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