]>
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 */ |
1a82e9e1 MS |
98 | struct list_head pending_alter; /* pending single-sop operations */ |
99 | /* that alter the semaphore */ | |
100 | struct list_head pending_const; /* pending single-sop operations */ | |
101 | /* that do not alter the semaphore*/ | |
f5c936c0 | 102 | } ____cacheline_aligned_in_smp; |
e57940d7 MS |
103 | |
104 | /* One queue for each sleeping process in the system. */ | |
105 | struct sem_queue { | |
e57940d7 MS |
106 | struct list_head list; /* queue of pending operations */ |
107 | struct task_struct *sleeper; /* this process */ | |
108 | struct sem_undo *undo; /* undo structure */ | |
109 | int pid; /* process id of requesting process */ | |
110 | int status; /* completion status of operation */ | |
111 | struct sembuf *sops; /* array of pending operations */ | |
112 | int nsops; /* number of operations */ | |
113 | int alter; /* does *sops alter the array? */ | |
114 | }; | |
115 | ||
116 | /* Each task has a list of undo requests. They are executed automatically | |
117 | * when the process exits. | |
118 | */ | |
119 | struct sem_undo { | |
120 | struct list_head list_proc; /* per-process list: * | |
121 | * all undos from one process | |
122 | * rcu protected */ | |
123 | struct rcu_head rcu; /* rcu struct for sem_undo */ | |
124 | struct sem_undo_list *ulp; /* back ptr to sem_undo_list */ | |
125 | struct list_head list_id; /* per semaphore array list: | |
126 | * all undos for one array */ | |
127 | int semid; /* semaphore set identifier */ | |
128 | short *semadj; /* array of adjustments */ | |
129 | /* one per semaphore */ | |
130 | }; | |
131 | ||
132 | /* sem_undo_list controls shared access to the list of sem_undo structures | |
133 | * that may be shared among all a CLONE_SYSVSEM task group. | |
134 | */ | |
135 | struct sem_undo_list { | |
136 | atomic_t refcnt; | |
137 | spinlock_t lock; | |
138 | struct list_head list_proc; | |
139 | }; | |
140 | ||
141 | ||
ed2ddbf8 | 142 | #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) |
e3893534 | 143 | |
1b531f21 | 144 | #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) |
1da177e4 | 145 | |
7748dbfa | 146 | static int newary(struct ipc_namespace *, struct ipc_params *); |
01b8b07a | 147 | static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); |
1da177e4 | 148 | #ifdef CONFIG_PROC_FS |
19b4946c | 149 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
1da177e4 LT |
150 | #endif |
151 | ||
152 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ | |
153 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ | |
154 | ||
155 | /* | |
156 | * linked list protection: | |
157 | * sem_undo.id_next, | |
1a82e9e1 | 158 | * sem_array.pending{_alter,_cont}, |
1da177e4 LT |
159 | * sem_array.sem_undo: sem_lock() for read/write |
160 | * sem_undo.proc_next: only "current" is allowed to read/write that field. | |
161 | * | |
162 | */ | |
163 | ||
e3893534 KK |
164 | #define sc_semmsl sem_ctls[0] |
165 | #define sc_semmns sem_ctls[1] | |
166 | #define sc_semopm sem_ctls[2] | |
167 | #define sc_semmni sem_ctls[3] | |
168 | ||
ed2ddbf8 | 169 | void sem_init_ns(struct ipc_namespace *ns) |
e3893534 | 170 | { |
e3893534 KK |
171 | ns->sc_semmsl = SEMMSL; |
172 | ns->sc_semmns = SEMMNS; | |
173 | ns->sc_semopm = SEMOPM; | |
174 | ns->sc_semmni = SEMMNI; | |
175 | ns->used_sems = 0; | |
ed2ddbf8 | 176 | ipc_init_ids(&ns->ids[IPC_SEM_IDS]); |
e3893534 KK |
177 | } |
178 | ||
ae5e1b22 | 179 | #ifdef CONFIG_IPC_NS |
e3893534 KK |
180 | void sem_exit_ns(struct ipc_namespace *ns) |
181 | { | |
01b8b07a | 182 | free_ipcs(ns, &sem_ids(ns), freeary); |
7d6feeb2 | 183 | idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); |
e3893534 | 184 | } |
ae5e1b22 | 185 | #endif |
1da177e4 LT |
186 | |
187 | void __init sem_init (void) | |
188 | { | |
ed2ddbf8 | 189 | sem_init_ns(&init_ipc_ns); |
19b4946c MW |
190 | ipc_init_proc_interface("sysvipc/sem", |
191 | " key semid perms nsems uid gid cuid cgid otime ctime\n", | |
e3893534 | 192 | IPC_SEM_IDS, sysvipc_sem_proc_show); |
1da177e4 LT |
193 | } |
194 | ||
6062a8dc RR |
195 | /* |
196 | * If the request contains only one semaphore operation, and there are | |
197 | * no complex transactions pending, lock only the semaphore involved. | |
198 | * Otherwise, lock the entire semaphore array, since we either have | |
199 | * multiple semaphores in our own semops, or we need to look at | |
200 | * semaphores from other pending complex operations. | |
201 | * | |
202 | * Carefully guard against sma->complex_count changing between zero | |
203 | * and non-zero while we are spinning for the lock. The value of | |
204 | * sma->complex_count cannot change while we are holding the lock, | |
205 | * so sem_unlock should be fine. | |
206 | * | |
207 | * The global lock path checks that all the local locks have been released, | |
208 | * checking each local lock once. This means that the local lock paths | |
209 | * cannot start their critical sections while the global lock is held. | |
210 | */ | |
211 | static inline int sem_lock(struct sem_array *sma, struct sembuf *sops, | |
212 | int nsops) | |
213 | { | |
214 | int locknum; | |
215 | again: | |
216 | if (nsops == 1 && !sma->complex_count) { | |
217 | struct sem *sem = sma->sem_base + sops->sem_num; | |
218 | ||
219 | /* Lock just the semaphore we are interested in. */ | |
220 | spin_lock(&sem->lock); | |
221 | ||
222 | /* | |
223 | * If sma->complex_count was set while we were spinning, | |
224 | * we may need to look at things we did not lock here. | |
225 | */ | |
226 | if (unlikely(sma->complex_count)) { | |
227 | spin_unlock(&sem->lock); | |
228 | goto lock_array; | |
229 | } | |
230 | ||
231 | /* | |
232 | * Another process is holding the global lock on the | |
233 | * sem_array; we cannot enter our critical section, | |
234 | * but have to wait for the global lock to be released. | |
235 | */ | |
236 | if (unlikely(spin_is_locked(&sma->sem_perm.lock))) { | |
237 | spin_unlock(&sem->lock); | |
238 | spin_unlock_wait(&sma->sem_perm.lock); | |
239 | goto again; | |
240 | } | |
241 | ||
242 | locknum = sops->sem_num; | |
243 | } else { | |
244 | int i; | |
245 | /* | |
246 | * Lock the semaphore array, and wait for all of the | |
247 | * individual semaphore locks to go away. The code | |
248 | * above ensures no new single-lock holders will enter | |
249 | * their critical section while the array lock is held. | |
250 | */ | |
251 | lock_array: | |
cf9d5d78 | 252 | ipc_lock_object(&sma->sem_perm); |
6062a8dc RR |
253 | for (i = 0; i < sma->sem_nsems; i++) { |
254 | struct sem *sem = sma->sem_base + i; | |
255 | spin_unlock_wait(&sem->lock); | |
256 | } | |
257 | locknum = -1; | |
258 | } | |
259 | return locknum; | |
260 | } | |
261 | ||
262 | static inline void sem_unlock(struct sem_array *sma, int locknum) | |
263 | { | |
264 | if (locknum == -1) { | |
cf9d5d78 | 265 | ipc_unlock_object(&sma->sem_perm); |
6062a8dc RR |
266 | } else { |
267 | struct sem *sem = sma->sem_base + locknum; | |
268 | spin_unlock(&sem->lock); | |
269 | } | |
6062a8dc RR |
270 | } |
271 | ||
3e148c79 ND |
272 | /* |
273 | * sem_lock_(check_) routines are called in the paths where the rw_mutex | |
274 | * is not held. | |
321310ce LT |
275 | * |
276 | * The caller holds the RCU read lock. | |
3e148c79 | 277 | */ |
6062a8dc RR |
278 | static inline struct sem_array *sem_obtain_lock(struct ipc_namespace *ns, |
279 | int id, struct sembuf *sops, int nsops, int *locknum) | |
023a5355 | 280 | { |
c460b662 RR |
281 | struct kern_ipc_perm *ipcp; |
282 | struct sem_array *sma; | |
03f02c76 | 283 | |
c460b662 | 284 | ipcp = ipc_obtain_object(&sem_ids(ns), id); |
321310ce LT |
285 | if (IS_ERR(ipcp)) |
286 | return ERR_CAST(ipcp); | |
b1ed88b4 | 287 | |
6062a8dc RR |
288 | sma = container_of(ipcp, struct sem_array, sem_perm); |
289 | *locknum = sem_lock(sma, sops, nsops); | |
c460b662 RR |
290 | |
291 | /* ipc_rmid() may have already freed the ID while sem_lock | |
292 | * was spinning: verify that the structure is still valid | |
293 | */ | |
294 | if (!ipcp->deleted) | |
295 | return container_of(ipcp, struct sem_array, sem_perm); | |
296 | ||
6062a8dc | 297 | sem_unlock(sma, *locknum); |
321310ce | 298 | return ERR_PTR(-EINVAL); |
023a5355 ND |
299 | } |
300 | ||
16df3674 DB |
301 | static inline struct sem_array *sem_obtain_object(struct ipc_namespace *ns, int id) |
302 | { | |
303 | struct kern_ipc_perm *ipcp = ipc_obtain_object(&sem_ids(ns), id); | |
304 | ||
305 | if (IS_ERR(ipcp)) | |
306 | return ERR_CAST(ipcp); | |
307 | ||
308 | return container_of(ipcp, struct sem_array, sem_perm); | |
309 | } | |
310 | ||
16df3674 DB |
311 | static inline struct sem_array *sem_obtain_object_check(struct ipc_namespace *ns, |
312 | int id) | |
313 | { | |
314 | struct kern_ipc_perm *ipcp = ipc_obtain_object_check(&sem_ids(ns), id); | |
315 | ||
316 | if (IS_ERR(ipcp)) | |
317 | return ERR_CAST(ipcp); | |
b1ed88b4 | 318 | |
03f02c76 | 319 | return container_of(ipcp, struct sem_array, sem_perm); |
023a5355 ND |
320 | } |
321 | ||
6ff37972 PP |
322 | static inline void sem_lock_and_putref(struct sem_array *sma) |
323 | { | |
6062a8dc | 324 | sem_lock(sma, NULL, -1); |
6ff37972 PP |
325 | ipc_rcu_putref(sma); |
326 | } | |
327 | ||
6ff37972 PP |
328 | static inline void sem_putref(struct sem_array *sma) |
329 | { | |
73b29505 | 330 | ipc_rcu_putref(sma); |
6ff37972 PP |
331 | } |
332 | ||
7ca7e564 ND |
333 | static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) |
334 | { | |
335 | ipc_rmid(&sem_ids(ns), &s->sem_perm); | |
336 | } | |
337 | ||
1da177e4 LT |
338 | /* |
339 | * Lockless wakeup algorithm: | |
340 | * Without the check/retry algorithm a lockless wakeup is possible: | |
341 | * - queue.status is initialized to -EINTR before blocking. | |
342 | * - wakeup is performed by | |
1a82e9e1 | 343 | * * unlinking the queue entry from the pending list |
1da177e4 LT |
344 | * * setting queue.status to IN_WAKEUP |
345 | * This is the notification for the blocked thread that a | |
346 | * result value is imminent. | |
347 | * * call wake_up_process | |
348 | * * set queue.status to the final value. | |
349 | * - the previously blocked thread checks queue.status: | |
350 | * * if it's IN_WAKEUP, then it must wait until the value changes | |
351 | * * if it's not -EINTR, then the operation was completed by | |
352 | * update_queue. semtimedop can return queue.status without | |
5f921ae9 | 353 | * performing any operation on the sem array. |
1da177e4 LT |
354 | * * otherwise it must acquire the spinlock and check what's up. |
355 | * | |
356 | * The two-stage algorithm is necessary to protect against the following | |
357 | * races: | |
358 | * - if queue.status is set after wake_up_process, then the woken up idle | |
359 | * thread could race forward and try (and fail) to acquire sma->lock | |
360 | * before update_queue had a chance to set queue.status | |
361 | * - if queue.status is written before wake_up_process and if the | |
362 | * blocked process is woken up by a signal between writing | |
363 | * queue.status and the wake_up_process, then the woken up | |
364 | * process could return from semtimedop and die by calling | |
365 | * sys_exit before wake_up_process is called. Then wake_up_process | |
366 | * will oops, because the task structure is already invalid. | |
367 | * (yes, this happened on s390 with sysv msg). | |
368 | * | |
369 | */ | |
370 | #define IN_WAKEUP 1 | |
371 | ||
f4566f04 ND |
372 | /** |
373 | * newary - Create a new semaphore set | |
374 | * @ns: namespace | |
375 | * @params: ptr to the structure that contains key, semflg and nsems | |
376 | * | |
3e148c79 | 377 | * Called with sem_ids.rw_mutex held (as a writer) |
f4566f04 ND |
378 | */ |
379 | ||
7748dbfa | 380 | static int newary(struct ipc_namespace *ns, struct ipc_params *params) |
1da177e4 LT |
381 | { |
382 | int id; | |
383 | int retval; | |
384 | struct sem_array *sma; | |
385 | int size; | |
7748dbfa ND |
386 | key_t key = params->key; |
387 | int nsems = params->u.nsems; | |
388 | int semflg = params->flg; | |
b97e820f | 389 | int i; |
1da177e4 LT |
390 | |
391 | if (!nsems) | |
392 | return -EINVAL; | |
e3893534 | 393 | if (ns->used_sems + nsems > ns->sc_semmns) |
1da177e4 LT |
394 | return -ENOSPC; |
395 | ||
396 | size = sizeof (*sma) + nsems * sizeof (struct sem); | |
397 | sma = ipc_rcu_alloc(size); | |
398 | if (!sma) { | |
399 | return -ENOMEM; | |
400 | } | |
401 | memset (sma, 0, size); | |
402 | ||
403 | sma->sem_perm.mode = (semflg & S_IRWXUGO); | |
404 | sma->sem_perm.key = key; | |
405 | ||
406 | sma->sem_perm.security = NULL; | |
407 | retval = security_sem_alloc(sma); | |
408 | if (retval) { | |
409 | ipc_rcu_putref(sma); | |
410 | return retval; | |
411 | } | |
412 | ||
e3893534 | 413 | id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); |
283bb7fa | 414 | if (id < 0) { |
1da177e4 LT |
415 | security_sem_free(sma); |
416 | ipc_rcu_putref(sma); | |
283bb7fa | 417 | return id; |
1da177e4 | 418 | } |
e3893534 | 419 | ns->used_sems += nsems; |
1da177e4 LT |
420 | |
421 | sma->sem_base = (struct sem *) &sma[1]; | |
b97e820f | 422 | |
6062a8dc | 423 | for (i = 0; i < nsems; i++) { |
1a82e9e1 MS |
424 | INIT_LIST_HEAD(&sma->sem_base[i].pending_alter); |
425 | INIT_LIST_HEAD(&sma->sem_base[i].pending_const); | |
6062a8dc RR |
426 | spin_lock_init(&sma->sem_base[i].lock); |
427 | } | |
b97e820f MS |
428 | |
429 | sma->complex_count = 0; | |
1a82e9e1 MS |
430 | INIT_LIST_HEAD(&sma->pending_alter); |
431 | INIT_LIST_HEAD(&sma->pending_const); | |
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 | |
1a82e9e1 MS |
617 | * modified the array. |
618 | * Note that wait-for-zero operations are handled without restart. | |
fd5db422 MS |
619 | */ |
620 | static int check_restart(struct sem_array *sma, struct sem_queue *q) | |
621 | { | |
1a82e9e1 MS |
622 | /* pending complex alter operations are too difficult to analyse */ |
623 | if (!list_empty(&sma->pending_alter)) | |
fd5db422 MS |
624 | return 1; |
625 | ||
626 | /* we were a sleeping complex operation. Too difficult */ | |
627 | if (q->nsops > 1) | |
628 | return 1; | |
629 | ||
1a82e9e1 MS |
630 | /* It is impossible that someone waits for the new value: |
631 | * - complex operations always restart. | |
632 | * - wait-for-zero are handled seperately. | |
633 | * - q is a previously sleeping simple operation that | |
634 | * altered the array. It must be a decrement, because | |
635 | * simple increments never sleep. | |
636 | * - If there are older (higher priority) decrements | |
637 | * in the queue, then they have observed the original | |
638 | * semval value and couldn't proceed. The operation | |
639 | * decremented to value - thus they won't proceed either. | |
640 | */ | |
641 | return 0; | |
642 | } | |
fd5db422 | 643 | |
1a82e9e1 MS |
644 | /** |
645 | * wake_const_ops(sma, semnum, pt) - Wake up non-alter tasks | |
646 | * @sma: semaphore array. | |
647 | * @semnum: semaphore that was modified. | |
648 | * @pt: list head for the tasks that must be woken up. | |
649 | * | |
650 | * wake_const_ops must be called after a semaphore in a semaphore array | |
651 | * was set to 0. If complex const operations are pending, wake_const_ops must | |
652 | * be called with semnum = -1, as well as with the number of each modified | |
653 | * semaphore. | |
654 | * The tasks that must be woken up are added to @pt. The return code | |
655 | * is stored in q->pid. | |
656 | * The function returns 1 if at least one operation was completed successfully. | |
657 | */ | |
658 | static int wake_const_ops(struct sem_array *sma, int semnum, | |
659 | struct list_head *pt) | |
660 | { | |
661 | struct sem_queue *q; | |
662 | struct list_head *walk; | |
663 | struct list_head *pending_list; | |
664 | int semop_completed = 0; | |
665 | ||
666 | if (semnum == -1) | |
667 | pending_list = &sma->pending_const; | |
668 | else | |
669 | pending_list = &sma->sem_base[semnum].pending_const; | |
fd5db422 | 670 | |
1a82e9e1 MS |
671 | walk = pending_list->next; |
672 | while (walk != pending_list) { | |
673 | int error; | |
674 | ||
675 | q = container_of(walk, struct sem_queue, list); | |
676 | walk = walk->next; | |
677 | ||
678 | error = try_atomic_semop(sma, q->sops, q->nsops, | |
679 | q->undo, q->pid); | |
680 | ||
681 | if (error <= 0) { | |
682 | /* operation completed, remove from queue & wakeup */ | |
683 | ||
684 | unlink_queue(sma, q); | |
685 | ||
686 | wake_up_sem_queue_prepare(pt, q, error); | |
687 | if (error == 0) | |
688 | semop_completed = 1; | |
689 | } | |
690 | } | |
691 | return semop_completed; | |
692 | } | |
693 | ||
694 | /** | |
695 | * do_smart_wakeup_zero(sma, sops, nsops, pt) - wakeup all wait for zero tasks | |
696 | * @sma: semaphore array | |
697 | * @sops: operations that were performed | |
698 | * @nsops: number of operations | |
699 | * @pt: list head of the tasks that must be woken up. | |
700 | * | |
701 | * do_smart_wakeup_zero() checks all required queue for wait-for-zero | |
702 | * operations, based on the actual changes that were performed on the | |
703 | * semaphore array. | |
704 | * The function returns 1 if at least one operation was completed successfully. | |
705 | */ | |
706 | static int do_smart_wakeup_zero(struct sem_array *sma, struct sembuf *sops, | |
707 | int nsops, struct list_head *pt) | |
708 | { | |
709 | int i; | |
710 | int semop_completed = 0; | |
711 | int got_zero = 0; | |
712 | ||
713 | /* first: the per-semaphore queues, if known */ | |
714 | if (sops) { | |
715 | for (i = 0; i < nsops; i++) { | |
716 | int num = sops[i].sem_num; | |
717 | ||
718 | if (sma->sem_base[num].semval == 0) { | |
719 | got_zero = 1; | |
720 | semop_completed |= wake_const_ops(sma, num, pt); | |
721 | } | |
722 | } | |
723 | } else { | |
724 | /* | |
725 | * No sops means modified semaphores not known. | |
726 | * Assume all were changed. | |
fd5db422 | 727 | */ |
1a82e9e1 MS |
728 | for (i = 0; i < sma->sem_nsems; i++) { |
729 | if (sma->sem_base[i].semval == 0) { | |
730 | got_zero = 1; | |
731 | semop_completed |= wake_const_ops(sma, i, pt); | |
732 | } | |
733 | } | |
fd5db422 MS |
734 | } |
735 | /* | |
1a82e9e1 MS |
736 | * If one of the modified semaphores got 0, |
737 | * then check the global queue, too. | |
fd5db422 | 738 | */ |
1a82e9e1 MS |
739 | if (got_zero) |
740 | semop_completed |= wake_const_ops(sma, -1, pt); | |
fd5db422 | 741 | |
1a82e9e1 | 742 | return semop_completed; |
fd5db422 MS |
743 | } |
744 | ||
636c6be8 MS |
745 | |
746 | /** | |
747 | * update_queue(sma, semnum): Look for tasks that can be completed. | |
748 | * @sma: semaphore array. | |
749 | * @semnum: semaphore that was modified. | |
0a2b9d4c | 750 | * @pt: list head for the tasks that must be woken up. |
636c6be8 MS |
751 | * |
752 | * update_queue must be called after a semaphore in a semaphore array | |
9f1bc2c9 RR |
753 | * was modified. If multiple semaphores were modified, update_queue must |
754 | * be called with semnum = -1, as well as with the number of each modified | |
755 | * semaphore. | |
0a2b9d4c MS |
756 | * The tasks that must be woken up are added to @pt. The return code |
757 | * is stored in q->pid. | |
1a82e9e1 MS |
758 | * The function internally checks if const operations can now succeed. |
759 | * | |
0a2b9d4c | 760 | * The function return 1 if at least one semop was completed successfully. |
1da177e4 | 761 | */ |
0a2b9d4c | 762 | static int update_queue(struct sem_array *sma, int semnum, struct list_head *pt) |
1da177e4 | 763 | { |
636c6be8 MS |
764 | struct sem_queue *q; |
765 | struct list_head *walk; | |
766 | struct list_head *pending_list; | |
0a2b9d4c | 767 | int semop_completed = 0; |
636c6be8 | 768 | |
9f1bc2c9 | 769 | if (semnum == -1) |
1a82e9e1 | 770 | pending_list = &sma->pending_alter; |
9f1bc2c9 | 771 | else |
1a82e9e1 | 772 | pending_list = &sma->sem_base[semnum].pending_alter; |
9cad200c NP |
773 | |
774 | again: | |
636c6be8 MS |
775 | walk = pending_list->next; |
776 | while (walk != pending_list) { | |
fd5db422 | 777 | int error, restart; |
636c6be8 | 778 | |
9f1bc2c9 | 779 | q = container_of(walk, struct sem_queue, list); |
636c6be8 | 780 | walk = walk->next; |
1da177e4 | 781 | |
d987f8b2 MS |
782 | /* If we are scanning the single sop, per-semaphore list of |
783 | * one semaphore and that semaphore is 0, then it is not | |
1a82e9e1 | 784 | * necessary to scan further: simple increments |
d987f8b2 MS |
785 | * that affect only one entry succeed immediately and cannot |
786 | * be in the per semaphore pending queue, and decrements | |
787 | * cannot be successful if the value is already 0. | |
788 | */ | |
1a82e9e1 | 789 | if (semnum != -1 && sma->sem_base[semnum].semval == 0) |
d987f8b2 MS |
790 | break; |
791 | ||
1da177e4 LT |
792 | error = try_atomic_semop(sma, q->sops, q->nsops, |
793 | q->undo, q->pid); | |
794 | ||
795 | /* Does q->sleeper still need to sleep? */ | |
9cad200c NP |
796 | if (error > 0) |
797 | continue; | |
798 | ||
b97e820f | 799 | unlink_queue(sma, q); |
9cad200c | 800 | |
0a2b9d4c | 801 | if (error) { |
fd5db422 | 802 | restart = 0; |
0a2b9d4c MS |
803 | } else { |
804 | semop_completed = 1; | |
1a82e9e1 | 805 | do_smart_wakeup_zero(sma, q->sops, q->nsops, pt); |
fd5db422 | 806 | restart = check_restart(sma, q); |
0a2b9d4c | 807 | } |
fd5db422 | 808 | |
0a2b9d4c | 809 | wake_up_sem_queue_prepare(pt, q, error); |
fd5db422 | 810 | if (restart) |
9cad200c | 811 | goto again; |
1da177e4 | 812 | } |
0a2b9d4c | 813 | return semop_completed; |
1da177e4 LT |
814 | } |
815 | ||
0a2b9d4c MS |
816 | /** |
817 | * do_smart_update(sma, sops, nsops, otime, pt) - optimized update_queue | |
fd5db422 MS |
818 | * @sma: semaphore array |
819 | * @sops: operations that were performed | |
820 | * @nsops: number of operations | |
0a2b9d4c MS |
821 | * @otime: force setting otime |
822 | * @pt: list head of the tasks that must be woken up. | |
fd5db422 | 823 | * |
1a82e9e1 MS |
824 | * do_smart_update() does the required calls to update_queue and wakeup_zero, |
825 | * based on the actual changes that were performed on the semaphore array. | |
0a2b9d4c MS |
826 | * Note that the function does not do the actual wake-up: the caller is |
827 | * responsible for calling wake_up_sem_queue_do(@pt). | |
828 | * It is safe to perform this call after dropping all locks. | |
fd5db422 | 829 | */ |
0a2b9d4c MS |
830 | static void do_smart_update(struct sem_array *sma, struct sembuf *sops, int nsops, |
831 | int otime, struct list_head *pt) | |
fd5db422 MS |
832 | { |
833 | int i; | |
ab465df9 | 834 | int progress; |
fd5db422 | 835 | |
1a82e9e1 MS |
836 | otime |= do_smart_wakeup_zero(sma, sops, nsops, pt); |
837 | ||
ab465df9 MS |
838 | progress = 1; |
839 | retry_global: | |
840 | if (sma->complex_count) { | |
841 | if (update_queue(sma, -1, pt)) { | |
842 | progress = 1; | |
0a2b9d4c | 843 | otime = 1; |
ab465df9 MS |
844 | sops = NULL; |
845 | } | |
9f1bc2c9 | 846 | } |
ab465df9 MS |
847 | if (!progress) |
848 | goto done; | |
9f1bc2c9 RR |
849 | |
850 | if (!sops) { | |
851 | /* No semops; something special is going on. */ | |
852 | for (i = 0; i < sma->sem_nsems; i++) { | |
ab465df9 | 853 | if (update_queue(sma, i, pt)) { |
9f1bc2c9 | 854 | otime = 1; |
ab465df9 MS |
855 | progress = 1; |
856 | } | |
9f1bc2c9 | 857 | } |
ab465df9 | 858 | goto done_checkretry; |
fd5db422 MS |
859 | } |
860 | ||
9f1bc2c9 | 861 | /* Check the semaphores that were modified. */ |
fd5db422 MS |
862 | for (i = 0; i < nsops; i++) { |
863 | if (sops[i].sem_op > 0 || | |
864 | (sops[i].sem_op < 0 && | |
865 | sma->sem_base[sops[i].sem_num].semval == 0)) | |
ab465df9 | 866 | if (update_queue(sma, sops[i].sem_num, pt)) { |
0a2b9d4c | 867 | otime = 1; |
ab465df9 MS |
868 | progress = 1; |
869 | } | |
870 | } | |
871 | done_checkretry: | |
872 | if (progress) { | |
873 | progress = 0; | |
874 | goto retry_global; | |
fd5db422 | 875 | } |
0a2b9d4c MS |
876 | done: |
877 | if (otime) | |
878 | sma->sem_otime = get_seconds(); | |
fd5db422 MS |
879 | } |
880 | ||
881 | ||
1da177e4 LT |
882 | /* The following counts are associated to each semaphore: |
883 | * semncnt number of tasks waiting on semval being nonzero | |
884 | * semzcnt number of tasks waiting on semval being zero | |
885 | * This model assumes that a task waits on exactly one semaphore. | |
886 | * Since semaphore operations are to be performed atomically, tasks actually | |
887 | * wait on a whole sequence of semaphores simultaneously. | |
888 | * The counts we return here are a rough approximation, but still | |
889 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. | |
890 | */ | |
891 | static int count_semncnt (struct sem_array * sma, ushort semnum) | |
892 | { | |
893 | int semncnt; | |
894 | struct sem_queue * q; | |
895 | ||
896 | semncnt = 0; | |
1a82e9e1 | 897 | list_for_each_entry(q, &sma->sem_base[semnum].pending_alter, list) { |
de2657f9 RR |
898 | struct sembuf * sops = q->sops; |
899 | BUG_ON(sops->sem_num != semnum); | |
900 | if ((sops->sem_op < 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
901 | semncnt++; | |
902 | } | |
903 | ||
1a82e9e1 | 904 | list_for_each_entry(q, &sma->pending_alter, list) { |
1da177e4 LT |
905 | struct sembuf * sops = q->sops; |
906 | int nsops = q->nsops; | |
907 | int i; | |
908 | for (i = 0; i < nsops; i++) | |
909 | if (sops[i].sem_num == semnum | |
910 | && (sops[i].sem_op < 0) | |
911 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
912 | semncnt++; | |
913 | } | |
914 | return semncnt; | |
915 | } | |
a1193f8e | 916 | |
1da177e4 LT |
917 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
918 | { | |
919 | int semzcnt; | |
920 | struct sem_queue * q; | |
921 | ||
922 | semzcnt = 0; | |
1a82e9e1 | 923 | list_for_each_entry(q, &sma->sem_base[semnum].pending_const, list) { |
ebc2e5e6 RR |
924 | struct sembuf * sops = q->sops; |
925 | BUG_ON(sops->sem_num != semnum); | |
926 | if ((sops->sem_op == 0) && !(sops->sem_flg & IPC_NOWAIT)) | |
927 | semzcnt++; | |
928 | } | |
929 | ||
1a82e9e1 | 930 | list_for_each_entry(q, &sma->pending_const, list) { |
1da177e4 LT |
931 | struct sembuf * sops = q->sops; |
932 | int nsops = q->nsops; | |
933 | int i; | |
934 | for (i = 0; i < nsops; i++) | |
935 | if (sops[i].sem_num == semnum | |
936 | && (sops[i].sem_op == 0) | |
937 | && !(sops[i].sem_flg & IPC_NOWAIT)) | |
938 | semzcnt++; | |
939 | } | |
940 | return semzcnt; | |
941 | } | |
942 | ||
3e148c79 ND |
943 | /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked |
944 | * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex | |
945 | * remains locked on exit. | |
1da177e4 | 946 | */ |
01b8b07a | 947 | static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) |
1da177e4 | 948 | { |
380af1b3 MS |
949 | struct sem_undo *un, *tu; |
950 | struct sem_queue *q, *tq; | |
01b8b07a | 951 | struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); |
0a2b9d4c | 952 | struct list_head tasks; |
9f1bc2c9 | 953 | int i; |
1da177e4 | 954 | |
380af1b3 | 955 | /* Free the existing undo structures for this semaphore set. */ |
cf9d5d78 | 956 | ipc_assert_locked_object(&sma->sem_perm); |
380af1b3 MS |
957 | list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { |
958 | list_del(&un->list_id); | |
959 | spin_lock(&un->ulp->lock); | |
1da177e4 | 960 | un->semid = -1; |
380af1b3 MS |
961 | list_del_rcu(&un->list_proc); |
962 | spin_unlock(&un->ulp->lock); | |
693a8b6e | 963 | kfree_rcu(un, rcu); |
380af1b3 | 964 | } |
1da177e4 LT |
965 | |
966 | /* Wake up all pending processes and let them fail with EIDRM. */ | |
0a2b9d4c | 967 | INIT_LIST_HEAD(&tasks); |
1a82e9e1 MS |
968 | list_for_each_entry_safe(q, tq, &sma->pending_const, list) { |
969 | unlink_queue(sma, q); | |
970 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
971 | } | |
972 | ||
973 | list_for_each_entry_safe(q, tq, &sma->pending_alter, list) { | |
b97e820f | 974 | unlink_queue(sma, q); |
0a2b9d4c | 975 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); |
1da177e4 | 976 | } |
9f1bc2c9 RR |
977 | for (i = 0; i < sma->sem_nsems; i++) { |
978 | struct sem *sem = sma->sem_base + i; | |
1a82e9e1 MS |
979 | list_for_each_entry_safe(q, tq, &sem->pending_const, list) { |
980 | unlink_queue(sma, q); | |
981 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
982 | } | |
983 | list_for_each_entry_safe(q, tq, &sem->pending_alter, list) { | |
9f1bc2c9 RR |
984 | unlink_queue(sma, q); |
985 | wake_up_sem_queue_prepare(&tasks, q, -EIDRM); | |
986 | } | |
987 | } | |
1da177e4 | 988 | |
7ca7e564 ND |
989 | /* Remove the semaphore set from the IDR */ |
990 | sem_rmid(ns, sma); | |
6062a8dc | 991 | sem_unlock(sma, -1); |
6d49dab8 | 992 | rcu_read_unlock(); |
1da177e4 | 993 | |
0a2b9d4c | 994 | wake_up_sem_queue_do(&tasks); |
e3893534 | 995 | ns->used_sems -= sma->sem_nsems; |
1da177e4 LT |
996 | security_sem_free(sma); |
997 | ipc_rcu_putref(sma); | |
998 | } | |
999 | ||
1000 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) | |
1001 | { | |
1002 | switch(version) { | |
1003 | case IPC_64: | |
1004 | return copy_to_user(buf, in, sizeof(*in)); | |
1005 | case IPC_OLD: | |
1006 | { | |
1007 | struct semid_ds out; | |
1008 | ||
982f7c2b DR |
1009 | memset(&out, 0, sizeof(out)); |
1010 | ||
1da177e4 LT |
1011 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
1012 | ||
1013 | out.sem_otime = in->sem_otime; | |
1014 | out.sem_ctime = in->sem_ctime; | |
1015 | out.sem_nsems = in->sem_nsems; | |
1016 | ||
1017 | return copy_to_user(buf, &out, sizeof(out)); | |
1018 | } | |
1019 | default: | |
1020 | return -EINVAL; | |
1021 | } | |
1022 | } | |
1023 | ||
4b9fcb0e | 1024 | static int semctl_nolock(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1025 | int cmd, int version, void __user *p) |
1da177e4 | 1026 | { |
e5cc9c7b | 1027 | int err; |
1da177e4 LT |
1028 | struct sem_array *sma; |
1029 | ||
1030 | switch(cmd) { | |
1031 | case IPC_INFO: | |
1032 | case SEM_INFO: | |
1033 | { | |
1034 | struct seminfo seminfo; | |
1035 | int max_id; | |
1036 | ||
1037 | err = security_sem_semctl(NULL, cmd); | |
1038 | if (err) | |
1039 | return err; | |
1040 | ||
1041 | memset(&seminfo,0,sizeof(seminfo)); | |
e3893534 KK |
1042 | seminfo.semmni = ns->sc_semmni; |
1043 | seminfo.semmns = ns->sc_semmns; | |
1044 | seminfo.semmsl = ns->sc_semmsl; | |
1045 | seminfo.semopm = ns->sc_semopm; | |
1da177e4 LT |
1046 | seminfo.semvmx = SEMVMX; |
1047 | seminfo.semmnu = SEMMNU; | |
1048 | seminfo.semmap = SEMMAP; | |
1049 | seminfo.semume = SEMUME; | |
3e148c79 | 1050 | down_read(&sem_ids(ns).rw_mutex); |
1da177e4 | 1051 | if (cmd == SEM_INFO) { |
e3893534 KK |
1052 | seminfo.semusz = sem_ids(ns).in_use; |
1053 | seminfo.semaem = ns->used_sems; | |
1da177e4 LT |
1054 | } else { |
1055 | seminfo.semusz = SEMUSZ; | |
1056 | seminfo.semaem = SEMAEM; | |
1057 | } | |
7ca7e564 | 1058 | max_id = ipc_get_maxid(&sem_ids(ns)); |
3e148c79 | 1059 | up_read(&sem_ids(ns).rw_mutex); |
e1fd1f49 | 1060 | if (copy_to_user(p, &seminfo, sizeof(struct seminfo))) |
1da177e4 LT |
1061 | return -EFAULT; |
1062 | return (max_id < 0) ? 0: max_id; | |
1063 | } | |
4b9fcb0e | 1064 | case IPC_STAT: |
1da177e4 LT |
1065 | case SEM_STAT: |
1066 | { | |
1067 | struct semid64_ds tbuf; | |
16df3674 DB |
1068 | int id = 0; |
1069 | ||
1070 | memset(&tbuf, 0, sizeof(tbuf)); | |
1da177e4 | 1071 | |
941b0304 | 1072 | rcu_read_lock(); |
4b9fcb0e | 1073 | if (cmd == SEM_STAT) { |
16df3674 DB |
1074 | sma = sem_obtain_object(ns, semid); |
1075 | if (IS_ERR(sma)) { | |
1076 | err = PTR_ERR(sma); | |
1077 | goto out_unlock; | |
1078 | } | |
4b9fcb0e PP |
1079 | id = sma->sem_perm.id; |
1080 | } else { | |
16df3674 DB |
1081 | sma = sem_obtain_object_check(ns, semid); |
1082 | if (IS_ERR(sma)) { | |
1083 | err = PTR_ERR(sma); | |
1084 | goto out_unlock; | |
1085 | } | |
4b9fcb0e | 1086 | } |
1da177e4 LT |
1087 | |
1088 | err = -EACCES; | |
b0e77598 | 1089 | if (ipcperms(ns, &sma->sem_perm, S_IRUGO)) |
1da177e4 LT |
1090 | goto out_unlock; |
1091 | ||
1092 | err = security_sem_semctl(sma, cmd); | |
1093 | if (err) | |
1094 | goto out_unlock; | |
1095 | ||
1da177e4 LT |
1096 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
1097 | tbuf.sem_otime = sma->sem_otime; | |
1098 | tbuf.sem_ctime = sma->sem_ctime; | |
1099 | tbuf.sem_nsems = sma->sem_nsems; | |
16df3674 | 1100 | rcu_read_unlock(); |
e1fd1f49 | 1101 | if (copy_semid_to_user(p, &tbuf, version)) |
1da177e4 LT |
1102 | return -EFAULT; |
1103 | return id; | |
1104 | } | |
1105 | default: | |
1106 | return -EINVAL; | |
1107 | } | |
1da177e4 | 1108 | out_unlock: |
16df3674 | 1109 | rcu_read_unlock(); |
1da177e4 LT |
1110 | return err; |
1111 | } | |
1112 | ||
e1fd1f49 AV |
1113 | static int semctl_setval(struct ipc_namespace *ns, int semid, int semnum, |
1114 | unsigned long arg) | |
1115 | { | |
1116 | struct sem_undo *un; | |
1117 | struct sem_array *sma; | |
1118 | struct sem* curr; | |
1119 | int err; | |
e1fd1f49 AV |
1120 | struct list_head tasks; |
1121 | int val; | |
1122 | #if defined(CONFIG_64BIT) && defined(__BIG_ENDIAN) | |
1123 | /* big-endian 64bit */ | |
1124 | val = arg >> 32; | |
1125 | #else | |
1126 | /* 32bit or little-endian 64bit */ | |
1127 | val = arg; | |
1128 | #endif | |
1129 | ||
6062a8dc RR |
1130 | if (val > SEMVMX || val < 0) |
1131 | return -ERANGE; | |
e1fd1f49 AV |
1132 | |
1133 | INIT_LIST_HEAD(&tasks); | |
e1fd1f49 | 1134 | |
6062a8dc RR |
1135 | rcu_read_lock(); |
1136 | sma = sem_obtain_object_check(ns, semid); | |
1137 | if (IS_ERR(sma)) { | |
1138 | rcu_read_unlock(); | |
1139 | return PTR_ERR(sma); | |
1140 | } | |
1141 | ||
1142 | if (semnum < 0 || semnum >= sma->sem_nsems) { | |
1143 | rcu_read_unlock(); | |
1144 | return -EINVAL; | |
1145 | } | |
1146 | ||
1147 | ||
1148 | if (ipcperms(ns, &sma->sem_perm, S_IWUGO)) { | |
1149 | rcu_read_unlock(); | |
1150 | return -EACCES; | |
1151 | } | |
e1fd1f49 AV |
1152 | |
1153 | err = security_sem_semctl(sma, SETVAL); | |
6062a8dc RR |
1154 | if (err) { |
1155 | rcu_read_unlock(); | |
1156 | return -EACCES; | |
1157 | } | |
e1fd1f49 | 1158 | |
6062a8dc | 1159 | sem_lock(sma, NULL, -1); |
e1fd1f49 AV |
1160 | |
1161 | curr = &sma->sem_base[semnum]; | |
1162 | ||
cf9d5d78 | 1163 | ipc_assert_locked_object(&sma->sem_perm); |
e1fd1f49 AV |
1164 | list_for_each_entry(un, &sma->list_id, list_id) |
1165 | un->semadj[semnum] = 0; | |
1166 | ||
1167 | curr->semval = val; | |
1168 | curr->sempid = task_tgid_vnr(current); | |
1169 | sma->sem_ctime = get_seconds(); | |
1170 | /* maybe some queued-up processes were waiting for this */ | |
1171 | do_smart_update(sma, NULL, 0, 0, &tasks); | |
6062a8dc | 1172 | sem_unlock(sma, -1); |
6d49dab8 | 1173 | rcu_read_unlock(); |
e1fd1f49 | 1174 | wake_up_sem_queue_do(&tasks); |
6062a8dc | 1175 | return 0; |
e1fd1f49 AV |
1176 | } |
1177 | ||
e3893534 | 1178 | static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, |
e1fd1f49 | 1179 | int cmd, void __user *p) |
1da177e4 LT |
1180 | { |
1181 | struct sem_array *sma; | |
1182 | struct sem* curr; | |
16df3674 | 1183 | int err, nsems; |
1da177e4 LT |
1184 | ushort fast_sem_io[SEMMSL_FAST]; |
1185 | ushort* sem_io = fast_sem_io; | |
0a2b9d4c | 1186 | struct list_head tasks; |
1da177e4 | 1187 | |
16df3674 DB |
1188 | INIT_LIST_HEAD(&tasks); |
1189 | ||
1190 | rcu_read_lock(); | |
1191 | sma = sem_obtain_object_check(ns, semid); | |
1192 | if (IS_ERR(sma)) { | |
1193 | rcu_read_unlock(); | |
023a5355 | 1194 | return PTR_ERR(sma); |
16df3674 | 1195 | } |
1da177e4 LT |
1196 | |
1197 | nsems = sma->sem_nsems; | |
1198 | ||
1da177e4 | 1199 | err = -EACCES; |
c728b9c8 LT |
1200 | if (ipcperms(ns, &sma->sem_perm, cmd == SETALL ? S_IWUGO : S_IRUGO)) |
1201 | goto out_rcu_wakeup; | |
1da177e4 LT |
1202 | |
1203 | err = security_sem_semctl(sma, cmd); | |
c728b9c8 LT |
1204 | if (err) |
1205 | goto out_rcu_wakeup; | |
1da177e4 LT |
1206 | |
1207 | err = -EACCES; | |
1208 | switch (cmd) { | |
1209 | case GETALL: | |
1210 | { | |
e1fd1f49 | 1211 | ushort __user *array = p; |
1da177e4 LT |
1212 | int i; |
1213 | ||
ce857229 | 1214 | sem_lock(sma, NULL, -1); |
1da177e4 | 1215 | if(nsems > SEMMSL_FAST) { |
ce857229 AV |
1216 | if (!ipc_rcu_getref(sma)) { |
1217 | sem_unlock(sma, -1); | |
6d49dab8 | 1218 | rcu_read_unlock(); |
ce857229 AV |
1219 | err = -EIDRM; |
1220 | goto out_free; | |
1221 | } | |
1222 | sem_unlock(sma, -1); | |
6d49dab8 | 1223 | rcu_read_unlock(); |
1da177e4 LT |
1224 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
1225 | if(sem_io == NULL) { | |
6ff37972 | 1226 | sem_putref(sma); |
1da177e4 LT |
1227 | return -ENOMEM; |
1228 | } | |
1229 | ||
4091fd94 | 1230 | rcu_read_lock(); |
6ff37972 | 1231 | sem_lock_and_putref(sma); |
1da177e4 | 1232 | if (sma->sem_perm.deleted) { |
6062a8dc | 1233 | sem_unlock(sma, -1); |
6d49dab8 | 1234 | rcu_read_unlock(); |
1da177e4 LT |
1235 | err = -EIDRM; |
1236 | goto out_free; | |
1237 | } | |
ce857229 | 1238 | } |
1da177e4 LT |
1239 | for (i = 0; i < sma->sem_nsems; i++) |
1240 | sem_io[i] = sma->sem_base[i].semval; | |
6062a8dc | 1241 | sem_unlock(sma, -1); |
6d49dab8 | 1242 | rcu_read_unlock(); |
1da177e4 LT |
1243 | err = 0; |
1244 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) | |
1245 | err = -EFAULT; | |
1246 | goto out_free; | |
1247 | } | |
1248 | case SETALL: | |
1249 | { | |
1250 | int i; | |
1251 | struct sem_undo *un; | |
1252 | ||
6062a8dc RR |
1253 | if (!ipc_rcu_getref(sma)) { |
1254 | rcu_read_unlock(); | |
1255 | return -EIDRM; | |
1256 | } | |
16df3674 | 1257 | rcu_read_unlock(); |
1da177e4 LT |
1258 | |
1259 | if(nsems > SEMMSL_FAST) { | |
1260 | sem_io = ipc_alloc(sizeof(ushort)*nsems); | |
1261 | if(sem_io == NULL) { | |
6ff37972 | 1262 | sem_putref(sma); |
1da177e4 LT |
1263 | return -ENOMEM; |
1264 | } | |
1265 | } | |
1266 | ||
e1fd1f49 | 1267 | if (copy_from_user (sem_io, p, nsems*sizeof(ushort))) { |
6ff37972 | 1268 | sem_putref(sma); |
1da177e4 LT |
1269 | err = -EFAULT; |
1270 | goto out_free; | |
1271 | } | |
1272 | ||
1273 | for (i = 0; i < nsems; i++) { | |
1274 | if (sem_io[i] > SEMVMX) { | |
6ff37972 | 1275 | sem_putref(sma); |
1da177e4 LT |
1276 | err = -ERANGE; |
1277 | goto out_free; | |
1278 | } | |
1279 | } | |
4091fd94 | 1280 | rcu_read_lock(); |
6ff37972 | 1281 | sem_lock_and_putref(sma); |
1da177e4 | 1282 | if (sma->sem_perm.deleted) { |
6062a8dc | 1283 | sem_unlock(sma, -1); |
6d49dab8 | 1284 | rcu_read_unlock(); |
1da177e4 LT |
1285 | err = -EIDRM; |
1286 | goto out_free; | |
1287 | } | |
1288 | ||
1289 | for (i = 0; i < nsems; i++) | |
1290 | sma->sem_base[i].semval = sem_io[i]; | |
4daa28f6 | 1291 | |
cf9d5d78 | 1292 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1293 | list_for_each_entry(un, &sma->list_id, list_id) { |
1da177e4 LT |
1294 | for (i = 0; i < nsems; i++) |
1295 | un->semadj[i] = 0; | |
4daa28f6 | 1296 | } |
1da177e4 LT |
1297 | sma->sem_ctime = get_seconds(); |
1298 | /* maybe some queued-up processes were waiting for this */ | |
0a2b9d4c | 1299 | do_smart_update(sma, NULL, 0, 0, &tasks); |
1da177e4 LT |
1300 | err = 0; |
1301 | goto out_unlock; | |
1302 | } | |
e1fd1f49 | 1303 | /* GETVAL, GETPID, GETNCTN, GETZCNT: fall-through */ |
1da177e4 LT |
1304 | } |
1305 | err = -EINVAL; | |
c728b9c8 LT |
1306 | if (semnum < 0 || semnum >= nsems) |
1307 | goto out_rcu_wakeup; | |
1da177e4 | 1308 | |
6062a8dc | 1309 | sem_lock(sma, NULL, -1); |
1da177e4 LT |
1310 | curr = &sma->sem_base[semnum]; |
1311 | ||
1312 | switch (cmd) { | |
1313 | case GETVAL: | |
1314 | err = curr->semval; | |
1315 | goto out_unlock; | |
1316 | case GETPID: | |
1317 | err = curr->sempid; | |
1318 | goto out_unlock; | |
1319 | case GETNCNT: | |
1320 | err = count_semncnt(sma,semnum); | |
1321 | goto out_unlock; | |
1322 | case GETZCNT: | |
1323 | err = count_semzcnt(sma,semnum); | |
1324 | goto out_unlock; | |
1da177e4 | 1325 | } |
16df3674 | 1326 | |
1da177e4 | 1327 | out_unlock: |
6062a8dc | 1328 | sem_unlock(sma, -1); |
c728b9c8 | 1329 | out_rcu_wakeup: |
6d49dab8 | 1330 | rcu_read_unlock(); |
0a2b9d4c | 1331 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
1332 | out_free: |
1333 | if(sem_io != fast_sem_io) | |
1334 | ipc_free(sem_io, sizeof(ushort)*nsems); | |
1335 | return err; | |
1336 | } | |
1337 | ||
016d7132 PP |
1338 | static inline unsigned long |
1339 | copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) | |
1da177e4 LT |
1340 | { |
1341 | switch(version) { | |
1342 | case IPC_64: | |
016d7132 | 1343 | if (copy_from_user(out, buf, sizeof(*out))) |
1da177e4 | 1344 | return -EFAULT; |
1da177e4 | 1345 | return 0; |
1da177e4 LT |
1346 | case IPC_OLD: |
1347 | { | |
1348 | struct semid_ds tbuf_old; | |
1349 | ||
1350 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) | |
1351 | return -EFAULT; | |
1352 | ||
016d7132 PP |
1353 | out->sem_perm.uid = tbuf_old.sem_perm.uid; |
1354 | out->sem_perm.gid = tbuf_old.sem_perm.gid; | |
1355 | out->sem_perm.mode = tbuf_old.sem_perm.mode; | |
1da177e4 LT |
1356 | |
1357 | return 0; | |
1358 | } | |
1359 | default: | |
1360 | return -EINVAL; | |
1361 | } | |
1362 | } | |
1363 | ||
522bb2a2 PP |
1364 | /* |
1365 | * This function handles some semctl commands which require the rw_mutex | |
1366 | * to be held in write mode. | |
1367 | * NOTE: no locks must be held, the rw_mutex is taken inside this function. | |
1368 | */ | |
21a4826a | 1369 | static int semctl_down(struct ipc_namespace *ns, int semid, |
e1fd1f49 | 1370 | int cmd, int version, void __user *p) |
1da177e4 LT |
1371 | { |
1372 | struct sem_array *sma; | |
1373 | int err; | |
016d7132 | 1374 | struct semid64_ds semid64; |
1da177e4 LT |
1375 | struct kern_ipc_perm *ipcp; |
1376 | ||
1377 | if(cmd == IPC_SET) { | |
e1fd1f49 | 1378 | if (copy_semid_from_user(&semid64, p, version)) |
1da177e4 | 1379 | return -EFAULT; |
1da177e4 | 1380 | } |
073115d6 | 1381 | |
7b4cc5d8 DB |
1382 | down_write(&sem_ids(ns).rw_mutex); |
1383 | rcu_read_lock(); | |
1384 | ||
16df3674 DB |
1385 | ipcp = ipcctl_pre_down_nolock(ns, &sem_ids(ns), semid, cmd, |
1386 | &semid64.sem_perm, 0); | |
7b4cc5d8 DB |
1387 | if (IS_ERR(ipcp)) { |
1388 | err = PTR_ERR(ipcp); | |
7b4cc5d8 DB |
1389 | goto out_unlock1; |
1390 | } | |
073115d6 | 1391 | |
a5f75e7f | 1392 | sma = container_of(ipcp, struct sem_array, sem_perm); |
1da177e4 LT |
1393 | |
1394 | err = security_sem_semctl(sma, cmd); | |
7b4cc5d8 DB |
1395 | if (err) |
1396 | goto out_unlock1; | |
1da177e4 | 1397 | |
7b4cc5d8 | 1398 | switch (cmd) { |
1da177e4 | 1399 | case IPC_RMID: |
6062a8dc | 1400 | sem_lock(sma, NULL, -1); |
7b4cc5d8 | 1401 | /* freeary unlocks the ipc object and rcu */ |
01b8b07a | 1402 | freeary(ns, ipcp); |
522bb2a2 | 1403 | goto out_up; |
1da177e4 | 1404 | case IPC_SET: |
6062a8dc | 1405 | sem_lock(sma, NULL, -1); |
1efdb69b EB |
1406 | err = ipc_update_perm(&semid64.sem_perm, ipcp); |
1407 | if (err) | |
7b4cc5d8 | 1408 | goto out_unlock0; |
1da177e4 | 1409 | sma->sem_ctime = get_seconds(); |
1da177e4 LT |
1410 | break; |
1411 | default: | |
1da177e4 | 1412 | err = -EINVAL; |
7b4cc5d8 | 1413 | goto out_unlock1; |
1da177e4 | 1414 | } |
1da177e4 | 1415 | |
7b4cc5d8 | 1416 | out_unlock0: |
6062a8dc | 1417 | sem_unlock(sma, -1); |
7b4cc5d8 | 1418 | out_unlock1: |
6d49dab8 | 1419 | rcu_read_unlock(); |
522bb2a2 PP |
1420 | out_up: |
1421 | up_write(&sem_ids(ns).rw_mutex); | |
1da177e4 LT |
1422 | return err; |
1423 | } | |
1424 | ||
e1fd1f49 | 1425 | SYSCALL_DEFINE4(semctl, int, semid, int, semnum, int, cmd, unsigned long, arg) |
1da177e4 | 1426 | { |
1da177e4 | 1427 | int version; |
e3893534 | 1428 | struct ipc_namespace *ns; |
e1fd1f49 | 1429 | void __user *p = (void __user *)arg; |
1da177e4 LT |
1430 | |
1431 | if (semid < 0) | |
1432 | return -EINVAL; | |
1433 | ||
1434 | version = ipc_parse_version(&cmd); | |
e3893534 | 1435 | ns = current->nsproxy->ipc_ns; |
1da177e4 LT |
1436 | |
1437 | switch(cmd) { | |
1438 | case IPC_INFO: | |
1439 | case SEM_INFO: | |
4b9fcb0e | 1440 | case IPC_STAT: |
1da177e4 | 1441 | case SEM_STAT: |
e1fd1f49 | 1442 | return semctl_nolock(ns, semid, cmd, version, p); |
1da177e4 LT |
1443 | case GETALL: |
1444 | case GETVAL: | |
1445 | case GETPID: | |
1446 | case GETNCNT: | |
1447 | case GETZCNT: | |
1da177e4 | 1448 | case SETALL: |
e1fd1f49 AV |
1449 | return semctl_main(ns, semid, semnum, cmd, p); |
1450 | case SETVAL: | |
1451 | return semctl_setval(ns, semid, semnum, arg); | |
1da177e4 LT |
1452 | case IPC_RMID: |
1453 | case IPC_SET: | |
e1fd1f49 | 1454 | return semctl_down(ns, semid, cmd, version, p); |
1da177e4 LT |
1455 | default: |
1456 | return -EINVAL; | |
1457 | } | |
1458 | } | |
1459 | ||
1da177e4 LT |
1460 | /* If the task doesn't already have a undo_list, then allocate one |
1461 | * here. We guarantee there is only one thread using this undo list, | |
1462 | * and current is THE ONE | |
1463 | * | |
1464 | * If this allocation and assignment succeeds, but later | |
1465 | * portions of this code fail, there is no need to free the sem_undo_list. | |
1466 | * Just let it stay associated with the task, and it'll be freed later | |
1467 | * at exit time. | |
1468 | * | |
1469 | * This can block, so callers must hold no locks. | |
1470 | */ | |
1471 | static inline int get_undo_list(struct sem_undo_list **undo_listp) | |
1472 | { | |
1473 | struct sem_undo_list *undo_list; | |
1da177e4 LT |
1474 | |
1475 | undo_list = current->sysvsem.undo_list; | |
1476 | if (!undo_list) { | |
2453a306 | 1477 | undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); |
1da177e4 LT |
1478 | if (undo_list == NULL) |
1479 | return -ENOMEM; | |
00a5dfdb | 1480 | spin_lock_init(&undo_list->lock); |
1da177e4 | 1481 | atomic_set(&undo_list->refcnt, 1); |
4daa28f6 MS |
1482 | INIT_LIST_HEAD(&undo_list->list_proc); |
1483 | ||
1da177e4 LT |
1484 | current->sysvsem.undo_list = undo_list; |
1485 | } | |
1486 | *undo_listp = undo_list; | |
1487 | return 0; | |
1488 | } | |
1489 | ||
bf17bb71 | 1490 | static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) |
1da177e4 | 1491 | { |
bf17bb71 | 1492 | struct sem_undo *un; |
4daa28f6 | 1493 | |
bf17bb71 NP |
1494 | list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { |
1495 | if (un->semid == semid) | |
1496 | return un; | |
1da177e4 | 1497 | } |
4daa28f6 | 1498 | return NULL; |
1da177e4 LT |
1499 | } |
1500 | ||
bf17bb71 NP |
1501 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
1502 | { | |
1503 | struct sem_undo *un; | |
1504 | ||
1505 | assert_spin_locked(&ulp->lock); | |
1506 | ||
1507 | un = __lookup_undo(ulp, semid); | |
1508 | if (un) { | |
1509 | list_del_rcu(&un->list_proc); | |
1510 | list_add_rcu(&un->list_proc, &ulp->list_proc); | |
1511 | } | |
1512 | return un; | |
1513 | } | |
1514 | ||
4daa28f6 MS |
1515 | /** |
1516 | * find_alloc_undo - Lookup (and if not present create) undo array | |
1517 | * @ns: namespace | |
1518 | * @semid: semaphore array id | |
1519 | * | |
1520 | * The function looks up (and if not present creates) the undo structure. | |
1521 | * The size of the undo structure depends on the size of the semaphore | |
1522 | * array, thus the alloc path is not that straightforward. | |
380af1b3 MS |
1523 | * Lifetime-rules: sem_undo is rcu-protected, on success, the function |
1524 | * performs a rcu_read_lock(). | |
4daa28f6 MS |
1525 | */ |
1526 | static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) | |
1da177e4 LT |
1527 | { |
1528 | struct sem_array *sma; | |
1529 | struct sem_undo_list *ulp; | |
1530 | struct sem_undo *un, *new; | |
6062a8dc | 1531 | int nsems, error; |
1da177e4 LT |
1532 | |
1533 | error = get_undo_list(&ulp); | |
1534 | if (error) | |
1535 | return ERR_PTR(error); | |
1536 | ||
380af1b3 | 1537 | rcu_read_lock(); |
c530c6ac | 1538 | spin_lock(&ulp->lock); |
1da177e4 | 1539 | un = lookup_undo(ulp, semid); |
c530c6ac | 1540 | spin_unlock(&ulp->lock); |
1da177e4 LT |
1541 | if (likely(un!=NULL)) |
1542 | goto out; | |
1543 | ||
1544 | /* no undo structure around - allocate one. */ | |
4daa28f6 | 1545 | /* step 1: figure out the size of the semaphore array */ |
16df3674 DB |
1546 | sma = sem_obtain_object_check(ns, semid); |
1547 | if (IS_ERR(sma)) { | |
1548 | rcu_read_unlock(); | |
4de85cd6 | 1549 | return ERR_CAST(sma); |
16df3674 | 1550 | } |
023a5355 | 1551 | |
1da177e4 | 1552 | nsems = sma->sem_nsems; |
6062a8dc RR |
1553 | if (!ipc_rcu_getref(sma)) { |
1554 | rcu_read_unlock(); | |
1555 | un = ERR_PTR(-EIDRM); | |
1556 | goto out; | |
1557 | } | |
16df3674 | 1558 | rcu_read_unlock(); |
1da177e4 | 1559 | |
4daa28f6 | 1560 | /* step 2: allocate new undo structure */ |
4668edc3 | 1561 | new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
1da177e4 | 1562 | if (!new) { |
6ff37972 | 1563 | sem_putref(sma); |
1da177e4 LT |
1564 | return ERR_PTR(-ENOMEM); |
1565 | } | |
1da177e4 | 1566 | |
380af1b3 | 1567 | /* step 3: Acquire the lock on semaphore array */ |
4091fd94 | 1568 | rcu_read_lock(); |
6ff37972 | 1569 | sem_lock_and_putref(sma); |
1da177e4 | 1570 | if (sma->sem_perm.deleted) { |
6062a8dc | 1571 | sem_unlock(sma, -1); |
6d49dab8 | 1572 | rcu_read_unlock(); |
1da177e4 LT |
1573 | kfree(new); |
1574 | un = ERR_PTR(-EIDRM); | |
1575 | goto out; | |
1576 | } | |
380af1b3 MS |
1577 | spin_lock(&ulp->lock); |
1578 | ||
1579 | /* | |
1580 | * step 4: check for races: did someone else allocate the undo struct? | |
1581 | */ | |
1582 | un = lookup_undo(ulp, semid); | |
1583 | if (un) { | |
1584 | kfree(new); | |
1585 | goto success; | |
1586 | } | |
4daa28f6 MS |
1587 | /* step 5: initialize & link new undo structure */ |
1588 | new->semadj = (short *) &new[1]; | |
380af1b3 | 1589 | new->ulp = ulp; |
4daa28f6 MS |
1590 | new->semid = semid; |
1591 | assert_spin_locked(&ulp->lock); | |
380af1b3 | 1592 | list_add_rcu(&new->list_proc, &ulp->list_proc); |
cf9d5d78 | 1593 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 | 1594 | list_add(&new->list_id, &sma->list_id); |
380af1b3 | 1595 | un = new; |
4daa28f6 | 1596 | |
380af1b3 | 1597 | success: |
c530c6ac | 1598 | spin_unlock(&ulp->lock); |
6062a8dc | 1599 | sem_unlock(sma, -1); |
1da177e4 LT |
1600 | out: |
1601 | return un; | |
1602 | } | |
1603 | ||
c61284e9 MS |
1604 | |
1605 | /** | |
1606 | * get_queue_result - Retrieve the result code from sem_queue | |
1607 | * @q: Pointer to queue structure | |
1608 | * | |
1609 | * Retrieve the return code from the pending queue. If IN_WAKEUP is found in | |
1610 | * q->status, then we must loop until the value is replaced with the final | |
1611 | * value: This may happen if a task is woken up by an unrelated event (e.g. | |
1612 | * signal) and in parallel the task is woken up by another task because it got | |
1613 | * the requested semaphores. | |
1614 | * | |
1615 | * The function can be called with or without holding the semaphore spinlock. | |
1616 | */ | |
1617 | static int get_queue_result(struct sem_queue *q) | |
1618 | { | |
1619 | int error; | |
1620 | ||
1621 | error = q->status; | |
1622 | while (unlikely(error == IN_WAKEUP)) { | |
1623 | cpu_relax(); | |
1624 | error = q->status; | |
1625 | } | |
1626 | ||
1627 | return error; | |
1628 | } | |
1629 | ||
1630 | ||
d5460c99 HC |
1631 | SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, |
1632 | unsigned, nsops, const struct timespec __user *, timeout) | |
1da177e4 LT |
1633 | { |
1634 | int error = -EINVAL; | |
1635 | struct sem_array *sma; | |
1636 | struct sembuf fast_sops[SEMOPM_FAST]; | |
1637 | struct sembuf* sops = fast_sops, *sop; | |
1638 | struct sem_undo *un; | |
6062a8dc | 1639 | int undos = 0, alter = 0, max, locknum; |
1da177e4 LT |
1640 | struct sem_queue queue; |
1641 | unsigned long jiffies_left = 0; | |
e3893534 | 1642 | struct ipc_namespace *ns; |
0a2b9d4c | 1643 | struct list_head tasks; |
e3893534 KK |
1644 | |
1645 | ns = current->nsproxy->ipc_ns; | |
1da177e4 LT |
1646 | |
1647 | if (nsops < 1 || semid < 0) | |
1648 | return -EINVAL; | |
e3893534 | 1649 | if (nsops > ns->sc_semopm) |
1da177e4 LT |
1650 | return -E2BIG; |
1651 | if(nsops > SEMOPM_FAST) { | |
1652 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); | |
1653 | if(sops==NULL) | |
1654 | return -ENOMEM; | |
1655 | } | |
1656 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { | |
1657 | error=-EFAULT; | |
1658 | goto out_free; | |
1659 | } | |
1660 | if (timeout) { | |
1661 | struct timespec _timeout; | |
1662 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { | |
1663 | error = -EFAULT; | |
1664 | goto out_free; | |
1665 | } | |
1666 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || | |
1667 | _timeout.tv_nsec >= 1000000000L) { | |
1668 | error = -EINVAL; | |
1669 | goto out_free; | |
1670 | } | |
1671 | jiffies_left = timespec_to_jiffies(&_timeout); | |
1672 | } | |
1673 | max = 0; | |
1674 | for (sop = sops; sop < sops + nsops; sop++) { | |
1675 | if (sop->sem_num >= max) | |
1676 | max = sop->sem_num; | |
1677 | if (sop->sem_flg & SEM_UNDO) | |
b78755ab MS |
1678 | undos = 1; |
1679 | if (sop->sem_op != 0) | |
1da177e4 LT |
1680 | alter = 1; |
1681 | } | |
1da177e4 | 1682 | |
6062a8dc RR |
1683 | INIT_LIST_HEAD(&tasks); |
1684 | ||
1da177e4 | 1685 | if (undos) { |
6062a8dc | 1686 | /* On success, find_alloc_undo takes the rcu_read_lock */ |
4daa28f6 | 1687 | un = find_alloc_undo(ns, semid); |
1da177e4 LT |
1688 | if (IS_ERR(un)) { |
1689 | error = PTR_ERR(un); | |
1690 | goto out_free; | |
1691 | } | |
6062a8dc | 1692 | } else { |
1da177e4 | 1693 | un = NULL; |
6062a8dc RR |
1694 | rcu_read_lock(); |
1695 | } | |
1da177e4 | 1696 | |
16df3674 | 1697 | sma = sem_obtain_object_check(ns, semid); |
023a5355 | 1698 | if (IS_ERR(sma)) { |
6062a8dc | 1699 | rcu_read_unlock(); |
023a5355 | 1700 | error = PTR_ERR(sma); |
1da177e4 | 1701 | goto out_free; |
023a5355 ND |
1702 | } |
1703 | ||
16df3674 | 1704 | error = -EFBIG; |
c728b9c8 LT |
1705 | if (max >= sma->sem_nsems) |
1706 | goto out_rcu_wakeup; | |
16df3674 DB |
1707 | |
1708 | error = -EACCES; | |
c728b9c8 LT |
1709 | if (ipcperms(ns, &sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
1710 | goto out_rcu_wakeup; | |
16df3674 DB |
1711 | |
1712 | error = security_sem_semop(sma, sops, nsops, alter); | |
c728b9c8 LT |
1713 | if (error) |
1714 | goto out_rcu_wakeup; | |
16df3674 | 1715 | |
1da177e4 | 1716 | /* |
4daa28f6 | 1717 | * semid identifiers are not unique - find_alloc_undo may have |
1da177e4 | 1718 | * allocated an undo structure, it was invalidated by an RMID |
4daa28f6 | 1719 | * and now a new array with received the same id. Check and fail. |
25985edc | 1720 | * This case can be detected checking un->semid. The existence of |
380af1b3 | 1721 | * "un" itself is guaranteed by rcu. |
1da177e4 | 1722 | */ |
4daa28f6 | 1723 | error = -EIDRM; |
6062a8dc RR |
1724 | locknum = sem_lock(sma, sops, nsops); |
1725 | if (un && un->semid == -1) | |
1726 | goto out_unlock_free; | |
4daa28f6 | 1727 | |
b488893a | 1728 | error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); |
1da177e4 LT |
1729 | if (error <= 0) { |
1730 | if (alter && error == 0) | |
0a2b9d4c | 1731 | do_smart_update(sma, sops, nsops, 1, &tasks); |
636c6be8 | 1732 | |
1da177e4 LT |
1733 | goto out_unlock_free; |
1734 | } | |
1735 | ||
1736 | /* We need to sleep on this operation, so we put the current | |
1737 | * task into the pending queue and go to sleep. | |
1738 | */ | |
1739 | ||
1da177e4 LT |
1740 | queue.sops = sops; |
1741 | queue.nsops = nsops; | |
1742 | queue.undo = un; | |
b488893a | 1743 | queue.pid = task_tgid_vnr(current); |
1da177e4 | 1744 | queue.alter = alter; |
1da177e4 | 1745 | |
b97e820f MS |
1746 | if (nsops == 1) { |
1747 | struct sem *curr; | |
1748 | curr = &sma->sem_base[sops->sem_num]; | |
1749 | ||
1750 | if (alter) | |
1a82e9e1 | 1751 | list_add_tail(&queue.list, &curr->pending_alter); |
b97e820f | 1752 | else |
1a82e9e1 | 1753 | list_add_tail(&queue.list, &curr->pending_const); |
b97e820f | 1754 | } else { |
9f1bc2c9 | 1755 | if (alter) |
1a82e9e1 | 1756 | list_add_tail(&queue.list, &sma->pending_alter); |
9f1bc2c9 | 1757 | else |
1a82e9e1 MS |
1758 | list_add_tail(&queue.list, &sma->pending_const); |
1759 | ||
b97e820f MS |
1760 | sma->complex_count++; |
1761 | } | |
1762 | ||
1da177e4 LT |
1763 | queue.status = -EINTR; |
1764 | queue.sleeper = current; | |
0b0577f6 MS |
1765 | |
1766 | sleep_again: | |
1da177e4 | 1767 | current->state = TASK_INTERRUPTIBLE; |
6062a8dc | 1768 | sem_unlock(sma, locknum); |
6d49dab8 | 1769 | rcu_read_unlock(); |
1da177e4 LT |
1770 | |
1771 | if (timeout) | |
1772 | jiffies_left = schedule_timeout(jiffies_left); | |
1773 | else | |
1774 | schedule(); | |
1775 | ||
c61284e9 | 1776 | error = get_queue_result(&queue); |
1da177e4 LT |
1777 | |
1778 | if (error != -EINTR) { | |
1779 | /* fast path: update_queue already obtained all requested | |
c61284e9 MS |
1780 | * resources. |
1781 | * Perform a smp_mb(): User space could assume that semop() | |
1782 | * is a memory barrier: Without the mb(), the cpu could | |
1783 | * speculatively read in user space stale data that was | |
1784 | * overwritten by the previous owner of the semaphore. | |
1785 | */ | |
1786 | smp_mb(); | |
1787 | ||
1da177e4 LT |
1788 | goto out_free; |
1789 | } | |
1790 | ||
321310ce | 1791 | rcu_read_lock(); |
6062a8dc | 1792 | sma = sem_obtain_lock(ns, semid, sops, nsops, &locknum); |
d694ad62 MS |
1793 | |
1794 | /* | |
1795 | * Wait until it's guaranteed that no wakeup_sem_queue_do() is ongoing. | |
1796 | */ | |
1797 | error = get_queue_result(&queue); | |
1798 | ||
1799 | /* | |
1800 | * Array removed? If yes, leave without sem_unlock(). | |
1801 | */ | |
023a5355 | 1802 | if (IS_ERR(sma)) { |
321310ce | 1803 | rcu_read_unlock(); |
1da177e4 LT |
1804 | goto out_free; |
1805 | } | |
1806 | ||
c61284e9 | 1807 | |
1da177e4 | 1808 | /* |
d694ad62 MS |
1809 | * If queue.status != -EINTR we are woken up by another process. |
1810 | * Leave without unlink_queue(), but with sem_unlock(). | |
1da177e4 | 1811 | */ |
c61284e9 | 1812 | |
1da177e4 LT |
1813 | if (error != -EINTR) { |
1814 | goto out_unlock_free; | |
1815 | } | |
1816 | ||
1817 | /* | |
1818 | * If an interrupt occurred we have to clean up the queue | |
1819 | */ | |
1820 | if (timeout && jiffies_left == 0) | |
1821 | error = -EAGAIN; | |
0b0577f6 MS |
1822 | |
1823 | /* | |
1824 | * If the wakeup was spurious, just retry | |
1825 | */ | |
1826 | if (error == -EINTR && !signal_pending(current)) | |
1827 | goto sleep_again; | |
1828 | ||
b97e820f | 1829 | unlink_queue(sma, &queue); |
1da177e4 LT |
1830 | |
1831 | out_unlock_free: | |
6062a8dc | 1832 | sem_unlock(sma, locknum); |
c728b9c8 | 1833 | out_rcu_wakeup: |
6d49dab8 | 1834 | rcu_read_unlock(); |
0a2b9d4c | 1835 | wake_up_sem_queue_do(&tasks); |
1da177e4 LT |
1836 | out_free: |
1837 | if(sops != fast_sops) | |
1838 | kfree(sops); | |
1839 | return error; | |
1840 | } | |
1841 | ||
d5460c99 HC |
1842 | SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, |
1843 | unsigned, nsops) | |
1da177e4 LT |
1844 | { |
1845 | return sys_semtimedop(semid, tsops, nsops, NULL); | |
1846 | } | |
1847 | ||
1848 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between | |
1849 | * parent and child tasks. | |
1da177e4 LT |
1850 | */ |
1851 | ||
1852 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) | |
1853 | { | |
1854 | struct sem_undo_list *undo_list; | |
1855 | int error; | |
1856 | ||
1857 | if (clone_flags & CLONE_SYSVSEM) { | |
1858 | error = get_undo_list(&undo_list); | |
1859 | if (error) | |
1860 | return error; | |
1da177e4 LT |
1861 | atomic_inc(&undo_list->refcnt); |
1862 | tsk->sysvsem.undo_list = undo_list; | |
1863 | } else | |
1864 | tsk->sysvsem.undo_list = NULL; | |
1865 | ||
1866 | return 0; | |
1867 | } | |
1868 | ||
1869 | /* | |
1870 | * add semadj values to semaphores, free undo structures. | |
1871 | * undo structures are not freed when semaphore arrays are destroyed | |
1872 | * so some of them may be out of date. | |
1873 | * IMPLEMENTATION NOTE: There is some confusion over whether the | |
1874 | * set of adjustments that needs to be done should be done in an atomic | |
1875 | * manner or not. That is, if we are attempting to decrement the semval | |
1876 | * should we queue up and wait until we can do so legally? | |
1877 | * The original implementation attempted to do this (queue and wait). | |
1878 | * The current implementation does not do so. The POSIX standard | |
1879 | * and SVID should be consulted to determine what behavior is mandated. | |
1880 | */ | |
1881 | void exit_sem(struct task_struct *tsk) | |
1882 | { | |
4daa28f6 | 1883 | struct sem_undo_list *ulp; |
1da177e4 | 1884 | |
4daa28f6 MS |
1885 | ulp = tsk->sysvsem.undo_list; |
1886 | if (!ulp) | |
1da177e4 | 1887 | return; |
9edff4ab | 1888 | tsk->sysvsem.undo_list = NULL; |
1da177e4 | 1889 | |
4daa28f6 | 1890 | if (!atomic_dec_and_test(&ulp->refcnt)) |
1da177e4 LT |
1891 | return; |
1892 | ||
380af1b3 | 1893 | for (;;) { |
1da177e4 | 1894 | struct sem_array *sma; |
380af1b3 | 1895 | struct sem_undo *un; |
0a2b9d4c | 1896 | struct list_head tasks; |
6062a8dc | 1897 | int semid, i; |
4daa28f6 | 1898 | |
380af1b3 | 1899 | rcu_read_lock(); |
05725f7e JP |
1900 | un = list_entry_rcu(ulp->list_proc.next, |
1901 | struct sem_undo, list_proc); | |
380af1b3 MS |
1902 | if (&un->list_proc == &ulp->list_proc) |
1903 | semid = -1; | |
1904 | else | |
1905 | semid = un->semid; | |
4daa28f6 | 1906 | |
6062a8dc RR |
1907 | if (semid == -1) { |
1908 | rcu_read_unlock(); | |
380af1b3 | 1909 | break; |
6062a8dc | 1910 | } |
1da177e4 | 1911 | |
6062a8dc | 1912 | sma = sem_obtain_object_check(tsk->nsproxy->ipc_ns, un->semid); |
380af1b3 | 1913 | /* exit_sem raced with IPC_RMID, nothing to do */ |
6062a8dc RR |
1914 | if (IS_ERR(sma)) { |
1915 | rcu_read_unlock(); | |
380af1b3 | 1916 | continue; |
6062a8dc | 1917 | } |
1da177e4 | 1918 | |
6062a8dc | 1919 | sem_lock(sma, NULL, -1); |
bf17bb71 | 1920 | un = __lookup_undo(ulp, semid); |
380af1b3 MS |
1921 | if (un == NULL) { |
1922 | /* exit_sem raced with IPC_RMID+semget() that created | |
1923 | * exactly the same semid. Nothing to do. | |
1924 | */ | |
6062a8dc | 1925 | sem_unlock(sma, -1); |
6d49dab8 | 1926 | rcu_read_unlock(); |
380af1b3 MS |
1927 | continue; |
1928 | } | |
1929 | ||
1930 | /* remove un from the linked lists */ | |
cf9d5d78 | 1931 | ipc_assert_locked_object(&sma->sem_perm); |
4daa28f6 MS |
1932 | list_del(&un->list_id); |
1933 | ||
380af1b3 MS |
1934 | spin_lock(&ulp->lock); |
1935 | list_del_rcu(&un->list_proc); | |
1936 | spin_unlock(&ulp->lock); | |
1937 | ||
4daa28f6 MS |
1938 | /* perform adjustments registered in un */ |
1939 | for (i = 0; i < sma->sem_nsems; i++) { | |
5f921ae9 | 1940 | struct sem * semaphore = &sma->sem_base[i]; |
4daa28f6 MS |
1941 | if (un->semadj[i]) { |
1942 | semaphore->semval += un->semadj[i]; | |
1da177e4 LT |
1943 | /* |
1944 | * Range checks of the new semaphore value, | |
1945 | * not defined by sus: | |
1946 | * - Some unices ignore the undo entirely | |
1947 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) | |
1948 | * - some cap the value (e.g. FreeBSD caps | |
1949 | * at 0, but doesn't enforce SEMVMX) | |
1950 | * | |
1951 | * Linux caps the semaphore value, both at 0 | |
1952 | * and at SEMVMX. | |
1953 | * | |
1954 | * Manfred <manfred@colorfullife.com> | |
1955 | */ | |
5f921ae9 IM |
1956 | if (semaphore->semval < 0) |
1957 | semaphore->semval = 0; | |
1958 | if (semaphore->semval > SEMVMX) | |
1959 | semaphore->semval = SEMVMX; | |
b488893a | 1960 | semaphore->sempid = task_tgid_vnr(current); |
1da177e4 LT |
1961 | } |
1962 | } | |
1da177e4 | 1963 | /* maybe some queued-up processes were waiting for this */ |
0a2b9d4c MS |
1964 | INIT_LIST_HEAD(&tasks); |
1965 | do_smart_update(sma, NULL, 0, 1, &tasks); | |
6062a8dc | 1966 | sem_unlock(sma, -1); |
6d49dab8 | 1967 | rcu_read_unlock(); |
0a2b9d4c | 1968 | wake_up_sem_queue_do(&tasks); |
380af1b3 | 1969 | |
693a8b6e | 1970 | kfree_rcu(un, rcu); |
1da177e4 | 1971 | } |
4daa28f6 | 1972 | kfree(ulp); |
1da177e4 LT |
1973 | } |
1974 | ||
1975 | #ifdef CONFIG_PROC_FS | |
19b4946c | 1976 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
1da177e4 | 1977 | { |
1efdb69b | 1978 | struct user_namespace *user_ns = seq_user_ns(s); |
19b4946c MW |
1979 | struct sem_array *sma = it; |
1980 | ||
1981 | return seq_printf(s, | |
b97e820f | 1982 | "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", |
19b4946c | 1983 | sma->sem_perm.key, |
7ca7e564 | 1984 | sma->sem_perm.id, |
19b4946c MW |
1985 | sma->sem_perm.mode, |
1986 | sma->sem_nsems, | |
1efdb69b EB |
1987 | from_kuid_munged(user_ns, sma->sem_perm.uid), |
1988 | from_kgid_munged(user_ns, sma->sem_perm.gid), | |
1989 | from_kuid_munged(user_ns, sma->sem_perm.cuid), | |
1990 | from_kgid_munged(user_ns, sma->sem_perm.cgid), | |
19b4946c MW |
1991 | sma->sem_otime, |
1992 | sma->sem_ctime); | |
1da177e4 LT |
1993 | } |
1994 | #endif |