&sma->sem_perm)) { | |
1252 | ipc_unlock_object(&sma->sem_perm); | |
1253 | err = -EIDRM; | |
1254 | goto out_unlock; | |
1255 | } | |
1256 | ||
45a4a64a AV |
1257 | kernel_to_ipc64_perm(&sma->sem_perm, &semid64->sem_perm); |
1258 | semid64->sem_otime = get_semotime(sma); | |
1259 | semid64->sem_ctime = sma->sem_ctime; | |
1260 | semid64->sem_nsems = sma->sem_nsems; | |
87ad4b0d PM |
1261 | |
1262 | ipc_unlock_object(&sma->sem_perm); | |
45a4a64a AV |
1263 | rcu_read_unlock(); |
1264 | return id; | |
1da177e4 | 1265 | |
1da177e4 | 1266 | out_unlock: |
16df3674 | 1267 | rcu_read_unlock(); |
1da177e4 LT |
1268 | return err; |
1269 | } | |
1270 | ||
45a4a64a AV |
1271 | static int semctl_info(struct ipc_namespace *ns, int semid, |
1272 | int cmd, void __user *p) | |
1273 | { | |
1274 | struct seminfo seminfo; | |
1275 | int max_id; | |
1276 | int err; | |
1277 | ||
1278 | err = security_sem_semctl(NULL, cmd); | |
1279 | if (err) | |
1280 | return err; | |
1281 | ||
1282 | memset(&seminfo, 0, sizeof(seminfo)); | |
1283 | seminfo.semmni = ns->sc_semmni; | |
1284 | seminfo.semmns = ns->sc_semmns; | |
1285 | seminfo.semmsl = ns->sc_semmsl; | |
1286 | seminfo.semopm = ns->sc_semopm; | |
1287 | seminfo.semvmx = SEMVMX; | |
1288 | seminfo.semmnu = SEMMNU; | |
1289 | seminfo.semmap = SEMMAP; | |
1290 | seminfo.semume = SEMUME; | |
1291 | down_read(&sem_ids(ns).rwsem); | |
1292 | if (cmd == SEM_INFO) { | |
1293 | seminfo.semusz = sem_ids(ns).in_use; | |
1294 | seminfo.semaem = ns->used_sems; | |
1295 | } else { | |
1296 | seminfo.semusz = SEMUSZ; | |
1297 | seminfo.semaem = SEMAEM; | |
1298 | } | |
1299 | max_id = ipc_get_maxid(&sem_ids(ns)); | |
1300 | up_read(&sem_ids(ns).rwsem); | |
1301 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) | |
1302 | return -EFAULT; | |
1303 | return (max_id < 0) ? 0 : max_id; | |
1304 | } | |
1305 | ||
e1fd1f49 | 1306 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
45a4a64a | 1307 | int val) |
e1fd1f49 AV |
1308 | { |
1309 | struct sem_undo *un; | |
1310 | struct sem_array *sma; | |
239521f3 | 1311 | struct sem *curr; |
45a4a64a | 1312 | int err; |
9ae949fa DB |
1313 | DEFINE_WAKE_Q(wake_q); |
1314 | ||
6062a8dc RR |
1315 | if (val > SEMVMX || val < 0) |
1316 | return -ERANGE; | |
e1fd1f49 | 1317 | |
6062a8dc RR |
1318 | rcu_read_lock(); |
1319 | sma = sem_obtain_object_check(ns, semid); | |
1320 | if (IS_ERR(sma)) { | |
1321 | rcu_read_unlock(); | |
1322 | return PTR_ERR(sma); | |
1323 | } | |
1324 | ||
1325 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1326 | rcu_read_unlock(); | |
1327 | return -EINVAL; | |
1328 | } | |
1329 | ||
1330 | ||
1331 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1332 | rcu_read_unlock(); | |
1333 | return -EACCES; | |
1334 | } | |
e1fd1f49 | 1335 | |
aefad959 | 1336 | err = security_sem_semctl(&sma->sem_perm, SETVAL); |
6062a8dc RR |
1337 | if (err) { |
1338 | rcu_read_unlock(); | |
1339 | return -EACCES; | |
1340 | } | |
e1fd1f49 | 1341 | |
6062a8dc | 1342 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1343 | |
0f3d2b01 | 1344 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1345 | sem_unlock(sma, -1); |
1346 | rcu_read_unlock(); | |
1347 | return -EIDRM; | |
1348 | } | |
1349 | ||
1a233956 | 1350 | curr = &sma->sems[semnum]; |
e1fd1f49 | 1351 | |
cf9d5d78 | 1352 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1353 | list_for_each_entry(un, &sma->list_id, list_id) |
1354 | un->semadj[semnum] = 0; | |
1355 | ||
1356 | curr->semval = val; | |
51d6f263 | 1357 | ipc_update_pid(&curr->sempid, task_tgid(current)); |
e54d02b2 | 1358 | sma->sem_ctime = ktime_get_real_seconds(); |
e1fd1f49 | 1359 | /* maybe some queued-up processes were waiting for this */ |
9ae949fa | 1360 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
6062a8dc | 1361 | sem_unlock(sma, -1); |
6d49dab8 | 1362 | rcu_read_unlock(); |
9ae949fa | 1363 | wake_up_q(&wake_q); |
6062a8dc | 1364 | return 0; |
e1fd1f49 AV |
1365 | } |
1366 | ||
e3893534 | 1367 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1368 | int cmd, void __user *p) |
1da177e4 LT |
1369 | { |
1370 | struct sem_array *sma; | |
239521f3 | 1371 | struct sem *curr; |
16df3674 | 1372 | int err, nsems; |
1da177e4 | 1373 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1374 | ushort *sem_io = fast_sem_io; |
9ae949fa | 1375 | DEFINE_WAKE_Q(wake_q); |
16df3674 DB |
1376 | |
1377 | rcu_read_lock(); | |
1378 | sma = sem_obtain_object_check(ns, semid); | |
1379 | if (IS_ERR(sma)) { | |
1380 | rcu_read_unlock(); | |
023a5355 | 1381 | return PTR_ERR(sma); |
16df3674 | 1382 | } |
1da177e4 LT |
1383 | |
1384 | nsems = sma->sem_nsems; | |
1385 | ||
1da177e4 | 1386 | err = -EACCES; |
c728b9c8 LT |
1387 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1388 | goto out_rcu_wakeup; | |
1da177e4 | 1389 | |
aefad959 | 1390 | err = security_sem_semctl(&sma->sem_perm, cmd); |
c728b9c8 LT |
1391 | if (err) |
1392 | goto out_rcu_wakeup; | |
1da177e4 LT |
1393 | |
1394 | err = -EACCES; | |
1395 | switch (cmd) { | |
1396 | case GETALL: | |
1397 | { | |
e1fd1f49 | 1398 | ushort __user *array = p; |
1da177e4 LT |
1399 | int i; |
1400 | ||
ce857229 | 1401 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1402 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1403 | err = -EIDRM; |
1404 | goto out_unlock; | |
1405 | } | |
239521f3 | 1406 | if (nsems > SEMMSL_FAST) { |
dba4cdd3 | 1407 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
ce857229 | 1408 | err = -EIDRM; |
6e224f94 | 1409 | goto out_unlock; |
ce857229 AV |
1410 | } |
1411 | sem_unlock(sma, -1); | |
6d49dab8 | 1412 | rcu_read_unlock(); |
f8dbe8d2 KC |
1413 | sem_io = kvmalloc_array(nsems, sizeof(ushort), |
1414 | GFP_KERNEL); | |
239521f3 | 1415 | if (sem_io == NULL) { |
dba4cdd3 | 1416 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1417 | return -ENOMEM; |
1418 | } | |
1419 | ||
4091fd94 | 1420 | rcu_read_lock(); |
6ff37972 | 1421 | sem_lock_and_putref(sma); |
0f3d2b01 | 1422 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1423 | err = -EIDRM; |
6e224f94 | 1424 | goto out_unlock; |
1da177e4 | 1425 | } |
ce857229 | 1426 | } |
1da177e4 | 1427 | for (i = 0; i < sma->sem_nsems; i++) |
1a233956 | 1428 | sem_io[i] = sma->sems[i].semval; |
6062a8dc | 1429 | sem_unlock(sma, -1); |
6d49dab8 | 1430 | rcu_read_unlock(); |
1da177e4 | 1431 | err = 0; |
239521f3 | 1432 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1433 | err = -EFAULT; |
1434 | goto out_free; | |
1435 | } | |
1436 | case SETALL: | |
1437 | { | |
1438 | int i; | |
1439 | struct sem_undo *un; | |
1440 | ||
dba4cdd3 | 1441 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
6e224f94 MS |
1442 | err = -EIDRM; |
1443 | goto out_rcu_wakeup; | |
6062a8dc | 1444 | } |
16df3674 | 1445 | rcu_read_unlock(); |
1da177e4 | 1446 | |
239521f3 | 1447 | if (nsems > SEMMSL_FAST) { |
f8dbe8d2 KC |
1448 | sem_io = kvmalloc_array(nsems, sizeof(ushort), |
1449 | GFP_KERNEL); | |
239521f3 | 1450 | if (sem_io == NULL) { |
dba4cdd3 | 1451 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1452 | return -ENOMEM; |
1453 | } | |
1454 | } | |
1455 | ||
239521f3 | 1456 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
dba4cdd3 | 1457 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1458 | err = -EFAULT; |
1459 | goto out_free; | |
1460 | } | |
1461 | ||
1462 | for (i = 0; i < nsems; i++) { | |
1463 | if (sem_io[i] > SEMVMX) { | |
dba4cdd3 | 1464 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1465 | err = -ERANGE; |
1466 | goto out_free; | |
1467 | } | |
1468 | } | |
4091fd94 | 1469 | rcu_read_lock(); |
6ff37972 | 1470 | sem_lock_and_putref(sma); |
0f3d2b01 | 1471 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1472 | err = -EIDRM; |
6e224f94 | 1473 | goto out_unlock; |
1da177e4 LT |
1474 | } |
1475 | ||
a5f4db87 | 1476 | for (i = 0; i < nsems; i++) { |
1a233956 | 1477 | sma->sems[i].semval = sem_io[i]; |
51d6f263 | 1478 | ipc_update_pid(&sma->sems[i].sempid, task_tgid(current)); |
a5f4db87 | 1479 | } |
4daa28f6 | 1480 | |
cf9d5d78 | 1481 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1482 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1483 | for (i = 0; i < nsems; i++) |
1484 | un->semadj[i] = 0; | |
4daa28f6 | 1485 | } |
e54d02b2 | 1486 | sma->sem_ctime = ktime_get_real_seconds(); |
1da177e4 | 1487 | /* maybe some queued-up processes were waiting for this */ |
9ae949fa | 1488 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
1da177e4 LT |
1489 | err = 0; |
1490 | goto out_unlock; | |
1491 | } | |
e1fd1f49 | 1492 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1493 | } |
1494 | err = -EINVAL; | |
c728b9c8 LT |
1495 | if (semnum < 0 || semnum >= nsems) |
1496 | goto out_rcu_wakeup; | |
1da177e4 | 1497 | |
6062a8dc | 1498 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1499 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1500 | err = -EIDRM; |
1501 | goto out_unlock; | |
1502 | } | |
1a233956 | 1503 | curr = &sma->sems[semnum]; |
1da177e4 LT |
1504 | |
1505 | switch (cmd) { | |
1506 | case GETVAL: | |
1507 | err = curr->semval; | |
1508 | goto out_unlock; | |
1509 | case GETPID: | |
51d6f263 | 1510 | err = pid_vnr(curr->sempid); |
1da177e4 LT |
1511 | goto out_unlock; |
1512 | case GETNCNT: | |
2f2ed41d | 1513 | err = count_semcnt(sma, semnum, 0); |
1da177e4 LT |
1514 | goto out_unlock; |
1515 | case GETZCNT: | |
2f2ed41d | 1516 | err = count_semcnt(sma, semnum, 1); |
1da177e4 | 1517 | goto out_unlock; |
1da177e4 | 1518 | } |
16df3674 | 1519 | |
1da177e4 | 1520 | out_unlock: |
6062a8dc | 1521 | sem_unlock(sma, -1); |
c728b9c8 | 1522 | out_rcu_wakeup: |
6d49dab8 | 1523 | rcu_read_unlock(); |
9ae949fa | 1524 | wake_up_q(&wake_q); |
1da177e4 | 1525 | out_free: |
239521f3 | 1526 | if (sem_io != fast_sem_io) |
f8dbe8d2 | 1527 | kvfree(sem_io); |
1da177e4 LT |
1528 | return err; |
1529 | } | |
1530 | ||
016d7132 PP |
1531 | static inline unsigned long |
1532 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1533 | { |
239521f3 | 1534 | switch (version) { |
1da177e4 | 1535 | case IPC_64: |
016d7132 | 1536 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1537 | return -EFAULT; |
1da177e4 | 1538 | return 0; |
1da177e4 LT |
1539 | case IPC_OLD: |
1540 | { | |
1541 | struct semid_ds tbuf_old; | |
1542 | ||
239521f3 | 1543 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1544 | return -EFAULT; |
1545 | ||
016d7132 PP |
1546 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1547 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1548 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1549 | |
1550 | return 0; | |
1551 | } | |
1552 | default: | |
1553 | return -EINVAL; | |
1554 | } | |
1555 | } | |
1556 | ||
522bb2a2 | 1557 | /* |
d9a605e4 | 1558 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1559 | * to be held in write mode. |
d9a605e4 | 1560 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1561 | */ |
21a4826a | 1562 | static int semctl_down(struct ipc_namespace *ns, int semid, |
45a4a64a | 1563 | int cmd, struct semid64_ds *semid64) |
1da177e4 LT |
1564 | { |
1565 | struct sem_array *sma; | |
1566 | int err; | |
1da177e4 LT |
1567 | struct kern_ipc_perm *ipcp; |
1568 | ||
d9a605e4 | 1569 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1570 | rcu_read_lock(); |
1571 | ||
16df3674 | 1572 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
45a4a64a | 1573 | &semid64->sem_perm, 0); |
7b4cc5d8 DB |
1574 | if (IS_ERR(ipcp)) { |
1575 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1576 | goto out_unlock1; |
1577 | } | |
073115d6 | 1578 | |
a5f75e7f | 1579 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 | 1580 | |
aefad959 | 1581 | err = security_sem_semctl(&sma->sem_perm, cmd); |
7b4cc5d8 DB |
1582 | if (err) |
1583 | goto out_unlock1; | |
1da177e4 | 1584 | |
7b4cc5d8 | 1585 | switch (cmd) { |
1da177e4 | 1586 | case IPC_RMID: |
6062a8dc | 1587 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1588 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1589 | freeary(ns, ipcp); |
522bb2a2 | 1590 | goto out_up; |
1da177e4 | 1591 | case IPC_SET: |
6062a8dc | 1592 | sem_lock(sma, NULL, -1); |
45a4a64a | 1593 | err = ipc_update_perm(&semid64->sem_perm, ipcp); |
1efdb69b | 1594 | if (err) |
7b4cc5d8 | 1595 | goto out_unlock0; |
e54d02b2 | 1596 | sma->sem_ctime = ktime_get_real_seconds(); |
1da177e4 LT |
1597 | break; |
1598 | default: | |
1da177e4 | 1599 | err = -EINVAL; |
7b4cc5d8 | 1600 | goto out_unlock1; |
1da177e4 | 1601 | } |
1da177e4 | 1602 | |
7b4cc5d8 | 1603 | out_unlock0: |
6062a8dc | 1604 | sem_unlock(sma, -1); |
7b4cc5d8 | 1605 | out_unlock1: |
6d49dab8 | 1606 | rcu_read_unlock(); |
522bb2a2 | 1607 | out_up: |
d9a605e4 | 1608 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1609 | return err; |
1610 | } | |
1611 | ||
e1fd1f49 | 1612 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1613 | { |
1da177e4 | 1614 | int version; |
e3893534 | 1615 | struct ipc_namespace *ns; |
e1fd1f49 | 1616 | void __user *p = (void __user *)arg; |
45a4a64a AV |
1617 | struct semid64_ds semid64; |
1618 | int err; | |
1da177e4 LT |
1619 | |
1620 | if (semid < 0) | |
1621 | return -EINVAL; | |
1622 | ||
1623 | version = ipc_parse_version(&cmd); | |
e3893534 | 1624 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1625 | |
239521f3 | 1626 | switch (cmd) { |
1da177e4 LT |
1627 | case IPC_INFO: |
1628 | case SEM_INFO: | |
45a4a64a | 1629 | return semctl_info(ns, semid, cmd, p); |
4b9fcb0e | 1630 | case IPC_STAT: |
1da177e4 | 1631 | case SEM_STAT: |
45a4a64a AV |
1632 | err = semctl_stat(ns, semid, cmd, &semid64); |
1633 | if (err < 0) | |
1634 | return err; | |
1635 | if (copy_semid_to_user(p, &semid64, version)) | |
1636 | err = -EFAULT; | |
1637 | return err; | |
1da177e4 LT |
1638 | case GETALL: |
1639 | case GETVAL: | |
1640 | case GETPID: | |
1641 | case GETNCNT: | |
1642 | case GETZCNT: | |
1da177e4 | 1643 | case SETALL: |
e1fd1f49 | 1644 | return semctl_main(ns, semid, semnum, cmd, p); |
45a4a64a AV |
1645 | case SETVAL: { |
1646 | int val; | |
1647 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1648 | /* big-endian 64bit */ | |
1649 | val = arg >> 32; | |
1650 | #else | |
1651 | /* 32bit or little-endian 64bit */ | |
1652 | val = arg; | |
1653 | #endif | |
1654 | return semctl_setval(ns, semid, semnum, val); | |
1655 | } | |
1da177e4 | 1656 | case IPC_SET: |
45a4a64a AV |
1657 | if (copy_semid_from_user(&semid64, p, version)) |
1658 | return -EFAULT; | |
1659 | case IPC_RMID: | |
1660 | return semctl_down(ns, semid, cmd, &semid64); | |
1da177e4 LT |
1661 | default: |
1662 | return -EINVAL; | |
1663 | } | |
1664 | } | |
1665 | ||
c0ebccb6 AV |
1666 | #ifdef CONFIG_COMPAT |
1667 | ||
1668 | struct compat_semid_ds { | |
1669 | struct compat_ipc_perm sem_perm; | |
1670 | compat_time_t sem_otime; | |
1671 | compat_time_t sem_ctime; | |
1672 | compat_uptr_t sem_base; | |
1673 | compat_uptr_t sem_pending; | |
1674 | compat_uptr_t sem_pending_last; | |
1675 | compat_uptr_t undo; | |
1676 | unsigned short sem_nsems; | |
1677 | }; | |
1678 | ||
1679 | static int copy_compat_semid_from_user(struct semid64_ds *out, void __user *buf, | |
1680 | int version) | |
1681 | { | |
1682 | memset(out, 0, sizeof(*out)); | |
1683 | if (version == IPC_64) { | |
6aa211e8 | 1684 | struct compat_semid64_ds __user *p = buf; |
c0ebccb6 AV |
1685 | return get_compat_ipc64_perm(&out->sem_perm, &p->sem_perm); |
1686 | } else { | |
6aa211e8 | 1687 | struct compat_semid_ds __user *p = buf; |
c0ebccb6 AV |
1688 | return get_compat_ipc_perm(&out->sem_perm, &p->sem_perm); |
1689 | } | |
1690 | } | |
1691 | ||
1692 | static int copy_compat_semid_to_user(void __user *buf, struct semid64_ds *in, | |
1693 | int version) | |
1694 | { | |
1695 | if (version == IPC_64) { | |
1696 | struct compat_semid64_ds v; | |
1697 | memset(&v, 0, sizeof(v)); | |
1698 | to_compat_ipc64_perm(&v.sem_perm, &in->sem_perm); | |
1699 | v.sem_otime = in->sem_otime; | |
1700 | v.sem_ctime = in->sem_ctime; | |
1701 | v.sem_nsems = in->sem_nsems; | |
1702 | return copy_to_user(buf, &v, sizeof(v)); | |
1703 | } else { | |
1704 | struct compat_semid_ds v; | |
1705 | memset(&v, 0, sizeof(v)); | |
1706 | to_compat_ipc_perm(&v.sem_perm, &in->sem_perm); | |
1707 | v.sem_otime = in->sem_otime; | |
1708 | v.sem_ctime = in->sem_ctime; | |
1709 | v.sem_nsems = in->sem_nsems; | |
1710 | return copy_to_user(buf, &v, sizeof(v)); | |
1711 | } | |
1712 | } | |
1713 | ||
1714 | COMPAT_SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, int, arg) | |
1715 | { | |
1716 | void __user *p = compat_ptr(arg); | |
1717 | struct ipc_namespace *ns; | |
1718 | struct semid64_ds semid64; | |
1719 | int version = compat_ipc_parse_version(&cmd); | |
1720 | int err; | |
1721 | ||
1722 | ns = current->nsproxy->ipc_ns; | |
1723 | ||
1724 | if (semid < 0) | |
1725 | return -EINVAL; | |
1726 | ||
1727 | switch (cmd & (~IPC_64)) { | |
1728 | case IPC_INFO: | |
1729 | case SEM_INFO: | |
1730 | return semctl_info(ns, semid, cmd, p); | |
1731 | case IPC_STAT: | |
1732 | case SEM_STAT: | |
1733 | err = semctl_stat(ns, semid, cmd, &semid64); | |
1734 | if (err < 0) | |
1735 | return err; | |
1736 | if (copy_compat_semid_to_user(p, &semid64, version)) | |
1737 | err = -EFAULT; | |
1738 | return err; | |
1739 | case GETVAL: | |
1740 | case GETPID: | |
1741 | case GETNCNT: | |
1742 | case GETZCNT: | |
1743 | case GETALL: | |
1da177e4 | 1744 | case SETALL: |
e1fd1f49 AV |
1745 | return semctl_main(ns, semid, semnum, cmd, p); |
1746 | case SETVAL: | |
1747 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 | 1748 | case IPC_SET: |
c0ebccb6 AV |
1749 | if (copy_compat_semid_from_user(&semid64, p, version)) |
1750 | return -EFAULT; | |
1751 | /* fallthru */ | |
1752 | case IPC_RMID: | |
1753 | return semctl_down(ns, semid, cmd, &semid64); | |
1da177e4 LT |
1754 | default: |
1755 | return -EINVAL; | |
1756 | } | |
1757 | } | |
c0ebccb6 | 1758 | #endif |
1da177e4 | 1759 | |
1da177e4 LT |
1760 | /* If the task doesn't already have a undo_list, then allocate one |
1761 | * here. We guarantee there is only one thread using this undo list, | |
1762 | * and current is THE ONE | |
1763 | * | |
1764 | * If this allocation and assignment succeeds, but later | |
1765 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1766 | * Just let it stay associated with the task, and it'll be freed later | |
1767 | * at exit time. | |
1768 | * | |
1769 | * This can block, so callers must hold no locks. | |
1770 | */ | |
1771 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1772 | { | |
1773 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1774 | |
1775 | undo_list = current->sysvsem.undo_list; | |
1776 | if (!undo_list) { | |
2453a306 | 1777 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1778 | if (undo_list == NULL) |
1779 | return -ENOMEM; | |
00a5dfdb | 1780 | spin_lock_init(&undo_list->lock); |
f74370b8 | 1781 | refcount_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1782 | INIT_LIST_HEAD(&undo_list->list_proc); |
1783 | ||
1da177e4 LT |
1784 | current->sysvsem.undo_list = undo_list; |
1785 | } | |
1786 | *undo_listp = undo_list; | |
1787 | return 0; | |
1788 | } | |
1789 | ||
bf17bb71 | 1790 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1791 | { |
bf17bb71 | 1792 | struct sem_undo *un; |
4daa28f6 | 1793 | |
bf17bb71 NP |
1794 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1795 | if (un->semid == semid) | |
1796 | return un; | |
1da177e4 | 1797 | } |
4daa28f6 | 1798 | return NULL; |
1da177e4 LT |
1799 | } |
1800 | ||
bf17bb71 NP |
1801 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1802 | { | |
1803 | struct sem_undo *un; | |
1804 | ||
239521f3 | 1805 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1806 | |
1807 | un = __lookup_undo(ulp, semid); | |
1808 | if (un) { | |
1809 | list_del_rcu(&un->list_proc); | |
1810 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1811 | } | |
1812 | return un; | |
1813 | } | |
1814 | ||
4daa28f6 | 1815 | /** |
8001c858 | 1816 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1817 | * @ns: namespace |
1818 | * @semid: semaphore array id | |
1819 | * | |
1820 | * The function looks up (and if not present creates) the undo structure. | |
1821 | * The size of the undo structure depends on the size of the semaphore | |
1822 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1823 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1824 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1825 | */ |
1826 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1827 | { |
1828 | struct sem_array *sma; | |
1829 | struct sem_undo_list *ulp; | |
1830 | struct sem_undo *un, *new; | |
6062a8dc | 1831 | int nsems, error; |
1da177e4 LT |
1832 | |
1833 | error = get_undo_list(&ulp); | |
1834 | if (error) | |
1835 | return ERR_PTR(error); | |
1836 | ||
380af1b3 | 1837 | rcu_read_lock(); |
c530c6ac | 1838 | spin_lock(&ulp->lock); |
1da177e4 | 1839 | un = lookup_undo(ulp, semid); |
c530c6ac | 1840 | spin_unlock(&ulp->lock); |
239521f3 | 1841 | if (likely(un != NULL)) |
1da177e4 LT |
1842 | goto out; |
1843 | ||
1844 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1845 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1846 | sma = sem_obtain_object_check(ns, semid); |
1847 | if (IS_ERR(sma)) { | |
1848 | rcu_read_unlock(); | |
4de85cd6 | 1849 | return ERR_CAST(sma); |
16df3674 | 1850 | } |
023a5355 | 1851 | |
1da177e4 | 1852 | nsems = sma->sem_nsems; |
dba4cdd3 | 1853 | if (!ipc_rcu_getref(&sma->sem_perm)) { |
6062a8dc RR |
1854 | rcu_read_unlock(); |
1855 | un = ERR_PTR(-EIDRM); | |
1856 | goto out; | |
1857 | } | |
16df3674 | 1858 | rcu_read_unlock(); |
1da177e4 | 1859 | |
4daa28f6 | 1860 | /* step 2: allocate new undo structure */ |
4668edc3 | 1861 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1862 | if (!new) { |
dba4cdd3 | 1863 | ipc_rcu_putref(&sma->sem_perm, sem_rcu_free); |
1da177e4 LT |
1864 | return ERR_PTR(-ENOMEM); |
1865 | } | |
1da177e4 | 1866 | |
380af1b3 | 1867 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1868 | rcu_read_lock(); |
6ff37972 | 1869 | sem_lock_and_putref(sma); |
0f3d2b01 | 1870 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1871 | sem_unlock(sma, -1); |
6d49dab8 | 1872 | rcu_read_unlock(); |
1da177e4 LT |
1873 | kfree(new); |
1874 | un = ERR_PTR(-EIDRM); | |
1875 | goto out; | |
1876 | } | |
380af1b3 MS |
1877 | spin_lock(&ulp->lock); |
1878 | ||
1879 | /* | |
1880 | * step 4: check for races: did someone else allocate the undo struct? | |
1881 | */ | |
1882 | un = lookup_undo(ulp, semid); | |
1883 | if (un) { | |
1884 | kfree(new); | |
1885 | goto success; | |
1886 | } | |
4daa28f6 MS |
1887 | /* step 5: initialize & link new undo structure */ |
1888 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1889 | new->ulp = ulp; |
4daa28f6 MS |
1890 | new->semid = semid; |
1891 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1892 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1893 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1894 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1895 | un = new; |
4daa28f6 | 1896 | |
380af1b3 | 1897 | success: |
c530c6ac | 1898 | spin_unlock(&ulp->lock); |
6062a8dc | 1899 | sem_unlock(sma, -1); |
1da177e4 LT |
1900 | out: |
1901 | return un; | |
1902 | } | |
1903 | ||
44ee4546 | 1904 | static long do_semtimedop(int semid, struct sembuf __user *tsops, |
3ef56dc2 | 1905 | unsigned nsops, const struct timespec64 *timeout) |
1da177e4 LT |
1906 | { |
1907 | int error = -EINVAL; | |
1908 | struct sem_array *sma; | |
1909 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1910 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1911 | struct sem_undo *un; |
4ce33ec2 DB |
1912 | int max, locknum; |
1913 | bool undos = false, alter = false, dupsop = false; | |
1da177e4 | 1914 | struct sem_queue queue; |
4ce33ec2 | 1915 | unsigned long dup = 0, jiffies_left = 0; |
e3893534 KK |
1916 | struct ipc_namespace *ns; |
1917 | ||
1918 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1919 | |
1920 | if (nsops < 1 || semid < 0) | |
1921 | return -EINVAL; | |
e3893534 | 1922 | if (nsops > ns->sc_semopm) |
1da177e4 | 1923 | return -E2BIG; |
239521f3 | 1924 | if (nsops > SEMOPM_FAST) { |
e4243b80 | 1925 | sops = kvmalloc(sizeof(*sops)*nsops, GFP_KERNEL); |
239521f3 | 1926 | if (sops == NULL) |
1da177e4 LT |
1927 | return -ENOMEM; |
1928 | } | |
4ce33ec2 | 1929 | |
239521f3 MS |
1930 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1931 | error = -EFAULT; | |
1da177e4 LT |
1932 | goto out_free; |
1933 | } | |
4ce33ec2 | 1934 | |
1da177e4 | 1935 | if (timeout) { |
44ee4546 AV |
1936 | if (timeout->tv_sec < 0 || timeout->tv_nsec < 0 || |
1937 | timeout->tv_nsec >= 1000000000L) { | |
1da177e4 LT |
1938 | error = -EINVAL; |
1939 | goto out_free; | |
1940 | } | |
3ef56dc2 | 1941 | jiffies_left = timespec64_to_jiffies(timeout); |
1da177e4 | 1942 | } |
4ce33ec2 | 1943 | |
1da177e4 LT |
1944 | max = 0; |
1945 | for (sop = sops; sop < sops + nsops; sop++) { | |
4ce33ec2 DB |
1946 | unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG); |
1947 | ||
1da177e4 LT |
1948 | if (sop->sem_num >= max) |
1949 | max = sop->sem_num; | |
1950 | if (sop->sem_flg & SEM_UNDO) | |
4ce33ec2 DB |
1951 | undos = true; |
1952 | if (dup & mask) { | |
1953 | /* | |
1954 | * There was a previous alter access that appears | |
1955 | * to have accessed the same semaphore, thus use | |
1956 | * the dupsop logic. "appears", because the detection | |
1957 | * can only check % BITS_PER_LONG. | |
1958 | */ | |
1959 | dupsop = true; | |
1960 | } | |
1961 | if (sop->sem_op != 0) { | |
1962 | alter = true; | |
1963 | dup |= mask; | |
1964 | } | |
1da177e4 | 1965 | } |
1da177e4 | 1966 | |
1da177e4 | 1967 | if (undos) { |
6062a8dc | 1968 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1969 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1970 | if (IS_ERR(un)) { |
1971 | error = PTR_ERR(un); | |
1972 | goto out_free; | |
1973 | } | |
6062a8dc | 1974 | } else { |
1da177e4 | 1975 | un = NULL; |
6062a8dc RR |
1976 | rcu_read_lock(); |
1977 | } | |
1da177e4 | 1978 | |
16df3674 | 1979 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1980 | if (IS_ERR(sma)) { |
6062a8dc | 1981 | rcu_read_unlock(); |
023a5355 | 1982 | error = PTR_ERR(sma); |
1da177e4 | 1983 | goto out_free; |
023a5355 ND |
1984 | } |
1985 | ||
16df3674 | 1986 | error = -EFBIG; |
248e7357 DB |
1987 | if (max >= sma->sem_nsems) { |
1988 | rcu_read_unlock(); | |
1989 | goto out_free; | |
1990 | } | |
16df3674 DB |
1991 | |
1992 | error = -EACCES; | |
248e7357 DB |
1993 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) { |
1994 | rcu_read_unlock(); | |
1995 | goto out_free; | |
1996 | } | |
16df3674 | 1997 | |
aefad959 | 1998 | error = security_sem_semop(&sma->sem_perm, sops, nsops, alter); |
248e7357 DB |
1999 | if (error) { |
2000 | rcu_read_unlock(); | |
2001 | goto out_free; | |
2002 | } | |
16df3674 | 2003 | |
6e224f94 MS |
2004 | error = -EIDRM; |
2005 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
2006 | /* |
2007 | * We eventually might perform the following check in a lockless | |
2008 | * fashion, considering ipc_valid_object() locking constraints. | |
2009 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
2010 | * only a per-semaphore lock is held and it's OK to proceed with the | |
2011 | * check below. More details on the fine grained locking scheme | |
2012 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
2013 | */ | |
2014 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 2015 | goto out_unlock_free; |
1da177e4 | 2016 | /* |
4daa28f6 | 2017 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 2018 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 2019 | * and now a new array with received the same id. Check and fail. |
25985edc | 2020 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 2021 | * "un" itself is guaranteed by rcu. |
1da177e4 | 2022 | */ |
6062a8dc RR |
2023 | if (un && un->semid == -1) |
2024 | goto out_unlock_free; | |
4daa28f6 | 2025 | |
d198cd6d MS |
2026 | queue.sops = sops; |
2027 | queue.nsops = nsops; | |
2028 | queue.undo = un; | |
51d6f263 | 2029 | queue.pid = task_tgid(current); |
d198cd6d | 2030 | queue.alter = alter; |
4ce33ec2 | 2031 | queue.dupsop = dupsop; |
d198cd6d MS |
2032 | |
2033 | error = perform_atomic_semop(sma, &queue); | |
9ae949fa DB |
2034 | if (error == 0) { /* non-blocking succesfull path */ |
2035 | DEFINE_WAKE_Q(wake_q); | |
2036 | ||
2037 | /* | |
2038 | * If the operation was successful, then do | |
0e8c6656 MS |
2039 | * the required updates. |
2040 | */ | |
2041 | if (alter) | |
9ae949fa | 2042 | do_smart_update(sma, sops, nsops, 1, &wake_q); |
0e8c6656 MS |
2043 | else |
2044 | set_semotime(sma, sops); | |
9ae949fa DB |
2045 | |
2046 | sem_unlock(sma, locknum); | |
2047 | rcu_read_unlock(); | |
2048 | wake_up_q(&wake_q); | |
2049 | ||
2050 | goto out_free; | |
1da177e4 | 2051 | } |
9ae949fa | 2052 | if (error < 0) /* non-blocking error path */ |
0e8c6656 | 2053 | goto out_unlock_free; |
1da177e4 | 2054 | |
9ae949fa DB |
2055 | /* |
2056 | * We need to sleep on this operation, so we put the current | |
1da177e4 LT |
2057 | * task into the pending queue and go to sleep. |
2058 | */ | |
b97e820f MS |
2059 | if (nsops == 1) { |
2060 | struct sem *curr; | |
1a233956 | 2061 | curr = &sma->sems[sops->sem_num]; |
b97e820f | 2062 | |
f269f40a MS |
2063 | if (alter) { |
2064 | if (sma->complex_count) { | |
2065 | list_add_tail(&queue.list, | |
2066 | &sma->pending_alter); | |
2067 | } else { | |
2068 | ||
2069 | list_add_tail(&queue.list, | |
2070 | &curr->pending_alter); | |
2071 | } | |
2072 | } else { | |
1a82e9e1 | 2073 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 2074 | } |
b97e820f | 2075 | } else { |
f269f40a MS |
2076 | if (!sma->complex_count) |
2077 | merge_queues(sma); | |
2078 | ||
9f1bc2c9 | 2079 | if (alter) |
1a82e9e1 | 2080 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 2081 | else |
1a82e9e1 MS |
2082 | list_add_tail(&queue.list, &sma->pending_const); |
2083 | ||
b97e820f MS |
2084 | sma->complex_count++; |
2085 | } | |
2086 | ||
b5fa01a2 DB |
2087 | do { |
2088 | queue.status = -EINTR; | |
2089 | queue.sleeper = current; | |
0b0577f6 | 2090 | |
b5fa01a2 DB |
2091 | __set_current_state(TASK_INTERRUPTIBLE); |
2092 | sem_unlock(sma, locknum); | |
2093 | rcu_read_unlock(); | |
1da177e4 | 2094 | |
b5fa01a2 DB |
2095 | if (timeout) |
2096 | jiffies_left = schedule_timeout(jiffies_left); | |
2097 | else | |
2098 | schedule(); | |
1da177e4 | 2099 | |
9ae949fa | 2100 | /* |
b5fa01a2 DB |
2101 | * fastpath: the semop has completed, either successfully or |
2102 | * not, from the syscall pov, is quite irrelevant to us at this | |
2103 | * point; we're done. | |
2104 | * | |
2105 | * We _do_ care, nonetheless, about being awoken by a signal or | |
2106 | * spuriously. The queue.status is checked again in the | |
2107 | * slowpath (aka after taking sem_lock), such that we can detect | |
2108 | * scenarios where we were awakened externally, during the | |
2109 | * window between wake_q_add() and wake_up_q(). | |
c61284e9 | 2110 | */ |
b5fa01a2 DB |
2111 | error = READ_ONCE(queue.status); |
2112 | if (error != -EINTR) { | |
2113 | /* | |
2114 | * User space could assume that semop() is a memory | |
2115 | * barrier: Without the mb(), the cpu could | |
2116 | * speculatively read in userspace stale data that was | |
2117 | * overwritten by the previous owner of the semaphore. | |
2118 | */ | |
2119 | smp_mb(); | |
2120 | goto out_free; | |
2121 | } | |
d694ad62 | 2122 | |
b5fa01a2 | 2123 | rcu_read_lock(); |
c626bc46 | 2124 | locknum = sem_lock(sma, sops, nsops); |
1da177e4 | 2125 | |
370b262c DB |
2126 | if (!ipc_valid_object(&sma->sem_perm)) |
2127 | goto out_unlock_free; | |
2128 | ||
2129 | error = READ_ONCE(queue.status); | |
1da177e4 | 2130 | |
b5fa01a2 DB |
2131 | /* |
2132 | * If queue.status != -EINTR we are woken up by another process. | |
2133 | * Leave without unlink_queue(), but with sem_unlock(). | |
2134 | */ | |
2135 | if (error != -EINTR) | |
2136 | goto out_unlock_free; | |
0b0577f6 | 2137 | |
b5fa01a2 DB |
2138 | /* |
2139 | * If an interrupt occurred we have to clean up the queue. | |
2140 | */ | |
2141 | if (timeout && jiffies_left == 0) | |
2142 | error = -EAGAIN; | |
2143 | } while (error == -EINTR && !signal_pending(current)); /* spurious */ | |
0b0577f6 | 2144 | |
b97e820f | 2145 | unlink_queue(sma, &queue); |
1da177e4 LT |
2146 | |
2147 | out_unlock_free: | |
6062a8dc | 2148 | sem_unlock(sma, locknum); |
6d49dab8 | 2149 | rcu_read_unlock(); |
1da177e4 | 2150 | out_free: |
239521f3 | 2151 | if (sops != fast_sops) |
e4243b80 | 2152 | kvfree(sops); |
1da177e4 LT |
2153 | return error; |
2154 | } | |
2155 | ||
44ee4546 AV |
2156 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
2157 | unsigned, nsops, const struct timespec __user *, timeout) | |
2158 | { | |
2159 | if (timeout) { | |
3ef56dc2 DD |
2160 | struct timespec64 ts; |
2161 | if (get_timespec64(&ts, timeout)) | |
44ee4546 AV |
2162 | return -EFAULT; |
2163 | return do_semtimedop(semid, tsops, nsops, &ts); | |
2164 | } | |
2165 | return do_semtimedop(semid, tsops, nsops, NULL); | |
2166 | } | |
2167 | ||
2168 | #ifdef CONFIG_COMPAT | |
2169 | COMPAT_SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsems, | |
2170 | unsigned, nsops, | |
2171 | const struct compat_timespec __user *, timeout) | |
2172 | { | |
2173 | if (timeout) { | |
3ef56dc2 DD |
2174 | struct timespec64 ts; |
2175 | if (compat_get_timespec64(&ts, timeout)) | |
44ee4546 AV |
2176 | return -EFAULT; |
2177 | return do_semtimedop(semid, tsems, nsops, &ts); | |
2178 | } | |
2179 | return do_semtimedop(semid, tsems, nsops, NULL); | |
2180 | } | |
2181 | #endif | |
2182 | ||
d5460c99 HC |
2183 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2184 | unsigned, nsops) | |
1da177e4 | 2185 | { |
44ee4546 | 2186 | return do_semtimedop(semid, tsops, nsops, NULL); |
1da177e4 LT |
2187 | } |
2188 | ||
2189 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2190 | * parent and child tasks. | |
1da177e4 LT |
2191 | */ |
2192 | ||
2193 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2194 | { | |
2195 | struct sem_undo_list *undo_list; | |
2196 | int error; | |
2197 | ||
2198 | if (clone_flags & CLONE_SYSVSEM) { | |
2199 | error = get_undo_list(&undo_list); | |
2200 | if (error) | |
2201 | return error; | |
f74370b8 | 2202 | refcount_inc(&undo_list->refcnt); |
1da177e4 | 2203 | tsk->sysvsem.undo_list = undo_list; |
46c0a8ca | 2204 | } else |
1da177e4 LT |
2205 | tsk->sysvsem.undo_list = NULL; |
2206 | ||
2207 | return 0; | |
2208 | } | |
2209 | ||
2210 | /* | |
2211 | * add semadj values to semaphores, free undo structures. | |
2212 | * undo structures are not freed when semaphore arrays are destroyed | |
2213 | * so some of them may be out of date. | |
2214 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2215 | * set of adjustments that needs to be done should be done in an atomic | |
2216 | * manner or not. That is, if we are attempting to decrement the semval | |
2217 | * should we queue up and wait until we can do so legally? | |
2218 | * The original implementation attempted to do this (queue and wait). | |
2219 | * The current implementation does not do so. The POSIX standard | |
2220 | * and SVID should be consulted to determine what behavior is mandated. | |
2221 | */ | |
2222 | void exit_sem(struct task_struct *tsk) | |
2223 | { | |
4daa28f6 | 2224 | struct sem_undo_list *ulp; |
1da177e4 | 2225 | |
4daa28f6 MS |
2226 | ulp = tsk->sysvsem.undo_list; |
2227 | if (!ulp) | |
1da177e4 | 2228 | return; |
9edff4ab | 2229 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2230 | |
f74370b8 | 2231 | if (!refcount_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2232 | return; |
2233 | ||
380af1b3 | 2234 | for (;;) { |
1da177e4 | 2235 | struct sem_array *sma; |
380af1b3 | 2236 | struct sem_undo *un; |
6062a8dc | 2237 | int semid, i; |
9ae949fa | 2238 | DEFINE_WAKE_Q(wake_q); |
4daa28f6 | 2239 | |
2a1613a5 NB |
2240 | cond_resched(); |
2241 | ||
380af1b3 | 2242 | rcu_read_lock(); |
05725f7e JP |
2243 | un = list_entry_rcu(ulp->list_proc.next, |
2244 | struct sem_undo, list_proc); | |
602b8593 HK |
2245 | if (&un->list_proc == &ulp->list_proc) { |
2246 | /* | |
2247 | * We must wait for freeary() before freeing this ulp, | |
2248 | * in case we raced with last sem_undo. There is a small | |
2249 | * possibility where we exit while freeary() didn't | |
2250 | * finish unlocking sem_undo_list. | |
2251 | */ | |
e0892e08 PM |
2252 | spin_lock(&ulp->lock); |
2253 | spin_unlock(&ulp->lock); | |
602b8593 HK |
2254 | rcu_read_unlock(); |
2255 | break; | |
2256 | } | |
2257 | spin_lock(&ulp->lock); | |
2258 | semid = un->semid; | |
2259 | spin_unlock(&ulp->lock); | |
4daa28f6 | 2260 | |
602b8593 | 2261 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2262 | if (semid == -1) { |
2263 | rcu_read_unlock(); | |
602b8593 | 2264 | continue; |
6062a8dc | 2265 | } |
1da177e4 | 2266 | |
602b8593 | 2267 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); |
380af1b3 | 2268 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2269 | if (IS_ERR(sma)) { |
2270 | rcu_read_unlock(); | |
380af1b3 | 2271 | continue; |
6062a8dc | 2272 | } |
1da177e4 | 2273 | |
6062a8dc | 2274 | sem_lock(sma, NULL, -1); |
6e224f94 | 2275 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2276 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2277 | sem_unlock(sma, -1); |
2278 | rcu_read_unlock(); | |
2279 | continue; | |
2280 | } | |
bf17bb71 | 2281 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2282 | if (un == NULL) { |
2283 | /* exit_sem raced with IPC_RMID+semget() that created | |
2284 | * exactly the same semid. Nothing to do. | |
2285 | */ | |
6062a8dc | 2286 | sem_unlock(sma, -1); |
6d49dab8 | 2287 | rcu_read_unlock(); |
380af1b3 MS |
2288 | continue; |
2289 | } | |
2290 | ||
2291 | /* remove un from the linked lists */ | |
cf9d5d78 | 2292 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2293 | list_del(&un->list_id); |
2294 | ||
a9795584 HK |
2295 | /* we are the last process using this ulp, acquiring ulp->lock |
2296 | * isn't required. Besides that, we are also protected against | |
2297 | * IPC_RMID as we hold sma->sem_perm lock now | |
2298 | */ | |
380af1b3 | 2299 | list_del_rcu(&un->list_proc); |
380af1b3 | 2300 | |
4daa28f6 MS |
2301 | /* perform adjustments registered in un */ |
2302 | for (i = 0; i < sma->sem_nsems; i++) { | |
1a233956 | 2303 | struct sem *semaphore = &sma->sems[i]; |
4daa28f6 MS |
2304 | if (un->semadj[i]) { |
2305 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2306 | /* |
2307 | * Range checks of the new semaphore value, | |
2308 | * not defined by sus: | |
2309 | * - Some unices ignore the undo entirely | |
2310 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2311 | * - some cap the value (e.g. FreeBSD caps | |
2312 | * at 0, but doesn't enforce SEMVMX) | |
2313 | * | |
2314 | * Linux caps the semaphore value, both at 0 | |
2315 | * and at SEMVMX. | |
2316 | * | |
239521f3 | 2317 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2318 | */ |
5f921ae9 IM |
2319 | if (semaphore->semval < 0) |
2320 | semaphore->semval = 0; | |
2321 | if (semaphore->semval > SEMVMX) | |
2322 | semaphore->semval = SEMVMX; | |
51d6f263 | 2323 | ipc_update_pid(&semaphore->sempid, task_tgid(current)); |
1da177e4 LT |
2324 | } |
2325 | } | |
1da177e4 | 2326 | /* maybe some queued-up processes were waiting for this */ |
9ae949fa | 2327 | do_smart_update(sma, NULL, 0, 1, &wake_q); |
6062a8dc | 2328 | sem_unlock(sma, -1); |
6d49dab8 | 2329 | rcu_read_unlock(); |
9ae949fa | 2330 | wake_up_q(&wake_q); |
380af1b3 | 2331 | |
693a8b6e | 2332 | kfree_rcu(un, rcu); |
1da177e4 | 2333 | } |
4daa28f6 | 2334 | kfree(ulp); |
1da177e4 LT |
2335 | } |
2336 | ||
2337 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2338 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2339 | { |
1efdb69b | 2340 | struct user_namespace *user_ns = seq_user_ns(s); |
ade9f91b KC |
2341 | struct kern_ipc_perm *ipcp = it; |
2342 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); | |
e54d02b2 | 2343 | time64_t sem_otime; |
d12e1e50 | 2344 | |
d8c63376 MS |
2345 | /* |
2346 | * The proc interface isn't aware of sem_lock(), it calls | |
2347 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
5864a2fd MS |
2348 | * In order to stay compatible with sem_lock(), we must |
2349 | * enter / leave complex_mode. | |
d8c63376 | 2350 | */ |
5864a2fd | 2351 | complexmode_enter(sma); |
d8c63376 | 2352 | |
d12e1e50 | 2353 | sem_otime = get_semotime(sma); |
19b4946c | 2354 | |
7f032d6e | 2355 | seq_printf(s, |
e54d02b2 | 2356 | "%10d %10d %4o %10u %5u %5u %5u %5u %10llu %10llu\n", |
7f032d6e JP |
2357 | sma->sem_perm.key, |
2358 | sma->sem_perm.id, | |
2359 | sma->sem_perm.mode, | |
2360 | sma->sem_nsems, | |
2361 | from_kuid_munged(user_ns, sma->sem_perm.uid), | |
2362 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2363 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2364 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
2365 | sem_otime, | |
2366 | sma->sem_ctime); | |
2367 | ||
5864a2fd MS |
2368 | complexmode_tryleave(sma); |
2369 | ||
7f032d6e | 2370 | return 0; |
1da177e4 LT |
2371 | } |
2372 | #endif |