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 59 * (c) 1999 Manfred Spraul <manfred@colorfullife.com> 60 * Enforced range limit on SEM_UNDO 61 * (c) 2001 Red Hat Inc 62 * Lockless wakeup 63 * (c) 2003 Manfred Spraul <manfred@colorfullife.com> 64 * 65 * support for audit of ipc object properties and permission changes 66 * Dustin Kirkland <dustin.kirkland@us.ibm.com> 67 * 68 * namespaces support 69 * OpenVZ, SWsoft Inc. 70 * Pavel Emelianov <xemul@openvz.org> 71 */ 72 73 #include <linux/slab.h> 74 #include <linux/spinlock.h> 75 #include <linux/init.h> 76 #include <linux/proc_fs.h> 77 #include <linux/time.h> 78 #include <linux/security.h> 79 #include <linux/syscalls.h> 80 #include <linux/audit.h> 81 #include <linux/capability.h> 82 #include <linux/seq_file.h> 83 #include <linux/rwsem.h> 84 #include <linux/nsproxy.h> 85 #include <linux/ipc_namespace.h> 86 87 #include <asm/uaccess.h> 88 #include "util.h" 89 90 #define sem_ids(ns) ((ns)->ids[IPC_SEM_IDS]) 91 92 #define sem_unlock(sma) ipc_unlock(&(sma)->sem_perm) 93 #define sem_checkid(sma, semid) ipc_checkid(&sma->sem_perm, semid) 94 95 static int newary(struct ipc_namespace *, struct ipc_params *); 96 static void freeary(struct ipc_namespace *, struct kern_ipc_perm *); 97 #ifdef CONFIG_PROC_FS 98 static int sysvipc_sem_proc_show(struct seq_file *s, void *it); 99 #endif 100 101 #define SEMMSL_FAST 256 /* 512 bytes on stack */ 102 #define SEMOPM_FAST 64 /* ~ 372 bytes on stack */ 103 104 /* 105 * linked list protection: 106 * sem_undo.id_next, 107 * sem_array.sem_pending{,last}, 108 * sem_array.sem_undo: sem_lock() for read/write 109 * sem_undo.proc_next: only "current" is allowed to read/write that field. 110 * 111 */ 112 113 #define sc_semmsl sem_ctls[0] 114 #define sc_semmns sem_ctls[1] 115 #define sc_semopm sem_ctls[2] 116 #define sc_semmni sem_ctls[3] 117 118 void sem_init_ns(struct ipc_namespace *ns) 119 { 120 ns->sc_semmsl = SEMMSL; 121 ns->sc_semmns = SEMMNS; 122 ns->sc_semopm = SEMOPM; 123 ns->sc_semmni = SEMMNI; 124 ns->used_sems = 0; 125 ipc_init_ids(&ns->ids[IPC_SEM_IDS]); 126 } 127 128 #ifdef CONFIG_IPC_NS 129 void sem_exit_ns(struct ipc_namespace *ns) 130 { 131 free_ipcs(ns, &sem_ids(ns), freeary); 132 idr_destroy(&ns->ids[IPC_SEM_IDS].ipcs_idr); 133 } 134 #endif 135 136 void __init sem_init (void) 137 { 138 sem_init_ns(&init_ipc_ns); 139 ipc_init_proc_interface("sysvipc/sem", 140 " key semid perms nsems uid gid cuid cgid otime ctime\n", 141 IPC_SEM_IDS, sysvipc_sem_proc_show); 142 } 143 144 /* 145 * sem_lock_(check_) routines are called in the paths where the rw_mutex 146 * is not held. 147 */ 148 static inline struct sem_array *sem_lock(struct ipc_namespace *ns, int id) 149 { 150 struct kern_ipc_perm *ipcp = ipc_lock(&sem_ids(ns), id); 151 152 if (IS_ERR(ipcp)) 153 return (struct sem_array *)ipcp; 154 155 return container_of(ipcp, struct sem_array, sem_perm); 156 } 157 158 static inline struct sem_array *sem_lock_check(struct ipc_namespace *ns, 159 int id) 160 { 161 struct kern_ipc_perm *ipcp = ipc_lock_check(&sem_ids(ns), id); 162 163 if (IS_ERR(ipcp)) 164 return (struct sem_array *)ipcp; 165 166 return container_of(ipcp, struct sem_array, sem_perm); 167 } 168 169 static inline void sem_lock_and_putref(struct sem_array *sma) 170 { 171 ipc_lock_by_ptr(&sma->sem_perm); 172 ipc_rcu_putref(sma); 173 } 174 175 static inline void sem_getref_and_unlock(struct sem_array *sma) 176 { 177 ipc_rcu_getref(sma); 178 ipc_unlock(&(sma)->sem_perm); 179 } 180 181 static inline void sem_putref(struct sem_array *sma) 182 { 183 ipc_lock_by_ptr(&sma->sem_perm); 184 ipc_rcu_putref(sma); 185 ipc_unlock(&(sma)->sem_perm); 186 } 187 188 static inline void sem_rmid(struct ipc_namespace *ns, struct sem_array *s) 189 { 190 ipc_rmid(&sem_ids(ns), &s->sem_perm); 191 } 192 193 /* 194 * Lockless wakeup algorithm: 195 * Without the check/retry algorithm a lockless wakeup is possible: 196 * - queue.status is initialized to -EINTR before blocking. 197 * - wakeup is performed by 198 * * unlinking the queue entry from sma->sem_pending 199 * * setting queue.status to IN_WAKEUP 200 * This is the notification for the blocked thread that a 201 * result value is imminent. 202 * * call wake_up_process 203 * * set queue.status to the final value. 204 * - the previously blocked thread checks queue.status: 205 * * if it's IN_WAKEUP, then it must wait until the value changes 206 * * if it's not -EINTR, then the operation was completed by 207 * update_queue. semtimedop can return queue.status without 208 * performing any operation on the sem array. 209 * * otherwise it must acquire the spinlock and check what's up. 210 * 211 * The two-stage algorithm is necessary to protect against the following 212 * races: 213 * - if queue.status is set after wake_up_process, then the woken up idle 214 * thread could race forward and try (and fail) to acquire sma->lock 215 * before update_queue had a chance to set queue.status 216 * - if queue.status is written before wake_up_process and if the 217 * blocked process is woken up by a signal between writing 218 * queue.status and the wake_up_process, then the woken up 219 * process could return from semtimedop and die by calling 220 * sys_exit before wake_up_process is called. Then wake_up_process 221 * will oops, because the task structure is already invalid. 222 * (yes, this happened on s390 with sysv msg). 223 * 224 */ 225 #define IN_WAKEUP 1 226 227 /** 228 * newary - Create a new semaphore set 229 * @ns: namespace 230 * @params: ptr to the structure that contains key, semflg and nsems 231 * 232 * Called with sem_ids.rw_mutex held (as a writer) 233 */ 234 235 static int newary(struct ipc_namespace *ns, struct ipc_params *params) 236 { 237 int id; 238 int retval; 239 struct sem_array *sma; 240 int size; 241 key_t key = params->key; 242 int nsems = params->u.nsems; 243 int semflg = params->flg; 244 int i; 245 246 if (!nsems) 247 return -EINVAL; 248 if (ns->used_sems + nsems > ns->sc_semmns) 249 return -ENOSPC; 250 251 size = sizeof (*sma) + nsems * sizeof (struct sem); 252 sma = ipc_rcu_alloc(size); 253 if (!sma) { 254 return -ENOMEM; 255 } 256 memset (sma, 0, size); 257 258 sma->sem_perm.mode = (semflg & S_IRWXUGO); 259 sma->sem_perm.key = key; 260 261 sma->sem_perm.security = NULL; 262 retval = security_sem_alloc(sma); 263 if (retval) { 264 ipc_rcu_putref(sma); 265 return retval; 266 } 267 268 id = ipc_addid(&sem_ids(ns), &sma->sem_perm, ns->sc_semmni); 269 if (id < 0) { 270 security_sem_free(sma); 271 ipc_rcu_putref(sma); 272 return id; 273 } 274 ns->used_sems += nsems; 275 276 sma->sem_base = (struct sem *) &sma[1]; 277 278 for (i = 0; i < nsems; i++) 279 INIT_LIST_HEAD(&sma->sem_base[i].sem_pending); 280 281 sma->complex_count = 0; 282 INIT_LIST_HEAD(&sma->sem_pending); 283 INIT_LIST_HEAD(&sma->list_id); 284 sma->sem_nsems = nsems; 285 sma->sem_ctime = get_seconds(); 286 sem_unlock(sma); 287 288 return sma->sem_perm.id; 289 } 290 291 292 /* 293 * Called with sem_ids.rw_mutex and ipcp locked. 294 */ 295 static inline int sem_security(struct kern_ipc_perm *ipcp, int semflg) 296 { 297 struct sem_array *sma; 298 299 sma = container_of(ipcp, struct sem_array, sem_perm); 300 return security_sem_associate(sma, semflg); 301 } 302 303 /* 304 * Called with sem_ids.rw_mutex and ipcp locked. 305 */ 306 static inline int sem_more_checks(struct kern_ipc_perm *ipcp, 307 struct ipc_params *params) 308 { 309 struct sem_array *sma; 310 311 sma = container_of(ipcp, struct sem_array, sem_perm); 312 if (params->u.nsems > sma->sem_nsems) 313 return -EINVAL; 314 315 return 0; 316 } 317 318 SYSCALL_DEFINE3(semget, key_t, key, int, nsems, int, semflg) 319 { 320 struct ipc_namespace *ns; 321 struct ipc_ops sem_ops; 322 struct ipc_params sem_params; 323 324 ns = current->nsproxy->ipc_ns; 325 326 if (nsems < 0 || nsems > ns->sc_semmsl) 327 return -EINVAL; 328 329 sem_ops.getnew = newary; 330 sem_ops.associate = sem_security; 331 sem_ops.more_checks = sem_more_checks; 332 333 sem_params.key = key; 334 sem_params.flg = semflg; 335 sem_params.u.nsems = nsems; 336 337 return ipcget(ns, &sem_ids(ns), &sem_ops, &sem_params); 338 } 339 340 /* 341 * Determine whether a sequence of semaphore operations would succeed 342 * all at once. Return 0 if yes, 1 if need to sleep, else return error code. 343 */ 344 345 static int try_atomic_semop (struct sem_array * sma, struct sembuf * sops, 346 int nsops, struct sem_undo *un, int pid) 347 { 348 int result, sem_op; 349 struct sembuf *sop; 350 struct sem * curr; 351 352 for (sop = sops; sop < sops + nsops; sop++) { 353 curr = sma->sem_base + sop->sem_num; 354 sem_op = sop->sem_op; 355 result = curr->semval; 356 357 if (!sem_op && result) 358 goto would_block; 359 360 result += sem_op; 361 if (result < 0) 362 goto would_block; 363 if (result > SEMVMX) 364 goto out_of_range; 365 if (sop->sem_flg & SEM_UNDO) { 366 int undo = un->semadj[sop->sem_num] - sem_op; 367 /* 368 * Exceeding the undo range is an error. 369 */ 370 if (undo < (-SEMAEM - 1) || undo > SEMAEM) 371 goto out_of_range; 372 } 373 curr->semval = result; 374 } 375 376 sop--; 377 while (sop >= sops) { 378 sma->sem_base[sop->sem_num].sempid = pid; 379 if (sop->sem_flg & SEM_UNDO) 380 un->semadj[sop->sem_num] -= sop->sem_op; 381 sop--; 382 } 383 384 sma->sem_otime = get_seconds(); 385 return 0; 386 387 out_of_range: 388 result = -ERANGE; 389 goto undo; 390 391 would_block: 392 if (sop->sem_flg & IPC_NOWAIT) 393 result = -EAGAIN; 394 else 395 result = 1; 396 397 undo: 398 sop--; 399 while (sop >= sops) { 400 sma->sem_base[sop->sem_num].semval -= sop->sem_op; 401 sop--; 402 } 403 404 return result; 405 } 406 407 /* 408 * Wake up a process waiting on the sem queue with a given error. 409 * The queue is invalid (may not be accessed) after the function returns. 410 */ 411 static void wake_up_sem_queue(struct sem_queue *q, int error) 412 { 413 /* 414 * Hold preempt off so that we don't get preempted and have the 415 * wakee busy-wait until we're scheduled back on. We're holding 416 * locks here so it may not strictly be needed, however if the 417 * locks become preemptible then this prevents such a problem. 418 */ 419 preempt_disable(); 420 q->status = IN_WAKEUP; 421 wake_up_process(q->sleeper); 422 /* hands-off: q can disappear immediately after writing q->status. */ 423 smp_wmb(); 424 q->status = error; 425 preempt_enable(); 426 } 427 428 static void unlink_queue(struct sem_array *sma, struct sem_queue *q) 429 { 430 list_del(&q->list); 431 if (q->nsops == 1) 432 list_del(&q->simple_list); 433 else 434 sma->complex_count--; 435 } 436 437 438 /** 439 * update_queue(sma, semnum): Look for tasks that can be completed. 440 * @sma: semaphore array. 441 * @semnum: semaphore that was modified. 442 * 443 * update_queue must be called after a semaphore in a semaphore array 444 * was modified. If multiple semaphore were modified, then @semnum 445 * must be set to -1. 446 */ 447 static void update_queue(struct sem_array *sma, int semnum) 448 { 449 struct sem_queue *q; 450 struct list_head *walk; 451 struct list_head *pending_list; 452 int offset; 453 454 /* if there are complex operations around, then knowing the semaphore 455 * that was modified doesn't help us. Assume that multiple semaphores 456 * were modified. 457 */ 458 if (sma->complex_count) 459 semnum = -1; 460 461 if (semnum == -1) { 462 pending_list = &sma->sem_pending; 463 offset = offsetof(struct sem_queue, list); 464 } else { 465 pending_list = &sma->sem_base[semnum].sem_pending; 466 offset = offsetof(struct sem_queue, simple_list); 467 } 468 469 again: 470 walk = pending_list->next; 471 while (walk != pending_list) { 472 int error, alter; 473 474 q = (struct sem_queue *)((char *)walk - offset); 475 walk = walk->next; 476 477 /* If we are scanning the single sop, per-semaphore list of 478 * one semaphore and that semaphore is 0, then it is not 479 * necessary to scan the "alter" entries: simple increments 480 * that affect only one entry succeed immediately and cannot 481 * be in the per semaphore pending queue, and decrements 482 * cannot be successful if the value is already 0. 483 */ 484 if (semnum != -1 && sma->sem_base[semnum].semval == 0 && 485 q->alter) 486 break; 487 488 error = try_atomic_semop(sma, q->sops, q->nsops, 489 q->undo, q->pid); 490 491 /* Does q->sleeper still need to sleep? */ 492 if (error > 0) 493 continue; 494 495 unlink_queue(sma, q); 496 497 /* 498 * The next operation that must be checked depends on the type 499 * of the completed operation: 500 * - if the operation modified the array, then restart from the 501 * head of the queue and check for threads that might be 502 * waiting for the new semaphore values. 503 * - if the operation didn't modify the array, then just 504 * continue. 505 */ 506 alter = q->alter; 507 wake_up_sem_queue(q, error); 508 if (alter && !error) 509 goto again; 510 } 511 } 512 513 /* The following counts are associated to each semaphore: 514 * semncnt number of tasks waiting on semval being nonzero 515 * semzcnt number of tasks waiting on semval being zero 516 * This model assumes that a task waits on exactly one semaphore. 517 * Since semaphore operations are to be performed atomically, tasks actually 518 * wait on a whole sequence of semaphores simultaneously. 519 * The counts we return here are a rough approximation, but still 520 * warrant that semncnt+semzcnt>0 if the task is on the pending queue. 521 */ 522 static int count_semncnt (struct sem_array * sma, ushort semnum) 523 { 524 int semncnt; 525 struct sem_queue * q; 526 527 semncnt = 0; 528 list_for_each_entry(q, &sma->sem_pending, list) { 529 struct sembuf * sops = q->sops; 530 int nsops = q->nsops; 531 int i; 532 for (i = 0; i < nsops; i++) 533 if (sops[i].sem_num == semnum 534 && (sops[i].sem_op < 0) 535 && !(sops[i].sem_flg & IPC_NOWAIT)) 536 semncnt++; 537 } 538 return semncnt; 539 } 540 541 static int count_semzcnt (struct sem_array * sma, ushort semnum) 542 { 543 int semzcnt; 544 struct sem_queue * q; 545 546 semzcnt = 0; 547 list_for_each_entry(q, &sma->sem_pending, list) { 548 struct sembuf * sops = q->sops; 549 int nsops = q->nsops; 550 int i; 551 for (i = 0; i < nsops; i++) 552 if (sops[i].sem_num == semnum 553 && (sops[i].sem_op == 0) 554 && !(sops[i].sem_flg & IPC_NOWAIT)) 555 semzcnt++; 556 } 557 return semzcnt; 558 } 559 560 static void free_un(struct rcu_head *head) 561 { 562 struct sem_undo *un = container_of(head, struct sem_undo, rcu); 563 kfree(un); 564 } 565 566 /* Free a semaphore set. freeary() is called with sem_ids.rw_mutex locked 567 * as a writer and the spinlock for this semaphore set hold. sem_ids.rw_mutex 568 * remains locked on exit. 569 */ 570 static void freeary(struct ipc_namespace *ns, struct kern_ipc_perm *ipcp) 571 { 572 struct sem_undo *un, *tu; 573 struct sem_queue *q, *tq; 574 struct sem_array *sma = container_of(ipcp, struct sem_array, sem_perm); 575 576 /* Free the existing undo structures for this semaphore set. */ 577 assert_spin_locked(&sma->sem_perm.lock); 578 list_for_each_entry_safe(un, tu, &sma->list_id, list_id) { 579 list_del(&un->list_id); 580 spin_lock(&un->ulp->lock); 581 un->semid = -1; 582 list_del_rcu(&un->list_proc); 583 spin_unlock(&un->ulp->lock); 584 call_rcu(&un->rcu, free_un); 585 } 586 587 /* Wake up all pending processes and let them fail with EIDRM. */ 588 list_for_each_entry_safe(q, tq, &sma->sem_pending, list) { 589 unlink_queue(sma, q); 590 wake_up_sem_queue(q, -EIDRM); 591 } 592 593 /* Remove the semaphore set from the IDR */ 594 sem_rmid(ns, sma); 595 sem_unlock(sma); 596 597 ns->used_sems -= sma->sem_nsems; 598 security_sem_free(sma); 599 ipc_rcu_putref(sma); 600 } 601 602 static unsigned long copy_semid_to_user(void __user *buf, struct semid64_ds *in, int version) 603 { 604 switch(version) { 605 case IPC_64: 606 return copy_to_user(buf, in, sizeof(*in)); 607 case IPC_OLD: 608 { 609 struct semid_ds out; 610 611 ipc64_perm_to_ipc_perm(&in->sem_perm, &out.sem_perm); 612 613 out.sem_otime = in->sem_otime; 614 out.sem_ctime = in->sem_ctime; 615 out.sem_nsems = in->sem_nsems; 616 617 return copy_to_user(buf, &out, sizeof(out)); 618 } 619 default: 620 return -EINVAL; 621 } 622 } 623 624 static int semctl_nolock(struct ipc_namespace *ns, int semid, 625 int cmd, int version, union semun arg) 626 { 627 int err; 628 struct sem_array *sma; 629 630 switch(cmd) { 631 case IPC_INFO: 632 case SEM_INFO: 633 { 634 struct seminfo seminfo; 635 int max_id; 636 637 err = security_sem_semctl(NULL, cmd); 638 if (err) 639 return err; 640 641 memset(&seminfo,0,sizeof(seminfo)); 642 seminfo.semmni = ns->sc_semmni; 643 seminfo.semmns = ns->sc_semmns; 644 seminfo.semmsl = ns->sc_semmsl; 645 seminfo.semopm = ns->sc_semopm; 646 seminfo.semvmx = SEMVMX; 647 seminfo.semmnu = SEMMNU; 648 seminfo.semmap = SEMMAP; 649 seminfo.semume = SEMUME; 650 down_read(&sem_ids(ns).rw_mutex); 651 if (cmd == SEM_INFO) { 652 seminfo.semusz = sem_ids(ns).in_use; 653 seminfo.semaem = ns->used_sems; 654 } else { 655 seminfo.semusz = SEMUSZ; 656 seminfo.semaem = SEMAEM; 657 } 658 max_id = ipc_get_maxid(&sem_ids(ns)); 659 up_read(&sem_ids(ns).rw_mutex); 660 if (copy_to_user (arg.__buf, &seminfo, sizeof(struct seminfo))) 661 return -EFAULT; 662 return (max_id < 0) ? 0: max_id; 663 } 664 case IPC_STAT: 665 case SEM_STAT: 666 { 667 struct semid64_ds tbuf; 668 int id; 669 670 if (cmd == SEM_STAT) { 671 sma = sem_lock(ns, semid); 672 if (IS_ERR(sma)) 673 return PTR_ERR(sma); 674 id = sma->sem_perm.id; 675 } else { 676 sma = sem_lock_check(ns, semid); 677 if (IS_ERR(sma)) 678 return PTR_ERR(sma); 679 id = 0; 680 } 681 682 err = -EACCES; 683 if (ipcperms (&sma->sem_perm, S_IRUGO)) 684 goto out_unlock; 685 686 err = security_sem_semctl(sma, cmd); 687 if (err) 688 goto out_unlock; 689 690 memset(&tbuf, 0, sizeof(tbuf)); 691 692 kernel_to_ipc64_perm(&sma->sem_perm, &tbuf.sem_perm); 693 tbuf.sem_otime = sma->sem_otime; 694 tbuf.sem_ctime = sma->sem_ctime; 695 tbuf.sem_nsems = sma->sem_nsems; 696 sem_unlock(sma); 697 if (copy_semid_to_user (arg.buf, &tbuf, version)) 698 return -EFAULT; 699 return id; 700 } 701 default: 702 return -EINVAL; 703 } 704 out_unlock: 705 sem_unlock(sma); 706 return err; 707 } 708 709 static int semctl_main(struct ipc_namespace *ns, int semid, int semnum, 710 int cmd, int version, union semun arg) 711 { 712 struct sem_array *sma; 713 struct sem* curr; 714 int err; 715 ushort fast_sem_io[SEMMSL_FAST]; 716 ushort* sem_io = fast_sem_io; 717 int nsems; 718 719 sma = sem_lock_check(ns, semid); 720 if (IS_ERR(sma)) 721 return PTR_ERR(sma); 722 723 nsems = sma->sem_nsems; 724 725 err = -EACCES; 726 if (ipcperms (&sma->sem_perm, (cmd==SETVAL||cmd==SETALL)?S_IWUGO:S_IRUGO)) 727 goto out_unlock; 728 729 err = security_sem_semctl(sma, cmd); 730 if (err) 731 goto out_unlock; 732 733 err = -EACCES; 734 switch (cmd) { 735 case GETALL: 736 { 737 ushort __user *array = arg.array; 738 int i; 739 740 if(nsems > SEMMSL_FAST) { 741 sem_getref_and_unlock(sma); 742 743 sem_io = ipc_alloc(sizeof(ushort)*nsems); 744 if(sem_io == NULL) { 745 sem_putref(sma); 746 return -ENOMEM; 747 } 748 749 sem_lock_and_putref(sma); 750 if (sma->sem_perm.deleted) { 751 sem_unlock(sma); 752 err = -EIDRM; 753 goto out_free; 754 } 755 } 756 757 for (i = 0; i < sma->sem_nsems; i++) 758 sem_io[i] = sma->sem_base[i].semval; 759 sem_unlock(sma); 760 err = 0; 761 if(copy_to_user(array, sem_io, nsems*sizeof(ushort))) 762 err = -EFAULT; 763 goto out_free; 764 } 765 case SETALL: 766 { 767 int i; 768 struct sem_undo *un; 769 770 sem_getref_and_unlock(sma); 771 772 if(nsems > SEMMSL_FAST) { 773 sem_io = ipc_alloc(sizeof(ushort)*nsems); 774 if(sem_io == NULL) { 775 sem_putref(sma); 776 return -ENOMEM; 777 } 778 } 779 780 if (copy_from_user (sem_io, arg.array, nsems*sizeof(ushort))) { 781 sem_putref(sma); 782 err = -EFAULT; 783 goto out_free; 784 } 785 786 for (i = 0; i < nsems; i++) { 787 if (sem_io[i] > SEMVMX) { 788 sem_putref(sma); 789 err = -ERANGE; 790 goto out_free; 791 } 792 } 793 sem_lock_and_putref(sma); 794 if (sma->sem_perm.deleted) { 795 sem_unlock(sma); 796 err = -EIDRM; 797 goto out_free; 798 } 799 800 for (i = 0; i < nsems; i++) 801 sma->sem_base[i].semval = sem_io[i]; 802 803 assert_spin_locked(&sma->sem_perm.lock); 804 list_for_each_entry(un, &sma->list_id, list_id) { 805 for (i = 0; i < nsems; i++) 806 un->semadj[i] = 0; 807 } 808 sma->sem_ctime = get_seconds(); 809 /* maybe some queued-up processes were waiting for this */ 810 update_queue(sma, -1); 811 err = 0; 812 goto out_unlock; 813 } 814 /* GETVAL, GETPID, GETNCTN, GETZCNT, SETVAL: fall-through */ 815 } 816 err = -EINVAL; 817 if(semnum < 0 || semnum >= nsems) 818 goto out_unlock; 819 820 curr = &sma->sem_base[semnum]; 821 822 switch (cmd) { 823 case GETVAL: 824 err = curr->semval; 825 goto out_unlock; 826 case GETPID: 827 err = curr->sempid; 828 goto out_unlock; 829 case GETNCNT: 830 err = count_semncnt(sma,semnum); 831 goto out_unlock; 832 case GETZCNT: 833 err = count_semzcnt(sma,semnum); 834 goto out_unlock; 835 case SETVAL: 836 { 837 int val = arg.val; 838 struct sem_undo *un; 839 840 err = -ERANGE; 841 if (val > SEMVMX || val < 0) 842 goto out_unlock; 843 844 assert_spin_locked(&sma->sem_perm.lock); 845 list_for_each_entry(un, &sma->list_id, list_id) 846 un->semadj[semnum] = 0; 847 848 curr->semval = val; 849 curr->sempid = task_tgid_vnr(current); 850 sma->sem_ctime = get_seconds(); 851 /* maybe some queued-up processes were waiting for this */ 852 update_queue(sma, semnum); 853 err = 0; 854 goto out_unlock; 855 } 856 } 857 out_unlock: 858 sem_unlock(sma); 859 out_free: 860 if(sem_io != fast_sem_io) 861 ipc_free(sem_io, sizeof(ushort)*nsems); 862 return err; 863 } 864 865 static inline unsigned long 866 copy_semid_from_user(struct semid64_ds *out, void __user *buf, int version) 867 { 868 switch(version) { 869 case IPC_64: 870 if (copy_from_user(out, buf, sizeof(*out))) 871 return -EFAULT; 872 return 0; 873 case IPC_OLD: 874 { 875 struct semid_ds tbuf_old; 876 877 if(copy_from_user(&tbuf_old, buf, sizeof(tbuf_old))) 878 return -EFAULT; 879 880 out->sem_perm.uid = tbuf_old.sem_perm.uid; 881 out->sem_perm.gid = tbuf_old.sem_perm.gid; 882 out->sem_perm.mode = tbuf_old.sem_perm.mode; 883 884 return 0; 885 } 886 default: 887 return -EINVAL; 888 } 889 } 890 891 /* 892 * This function handles some semctl commands which require the rw_mutex 893 * to be held in write mode. 894 * NOTE: no locks must be held, the rw_mutex is taken inside this function. 895 */ 896 static int semctl_down(struct ipc_namespace *ns, int semid, 897 int cmd, int version, union semun arg) 898 { 899 struct sem_array *sma; 900 int err; 901 struct semid64_ds semid64; 902 struct kern_ipc_perm *ipcp; 903 904 if(cmd == IPC_SET) { 905 if (copy_semid_from_user(&semid64, arg.buf, version)) 906 return -EFAULT; 907 } 908 909 ipcp = ipcctl_pre_down(&sem_ids(ns), semid, cmd, &semid64.sem_perm, 0); 910 if (IS_ERR(ipcp)) 911 return PTR_ERR(ipcp); 912 913 sma = container_of(ipcp, struct sem_array, sem_perm); 914 915 err = security_sem_semctl(sma, cmd); 916 if (err) 917 goto out_unlock; 918 919 switch(cmd){ 920 case IPC_RMID: 921 freeary(ns, ipcp); 922 goto out_up; 923 case IPC_SET: 924 ipc_update_perm(&semid64.sem_perm, ipcp); 925 sma->sem_ctime = get_seconds(); 926 break; 927 default: 928 err = -EINVAL; 929 } 930 931 out_unlock: 932 sem_unlock(sma); 933 out_up: 934 up_write(&sem_ids(ns).rw_mutex); 935 return err; 936 } 937 938 SYSCALL_DEFINE(semctl)(int semid, int semnum, int cmd, union semun arg) 939 { 940 int err = -EINVAL; 941 int version; 942 struct ipc_namespace *ns; 943 944 if (semid < 0) 945 return -EINVAL; 946 947 version = ipc_parse_version(&cmd); 948 ns = current->nsproxy->ipc_ns; 949 950 switch(cmd) { 951 case IPC_INFO: 952 case SEM_INFO: 953 case IPC_STAT: 954 case SEM_STAT: 955 err = semctl_nolock(ns, semid, cmd, version, arg); 956 return err; 957 case GETALL: 958 case GETVAL: 959 case GETPID: 960 case GETNCNT: 961 case GETZCNT: 962 case SETVAL: 963 case SETALL: 964 err = semctl_main(ns,semid,semnum,cmd,version,arg); 965 return err; 966 case IPC_RMID: 967 case IPC_SET: 968 err = semctl_down(ns, semid, cmd, version, arg); 969 return err; 970 default: 971 return -EINVAL; 972 } 973 } 974 #ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 975 asmlinkage long SyS_semctl(int semid, int semnum, int cmd, union semun arg) 976 { 977 return SYSC_semctl((int) semid, (int) semnum, (int) cmd, arg); 978 } 979 SYSCALL_ALIAS(sys_semctl, SyS_semctl); 980 #endif 981 982 /* If the task doesn't already have a undo_list, then allocate one 983 * here. We guarantee there is only one thread using this undo list, 984 * and current is THE ONE 985 * 986 * If this allocation and assignment succeeds, but later 987 * portions of this code fail, there is no need to free the sem_undo_list. 988 * Just let it stay associated with the task, and it'll be freed later 989 * at exit time. 990 * 991 * This can block, so callers must hold no locks. 992 */ 993 static inline int get_undo_list(struct sem_undo_list **undo_listp) 994 { 995 struct sem_undo_list *undo_list; 996 997 undo_list = current->sysvsem.undo_list; 998 if (!undo_list) { 999 undo_list = kzalloc(sizeof(*undo_list), GFP_KERNEL); 1000 if (undo_list == NULL) 1001 return -ENOMEM; 1002 spin_lock_init(&undo_list->lock); 1003 atomic_set(&undo_list->refcnt, 1); 1004 INIT_LIST_HEAD(&undo_list->list_proc); 1005 1006 current->sysvsem.undo_list = undo_list; 1007 } 1008 *undo_listp = undo_list; 1009 return 0; 1010 } 1011 1012 static struct sem_undo *__lookup_undo(struct sem_undo_list *ulp, int semid) 1013 { 1014 struct sem_undo *un; 1015 1016 list_for_each_entry_rcu(un, &ulp->list_proc, list_proc) { 1017 if (un->semid == semid) 1018 return un; 1019 } 1020 return NULL; 1021 } 1022 1023 static struct sem_undo *lookup_undo(struct sem_undo_list *ulp, int semid) 1024 { 1025 struct sem_undo *un; 1026 1027 assert_spin_locked(&ulp->lock); 1028 1029 un = __lookup_undo(ulp, semid); 1030 if (un) { 1031 list_del_rcu(&un->list_proc); 1032 list_add_rcu(&un->list_proc, &ulp->list_proc); 1033 } 1034 return un; 1035 } 1036 1037 /** 1038 * find_alloc_undo - Lookup (and if not present create) undo array 1039 * @ns: namespace 1040 * @semid: semaphore array id 1041 * 1042 * The function looks up (and if not present creates) the undo structure. 1043 * The size of the undo structure depends on the size of the semaphore 1044 * array, thus the alloc path is not that straightforward. 1045 * Lifetime-rules: sem_undo is rcu-protected, on success, the function 1046 * performs a rcu_read_lock(). 1047 */ 1048 static struct sem_undo *find_alloc_undo(struct ipc_namespace *ns, int semid) 1049 { 1050 struct sem_array *sma; 1051 struct sem_undo_list *ulp; 1052 struct sem_undo *un, *new; 1053 int nsems; 1054 int error; 1055 1056 error = get_undo_list(&ulp); 1057 if (error) 1058 return ERR_PTR(error); 1059 1060 rcu_read_lock(); 1061 spin_lock(&ulp->lock); 1062 un = lookup_undo(ulp, semid); 1063 spin_unlock(&ulp->lock); 1064 if (likely(un!=NULL)) 1065 goto out; 1066 rcu_read_unlock(); 1067 1068 /* no undo structure around - allocate one. */ 1069 /* step 1: figure out the size of the semaphore array */ 1070 sma = sem_lock_check(ns, semid); 1071 if (IS_ERR(sma)) 1072 return ERR_PTR(PTR_ERR(sma)); 1073 1074 nsems = sma->sem_nsems; 1075 sem_getref_and_unlock(sma); 1076 1077 /* step 2: allocate new undo structure */ 1078 new = kzalloc(sizeof(struct sem_undo) + sizeof(short)*nsems, GFP_KERNEL); 1079 if (!new) { 1080 sem_putref(sma); 1081 return ERR_PTR(-ENOMEM); 1082 } 1083 1084 /* step 3: Acquire the lock on semaphore array */ 1085 sem_lock_and_putref(sma); 1086 if (sma->sem_perm.deleted) { 1087 sem_unlock(sma); 1088 kfree(new); 1089 un = ERR_PTR(-EIDRM); 1090 goto out; 1091 } 1092 spin_lock(&ulp->lock); 1093 1094 /* 1095 * step 4: check for races: did someone else allocate the undo struct? 1096 */ 1097 un = lookup_undo(ulp, semid); 1098 if (un) { 1099 kfree(new); 1100 goto success; 1101 } 1102 /* step 5: initialize & link new undo structure */ 1103 new->semadj = (short *) &new[1]; 1104 new->ulp = ulp; 1105 new->semid = semid; 1106 assert_spin_locked(&ulp->lock); 1107 list_add_rcu(&new->list_proc, &ulp->list_proc); 1108 assert_spin_locked(&sma->sem_perm.lock); 1109 list_add(&new->list_id, &sma->list_id); 1110 un = new; 1111 1112 success: 1113 spin_unlock(&ulp->lock); 1114 rcu_read_lock(); 1115 sem_unlock(sma); 1116 out: 1117 return un; 1118 } 1119 1120 SYSCALL_DEFINE4(semtimedop, int, semid, struct sembuf __user *, tsops, 1121 unsigned, nsops, const struct timespec __user *, timeout) 1122 { 1123 int error = -EINVAL; 1124 struct sem_array *sma; 1125 struct sembuf fast_sops[SEMOPM_FAST]; 1126 struct sembuf* sops = fast_sops, *sop; 1127 struct sem_undo *un; 1128 int undos = 0, alter = 0, max; 1129 struct sem_queue queue; 1130 unsigned long jiffies_left = 0; 1131 struct ipc_namespace *ns; 1132 1133 ns = current->nsproxy->ipc_ns; 1134 1135 if (nsops < 1 || semid < 0) 1136 return -EINVAL; 1137 if (nsops > ns->sc_semopm) 1138 return -E2BIG; 1139 if(nsops > SEMOPM_FAST) { 1140 sops = kmalloc(sizeof(*sops)*nsops,GFP_KERNEL); 1141 if(sops==NULL) 1142 return -ENOMEM; 1143 } 1144 if (copy_from_user (sops, tsops, nsops * sizeof(*tsops))) { 1145 error=-EFAULT; 1146 goto out_free; 1147 } 1148 if (timeout) { 1149 struct timespec _timeout; 1150 if (copy_from_user(&_timeout, timeout, sizeof(*timeout))) { 1151 error = -EFAULT; 1152 goto out_free; 1153 } 1154 if (_timeout.tv_sec < 0 || _timeout.tv_nsec < 0 || 1155 _timeout.tv_nsec >= 1000000000L) { 1156 error = -EINVAL; 1157 goto out_free; 1158 } 1159 jiffies_left = timespec_to_jiffies(&_timeout); 1160 } 1161 max = 0; 1162 for (sop = sops; sop < sops + nsops; sop++) { 1163 if (sop->sem_num >= max) 1164 max = sop->sem_num; 1165 if (sop->sem_flg & SEM_UNDO) 1166 undos = 1; 1167 if (sop->sem_op != 0) 1168 alter = 1; 1169 } 1170 1171 if (undos) { 1172 un = find_alloc_undo(ns, semid); 1173 if (IS_ERR(un)) { 1174 error = PTR_ERR(un); 1175 goto out_free; 1176 } 1177 } else 1178 un = NULL; 1179 1180 sma = sem_lock_check(ns, semid); 1181 if (IS_ERR(sma)) { 1182 if (un) 1183 rcu_read_unlock(); 1184 error = PTR_ERR(sma); 1185 goto out_free; 1186 } 1187 1188 /* 1189 * semid identifiers are not unique - find_alloc_undo may have 1190 * allocated an undo structure, it was invalidated by an RMID 1191 * and now a new array with received the same id. Check and fail. 1192 * This case can be detected checking un->semid. The existance of 1193 * "un" itself is guaranteed by rcu. 1194 */ 1195 error = -EIDRM; 1196 if (un) { 1197 if (un->semid == -1) { 1198 rcu_read_unlock(); 1199 goto out_unlock_free; 1200 } else { 1201 /* 1202 * rcu lock can be released, "un" cannot disappear: 1203 * - sem_lock is acquired, thus IPC_RMID is 1204 * impossible. 1205 * - exit_sem is impossible, it always operates on 1206 * current (or a dead task). 1207 */ 1208 1209 rcu_read_unlock(); 1210 } 1211 } 1212 1213 error = -EFBIG; 1214 if (max >= sma->sem_nsems) 1215 goto out_unlock_free; 1216 1217 error = -EACCES; 1218 if (ipcperms(&sma->sem_perm, alter ? S_IWUGO : S_IRUGO)) 1219 goto out_unlock_free; 1220 1221 error = security_sem_semop(sma, sops, nsops, alter); 1222 if (error) 1223 goto out_unlock_free; 1224 1225 error = try_atomic_semop (sma, sops, nsops, un, task_tgid_vnr(current)); 1226 if (error <= 0) { 1227 if (alter && error == 0) 1228 update_queue(sma, (nsops == 1) ? sops[0].sem_num : -1); 1229 1230 goto out_unlock_free; 1231 } 1232 1233 /* We need to sleep on this operation, so we put the current 1234 * task into the pending queue and go to sleep. 1235 */ 1236 1237 queue.sops = sops; 1238 queue.nsops = nsops; 1239 queue.undo = un; 1240 queue.pid = task_tgid_vnr(current); 1241 queue.alter = alter; 1242 if (alter) 1243 list_add_tail(&queue.list, &sma->sem_pending); 1244 else 1245 list_add(&queue.list, &sma->sem_pending); 1246 1247 if (nsops == 1) { 1248 struct sem *curr; 1249 curr = &sma->sem_base[sops->sem_num]; 1250 1251 if (alter) 1252 list_add_tail(&queue.simple_list, &curr->sem_pending); 1253 else 1254 list_add(&queue.simple_list, &curr->sem_pending); 1255 } else { 1256 INIT_LIST_HEAD(&queue.simple_list); 1257 sma->complex_count++; 1258 } 1259 1260 queue.status = -EINTR; 1261 queue.sleeper = current; 1262 current->state = TASK_INTERRUPTIBLE; 1263 sem_unlock(sma); 1264 1265 if (timeout) 1266 jiffies_left = schedule_timeout(jiffies_left); 1267 else 1268 schedule(); 1269 1270 error = queue.status; 1271 while(unlikely(error == IN_WAKEUP)) { 1272 cpu_relax(); 1273 error = queue.status; 1274 } 1275 1276 if (error != -EINTR) { 1277 /* fast path: update_queue already obtained all requested 1278 * resources */ 1279 goto out_free; 1280 } 1281 1282 sma = sem_lock(ns, semid); 1283 if (IS_ERR(sma)) { 1284 error = -EIDRM; 1285 goto out_free; 1286 } 1287 1288 /* 1289 * If queue.status != -EINTR we are woken up by another process 1290 */ 1291 error = queue.status; 1292 if (error != -EINTR) { 1293 goto out_unlock_free; 1294 } 1295 1296 /* 1297 * If an interrupt occurred we have to clean up the queue 1298 */ 1299 if (timeout && jiffies_left == 0) 1300 error = -EAGAIN; 1301 unlink_queue(sma, &queue); 1302 1303 out_unlock_free: 1304 sem_unlock(sma); 1305 out_free: 1306 if(sops != fast_sops) 1307 kfree(sops); 1308 return error; 1309 } 1310 1311 SYSCALL_DEFINE3(semop, int, semid, struct sembuf __user *, tsops, 1312 unsigned, nsops) 1313 { 1314 return sys_semtimedop(semid, tsops, nsops, NULL); 1315 } 1316 1317 /* If CLONE_SYSVSEM is set, establish sharing of SEM_UNDO state between 1318 * parent and child tasks. 1319 */ 1320 1321 int copy_semundo(unsigned long clone_flags, struct task_struct *tsk) 1322 { 1323 struct sem_undo_list *undo_list; 1324 int error; 1325 1326 if (clone_flags & CLONE_SYSVSEM) { 1327 error = get_undo_list(&undo_list); 1328 if (error) 1329 return error; 1330 atomic_inc(&undo_list->refcnt); 1331 tsk->sysvsem.undo_list = undo_list; 1332 } else 1333 tsk->sysvsem.undo_list = NULL; 1334 1335 return 0; 1336 } 1337 1338 /* 1339 * add semadj values to semaphores, free undo structures. 1340 * undo structures are not freed when semaphore arrays are destroyed 1341 * so some of them may be out of date. 1342 * IMPLEMENTATION NOTE: There is some confusion over whether the 1343 * set of adjustments that needs to be done should be done in an atomic 1344 * manner or not. That is, if we are attempting to decrement the semval 1345 * should we queue up and wait until we can do so legally? 1346 * The original implementation attempted to do this (queue and wait). 1347 * The current implementation does not do so. The POSIX standard 1348 * and SVID should be consulted to determine what behavior is mandated. 1349 */ 1350 void exit_sem(struct task_struct *tsk) 1351 { 1352 struct sem_undo_list *ulp; 1353 1354 ulp = tsk->sysvsem.undo_list; 1355 if (!ulp) 1356 return; 1357 tsk->sysvsem.undo_list = NULL; 1358 1359 if (!atomic_dec_and_test(&ulp->refcnt)) 1360 return; 1361 1362 for (;;) { 1363 struct sem_array *sma; 1364 struct sem_undo *un; 1365 int semid; 1366 int i; 1367 1368 rcu_read_lock(); 1369 un = list_entry_rcu(ulp->list_proc.next, 1370 struct sem_undo, list_proc); 1371 if (&un->list_proc == &ulp->list_proc) 1372 semid = -1; 1373 else 1374 semid = un->semid; 1375 rcu_read_unlock(); 1376 1377 if (semid == -1) 1378 break; 1379 1380 sma = sem_lock_check(tsk->nsproxy->ipc_ns, un->semid); 1381 1382 /* exit_sem raced with IPC_RMID, nothing to do */ 1383 if (IS_ERR(sma)) 1384 continue; 1385 1386 un = __lookup_undo(ulp, semid); 1387 if (un == NULL) { 1388 /* exit_sem raced with IPC_RMID+semget() that created 1389 * exactly the same semid. Nothing to do. 1390 */ 1391 sem_unlock(sma); 1392 continue; 1393 } 1394 1395 /* remove un from the linked lists */ 1396 assert_spin_locked(&sma->sem_perm.lock); 1397 list_del(&un->list_id); 1398 1399 spin_lock(&ulp->lock); 1400 list_del_rcu(&un->list_proc); 1401 spin_unlock(&ulp->lock); 1402 1403 /* perform adjustments registered in un */ 1404 for (i = 0; i < sma->sem_nsems; i++) { 1405 struct sem * semaphore = &sma->sem_base[i]; 1406 if (un->semadj[i]) { 1407 semaphore->semval += un->semadj[i]; 1408 /* 1409 * Range checks of the new semaphore value, 1410 * not defined by sus: 1411 * - Some unices ignore the undo entirely 1412 * (e.g. HP UX 11i 11.22, Tru64 V5.1) 1413 * - some cap the value (e.g. FreeBSD caps 1414 * at 0, but doesn't enforce SEMVMX) 1415 * 1416 * Linux caps the semaphore value, both at 0 1417 * and at SEMVMX. 1418 * 1419 * Manfred <manfred@colorfullife.com> 1420 */ 1421 if (semaphore->semval < 0) 1422 semaphore->semval = 0; 1423 if (semaphore->semval > SEMVMX) 1424 semaphore->semval = SEMVMX; 1425 semaphore->sempid = task_tgid_vnr(current); 1426 } 1427 } 1428 sma->sem_otime = get_seconds(); 1429 /* maybe some queued-up processes were waiting for this */ 1430 update_queue(sma, -1); 1431 sem_unlock(sma); 1432 1433 call_rcu(&un->rcu, free_un); 1434 } 1435 kfree(ulp); 1436 } 1437 1438 #ifdef CONFIG_PROC_FS 1439 static int sysvipc_sem_proc_show(struct seq_file *s, void *it) 1440 { 1441 struct sem_array *sma = it; 1442 1443 return seq_printf(s, 1444 "%10d %10d %4o %10u %5u %5u %5u %5u %10lu %10lu\n", 1445 sma->sem_perm.key, 1446 sma->sem_perm.id, 1447 sma->sem_perm.mode, 1448 sma->sem_nsems, 1449 sma->sem_perm.uid, 1450 sma->sem_perm.gid, 1451 sma->sem_perm.cuid, 1452 sma->sem_perm.cgid, 1453 sma->sem_otime, 1454 sma->sem_ctime); 1455 } 1456 #endif 1457