Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * linux/ipc/sem.c |
| 3 | * Copyright (C) 1992 Krishna Balasubramanian |
| 4 | * Copyright (C) 1995 Eric Schenk, Bruno Haible |
| 5 | * |
| 6 | * IMPLEMENTATION NOTES ON CODE REWRITE (Eric Schenk, January 1995): |
| 7 | * This code underwent a massive rewrite in order to solve some problems |
| 8 | * with the original code. In particular the original code failed to |
| 9 | * wake up processes that were waiting for semval to go to 0 if the |
| 10 | * value went to 0 and was then incremented rapidly enough. In solving |
| 11 | * this problem I have also modified the implementation so that it |
| 12 | * processes pending operations in a FIFO manner, thus give a guarantee |
| 13 | * that processes waiting for a lock on the semaphore won't starve |
| 14 | * unless another locking process fails to unlock. |
| 15 | * In addition the following two changes in behavior have been introduced: |
| 16 | * - The original implementation of semop returned the value |
| 17 | * last semaphore element examined on success. This does not |
| 18 | * match the manual page specifications, and effectively |
| 19 | * allows the user to read the semaphore even if they do not |
| 20 | * have read permissions. The implementation now returns 0 |
| 21 | * on success as stated in the manual page. |
| 22 | * - There is some confusion over whether the set of undo adjustments |
| 23 | * to be performed at exit should be done in an atomic manner. |
| 24 | * That is, if we are attempting to decrement the semval should we queue |
| 25 | * up and wait until we can do so legally? |
| 26 | * The original implementation attempted to do this. |
| 27 | * The current implementation does not do so. This is because I don't |
| 28 | * think it is the right thing (TM) to do, and because I couldn't |
| 29 | * see a clean way to get the old behavior with the new design. |
| 30 | * The POSIX standard and SVID should be consulted to determine |
| 31 | * what behavior is mandated. |
| 32 | * |
| 33 | * Further notes on refinement (Christoph Rohland, December 1998): |
| 34 | * - The POSIX standard says, that the undo adjustments simply should |
| 35 | * redo. So the current implementation is o.K. |
| 36 | * - The previous code had two flaws: |
| 37 | * 1) It actively gave the semaphore to the next waiting process |
| 38 | * sleeping on the semaphore. Since this process did not have the |
| 39 | * cpu this led to many unnecessary context switches and bad |
| 40 | * performance. Now we only check which process should be able to |
| 41 | * get the semaphore and if this process wants to reduce some |
| 42 | * semaphore value we simply wake it up without doing the |
| 43 | * operation. So it has to try to get it later. Thus e.g. the |
| 44 | * running process may reacquire the semaphore during the current |
| 45 | * time slice. If it only waits for zero or increases the semaphore, |
| 46 | * we do the operation in advance and wake it up. |
| 47 | * 2) It did not wake up all zero waiting processes. We try to do |
| 48 | * better but only get the semops right which only wait for zero or |
| 49 | * increase. If there are decrement operations in the operations |
| 50 | * array we do the same as before. |
| 51 | * |
| 52 | * With the incarnation of O(1) scheduler, it becomes unnecessary to perform |
| 53 | * check/retry algorithm for waking up blocked processes as the new scheduler |
| 54 | * is better at handling thread switch than the old one. |
| 55 | * |
| 56 | * /proc/sysvipc/sem support (c) 1999 Dragos Acostachioaie <dragos@iname.com> |
| 57 | * |
| 58 | * SMP-threaded, sysctl's added |
Christian Kujau | 624dffc | 2006-01-15 02:43:54 +0100 | [diff] [blame] | 59 | * (c) 1999 Manfred Spraul <manfred@colorfullife.com> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 60 | * Enforced range limit on SEM_UNDO |
| 61 | * (c) 2001 Red Hat Inc <alan@redhat.com> |
| 62 | * Lockless wakeup |
| 63 | * (c) 2003 Manfred Spraul <manfred@colorfullife.com> |
| 64 | */ |
| 65 | |
| 66 | #include <linux/config.h> |
| 67 | #include <linux/slab.h> |
| 68 | #include <linux/spinlock.h> |
| 69 | #include <linux/init.h> |
| 70 | #include <linux/proc_fs.h> |
| 71 | #include <linux/time.h> |
| 72 | #include <linux/smp_lock.h> |
| 73 | #include <linux/security.h> |
| 74 | #include <linux/syscalls.h> |
| 75 | #include <linux/audit.h> |
Randy.Dunlap | c59ede7 | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 76 | #include <linux/capability.h> |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 77 | #include <linux/seq_file.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 78 | #include <asm/uaccess.h> |
| 79 | #include "util.h" |
| 80 | |
| 81 | |
| 82 | #define sem_lock(id) ((struct sem_array*)ipc_lock(&sem_ids,id)) |
| 83 | #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm) |
| 84 | #define sem_rmid(id) ((struct sem_array*)ipc_rmid(&sem_ids,id)) |
| 85 | #define sem_checkid(sma, semid) \ |
| 86 | ipc_checkid(&sem_ids,&sma->sem_perm,semid) |
| 87 | #define sem_buildid(id, seq) \ |
| 88 | ipc_buildid(&sem_ids, id, seq) |
| 89 | static struct ipc_ids sem_ids; |
| 90 | |
| 91 | static int newary (key_t, int, int); |
| 92 | static void freeary (struct sem_array *sma, int id); |
| 93 | #ifdef CONFIG_PROC_FS |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 94 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | #endif |
| 96 | |
| 97 | #define SEMMSL_FAST 256 /* 512 bytes on stack */ |
| 98 | #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ |
| 99 | |
| 100 | /* |
| 101 | * linked list protection: |
| 102 | * sem_undo.id_next, |
| 103 | * sem_array.sem_pending{,last}, |
| 104 | * sem_array.sem_undo: sem_lock() for read/write |
| 105 | * sem_undo.proc_next: only "current" is allowed to read/write that field. |
| 106 | * |
| 107 | */ |
| 108 | |
| 109 | int sem_ctls[4] = {SEMMSL, SEMMNS, SEMOPM, SEMMNI}; |
| 110 | #define sc_semmsl (sem_ctls[0]) |
| 111 | #define sc_semmns (sem_ctls[1]) |
| 112 | #define sc_semopm (sem_ctls[2]) |
| 113 | #define sc_semmni (sem_ctls[3]) |
| 114 | |
| 115 | static int used_sems; |
| 116 | |
| 117 | void __init sem_init (void) |
| 118 | { |
| 119 | used_sems = 0; |
| 120 | ipc_init_ids(&sem_ids,sc_semmni); |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 121 | ipc_init_proc_interface("sysvipc/sem", |
| 122 | " key semid perms nsems uid gid cuid cgid otime ctime\n", |
| 123 | &sem_ids, |
| 124 | sysvipc_sem_proc_show); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 125 | } |
| 126 | |
| 127 | /* |
| 128 | * Lockless wakeup algorithm: |
| 129 | * Without the check/retry algorithm a lockless wakeup is possible: |
| 130 | * - queue.status is initialized to -EINTR before blocking. |
| 131 | * - wakeup is performed by |
| 132 | * * unlinking the queue entry from sma->sem_pending |
| 133 | * * setting queue.status to IN_WAKEUP |
| 134 | * This is the notification for the blocked thread that a |
| 135 | * result value is imminent. |
| 136 | * * call wake_up_process |
| 137 | * * set queue.status to the final value. |
| 138 | * - the previously blocked thread checks queue.status: |
| 139 | * * if it's IN_WAKEUP, then it must wait until the value changes |
| 140 | * * if it's not -EINTR, then the operation was completed by |
| 141 | * update_queue. semtimedop can return queue.status without |
| 142 | * performing any operation on the semaphore array. |
| 143 | * * otherwise it must acquire the spinlock and check what's up. |
| 144 | * |
| 145 | * The two-stage algorithm is necessary to protect against the following |
| 146 | * races: |
| 147 | * - if queue.status is set after wake_up_process, then the woken up idle |
| 148 | * thread could race forward and try (and fail) to acquire sma->lock |
| 149 | * before update_queue had a chance to set queue.status |
| 150 | * - if queue.status is written before wake_up_process and if the |
| 151 | * blocked process is woken up by a signal between writing |
| 152 | * queue.status and the wake_up_process, then the woken up |
| 153 | * process could return from semtimedop and die by calling |
| 154 | * sys_exit before wake_up_process is called. Then wake_up_process |
| 155 | * will oops, because the task structure is already invalid. |
| 156 | * (yes, this happened on s390 with sysv msg). |
| 157 | * |
| 158 | */ |
| 159 | #define IN_WAKEUP 1 |
| 160 | |
| 161 | static int newary (key_t key, int nsems, int semflg) |
| 162 | { |
| 163 | int id; |
| 164 | int retval; |
| 165 | struct sem_array *sma; |
| 166 | int size; |
| 167 | |
| 168 | if (!nsems) |
| 169 | return -EINVAL; |
| 170 | if (used_sems + nsems > sc_semmns) |
| 171 | return -ENOSPC; |
| 172 | |
| 173 | size = sizeof (*sma) + nsems * sizeof (struct sem); |
| 174 | sma = ipc_rcu_alloc(size); |
| 175 | if (!sma) { |
| 176 | return -ENOMEM; |
| 177 | } |
| 178 | memset (sma, 0, size); |
| 179 | |
| 180 | sma->sem_perm.mode = (semflg & S_IRWXUGO); |
| 181 | sma->sem_perm.key = key; |
| 182 | |
| 183 | sma->sem_perm.security = NULL; |
| 184 | retval = security_sem_alloc(sma); |
| 185 | if (retval) { |
| 186 | ipc_rcu_putref(sma); |
| 187 | return retval; |
| 188 | } |
| 189 | |
| 190 | id = ipc_addid(&sem_ids, &sma->sem_perm, sc_semmni); |
| 191 | if(id == -1) { |
| 192 | security_sem_free(sma); |
| 193 | ipc_rcu_putref(sma); |
| 194 | return -ENOSPC; |
| 195 | } |
| 196 | used_sems += nsems; |
| 197 | |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 198 | sma->sem_id = sem_buildid(id, sma->sem_perm.seq); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | sma->sem_base = (struct sem *) &sma[1]; |
| 200 | /* sma->sem_pending = NULL; */ |
| 201 | sma->sem_pending_last = &sma->sem_pending; |
| 202 | /* sma->undo = NULL; */ |
| 203 | sma->sem_nsems = nsems; |
| 204 | sma->sem_ctime = get_seconds(); |
| 205 | sem_unlock(sma); |
| 206 | |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 207 | return sma->sem_id; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 208 | } |
| 209 | |
| 210 | asmlinkage long sys_semget (key_t key, int nsems, int semflg) |
| 211 | { |
| 212 | int id, err = -EINVAL; |
| 213 | struct sem_array *sma; |
| 214 | |
| 215 | if (nsems < 0 || nsems > sc_semmsl) |
| 216 | return -EINVAL; |
| 217 | down(&sem_ids.sem); |
| 218 | |
| 219 | if (key == IPC_PRIVATE) { |
| 220 | err = newary(key, nsems, semflg); |
| 221 | } else if ((id = ipc_findkey(&sem_ids, key)) == -1) { /* key not used */ |
| 222 | if (!(semflg & IPC_CREAT)) |
| 223 | err = -ENOENT; |
| 224 | else |
| 225 | err = newary(key, nsems, semflg); |
| 226 | } else if (semflg & IPC_CREAT && semflg & IPC_EXCL) { |
| 227 | err = -EEXIST; |
| 228 | } else { |
| 229 | sma = sem_lock(id); |
| 230 | if(sma==NULL) |
| 231 | BUG(); |
| 232 | if (nsems > sma->sem_nsems) |
| 233 | err = -EINVAL; |
| 234 | else if (ipcperms(&sma->sem_perm, semflg)) |
| 235 | err = -EACCES; |
| 236 | else { |
| 237 | int semid = sem_buildid(id, sma->sem_perm.seq); |
| 238 | err = security_sem_associate(sma, semflg); |
| 239 | if (!err) |
| 240 | err = semid; |
| 241 | } |
| 242 | sem_unlock(sma); |
| 243 | } |
| 244 | |
| 245 | up(&sem_ids.sem); |
| 246 | return err; |
| 247 | } |
| 248 | |
| 249 | /* Manage the doubly linked list sma->sem_pending as a FIFO: |
| 250 | * insert new queue elements at the tail sma->sem_pending_last. |
| 251 | */ |
| 252 | static inline void append_to_queue (struct sem_array * sma, |
| 253 | struct sem_queue * q) |
| 254 | { |
| 255 | *(q->prev = sma->sem_pending_last) = q; |
| 256 | *(sma->sem_pending_last = &q->next) = NULL; |
| 257 | } |
| 258 | |
| 259 | static inline void prepend_to_queue (struct sem_array * sma, |
| 260 | struct sem_queue * q) |
| 261 | { |
| 262 | q->next = sma->sem_pending; |
| 263 | *(q->prev = &sma->sem_pending) = q; |
| 264 | if (q->next) |
| 265 | q->next->prev = &q->next; |
| 266 | else /* sma->sem_pending_last == &sma->sem_pending */ |
| 267 | sma->sem_pending_last = &q->next; |
| 268 | } |
| 269 | |
| 270 | static inline void remove_from_queue (struct sem_array * sma, |
| 271 | struct sem_queue * q) |
| 272 | { |
| 273 | *(q->prev) = q->next; |
| 274 | if (q->next) |
| 275 | q->next->prev = q->prev; |
| 276 | else /* sma->sem_pending_last == &q->next */ |
| 277 | sma->sem_pending_last = q->prev; |
| 278 | q->prev = NULL; /* mark as removed */ |
| 279 | } |
| 280 | |
| 281 | /* |
| 282 | * Determine whether a sequence of semaphore operations would succeed |
| 283 | * all at once. Return 0 if yes, 1 if need to sleep, else return error code. |
| 284 | */ |
| 285 | |
| 286 | static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, |
| 287 | int nsops, struct sem_undo *un, int pid) |
| 288 | { |
| 289 | int result, sem_op; |
| 290 | struct sembuf *sop; |
| 291 | struct sem * curr; |
| 292 | |
| 293 | for (sop = sops; sop < sops + nsops; sop++) { |
| 294 | curr = sma->sem_base + sop->sem_num; |
| 295 | sem_op = sop->sem_op; |
| 296 | result = curr->semval; |
| 297 | |
| 298 | if (!sem_op && result) |
| 299 | goto would_block; |
| 300 | |
| 301 | result += sem_op; |
| 302 | if (result < 0) |
| 303 | goto would_block; |
| 304 | if (result > SEMVMX) |
| 305 | goto out_of_range; |
| 306 | if (sop->sem_flg & SEM_UNDO) { |
| 307 | int undo = un->semadj[sop->sem_num] - sem_op; |
| 308 | /* |
| 309 | * Exceeding the undo range is an error. |
| 310 | */ |
| 311 | if (undo < (-SEMAEM - 1) || undo > SEMAEM) |
| 312 | goto out_of_range; |
| 313 | } |
| 314 | curr->semval = result; |
| 315 | } |
| 316 | |
| 317 | sop--; |
| 318 | while (sop >= sops) { |
| 319 | sma->sem_base[sop->sem_num].sempid = pid; |
| 320 | if (sop->sem_flg & SEM_UNDO) |
| 321 | un->semadj[sop->sem_num] -= sop->sem_op; |
| 322 | sop--; |
| 323 | } |
| 324 | |
| 325 | sma->sem_otime = get_seconds(); |
| 326 | return 0; |
| 327 | |
| 328 | out_of_range: |
| 329 | result = -ERANGE; |
| 330 | goto undo; |
| 331 | |
| 332 | would_block: |
| 333 | if (sop->sem_flg & IPC_NOWAIT) |
| 334 | result = -EAGAIN; |
| 335 | else |
| 336 | result = 1; |
| 337 | |
| 338 | undo: |
| 339 | sop--; |
| 340 | while (sop >= sops) { |
| 341 | sma->sem_base[sop->sem_num].semval -= sop->sem_op; |
| 342 | sop--; |
| 343 | } |
| 344 | |
| 345 | return result; |
| 346 | } |
| 347 | |
| 348 | /* Go through the pending queue for the indicated semaphore |
| 349 | * looking for tasks that can be completed. |
| 350 | */ |
| 351 | static void update_queue (struct sem_array * sma) |
| 352 | { |
| 353 | int error; |
| 354 | struct sem_queue * q; |
| 355 | |
| 356 | q = sma->sem_pending; |
| 357 | while(q) { |
| 358 | error = try_atomic_semop(sma, q->sops, q->nsops, |
| 359 | q->undo, q->pid); |
| 360 | |
| 361 | /* Does q->sleeper still need to sleep? */ |
| 362 | if (error <= 0) { |
| 363 | struct sem_queue *n; |
| 364 | remove_from_queue(sma,q); |
| 365 | q->status = IN_WAKEUP; |
| 366 | /* |
| 367 | * Continue scanning. The next operation |
| 368 | * that must be checked depends on the type of the |
| 369 | * completed operation: |
| 370 | * - if the operation modified the array, then |
| 371 | * restart from the head of the queue and |
| 372 | * check for threads that might be waiting |
| 373 | * for semaphore values to become 0. |
| 374 | * - if the operation didn't modify the array, |
| 375 | * then just continue. |
| 376 | */ |
| 377 | if (q->alter) |
| 378 | n = sma->sem_pending; |
| 379 | else |
| 380 | n = q->next; |
| 381 | wake_up_process(q->sleeper); |
| 382 | /* hands-off: q will disappear immediately after |
| 383 | * writing q->status. |
| 384 | */ |
Linus Torvalds | 1224b37 | 2005-12-24 12:19:38 -0800 | [diff] [blame] | 385 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 386 | q->status = error; |
| 387 | q = n; |
| 388 | } else { |
| 389 | q = q->next; |
| 390 | } |
| 391 | } |
| 392 | } |
| 393 | |
| 394 | /* The following counts are associated to each semaphore: |
| 395 | * semncnt number of tasks waiting on semval being nonzero |
| 396 | * semzcnt number of tasks waiting on semval being zero |
| 397 | * This model assumes that a task waits on exactly one semaphore. |
| 398 | * Since semaphore operations are to be performed atomically, tasks actually |
| 399 | * wait on a whole sequence of semaphores simultaneously. |
| 400 | * The counts we return here are a rough approximation, but still |
| 401 | * warrant that semncnt+semzcnt>0 if the task is on the pending queue. |
| 402 | */ |
| 403 | static int count_semncnt (struct sem_array * sma, ushort semnum) |
| 404 | { |
| 405 | int semncnt; |
| 406 | struct sem_queue * q; |
| 407 | |
| 408 | semncnt = 0; |
| 409 | for (q = sma->sem_pending; q; q = q->next) { |
| 410 | struct sembuf * sops = q->sops; |
| 411 | int nsops = q->nsops; |
| 412 | int i; |
| 413 | for (i = 0; i < nsops; i++) |
| 414 | if (sops[i].sem_num == semnum |
| 415 | && (sops[i].sem_op < 0) |
| 416 | && !(sops[i].sem_flg & IPC_NOWAIT)) |
| 417 | semncnt++; |
| 418 | } |
| 419 | return semncnt; |
| 420 | } |
| 421 | static int count_semzcnt (struct sem_array * sma, ushort semnum) |
| 422 | { |
| 423 | int semzcnt; |
| 424 | struct sem_queue * q; |
| 425 | |
| 426 | semzcnt = 0; |
| 427 | for (q = sma->sem_pending; q; q = q->next) { |
| 428 | struct sembuf * sops = q->sops; |
| 429 | int nsops = q->nsops; |
| 430 | int i; |
| 431 | for (i = 0; i < nsops; i++) |
| 432 | if (sops[i].sem_num == semnum |
| 433 | && (sops[i].sem_op == 0) |
| 434 | && !(sops[i].sem_flg & IPC_NOWAIT)) |
| 435 | semzcnt++; |
| 436 | } |
| 437 | return semzcnt; |
| 438 | } |
| 439 | |
| 440 | /* Free a semaphore set. freeary() is called with sem_ids.sem down and |
| 441 | * the spinlock for this semaphore set hold. sem_ids.sem remains locked |
| 442 | * on exit. |
| 443 | */ |
| 444 | static void freeary (struct sem_array *sma, int id) |
| 445 | { |
| 446 | struct sem_undo *un; |
| 447 | struct sem_queue *q; |
| 448 | int size; |
| 449 | |
| 450 | /* Invalidate the existing undo structures for this semaphore set. |
| 451 | * (They will be freed without any further action in exit_sem() |
| 452 | * or during the next semop.) |
| 453 | */ |
| 454 | for (un = sma->undo; un; un = un->id_next) |
| 455 | un->semid = -1; |
| 456 | |
| 457 | /* Wake up all pending processes and let them fail with EIDRM. */ |
| 458 | q = sma->sem_pending; |
| 459 | while(q) { |
| 460 | struct sem_queue *n; |
| 461 | /* lazy remove_from_queue: we are killing the whole queue */ |
| 462 | q->prev = NULL; |
| 463 | n = q->next; |
| 464 | q->status = IN_WAKEUP; |
| 465 | wake_up_process(q->sleeper); /* doesn't sleep */ |
Manfred Spraul | 6003a93 | 2005-12-23 23:57:41 +0100 | [diff] [blame] | 466 | smp_wmb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 467 | q->status = -EIDRM; /* hands-off q */ |
| 468 | q = n; |
| 469 | } |
| 470 | |
| 471 | /* Remove the semaphore set from the ID array*/ |
| 472 | sma = sem_rmid(id); |
| 473 | sem_unlock(sma); |
| 474 | |
| 475 | used_sems -= sma->sem_nsems; |
| 476 | size = sizeof (*sma) + sma->sem_nsems * sizeof (struct sem); |
| 477 | security_sem_free(sma); |
| 478 | ipc_rcu_putref(sma); |
| 479 | } |
| 480 | |
| 481 | static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) |
| 482 | { |
| 483 | switch(version) { |
| 484 | case IPC_64: |
| 485 | return copy_to_user(buf, in, sizeof(*in)); |
| 486 | case IPC_OLD: |
| 487 | { |
| 488 | struct semid_ds out; |
| 489 | |
| 490 | ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); |
| 491 | |
| 492 | out.sem_otime = in->sem_otime; |
| 493 | out.sem_ctime = in->sem_ctime; |
| 494 | out.sem_nsems = in->sem_nsems; |
| 495 | |
| 496 | return copy_to_user(buf, &out, sizeof(out)); |
| 497 | } |
| 498 | default: |
| 499 | return -EINVAL; |
| 500 | } |
| 501 | } |
| 502 | |
| 503 | static int semctl_nolock(int semid, int semnum, int cmd, int version, union semun arg) |
| 504 | { |
| 505 | int err = -EINVAL; |
| 506 | struct sem_array *sma; |
| 507 | |
| 508 | switch(cmd) { |
| 509 | case IPC_INFO: |
| 510 | case SEM_INFO: |
| 511 | { |
| 512 | struct seminfo seminfo; |
| 513 | int max_id; |
| 514 | |
| 515 | err = security_sem_semctl(NULL, cmd); |
| 516 | if (err) |
| 517 | return err; |
| 518 | |
| 519 | memset(&seminfo,0,sizeof(seminfo)); |
| 520 | seminfo.semmni = sc_semmni; |
| 521 | seminfo.semmns = sc_semmns; |
| 522 | seminfo.semmsl = sc_semmsl; |
| 523 | seminfo.semopm = sc_semopm; |
| 524 | seminfo.semvmx = SEMVMX; |
| 525 | seminfo.semmnu = SEMMNU; |
| 526 | seminfo.semmap = SEMMAP; |
| 527 | seminfo.semume = SEMUME; |
| 528 | down(&sem_ids.sem); |
| 529 | if (cmd == SEM_INFO) { |
| 530 | seminfo.semusz = sem_ids.in_use; |
| 531 | seminfo.semaem = used_sems; |
| 532 | } else { |
| 533 | seminfo.semusz = SEMUSZ; |
| 534 | seminfo.semaem = SEMAEM; |
| 535 | } |
| 536 | max_id = sem_ids.max_id; |
| 537 | up(&sem_ids.sem); |
| 538 | if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) |
| 539 | return -EFAULT; |
| 540 | return (max_id < 0) ? 0: max_id; |
| 541 | } |
| 542 | case SEM_STAT: |
| 543 | { |
| 544 | struct semid64_ds tbuf; |
| 545 | int id; |
| 546 | |
| 547 | if(semid >= sem_ids.entries->size) |
| 548 | return -EINVAL; |
| 549 | |
| 550 | memset(&tbuf,0,sizeof(tbuf)); |
| 551 | |
| 552 | sma = sem_lock(semid); |
| 553 | if(sma == NULL) |
| 554 | return -EINVAL; |
| 555 | |
| 556 | err = -EACCES; |
| 557 | if (ipcperms (&sma->sem_perm, S_IRUGO)) |
| 558 | goto out_unlock; |
| 559 | |
| 560 | err = security_sem_semctl(sma, cmd); |
| 561 | if (err) |
| 562 | goto out_unlock; |
| 563 | |
| 564 | id = sem_buildid(semid, sma->sem_perm.seq); |
| 565 | |
| 566 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
| 567 | tbuf.sem_otime = sma->sem_otime; |
| 568 | tbuf.sem_ctime = sma->sem_ctime; |
| 569 | tbuf.sem_nsems = sma->sem_nsems; |
| 570 | sem_unlock(sma); |
| 571 | if (copy_semid_to_user (arg.buf, &tbuf, version)) |
| 572 | return -EFAULT; |
| 573 | return id; |
| 574 | } |
| 575 | default: |
| 576 | return -EINVAL; |
| 577 | } |
| 578 | return err; |
| 579 | out_unlock: |
| 580 | sem_unlock(sma); |
| 581 | return err; |
| 582 | } |
| 583 | |
| 584 | static int semctl_main(int semid, int semnum, int cmd, int version, union semun arg) |
| 585 | { |
| 586 | struct sem_array *sma; |
| 587 | struct sem* curr; |
| 588 | int err; |
| 589 | ushort fast_sem_io[SEMMSL_FAST]; |
| 590 | ushort* sem_io = fast_sem_io; |
| 591 | int nsems; |
| 592 | |
| 593 | sma = sem_lock(semid); |
| 594 | if(sma==NULL) |
| 595 | return -EINVAL; |
| 596 | |
| 597 | nsems = sma->sem_nsems; |
| 598 | |
| 599 | err=-EIDRM; |
| 600 | if (sem_checkid(sma,semid)) |
| 601 | goto out_unlock; |
| 602 | |
| 603 | err = -EACCES; |
| 604 | if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) |
| 605 | goto out_unlock; |
| 606 | |
| 607 | err = security_sem_semctl(sma, cmd); |
| 608 | if (err) |
| 609 | goto out_unlock; |
| 610 | |
| 611 | err = -EACCES; |
| 612 | switch (cmd) { |
| 613 | case GETALL: |
| 614 | { |
| 615 | ushort __user *array = arg.array; |
| 616 | int i; |
| 617 | |
| 618 | if(nsems > SEMMSL_FAST) { |
| 619 | ipc_rcu_getref(sma); |
| 620 | sem_unlock(sma); |
| 621 | |
| 622 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
| 623 | if(sem_io == NULL) { |
| 624 | ipc_lock_by_ptr(&sma->sem_perm); |
| 625 | ipc_rcu_putref(sma); |
| 626 | sem_unlock(sma); |
| 627 | return -ENOMEM; |
| 628 | } |
| 629 | |
| 630 | ipc_lock_by_ptr(&sma->sem_perm); |
| 631 | ipc_rcu_putref(sma); |
| 632 | if (sma->sem_perm.deleted) { |
| 633 | sem_unlock(sma); |
| 634 | err = -EIDRM; |
| 635 | goto out_free; |
| 636 | } |
| 637 | } |
| 638 | |
| 639 | for (i = 0; i < sma->sem_nsems; i++) |
| 640 | sem_io[i] = sma->sem_base[i].semval; |
| 641 | sem_unlock(sma); |
| 642 | err = 0; |
| 643 | if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) |
| 644 | err = -EFAULT; |
| 645 | goto out_free; |
| 646 | } |
| 647 | case SETALL: |
| 648 | { |
| 649 | int i; |
| 650 | struct sem_undo *un; |
| 651 | |
| 652 | ipc_rcu_getref(sma); |
| 653 | sem_unlock(sma); |
| 654 | |
| 655 | if(nsems > SEMMSL_FAST) { |
| 656 | sem_io = ipc_alloc(sizeof(ushort)*nsems); |
| 657 | if(sem_io == NULL) { |
| 658 | ipc_lock_by_ptr(&sma->sem_perm); |
| 659 | ipc_rcu_putref(sma); |
| 660 | sem_unlock(sma); |
| 661 | return -ENOMEM; |
| 662 | } |
| 663 | } |
| 664 | |
| 665 | if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { |
| 666 | ipc_lock_by_ptr(&sma->sem_perm); |
| 667 | ipc_rcu_putref(sma); |
| 668 | sem_unlock(sma); |
| 669 | err = -EFAULT; |
| 670 | goto out_free; |
| 671 | } |
| 672 | |
| 673 | for (i = 0; i < nsems; i++) { |
| 674 | if (sem_io[i] > SEMVMX) { |
| 675 | ipc_lock_by_ptr(&sma->sem_perm); |
| 676 | ipc_rcu_putref(sma); |
| 677 | sem_unlock(sma); |
| 678 | err = -ERANGE; |
| 679 | goto out_free; |
| 680 | } |
| 681 | } |
| 682 | ipc_lock_by_ptr(&sma->sem_perm); |
| 683 | ipc_rcu_putref(sma); |
| 684 | if (sma->sem_perm.deleted) { |
| 685 | sem_unlock(sma); |
| 686 | err = -EIDRM; |
| 687 | goto out_free; |
| 688 | } |
| 689 | |
| 690 | for (i = 0; i < nsems; i++) |
| 691 | sma->sem_base[i].semval = sem_io[i]; |
| 692 | for (un = sma->undo; un; un = un->id_next) |
| 693 | for (i = 0; i < nsems; i++) |
| 694 | un->semadj[i] = 0; |
| 695 | sma->sem_ctime = get_seconds(); |
| 696 | /* maybe some queued-up processes were waiting for this */ |
| 697 | update_queue(sma); |
| 698 | err = 0; |
| 699 | goto out_unlock; |
| 700 | } |
| 701 | case IPC_STAT: |
| 702 | { |
| 703 | struct semid64_ds tbuf; |
| 704 | memset(&tbuf,0,sizeof(tbuf)); |
| 705 | kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); |
| 706 | tbuf.sem_otime = sma->sem_otime; |
| 707 | tbuf.sem_ctime = sma->sem_ctime; |
| 708 | tbuf.sem_nsems = sma->sem_nsems; |
| 709 | sem_unlock(sma); |
| 710 | if (copy_semid_to_user (arg.buf, &tbuf, version)) |
| 711 | return -EFAULT; |
| 712 | return 0; |
| 713 | } |
| 714 | /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ |
| 715 | } |
| 716 | err = -EINVAL; |
| 717 | if(semnum < 0 || semnum >= nsems) |
| 718 | goto out_unlock; |
| 719 | |
| 720 | curr = &sma->sem_base[semnum]; |
| 721 | |
| 722 | switch (cmd) { |
| 723 | case GETVAL: |
| 724 | err = curr->semval; |
| 725 | goto out_unlock; |
| 726 | case GETPID: |
| 727 | err = curr->sempid; |
| 728 | goto out_unlock; |
| 729 | case GETNCNT: |
| 730 | err = count_semncnt(sma,semnum); |
| 731 | goto out_unlock; |
| 732 | case GETZCNT: |
| 733 | err = count_semzcnt(sma,semnum); |
| 734 | goto out_unlock; |
| 735 | case SETVAL: |
| 736 | { |
| 737 | int val = arg.val; |
| 738 | struct sem_undo *un; |
| 739 | err = -ERANGE; |
| 740 | if (val > SEMVMX || val < 0) |
| 741 | goto out_unlock; |
| 742 | |
| 743 | for (un = sma->undo; un; un = un->id_next) |
| 744 | un->semadj[semnum] = 0; |
| 745 | curr->semval = val; |
| 746 | curr->sempid = current->tgid; |
| 747 | sma->sem_ctime = get_seconds(); |
| 748 | /* maybe some queued-up processes were waiting for this */ |
| 749 | update_queue(sma); |
| 750 | err = 0; |
| 751 | goto out_unlock; |
| 752 | } |
| 753 | } |
| 754 | out_unlock: |
| 755 | sem_unlock(sma); |
| 756 | out_free: |
| 757 | if(sem_io != fast_sem_io) |
| 758 | ipc_free(sem_io, sizeof(ushort)*nsems); |
| 759 | return err; |
| 760 | } |
| 761 | |
| 762 | struct sem_setbuf { |
| 763 | uid_t uid; |
| 764 | gid_t gid; |
| 765 | mode_t mode; |
| 766 | }; |
| 767 | |
| 768 | static inline unsigned long copy_semid_from_user(struct sem_setbuf *out, void __user *buf, int version) |
| 769 | { |
| 770 | switch(version) { |
| 771 | case IPC_64: |
| 772 | { |
| 773 | struct semid64_ds tbuf; |
| 774 | |
| 775 | if(copy_from_user(&tbuf, buf, sizeof(tbuf))) |
| 776 | return -EFAULT; |
| 777 | |
| 778 | out->uid = tbuf.sem_perm.uid; |
| 779 | out->gid = tbuf.sem_perm.gid; |
| 780 | out->mode = tbuf.sem_perm.mode; |
| 781 | |
| 782 | return 0; |
| 783 | } |
| 784 | case IPC_OLD: |
| 785 | { |
| 786 | struct semid_ds tbuf_old; |
| 787 | |
| 788 | if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) |
| 789 | return -EFAULT; |
| 790 | |
| 791 | out->uid = tbuf_old.sem_perm.uid; |
| 792 | out->gid = tbuf_old.sem_perm.gid; |
| 793 | out->mode = tbuf_old.sem_perm.mode; |
| 794 | |
| 795 | return 0; |
| 796 | } |
| 797 | default: |
| 798 | return -EINVAL; |
| 799 | } |
| 800 | } |
| 801 | |
| 802 | static int semctl_down(int semid, int semnum, int cmd, int version, union semun arg) |
| 803 | { |
| 804 | struct sem_array *sma; |
| 805 | int err; |
| 806 | struct sem_setbuf setbuf; |
| 807 | struct kern_ipc_perm *ipcp; |
| 808 | |
| 809 | if(cmd == IPC_SET) { |
| 810 | if(copy_semid_from_user (&setbuf, arg.buf, version)) |
| 811 | return -EFAULT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 812 | } |
| 813 | sma = sem_lock(semid); |
| 814 | if(sma==NULL) |
| 815 | return -EINVAL; |
| 816 | |
| 817 | if (sem_checkid(sma,semid)) { |
| 818 | err=-EIDRM; |
| 819 | goto out_unlock; |
| 820 | } |
| 821 | ipcp = &sma->sem_perm; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 822 | if (current->euid != ipcp->cuid && |
| 823 | current->euid != ipcp->uid && !capable(CAP_SYS_ADMIN)) { |
| 824 | err=-EPERM; |
| 825 | goto out_unlock; |
| 826 | } |
| 827 | |
| 828 | err = security_sem_semctl(sma, cmd); |
| 829 | if (err) |
| 830 | goto out_unlock; |
| 831 | |
| 832 | switch(cmd){ |
| 833 | case IPC_RMID: |
| 834 | freeary(sma, semid); |
| 835 | err = 0; |
| 836 | break; |
| 837 | case IPC_SET: |
Dustin Kirkland | 8c8570f | 2005-11-03 17:15:16 +0000 | [diff] [blame] | 838 | if ((err = audit_ipc_perms(0, setbuf.uid, setbuf.gid, setbuf.mode, ipcp))) |
| 839 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 840 | ipcp->uid = setbuf.uid; |
| 841 | ipcp->gid = setbuf.gid; |
| 842 | ipcp->mode = (ipcp->mode & ~S_IRWXUGO) |
| 843 | | (setbuf.mode & S_IRWXUGO); |
| 844 | sma->sem_ctime = get_seconds(); |
| 845 | sem_unlock(sma); |
| 846 | err = 0; |
| 847 | break; |
| 848 | default: |
| 849 | sem_unlock(sma); |
| 850 | err = -EINVAL; |
| 851 | break; |
| 852 | } |
| 853 | return err; |
| 854 | |
| 855 | out_unlock: |
| 856 | sem_unlock(sma); |
| 857 | return err; |
| 858 | } |
| 859 | |
| 860 | asmlinkage long sys_semctl (int semid, int semnum, int cmd, union semun arg) |
| 861 | { |
| 862 | int err = -EINVAL; |
| 863 | int version; |
| 864 | |
| 865 | if (semid < 0) |
| 866 | return -EINVAL; |
| 867 | |
| 868 | version = ipc_parse_version(&cmd); |
| 869 | |
| 870 | switch(cmd) { |
| 871 | case IPC_INFO: |
| 872 | case SEM_INFO: |
| 873 | case SEM_STAT: |
| 874 | err = semctl_nolock(semid,semnum,cmd,version,arg); |
| 875 | return err; |
| 876 | case GETALL: |
| 877 | case GETVAL: |
| 878 | case GETPID: |
| 879 | case GETNCNT: |
| 880 | case GETZCNT: |
| 881 | case IPC_STAT: |
| 882 | case SETVAL: |
| 883 | case SETALL: |
| 884 | err = semctl_main(semid,semnum,cmd,version,arg); |
| 885 | return err; |
| 886 | case IPC_RMID: |
| 887 | case IPC_SET: |
| 888 | down(&sem_ids.sem); |
| 889 | err = semctl_down(semid,semnum,cmd,version,arg); |
| 890 | up(&sem_ids.sem); |
| 891 | return err; |
| 892 | default: |
| 893 | return -EINVAL; |
| 894 | } |
| 895 | } |
| 896 | |
| 897 | static inline void lock_semundo(void) |
| 898 | { |
| 899 | struct sem_undo_list *undo_list; |
| 900 | |
| 901 | undo_list = current->sysvsem.undo_list; |
Ingo Molnar | 00a5dfd | 2005-08-05 23:05:27 +0200 | [diff] [blame] | 902 | if (undo_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 903 | spin_lock(&undo_list->lock); |
| 904 | } |
| 905 | |
| 906 | /* This code has an interaction with copy_semundo(). |
| 907 | * Consider; two tasks are sharing the undo_list. task1 |
| 908 | * acquires the undo_list lock in lock_semundo(). If task2 now |
| 909 | * exits before task1 releases the lock (by calling |
| 910 | * unlock_semundo()), then task1 will never call spin_unlock(). |
| 911 | * This leave the sem_undo_list in a locked state. If task1 now creats task3 |
| 912 | * and once again shares the sem_undo_list, the sem_undo_list will still be |
| 913 | * locked, and future SEM_UNDO operations will deadlock. This case is |
| 914 | * dealt with in copy_semundo() by having it reinitialize the spin lock when |
| 915 | * the refcnt goes from 1 to 2. |
| 916 | */ |
| 917 | static inline void unlock_semundo(void) |
| 918 | { |
| 919 | struct sem_undo_list *undo_list; |
| 920 | |
| 921 | undo_list = current->sysvsem.undo_list; |
Ingo Molnar | 00a5dfd | 2005-08-05 23:05:27 +0200 | [diff] [blame] | 922 | if (undo_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | spin_unlock(&undo_list->lock); |
| 924 | } |
| 925 | |
| 926 | |
| 927 | /* If the task doesn't already have a undo_list, then allocate one |
| 928 | * here. We guarantee there is only one thread using this undo list, |
| 929 | * and current is THE ONE |
| 930 | * |
| 931 | * If this allocation and assignment succeeds, but later |
| 932 | * portions of this code fail, there is no need to free the sem_undo_list. |
| 933 | * Just let it stay associated with the task, and it'll be freed later |
| 934 | * at exit time. |
| 935 | * |
| 936 | * This can block, so callers must hold no locks. |
| 937 | */ |
| 938 | static inline int get_undo_list(struct sem_undo_list **undo_listp) |
| 939 | { |
| 940 | struct sem_undo_list *undo_list; |
| 941 | int size; |
| 942 | |
| 943 | undo_list = current->sysvsem.undo_list; |
| 944 | if (!undo_list) { |
| 945 | size = sizeof(struct sem_undo_list); |
| 946 | undo_list = (struct sem_undo_list *) kmalloc(size, GFP_KERNEL); |
| 947 | if (undo_list == NULL) |
| 948 | return -ENOMEM; |
| 949 | memset(undo_list, 0, size); |
Ingo Molnar | 00a5dfd | 2005-08-05 23:05:27 +0200 | [diff] [blame] | 950 | spin_lock_init(&undo_list->lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 951 | atomic_set(&undo_list->refcnt, 1); |
| 952 | current->sysvsem.undo_list = undo_list; |
| 953 | } |
| 954 | *undo_listp = undo_list; |
| 955 | return 0; |
| 956 | } |
| 957 | |
| 958 | static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) |
| 959 | { |
| 960 | struct sem_undo **last, *un; |
| 961 | |
| 962 | last = &ulp->proc_list; |
| 963 | un = *last; |
| 964 | while(un != NULL) { |
| 965 | if(un->semid==semid) |
| 966 | break; |
| 967 | if(un->semid==-1) { |
| 968 | *last=un->proc_next; |
| 969 | kfree(un); |
| 970 | } else { |
| 971 | last=&un->proc_next; |
| 972 | } |
| 973 | un=*last; |
| 974 | } |
| 975 | return un; |
| 976 | } |
| 977 | |
| 978 | static struct sem_undo *find_undo(int semid) |
| 979 | { |
| 980 | struct sem_array *sma; |
| 981 | struct sem_undo_list *ulp; |
| 982 | struct sem_undo *un, *new; |
| 983 | int nsems; |
| 984 | int error; |
| 985 | |
| 986 | error = get_undo_list(&ulp); |
| 987 | if (error) |
| 988 | return ERR_PTR(error); |
| 989 | |
| 990 | lock_semundo(); |
| 991 | un = lookup_undo(ulp, semid); |
| 992 | unlock_semundo(); |
| 993 | if (likely(un!=NULL)) |
| 994 | goto out; |
| 995 | |
| 996 | /* no undo structure around - allocate one. */ |
| 997 | sma = sem_lock(semid); |
| 998 | un = ERR_PTR(-EINVAL); |
| 999 | if(sma==NULL) |
| 1000 | goto out; |
| 1001 | un = ERR_PTR(-EIDRM); |
| 1002 | if (sem_checkid(sma,semid)) { |
| 1003 | sem_unlock(sma); |
| 1004 | goto out; |
| 1005 | } |
| 1006 | nsems = sma->sem_nsems; |
| 1007 | ipc_rcu_getref(sma); |
| 1008 | sem_unlock(sma); |
| 1009 | |
| 1010 | new = (struct sem_undo *) kmalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); |
| 1011 | if (!new) { |
| 1012 | ipc_lock_by_ptr(&sma->sem_perm); |
| 1013 | ipc_rcu_putref(sma); |
| 1014 | sem_unlock(sma); |
| 1015 | return ERR_PTR(-ENOMEM); |
| 1016 | } |
| 1017 | memset(new, 0, sizeof(struct sem_undo) + sizeof(short)*nsems); |
| 1018 | new->semadj = (short *) &new[1]; |
| 1019 | new->semid = semid; |
| 1020 | |
| 1021 | lock_semundo(); |
| 1022 | un = lookup_undo(ulp, semid); |
| 1023 | if (un) { |
| 1024 | unlock_semundo(); |
| 1025 | kfree(new); |
| 1026 | ipc_lock_by_ptr(&sma->sem_perm); |
| 1027 | ipc_rcu_putref(sma); |
| 1028 | sem_unlock(sma); |
| 1029 | goto out; |
| 1030 | } |
| 1031 | ipc_lock_by_ptr(&sma->sem_perm); |
| 1032 | ipc_rcu_putref(sma); |
| 1033 | if (sma->sem_perm.deleted) { |
| 1034 | sem_unlock(sma); |
| 1035 | unlock_semundo(); |
| 1036 | kfree(new); |
| 1037 | un = ERR_PTR(-EIDRM); |
| 1038 | goto out; |
| 1039 | } |
| 1040 | new->proc_next = ulp->proc_list; |
| 1041 | ulp->proc_list = new; |
| 1042 | new->id_next = sma->undo; |
| 1043 | sma->undo = new; |
| 1044 | sem_unlock(sma); |
| 1045 | un = new; |
| 1046 | unlock_semundo(); |
| 1047 | out: |
| 1048 | return un; |
| 1049 | } |
| 1050 | |
| 1051 | asmlinkage long sys_semtimedop(int semid, struct sembuf __user *tsops, |
| 1052 | unsigned nsops, const struct timespec __user *timeout) |
| 1053 | { |
| 1054 | int error = -EINVAL; |
| 1055 | struct sem_array *sma; |
| 1056 | struct sembuf fast_sops[SEMOPM_FAST]; |
| 1057 | struct sembuf* sops = fast_sops, *sop; |
| 1058 | struct sem_undo *un; |
Manfred Spraul | b78755a | 2005-06-23 00:10:06 -0700 | [diff] [blame] | 1059 | int undos = 0, alter = 0, max; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1060 | struct sem_queue queue; |
| 1061 | unsigned long jiffies_left = 0; |
| 1062 | |
| 1063 | if (nsops < 1 || semid < 0) |
| 1064 | return -EINVAL; |
| 1065 | if (nsops > sc_semopm) |
| 1066 | return -E2BIG; |
| 1067 | if(nsops > SEMOPM_FAST) { |
| 1068 | sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); |
| 1069 | if(sops==NULL) |
| 1070 | return -ENOMEM; |
| 1071 | } |
| 1072 | if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { |
| 1073 | error=-EFAULT; |
| 1074 | goto out_free; |
| 1075 | } |
| 1076 | if (timeout) { |
| 1077 | struct timespec _timeout; |
| 1078 | if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { |
| 1079 | error = -EFAULT; |
| 1080 | goto out_free; |
| 1081 | } |
| 1082 | if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || |
| 1083 | _timeout.tv_nsec >= 1000000000L) { |
| 1084 | error = -EINVAL; |
| 1085 | goto out_free; |
| 1086 | } |
| 1087 | jiffies_left = timespec_to_jiffies(&_timeout); |
| 1088 | } |
| 1089 | max = 0; |
| 1090 | for (sop = sops; sop < sops + nsops; sop++) { |
| 1091 | if (sop->sem_num >= max) |
| 1092 | max = sop->sem_num; |
| 1093 | if (sop->sem_flg & SEM_UNDO) |
Manfred Spraul | b78755a | 2005-06-23 00:10:06 -0700 | [diff] [blame] | 1094 | undos = 1; |
| 1095 | if (sop->sem_op != 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1096 | alter = 1; |
| 1097 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1098 | |
| 1099 | retry_undos: |
| 1100 | if (undos) { |
| 1101 | un = find_undo(semid); |
| 1102 | if (IS_ERR(un)) { |
| 1103 | error = PTR_ERR(un); |
| 1104 | goto out_free; |
| 1105 | } |
| 1106 | } else |
| 1107 | un = NULL; |
| 1108 | |
| 1109 | sma = sem_lock(semid); |
| 1110 | error=-EINVAL; |
| 1111 | if(sma==NULL) |
| 1112 | goto out_free; |
| 1113 | error = -EIDRM; |
| 1114 | if (sem_checkid(sma,semid)) |
| 1115 | goto out_unlock_free; |
| 1116 | /* |
| 1117 | * semid identifies are not unique - find_undo may have |
| 1118 | * allocated an undo structure, it was invalidated by an RMID |
| 1119 | * and now a new array with received the same id. Check and retry. |
| 1120 | */ |
| 1121 | if (un && un->semid == -1) { |
| 1122 | sem_unlock(sma); |
| 1123 | goto retry_undos; |
| 1124 | } |
| 1125 | error = -EFBIG; |
| 1126 | if (max >= sma->sem_nsems) |
| 1127 | goto out_unlock_free; |
| 1128 | |
| 1129 | error = -EACCES; |
| 1130 | if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) |
| 1131 | goto out_unlock_free; |
| 1132 | |
| 1133 | error = security_sem_semop(sma, sops, nsops, alter); |
| 1134 | if (error) |
| 1135 | goto out_unlock_free; |
| 1136 | |
| 1137 | error = try_atomic_semop (sma, sops, nsops, un, current->tgid); |
| 1138 | if (error <= 0) { |
| 1139 | if (alter && error == 0) |
| 1140 | update_queue (sma); |
| 1141 | goto out_unlock_free; |
| 1142 | } |
| 1143 | |
| 1144 | /* We need to sleep on this operation, so we put the current |
| 1145 | * task into the pending queue and go to sleep. |
| 1146 | */ |
| 1147 | |
| 1148 | queue.sma = sma; |
| 1149 | queue.sops = sops; |
| 1150 | queue.nsops = nsops; |
| 1151 | queue.undo = un; |
| 1152 | queue.pid = current->tgid; |
| 1153 | queue.id = semid; |
| 1154 | queue.alter = alter; |
| 1155 | if (alter) |
| 1156 | append_to_queue(sma ,&queue); |
| 1157 | else |
| 1158 | prepend_to_queue(sma ,&queue); |
| 1159 | |
| 1160 | queue.status = -EINTR; |
| 1161 | queue.sleeper = current; |
| 1162 | current->state = TASK_INTERRUPTIBLE; |
| 1163 | sem_unlock(sma); |
| 1164 | |
| 1165 | if (timeout) |
| 1166 | jiffies_left = schedule_timeout(jiffies_left); |
| 1167 | else |
| 1168 | schedule(); |
| 1169 | |
| 1170 | error = queue.status; |
| 1171 | while(unlikely(error == IN_WAKEUP)) { |
| 1172 | cpu_relax(); |
| 1173 | error = queue.status; |
| 1174 | } |
| 1175 | |
| 1176 | if (error != -EINTR) { |
| 1177 | /* fast path: update_queue already obtained all requested |
| 1178 | * resources */ |
| 1179 | goto out_free; |
| 1180 | } |
| 1181 | |
| 1182 | sma = sem_lock(semid); |
| 1183 | if(sma==NULL) { |
| 1184 | if(queue.prev != NULL) |
| 1185 | BUG(); |
| 1186 | error = -EIDRM; |
| 1187 | goto out_free; |
| 1188 | } |
| 1189 | |
| 1190 | /* |
| 1191 | * If queue.status != -EINTR we are woken up by another process |
| 1192 | */ |
| 1193 | error = queue.status; |
| 1194 | if (error != -EINTR) { |
| 1195 | goto out_unlock_free; |
| 1196 | } |
| 1197 | |
| 1198 | /* |
| 1199 | * If an interrupt occurred we have to clean up the queue |
| 1200 | */ |
| 1201 | if (timeout && jiffies_left == 0) |
| 1202 | error = -EAGAIN; |
| 1203 | remove_from_queue(sma,&queue); |
| 1204 | goto out_unlock_free; |
| 1205 | |
| 1206 | out_unlock_free: |
| 1207 | sem_unlock(sma); |
| 1208 | out_free: |
| 1209 | if(sops != fast_sops) |
| 1210 | kfree(sops); |
| 1211 | return error; |
| 1212 | } |
| 1213 | |
| 1214 | asmlinkage long sys_semop (int semid, struct sembuf __user *tsops, unsigned nsops) |
| 1215 | { |
| 1216 | return sys_semtimedop(semid, tsops, nsops, NULL); |
| 1217 | } |
| 1218 | |
| 1219 | /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between |
| 1220 | * parent and child tasks. |
| 1221 | * |
| 1222 | * See the notes above unlock_semundo() regarding the spin_lock_init() |
| 1223 | * in this code. Initialize the undo_list->lock here instead of get_undo_list() |
| 1224 | * because of the reasoning in the comment above unlock_semundo. |
| 1225 | */ |
| 1226 | |
| 1227 | int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) |
| 1228 | { |
| 1229 | struct sem_undo_list *undo_list; |
| 1230 | int error; |
| 1231 | |
| 1232 | if (clone_flags & CLONE_SYSVSEM) { |
| 1233 | error = get_undo_list(&undo_list); |
| 1234 | if (error) |
| 1235 | return error; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | atomic_inc(&undo_list->refcnt); |
| 1237 | tsk->sysvsem.undo_list = undo_list; |
| 1238 | } else |
| 1239 | tsk->sysvsem.undo_list = NULL; |
| 1240 | |
| 1241 | return 0; |
| 1242 | } |
| 1243 | |
| 1244 | /* |
| 1245 | * add semadj values to semaphores, free undo structures. |
| 1246 | * undo structures are not freed when semaphore arrays are destroyed |
| 1247 | * so some of them may be out of date. |
| 1248 | * IMPLEMENTATION NOTE: There is some confusion over whether the |
| 1249 | * set of adjustments that needs to be done should be done in an atomic |
| 1250 | * manner or not. That is, if we are attempting to decrement the semval |
| 1251 | * should we queue up and wait until we can do so legally? |
| 1252 | * The original implementation attempted to do this (queue and wait). |
| 1253 | * The current implementation does not do so. The POSIX standard |
| 1254 | * and SVID should be consulted to determine what behavior is mandated. |
| 1255 | */ |
| 1256 | void exit_sem(struct task_struct *tsk) |
| 1257 | { |
| 1258 | struct sem_undo_list *undo_list; |
| 1259 | struct sem_undo *u, **up; |
| 1260 | |
| 1261 | undo_list = tsk->sysvsem.undo_list; |
| 1262 | if (!undo_list) |
| 1263 | return; |
| 1264 | |
| 1265 | if (!atomic_dec_and_test(&undo_list->refcnt)) |
| 1266 | return; |
| 1267 | |
| 1268 | /* There's no need to hold the semundo list lock, as current |
| 1269 | * is the last task exiting for this undo list. |
| 1270 | */ |
| 1271 | for (up = &undo_list->proc_list; (u = *up); *up = u->proc_next, kfree(u)) { |
| 1272 | struct sem_array *sma; |
| 1273 | int nsems, i; |
| 1274 | struct sem_undo *un, **unp; |
| 1275 | int semid; |
| 1276 | |
| 1277 | semid = u->semid; |
| 1278 | |
| 1279 | if(semid == -1) |
| 1280 | continue; |
| 1281 | sma = sem_lock(semid); |
| 1282 | if (sma == NULL) |
| 1283 | continue; |
| 1284 | |
| 1285 | if (u->semid == -1) |
| 1286 | goto next_entry; |
| 1287 | |
| 1288 | BUG_ON(sem_checkid(sma,u->semid)); |
| 1289 | |
| 1290 | /* remove u from the sma->undo list */ |
| 1291 | for (unp = &sma->undo; (un = *unp); unp = &un->id_next) { |
| 1292 | if (u == un) |
| 1293 | goto found; |
| 1294 | } |
| 1295 | printk ("exit_sem undo list error id=%d\n", u->semid); |
| 1296 | goto next_entry; |
| 1297 | found: |
| 1298 | *unp = un->id_next; |
| 1299 | /* perform adjustments registered in u */ |
| 1300 | nsems = sma->sem_nsems; |
| 1301 | for (i = 0; i < nsems; i++) { |
| 1302 | struct sem * sem = &sma->sem_base[i]; |
| 1303 | if (u->semadj[i]) { |
| 1304 | sem->semval += u->semadj[i]; |
| 1305 | /* |
| 1306 | * Range checks of the new semaphore value, |
| 1307 | * not defined by sus: |
| 1308 | * - Some unices ignore the undo entirely |
| 1309 | * (e.g. HP UX 11i 11.22, Tru64 V5.1) |
| 1310 | * - some cap the value (e.g. FreeBSD caps |
| 1311 | * at 0, but doesn't enforce SEMVMX) |
| 1312 | * |
| 1313 | * Linux caps the semaphore value, both at 0 |
| 1314 | * and at SEMVMX. |
| 1315 | * |
| 1316 | * Manfred <manfred@colorfullife.com> |
| 1317 | */ |
| 1318 | if (sem->semval < 0) |
| 1319 | sem->semval = 0; |
| 1320 | if (sem->semval > SEMVMX) |
| 1321 | sem->semval = SEMVMX; |
| 1322 | sem->sempid = current->tgid; |
| 1323 | } |
| 1324 | } |
| 1325 | sma->sem_otime = get_seconds(); |
| 1326 | /* maybe some queued-up processes were waiting for this */ |
| 1327 | update_queue(sma); |
| 1328 | next_entry: |
| 1329 | sem_unlock(sma); |
| 1330 | } |
| 1331 | kfree(undo_list); |
| 1332 | } |
| 1333 | |
| 1334 | #ifdef CONFIG_PROC_FS |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 1335 | static int sysvipc_sem_proc_show(struct seq_file *s, void *it) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1336 | { |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 1337 | struct sem_array *sma = it; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1338 | |
Mike Waychison | 19b4946 | 2005-09-06 15:17:10 -0700 | [diff] [blame] | 1339 | return seq_printf(s, |
| 1340 | "%10d %10d %4o %10lu %5u %5u %5u %5u %10lu %10lu\n", |
| 1341 | sma->sem_perm.key, |
| 1342 | sma->sem_id, |
| 1343 | sma->sem_perm.mode, |
| 1344 | sma->sem_nsems, |
| 1345 | sma->sem_perm.uid, |
| 1346 | sma->sem_perm.gid, |
| 1347 | sma->sem_perm.cuid, |
| 1348 | sma->sem_perm.cgid, |
| 1349 | sma->sem_otime, |
| 1350 | sma->sem_ctime); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1351 | } |
| 1352 | #endif |