Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/ipc/sem.c | |
3 | * Copyright (C) 1992 Krishna Balasubramanian | |
4 | * Copyright (C) 1995 Eric Schenk, Bruno Haible | |
5 | * | |
1da177e4 LT |
6 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
7 | * | |
8 | * SMP-threaded, sysctl's added | |
624dffcb | 9 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
1da177e4 | 10 | * Enforced range limit on SEM_UNDO |
046c6884 | 11 | * (c) 2001 Red Hat Inc |
1da177e4 LT |
12 | * Lockless wakeup |
13 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> | |
c5cf6359 MS |
14 | * Further wakeup optimizations, documentation |
15 | * (c) 2010 Manfred Spraul <manfred@colorfullife.com> | |
073115d6 SG |
16 | * |
17 | * support for audit of ipc object properties and permission changes | |
18 | * Dustin Kirkland <dustin.kirkland@us.ibm.com> | |
e3893534 KK |
19 | * |
20 | * namespaces support | |
21 | * OpenVZ, SWsoft Inc. | |
22 | * Pavel Emelianov <xemul@openvz.org> | |
c5cf6359 MS |
23 | * |
24 | * Implementation notes: (May 2010) | |
25 | * This file implements System V semaphores. | |
26 | * | |
27 | * User space visible behavior: | |
28 | * - FIFO ordering for semop() operations (just FIFO, not starvation | |
29 | * protection) | |
30 | * - multiple semaphore operations that alter the same semaphore in | |
31 | * one semop() are handled. | |
32 | * - sem_ctime (time of last semctl()) is updated in the IPC_SET, SETVAL and | |
33 | * SETALL calls. | |
34 | * - two Linux specific semctl() commands: SEM_STAT, SEM_INFO. | |
35 | * - undo adjustments at process exit are limited to 0..SEMVMX. | |
36 | * - namespace are supported. | |
37 | * - SEMMSL, SEMMNS, SEMOPM and SEMMNI can be configured at runtine by writing | |
38 | * to /proc/sys/kernel/sem. | |
39 | * - statistics about the usage are reported in /proc/sysvipc/sem. | |
40 | * | |
41 | * Internals: | |
42 | * - scalability: | |
43 | * - all global variables are read-mostly. | |
44 | * - semop() calls and semctl(RMID) are synchronized by RCU. | |
45 | * - most operations do write operations (actually: spin_lock calls) to | |
46 | * the per-semaphore array structure. | |
47 | * Thus: Perfect SMP scaling between independent semaphore arrays. | |
48 | * If multiple semaphores in one array are used, then cache line | |
49 | * trashing on the semaphore array spinlock will limit the scaling. | |
50 | * - semncnt and semzcnt are calculated on demand in count_semncnt() and | |
51 | * count_semzcnt() | |
52 | * - the task that performs a successful semop() scans the list of all | |
53 | * sleeping tasks and completes any pending operations that can be fulfilled. | |
54 | * Semaphores are actively given to waiting tasks (necessary for FIFO). | |
55 | * (see update_queue()) | |
56 | * - To improve the scalability, the actual wake-up calls are performed after | |
57 | * dropping all locks. (see wake_up_sem_queue_prepare(), | |
58 | * wake_up_sem_queue_do()) | |
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). | |
63 | * - The synchronizations between wake-ups due to a timeout/signal and a | |
64 | * wake-up due to a completed semaphore operation is achieved by using an | |
65 | * intermediate state (IN_WAKEUP). | |
66 | * - UNDO values are stored in an array (one per process and per | |
67 | * semaphore array, lazily allocated). For backwards compatibility, multiple | |
68 | * modes for the UNDO variables are supported (per process, per thread) | |
69 | * (see copy_semundo, CLONE_SYSVSEM) | |
70 | * - There are two lists of the pending operations: a per-array list | |
71 | * and per-semaphore list (stored in the array). This allows to achieve FIFO | |
72 | * ordering without always scanning all pending operations. | |
73 | * The worst-case behavior is nevertheless O(N^2) for N wakeups. | |
1da177e4 LT |
74 | */ |
75 | ||
1da177e4 LT |
76 | #include <linux/slab.h> |
77 | #include <linux/spinlock.h> | |
78 | #include <linux/init.h> | |
79 | #include <linux/proc_fs.h> | |
80 | #include <linux/time.h> | |
1da177e4 LT |
81 | #include <linux/security.h> |
82 | #include <linux/syscalls.h> | |
83 | #include <linux/audit.h> | |
c59ede7b | 84 | #include <linux/capability.h> |
19b4946c | 85 | #include <linux/seq_file.h> |
3e148c79 | 86 | #include <linux/rwsem.h> |
e3893534 | 87 | #include <linux/nsproxy.h> |
ae5e1b22 | 88 | #include <linux/ipc_namespace.h> |
5f921ae9 | 89 | |
1da177e4 LT |
90 | #include <asm/uaccess.h> |
91 | #include "util.h" | |
92 | ||
e57940d7 MS |
93 | /* One semaphore structure for each semaphore in the system. */ |
94 | struct sem { | |
95 | int semval; /* current value */ | |
96 | int sempid; /* pid of last operation */ | |
6062a8dc | 97 | spinlock_t lock; /* spinlock for fine-grained semtimedop */ |
e57940d7 MS |
98 | struct list_head sem_pending; /* pending single-sop operations */ |
99 | }; | |
100 | ||
101 | /* One queue for each sleeping process in the system. */ | |
102 | struct sem_queue { | |
e57940d7 MS |
103 | struct list_head list; /* queue of pending operations */ |
104 | struct task_struct *sleeper; /* this process */ | |
105 | struct sem_undo *undo; /* undo structure */ | |
106 | int pid; /* process id of requesting process */ | |
107 | int status; /* completion status of operation */ | |
108 | struct sembuf *sops; /* array of pending operations */ | |
109 | int nsops; /* number of operations */ | |
110 | int alter; /* does *sops alter the array? */ | |
111 | }; | |
112 | ||
113 | /* Each task has a list of undo requests. They are executed automatically | |
114 | * when the process exits. | |
115 | */ | |
116 | struct sem_undo { | |
117 | struct list_head list_proc; /* per-process list: * | |
118 | * all undos from one process | |
119 | * rcu protected */ | |
120 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
121 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
122 | struct list_head list_id; /* per semaphore array list: | |
123 | * all undos for one array */ | |
124 | int semid; /* semaphore set identifier */ | |
125 | short *semadj; /* array of adjustments */ | |
126 | /* one per semaphore */ | |
127 | }; | |
128 | ||
129 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
130 | * that may be shared among all a CLONE_SYSVSEM task group. | |
131 | */ | |
132 | struct sem_undo_list { | |
133 | atomic_t refcnt; | |
134 | spinlock_t lock; | |
135 | struct list_head list_proc; | |
136 | }; | |
137 | ||
138 | ||
ed2ddbf8 | 139 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 140 | |
1b531f21 | 141 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 142 | |
7748dbfa | 143 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 144 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 145 | #ifdef CONFIG_PROC_FS |
19b4946c | 146 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
147 | #endif |
148 | ||
149 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
150 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
151 | ||
152 | /* | |
153 | * linked list protection: | |
154 | * sem_undo.id_next, | |
155 | * sem_array.sem_pending{,last}, | |
156 | * sem_array.sem_undo: sem_lock() for read/write | |
157 | * sem_undo.proc_next: only "current" is allowed to read/write that field. | |
158 | * | |
159 | */ | |
160 | ||
e3893534 KK |
161 | #define sc_semmsl sem_ctls[0] |
162 | #define sc_semmns sem_ctls[1] | |
163 | #define sc_semopm sem_ctls[2] | |
164 | #define sc_semmni sem_ctls[3] | |
165 | ||
ed2ddbf8 | 166 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 167 | { |
e3893534 KK |
168 | ns->sc_semmsl = SEMMSL; |
169 | ns->sc_semmns = SEMMNS; | |
170 | ns->sc_semopm = SEMOPM; | |
171 | ns->sc_semmni = SEMMNI; | |
172 | ns->used_sems = 0; | |
ed2ddbf8 | 173 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
174 | } |
175 | ||
ae5e1b22 | 176 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
177 | void sem_exit_ns(struct ipc_namespace *ns) |
178 | { | |
01b8b07a | 179 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 180 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 181 | } |
ae5e1b22 | 182 | #endif |
1da177e4 LT |
183 | |
184 | void __init sem_init (void) | |
185 | { | |
ed2ddbf8 | 186 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
187 | ipc_init_proc_interface("sysvipc/sem", |
188 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 189 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
190 | } |
191 | ||
6062a8dc RR |
192 | /* |
193 | * If the request contains only one semaphore operation, and there are | |
194 | * no complex transactions pending, lock only the semaphore involved. | |
195 | * Otherwise, lock the entire semaphore array, since we either have | |
196 | * multiple semaphores in our own semops, or we need to look at | |
197 | * semaphores from other pending complex operations. | |
198 | * | |
199 | * Carefully guard against sma->complex_count changing between zero | |
200 | * and non-zero while we are spinning for the lock. The value of | |
201 | * sma->complex_count cannot change while we are holding the lock, | |
202 | * so sem_unlock should be fine. | |
203 | * | |
204 | * The global lock path checks that all the local locks have been released, | |
205 | * checking each local lock once. This means that the local lock paths | |
206 | * cannot start their critical sections while the global lock is held. | |
207 | */ | |
208 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
209 | int nsops) | |
210 | { | |
211 | int locknum; | |
212 | again: | |
213 | if (nsops == 1 && !sma->complex_count) { | |
214 | struct sem *sem = sma->sem_base + sops->sem_num; | |
215 | ||
216 | /* Lock just the semaphore we are interested in. */ | |
217 | spin_lock(&sem->lock); | |
218 | ||
219 | /* | |
220 | * If sma->complex_count was set while we were spinning, | |
221 | * we may need to look at things we did not lock here. | |
222 | */ | |
223 | if (unlikely(sma->complex_count)) { | |
224 | spin_unlock(&sem->lock); | |
225 | goto lock_array; | |
226 | } | |
227 | ||
228 | /* | |
229 | * Another process is holding the global lock on the | |
230 | * sem_array; we cannot enter our critical section, | |
231 | * but have to wait for the global lock to be released. | |
232 | */ | |
233 | if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { | |
234 | spin_unlock(&sem->lock); | |
235 | spin_unlock_wait(&sma->sem_perm.lock); | |
236 | goto again; | |
237 | } | |
238 | ||
239 | locknum = sops->sem_num; | |
240 | } else { | |
241 | int i; | |
242 | /* | |
243 | * Lock the semaphore array, and wait for all of the | |
244 | * individual semaphore locks to go away. The code | |
245 | * above ensures no new single-lock holders will enter | |
246 | * their critical section while the array lock is held. | |
247 | */ | |
248 | lock_array: | |
249 | spin_lock(&sma->sem_perm.lock); | |
250 | for (i = 0; i < sma->sem_nsems; i++) { | |
251 | struct sem *sem = sma->sem_base + i; | |
252 | spin_unlock_wait(&sem->lock); | |
253 | } | |
254 | locknum = -1; | |
255 | } | |
256 | return locknum; | |
257 | } | |
258 | ||
259 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
260 | { | |
261 | if (locknum == -1) { | |
262 | spin_unlock(&sma->sem_perm.lock); | |
263 | } else { | |
264 | struct sem *sem = sma->sem_base + locknum; | |
265 | spin_unlock(&sem->lock); | |
266 | } | |
6062a8dc RR |
267 | } |
268 | ||
3e148c79 ND |
269 | /* |
270 | * sem_lock_(check_) routines are called in the paths where the rw_mutex | |
271 | * is not held. | |
272 | */ | |
6062a8dc RR |
273 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
274 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 275 | { |
c460b662 RR |
276 | struct kern_ipc_perm *ipcp; |
277 | struct sem_array *sma; | |
03f02c76 | 278 | |
c460b662 RR |
279 | rcu_read_lock(); |
280 | ipcp = ipc_obtain_object(&sem_ids(ns), id); | |
281 | if (IS_ERR(ipcp)) { | |
282 | sma = ERR_CAST(ipcp); | |
283 | goto err; | |
284 | } | |
b1ed88b4 | 285 | |
6062a8dc RR |
286 | sma = container_of(ipcp, struct sem_array, sem_perm); |
287 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
288 | |
289 | /* ipc_rmid() may have already freed the ID while sem_lock | |
290 | * was spinning: verify that the structure is still valid | |
291 | */ | |
292 | if (!ipcp->deleted) | |
293 | return container_of(ipcp, struct sem_array, sem_perm); | |
294 | ||
6062a8dc | 295 | sem_unlock(sma, *locknum); |
c460b662 RR |
296 | sma = ERR_PTR(-EINVAL); |
297 | err: | |
298 | rcu_read_unlock(); | |
299 | return sma; | |
023a5355 ND |
300 | } |
301 | ||
16df3674 DB |
302 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
303 | { | |
304 | struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id); | |
305 | ||
306 | if (IS_ERR(ipcp)) | |
307 | return ERR_CAST(ipcp); | |
308 | ||
309 | return container_of(ipcp, struct sem_array, sem_perm); | |
310 | } | |
311 | ||
16df3674 DB |
312 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
313 | int id) | |
314 | { | |
315 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
316 | ||
317 | if (IS_ERR(ipcp)) | |
318 | return ERR_CAST(ipcp); | |
b1ed88b4 | 319 | |
03f02c76 | 320 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
321 | } |
322 | ||
6ff37972 PP |
323 | static inline void sem_lock_and_putref(struct sem_array *sma) |
324 | { | |
6062a8dc RR |
325 | rcu_read_lock(); |
326 | sem_lock(sma, NULL, -1); | |
6ff37972 PP |
327 | ipc_rcu_putref(sma); |
328 | } | |
329 | ||
6ff37972 PP |
330 | static inline void sem_putref(struct sem_array *sma) |
331 | { | |
73b29505 | 332 | ipc_rcu_putref(sma); |
6ff37972 PP |
333 | } |
334 | ||
7ca7e564 ND |
335 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
336 | { | |
337 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
338 | } | |
339 | ||
1da177e4 LT |
340 | /* |
341 | * Lockless wakeup algorithm: | |
342 | * Without the check/retry algorithm a lockless wakeup is possible: | |
343 | * - queue.status is initialized to -EINTR before blocking. | |
344 | * - wakeup is performed by | |
345 | * * unlinking the queue entry from sma->sem_pending | |
346 | * * setting queue.status to IN_WAKEUP | |
347 | * This is the notification for the blocked thread that a | |
348 | * result value is imminent. | |
349 | * * call wake_up_process | |
350 | * * set queue.status to the final value. | |
351 | * - the previously blocked thread checks queue.status: | |
352 | * * if it's IN_WAKEUP, then it must wait until the value changes | |
353 | * * if it's not -EINTR, then the operation was completed by | |
354 | * update_queue. semtimedop can return queue.status without | |
5f921ae9 | 355 | * performing any operation on the sem array. |
1da177e4 LT |
356 | * * otherwise it must acquire the spinlock and check what's up. |
357 | * | |
358 | * The two-stage algorithm is necessary to protect against the following | |
359 | * races: | |
360 | * - if queue.status is set after wake_up_process, then the woken up idle | |
361 | * thread could race forward and try (and fail) to acquire sma->lock | |
362 | * before update_queue had a chance to set queue.status | |
363 | * - if queue.status is written before wake_up_process and if the | |
364 | * blocked process is woken up by a signal between writing | |
365 | * queue.status and the wake_up_process, then the woken up | |
366 | * process could return from semtimedop and die by calling | |
367 | * sys_exit before wake_up_process is called. Then wake_up_process | |
368 | * will oops, because the task structure is already invalid. | |
369 | * (yes, this happened on s390 with sysv msg). | |
370 | * | |
371 | */ | |
372 | #define IN_WAKEUP 1 | |
373 | ||
f4566f04 ND |
374 | /** |
375 | * newary - Create a new semaphore set | |
376 | * @ns: namespace | |
377 | * @params: ptr to the structure that contains key, semflg and nsems | |
378 | * | |
3e148c79 | 379 | * Called with sem_ids.rw_mutex held (as a writer) |
f4566f04 ND |
380 | */ |
381 | ||
7748dbfa | 382 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
383 | { |
384 | int id; | |
385 | int retval; | |
386 | struct sem_array *sma; | |
387 | int size; | |
7748dbfa ND |
388 | key_t key = params->key; |
389 | int nsems = params->u.nsems; | |
390 | int semflg = params->flg; | |
b97e820f | 391 | int i; |
1da177e4 LT |
392 | |
393 | if (!nsems) | |
394 | return -EINVAL; | |
e3893534 | 395 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
396 | return -ENOSPC; |
397 | ||
398 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
399 | sma = ipc_rcu_alloc(size); | |
400 | if (!sma) { | |
401 | return -ENOMEM; | |
402 | } | |
403 | memset (sma, 0, size); | |
404 | ||
405 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
406 | sma->sem_perm.key = key; | |
407 | ||
408 | sma->sem_perm.security = NULL; | |
409 | retval = security_sem_alloc(sma); | |
410 | if (retval) { | |
411 | ipc_rcu_putref(sma); | |
412 | return retval; | |
413 | } | |
414 | ||
e3893534 | 415 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 416 | if (id < 0) { |
1da177e4 LT |
417 | security_sem_free(sma); |
418 | ipc_rcu_putref(sma); | |
283bb7fa | 419 | return id; |
1da177e4 | 420 | } |
e3893534 | 421 | ns->used_sems += nsems; |
1da177e4 LT |
422 | |
423 | sma->sem_base = (struct sem *) &sma[1]; | |
b97e820f | 424 | |
6062a8dc | 425 | for (i = 0; i < nsems; i++) { |
b97e820f | 426 | INIT_LIST_HEAD(&sma->sem_base[i].sem_pending); |
6062a8dc RR |
427 | spin_lock_init(&sma->sem_base[i].lock); |
428 | } | |
b97e820f MS |
429 | |
430 | sma->complex_count = 0; | |
a1193f8e | 431 | INIT_LIST_HEAD(&sma->sem_pending); |
4daa28f6 | 432 | INIT_LIST_HEAD(&sma->list_id); |
1da177e4 LT |
433 | sma->sem_nsems = nsems; |
434 | sma->sem_ctime = get_seconds(); | |
6062a8dc | 435 | sem_unlock(sma, -1); |
6d49dab8 | 436 | rcu_read_unlock(); |
1da177e4 | 437 | |
7ca7e564 | 438 | return sma->sem_perm.id; |
1da177e4 LT |
439 | } |
440 | ||
7748dbfa | 441 | |
f4566f04 | 442 | /* |
3e148c79 | 443 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 444 | */ |
03f02c76 | 445 | static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) |
7748dbfa | 446 | { |
03f02c76 ND |
447 | struct sem_array *sma; |
448 | ||
449 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
450 | return security_sem_associate(sma, semflg); | |
7748dbfa ND |
451 | } |
452 | ||
f4566f04 | 453 | /* |
3e148c79 | 454 | * Called with sem_ids.rw_mutex and ipcp locked. |
f4566f04 | 455 | */ |
03f02c76 ND |
456 | static inline int sem_more_checks(struct kern_ipc_perm *ipcp, |
457 | struct ipc_params *params) | |
7748dbfa | 458 | { |
03f02c76 ND |
459 | struct sem_array *sma; |
460 | ||
461 | sma = container_of(ipcp, struct sem_array, sem_perm); | |
462 | if (params->u.nsems > sma->sem_nsems) | |
7748dbfa ND |
463 | return -EINVAL; |
464 | ||
465 | return 0; | |
466 | } | |
467 | ||
d5460c99 | 468 | SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) |
1da177e4 | 469 | { |
e3893534 | 470 | struct ipc_namespace *ns; |
7748dbfa ND |
471 | struct ipc_ops sem_ops; |
472 | struct ipc_params sem_params; | |
e3893534 KK |
473 | |
474 | ns = current->nsproxy->ipc_ns; | |
1da177e4 | 475 | |
e3893534 | 476 | if (nsems < 0 || nsems > ns->sc_semmsl) |
1da177e4 | 477 | return -EINVAL; |
7ca7e564 | 478 | |
7748dbfa ND |
479 | sem_ops.getnew = newary; |
480 | sem_ops.associate = sem_security; | |
481 | sem_ops.more_checks = sem_more_checks; | |
482 | ||
483 | sem_params.key = key; | |
484 | sem_params.flg = semflg; | |
485 | sem_params.u.nsems = nsems; | |
1da177e4 | 486 | |
7748dbfa | 487 | return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); |
1da177e4 LT |
488 | } |
489 | ||
1da177e4 LT |
490 | /* |
491 | * Determine whether a sequence of semaphore operations would succeed | |
492 | * all at once. Return 0 if yes, 1 if need to sleep, else return error code. | |
493 | */ | |
494 | ||
495 | static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, | |
496 | int nsops, struct sem_undo *un, int pid) | |
497 | { | |
498 | int result, sem_op; | |
499 | struct sembuf *sop; | |
500 | struct sem * curr; | |
501 | ||
502 | for (sop = sops; sop < sops + nsops; sop++) { | |
503 | curr = sma->sem_base + sop->sem_num; | |
504 | sem_op = sop->sem_op; | |
505 | result = curr->semval; | |
506 | ||
507 | if (!sem_op && result) | |
508 | goto would_block; | |
509 | ||
510 | result += sem_op; | |
511 | if (result < 0) | |
512 | goto would_block; | |
513 | if (result > SEMVMX) | |
514 | goto out_of_range; | |
515 | if (sop->sem_flg & SEM_UNDO) { | |
516 | int undo = un->semadj[sop->sem_num] - sem_op; | |
517 | /* | |
518 | * Exceeding the undo range is an error. | |
519 | */ | |
520 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) | |
521 | goto out_of_range; | |
522 | } | |
523 | curr->semval = result; | |
524 | } | |
525 | ||
526 | sop--; | |
527 | while (sop >= sops) { | |
528 | sma->sem_base[sop->sem_num].sempid = pid; | |
529 | if (sop->sem_flg & SEM_UNDO) | |
530 | un->semadj[sop->sem_num] -= sop->sem_op; | |
531 | sop--; | |
532 | } | |
533 | ||
1da177e4 LT |
534 | return 0; |
535 | ||
536 | out_of_range: | |
537 | result = -ERANGE; | |
538 | goto undo; | |
539 | ||
540 | would_block: | |
541 | if (sop->sem_flg & IPC_NOWAIT) | |
542 | result = -EAGAIN; | |
543 | else | |
544 | result = 1; | |
545 | ||
546 | undo: | |
547 | sop--; | |
548 | while (sop >= sops) { | |
549 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; | |
550 | sop--; | |
551 | } | |
552 | ||
553 | return result; | |
554 | } | |
555 | ||
0a2b9d4c MS |
556 | /** wake_up_sem_queue_prepare(q, error): Prepare wake-up |
557 | * @q: queue entry that must be signaled | |
558 | * @error: Error value for the signal | |
559 | * | |
560 | * Prepare the wake-up of the queue entry q. | |
d4212093 | 561 | */ |
0a2b9d4c MS |
562 | static void wake_up_sem_queue_prepare(struct list_head *pt, |
563 | struct sem_queue *q, int error) | |
d4212093 | 564 | { |
0a2b9d4c MS |
565 | if (list_empty(pt)) { |
566 | /* | |
567 | * Hold preempt off so that we don't get preempted and have the | |
568 | * wakee busy-wait until we're scheduled back on. | |
569 | */ | |
570 | preempt_disable(); | |
571 | } | |
d4212093 | 572 | q->status = IN_WAKEUP; |
0a2b9d4c MS |
573 | q->pid = error; |
574 | ||
9f1bc2c9 | 575 | list_add_tail(&q->list, pt); |
0a2b9d4c MS |
576 | } |
577 | ||
578 | /** | |
579 | * wake_up_sem_queue_do(pt) - do the actual wake-up | |
580 | * @pt: list of tasks to be woken up | |
581 | * | |
582 | * Do the actual wake-up. | |
583 | * The function is called without any locks held, thus the semaphore array | |
584 | * could be destroyed already and the tasks can disappear as soon as the | |
585 | * status is set to the actual return code. | |
586 | */ | |
587 | static void wake_up_sem_queue_do(struct list_head *pt) | |
588 | { | |
589 | struct sem_queue *q, *t; | |
590 | int did_something; | |
591 | ||
592 | did_something = !list_empty(pt); | |
9f1bc2c9 | 593 | list_for_each_entry_safe(q, t, pt, list) { |
0a2b9d4c MS |
594 | wake_up_process(q->sleeper); |
595 | /* q can disappear immediately after writing q->status. */ | |
596 | smp_wmb(); | |
597 | q->status = q->pid; | |
598 | } | |
599 | if (did_something) | |
600 | preempt_enable(); | |
d4212093 NP |
601 | } |
602 | ||
b97e820f MS |
603 | static void unlink_queue(struct sem_array *sma, struct sem_queue *q) |
604 | { | |
605 | list_del(&q->list); | |
9f1bc2c9 | 606 | if (q->nsops > 1) |
b97e820f MS |
607 | sma->complex_count--; |
608 | } | |
609 | ||
fd5db422 MS |
610 | /** check_restart(sma, q) |
611 | * @sma: semaphore array | |
612 | * @q: the operation that just completed | |
613 | * | |
614 | * update_queue is O(N^2) when it restarts scanning the whole queue of | |
615 | * waiting operations. Therefore this function checks if the restart is | |
616 | * really necessary. It is called after a previously waiting operation | |
617 | * was completed. | |
618 | */ | |
619 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
620 | { | |
621 | struct sem *curr; | |
622 | struct sem_queue *h; | |
623 | ||
624 | /* if the operation didn't modify the array, then no restart */ | |
625 | if (q->alter == 0) | |
626 | return 0; | |
627 | ||
628 | /* pending complex operations are too difficult to analyse */ | |
629 | if (sma->complex_count) | |
630 | return 1; | |
631 | ||
632 | /* we were a sleeping complex operation. Too difficult */ | |
633 | if (q->nsops > 1) | |
634 | return 1; | |
635 | ||
636 | curr = sma->sem_base + q->sops[0].sem_num; | |
637 | ||
638 | /* No-one waits on this queue */ | |
639 | if (list_empty(&curr->sem_pending)) | |
640 | return 0; | |
641 | ||
642 | /* the new semaphore value */ | |
643 | if (curr->semval) { | |
644 | /* It is impossible that someone waits for the new value: | |
645 | * - q is a previously sleeping simple operation that | |
646 | * altered the array. It must be a decrement, because | |
647 | * simple increments never sleep. | |
648 | * - The value is not 0, thus wait-for-zero won't proceed. | |
649 | * - If there are older (higher priority) decrements | |
650 | * in the queue, then they have observed the original | |
651 | * semval value and couldn't proceed. The operation | |
652 | * decremented to value - thus they won't proceed either. | |
653 | */ | |
654 | BUG_ON(q->sops[0].sem_op >= 0); | |
655 | return 0; | |
656 | } | |
657 | /* | |
658 | * semval is 0. Check if there are wait-for-zero semops. | |
9f1bc2c9 | 659 | * They must be the first entries in the per-semaphore queue |
fd5db422 | 660 | */ |
9f1bc2c9 | 661 | h = list_first_entry(&curr->sem_pending, struct sem_queue, list); |
fd5db422 MS |
662 | BUG_ON(h->nsops != 1); |
663 | BUG_ON(h->sops[0].sem_num != q->sops[0].sem_num); | |
664 | ||
665 | /* Yes, there is a wait-for-zero semop. Restart */ | |
666 | if (h->sops[0].sem_op == 0) | |
667 | return 1; | |
668 | ||
669 | /* Again - no-one is waiting for the new value. */ | |
670 | return 0; | |
671 | } | |
672 | ||
636c6be8 MS |
673 | |
674 | /** | |
675 | * update_queue(sma, semnum): Look for tasks that can be completed. | |
676 | * @sma: semaphore array. | |
677 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 678 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
679 | * |
680 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
681 | * was modified. If multiple semaphores were modified, update_queue must |
682 | * be called with semnum = -1, as well as with the number of each modified | |
683 | * semaphore. | |
0a2b9d4c MS |
684 | * The tasks that must be woken up are added to @pt. The return code |
685 | * is stored in q->pid. | |
686 | * The function return 1 if at least one semop was completed successfully. | |
1da177e4 | 687 | */ |
0a2b9d4c | 688 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 689 | { |
636c6be8 MS |
690 | struct sem_queue *q; |
691 | struct list_head *walk; | |
692 | struct list_head *pending_list; | |
0a2b9d4c | 693 | int semop_completed = 0; |
636c6be8 | 694 | |
9f1bc2c9 | 695 | if (semnum == -1) |
636c6be8 | 696 | pending_list = &sma->sem_pending; |
9f1bc2c9 | 697 | else |
636c6be8 | 698 | pending_list = &sma->sem_base[semnum].sem_pending; |
9cad200c NP |
699 | |
700 | again: | |
636c6be8 MS |
701 | walk = pending_list->next; |
702 | while (walk != pending_list) { | |
fd5db422 | 703 | int error, restart; |
636c6be8 | 704 | |
9f1bc2c9 | 705 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 706 | walk = walk->next; |
1da177e4 | 707 | |
d987f8b2 MS |
708 | /* If we are scanning the single sop, per-semaphore list of |
709 | * one semaphore and that semaphore is 0, then it is not | |
710 | * necessary to scan the "alter" entries: simple increments | |
711 | * that affect only one entry succeed immediately and cannot | |
712 | * be in the per semaphore pending queue, and decrements | |
713 | * cannot be successful if the value is already 0. | |
714 | */ | |
715 | if (semnum != -1 && sma->sem_base[semnum].semval == 0 && | |
716 | q->alter) | |
717 | break; | |
718 | ||
1da177e4 LT |
719 | error = try_atomic_semop(sma, q->sops, q->nsops, |
720 | q->undo, q->pid); | |
721 | ||
722 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
723 | if (error > 0) |
724 | continue; | |
725 | ||
b97e820f | 726 | unlink_queue(sma, q); |
9cad200c | 727 | |
0a2b9d4c | 728 | if (error) { |
fd5db422 | 729 | restart = 0; |
0a2b9d4c MS |
730 | } else { |
731 | semop_completed = 1; | |
fd5db422 | 732 | restart = check_restart(sma, q); |
0a2b9d4c | 733 | } |
fd5db422 | 734 | |
0a2b9d4c | 735 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 736 | if (restart) |
9cad200c | 737 | goto again; |
1da177e4 | 738 | } |
0a2b9d4c | 739 | return semop_completed; |
1da177e4 LT |
740 | } |
741 | ||
0a2b9d4c MS |
742 | /** |
743 | * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue | |
fd5db422 MS |
744 | * @sma: semaphore array |
745 | * @sops: operations that were performed | |
746 | * @nsops: number of operations | |
0a2b9d4c MS |
747 | * @otime: force setting otime |
748 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 MS |
749 | * |
750 | * do_smart_update() does the required called to update_queue, based on the | |
751 | * actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
752 | * Note that the function does not do the actual wake-up: the caller is |
753 | * responsible for calling wake_up_sem_queue_do(@pt). | |
754 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 755 | */ |
0a2b9d4c MS |
756 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
757 | int otime, struct list_head *pt) | |
fd5db422 MS |
758 | { |
759 | int i; | |
760 | ||
761 | if (sma->complex_count || sops == NULL) { | |
0a2b9d4c MS |
762 | if (update_queue(sma, -1, pt)) |
763 | otime = 1; | |
9f1bc2c9 RR |
764 | } |
765 | ||
766 | if (!sops) { | |
767 | /* No semops; something special is going on. */ | |
768 | for (i = 0; i < sma->sem_nsems; i++) { | |
769 | if (update_queue(sma, i, pt)) | |
770 | otime = 1; | |
771 | } | |
0a2b9d4c | 772 | goto done; |
fd5db422 MS |
773 | } |
774 | ||
9f1bc2c9 | 775 | /* Check the semaphores that were modified. */ |
fd5db422 MS |
776 | for (i = 0; i < nsops; i++) { |
777 | if (sops[i].sem_op > 0 || | |
778 | (sops[i].sem_op < 0 && | |
779 | sma->sem_base[sops[i].sem_num].semval == 0)) | |
0a2b9d4c MS |
780 | if (update_queue(sma, sops[i].sem_num, pt)) |
781 | otime = 1; | |
fd5db422 | 782 | } |
0a2b9d4c MS |
783 | done: |
784 | if (otime) | |
785 | sma->sem_otime = get_seconds(); | |
fd5db422 MS |
786 | } |
787 | ||
788 | ||
1da177e4 LT |
789 | /* The following counts are associated to each semaphore: |
790 | * semncnt number of tasks waiting on semval being nonzero | |
791 | * semzcnt number of tasks waiting on semval being zero | |
792 | * This model assumes that a task waits on exactly one semaphore. | |
793 | * Since semaphore operations are to be performed atomically, tasks actually | |
794 | * wait on a whole sequence of semaphores simultaneously. | |
795 | * The counts we return here are a rough approximation, but still | |
796 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
797 | */ | |
798 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
799 | { | |
800 | int semncnt; | |
801 | struct sem_queue * q; | |
802 | ||
803 | semncnt = 0; | |
a1193f8e | 804 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
805 | struct sembuf * sops = q->sops; |
806 | int nsops = q->nsops; | |
807 | int i; | |
808 | for (i = 0; i < nsops; i++) | |
809 | if (sops[i].sem_num == semnum | |
810 | && (sops[i].sem_op < 0) | |
811 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
812 | semncnt++; | |
813 | } | |
814 | return semncnt; | |
815 | } | |
a1193f8e | 816 | |
1da177e4 LT |
817 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
818 | { | |
819 | int semzcnt; | |
820 | struct sem_queue * q; | |
821 | ||
822 | semzcnt = 0; | |
a1193f8e | 823 | list_for_each_entry(q, &sma->sem_pending, list) { |
1da177e4 LT |
824 | struct sembuf * sops = q->sops; |
825 | int nsops = q->nsops; | |
826 | int i; | |
827 | for (i = 0; i < nsops; i++) | |
828 | if (sops[i].sem_num == semnum | |
829 | && (sops[i].sem_op == 0) | |
830 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
831 | semzcnt++; | |
832 | } | |
833 | return semzcnt; | |
834 | } | |
835 | ||
3e148c79 ND |
836 | /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked |
837 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex | |
838 | * remains locked on exit. | |
1da177e4 | 839 | */ |
01b8b07a | 840 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 841 | { |
380af1b3 MS |
842 | struct sem_undo *un, *tu; |
843 | struct sem_queue *q, *tq; | |
01b8b07a | 844 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 845 | struct list_head tasks; |
9f1bc2c9 | 846 | int i; |
1da177e4 | 847 | |
380af1b3 | 848 | /* Free the existing undo structures for this semaphore set. */ |
4daa28f6 | 849 | assert_spin_locked(&sma->sem_perm.lock); |
380af1b3 MS |
850 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
851 | list_del(&un->list_id); | |
852 | spin_lock(&un->ulp->lock); | |
1da177e4 | 853 | un->semid = -1; |
380af1b3 MS |
854 | list_del_rcu(&un->list_proc); |
855 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 856 | kfree_rcu(un, rcu); |
380af1b3 | 857 | } |
1da177e4 LT |
858 | |
859 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 860 | INIT_LIST_HEAD(&tasks); |
380af1b3 | 861 | list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { |
b97e820f | 862 | unlink_queue(sma, q); |
0a2b9d4c | 863 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 864 | } |
9f1bc2c9 RR |
865 | for (i = 0; i < sma->sem_nsems; i++) { |
866 | struct sem *sem = sma->sem_base + i; | |
867 | list_for_each_entry_safe(q, tq, &sem->sem_pending, list) { | |
868 | unlink_queue(sma, q); | |
869 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
870 | } | |
871 | } | |
1da177e4 | 872 | |
7ca7e564 ND |
873 | /* Remove the semaphore set from the IDR */ |
874 | sem_rmid(ns, sma); | |
6062a8dc | 875 | sem_unlock(sma, -1); |
6d49dab8 | 876 | rcu_read_unlock(); |
1da177e4 | 877 | |
0a2b9d4c | 878 | wake_up_sem_queue_do(&tasks); |
e3893534 | 879 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
880 | security_sem_free(sma); |
881 | ipc_rcu_putref(sma); | |
882 | } | |
883 | ||
884 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
885 | { | |
886 | switch(version) { | |
887 | case IPC_64: | |
888 | return copy_to_user(buf, in, sizeof(*in)); | |
889 | case IPC_OLD: | |
890 | { | |
891 | struct semid_ds out; | |
892 | ||
982f7c2b DR |
893 | memset(&out, 0, sizeof(out)); |
894 | ||
1da177e4 LT |
895 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
896 | ||
897 | out.sem_otime = in->sem_otime; | |
898 | out.sem_ctime = in->sem_ctime; | |
899 | out.sem_nsems = in->sem_nsems; | |
900 | ||
901 | return copy_to_user(buf, &out, sizeof(out)); | |
902 | } | |
903 | default: | |
904 | return -EINVAL; | |
905 | } | |
906 | } | |
907 | ||
4b9fcb0e | 908 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 909 | int cmd, int version, void __user *p) |
1da177e4 | 910 | { |
e5cc9c7b | 911 | int err; |
1da177e4 LT |
912 | struct sem_array *sma; |
913 | ||
914 | switch(cmd) { | |
915 | case IPC_INFO: | |
916 | case SEM_INFO: | |
917 | { | |
918 | struct seminfo seminfo; | |
919 | int max_id; | |
920 | ||
921 | err = security_sem_semctl(NULL, cmd); | |
922 | if (err) | |
923 | return err; | |
924 | ||
925 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
926 | seminfo.semmni = ns->sc_semmni; |
927 | seminfo.semmns = ns->sc_semmns; | |
928 | seminfo.semmsl = ns->sc_semmsl; | |
929 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
930 | seminfo.semvmx = SEMVMX; |
931 | seminfo.semmnu = SEMMNU; | |
932 | seminfo.semmap = SEMMAP; | |
933 | seminfo.semume = SEMUME; | |
3e148c79 | 934 | down_read(&sem_ids(ns).rw_mutex); |
1da177e4 | 935 | if (cmd == SEM_INFO) { |
e3893534 KK |
936 | seminfo.semusz = sem_ids(ns).in_use; |
937 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
938 | } else { |
939 | seminfo.semusz = SEMUSZ; | |
940 | seminfo.semaem = SEMAEM; | |
941 | } | |
7ca7e564 | 942 | max_id = ipc_get_maxid(&sem_ids(ns)); |
3e148c79 | 943 | up_read(&sem_ids(ns).rw_mutex); |
e1fd1f49 | 944 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 LT |
945 | return -EFAULT; |
946 | return (max_id < 0) ? 0: max_id; | |
947 | } | |
4b9fcb0e | 948 | case IPC_STAT: |
1da177e4 LT |
949 | case SEM_STAT: |
950 | { | |
951 | struct semid64_ds tbuf; | |
16df3674 DB |
952 | int id = 0; |
953 | ||
954 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 955 | |
4b9fcb0e | 956 | if (cmd == SEM_STAT) { |
16df3674 DB |
957 | rcu_read_lock(); |
958 | sma = sem_obtain_object(ns, semid); | |
959 | if (IS_ERR(sma)) { | |
960 | err = PTR_ERR(sma); | |
961 | goto out_unlock; | |
962 | } | |
4b9fcb0e PP |
963 | id = sma->sem_perm.id; |
964 | } else { | |
16df3674 DB |
965 | rcu_read_lock(); |
966 | sma = sem_obtain_object_check(ns, semid); | |
967 | if (IS_ERR(sma)) { | |
968 | err = PTR_ERR(sma); | |
969 | goto out_unlock; | |
970 | } | |
4b9fcb0e | 971 | } |
1da177e4 LT |
972 | |
973 | err = -EACCES; | |
b0e77598 | 974 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
975 | goto out_unlock; |
976 | ||
977 | err = security_sem_semctl(sma, cmd); | |
978 | if (err) | |
979 | goto out_unlock; | |
980 | ||
1da177e4 LT |
981 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
982 | tbuf.sem_otime = sma->sem_otime; | |
983 | tbuf.sem_ctime = sma->sem_ctime; | |
984 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 985 | rcu_read_unlock(); |
e1fd1f49 | 986 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
987 | return -EFAULT; |
988 | return id; | |
989 | } | |
990 | default: | |
991 | return -EINVAL; | |
992 | } | |
1da177e4 | 993 | out_unlock: |
16df3674 | 994 | rcu_read_unlock(); |
1da177e4 LT |
995 | return err; |
996 | } | |
997 | ||
e1fd1f49 AV |
998 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
999 | unsigned long arg) | |
1000 | { | |
1001 | struct sem_undo *un; | |
1002 | struct sem_array *sma; | |
1003 | struct sem* curr; | |
1004 | int err; | |
e1fd1f49 AV |
1005 | struct list_head tasks; |
1006 | int val; | |
1007 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1008 | /* big-endian 64bit */ | |
1009 | val = arg >> 32; | |
1010 | #else | |
1011 | /* 32bit or little-endian 64bit */ | |
1012 | val = arg; | |
1013 | #endif | |
1014 | ||
6062a8dc RR |
1015 | if (val > SEMVMX || val < 0) |
1016 | return -ERANGE; | |
e1fd1f49 AV |
1017 | |
1018 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1019 | |
6062a8dc RR |
1020 | rcu_read_lock(); |
1021 | sma = sem_obtain_object_check(ns, semid); | |
1022 | if (IS_ERR(sma)) { | |
1023 | rcu_read_unlock(); | |
1024 | return PTR_ERR(sma); | |
1025 | } | |
1026 | ||
1027 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1028 | rcu_read_unlock(); | |
1029 | return -EINVAL; | |
1030 | } | |
1031 | ||
1032 | ||
1033 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1034 | rcu_read_unlock(); | |
1035 | return -EACCES; | |
1036 | } | |
e1fd1f49 AV |
1037 | |
1038 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1039 | if (err) { |
1040 | rcu_read_unlock(); | |
1041 | return -EACCES; | |
1042 | } | |
e1fd1f49 | 1043 | |
6062a8dc | 1044 | sem_lock(sma, NULL, -1); |
e1fd1f49 AV |
1045 | |
1046 | curr = &sma->sem_base[semnum]; | |
1047 | ||
e1fd1f49 AV |
1048 | assert_spin_locked(&sma->sem_perm.lock); |
1049 | list_for_each_entry(un, &sma->list_id, list_id) | |
1050 | un->semadj[semnum] = 0; | |
1051 | ||
1052 | curr->semval = val; | |
1053 | curr->sempid = task_tgid_vnr(current); | |
1054 | sma->sem_ctime = get_seconds(); | |
1055 | /* maybe some queued-up processes were waiting for this */ | |
1056 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1057 | sem_unlock(sma, -1); |
6d49dab8 | 1058 | rcu_read_unlock(); |
e1fd1f49 | 1059 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1060 | return 0; |
e1fd1f49 AV |
1061 | } |
1062 | ||
e3893534 | 1063 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1064 | int cmd, void __user *p) |
1da177e4 LT |
1065 | { |
1066 | struct sem_array *sma; | |
1067 | struct sem* curr; | |
16df3674 | 1068 | int err, nsems; |
1da177e4 LT |
1069 | ushort fast_sem_io[SEMMSL_FAST]; |
1070 | ushort* sem_io = fast_sem_io; | |
0a2b9d4c | 1071 | struct list_head tasks; |
1da177e4 | 1072 | |
16df3674 DB |
1073 | INIT_LIST_HEAD(&tasks); |
1074 | ||
1075 | rcu_read_lock(); | |
1076 | sma = sem_obtain_object_check(ns, semid); | |
1077 | if (IS_ERR(sma)) { | |
1078 | rcu_read_unlock(); | |
023a5355 | 1079 | return PTR_ERR(sma); |
16df3674 | 1080 | } |
1da177e4 LT |
1081 | |
1082 | nsems = sma->sem_nsems; | |
1083 | ||
1da177e4 | 1084 | err = -EACCES; |
b0e77598 | 1085 | if (ipcperms(ns, &sma->sem_perm, |
16df3674 DB |
1086 | cmd == SETALL ? S_IWUGO : S_IRUGO)) { |
1087 | rcu_read_unlock(); | |
1088 | goto out_wakeup; | |
1089 | } | |
1da177e4 LT |
1090 | |
1091 | err = security_sem_semctl(sma, cmd); | |
16df3674 DB |
1092 | if (err) { |
1093 | rcu_read_unlock(); | |
1094 | goto out_wakeup; | |
1095 | } | |
1da177e4 LT |
1096 | |
1097 | err = -EACCES; | |
1098 | switch (cmd) { | |
1099 | case GETALL: | |
1100 | { | |
e1fd1f49 | 1101 | ushort __user *array = p; |
1da177e4 LT |
1102 | int i; |
1103 | ||
ce857229 | 1104 | sem_lock(sma, NULL, -1); |
1da177e4 | 1105 | if(nsems > SEMMSL_FAST) { |
ce857229 AV |
1106 | if (!ipc_rcu_getref(sma)) { |
1107 | sem_unlock(sma, -1); | |
6d49dab8 | 1108 | rcu_read_unlock(); |
ce857229 AV |
1109 | err = -EIDRM; |
1110 | goto out_free; | |
1111 | } | |
1112 | sem_unlock(sma, -1); | |
6d49dab8 | 1113 | rcu_read_unlock(); |
1da177e4 LT |
1114 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
1115 | if(sem_io == NULL) { | |
6ff37972 | 1116 | sem_putref(sma); |
1da177e4 LT |
1117 | return -ENOMEM; |
1118 | } | |
1119 | ||
6ff37972 | 1120 | sem_lock_and_putref(sma); |
1da177e4 | 1121 | if (sma->sem_perm.deleted) { |
6062a8dc | 1122 | sem_unlock(sma, -1); |
6d49dab8 | 1123 | rcu_read_unlock(); |
1da177e4 LT |
1124 | err = -EIDRM; |
1125 | goto out_free; | |
1126 | } | |
ce857229 | 1127 | } |
1da177e4 LT |
1128 | for (i = 0; i < sma->sem_nsems; i++) |
1129 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1130 | sem_unlock(sma, -1); |
6d49dab8 | 1131 | rcu_read_unlock(); |
1da177e4 LT |
1132 | err = 0; |
1133 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
1134 | err = -EFAULT; | |
1135 | goto out_free; | |
1136 | } | |
1137 | case SETALL: | |
1138 | { | |
1139 | int i; | |
1140 | struct sem_undo *un; | |
1141 | ||
6062a8dc RR |
1142 | if (!ipc_rcu_getref(sma)) { |
1143 | rcu_read_unlock(); | |
1144 | return -EIDRM; | |
1145 | } | |
16df3674 | 1146 | rcu_read_unlock(); |
1da177e4 LT |
1147 | |
1148 | if(nsems > SEMMSL_FAST) { | |
1149 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
1150 | if(sem_io == NULL) { | |
6ff37972 | 1151 | sem_putref(sma); |
1da177e4 LT |
1152 | return -ENOMEM; |
1153 | } | |
1154 | } | |
1155 | ||
e1fd1f49 | 1156 | if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) { |
6ff37972 | 1157 | sem_putref(sma); |
1da177e4 LT |
1158 | err = -EFAULT; |
1159 | goto out_free; | |
1160 | } | |
1161 | ||
1162 | for (i = 0; i < nsems; i++) { | |
1163 | if (sem_io[i] > SEMVMX) { | |
6ff37972 | 1164 | sem_putref(sma); |
1da177e4 LT |
1165 | err = -ERANGE; |
1166 | goto out_free; | |
1167 | } | |
1168 | } | |
6ff37972 | 1169 | sem_lock_and_putref(sma); |
1da177e4 | 1170 | if (sma->sem_perm.deleted) { |
6062a8dc | 1171 | sem_unlock(sma, -1); |
6d49dab8 | 1172 | rcu_read_unlock(); |
1da177e4 LT |
1173 | err = -EIDRM; |
1174 | goto out_free; | |
1175 | } | |
1176 | ||
1177 | for (i = 0; i < nsems; i++) | |
1178 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 MS |
1179 | |
1180 | assert_spin_locked(&sma->sem_perm.lock); | |
1181 | list_for_each_entry(un, &sma->list_id, list_id) { | |
1da177e4 LT |
1182 | for (i = 0; i < nsems; i++) |
1183 | un->semadj[i] = 0; | |
4daa28f6 | 1184 | } |
1da177e4 LT |
1185 | sma->sem_ctime = get_seconds(); |
1186 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1187 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1188 | err = 0; |
1189 | goto out_unlock; | |
1190 | } | |
e1fd1f49 | 1191 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1192 | } |
1193 | err = -EINVAL; | |
16df3674 DB |
1194 | if (semnum < 0 || semnum >= nsems) { |
1195 | rcu_read_unlock(); | |
1196 | goto out_wakeup; | |
1197 | } | |
1da177e4 | 1198 | |
6062a8dc | 1199 | sem_lock(sma, NULL, -1); |
1da177e4 LT |
1200 | curr = &sma->sem_base[semnum]; |
1201 | ||
1202 | switch (cmd) { | |
1203 | case GETVAL: | |
1204 | err = curr->semval; | |
1205 | goto out_unlock; | |
1206 | case GETPID: | |
1207 | err = curr->sempid; | |
1208 | goto out_unlock; | |
1209 | case GETNCNT: | |
1210 | err = count_semncnt(sma,semnum); | |
1211 | goto out_unlock; | |
1212 | case GETZCNT: | |
1213 | err = count_semzcnt(sma,semnum); | |
1214 | goto out_unlock; | |
1da177e4 | 1215 | } |
16df3674 | 1216 | |
1da177e4 | 1217 | out_unlock: |
6062a8dc | 1218 | sem_unlock(sma, -1); |
6d49dab8 | 1219 | rcu_read_unlock(); |
16df3674 | 1220 | out_wakeup: |
0a2b9d4c | 1221 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
1222 | out_free: |
1223 | if(sem_io != fast_sem_io) | |
1224 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
1225 | return err; | |
1226 | } | |
1227 | ||
016d7132 PP |
1228 | static inline unsigned long |
1229 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
1230 | { |
1231 | switch(version) { | |
1232 | case IPC_64: | |
016d7132 | 1233 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1234 | return -EFAULT; |
1da177e4 | 1235 | return 0; |
1da177e4 LT |
1236 | case IPC_OLD: |
1237 | { | |
1238 | struct semid_ds tbuf_old; | |
1239 | ||
1240 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
1241 | return -EFAULT; | |
1242 | ||
016d7132 PP |
1243 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1244 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1245 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1246 | |
1247 | return 0; | |
1248 | } | |
1249 | default: | |
1250 | return -EINVAL; | |
1251 | } | |
1252 | } | |
1253 | ||
522bb2a2 PP |
1254 | /* |
1255 | * This function handles some semctl commands which require the rw_mutex | |
1256 | * to be held in write mode. | |
1257 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
1258 | */ | |
21a4826a | 1259 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1260 | int cmd, int version, void __user *p) |
1da177e4 LT |
1261 | { |
1262 | struct sem_array *sma; | |
1263 | int err; | |
016d7132 | 1264 | struct semid64_ds semid64; |
1da177e4 LT |
1265 | struct kern_ipc_perm *ipcp; |
1266 | ||
1267 | if(cmd == IPC_SET) { | |
e1fd1f49 | 1268 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1269 | return -EFAULT; |
1da177e4 | 1270 | } |
073115d6 | 1271 | |
16df3674 DB |
1272 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1273 | &semid64.sem_perm, 0); | |
a5f75e7f PP |
1274 | if (IS_ERR(ipcp)) |
1275 | return PTR_ERR(ipcp); | |
073115d6 | 1276 | |
a5f75e7f | 1277 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1278 | |
1279 | err = security_sem_semctl(sma, cmd); | |
16df3674 DB |
1280 | if (err) { |
1281 | rcu_read_unlock(); | |
1da177e4 | 1282 | goto out_unlock; |
16df3674 | 1283 | } |
1da177e4 LT |
1284 | |
1285 | switch(cmd){ | |
1286 | case IPC_RMID: | |
6062a8dc | 1287 | sem_lock(sma, NULL, -1); |
01b8b07a | 1288 | freeary(ns, ipcp); |
522bb2a2 | 1289 | goto out_up; |
1da177e4 | 1290 | case IPC_SET: |
6062a8dc | 1291 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1292 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1293 | if (err) | |
1294 | goto out_unlock; | |
1da177e4 | 1295 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1296 | break; |
1297 | default: | |
16df3674 | 1298 | rcu_read_unlock(); |
1da177e4 | 1299 | err = -EINVAL; |
16df3674 | 1300 | goto out_up; |
1da177e4 | 1301 | } |
1da177e4 LT |
1302 | |
1303 | out_unlock: | |
6062a8dc | 1304 | sem_unlock(sma, -1); |
6d49dab8 | 1305 | rcu_read_unlock(); |
522bb2a2 PP |
1306 | out_up: |
1307 | up_write(&sem_ids(ns).rw_mutex); | |
1da177e4 LT |
1308 | return err; |
1309 | } | |
1310 | ||
e1fd1f49 | 1311 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1312 | { |
1da177e4 | 1313 | int version; |
e3893534 | 1314 | struct ipc_namespace *ns; |
e1fd1f49 | 1315 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1316 | |
1317 | if (semid < 0) | |
1318 | return -EINVAL; | |
1319 | ||
1320 | version = ipc_parse_version(&cmd); | |
e3893534 | 1321 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
1322 | |
1323 | switch(cmd) { | |
1324 | case IPC_INFO: | |
1325 | case SEM_INFO: | |
4b9fcb0e | 1326 | case IPC_STAT: |
1da177e4 | 1327 | case SEM_STAT: |
e1fd1f49 | 1328 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1329 | case GETALL: |
1330 | case GETVAL: | |
1331 | case GETPID: | |
1332 | case GETNCNT: | |
1333 | case GETZCNT: | |
1da177e4 | 1334 | case SETALL: |
e1fd1f49 AV |
1335 | return semctl_main(ns, semid, semnum, cmd, p); |
1336 | case SETVAL: | |
1337 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1338 | case IPC_RMID: |
1339 | case IPC_SET: | |
e1fd1f49 | 1340 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1341 | default: |
1342 | return -EINVAL; | |
1343 | } | |
1344 | } | |
1345 | ||
1da177e4 LT |
1346 | /* If the task doesn't already have a undo_list, then allocate one |
1347 | * here. We guarantee there is only one thread using this undo list, | |
1348 | * and current is THE ONE | |
1349 | * | |
1350 | * If this allocation and assignment succeeds, but later | |
1351 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1352 | * Just let it stay associated with the task, and it'll be freed later | |
1353 | * at exit time. | |
1354 | * | |
1355 | * This can block, so callers must hold no locks. | |
1356 | */ | |
1357 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1358 | { | |
1359 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1360 | |
1361 | undo_list = current->sysvsem.undo_list; | |
1362 | if (!undo_list) { | |
2453a306 | 1363 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1364 | if (undo_list == NULL) |
1365 | return -ENOMEM; | |
00a5dfdb | 1366 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1367 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1368 | INIT_LIST_HEAD(&undo_list->list_proc); |
1369 | ||
1da177e4 LT |
1370 | current->sysvsem.undo_list = undo_list; |
1371 | } | |
1372 | *undo_listp = undo_list; | |
1373 | return 0; | |
1374 | } | |
1375 | ||
bf17bb71 | 1376 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1377 | { |
bf17bb71 | 1378 | struct sem_undo *un; |
4daa28f6 | 1379 | |
bf17bb71 NP |
1380 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1381 | if (un->semid == semid) | |
1382 | return un; | |
1da177e4 | 1383 | } |
4daa28f6 | 1384 | return NULL; |
1da177e4 LT |
1385 | } |
1386 | ||
bf17bb71 NP |
1387 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1388 | { | |
1389 | struct sem_undo *un; | |
1390 | ||
1391 | assert_spin_locked(&ulp->lock); | |
1392 | ||
1393 | un = __lookup_undo(ulp, semid); | |
1394 | if (un) { | |
1395 | list_del_rcu(&un->list_proc); | |
1396 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1397 | } | |
1398 | return un; | |
1399 | } | |
1400 | ||
4daa28f6 MS |
1401 | /** |
1402 | * find_alloc_undo - Lookup (and if not present create) undo array | |
1403 | * @ns: namespace | |
1404 | * @semid: semaphore array id | |
1405 | * | |
1406 | * The function looks up (and if not present creates) the undo structure. | |
1407 | * The size of the undo structure depends on the size of the semaphore | |
1408 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1409 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1410 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1411 | */ |
1412 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1413 | { |
1414 | struct sem_array *sma; | |
1415 | struct sem_undo_list *ulp; | |
1416 | struct sem_undo *un, *new; | |
6062a8dc | 1417 | int nsems, error; |
1da177e4 LT |
1418 | |
1419 | error = get_undo_list(&ulp); | |
1420 | if (error) | |
1421 | return ERR_PTR(error); | |
1422 | ||
380af1b3 | 1423 | rcu_read_lock(); |
c530c6ac | 1424 | spin_lock(&ulp->lock); |
1da177e4 | 1425 | un = lookup_undo(ulp, semid); |
c530c6ac | 1426 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1427 | if (likely(un!=NULL)) |
1428 | goto out; | |
1429 | ||
1430 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1431 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1432 | sma = sem_obtain_object_check(ns, semid); |
1433 | if (IS_ERR(sma)) { | |
1434 | rcu_read_unlock(); | |
4de85cd6 | 1435 | return ERR_CAST(sma); |
16df3674 | 1436 | } |
023a5355 | 1437 | |
1da177e4 | 1438 | nsems = sma->sem_nsems; |
6062a8dc RR |
1439 | if (!ipc_rcu_getref(sma)) { |
1440 | rcu_read_unlock(); | |
1441 | un = ERR_PTR(-EIDRM); | |
1442 | goto out; | |
1443 | } | |
16df3674 | 1444 | rcu_read_unlock(); |
1da177e4 | 1445 | |
4daa28f6 | 1446 | /* step 2: allocate new undo structure */ |
4668edc3 | 1447 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1448 | if (!new) { |
6ff37972 | 1449 | sem_putref(sma); |
1da177e4 LT |
1450 | return ERR_PTR(-ENOMEM); |
1451 | } | |
1da177e4 | 1452 | |
380af1b3 | 1453 | /* step 3: Acquire the lock on semaphore array */ |
6d49dab8 | 1454 | /* This also does the rcu_read_lock() */ |
6ff37972 | 1455 | sem_lock_and_putref(sma); |
1da177e4 | 1456 | if (sma->sem_perm.deleted) { |
6062a8dc | 1457 | sem_unlock(sma, -1); |
6d49dab8 | 1458 | rcu_read_unlock(); |
1da177e4 LT |
1459 | kfree(new); |
1460 | un = ERR_PTR(-EIDRM); | |
1461 | goto out; | |
1462 | } | |
380af1b3 MS |
1463 | spin_lock(&ulp->lock); |
1464 | ||
1465 | /* | |
1466 | * step 4: check for races: did someone else allocate the undo struct? | |
1467 | */ | |
1468 | un = lookup_undo(ulp, semid); | |
1469 | if (un) { | |
1470 | kfree(new); | |
1471 | goto success; | |
1472 | } | |
4daa28f6 MS |
1473 | /* step 5: initialize & link new undo structure */ |
1474 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1475 | new->ulp = ulp; |
4daa28f6 MS |
1476 | new->semid = semid; |
1477 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1478 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
4daa28f6 MS |
1479 | assert_spin_locked(&sma->sem_perm.lock); |
1480 | list_add(&new->list_id, &sma->list_id); | |
380af1b3 | 1481 | un = new; |
4daa28f6 | 1482 | |
380af1b3 | 1483 | success: |
c530c6ac | 1484 | spin_unlock(&ulp->lock); |
6062a8dc | 1485 | sem_unlock(sma, -1); |
1da177e4 LT |
1486 | out: |
1487 | return un; | |
1488 | } | |
1489 | ||
c61284e9 MS |
1490 | |
1491 | /** | |
1492 | * get_queue_result - Retrieve the result code from sem_queue | |
1493 | * @q: Pointer to queue structure | |
1494 | * | |
1495 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1496 | * q->status, then we must loop until the value is replaced with the final | |
1497 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1498 | * signal) and in parallel the task is woken up by another task because it got | |
1499 | * the requested semaphores. | |
1500 | * | |
1501 | * The function can be called with or without holding the semaphore spinlock. | |
1502 | */ | |
1503 | static int get_queue_result(struct sem_queue *q) | |
1504 | { | |
1505 | int error; | |
1506 | ||
1507 | error = q->status; | |
1508 | while (unlikely(error == IN_WAKEUP)) { | |
1509 | cpu_relax(); | |
1510 | error = q->status; | |
1511 | } | |
1512 | ||
1513 | return error; | |
1514 | } | |
1515 | ||
1516 | ||
d5460c99 HC |
1517 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1518 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1519 | { |
1520 | int error = -EINVAL; | |
1521 | struct sem_array *sma; | |
1522 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1523 | struct sembuf* sops = fast_sops, *sop; | |
1524 | struct sem_undo *un; | |
6062a8dc | 1525 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1526 | struct sem_queue queue; |
1527 | unsigned long jiffies_left = 0; | |
e3893534 | 1528 | struct ipc_namespace *ns; |
0a2b9d4c | 1529 | struct list_head tasks; |
e3893534 KK |
1530 | |
1531 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1532 | |
1533 | if (nsops < 1 || semid < 0) | |
1534 | return -EINVAL; | |
e3893534 | 1535 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1536 | return -E2BIG; |
1537 | if(nsops > SEMOPM_FAST) { | |
1538 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1539 | if(sops==NULL) | |
1540 | return -ENOMEM; | |
1541 | } | |
1542 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1543 | error=-EFAULT; | |
1544 | goto out_free; | |
1545 | } | |
1546 | if (timeout) { | |
1547 | struct timespec _timeout; | |
1548 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1549 | error = -EFAULT; | |
1550 | goto out_free; | |
1551 | } | |
1552 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1553 | _timeout.tv_nsec >= 1000000000L) { | |
1554 | error = -EINVAL; | |
1555 | goto out_free; | |
1556 | } | |
1557 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1558 | } | |
1559 | max = 0; | |
1560 | for (sop = sops; sop < sops + nsops; sop++) { | |
1561 | if (sop->sem_num >= max) | |
1562 | max = sop->sem_num; | |
1563 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1564 | undos = 1; |
1565 | if (sop->sem_op != 0) | |
1da177e4 LT |
1566 | alter = 1; |
1567 | } | |
1da177e4 | 1568 | |
6062a8dc RR |
1569 | INIT_LIST_HEAD(&tasks); |
1570 | ||
1da177e4 | 1571 | if (undos) { |
6062a8dc | 1572 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1573 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1574 | if (IS_ERR(un)) { |
1575 | error = PTR_ERR(un); | |
1576 | goto out_free; | |
1577 | } | |
6062a8dc | 1578 | } else { |
1da177e4 | 1579 | un = NULL; |
6062a8dc RR |
1580 | rcu_read_lock(); |
1581 | } | |
1da177e4 | 1582 | |
16df3674 | 1583 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1584 | if (IS_ERR(sma)) { |
6062a8dc | 1585 | rcu_read_unlock(); |
023a5355 | 1586 | error = PTR_ERR(sma); |
1da177e4 | 1587 | goto out_free; |
023a5355 ND |
1588 | } |
1589 | ||
16df3674 DB |
1590 | error = -EFBIG; |
1591 | if (max >= sma->sem_nsems) { | |
1592 | rcu_read_unlock(); | |
1593 | goto out_wakeup; | |
1594 | } | |
1595 | ||
1596 | error = -EACCES; | |
1597 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) { | |
1598 | rcu_read_unlock(); | |
1599 | goto out_wakeup; | |
1600 | } | |
1601 | ||
1602 | error = security_sem_semop(sma, sops, nsops, alter); | |
1603 | if (error) { | |
1604 | rcu_read_unlock(); | |
1605 | goto out_wakeup; | |
1606 | } | |
1607 | ||
1da177e4 | 1608 | /* |
4daa28f6 | 1609 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1610 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1611 | * and now a new array with received the same id. Check and fail. |
25985edc | 1612 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1613 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1614 | */ |
4daa28f6 | 1615 | error = -EIDRM; |
6062a8dc RR |
1616 | locknum = sem_lock(sma, sops, nsops); |
1617 | if (un && un->semid == -1) | |
1618 | goto out_unlock_free; | |
4daa28f6 | 1619 | |
b488893a | 1620 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1621 | if (error <= 0) { |
1622 | if (alter && error == 0) | |
0a2b9d4c | 1623 | do_smart_update(sma, sops, nsops, 1, &tasks); |
636c6be8 | 1624 | |
1da177e4 LT |
1625 | goto out_unlock_free; |
1626 | } | |
1627 | ||
1628 | /* We need to sleep on this operation, so we put the current | |
1629 | * task into the pending queue and go to sleep. | |
1630 | */ | |
1631 | ||
1da177e4 LT |
1632 | queue.sops = sops; |
1633 | queue.nsops = nsops; | |
1634 | queue.undo = un; | |
b488893a | 1635 | queue.pid = task_tgid_vnr(current); |
1da177e4 | 1636 | queue.alter = alter; |
1da177e4 | 1637 | |
b97e820f MS |
1638 | if (nsops == 1) { |
1639 | struct sem *curr; | |
1640 | curr = &sma->sem_base[sops->sem_num]; | |
1641 | ||
1642 | if (alter) | |
9f1bc2c9 | 1643 | list_add_tail(&queue.list, &curr->sem_pending); |
b97e820f | 1644 | else |
9f1bc2c9 | 1645 | list_add(&queue.list, &curr->sem_pending); |
b97e820f | 1646 | } else { |
9f1bc2c9 RR |
1647 | if (alter) |
1648 | list_add_tail(&queue.list, &sma->sem_pending); | |
1649 | else | |
1650 | list_add(&queue.list, &sma->sem_pending); | |
b97e820f MS |
1651 | sma->complex_count++; |
1652 | } | |
1653 | ||
1da177e4 LT |
1654 | queue.status = -EINTR; |
1655 | queue.sleeper = current; | |
0b0577f6 MS |
1656 | |
1657 | sleep_again: | |
1da177e4 | 1658 | current->state = TASK_INTERRUPTIBLE; |
6062a8dc | 1659 | sem_unlock(sma, locknum); |
6d49dab8 | 1660 | rcu_read_unlock(); |
1da177e4 LT |
1661 | |
1662 | if (timeout) | |
1663 | jiffies_left = schedule_timeout(jiffies_left); | |
1664 | else | |
1665 | schedule(); | |
1666 | ||
c61284e9 | 1667 | error = get_queue_result(&queue); |
1da177e4 LT |
1668 | |
1669 | if (error != -EINTR) { | |
1670 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1671 | * resources. |
1672 | * Perform a smp_mb(): User space could assume that semop() | |
1673 | * is a memory barrier: Without the mb(), the cpu could | |
1674 | * speculatively read in user space stale data that was | |
1675 | * overwritten by the previous owner of the semaphore. | |
1676 | */ | |
1677 | smp_mb(); | |
1678 | ||
1da177e4 LT |
1679 | goto out_free; |
1680 | } | |
1681 | ||
6062a8dc | 1682 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1683 | |
1684 | /* | |
1685 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
1686 | */ | |
1687 | error = get_queue_result(&queue); | |
1688 | ||
1689 | /* | |
1690 | * Array removed? If yes, leave without sem_unlock(). | |
1691 | */ | |
023a5355 | 1692 | if (IS_ERR(sma)) { |
1da177e4 LT |
1693 | goto out_free; |
1694 | } | |
1695 | ||
c61284e9 | 1696 | |
1da177e4 | 1697 | /* |
d694ad62 MS |
1698 | * If queue.status != -EINTR we are woken up by another process. |
1699 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 1700 | */ |
c61284e9 | 1701 | |
1da177e4 LT |
1702 | if (error != -EINTR) { |
1703 | goto out_unlock_free; | |
1704 | } | |
1705 | ||
1706 | /* | |
1707 | * If an interrupt occurred we have to clean up the queue | |
1708 | */ | |
1709 | if (timeout && jiffies_left == 0) | |
1710 | error = -EAGAIN; | |
0b0577f6 MS |
1711 | |
1712 | /* | |
1713 | * If the wakeup was spurious, just retry | |
1714 | */ | |
1715 | if (error == -EINTR && !signal_pending(current)) | |
1716 | goto sleep_again; | |
1717 | ||
b97e820f | 1718 | unlink_queue(sma, &queue); |
1da177e4 LT |
1719 | |
1720 | out_unlock_free: | |
6062a8dc | 1721 | sem_unlock(sma, locknum); |
6d49dab8 | 1722 | rcu_read_unlock(); |
16df3674 | 1723 | out_wakeup: |
0a2b9d4c | 1724 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
1725 | out_free: |
1726 | if(sops != fast_sops) | |
1727 | kfree(sops); | |
1728 | return error; | |
1729 | } | |
1730 | ||
d5460c99 HC |
1731 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
1732 | unsigned, nsops) | |
1da177e4 LT |
1733 | { |
1734 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1735 | } | |
1736 | ||
1737 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1738 | * parent and child tasks. | |
1da177e4 LT |
1739 | */ |
1740 | ||
1741 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1742 | { | |
1743 | struct sem_undo_list *undo_list; | |
1744 | int error; | |
1745 | ||
1746 | if (clone_flags & CLONE_SYSVSEM) { | |
1747 | error = get_undo_list(&undo_list); | |
1748 | if (error) | |
1749 | return error; | |
1da177e4 LT |
1750 | atomic_inc(&undo_list->refcnt); |
1751 | tsk->sysvsem.undo_list = undo_list; | |
1752 | } else | |
1753 | tsk->sysvsem.undo_list = NULL; | |
1754 | ||
1755 | return 0; | |
1756 | } | |
1757 | ||
1758 | /* | |
1759 | * add semadj values to semaphores, free undo structures. | |
1760 | * undo structures are not freed when semaphore arrays are destroyed | |
1761 | * so some of them may be out of date. | |
1762 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1763 | * set of adjustments that needs to be done should be done in an atomic | |
1764 | * manner or not. That is, if we are attempting to decrement the semval | |
1765 | * should we queue up and wait until we can do so legally? | |
1766 | * The original implementation attempted to do this (queue and wait). | |
1767 | * The current implementation does not do so. The POSIX standard | |
1768 | * and SVID should be consulted to determine what behavior is mandated. | |
1769 | */ | |
1770 | void exit_sem(struct task_struct *tsk) | |
1771 | { | |
4daa28f6 | 1772 | struct sem_undo_list *ulp; |
1da177e4 | 1773 | |
4daa28f6 MS |
1774 | ulp = tsk->sysvsem.undo_list; |
1775 | if (!ulp) | |
1da177e4 | 1776 | return; |
9edff4ab | 1777 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 1778 | |
4daa28f6 | 1779 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
1780 | return; |
1781 | ||
380af1b3 | 1782 | for (;;) { |
1da177e4 | 1783 | struct sem_array *sma; |
380af1b3 | 1784 | struct sem_undo *un; |
0a2b9d4c | 1785 | struct list_head tasks; |
6062a8dc | 1786 | int semid, i; |
4daa28f6 | 1787 | |
380af1b3 | 1788 | rcu_read_lock(); |
05725f7e JP |
1789 | un = list_entry_rcu(ulp->list_proc.next, |
1790 | struct sem_undo, list_proc); | |
380af1b3 MS |
1791 | if (&un->list_proc == &ulp->list_proc) |
1792 | semid = -1; | |
1793 | else | |
1794 | semid = un->semid; | |
4daa28f6 | 1795 | |
6062a8dc RR |
1796 | if (semid == -1) { |
1797 | rcu_read_unlock(); | |
380af1b3 | 1798 | break; |
6062a8dc | 1799 | } |
1da177e4 | 1800 | |
6062a8dc | 1801 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid); |
380af1b3 | 1802 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
1803 | if (IS_ERR(sma)) { |
1804 | rcu_read_unlock(); | |
380af1b3 | 1805 | continue; |
6062a8dc | 1806 | } |
1da177e4 | 1807 | |
6062a8dc | 1808 | sem_lock(sma, NULL, -1); |
bf17bb71 | 1809 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
1810 | if (un == NULL) { |
1811 | /* exit_sem raced with IPC_RMID+semget() that created | |
1812 | * exactly the same semid. Nothing to do. | |
1813 | */ | |
6062a8dc | 1814 | sem_unlock(sma, -1); |
6d49dab8 | 1815 | rcu_read_unlock(); |
380af1b3 MS |
1816 | continue; |
1817 | } | |
1818 | ||
1819 | /* remove un from the linked lists */ | |
4daa28f6 MS |
1820 | assert_spin_locked(&sma->sem_perm.lock); |
1821 | list_del(&un->list_id); | |
1822 | ||
380af1b3 MS |
1823 | spin_lock(&ulp->lock); |
1824 | list_del_rcu(&un->list_proc); | |
1825 | spin_unlock(&ulp->lock); | |
1826 | ||
4daa28f6 MS |
1827 | /* perform adjustments registered in un */ |
1828 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 1829 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
1830 | if (un->semadj[i]) { |
1831 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
1832 | /* |
1833 | * Range checks of the new semaphore value, | |
1834 | * not defined by sus: | |
1835 | * - Some unices ignore the undo entirely | |
1836 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
1837 | * - some cap the value (e.g. FreeBSD caps | |
1838 | * at 0, but doesn't enforce SEMVMX) | |
1839 | * | |
1840 | * Linux caps the semaphore value, both at 0 | |
1841 | * and at SEMVMX. | |
1842 | * | |
1843 | * Manfred <manfred@colorfullife.com> | |
1844 | */ | |
5f921ae9 IM |
1845 | if (semaphore->semval < 0) |
1846 | semaphore->semval = 0; | |
1847 | if (semaphore->semval > SEMVMX) | |
1848 | semaphore->semval = SEMVMX; | |
b488893a | 1849 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
1850 | } |
1851 | } | |
1da177e4 | 1852 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
1853 | INIT_LIST_HEAD(&tasks); |
1854 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 1855 | sem_unlock(sma, -1); |
6d49dab8 | 1856 | rcu_read_unlock(); |
0a2b9d4c | 1857 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 1858 | |
693a8b6e | 1859 | kfree_rcu(un, rcu); |
1da177e4 | 1860 | } |
4daa28f6 | 1861 | kfree(ulp); |
1da177e4 LT |
1862 | } |
1863 | ||
1864 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1865 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1866 | { |
1efdb69b | 1867 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c MW |
1868 | struct sem_array *sma = it; |
1869 | ||
1870 | return seq_printf(s, | |
b97e820f | 1871 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 1872 | sma->sem_perm.key, |
7ca7e564 | 1873 | sma->sem_perm.id, |
19b4946c MW |
1874 | sma->sem_perm.mode, |
1875 | sma->sem_nsems, | |
1efdb69b EB |
1876 | from_kuid_munged(user_ns, sma->sem_perm.uid), |
1877 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
1878 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
1879 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
19b4946c MW |
1880 | sma->sem_otime, |
1881 | sma->sem_ctime); | |
1da177e4 LT |
1882 | } |
1883 | #endif |