]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/ipc/sem.c | |
3 | * Copyright (C) 1992 Krishna Balasubramanian | |
4 | * Copyright (C) 1995 Eric Schenk, Bruno Haible | |
5 | * | |
1da177e4 LT |
6 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
7 | * | |
8 | * SMP-threaded, sysctl's added | |
624dffcb | 9 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 10 | * Enforced range limit on SEM_UNDO |
046c6884 | 11 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
12 | * Lockless wakeup |
13 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
9ae949fa | 14 | * (c) 2016 Davidlohr Bueso <dave@stgolabs.net> |
c5cf6359 MS |
15 | * Further wakeup optimizations, documentation |
16 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
17 | * |
18 | * support for audit of ipc object properties and permission changes | |
19 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
20 | * |
21 | * namespaces support | |
22 | * OpenVZ, SWsoft Inc. | |
23 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
24 | * |
25 | * Implementation notes: (May 2010) | |
26 | * This file implements System V semaphores. | |
27 | * | |
28 | * User space visible behavior: | |
29 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
30 | * protection) | |
31 | * - multiple semaphore operations that alter the same semaphore in | |
32 | * one semop() are handled. | |
33 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
34 | * SETALL calls. | |
35 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
36 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
37 | * - namespace are supported. | |
38 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
39 | * to /proc/sys/kernel/sem. | |
40 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
41 | * | |
42 | * Internals: | |
43 | * - scalability: | |
44 | * - all global variables are read-mostly. | |
45 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
46 | * - most operations do write operations (actually: spin_lock calls) to | |
47 | * the per-semaphore array structure. | |
48 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
49 | * If multiple semaphores in one array are used, then cache line | |
50 | * trashing on the semaphore array spinlock will limit the scaling. | |
2f2ed41d | 51 | * - semncnt and semzcnt are calculated on demand in count_semcnt() |
c5cf6359 MS |
52 | * - the task that performs a successful semop() scans the list of all |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
9ae949fa | 57 | * dropping all locks. (see wake_up_sem_queue_prepare()) |
c5cf6359 MS |
58 | * - All work is done by the waker, the woken up task does not have to do |
59 | * anything - not even acquiring a lock or dropping a refcount. | |
60 | * - A woken up task may not even touch the semaphore array anymore, it may | |
61 | * have been destroyed already by a semctl(RMID). | |
c5cf6359 MS |
62 | * - UNDO values are stored in an array (one per process and per |
63 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
64 | * modes for the UNDO variables are supported (per process, per thread) | |
65 | * (see copy_semundo, CLONE_SYSVSEM) | |
66 | * - There are two lists of the pending operations: a per-array list | |
67 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
68 | * ordering without always scanning all pending operations. | |
69 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
70 | */ |
71 | ||
1da177e4 LT |
72 | #include <linux/slab.h> |
73 | #include <linux/spinlock.h> | |
74 | #include <linux/init.h> | |
75 | #include <linux/proc_fs.h> | |
76 | #include <linux/time.h> | |
1da177e4 LT |
77 | #include <linux/security.h> |
78 | #include <linux/syscalls.h> | |
79 | #include <linux/audit.h> | |
c59ede7b | 80 | #include <linux/capability.h> |
19b4946c | 81 | #include <linux/seq_file.h> |
3e148c79 | 82 | #include <linux/rwsem.h> |
e3893534 | 83 | #include <linux/nsproxy.h> |
ae5e1b22 | 84 | #include <linux/ipc_namespace.h> |
5f921ae9 | 85 | |
7153e402 | 86 | #include <linux/uaccess.h> |
1da177e4 LT |
87 | #include "util.h" |
88 | ||
e57940d7 MS |
89 | /* One semaphore structure for each semaphore in the system. */ |
90 | struct sem { | |
91 | int semval; /* current value */ | |
a5f4db87 DB |
92 | /* |
93 | * PID of the process that last modified the semaphore. For | |
94 | * Linux, specifically these are: | |
95 | * - semop | |
96 | * - semctl, via SETVAL and SETALL. | |
97 | * - at task exit when performing undo adjustments (see exit_sem). | |
98 | */ | |
99 | int sempid; | |
6062a8dc | 100 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
1a82e9e1 MS |
101 | struct list_head pending_alter; /* pending single-sop operations */ |
102 | /* that alter the semaphore */ | |
103 | struct list_head pending_const; /* pending single-sop operations */ | |
104 | /* that do not alter the semaphore*/ | |
d12e1e50 | 105 | time_t sem_otime; /* candidate for sem_otime */ |
f5c936c0 | 106 | } ____cacheline_aligned_in_smp; |
e57940d7 MS |
107 | |
108 | /* One queue for each sleeping process in the system. */ | |
109 | struct sem_queue { | |
e57940d7 MS |
110 | struct list_head list; /* queue of pending operations */ |
111 | struct task_struct *sleeper; /* this process */ | |
112 | struct sem_undo *undo; /* undo structure */ | |
113 | int pid; /* process id of requesting process */ | |
114 | int status; /* completion status of operation */ | |
115 | struct sembuf *sops; /* array of pending operations */ | |
ed247b7c | 116 | struct sembuf *blocking; /* the operation that blocked */ |
e57940d7 | 117 | int nsops; /* number of operations */ |
4ce33ec2 DB |
118 | bool alter; /* does *sops alter the array? */ |
119 | bool dupsop; /* sops on more than one sem_num */ | |
e57940d7 MS |
120 | }; |
121 | ||
122 | /* Each task has a list of undo requests. They are executed automatically | |
123 | * when the process exits. | |
124 | */ | |
125 | struct sem_undo { | |
126 | struct list_head list_proc; /* per-process list: * | |
127 | * all undos from one process | |
128 | * rcu protected */ | |
129 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
130 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
131 | struct list_head list_id; /* per semaphore array list: | |
132 | * all undos for one array */ | |
133 | int semid; /* semaphore set identifier */ | |
134 | short *semadj; /* array of adjustments */ | |
135 | /* one per semaphore */ | |
136 | }; | |
137 | ||
138 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
139 | * that may be shared among all a CLONE_SYSVSEM task group. | |
140 | */ | |
141 | struct sem_undo_list { | |
142 | atomic_t refcnt; | |
143 | spinlock_t lock; | |
144 | struct list_head list_proc; | |
145 | }; | |
146 | ||
147 | ||
ed2ddbf8 | 148 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 149 | |
1b531f21 | 150 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 151 | |
7748dbfa | 152 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 153 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 154 | #ifdef CONFIG_PROC_FS |
19b4946c | 155 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
156 | #endif |
157 | ||
158 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
159 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
160 | ||
161 | /* | |
758a6ba3 | 162 | * Locking: |
5864a2fd | 163 | * a) global sem_lock() for read/write |
1da177e4 | 164 | * sem_undo.id_next, |
758a6ba3 | 165 | * sem_array.complex_count, |
5864a2fd MS |
166 | * sem_array.complex_mode |
167 | * sem_array.pending{_alter,_const}, | |
168 | * sem_array.sem_undo | |
46c0a8ca | 169 | * |
5864a2fd | 170 | * b) global or semaphore sem_lock() for read/write: |
758a6ba3 | 171 | * sem_array.sem_base[i].pending_{const,alter}: |
5864a2fd MS |
172 | * sem_array.complex_mode (for read) |
173 | * | |
174 | * c) special: | |
175 | * sem_undo_list.list_proc: | |
176 | * * undo_list->lock for write | |
177 | * * rcu for read | |
1da177e4 LT |
178 | */ |
179 | ||
e3893534 KK |
180 | #define sc_semmsl sem_ctls[0] |
181 | #define sc_semmns sem_ctls[1] | |
182 | #define sc_semopm sem_ctls[2] | |
183 | #define sc_semmni sem_ctls[3] | |
184 | ||
ed2ddbf8 | 185 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 186 | { |
e3893534 KK |
187 | ns->sc_semmsl = SEMMSL; |
188 | ns->sc_semmns = SEMMNS; | |
189 | ns->sc_semopm = SEMOPM; | |
190 | ns->sc_semmni = SEMMNI; | |
191 | ns->used_sems = 0; | |
ed2ddbf8 | 192 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
193 | } |
194 | ||
ae5e1b22 | 195 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
196 | void sem_exit_ns(struct ipc_namespace *ns) |
197 | { | |
01b8b07a | 198 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 199 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 200 | } |
ae5e1b22 | 201 | #endif |
1da177e4 | 202 | |
239521f3 | 203 | void __init sem_init(void) |
1da177e4 | 204 | { |
ed2ddbf8 | 205 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
206 | ipc_init_proc_interface("sysvipc/sem", |
207 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 208 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
209 | } |
210 | ||
f269f40a MS |
211 | /** |
212 | * unmerge_queues - unmerge queues, if possible. | |
213 | * @sma: semaphore array | |
214 | * | |
215 | * The function unmerges the wait queues if complex_count is 0. | |
216 | * It must be called prior to dropping the global semaphore array lock. | |
217 | */ | |
218 | static void unmerge_queues(struct sem_array *sma) | |
219 | { | |
220 | struct sem_queue *q, *tq; | |
221 | ||
222 | /* complex operations still around? */ | |
223 | if (sma->complex_count) | |
224 | return; | |
225 | /* | |
226 | * We will switch back to simple mode. | |
227 | * Move all pending operation back into the per-semaphore | |
228 | * queues. | |
229 | */ | |
230 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
231 | struct sem *curr; | |
232 | curr = &sma->sem_base[q->sops[0].sem_num]; | |
233 | ||
234 | list_add_tail(&q->list, &curr->pending_alter); | |
235 | } | |
236 | INIT_LIST_HEAD(&sma->pending_alter); | |
237 | } | |
238 | ||
239 | /** | |
8001c858 | 240 | * merge_queues - merge single semop queues into global queue |
f269f40a MS |
241 | * @sma: semaphore array |
242 | * | |
243 | * This function merges all per-semaphore queues into the global queue. | |
244 | * It is necessary to achieve FIFO ordering for the pending single-sop | |
245 | * operations when a multi-semop operation must sleep. | |
246 | * Only the alter operations must be moved, the const operations can stay. | |
247 | */ | |
248 | static void merge_queues(struct sem_array *sma) | |
249 | { | |
250 | int i; | |
251 | for (i = 0; i < sma->sem_nsems; i++) { | |
252 | struct sem *sem = sma->sem_base + i; | |
253 | ||
254 | list_splice_init(&sem->pending_alter, &sma->pending_alter); | |
255 | } | |
256 | } | |
257 | ||
53dad6d3 DB |
258 | static void sem_rcu_free(struct rcu_head *head) |
259 | { | |
260 | struct ipc_rcu *p = container_of(head, struct ipc_rcu, rcu); | |
261 | struct sem_array *sma = ipc_rcu_to_struct(p); | |
262 | ||
263 | security_sem_free(sma); | |
264 | ipc_rcu_free(head); | |
265 | } | |
266 | ||
5e9d5275 | 267 | /* |
5864a2fd | 268 | * Enter the mode suitable for non-simple operations: |
5e9d5275 | 269 | * Caller must own sem_perm.lock. |
5e9d5275 | 270 | */ |
5864a2fd | 271 | static void complexmode_enter(struct sem_array *sma) |
5e9d5275 MS |
272 | { |
273 | int i; | |
274 | struct sem *sem; | |
275 | ||
5864a2fd MS |
276 | if (sma->complex_mode) { |
277 | /* We are already in complex_mode. Nothing to do */ | |
6d07b68c MS |
278 | return; |
279 | } | |
280 | ||
5864a2fd MS |
281 | /* We need a full barrier after seting complex_mode: |
282 | * The write to complex_mode must be visible | |
283 | * before we read the first sem->lock spinlock state. | |
284 | */ | |
285 | smp_store_mb(sma->complex_mode, true); | |
286 | ||
5e9d5275 MS |
287 | for (i = 0; i < sma->sem_nsems; i++) { |
288 | sem = sma->sem_base + i; | |
289 | spin_unlock_wait(&sem->lock); | |
290 | } | |
5864a2fd MS |
291 | /* |
292 | * spin_unlock_wait() is not a memory barriers, it is only a | |
293 | * control barrier. The code must pair with spin_unlock(&sem->lock), | |
294 | * thus just the control barrier is insufficient. | |
295 | * | |
296 | * smp_rmb() is sufficient, as writes cannot pass the control barrier. | |
297 | */ | |
298 | smp_rmb(); | |
299 | } | |
300 | ||
301 | /* | |
302 | * Try to leave the mode that disallows simple operations: | |
303 | * Caller must own sem_perm.lock. | |
304 | */ | |
305 | static void complexmode_tryleave(struct sem_array *sma) | |
306 | { | |
307 | if (sma->complex_count) { | |
308 | /* Complex ops are sleeping. | |
309 | * We must stay in complex mode | |
310 | */ | |
311 | return; | |
312 | } | |
313 | /* | |
314 | * Immediately after setting complex_mode to false, | |
315 | * a simple op can start. Thus: all memory writes | |
316 | * performed by the current operation must be visible | |
317 | * before we set complex_mode to false. | |
318 | */ | |
319 | smp_store_release(&sma->complex_mode, false); | |
5e9d5275 MS |
320 | } |
321 | ||
5864a2fd | 322 | #define SEM_GLOBAL_LOCK (-1) |
6062a8dc RR |
323 | /* |
324 | * If the request contains only one semaphore operation, and there are | |
325 | * no complex transactions pending, lock only the semaphore involved. | |
326 | * Otherwise, lock the entire semaphore array, since we either have | |
327 | * multiple semaphores in our own semops, or we need to look at | |
328 | * semaphores from other pending complex operations. | |
6062a8dc RR |
329 | */ |
330 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
331 | int nsops) | |
332 | { | |
5e9d5275 | 333 | struct sem *sem; |
6062a8dc | 334 | |
5e9d5275 MS |
335 | if (nsops != 1) { |
336 | /* Complex operation - acquire a full lock */ | |
337 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 338 | |
5864a2fd MS |
339 | /* Prevent parallel simple ops */ |
340 | complexmode_enter(sma); | |
341 | return SEM_GLOBAL_LOCK; | |
5e9d5275 MS |
342 | } |
343 | ||
344 | /* | |
345 | * Only one semaphore affected - try to optimize locking. | |
5864a2fd MS |
346 | * Optimized locking is possible if no complex operation |
347 | * is either enqueued or processed right now. | |
348 | * | |
349 | * Both facts are tracked by complex_mode. | |
5e9d5275 MS |
350 | */ |
351 | sem = sma->sem_base + sops->sem_num; | |
6062a8dc | 352 | |
5864a2fd MS |
353 | /* |
354 | * Initial check for complex_mode. Just an optimization, | |
355 | * no locking, no memory barrier. | |
356 | */ | |
357 | if (!sma->complex_mode) { | |
6062a8dc | 358 | /* |
5e9d5275 MS |
359 | * It appears that no complex operation is around. |
360 | * Acquire the per-semaphore lock. | |
6062a8dc | 361 | */ |
5e9d5275 MS |
362 | spin_lock(&sem->lock); |
363 | ||
5864a2fd MS |
364 | /* |
365 | * See 51d7d5205d33 | |
366 | * ("powerpc: Add smp_mb() to arch_spin_is_locked()"): | |
367 | * A full barrier is required: the write of sem->lock | |
368 | * must be visible before the read is executed | |
369 | */ | |
370 | smp_mb(); | |
5e9d5275 | 371 | |
5864a2fd MS |
372 | if (!smp_load_acquire(&sma->complex_mode)) { |
373 | /* fast path successful! */ | |
374 | return sops->sem_num; | |
6062a8dc | 375 | } |
5e9d5275 MS |
376 | spin_unlock(&sem->lock); |
377 | } | |
378 | ||
379 | /* slow path: acquire the full lock */ | |
380 | ipc_lock_object(&sma->sem_perm); | |
6062a8dc | 381 | |
5e9d5275 MS |
382 | if (sma->complex_count == 0) { |
383 | /* False alarm: | |
384 | * There is no complex operation, thus we can switch | |
385 | * back to the fast path. | |
386 | */ | |
387 | spin_lock(&sem->lock); | |
388 | ipc_unlock_object(&sma->sem_perm); | |
389 | return sops->sem_num; | |
6062a8dc | 390 | } else { |
5e9d5275 MS |
391 | /* Not a false alarm, thus complete the sequence for a |
392 | * full lock. | |
6062a8dc | 393 | */ |
5864a2fd MS |
394 | complexmode_enter(sma); |
395 | return SEM_GLOBAL_LOCK; | |
6062a8dc | 396 | } |
6062a8dc RR |
397 | } |
398 | ||
399 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
400 | { | |
5864a2fd | 401 | if (locknum == SEM_GLOBAL_LOCK) { |
f269f40a | 402 | unmerge_queues(sma); |
5864a2fd | 403 | complexmode_tryleave(sma); |
cf9d5d78 | 404 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc RR |
405 | } else { |
406 | struct sem *sem = sma->sem_base + locknum; | |
407 | spin_unlock(&sem->lock); | |
408 | } | |
6062a8dc RR |
409 | } |
410 | ||
3e148c79 | 411 | /* |
d9a605e4 | 412 | * sem_lock_(check_) routines are called in the paths where the rwsem |
3e148c79 | 413 | * is not held. |
321310ce LT |
414 | * |
415 | * The caller holds the RCU read lock. | |
3e148c79 | 416 | */ |
6062a8dc RR |
417 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
418 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 419 | { |
c460b662 RR |
420 | struct kern_ipc_perm *ipcp; |
421 | struct sem_array *sma; | |
03f02c76 | 422 | |
55b7ae50 | 423 | ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
321310ce LT |
424 | if (IS_ERR(ipcp)) |
425 | return ERR_CAST(ipcp); | |
b1ed88b4 | 426 | |
6062a8dc RR |
427 | sma = container_of(ipcp, struct sem_array, sem_perm); |
428 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
429 | |
430 | /* ipc_rmid() may have already freed the ID while sem_lock | |
431 | * was spinning: verify that the structure is still valid | |
432 | */ | |
72a8ff2f | 433 | if (ipc_valid_object(ipcp)) |
c460b662 RR |
434 | return container_of(ipcp, struct sem_array, sem_perm); |
435 | ||
6062a8dc | 436 | sem_unlock(sma, *locknum); |
321310ce | 437 | return ERR_PTR(-EINVAL); |
023a5355 ND |
438 | } |
439 | ||
16df3674 DB |
440 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
441 | { | |
55b7ae50 | 442 | struct kern_ipc_perm *ipcp = ipc_obtain_object_idr(&sem_ids(ns), id); |
16df3674 DB |
443 | |
444 | if (IS_ERR(ipcp)) | |
445 | return ERR_CAST(ipcp); | |
446 | ||
447 | return container_of(ipcp, struct sem_array, sem_perm); | |
448 | } | |
449 | ||
16df3674 DB |
450 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
451 | int id) | |
452 | { | |
453 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
454 | ||
455 | if (IS_ERR(ipcp)) | |
456 | return ERR_CAST(ipcp); | |
b1ed88b4 | 457 | |
03f02c76 | 458 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
459 | } |
460 | ||
6ff37972 PP |
461 | static inline void sem_lock_and_putref(struct sem_array *sma) |
462 | { | |
6062a8dc | 463 | sem_lock(sma, NULL, -1); |
9b24fef9 | 464 | ipc_rcu_putref(sma, sem_rcu_free); |
6ff37972 PP |
465 | } |
466 | ||
7ca7e564 ND |
467 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
468 | { | |
469 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
470 | } | |
471 | ||
f4566f04 ND |
472 | /** |
473 | * newary - Create a new semaphore set | |
474 | * @ns: namespace | |
475 | * @params: ptr to the structure that contains key, semflg and nsems | |
476 | * | |
d9a605e4 | 477 | * Called with sem_ids.rwsem held (as a writer) |
f4566f04 | 478 | */ |
7748dbfa | 479 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
480 | { |
481 | int id; | |
482 | int retval; | |
483 | struct sem_array *sma; | |
484 | int size; | |
7748dbfa ND |
485 | key_t key = params->key; |
486 | int nsems = params->u.nsems; | |
487 | int semflg = params->flg; | |
b97e820f | 488 | int i; |
1da177e4 LT |
489 | |
490 | if (!nsems) | |
491 | return -EINVAL; | |
e3893534 | 492 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
493 | return -ENOSPC; |
494 | ||
239521f3 | 495 | size = sizeof(*sma) + nsems * sizeof(struct sem); |
1da177e4 | 496 | sma = ipc_rcu_alloc(size); |
3ab08fe2 | 497 | if (!sma) |
1da177e4 | 498 | return -ENOMEM; |
3ab08fe2 | 499 | |
239521f3 | 500 | memset(sma, 0, size); |
1da177e4 LT |
501 | |
502 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
503 | sma->sem_perm.key = key; | |
504 | ||
505 | sma->sem_perm.security = NULL; | |
506 | retval = security_sem_alloc(sma); | |
507 | if (retval) { | |
53dad6d3 | 508 | ipc_rcu_putref(sma, ipc_rcu_free); |
1da177e4 LT |
509 | return retval; |
510 | } | |
511 | ||
1da177e4 | 512 | sma->sem_base = (struct sem *) &sma[1]; |
b97e820f | 513 | |
6062a8dc | 514 | for (i = 0; i < nsems; i++) { |
1a82e9e1 MS |
515 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
516 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
517 | spin_lock_init(&sma->sem_base[i].lock); |
518 | } | |
b97e820f MS |
519 | |
520 | sma->complex_count = 0; | |
5864a2fd | 521 | sma->complex_mode = true; /* dropped by sem_unlock below */ |
1a82e9e1 MS |
522 | INIT_LIST_HEAD(&sma->pending_alter); |
523 | INIT_LIST_HEAD(&sma->pending_const); | |
4daa28f6 | 524 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
525 | sma->sem_nsems = nsems; |
526 | sma->sem_ctime = get_seconds(); | |
e8577d1f MS |
527 | |
528 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); | |
529 | if (id < 0) { | |
530 | ipc_rcu_putref(sma, sem_rcu_free); | |
531 | return id; | |
532 | } | |
533 | ns->used_sems += nsems; | |
534 | ||
6062a8dc | 535 | sem_unlock(sma, -1); |
6d49dab8 | 536 | rcu_read_unlock(); |
1da177e4 | 537 | |
7ca7e564 | 538 | return sma->sem_perm.id; |
1da177e4 LT |
539 | } |
540 | ||
7748dbfa | 541 | |
f4566f04 | 542 | /* |
d9a605e4 | 543 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 544 | */ |
03f02c76 | 545 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 546 | { |
03f02c76 ND |
547 | struct sem_array *sma; |
548 | ||
549 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
550 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
551 | } |
552 | ||
f4566f04 | 553 | /* |
d9a605e4 | 554 | * Called with sem_ids.rwsem and ipcp locked. |
f4566f04 | 555 | */ |
03f02c76 ND |
556 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
557 | struct ipc_params *params) | |
7748dbfa | 558 | { |
03f02c76 ND |
559 | struct sem_array *sma; |
560 | ||
561 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
562 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
563 | return -EINVAL; |
564 | ||
565 | return 0; | |
566 | } | |
567 | ||
d5460c99 | 568 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 569 | { |
e3893534 | 570 | struct ipc_namespace *ns; |
eb66ec44 MK |
571 | static const struct ipc_ops sem_ops = { |
572 | .getnew = newary, | |
573 | .associate = sem_security, | |
574 | .more_checks = sem_more_checks, | |
575 | }; | |
7748dbfa | 576 | struct ipc_params sem_params; |
e3893534 KK |
577 | |
578 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 579 | |
e3893534 | 580 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 581 | return -EINVAL; |
7ca7e564 | 582 | |
7748dbfa ND |
583 | sem_params.key = key; |
584 | sem_params.flg = semflg; | |
585 | sem_params.u.nsems = nsems; | |
1da177e4 | 586 | |
7748dbfa | 587 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
588 | } |
589 | ||
78f5009c | 590 | /** |
4ce33ec2 DB |
591 | * perform_atomic_semop[_slow] - Attempt to perform semaphore |
592 | * operations on a given array. | |
758a6ba3 | 593 | * @sma: semaphore array |
d198cd6d | 594 | * @q: struct sem_queue that describes the operation |
758a6ba3 | 595 | * |
4ce33ec2 DB |
596 | * Caller blocking are as follows, based the value |
597 | * indicated by the semaphore operation (sem_op): | |
598 | * | |
599 | * (1) >0 never blocks. | |
600 | * (2) 0 (wait-for-zero operation): semval is non-zero. | |
601 | * (3) <0 attempting to decrement semval to a value smaller than zero. | |
602 | * | |
758a6ba3 MS |
603 | * Returns 0 if the operation was possible. |
604 | * Returns 1 if the operation is impossible, the caller must sleep. | |
4ce33ec2 | 605 | * Returns <0 for error codes. |
1da177e4 | 606 | */ |
4ce33ec2 | 607 | static int perform_atomic_semop_slow(struct sem_array *sma, struct sem_queue *q) |
1da177e4 | 608 | { |
d198cd6d | 609 | int result, sem_op, nsops, pid; |
1da177e4 | 610 | struct sembuf *sop; |
239521f3 | 611 | struct sem *curr; |
d198cd6d MS |
612 | struct sembuf *sops; |
613 | struct sem_undo *un; | |
614 | ||
615 | sops = q->sops; | |
616 | nsops = q->nsops; | |
617 | un = q->undo; | |
1da177e4 LT |
618 | |
619 | for (sop = sops; sop < sops + nsops; sop++) { | |
620 | curr = sma->sem_base + sop->sem_num; | |
621 | sem_op = sop->sem_op; | |
622 | result = curr->semval; | |
78f5009c | 623 | |
1da177e4 LT |
624 | if (!sem_op && result) |
625 | goto would_block; | |
626 | ||
627 | result += sem_op; | |
628 | if (result < 0) | |
629 | goto would_block; | |
630 | if (result > SEMVMX) | |
631 | goto out_of_range; | |
78f5009c | 632 | |
1da177e4 LT |
633 | if (sop->sem_flg & SEM_UNDO) { |
634 | int undo = un->semadj[sop->sem_num] - sem_op; | |
78f5009c | 635 | /* Exceeding the undo range is an error. */ |
1da177e4 LT |
636 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
637 | goto out_of_range; | |
78f5009c | 638 | un->semadj[sop->sem_num] = undo; |
1da177e4 | 639 | } |
78f5009c | 640 | |
1da177e4 LT |
641 | curr->semval = result; |
642 | } | |
643 | ||
644 | sop--; | |
d198cd6d | 645 | pid = q->pid; |
1da177e4 LT |
646 | while (sop >= sops) { |
647 | sma->sem_base[sop->sem_num].sempid = pid; | |
1da177e4 LT |
648 | sop--; |
649 | } | |
78f5009c | 650 | |
1da177e4 LT |
651 | return 0; |
652 | ||
653 | out_of_range: | |
654 | result = -ERANGE; | |
655 | goto undo; | |
656 | ||
657 | would_block: | |
ed247b7c MS |
658 | q->blocking = sop; |
659 | ||
1da177e4 LT |
660 | if (sop->sem_flg & IPC_NOWAIT) |
661 | result = -EAGAIN; | |
662 | else | |
663 | result = 1; | |
664 | ||
665 | undo: | |
666 | sop--; | |
667 | while (sop >= sops) { | |
78f5009c PM |
668 | sem_op = sop->sem_op; |
669 | sma->sem_base[sop->sem_num].semval -= sem_op; | |
670 | if (sop->sem_flg & SEM_UNDO) | |
671 | un->semadj[sop->sem_num] += sem_op; | |
1da177e4 LT |
672 | sop--; |
673 | } | |
674 | ||
675 | return result; | |
676 | } | |
677 | ||
4ce33ec2 DB |
678 | static int perform_atomic_semop(struct sem_array *sma, struct sem_queue *q) |
679 | { | |
680 | int result, sem_op, nsops; | |
681 | struct sembuf *sop; | |
682 | struct sem *curr; | |
683 | struct sembuf *sops; | |
684 | struct sem_undo *un; | |
685 | ||
686 | sops = q->sops; | |
687 | nsops = q->nsops; | |
688 | un = q->undo; | |
689 | ||
690 | if (unlikely(q->dupsop)) | |
691 | return perform_atomic_semop_slow(sma, q); | |
692 | ||
693 | /* | |
694 | * We scan the semaphore set twice, first to ensure that the entire | |
695 | * operation can succeed, therefore avoiding any pointless writes | |
696 | * to shared memory and having to undo such changes in order to block | |
697 | * until the operations can go through. | |
698 | */ | |
699 | for (sop = sops; sop < sops + nsops; sop++) { | |
700 | curr = sma->sem_base + sop->sem_num; | |
701 | sem_op = sop->sem_op; | |
702 | result = curr->semval; | |
703 | ||
704 | if (!sem_op && result) | |
705 | goto would_block; /* wait-for-zero */ | |
706 | ||
707 | result += sem_op; | |
708 | if (result < 0) | |
709 | goto would_block; | |
710 | ||
711 | if (result > SEMVMX) | |
712 | return -ERANGE; | |
713 | ||
714 | if (sop->sem_flg & SEM_UNDO) { | |
715 | int undo = un->semadj[sop->sem_num] - sem_op; | |
716 | ||
717 | /* Exceeding the undo range is an error. */ | |
718 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
719 | return -ERANGE; | |
720 | } | |
721 | } | |
722 | ||
723 | for (sop = sops; sop < sops + nsops; sop++) { | |
724 | curr = sma->sem_base + sop->sem_num; | |
725 | sem_op = sop->sem_op; | |
726 | result = curr->semval; | |
727 | ||
728 | if (sop->sem_flg & SEM_UNDO) { | |
729 | int undo = un->semadj[sop->sem_num] - sem_op; | |
730 | ||
731 | un->semadj[sop->sem_num] = undo; | |
732 | } | |
733 | curr->semval += sem_op; | |
734 | curr->sempid = q->pid; | |
735 | } | |
736 | ||
737 | return 0; | |
738 | ||
739 | would_block: | |
740 | q->blocking = sop; | |
741 | return sop->sem_flg & IPC_NOWAIT ? -EAGAIN : 1; | |
742 | } | |
743 | ||
9ae949fa DB |
744 | static inline void wake_up_sem_queue_prepare(struct sem_queue *q, int error, |
745 | struct wake_q_head *wake_q) | |
0a2b9d4c | 746 | { |
9ae949fa DB |
747 | wake_q_add(wake_q, q->sleeper); |
748 | /* | |
749 | * Rely on the above implicit barrier, such that we can | |
750 | * ensure that we hold reference to the task before setting | |
751 | * q->status. Otherwise we could race with do_exit if the | |
752 | * task is awoken by an external event before calling | |
753 | * wake_up_process(). | |
754 | */ | |
755 | WRITE_ONCE(q->status, error); | |
d4212093 NP |
756 | } |
757 | ||
b97e820f MS |
758 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
759 | { | |
760 | list_del(&q->list); | |
9f1bc2c9 | 761 | if (q->nsops > 1) |
b97e820f MS |
762 | sma->complex_count--; |
763 | } | |
764 | ||
fd5db422 MS |
765 | /** check_restart(sma, q) |
766 | * @sma: semaphore array | |
767 | * @q: the operation that just completed | |
768 | * | |
769 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
770 | * waiting operations. Therefore this function checks if the restart is | |
771 | * really necessary. It is called after a previously waiting operation | |
1a82e9e1 MS |
772 | * modified the array. |
773 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 | 774 | */ |
4663d3e8 | 775 | static inline int check_restart(struct sem_array *sma, struct sem_queue *q) |
fd5db422 | 776 | { |
1a82e9e1 MS |
777 | /* pending complex alter operations are too difficult to analyse */ |
778 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
779 | return 1; |
780 | ||
781 | /* we were a sleeping complex operation. Too difficult */ | |
782 | if (q->nsops > 1) | |
783 | return 1; | |
784 | ||
1a82e9e1 MS |
785 | /* It is impossible that someone waits for the new value: |
786 | * - complex operations always restart. | |
787 | * - wait-for-zero are handled seperately. | |
788 | * - q is a previously sleeping simple operation that | |
789 | * altered the array. It must be a decrement, because | |
790 | * simple increments never sleep. | |
791 | * - If there are older (higher priority) decrements | |
792 | * in the queue, then they have observed the original | |
793 | * semval value and couldn't proceed. The operation | |
794 | * decremented to value - thus they won't proceed either. | |
795 | */ | |
796 | return 0; | |
797 | } | |
fd5db422 | 798 | |
1a82e9e1 | 799 | /** |
8001c858 | 800 | * wake_const_ops - wake up non-alter tasks |
1a82e9e1 MS |
801 | * @sma: semaphore array. |
802 | * @semnum: semaphore that was modified. | |
9ae949fa | 803 | * @wake_q: lockless wake-queue head. |
1a82e9e1 MS |
804 | * |
805 | * wake_const_ops must be called after a semaphore in a semaphore array | |
806 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
807 | * be called with semnum = -1, as well as with the number of each modified | |
808 | * semaphore. | |
9ae949fa | 809 | * The tasks that must be woken up are added to @wake_q. The return code |
1a82e9e1 MS |
810 | * is stored in q->pid. |
811 | * The function returns 1 if at least one operation was completed successfully. | |
812 | */ | |
813 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
9ae949fa | 814 | struct wake_q_head *wake_q) |
1a82e9e1 | 815 | { |
f150f02c | 816 | struct sem_queue *q, *tmp; |
1a82e9e1 MS |
817 | struct list_head *pending_list; |
818 | int semop_completed = 0; | |
819 | ||
820 | if (semnum == -1) | |
821 | pending_list = &sma->pending_const; | |
822 | else | |
823 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 824 | |
f150f02c DB |
825 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
826 | int error = perform_atomic_semop(sma, q); | |
1a82e9e1 | 827 | |
f150f02c DB |
828 | if (error > 0) |
829 | continue; | |
830 | /* operation completed, remove from queue & wakeup */ | |
831 | unlink_queue(sma, q); | |
1a82e9e1 | 832 | |
f150f02c DB |
833 | wake_up_sem_queue_prepare(q, error, wake_q); |
834 | if (error == 0) | |
835 | semop_completed = 1; | |
1a82e9e1 | 836 | } |
f150f02c | 837 | |
1a82e9e1 MS |
838 | return semop_completed; |
839 | } | |
840 | ||
841 | /** | |
8001c858 | 842 | * do_smart_wakeup_zero - wakeup all wait for zero tasks |
1a82e9e1 MS |
843 | * @sma: semaphore array |
844 | * @sops: operations that were performed | |
845 | * @nsops: number of operations | |
9ae949fa | 846 | * @wake_q: lockless wake-queue head |
1a82e9e1 | 847 | * |
8001c858 DB |
848 | * Checks all required queue for wait-for-zero operations, based |
849 | * on the actual changes that were performed on the semaphore array. | |
1a82e9e1 MS |
850 | * The function returns 1 if at least one operation was completed successfully. |
851 | */ | |
852 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
9ae949fa | 853 | int nsops, struct wake_q_head *wake_q) |
1a82e9e1 MS |
854 | { |
855 | int i; | |
856 | int semop_completed = 0; | |
857 | int got_zero = 0; | |
858 | ||
859 | /* first: the per-semaphore queues, if known */ | |
860 | if (sops) { | |
861 | for (i = 0; i < nsops; i++) { | |
862 | int num = sops[i].sem_num; | |
863 | ||
864 | if (sma->sem_base[num].semval == 0) { | |
865 | got_zero = 1; | |
9ae949fa | 866 | semop_completed |= wake_const_ops(sma, num, wake_q); |
1a82e9e1 MS |
867 | } |
868 | } | |
869 | } else { | |
870 | /* | |
871 | * No sops means modified semaphores not known. | |
872 | * Assume all were changed. | |
fd5db422 | 873 | */ |
1a82e9e1 MS |
874 | for (i = 0; i < sma->sem_nsems; i++) { |
875 | if (sma->sem_base[i].semval == 0) { | |
876 | got_zero = 1; | |
9ae949fa | 877 | semop_completed |= wake_const_ops(sma, i, wake_q); |
1a82e9e1 MS |
878 | } |
879 | } | |
fd5db422 MS |
880 | } |
881 | /* | |
1a82e9e1 MS |
882 | * If one of the modified semaphores got 0, |
883 | * then check the global queue, too. | |
fd5db422 | 884 | */ |
1a82e9e1 | 885 | if (got_zero) |
9ae949fa | 886 | semop_completed |= wake_const_ops(sma, -1, wake_q); |
fd5db422 | 887 | |
1a82e9e1 | 888 | return semop_completed; |
fd5db422 MS |
889 | } |
890 | ||
636c6be8 MS |
891 | |
892 | /** | |
8001c858 | 893 | * update_queue - look for tasks that can be completed. |
636c6be8 MS |
894 | * @sma: semaphore array. |
895 | * @semnum: semaphore that was modified. | |
9ae949fa | 896 | * @wake_q: lockless wake-queue head. |
636c6be8 MS |
897 | * |
898 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
899 | * was modified. If multiple semaphores were modified, update_queue must |
900 | * be called with semnum = -1, as well as with the number of each modified | |
901 | * semaphore. | |
9ae949fa | 902 | * The tasks that must be woken up are added to @wake_q. The return code |
0a2b9d4c | 903 | * is stored in q->pid. |
1a82e9e1 MS |
904 | * The function internally checks if const operations can now succeed. |
905 | * | |
0a2b9d4c | 906 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 907 | */ |
9ae949fa | 908 | static int update_queue(struct sem_array *sma, int semnum, struct wake_q_head *wake_q) |
1da177e4 | 909 | { |
f150f02c | 910 | struct sem_queue *q, *tmp; |
636c6be8 | 911 | struct list_head *pending_list; |
0a2b9d4c | 912 | int semop_completed = 0; |
636c6be8 | 913 | |
9f1bc2c9 | 914 | if (semnum == -1) |
1a82e9e1 | 915 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 916 | else |
1a82e9e1 | 917 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
918 | |
919 | again: | |
f150f02c | 920 | list_for_each_entry_safe(q, tmp, pending_list, list) { |
fd5db422 | 921 | int error, restart; |
636c6be8 | 922 | |
d987f8b2 MS |
923 | /* If we are scanning the single sop, per-semaphore list of |
924 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 925 | * necessary to scan further: simple increments |
d987f8b2 MS |
926 | * that affect only one entry succeed immediately and cannot |
927 | * be in the per semaphore pending queue, and decrements | |
928 | * cannot be successful if the value is already 0. | |
929 | */ | |
1a82e9e1 | 930 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
931 | break; |
932 | ||
d198cd6d | 933 | error = perform_atomic_semop(sma, q); |
1da177e4 LT |
934 | |
935 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
936 | if (error > 0) |
937 | continue; | |
938 | ||
b97e820f | 939 | unlink_queue(sma, q); |
9cad200c | 940 | |
0a2b9d4c | 941 | if (error) { |
fd5db422 | 942 | restart = 0; |
0a2b9d4c MS |
943 | } else { |
944 | semop_completed = 1; | |
9ae949fa | 945 | do_smart_wakeup_zero(sma, q->sops, q->nsops, wake_q); |
fd5db422 | 946 | restart = check_restart(sma, q); |
0a2b9d4c | 947 | } |
fd5db422 | 948 | |
9ae949fa | 949 | wake_up_sem_queue_prepare(q, error, wake_q); |
fd5db422 | 950 | if (restart) |
9cad200c | 951 | goto again; |
1da177e4 | 952 | } |
0a2b9d4c | 953 | return semop_completed; |
1da177e4 LT |
954 | } |
955 | ||
0e8c6656 | 956 | /** |
8001c858 | 957 | * set_semotime - set sem_otime |
0e8c6656 MS |
958 | * @sma: semaphore array |
959 | * @sops: operations that modified the array, may be NULL | |
960 | * | |
961 | * sem_otime is replicated to avoid cache line trashing. | |
962 | * This function sets one instance to the current time. | |
963 | */ | |
964 | static void set_semotime(struct sem_array *sma, struct sembuf *sops) | |
965 | { | |
966 | if (sops == NULL) { | |
967 | sma->sem_base[0].sem_otime = get_seconds(); | |
968 | } else { | |
969 | sma->sem_base[sops[0].sem_num].sem_otime = | |
970 | get_seconds(); | |
971 | } | |
972 | } | |
973 | ||
0a2b9d4c | 974 | /** |
8001c858 | 975 | * do_smart_update - optimized update_queue |
fd5db422 MS |
976 | * @sma: semaphore array |
977 | * @sops: operations that were performed | |
978 | * @nsops: number of operations | |
0a2b9d4c | 979 | * @otime: force setting otime |
9ae949fa | 980 | * @wake_q: lockless wake-queue head |
fd5db422 | 981 | * |
1a82e9e1 MS |
982 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
983 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c | 984 | * Note that the function does not do the actual wake-up: the caller is |
9ae949fa | 985 | * responsible for calling wake_up_q(). |
0a2b9d4c | 986 | * It is safe to perform this call after dropping all locks. |
fd5db422 | 987 | */ |
0a2b9d4c | 988 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
9ae949fa | 989 | int otime, struct wake_q_head *wake_q) |
fd5db422 MS |
990 | { |
991 | int i; | |
992 | ||
9ae949fa | 993 | otime |= do_smart_wakeup_zero(sma, sops, nsops, wake_q); |
1a82e9e1 | 994 | |
f269f40a MS |
995 | if (!list_empty(&sma->pending_alter)) { |
996 | /* semaphore array uses the global queue - just process it. */ | |
9ae949fa | 997 | otime |= update_queue(sma, -1, wake_q); |
f269f40a MS |
998 | } else { |
999 | if (!sops) { | |
1000 | /* | |
1001 | * No sops, thus the modified semaphores are not | |
1002 | * known. Check all. | |
1003 | */ | |
1004 | for (i = 0; i < sma->sem_nsems; i++) | |
9ae949fa | 1005 | otime |= update_queue(sma, i, wake_q); |
f269f40a MS |
1006 | } else { |
1007 | /* | |
1008 | * Check the semaphores that were increased: | |
1009 | * - No complex ops, thus all sleeping ops are | |
1010 | * decrease. | |
1011 | * - if we decreased the value, then any sleeping | |
1012 | * semaphore ops wont be able to run: If the | |
1013 | * previous value was too small, then the new | |
1014 | * value will be too small, too. | |
1015 | */ | |
1016 | for (i = 0; i < nsops; i++) { | |
1017 | if (sops[i].sem_op > 0) { | |
1018 | otime |= update_queue(sma, | |
9ae949fa | 1019 | sops[i].sem_num, wake_q); |
f269f40a | 1020 | } |
ab465df9 | 1021 | } |
9f1bc2c9 | 1022 | } |
fd5db422 | 1023 | } |
0e8c6656 MS |
1024 | if (otime) |
1025 | set_semotime(sma, sops); | |
fd5db422 MS |
1026 | } |
1027 | ||
2f2ed41d | 1028 | /* |
b220c57a | 1029 | * check_qop: Test if a queued operation sleeps on the semaphore semnum |
2f2ed41d MS |
1030 | */ |
1031 | static int check_qop(struct sem_array *sma, int semnum, struct sem_queue *q, | |
1032 | bool count_zero) | |
1033 | { | |
b220c57a | 1034 | struct sembuf *sop = q->blocking; |
2f2ed41d | 1035 | |
9b44ee2e MS |
1036 | /* |
1037 | * Linux always (since 0.99.10) reported a task as sleeping on all | |
1038 | * semaphores. This violates SUS, therefore it was changed to the | |
1039 | * standard compliant behavior. | |
1040 | * Give the administrators a chance to notice that an application | |
1041 | * might misbehave because it relies on the Linux behavior. | |
1042 | */ | |
1043 | pr_info_once("semctl(GETNCNT/GETZCNT) is since 3.16 Single Unix Specification compliant.\n" | |
1044 | "The task %s (%d) triggered the difference, watch for misbehavior.\n", | |
1045 | current->comm, task_pid_nr(current)); | |
1046 | ||
b220c57a MS |
1047 | if (sop->sem_num != semnum) |
1048 | return 0; | |
2f2ed41d | 1049 | |
b220c57a MS |
1050 | if (count_zero && sop->sem_op == 0) |
1051 | return 1; | |
1052 | if (!count_zero && sop->sem_op < 0) | |
1053 | return 1; | |
1054 | ||
1055 | return 0; | |
2f2ed41d MS |
1056 | } |
1057 | ||
1da177e4 LT |
1058 | /* The following counts are associated to each semaphore: |
1059 | * semncnt number of tasks waiting on semval being nonzero | |
1060 | * semzcnt number of tasks waiting on semval being zero | |
b220c57a MS |
1061 | * |
1062 | * Per definition, a task waits only on the semaphore of the first semop | |
1063 | * that cannot proceed, even if additional operation would block, too. | |
1da177e4 | 1064 | */ |
2f2ed41d MS |
1065 | static int count_semcnt(struct sem_array *sma, ushort semnum, |
1066 | bool count_zero) | |
1da177e4 | 1067 | { |
2f2ed41d | 1068 | struct list_head *l; |
239521f3 | 1069 | struct sem_queue *q; |
2f2ed41d | 1070 | int semcnt; |
1da177e4 | 1071 | |
2f2ed41d MS |
1072 | semcnt = 0; |
1073 | /* First: check the simple operations. They are easy to evaluate */ | |
1074 | if (count_zero) | |
1075 | l = &sma->sem_base[semnum].pending_const; | |
1076 | else | |
1077 | l = &sma->sem_base[semnum].pending_alter; | |
1da177e4 | 1078 | |
2f2ed41d MS |
1079 | list_for_each_entry(q, l, list) { |
1080 | /* all task on a per-semaphore list sleep on exactly | |
1081 | * that semaphore | |
1082 | */ | |
1083 | semcnt++; | |
ebc2e5e6 RR |
1084 | } |
1085 | ||
2f2ed41d | 1086 | /* Then: check the complex operations. */ |
1994862d | 1087 | list_for_each_entry(q, &sma->pending_alter, list) { |
2f2ed41d MS |
1088 | semcnt += check_qop(sma, semnum, q, count_zero); |
1089 | } | |
1090 | if (count_zero) { | |
1091 | list_for_each_entry(q, &sma->pending_const, list) { | |
1092 | semcnt += check_qop(sma, semnum, q, count_zero); | |
1093 | } | |
1994862d | 1094 | } |
2f2ed41d | 1095 | return semcnt; |
1da177e4 LT |
1096 | } |
1097 | ||
d9a605e4 DB |
1098 | /* Free a semaphore set. freeary() is called with sem_ids.rwsem locked |
1099 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rwsem | |
3e148c79 | 1100 | * remains locked on exit. |
1da177e4 | 1101 | */ |
01b8b07a | 1102 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 1103 | { |
380af1b3 MS |
1104 | struct sem_undo *un, *tu; |
1105 | struct sem_queue *q, *tq; | |
01b8b07a | 1106 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
9f1bc2c9 | 1107 | int i; |
9ae949fa | 1108 | DEFINE_WAKE_Q(wake_q); |
1da177e4 | 1109 | |
380af1b3 | 1110 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 1111 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
1112 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
1113 | list_del(&un->list_id); | |
1114 | spin_lock(&un->ulp->lock); | |
1da177e4 | 1115 | un->semid = -1; |
380af1b3 MS |
1116 | list_del_rcu(&un->list_proc); |
1117 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 1118 | kfree_rcu(un, rcu); |
380af1b3 | 1119 | } |
1da177e4 LT |
1120 | |
1121 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
1a82e9e1 MS |
1122 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
1123 | unlink_queue(sma, q); | |
9ae949fa | 1124 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1125 | } |
1126 | ||
1127 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 1128 | unlink_queue(sma, q); |
9ae949fa | 1129 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1da177e4 | 1130 | } |
9f1bc2c9 RR |
1131 | for (i = 0; i < sma->sem_nsems; i++) { |
1132 | struct sem *sem = sma->sem_base + i; | |
1a82e9e1 MS |
1133 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
1134 | unlink_queue(sma, q); | |
9ae949fa | 1135 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
1a82e9e1 MS |
1136 | } |
1137 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 | 1138 | unlink_queue(sma, q); |
9ae949fa | 1139 | wake_up_sem_queue_prepare(q, -EIDRM, &wake_q); |
9f1bc2c9 RR |
1140 | } |
1141 | } | |
1da177e4 | 1142 | |
7ca7e564 ND |
1143 | /* Remove the semaphore set from the IDR */ |
1144 | sem_rmid(ns, sma); | |
6062a8dc | 1145 | sem_unlock(sma, -1); |
6d49dab8 | 1146 | rcu_read_unlock(); |
1da177e4 | 1147 | |
9ae949fa | 1148 | wake_up_q(&wake_q); |
e3893534 | 1149 | ns->used_sems -= sma->sem_nsems; |
53dad6d3 | 1150 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1151 | } |
1152 | ||
1153 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1154 | { | |
239521f3 | 1155 | switch (version) { |
1da177e4 LT |
1156 | case IPC_64: |
1157 | return copy_to_user(buf, in, sizeof(*in)); | |
1158 | case IPC_OLD: | |
1159 | { | |
1160 | struct semid_ds out; | |
1161 | ||
982f7c2b DR |
1162 | memset(&out, 0, sizeof(out)); |
1163 | ||
1da177e4 LT |
1164 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1165 | ||
1166 | out.sem_otime = in->sem_otime; | |
1167 | out.sem_ctime = in->sem_ctime; | |
1168 | out.sem_nsems = in->sem_nsems; | |
1169 | ||
1170 | return copy_to_user(buf, &out, sizeof(out)); | |
1171 | } | |
1172 | default: | |
1173 | return -EINVAL; | |
1174 | } | |
1175 | } | |
1176 | ||
d12e1e50 MS |
1177 | static time_t get_semotime(struct sem_array *sma) |
1178 | { | |
1179 | int i; | |
1180 | time_t res; | |
1181 | ||
1182 | res = sma->sem_base[0].sem_otime; | |
1183 | for (i = 1; i < sma->sem_nsems; i++) { | |
1184 | time_t to = sma->sem_base[i].sem_otime; | |
1185 | ||
1186 | if (to > res) | |
1187 | res = to; | |
1188 | } | |
1189 | return res; | |
1190 | } | |
1191 | ||
4b9fcb0e | 1192 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1193 | int cmd, int version, void __user *p) |
1da177e4 | 1194 | { |
e5cc9c7b | 1195 | int err; |
1da177e4 LT |
1196 | struct sem_array *sma; |
1197 | ||
239521f3 | 1198 | switch (cmd) { |
1da177e4 LT |
1199 | case IPC_INFO: |
1200 | case SEM_INFO: | |
1201 | { | |
1202 | struct seminfo seminfo; | |
1203 | int max_id; | |
1204 | ||
1205 | err = security_sem_semctl(NULL, cmd); | |
1206 | if (err) | |
1207 | return err; | |
46c0a8ca | 1208 | |
239521f3 | 1209 | memset(&seminfo, 0, sizeof(seminfo)); |
e3893534 KK |
1210 | seminfo.semmni = ns->sc_semmni; |
1211 | seminfo.semmns = ns->sc_semmns; | |
1212 | seminfo.semmsl = ns->sc_semmsl; | |
1213 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1214 | seminfo.semvmx = SEMVMX; |
1215 | seminfo.semmnu = SEMMNU; | |
1216 | seminfo.semmap = SEMMAP; | |
1217 | seminfo.semume = SEMUME; | |
d9a605e4 | 1218 | down_read(&sem_ids(ns).rwsem); |
1da177e4 | 1219 | if (cmd == SEM_INFO) { |
e3893534 KK |
1220 | seminfo.semusz = sem_ids(ns).in_use; |
1221 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1222 | } else { |
1223 | seminfo.semusz = SEMUSZ; | |
1224 | seminfo.semaem = SEMAEM; | |
1225 | } | |
7ca7e564 | 1226 | max_id = ipc_get_maxid(&sem_ids(ns)); |
d9a605e4 | 1227 | up_read(&sem_ids(ns).rwsem); |
46c0a8ca | 1228 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 | 1229 | return -EFAULT; |
239521f3 | 1230 | return (max_id < 0) ? 0 : max_id; |
1da177e4 | 1231 | } |
4b9fcb0e | 1232 | case IPC_STAT: |
1da177e4 LT |
1233 | case SEM_STAT: |
1234 | { | |
1235 | struct semid64_ds tbuf; | |
16df3674 DB |
1236 | int id = 0; |
1237 | ||
1238 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1239 | |
941b0304 | 1240 | rcu_read_lock(); |
4b9fcb0e | 1241 | if (cmd == SEM_STAT) { |
16df3674 DB |
1242 | sma = sem_obtain_object(ns, semid); |
1243 | if (IS_ERR(sma)) { | |
1244 | err = PTR_ERR(sma); | |
1245 | goto out_unlock; | |
1246 | } | |
4b9fcb0e PP |
1247 | id = sma->sem_perm.id; |
1248 | } else { | |
16df3674 DB |
1249 | sma = sem_obtain_object_check(ns, semid); |
1250 | if (IS_ERR(sma)) { | |
1251 | err = PTR_ERR(sma); | |
1252 | goto out_unlock; | |
1253 | } | |
4b9fcb0e | 1254 | } |
1da177e4 LT |
1255 | |
1256 | err = -EACCES; | |
b0e77598 | 1257 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1258 | goto out_unlock; |
1259 | ||
1260 | err = security_sem_semctl(sma, cmd); | |
1261 | if (err) | |
1262 | goto out_unlock; | |
1263 | ||
1da177e4 | 1264 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
d12e1e50 MS |
1265 | tbuf.sem_otime = get_semotime(sma); |
1266 | tbuf.sem_ctime = sma->sem_ctime; | |
1267 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1268 | rcu_read_unlock(); |
e1fd1f49 | 1269 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1270 | return -EFAULT; |
1271 | return id; | |
1272 | } | |
1273 | default: | |
1274 | return -EINVAL; | |
1275 | } | |
1da177e4 | 1276 | out_unlock: |
16df3674 | 1277 | rcu_read_unlock(); |
1da177e4 LT |
1278 | return err; |
1279 | } | |
1280 | ||
e1fd1f49 AV |
1281 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1282 | unsigned long arg) | |
1283 | { | |
1284 | struct sem_undo *un; | |
1285 | struct sem_array *sma; | |
239521f3 | 1286 | struct sem *curr; |
9ae949fa DB |
1287 | int err, val; |
1288 | DEFINE_WAKE_Q(wake_q); | |
1289 | ||
e1fd1f49 AV |
1290 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) |
1291 | /* big-endian 64bit */ | |
1292 | val = arg >> 32; | |
1293 | #else | |
1294 | /* 32bit or little-endian 64bit */ | |
1295 | val = arg; | |
1296 | #endif | |
1297 | ||
6062a8dc RR |
1298 | if (val > SEMVMX || val < 0) |
1299 | return -ERANGE; | |
e1fd1f49 | 1300 | |
6062a8dc RR |
1301 | rcu_read_lock(); |
1302 | sma = sem_obtain_object_check(ns, semid); | |
1303 | if (IS_ERR(sma)) { | |
1304 | rcu_read_unlock(); | |
1305 | return PTR_ERR(sma); | |
1306 | } | |
1307 | ||
1308 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1309 | rcu_read_unlock(); | |
1310 | return -EINVAL; | |
1311 | } | |
1312 | ||
1313 | ||
1314 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1315 | rcu_read_unlock(); | |
1316 | return -EACCES; | |
1317 | } | |
e1fd1f49 AV |
1318 | |
1319 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1320 | if (err) { |
1321 | rcu_read_unlock(); | |
1322 | return -EACCES; | |
1323 | } | |
e1fd1f49 | 1324 | |
6062a8dc | 1325 | sem_lock(sma, NULL, -1); |
e1fd1f49 | 1326 | |
0f3d2b01 | 1327 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1328 | sem_unlock(sma, -1); |
1329 | rcu_read_unlock(); | |
1330 | return -EIDRM; | |
1331 | } | |
1332 | ||
e1fd1f49 AV |
1333 | curr = &sma->sem_base[semnum]; |
1334 | ||
cf9d5d78 | 1335 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1336 | list_for_each_entry(un, &sma->list_id, list_id) |
1337 | un->semadj[semnum] = 0; | |
1338 | ||
1339 | curr->semval = val; | |
1340 | curr->sempid = task_tgid_vnr(current); | |
1341 | sma->sem_ctime = get_seconds(); | |
1342 | /* maybe some queued-up processes were waiting for this */ | |
9ae949fa | 1343 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
6062a8dc | 1344 | sem_unlock(sma, -1); |
6d49dab8 | 1345 | rcu_read_unlock(); |
9ae949fa | 1346 | wake_up_q(&wake_q); |
6062a8dc | 1347 | return 0; |
e1fd1f49 AV |
1348 | } |
1349 | ||
e3893534 | 1350 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1351 | int cmd, void __user *p) |
1da177e4 LT |
1352 | { |
1353 | struct sem_array *sma; | |
239521f3 | 1354 | struct sem *curr; |
16df3674 | 1355 | int err, nsems; |
1da177e4 | 1356 | ushort fast_sem_io[SEMMSL_FAST]; |
239521f3 | 1357 | ushort *sem_io = fast_sem_io; |
9ae949fa | 1358 | DEFINE_WAKE_Q(wake_q); |
16df3674 DB |
1359 | |
1360 | rcu_read_lock(); | |
1361 | sma = sem_obtain_object_check(ns, semid); | |
1362 | if (IS_ERR(sma)) { | |
1363 | rcu_read_unlock(); | |
023a5355 | 1364 | return PTR_ERR(sma); |
16df3674 | 1365 | } |
1da177e4 LT |
1366 | |
1367 | nsems = sma->sem_nsems; | |
1368 | ||
1da177e4 | 1369 | err = -EACCES; |
c728b9c8 LT |
1370 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1371 | goto out_rcu_wakeup; | |
1da177e4 LT |
1372 | |
1373 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1374 | if (err) |
1375 | goto out_rcu_wakeup; | |
1da177e4 LT |
1376 | |
1377 | err = -EACCES; | |
1378 | switch (cmd) { | |
1379 | case GETALL: | |
1380 | { | |
e1fd1f49 | 1381 | ushort __user *array = p; |
1da177e4 LT |
1382 | int i; |
1383 | ||
ce857229 | 1384 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1385 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1386 | err = -EIDRM; |
1387 | goto out_unlock; | |
1388 | } | |
239521f3 | 1389 | if (nsems > SEMMSL_FAST) { |
ce857229 | 1390 | if (!ipc_rcu_getref(sma)) { |
ce857229 | 1391 | err = -EIDRM; |
6e224f94 | 1392 | goto out_unlock; |
ce857229 AV |
1393 | } |
1394 | sem_unlock(sma, -1); | |
6d49dab8 | 1395 | rcu_read_unlock(); |
1da177e4 | 1396 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1397 | if (sem_io == NULL) { |
9b24fef9 | 1398 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1399 | return -ENOMEM; |
1400 | } | |
1401 | ||
4091fd94 | 1402 | rcu_read_lock(); |
6ff37972 | 1403 | sem_lock_and_putref(sma); |
0f3d2b01 | 1404 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1405 | err = -EIDRM; |
6e224f94 | 1406 | goto out_unlock; |
1da177e4 | 1407 | } |
ce857229 | 1408 | } |
1da177e4 LT |
1409 | for (i = 0; i < sma->sem_nsems; i++) |
1410 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1411 | sem_unlock(sma, -1); |
6d49dab8 | 1412 | rcu_read_unlock(); |
1da177e4 | 1413 | err = 0; |
239521f3 | 1414 | if (copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
1da177e4 LT |
1415 | err = -EFAULT; |
1416 | goto out_free; | |
1417 | } | |
1418 | case SETALL: | |
1419 | { | |
1420 | int i; | |
1421 | struct sem_undo *un; | |
1422 | ||
6062a8dc | 1423 | if (!ipc_rcu_getref(sma)) { |
6e224f94 MS |
1424 | err = -EIDRM; |
1425 | goto out_rcu_wakeup; | |
6062a8dc | 1426 | } |
16df3674 | 1427 | rcu_read_unlock(); |
1da177e4 | 1428 | |
239521f3 | 1429 | if (nsems > SEMMSL_FAST) { |
1da177e4 | 1430 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
239521f3 | 1431 | if (sem_io == NULL) { |
9b24fef9 | 1432 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1433 | return -ENOMEM; |
1434 | } | |
1435 | } | |
1436 | ||
239521f3 | 1437 | if (copy_from_user(sem_io, p, nsems*sizeof(ushort))) { |
9b24fef9 | 1438 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1439 | err = -EFAULT; |
1440 | goto out_free; | |
1441 | } | |
1442 | ||
1443 | for (i = 0; i < nsems; i++) { | |
1444 | if (sem_io[i] > SEMVMX) { | |
9b24fef9 | 1445 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1446 | err = -ERANGE; |
1447 | goto out_free; | |
1448 | } | |
1449 | } | |
4091fd94 | 1450 | rcu_read_lock(); |
6ff37972 | 1451 | sem_lock_and_putref(sma); |
0f3d2b01 | 1452 | if (!ipc_valid_object(&sma->sem_perm)) { |
1da177e4 | 1453 | err = -EIDRM; |
6e224f94 | 1454 | goto out_unlock; |
1da177e4 LT |
1455 | } |
1456 | ||
a5f4db87 | 1457 | for (i = 0; i < nsems; i++) { |
1da177e4 | 1458 | sma->sem_base[i].semval = sem_io[i]; |
a5f4db87 DB |
1459 | sma->sem_base[i].sempid = task_tgid_vnr(current); |
1460 | } | |
4daa28f6 | 1461 | |
cf9d5d78 | 1462 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1463 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1464 | for (i = 0; i < nsems; i++) |
1465 | un->semadj[i] = 0; | |
4daa28f6 | 1466 | } |
1da177e4 LT |
1467 | sma->sem_ctime = get_seconds(); |
1468 | /* maybe some queued-up processes were waiting for this */ | |
9ae949fa | 1469 | do_smart_update(sma, NULL, 0, 0, &wake_q); |
1da177e4 LT |
1470 | err = 0; |
1471 | goto out_unlock; | |
1472 | } | |
e1fd1f49 | 1473 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1474 | } |
1475 | err = -EINVAL; | |
c728b9c8 LT |
1476 | if (semnum < 0 || semnum >= nsems) |
1477 | goto out_rcu_wakeup; | |
1da177e4 | 1478 | |
6062a8dc | 1479 | sem_lock(sma, NULL, -1); |
0f3d2b01 | 1480 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
1481 | err = -EIDRM; |
1482 | goto out_unlock; | |
1483 | } | |
1da177e4 LT |
1484 | curr = &sma->sem_base[semnum]; |
1485 | ||
1486 | switch (cmd) { | |
1487 | case GETVAL: | |
1488 | err = curr->semval; | |
1489 | goto out_unlock; | |
1490 | case GETPID: | |
1491 | err = curr->sempid; | |
1492 | goto out_unlock; | |
1493 | case GETNCNT: | |
2f2ed41d | 1494 | err = count_semcnt(sma, semnum, 0); |
1da177e4 LT |
1495 | goto out_unlock; |
1496 | case GETZCNT: | |
2f2ed41d | 1497 | err = count_semcnt(sma, semnum, 1); |
1da177e4 | 1498 | goto out_unlock; |
1da177e4 | 1499 | } |
16df3674 | 1500 | |
1da177e4 | 1501 | out_unlock: |
6062a8dc | 1502 | sem_unlock(sma, -1); |
c728b9c8 | 1503 | out_rcu_wakeup: |
6d49dab8 | 1504 | rcu_read_unlock(); |
9ae949fa | 1505 | wake_up_q(&wake_q); |
1da177e4 | 1506 | out_free: |
239521f3 | 1507 | if (sem_io != fast_sem_io) |
1d5cfdb0 | 1508 | ipc_free(sem_io); |
1da177e4 LT |
1509 | return err; |
1510 | } | |
1511 | ||
016d7132 PP |
1512 | static inline unsigned long |
1513 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 | 1514 | { |
239521f3 | 1515 | switch (version) { |
1da177e4 | 1516 | case IPC_64: |
016d7132 | 1517 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1518 | return -EFAULT; |
1da177e4 | 1519 | return 0; |
1da177e4 LT |
1520 | case IPC_OLD: |
1521 | { | |
1522 | struct semid_ds tbuf_old; | |
1523 | ||
239521f3 | 1524 | if (copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
1da177e4 LT |
1525 | return -EFAULT; |
1526 | ||
016d7132 PP |
1527 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1528 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1529 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1530 | |
1531 | return 0; | |
1532 | } | |
1533 | default: | |
1534 | return -EINVAL; | |
1535 | } | |
1536 | } | |
1537 | ||
522bb2a2 | 1538 | /* |
d9a605e4 | 1539 | * This function handles some semctl commands which require the rwsem |
522bb2a2 | 1540 | * to be held in write mode. |
d9a605e4 | 1541 | * NOTE: no locks must be held, the rwsem is taken inside this function. |
522bb2a2 | 1542 | */ |
21a4826a | 1543 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1544 | int cmd, int version, void __user *p) |
1da177e4 LT |
1545 | { |
1546 | struct sem_array *sma; | |
1547 | int err; | |
016d7132 | 1548 | struct semid64_ds semid64; |
1da177e4 LT |
1549 | struct kern_ipc_perm *ipcp; |
1550 | ||
239521f3 | 1551 | if (cmd == IPC_SET) { |
e1fd1f49 | 1552 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1553 | return -EFAULT; |
1da177e4 | 1554 | } |
073115d6 | 1555 | |
d9a605e4 | 1556 | down_write(&sem_ids(ns).rwsem); |
7b4cc5d8 DB |
1557 | rcu_read_lock(); |
1558 | ||
16df3674 DB |
1559 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1560 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1561 | if (IS_ERR(ipcp)) { |
1562 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1563 | goto out_unlock1; |
1564 | } | |
073115d6 | 1565 | |
a5f75e7f | 1566 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1567 | |
1568 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1569 | if (err) |
1570 | goto out_unlock1; | |
1da177e4 | 1571 | |
7b4cc5d8 | 1572 | switch (cmd) { |
1da177e4 | 1573 | case IPC_RMID: |
6062a8dc | 1574 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1575 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1576 | freeary(ns, ipcp); |
522bb2a2 | 1577 | goto out_up; |
1da177e4 | 1578 | case IPC_SET: |
6062a8dc | 1579 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1580 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1581 | if (err) | |
7b4cc5d8 | 1582 | goto out_unlock0; |
1da177e4 | 1583 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1584 | break; |
1585 | default: | |
1da177e4 | 1586 | err = -EINVAL; |
7b4cc5d8 | 1587 | goto out_unlock1; |
1da177e4 | 1588 | } |
1da177e4 | 1589 | |
7b4cc5d8 | 1590 | out_unlock0: |
6062a8dc | 1591 | sem_unlock(sma, -1); |
7b4cc5d8 | 1592 | out_unlock1: |
6d49dab8 | 1593 | rcu_read_unlock(); |
522bb2a2 | 1594 | out_up: |
d9a605e4 | 1595 | up_write(&sem_ids(ns).rwsem); |
1da177e4 LT |
1596 | return err; |
1597 | } | |
1598 | ||
e1fd1f49 | 1599 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1600 | { |
1da177e4 | 1601 | int version; |
e3893534 | 1602 | struct ipc_namespace *ns; |
e1fd1f49 | 1603 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1604 | |
1605 | if (semid < 0) | |
1606 | return -EINVAL; | |
1607 | ||
1608 | version = ipc_parse_version(&cmd); | |
e3893534 | 1609 | ns = current->nsproxy->ipc_ns; |
1da177e4 | 1610 | |
239521f3 | 1611 | switch (cmd) { |
1da177e4 LT |
1612 | case IPC_INFO: |
1613 | case SEM_INFO: | |
4b9fcb0e | 1614 | case IPC_STAT: |
1da177e4 | 1615 | case SEM_STAT: |
e1fd1f49 | 1616 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1617 | case GETALL: |
1618 | case GETVAL: | |
1619 | case GETPID: | |
1620 | case GETNCNT: | |
1621 | case GETZCNT: | |
1da177e4 | 1622 | case SETALL: |
e1fd1f49 AV |
1623 | return semctl_main(ns, semid, semnum, cmd, p); |
1624 | case SETVAL: | |
1625 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1626 | case IPC_RMID: |
1627 | case IPC_SET: | |
e1fd1f49 | 1628 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1629 | default: |
1630 | return -EINVAL; | |
1631 | } | |
1632 | } | |
1633 | ||
1da177e4 LT |
1634 | /* If the task doesn't already have a undo_list, then allocate one |
1635 | * here. We guarantee there is only one thread using this undo list, | |
1636 | * and current is THE ONE | |
1637 | * | |
1638 | * If this allocation and assignment succeeds, but later | |
1639 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1640 | * Just let it stay associated with the task, and it'll be freed later | |
1641 | * at exit time. | |
1642 | * | |
1643 | * This can block, so callers must hold no locks. | |
1644 | */ | |
1645 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1646 | { | |
1647 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1648 | |
1649 | undo_list = current->sysvsem.undo_list; | |
1650 | if (!undo_list) { | |
2453a306 | 1651 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1652 | if (undo_list == NULL) |
1653 | return -ENOMEM; | |
00a5dfdb | 1654 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1655 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1656 | INIT_LIST_HEAD(&undo_list->list_proc); |
1657 | ||
1da177e4 LT |
1658 | current->sysvsem.undo_list = undo_list; |
1659 | } | |
1660 | *undo_listp = undo_list; | |
1661 | return 0; | |
1662 | } | |
1663 | ||
bf17bb71 | 1664 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1665 | { |
bf17bb71 | 1666 | struct sem_undo *un; |
4daa28f6 | 1667 | |
bf17bb71 NP |
1668 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1669 | if (un->semid == semid) | |
1670 | return un; | |
1da177e4 | 1671 | } |
4daa28f6 | 1672 | return NULL; |
1da177e4 LT |
1673 | } |
1674 | ||
bf17bb71 NP |
1675 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1676 | { | |
1677 | struct sem_undo *un; | |
1678 | ||
239521f3 | 1679 | assert_spin_locked(&ulp->lock); |
bf17bb71 NP |
1680 | |
1681 | un = __lookup_undo(ulp, semid); | |
1682 | if (un) { | |
1683 | list_del_rcu(&un->list_proc); | |
1684 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1685 | } | |
1686 | return un; | |
1687 | } | |
1688 | ||
4daa28f6 | 1689 | /** |
8001c858 | 1690 | * find_alloc_undo - lookup (and if not present create) undo array |
4daa28f6 MS |
1691 | * @ns: namespace |
1692 | * @semid: semaphore array id | |
1693 | * | |
1694 | * The function looks up (and if not present creates) the undo structure. | |
1695 | * The size of the undo structure depends on the size of the semaphore | |
1696 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1697 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1698 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1699 | */ |
1700 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1701 | { |
1702 | struct sem_array *sma; | |
1703 | struct sem_undo_list *ulp; | |
1704 | struct sem_undo *un, *new; | |
6062a8dc | 1705 | int nsems, error; |
1da177e4 LT |
1706 | |
1707 | error = get_undo_list(&ulp); | |
1708 | if (error) | |
1709 | return ERR_PTR(error); | |
1710 | ||
380af1b3 | 1711 | rcu_read_lock(); |
c530c6ac | 1712 | spin_lock(&ulp->lock); |
1da177e4 | 1713 | un = lookup_undo(ulp, semid); |
c530c6ac | 1714 | spin_unlock(&ulp->lock); |
239521f3 | 1715 | if (likely(un != NULL)) |
1da177e4 LT |
1716 | goto out; |
1717 | ||
1718 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1719 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1720 | sma = sem_obtain_object_check(ns, semid); |
1721 | if (IS_ERR(sma)) { | |
1722 | rcu_read_unlock(); | |
4de85cd6 | 1723 | return ERR_CAST(sma); |
16df3674 | 1724 | } |
023a5355 | 1725 | |
1da177e4 | 1726 | nsems = sma->sem_nsems; |
6062a8dc RR |
1727 | if (!ipc_rcu_getref(sma)) { |
1728 | rcu_read_unlock(); | |
1729 | un = ERR_PTR(-EIDRM); | |
1730 | goto out; | |
1731 | } | |
16df3674 | 1732 | rcu_read_unlock(); |
1da177e4 | 1733 | |
4daa28f6 | 1734 | /* step 2: allocate new undo structure */ |
4668edc3 | 1735 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1736 | if (!new) { |
9b24fef9 | 1737 | ipc_rcu_putref(sma, sem_rcu_free); |
1da177e4 LT |
1738 | return ERR_PTR(-ENOMEM); |
1739 | } | |
1da177e4 | 1740 | |
380af1b3 | 1741 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1742 | rcu_read_lock(); |
6ff37972 | 1743 | sem_lock_and_putref(sma); |
0f3d2b01 | 1744 | if (!ipc_valid_object(&sma->sem_perm)) { |
6062a8dc | 1745 | sem_unlock(sma, -1); |
6d49dab8 | 1746 | rcu_read_unlock(); |
1da177e4 LT |
1747 | kfree(new); |
1748 | un = ERR_PTR(-EIDRM); | |
1749 | goto out; | |
1750 | } | |
380af1b3 MS |
1751 | spin_lock(&ulp->lock); |
1752 | ||
1753 | /* | |
1754 | * step 4: check for races: did someone else allocate the undo struct? | |
1755 | */ | |
1756 | un = lookup_undo(ulp, semid); | |
1757 | if (un) { | |
1758 | kfree(new); | |
1759 | goto success; | |
1760 | } | |
4daa28f6 MS |
1761 | /* step 5: initialize & link new undo structure */ |
1762 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1763 | new->ulp = ulp; |
4daa28f6 MS |
1764 | new->semid = semid; |
1765 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1766 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1767 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1768 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1769 | un = new; |
4daa28f6 | 1770 | |
380af1b3 | 1771 | success: |
c530c6ac | 1772 | spin_unlock(&ulp->lock); |
6062a8dc | 1773 | sem_unlock(sma, -1); |
1da177e4 LT |
1774 | out: |
1775 | return un; | |
1776 | } | |
1777 | ||
d5460c99 HC |
1778 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1779 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1780 | { |
1781 | int error = -EINVAL; | |
1782 | struct sem_array *sma; | |
1783 | struct sembuf fast_sops[SEMOPM_FAST]; | |
239521f3 | 1784 | struct sembuf *sops = fast_sops, *sop; |
1da177e4 | 1785 | struct sem_undo *un; |
4ce33ec2 DB |
1786 | int max, locknum; |
1787 | bool undos = false, alter = false, dupsop = false; | |
1da177e4 | 1788 | struct sem_queue queue; |
4ce33ec2 | 1789 | unsigned long dup = 0, jiffies_left = 0; |
e3893534 KK |
1790 | struct ipc_namespace *ns; |
1791 | ||
1792 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1793 | |
1794 | if (nsops < 1 || semid < 0) | |
1795 | return -EINVAL; | |
e3893534 | 1796 | if (nsops > ns->sc_semopm) |
1da177e4 | 1797 | return -E2BIG; |
239521f3 MS |
1798 | if (nsops > SEMOPM_FAST) { |
1799 | sops = kmalloc(sizeof(*sops)*nsops, GFP_KERNEL); | |
1800 | if (sops == NULL) | |
1da177e4 LT |
1801 | return -ENOMEM; |
1802 | } | |
4ce33ec2 | 1803 | |
239521f3 MS |
1804 | if (copy_from_user(sops, tsops, nsops * sizeof(*tsops))) { |
1805 | error = -EFAULT; | |
1da177e4 LT |
1806 | goto out_free; |
1807 | } | |
4ce33ec2 | 1808 | |
1da177e4 LT |
1809 | if (timeout) { |
1810 | struct timespec _timeout; | |
1811 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1812 | error = -EFAULT; | |
1813 | goto out_free; | |
1814 | } | |
1815 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1816 | _timeout.tv_nsec >= 1000000000L) { | |
1817 | error = -EINVAL; | |
1818 | goto out_free; | |
1819 | } | |
1820 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1821 | } | |
4ce33ec2 | 1822 | |
1da177e4 LT |
1823 | max = 0; |
1824 | for (sop = sops; sop < sops + nsops; sop++) { | |
4ce33ec2 DB |
1825 | unsigned long mask = 1ULL << ((sop->sem_num) % BITS_PER_LONG); |
1826 | ||
1da177e4 LT |
1827 | if (sop->sem_num >= max) |
1828 | max = sop->sem_num; | |
1829 | if (sop->sem_flg & SEM_UNDO) | |
4ce33ec2 DB |
1830 | undos = true; |
1831 | if (dup & mask) { | |
1832 | /* | |
1833 | * There was a previous alter access that appears | |
1834 | * to have accessed the same semaphore, thus use | |
1835 | * the dupsop logic. "appears", because the detection | |
1836 | * can only check % BITS_PER_LONG. | |
1837 | */ | |
1838 | dupsop = true; | |
1839 | } | |
1840 | if (sop->sem_op != 0) { | |
1841 | alter = true; | |
1842 | dup |= mask; | |
1843 | } | |
1da177e4 | 1844 | } |
1da177e4 | 1845 | |
1da177e4 | 1846 | if (undos) { |
6062a8dc | 1847 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1848 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1849 | if (IS_ERR(un)) { |
1850 | error = PTR_ERR(un); | |
1851 | goto out_free; | |
1852 | } | |
6062a8dc | 1853 | } else { |
1da177e4 | 1854 | un = NULL; |
6062a8dc RR |
1855 | rcu_read_lock(); |
1856 | } | |
1da177e4 | 1857 | |
16df3674 | 1858 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1859 | if (IS_ERR(sma)) { |
6062a8dc | 1860 | rcu_read_unlock(); |
023a5355 | 1861 | error = PTR_ERR(sma); |
1da177e4 | 1862 | goto out_free; |
023a5355 ND |
1863 | } |
1864 | ||
16df3674 | 1865 | error = -EFBIG; |
248e7357 DB |
1866 | if (max >= sma->sem_nsems) { |
1867 | rcu_read_unlock(); | |
1868 | goto out_free; | |
1869 | } | |
16df3674 DB |
1870 | |
1871 | error = -EACCES; | |
248e7357 DB |
1872 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) { |
1873 | rcu_read_unlock(); | |
1874 | goto out_free; | |
1875 | } | |
16df3674 DB |
1876 | |
1877 | error = security_sem_semop(sma, sops, nsops, alter); | |
248e7357 DB |
1878 | if (error) { |
1879 | rcu_read_unlock(); | |
1880 | goto out_free; | |
1881 | } | |
16df3674 | 1882 | |
6e224f94 MS |
1883 | error = -EIDRM; |
1884 | locknum = sem_lock(sma, sops, nsops); | |
0f3d2b01 RA |
1885 | /* |
1886 | * We eventually might perform the following check in a lockless | |
1887 | * fashion, considering ipc_valid_object() locking constraints. | |
1888 | * If nsops == 1 and there is no contention for sem_perm.lock, then | |
1889 | * only a per-semaphore lock is held and it's OK to proceed with the | |
1890 | * check below. More details on the fine grained locking scheme | |
1891 | * entangled here and why it's RMID race safe on comments at sem_lock() | |
1892 | */ | |
1893 | if (!ipc_valid_object(&sma->sem_perm)) | |
6e224f94 | 1894 | goto out_unlock_free; |
1da177e4 | 1895 | /* |
4daa28f6 | 1896 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1897 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1898 | * and now a new array with received the same id. Check and fail. |
25985edc | 1899 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1900 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1901 | */ |
6062a8dc RR |
1902 | if (un && un->semid == -1) |
1903 | goto out_unlock_free; | |
4daa28f6 | 1904 | |
d198cd6d MS |
1905 | queue.sops = sops; |
1906 | queue.nsops = nsops; | |
1907 | queue.undo = un; | |
1908 | queue.pid = task_tgid_vnr(current); | |
1909 | queue.alter = alter; | |
4ce33ec2 | 1910 | queue.dupsop = dupsop; |
d198cd6d MS |
1911 | |
1912 | error = perform_atomic_semop(sma, &queue); | |
9ae949fa DB |
1913 | if (error == 0) { /* non-blocking succesfull path */ |
1914 | DEFINE_WAKE_Q(wake_q); | |
1915 | ||
1916 | /* | |
1917 | * If the operation was successful, then do | |
0e8c6656 MS |
1918 | * the required updates. |
1919 | */ | |
1920 | if (alter) | |
9ae949fa | 1921 | do_smart_update(sma, sops, nsops, 1, &wake_q); |
0e8c6656 MS |
1922 | else |
1923 | set_semotime(sma, sops); | |
9ae949fa DB |
1924 | |
1925 | sem_unlock(sma, locknum); | |
1926 | rcu_read_unlock(); | |
1927 | wake_up_q(&wake_q); | |
1928 | ||
1929 | goto out_free; | |
1da177e4 | 1930 | } |
9ae949fa | 1931 | if (error < 0) /* non-blocking error path */ |
0e8c6656 | 1932 | goto out_unlock_free; |
1da177e4 | 1933 | |
9ae949fa DB |
1934 | /* |
1935 | * We need to sleep on this operation, so we put the current | |
1da177e4 LT |
1936 | * task into the pending queue and go to sleep. |
1937 | */ | |
b97e820f MS |
1938 | if (nsops == 1) { |
1939 | struct sem *curr; | |
1940 | curr = &sma->sem_base[sops->sem_num]; | |
1941 | ||
f269f40a MS |
1942 | if (alter) { |
1943 | if (sma->complex_count) { | |
1944 | list_add_tail(&queue.list, | |
1945 | &sma->pending_alter); | |
1946 | } else { | |
1947 | ||
1948 | list_add_tail(&queue.list, | |
1949 | &curr->pending_alter); | |
1950 | } | |
1951 | } else { | |
1a82e9e1 | 1952 | list_add_tail(&queue.list, &curr->pending_const); |
f269f40a | 1953 | } |
b97e820f | 1954 | } else { |
f269f40a MS |
1955 | if (!sma->complex_count) |
1956 | merge_queues(sma); | |
1957 | ||
9f1bc2c9 | 1958 | if (alter) |
1a82e9e1 | 1959 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1960 | else |
1a82e9e1 MS |
1961 | list_add_tail(&queue.list, &sma->pending_const); |
1962 | ||
b97e820f MS |
1963 | sma->complex_count++; |
1964 | } | |
1965 | ||
b5fa01a2 DB |
1966 | do { |
1967 | queue.status = -EINTR; | |
1968 | queue.sleeper = current; | |
0b0577f6 | 1969 | |
b5fa01a2 DB |
1970 | __set_current_state(TASK_INTERRUPTIBLE); |
1971 | sem_unlock(sma, locknum); | |
1972 | rcu_read_unlock(); | |
1da177e4 | 1973 | |
b5fa01a2 DB |
1974 | if (timeout) |
1975 | jiffies_left = schedule_timeout(jiffies_left); | |
1976 | else | |
1977 | schedule(); | |
1da177e4 | 1978 | |
9ae949fa | 1979 | /* |
b5fa01a2 DB |
1980 | * fastpath: the semop has completed, either successfully or |
1981 | * not, from the syscall pov, is quite irrelevant to us at this | |
1982 | * point; we're done. | |
1983 | * | |
1984 | * We _do_ care, nonetheless, about being awoken by a signal or | |
1985 | * spuriously. The queue.status is checked again in the | |
1986 | * slowpath (aka after taking sem_lock), such that we can detect | |
1987 | * scenarios where we were awakened externally, during the | |
1988 | * window between wake_q_add() and wake_up_q(). | |
c61284e9 | 1989 | */ |
b5fa01a2 DB |
1990 | error = READ_ONCE(queue.status); |
1991 | if (error != -EINTR) { | |
1992 | /* | |
1993 | * User space could assume that semop() is a memory | |
1994 | * barrier: Without the mb(), the cpu could | |
1995 | * speculatively read in userspace stale data that was | |
1996 | * overwritten by the previous owner of the semaphore. | |
1997 | */ | |
1998 | smp_mb(); | |
1999 | goto out_free; | |
2000 | } | |
d694ad62 | 2001 | |
b5fa01a2 DB |
2002 | rcu_read_lock(); |
2003 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); | |
2004 | error = READ_ONCE(queue.status); | |
1da177e4 | 2005 | |
b5fa01a2 DB |
2006 | /* |
2007 | * Array removed? If yes, leave without sem_unlock(). | |
2008 | */ | |
2009 | if (IS_ERR(sma)) { | |
2010 | rcu_read_unlock(); | |
2011 | goto out_free; | |
2012 | } | |
1da177e4 | 2013 | |
b5fa01a2 DB |
2014 | /* |
2015 | * If queue.status != -EINTR we are woken up by another process. | |
2016 | * Leave without unlink_queue(), but with sem_unlock(). | |
2017 | */ | |
2018 | if (error != -EINTR) | |
2019 | goto out_unlock_free; | |
0b0577f6 | 2020 | |
b5fa01a2 DB |
2021 | /* |
2022 | * If an interrupt occurred we have to clean up the queue. | |
2023 | */ | |
2024 | if (timeout && jiffies_left == 0) | |
2025 | error = -EAGAIN; | |
2026 | } while (error == -EINTR && !signal_pending(current)); /* spurious */ | |
0b0577f6 | 2027 | |
b97e820f | 2028 | unlink_queue(sma, &queue); |
1da177e4 LT |
2029 | |
2030 | out_unlock_free: | |
6062a8dc | 2031 | sem_unlock(sma, locknum); |
6d49dab8 | 2032 | rcu_read_unlock(); |
1da177e4 | 2033 | out_free: |
239521f3 | 2034 | if (sops != fast_sops) |
1da177e4 LT |
2035 | kfree(sops); |
2036 | return error; | |
2037 | } | |
2038 | ||
d5460c99 HC |
2039 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
2040 | unsigned, nsops) | |
1da177e4 LT |
2041 | { |
2042 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
2043 | } | |
2044 | ||
2045 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
2046 | * parent and child tasks. | |
1da177e4 LT |
2047 | */ |
2048 | ||
2049 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
2050 | { | |
2051 | struct sem_undo_list *undo_list; | |
2052 | int error; | |
2053 | ||
2054 | if (clone_flags & CLONE_SYSVSEM) { | |
2055 | error = get_undo_list(&undo_list); | |
2056 | if (error) | |
2057 | return error; | |
1da177e4 LT |
2058 | atomic_inc(&undo_list->refcnt); |
2059 | tsk->sysvsem.undo_list = undo_list; | |
46c0a8ca | 2060 | } else |
1da177e4 LT |
2061 | tsk->sysvsem.undo_list = NULL; |
2062 | ||
2063 | return 0; | |
2064 | } | |
2065 | ||
2066 | /* | |
2067 | * add semadj values to semaphores, free undo structures. | |
2068 | * undo structures are not freed when semaphore arrays are destroyed | |
2069 | * so some of them may be out of date. | |
2070 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
2071 | * set of adjustments that needs to be done should be done in an atomic | |
2072 | * manner or not. That is, if we are attempting to decrement the semval | |
2073 | * should we queue up and wait until we can do so legally? | |
2074 | * The original implementation attempted to do this (queue and wait). | |
2075 | * The current implementation does not do so. The POSIX standard | |
2076 | * and SVID should be consulted to determine what behavior is mandated. | |
2077 | */ | |
2078 | void exit_sem(struct task_struct *tsk) | |
2079 | { | |
4daa28f6 | 2080 | struct sem_undo_list *ulp; |
1da177e4 | 2081 | |
4daa28f6 MS |
2082 | ulp = tsk->sysvsem.undo_list; |
2083 | if (!ulp) | |
1da177e4 | 2084 | return; |
9edff4ab | 2085 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 2086 | |
4daa28f6 | 2087 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
2088 | return; |
2089 | ||
380af1b3 | 2090 | for (;;) { |
1da177e4 | 2091 | struct sem_array *sma; |
380af1b3 | 2092 | struct sem_undo *un; |
6062a8dc | 2093 | int semid, i; |
9ae949fa | 2094 | DEFINE_WAKE_Q(wake_q); |
4daa28f6 | 2095 | |
2a1613a5 NB |
2096 | cond_resched(); |
2097 | ||
380af1b3 | 2098 | rcu_read_lock(); |
05725f7e JP |
2099 | un = list_entry_rcu(ulp->list_proc.next, |
2100 | struct sem_undo, list_proc); | |
602b8593 HK |
2101 | if (&un->list_proc == &ulp->list_proc) { |
2102 | /* | |
2103 | * We must wait for freeary() before freeing this ulp, | |
2104 | * in case we raced with last sem_undo. There is a small | |
2105 | * possibility where we exit while freeary() didn't | |
2106 | * finish unlocking sem_undo_list. | |
2107 | */ | |
2108 | spin_unlock_wait(&ulp->lock); | |
2109 | rcu_read_unlock(); | |
2110 | break; | |
2111 | } | |
2112 | spin_lock(&ulp->lock); | |
2113 | semid = un->semid; | |
2114 | spin_unlock(&ulp->lock); | |
4daa28f6 | 2115 | |
602b8593 | 2116 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2117 | if (semid == -1) { |
2118 | rcu_read_unlock(); | |
602b8593 | 2119 | continue; |
6062a8dc | 2120 | } |
1da177e4 | 2121 | |
602b8593 | 2122 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, semid); |
380af1b3 | 2123 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
2124 | if (IS_ERR(sma)) { |
2125 | rcu_read_unlock(); | |
380af1b3 | 2126 | continue; |
6062a8dc | 2127 | } |
1da177e4 | 2128 | |
6062a8dc | 2129 | sem_lock(sma, NULL, -1); |
6e224f94 | 2130 | /* exit_sem raced with IPC_RMID, nothing to do */ |
0f3d2b01 | 2131 | if (!ipc_valid_object(&sma->sem_perm)) { |
6e224f94 MS |
2132 | sem_unlock(sma, -1); |
2133 | rcu_read_unlock(); | |
2134 | continue; | |
2135 | } | |
bf17bb71 | 2136 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
2137 | if (un == NULL) { |
2138 | /* exit_sem raced with IPC_RMID+semget() that created | |
2139 | * exactly the same semid. Nothing to do. | |
2140 | */ | |
6062a8dc | 2141 | sem_unlock(sma, -1); |
6d49dab8 | 2142 | rcu_read_unlock(); |
380af1b3 MS |
2143 | continue; |
2144 | } | |
2145 | ||
2146 | /* remove un from the linked lists */ | |
cf9d5d78 | 2147 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
2148 | list_del(&un->list_id); |
2149 | ||
a9795584 HK |
2150 | /* we are the last process using this ulp, acquiring ulp->lock |
2151 | * isn't required. Besides that, we are also protected against | |
2152 | * IPC_RMID as we hold sma->sem_perm lock now | |
2153 | */ | |
380af1b3 | 2154 | list_del_rcu(&un->list_proc); |
380af1b3 | 2155 | |
4daa28f6 MS |
2156 | /* perform adjustments registered in un */ |
2157 | for (i = 0; i < sma->sem_nsems; i++) { | |
239521f3 | 2158 | struct sem *semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
2159 | if (un->semadj[i]) { |
2160 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
2161 | /* |
2162 | * Range checks of the new semaphore value, | |
2163 | * not defined by sus: | |
2164 | * - Some unices ignore the undo entirely | |
2165 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
2166 | * - some cap the value (e.g. FreeBSD caps | |
2167 | * at 0, but doesn't enforce SEMVMX) | |
2168 | * | |
2169 | * Linux caps the semaphore value, both at 0 | |
2170 | * and at SEMVMX. | |
2171 | * | |
239521f3 | 2172 | * Manfred <manfred@colorfullife.com> |
1da177e4 | 2173 | */ |
5f921ae9 IM |
2174 | if (semaphore->semval < 0) |
2175 | semaphore->semval = 0; | |
2176 | if (semaphore->semval > SEMVMX) | |
2177 | semaphore->semval = SEMVMX; | |
b488893a | 2178 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
2179 | } |
2180 | } | |
1da177e4 | 2181 | /* maybe some queued-up processes were waiting for this */ |
9ae949fa | 2182 | do_smart_update(sma, NULL, 0, 1, &wake_q); |
6062a8dc | 2183 | sem_unlock(sma, -1); |
6d49dab8 | 2184 | rcu_read_unlock(); |
9ae949fa | 2185 | wake_up_q(&wake_q); |
380af1b3 | 2186 | |
693a8b6e | 2187 | kfree_rcu(un, rcu); |
1da177e4 | 2188 | } |
4daa28f6 | 2189 | kfree(ulp); |
1da177e4 LT |
2190 | } |
2191 | ||
2192 | #ifdef CONFIG_PROC_FS | |
19b4946c | 2193 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 2194 | { |
1efdb69b | 2195 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c | 2196 | struct sem_array *sma = it; |
d12e1e50 MS |
2197 | time_t sem_otime; |
2198 | ||
d8c63376 MS |
2199 | /* |
2200 | * The proc interface isn't aware of sem_lock(), it calls | |
2201 | * ipc_lock_object() directly (in sysvipc_find_ipc). | |
5864a2fd MS |
2202 | * In order to stay compatible with sem_lock(), we must |
2203 | * enter / leave complex_mode. | |
d8c63376 | 2204 | */ |
5864a2fd | 2205 | complexmode_enter(sma); |
d8c63376 | 2206 | |
d12e1e50 | 2207 | sem_otime = get_semotime(sma); |
19b4946c | 2208 | |
7f032d6e JP |
2209 | seq_printf(s, |
2210 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", | |
2211 | sma->sem_perm.key, | |
2212 | sma->sem_perm.id, | |
2213 | sma->sem_perm.mode, | |
2214 | sma->sem_nsems, | |
2215 | from_kuid_munged(user_ns, sma->sem_perm.uid), | |
2216 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
2217 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
2218 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
2219 | sem_otime, | |
2220 | sma->sem_ctime); | |
2221 | ||
5864a2fd MS |
2222 | complexmode_tryleave(sma); |
2223 | ||
7f032d6e | 2224 | return 0; |
1da177e4 LT |
2225 | } |
2226 | #endif |