1 /* 2 * linux/kernel/posix-timers.c 3 * 4 * 5 * 2002-10-15 Posix Clocks & timers 6 * by George Anzinger george@mvista.com 7 * 8 * Copyright (C) 2002 2003 by MontaVista Software. 9 * 10 * 2004-06-01 Fix CLOCK_REALTIME clock/timer TIMER_ABSTIME bug. 11 * Copyright (C) 2004 Boris Hu 12 * 13 * This program is free software; you can redistribute it and/or modify 14 * it under the terms of the GNU General Public License as published by 15 * the Free Software Foundation; either version 2 of the License, or (at 16 * your option) any later version. 17 * 18 * This program is distributed in the hope that it will be useful, but 19 * WITHOUT ANY WARRANTY; without even the implied warranty of 20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 21 * General Public License for more details. 22 23 * You should have received a copy of the GNU General Public License 24 * along with this program; if not, write to the Free Software 25 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA. 26 * 27 * MontaVista Software | 1237 East Arques Avenue | Sunnyvale | CA 94085 | USA 28 */ 29 30 /* These are all the functions necessary to implement 31 * POSIX clocks & timers 32 */ 33 #include <linux/mm.h> 34 #include <linux/interrupt.h> 35 #include <linux/slab.h> 36 #include <linux/time.h> 37 #include <linux/mutex.h> 38 39 #include <asm/uaccess.h> 40 #include <linux/list.h> 41 #include <linux/init.h> 42 #include <linux/compiler.h> 43 #include <linux/hash.h> 44 #include <linux/posix-clock.h> 45 #include <linux/posix-timers.h> 46 #include <linux/syscalls.h> 47 #include <linux/wait.h> 48 #include <linux/workqueue.h> 49 #include <linux/export.h> 50 #include <linux/hashtable.h> 51 52 #include "timekeeping.h" 53 54 /* 55 * Management arrays for POSIX timers. Timers are now kept in static hash table 56 * with 512 entries. 57 * Timer ids are allocated by local routine, which selects proper hash head by 58 * key, constructed from current->signal address and per signal struct counter. 59 * This keeps timer ids unique per process, but now they can intersect between 60 * processes. 61 */ 62 63 /* 64 * Lets keep our timers in a slab cache :-) 65 */ 66 static struct kmem_cache *posix_timers_cache; 67 68 static DEFINE_HASHTABLE(posix_timers_hashtable, 9); 69 static DEFINE_SPINLOCK(hash_lock); 70 71 /* 72 * we assume that the new SIGEV_THREAD_ID shares no bits with the other 73 * SIGEV values. Here we put out an error if this assumption fails. 74 */ 75 #if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \ 76 ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD)) 77 #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!" 78 #endif 79 80 /* 81 * parisc wants ENOTSUP instead of EOPNOTSUPP 82 */ 83 #ifndef ENOTSUP 84 # define ENANOSLEEP_NOTSUP EOPNOTSUPP 85 #else 86 # define ENANOSLEEP_NOTSUP ENOTSUP 87 #endif 88 89 /* 90 * The timer ID is turned into a timer address by idr_find(). 91 * Verifying a valid ID consists of: 92 * 93 * a) checking that idr_find() returns other than -1. 94 * b) checking that the timer id matches the one in the timer itself. 95 * c) that the timer owner is in the callers thread group. 96 */ 97 98 /* 99 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us 100 * to implement others. This structure defines the various 101 * clocks. 102 * 103 * RESOLUTION: Clock resolution is used to round up timer and interval 104 * times, NOT to report clock times, which are reported with as 105 * much resolution as the system can muster. In some cases this 106 * resolution may depend on the underlying clock hardware and 107 * may not be quantifiable until run time, and only then is the 108 * necessary code is written. The standard says we should say 109 * something about this issue in the documentation... 110 * 111 * FUNCTIONS: The CLOCKs structure defines possible functions to 112 * handle various clock functions. 113 * 114 * The standard POSIX timer management code assumes the 115 * following: 1.) The k_itimer struct (sched.h) is used for 116 * the timer. 2.) The list, it_lock, it_clock, it_id and 117 * it_pid fields are not modified by timer code. 118 * 119 * Permissions: It is assumed that the clock_settime() function defined 120 * for each clock will take care of permission checks. Some 121 * clocks may be set able by any user (i.e. local process 122 * clocks) others not. Currently the only set able clock we 123 * have is CLOCK_REALTIME and its high res counter part, both of 124 * which we beg off on and pass to do_sys_settimeofday(). 125 */ 126 127 static struct k_clock posix_clocks[MAX_CLOCKS]; 128 129 /* 130 * These ones are defined below. 131 */ 132 static int common_nsleep(const clockid_t, int flags, struct timespec *t, 133 struct timespec __user *rmtp); 134 static int common_timer_create(struct k_itimer *new_timer); 135 static void common_timer_get(struct k_itimer *, struct itimerspec *); 136 static int common_timer_set(struct k_itimer *, int, 137 struct itimerspec *, struct itimerspec *); 138 static int common_timer_del(struct k_itimer *timer); 139 140 static enum hrtimer_restart posix_timer_fn(struct hrtimer *data); 141 142 static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags); 143 144 #define lock_timer(tid, flags) \ 145 ({ struct k_itimer *__timr; \ 146 __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags)); \ 147 __timr; \ 148 }) 149 150 static int hash(struct signal_struct *sig, unsigned int nr) 151 { 152 return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable)); 153 } 154 155 static struct k_itimer *__posix_timers_find(struct hlist_head *head, 156 struct signal_struct *sig, 157 timer_t id) 158 { 159 struct k_itimer *timer; 160 161 hlist_for_each_entry_rcu(timer, head, t_hash) { 162 if ((timer->it_signal == sig) && (timer->it_id == id)) 163 return timer; 164 } 165 return NULL; 166 } 167 168 static struct k_itimer *posix_timer_by_id(timer_t id) 169 { 170 struct signal_struct *sig = current->signal; 171 struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)]; 172 173 return __posix_timers_find(head, sig, id); 174 } 175 176 static int posix_timer_add(struct k_itimer *timer) 177 { 178 struct signal_struct *sig = current->signal; 179 int first_free_id = sig->posix_timer_id; 180 struct hlist_head *head; 181 int ret = -ENOENT; 182 183 do { 184 spin_lock(&hash_lock); 185 head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)]; 186 if (!__posix_timers_find(head, sig, sig->posix_timer_id)) { 187 hlist_add_head_rcu(&timer->t_hash, head); 188 ret = sig->posix_timer_id; 189 } 190 if (++sig->posix_timer_id < 0) 191 sig->posix_timer_id = 0; 192 if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT)) 193 /* Loop over all possible ids completed */ 194 ret = -EAGAIN; 195 spin_unlock(&hash_lock); 196 } while (ret == -ENOENT); 197 return ret; 198 } 199 200 static inline void unlock_timer(struct k_itimer *timr, unsigned long flags) 201 { 202 spin_unlock_irqrestore(&timr->it_lock, flags); 203 } 204 205 /* Get clock_realtime */ 206 static int posix_clock_realtime_get(clockid_t which_clock, struct timespec *tp) 207 { 208 ktime_get_real_ts(tp); 209 return 0; 210 } 211 212 /* Set clock_realtime */ 213 static int posix_clock_realtime_set(const clockid_t which_clock, 214 const struct timespec *tp) 215 { 216 return do_sys_settimeofday(tp, NULL); 217 } 218 219 static int posix_clock_realtime_adj(const clockid_t which_clock, 220 struct timex *t) 221 { 222 return do_adjtimex(t); 223 } 224 225 /* 226 * Get monotonic time for posix timers 227 */ 228 static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp) 229 { 230 ktime_get_ts(tp); 231 return 0; 232 } 233 234 /* 235 * Get monotonic-raw time for posix timers 236 */ 237 static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp) 238 { 239 getrawmonotonic(tp); 240 return 0; 241 } 242 243 244 static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp) 245 { 246 *tp = current_kernel_time(); 247 return 0; 248 } 249 250 static int posix_get_monotonic_coarse(clockid_t which_clock, 251 struct timespec *tp) 252 { 253 *tp = get_monotonic_coarse(); 254 return 0; 255 } 256 257 static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp) 258 { 259 *tp = ktime_to_timespec(KTIME_LOW_RES); 260 return 0; 261 } 262 263 static int posix_get_boottime(const clockid_t which_clock, struct timespec *tp) 264 { 265 get_monotonic_boottime(tp); 266 return 0; 267 } 268 269 static int posix_get_tai(clockid_t which_clock, struct timespec *tp) 270 { 271 timekeeping_clocktai(tp); 272 return 0; 273 } 274 275 static int posix_get_hrtimer_res(clockid_t which_clock, struct timespec *tp) 276 { 277 tp->tv_sec = 0; 278 tp->tv_nsec = hrtimer_resolution; 279 return 0; 280 } 281 282 /* 283 * Initialize everything, well, just everything in Posix clocks/timers ;) 284 */ 285 static __init int init_posix_timers(void) 286 { 287 struct k_clock clock_realtime = { 288 .clock_getres = posix_get_hrtimer_res, 289 .clock_get = posix_clock_realtime_get, 290 .clock_set = posix_clock_realtime_set, 291 .clock_adj = posix_clock_realtime_adj, 292 .nsleep = common_nsleep, 293 .nsleep_restart = hrtimer_nanosleep_restart, 294 .timer_create = common_timer_create, 295 .timer_set = common_timer_set, 296 .timer_get = common_timer_get, 297 .timer_del = common_timer_del, 298 }; 299 struct k_clock clock_monotonic = { 300 .clock_getres = posix_get_hrtimer_res, 301 .clock_get = posix_ktime_get_ts, 302 .nsleep = common_nsleep, 303 .nsleep_restart = hrtimer_nanosleep_restart, 304 .timer_create = common_timer_create, 305 .timer_set = common_timer_set, 306 .timer_get = common_timer_get, 307 .timer_del = common_timer_del, 308 }; 309 struct k_clock clock_monotonic_raw = { 310 .clock_getres = posix_get_hrtimer_res, 311 .clock_get = posix_get_monotonic_raw, 312 }; 313 struct k_clock clock_realtime_coarse = { 314 .clock_getres = posix_get_coarse_res, 315 .clock_get = posix_get_realtime_coarse, 316 }; 317 struct k_clock clock_monotonic_coarse = { 318 .clock_getres = posix_get_coarse_res, 319 .clock_get = posix_get_monotonic_coarse, 320 }; 321 struct k_clock clock_tai = { 322 .clock_getres = posix_get_hrtimer_res, 323 .clock_get = posix_get_tai, 324 .nsleep = common_nsleep, 325 .nsleep_restart = hrtimer_nanosleep_restart, 326 .timer_create = common_timer_create, 327 .timer_set = common_timer_set, 328 .timer_get = common_timer_get, 329 .timer_del = common_timer_del, 330 }; 331 struct k_clock clock_boottime = { 332 .clock_getres = posix_get_hrtimer_res, 333 .clock_get = posix_get_boottime, 334 .nsleep = common_nsleep, 335 .nsleep_restart = hrtimer_nanosleep_restart, 336 .timer_create = common_timer_create, 337 .timer_set = common_timer_set, 338 .timer_get = common_timer_get, 339 .timer_del = common_timer_del, 340 }; 341 342 posix_timers_register_clock(CLOCK_REALTIME, &clock_realtime); 343 posix_timers_register_clock(CLOCK_MONOTONIC, &clock_monotonic); 344 posix_timers_register_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw); 345 posix_timers_register_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse); 346 posix_timers_register_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse); 347 posix_timers_register_clock(CLOCK_BOOTTIME, &clock_boottime); 348 posix_timers_register_clock(CLOCK_TAI, &clock_tai); 349 350 posix_timers_cache = kmem_cache_create("posix_timers_cache", 351 sizeof (struct k_itimer), 0, SLAB_PANIC, 352 NULL); 353 return 0; 354 } 355 356 __initcall(init_posix_timers); 357 358 static void schedule_next_timer(struct k_itimer *timr) 359 { 360 struct hrtimer *timer = &timr->it.real.timer; 361 362 if (timr->it.real.interval.tv64 == 0) 363 return; 364 365 timr->it_overrun += (unsigned int) hrtimer_forward(timer, 366 timer->base->get_time(), 367 timr->it.real.interval); 368 369 timr->it_overrun_last = timr->it_overrun; 370 timr->it_overrun = -1; 371 ++timr->it_requeue_pending; 372 hrtimer_restart(timer); 373 } 374 375 /* 376 * This function is exported for use by the signal deliver code. It is 377 * called just prior to the info block being released and passes that 378 * block to us. It's function is to update the overrun entry AND to 379 * restart the timer. It should only be called if the timer is to be 380 * restarted (i.e. we have flagged this in the sys_private entry of the 381 * info block). 382 * 383 * To protect against the timer going away while the interrupt is queued, 384 * we require that the it_requeue_pending flag be set. 385 */ 386 void do_schedule_next_timer(struct siginfo *info) 387 { 388 struct k_itimer *timr; 389 unsigned long flags; 390 391 timr = lock_timer(info->si_tid, &flags); 392 393 if (timr && timr->it_requeue_pending == info->si_sys_private) { 394 if (timr->it_clock < 0) 395 posix_cpu_timer_schedule(timr); 396 else 397 schedule_next_timer(timr); 398 399 info->si_overrun += timr->it_overrun_last; 400 } 401 402 if (timr) 403 unlock_timer(timr, flags); 404 } 405 406 int posix_timer_event(struct k_itimer *timr, int si_private) 407 { 408 struct task_struct *task; 409 int shared, ret = -1; 410 /* 411 * FIXME: if ->sigq is queued we can race with 412 * dequeue_signal()->do_schedule_next_timer(). 413 * 414 * If dequeue_signal() sees the "right" value of 415 * si_sys_private it calls do_schedule_next_timer(). 416 * We re-queue ->sigq and drop ->it_lock(). 417 * do_schedule_next_timer() locks the timer 418 * and re-schedules it while ->sigq is pending. 419 * Not really bad, but not that we want. 420 */ 421 timr->sigq->info.si_sys_private = si_private; 422 423 rcu_read_lock(); 424 task = pid_task(timr->it_pid, PIDTYPE_PID); 425 if (task) { 426 shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID); 427 ret = send_sigqueue(timr->sigq, task, shared); 428 } 429 rcu_read_unlock(); 430 /* If we failed to send the signal the timer stops. */ 431 return ret > 0; 432 } 433 EXPORT_SYMBOL_GPL(posix_timer_event); 434 435 /* 436 * This function gets called when a POSIX.1b interval timer expires. It 437 * is used as a callback from the kernel internal timer. The 438 * run_timer_list code ALWAYS calls with interrupts on. 439 440 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers. 441 */ 442 static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer) 443 { 444 struct k_itimer *timr; 445 unsigned long flags; 446 int si_private = 0; 447 enum hrtimer_restart ret = HRTIMER_NORESTART; 448 449 timr = container_of(timer, struct k_itimer, it.real.timer); 450 spin_lock_irqsave(&timr->it_lock, flags); 451 452 if (timr->it.real.interval.tv64 != 0) 453 si_private = ++timr->it_requeue_pending; 454 455 if (posix_timer_event(timr, si_private)) { 456 /* 457 * signal was not sent because of sig_ignor 458 * we will not get a call back to restart it AND 459 * it should be restarted. 460 */ 461 if (timr->it.real.interval.tv64 != 0) { 462 ktime_t now = hrtimer_cb_get_time(timer); 463 464 /* 465 * FIXME: What we really want, is to stop this 466 * timer completely and restart it in case the 467 * SIG_IGN is removed. This is a non trivial 468 * change which involves sighand locking 469 * (sigh !), which we don't want to do late in 470 * the release cycle. 471 * 472 * For now we just let timers with an interval 473 * less than a jiffie expire every jiffie to 474 * avoid softirq starvation in case of SIG_IGN 475 * and a very small interval, which would put 476 * the timer right back on the softirq pending 477 * list. By moving now ahead of time we trick 478 * hrtimer_forward() to expire the timer 479 * later, while we still maintain the overrun 480 * accuracy, but have some inconsistency in 481 * the timer_gettime() case. This is at least 482 * better than a starved softirq. A more 483 * complex fix which solves also another related 484 * inconsistency is already in the pipeline. 485 */ 486 #ifdef CONFIG_HIGH_RES_TIMERS 487 { 488 ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ); 489 490 if (timr->it.real.interval.tv64 < kj.tv64) 491 now = ktime_add(now, kj); 492 } 493 #endif 494 timr->it_overrun += (unsigned int) 495 hrtimer_forward(timer, now, 496 timr->it.real.interval); 497 ret = HRTIMER_RESTART; 498 ++timr->it_requeue_pending; 499 } 500 } 501 502 unlock_timer(timr, flags); 503 return ret; 504 } 505 506 static struct pid *good_sigevent(sigevent_t * event) 507 { 508 struct task_struct *rtn = current->group_leader; 509 510 if ((event->sigev_notify & SIGEV_THREAD_ID ) && 511 (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) || 512 !same_thread_group(rtn, current) || 513 (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL)) 514 return NULL; 515 516 if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) && 517 ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX))) 518 return NULL; 519 520 return task_pid(rtn); 521 } 522 523 void posix_timers_register_clock(const clockid_t clock_id, 524 struct k_clock *new_clock) 525 { 526 if ((unsigned) clock_id >= MAX_CLOCKS) { 527 printk(KERN_WARNING "POSIX clock register failed for clock_id %d\n", 528 clock_id); 529 return; 530 } 531 532 if (!new_clock->clock_get) { 533 printk(KERN_WARNING "POSIX clock id %d lacks clock_get()\n", 534 clock_id); 535 return; 536 } 537 if (!new_clock->clock_getres) { 538 printk(KERN_WARNING "POSIX clock id %d lacks clock_getres()\n", 539 clock_id); 540 return; 541 } 542 543 posix_clocks[clock_id] = *new_clock; 544 } 545 EXPORT_SYMBOL_GPL(posix_timers_register_clock); 546 547 static struct k_itimer * alloc_posix_timer(void) 548 { 549 struct k_itimer *tmr; 550 tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL); 551 if (!tmr) 552 return tmr; 553 if (unlikely(!(tmr->sigq = sigqueue_alloc()))) { 554 kmem_cache_free(posix_timers_cache, tmr); 555 return NULL; 556 } 557 memset(&tmr->sigq->info, 0, sizeof(siginfo_t)); 558 return tmr; 559 } 560 561 static void k_itimer_rcu_free(struct rcu_head *head) 562 { 563 struct k_itimer *tmr = container_of(head, struct k_itimer, it.rcu); 564 565 kmem_cache_free(posix_timers_cache, tmr); 566 } 567 568 #define IT_ID_SET 1 569 #define IT_ID_NOT_SET 0 570 static void release_posix_timer(struct k_itimer *tmr, int it_id_set) 571 { 572 if (it_id_set) { 573 unsigned long flags; 574 spin_lock_irqsave(&hash_lock, flags); 575 hlist_del_rcu(&tmr->t_hash); 576 spin_unlock_irqrestore(&hash_lock, flags); 577 } 578 put_pid(tmr->it_pid); 579 sigqueue_free(tmr->sigq); 580 call_rcu(&tmr->it.rcu, k_itimer_rcu_free); 581 } 582 583 static struct k_clock *clockid_to_kclock(const clockid_t id) 584 { 585 if (id < 0) 586 return (id & CLOCKFD_MASK) == CLOCKFD ? 587 &clock_posix_dynamic : &clock_posix_cpu; 588 589 if (id >= MAX_CLOCKS || !posix_clocks[id].clock_getres) 590 return NULL; 591 return &posix_clocks[id]; 592 } 593 594 static int common_timer_create(struct k_itimer *new_timer) 595 { 596 hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0); 597 return 0; 598 } 599 600 /* Create a POSIX.1b interval timer. */ 601 602 SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock, 603 struct sigevent __user *, timer_event_spec, 604 timer_t __user *, created_timer_id) 605 { 606 struct k_clock *kc = clockid_to_kclock(which_clock); 607 struct k_itimer *new_timer; 608 int error, new_timer_id; 609 sigevent_t event; 610 int it_id_set = IT_ID_NOT_SET; 611 612 if (!kc) 613 return -EINVAL; 614 if (!kc->timer_create) 615 return -EOPNOTSUPP; 616 617 new_timer = alloc_posix_timer(); 618 if (unlikely(!new_timer)) 619 return -EAGAIN; 620 621 spin_lock_init(&new_timer->it_lock); 622 new_timer_id = posix_timer_add(new_timer); 623 if (new_timer_id < 0) { 624 error = new_timer_id; 625 goto out; 626 } 627 628 it_id_set = IT_ID_SET; 629 new_timer->it_id = (timer_t) new_timer_id; 630 new_timer->it_clock = which_clock; 631 new_timer->it_overrun = -1; 632 633 if (timer_event_spec) { 634 if (copy_from_user(&event, timer_event_spec, sizeof (event))) { 635 error = -EFAULT; 636 goto out; 637 } 638 rcu_read_lock(); 639 new_timer->it_pid = get_pid(good_sigevent(&event)); 640 rcu_read_unlock(); 641 if (!new_timer->it_pid) { 642 error = -EINVAL; 643 goto out; 644 } 645 } else { 646 memset(&event.sigev_value, 0, sizeof(event.sigev_value)); 647 event.sigev_notify = SIGEV_SIGNAL; 648 event.sigev_signo = SIGALRM; 649 event.sigev_value.sival_int = new_timer->it_id; 650 new_timer->it_pid = get_pid(task_tgid(current)); 651 } 652 653 new_timer->it_sigev_notify = event.sigev_notify; 654 new_timer->sigq->info.si_signo = event.sigev_signo; 655 new_timer->sigq->info.si_value = event.sigev_value; 656 new_timer->sigq->info.si_tid = new_timer->it_id; 657 new_timer->sigq->info.si_code = SI_TIMER; 658 659 if (copy_to_user(created_timer_id, 660 &new_timer_id, sizeof (new_timer_id))) { 661 error = -EFAULT; 662 goto out; 663 } 664 665 error = kc->timer_create(new_timer); 666 if (error) 667 goto out; 668 669 spin_lock_irq(¤t->sighand->siglock); 670 new_timer->it_signal = current->signal; 671 list_add(&new_timer->list, ¤t->signal->posix_timers); 672 spin_unlock_irq(¤t->sighand->siglock); 673 674 return 0; 675 /* 676 * In the case of the timer belonging to another task, after 677 * the task is unlocked, the timer is owned by the other task 678 * and may cease to exist at any time. Don't use or modify 679 * new_timer after the unlock call. 680 */ 681 out: 682 release_posix_timer(new_timer, it_id_set); 683 return error; 684 } 685 686 /* 687 * Locking issues: We need to protect the result of the id look up until 688 * we get the timer locked down so it is not deleted under us. The 689 * removal is done under the idr spinlock so we use that here to bridge 690 * the find to the timer lock. To avoid a dead lock, the timer id MUST 691 * be release with out holding the timer lock. 692 */ 693 static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags) 694 { 695 struct k_itimer *timr; 696 697 /* 698 * timer_t could be any type >= int and we want to make sure any 699 * @timer_id outside positive int range fails lookup. 700 */ 701 if ((unsigned long long)timer_id > INT_MAX) 702 return NULL; 703 704 rcu_read_lock(); 705 timr = posix_timer_by_id(timer_id); 706 if (timr) { 707 spin_lock_irqsave(&timr->it_lock, *flags); 708 if (timr->it_signal == current->signal) { 709 rcu_read_unlock(); 710 return timr; 711 } 712 spin_unlock_irqrestore(&timr->it_lock, *flags); 713 } 714 rcu_read_unlock(); 715 716 return NULL; 717 } 718 719 /* 720 * Get the time remaining on a POSIX.1b interval timer. This function 721 * is ALWAYS called with spin_lock_irq on the timer, thus it must not 722 * mess with irq. 723 * 724 * We have a couple of messes to clean up here. First there is the case 725 * of a timer that has a requeue pending. These timers should appear to 726 * be in the timer list with an expiry as if we were to requeue them 727 * now. 728 * 729 * The second issue is the SIGEV_NONE timer which may be active but is 730 * not really ever put in the timer list (to save system resources). 731 * This timer may be expired, and if so, we will do it here. Otherwise 732 * it is the same as a requeue pending timer WRT to what we should 733 * report. 734 */ 735 static void 736 common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting) 737 { 738 ktime_t now, remaining, iv; 739 struct hrtimer *timer = &timr->it.real.timer; 740 741 memset(cur_setting, 0, sizeof(struct itimerspec)); 742 743 iv = timr->it.real.interval; 744 745 /* interval timer ? */ 746 if (iv.tv64) 747 cur_setting->it_interval = ktime_to_timespec(iv); 748 else if (!hrtimer_active(timer) && 749 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) 750 return; 751 752 now = timer->base->get_time(); 753 754 /* 755 * When a requeue is pending or this is a SIGEV_NONE 756 * timer move the expiry time forward by intervals, so 757 * expiry is > now. 758 */ 759 if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING || 760 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) 761 timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv); 762 763 remaining = __hrtimer_expires_remaining_adjusted(timer, now); 764 /* Return 0 only, when the timer is expired and not pending */ 765 if (remaining.tv64 <= 0) { 766 /* 767 * A single shot SIGEV_NONE timer must return 0, when 768 * it is expired ! 769 */ 770 if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) 771 cur_setting->it_value.tv_nsec = 1; 772 } else 773 cur_setting->it_value = ktime_to_timespec(remaining); 774 } 775 776 /* Get the time remaining on a POSIX.1b interval timer. */ 777 SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id, 778 struct itimerspec __user *, setting) 779 { 780 struct itimerspec cur_setting; 781 struct k_itimer *timr; 782 struct k_clock *kc; 783 unsigned long flags; 784 int ret = 0; 785 786 timr = lock_timer(timer_id, &flags); 787 if (!timr) 788 return -EINVAL; 789 790 kc = clockid_to_kclock(timr->it_clock); 791 if (WARN_ON_ONCE(!kc || !kc->timer_get)) 792 ret = -EINVAL; 793 else 794 kc->timer_get(timr, &cur_setting); 795 796 unlock_timer(timr, flags); 797 798 if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting))) 799 return -EFAULT; 800 801 return ret; 802 } 803 804 /* 805 * Get the number of overruns of a POSIX.1b interval timer. This is to 806 * be the overrun of the timer last delivered. At the same time we are 807 * accumulating overruns on the next timer. The overrun is frozen when 808 * the signal is delivered, either at the notify time (if the info block 809 * is not queued) or at the actual delivery time (as we are informed by 810 * the call back to do_schedule_next_timer(). So all we need to do is 811 * to pick up the frozen overrun. 812 */ 813 SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id) 814 { 815 struct k_itimer *timr; 816 int overrun; 817 unsigned long flags; 818 819 timr = lock_timer(timer_id, &flags); 820 if (!timr) 821 return -EINVAL; 822 823 overrun = timr->it_overrun_last; 824 unlock_timer(timr, flags); 825 826 return overrun; 827 } 828 829 /* Set a POSIX.1b interval timer. */ 830 /* timr->it_lock is taken. */ 831 static int 832 common_timer_set(struct k_itimer *timr, int flags, 833 struct itimerspec *new_setting, struct itimerspec *old_setting) 834 { 835 struct hrtimer *timer = &timr->it.real.timer; 836 enum hrtimer_mode mode; 837 838 if (old_setting) 839 common_timer_get(timr, old_setting); 840 841 /* disable the timer */ 842 timr->it.real.interval.tv64 = 0; 843 /* 844 * careful here. If smp we could be in the "fire" routine which will 845 * be spinning as we hold the lock. But this is ONLY an SMP issue. 846 */ 847 if (hrtimer_try_to_cancel(timer) < 0) 848 return TIMER_RETRY; 849 850 timr->it_requeue_pending = (timr->it_requeue_pending + 2) & 851 ~REQUEUE_PENDING; 852 timr->it_overrun_last = 0; 853 854 /* switch off the timer when it_value is zero */ 855 if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) 856 return 0; 857 858 mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL; 859 hrtimer_init(&timr->it.real.timer, timr->it_clock, mode); 860 timr->it.real.timer.function = posix_timer_fn; 861 862 hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value)); 863 864 /* Convert interval */ 865 timr->it.real.interval = timespec_to_ktime(new_setting->it_interval); 866 867 /* SIGEV_NONE timers are not queued ! See common_timer_get */ 868 if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) { 869 /* Setup correct expiry time for relative timers */ 870 if (mode == HRTIMER_MODE_REL) { 871 hrtimer_add_expires(timer, timer->base->get_time()); 872 } 873 return 0; 874 } 875 876 hrtimer_start_expires(timer, mode); 877 return 0; 878 } 879 880 /* Set a POSIX.1b interval timer */ 881 SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags, 882 const struct itimerspec __user *, new_setting, 883 struct itimerspec __user *, old_setting) 884 { 885 struct k_itimer *timr; 886 struct itimerspec new_spec, old_spec; 887 int error = 0; 888 unsigned long flag; 889 struct itimerspec *rtn = old_setting ? &old_spec : NULL; 890 struct k_clock *kc; 891 892 if (!new_setting) 893 return -EINVAL; 894 895 if (copy_from_user(&new_spec, new_setting, sizeof (new_spec))) 896 return -EFAULT; 897 898 if (!timespec_valid(&new_spec.it_interval) || 899 !timespec_valid(&new_spec.it_value)) 900 return -EINVAL; 901 retry: 902 timr = lock_timer(timer_id, &flag); 903 if (!timr) 904 return -EINVAL; 905 906 kc = clockid_to_kclock(timr->it_clock); 907 if (WARN_ON_ONCE(!kc || !kc->timer_set)) 908 error = -EINVAL; 909 else 910 error = kc->timer_set(timr, flags, &new_spec, rtn); 911 912 unlock_timer(timr, flag); 913 if (error == TIMER_RETRY) { 914 rtn = NULL; // We already got the old time... 915 goto retry; 916 } 917 918 if (old_setting && !error && 919 copy_to_user(old_setting, &old_spec, sizeof (old_spec))) 920 error = -EFAULT; 921 922 return error; 923 } 924 925 static int common_timer_del(struct k_itimer *timer) 926 { 927 timer->it.real.interval.tv64 = 0; 928 929 if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0) 930 return TIMER_RETRY; 931 return 0; 932 } 933 934 static inline int timer_delete_hook(struct k_itimer *timer) 935 { 936 struct k_clock *kc = clockid_to_kclock(timer->it_clock); 937 938 if (WARN_ON_ONCE(!kc || !kc->timer_del)) 939 return -EINVAL; 940 return kc->timer_del(timer); 941 } 942 943 /* Delete a POSIX.1b interval timer. */ 944 SYSCALL_DEFINE1(timer_delete, timer_t, timer_id) 945 { 946 struct k_itimer *timer; 947 unsigned long flags; 948 949 retry_delete: 950 timer = lock_timer(timer_id, &flags); 951 if (!timer) 952 return -EINVAL; 953 954 if (timer_delete_hook(timer) == TIMER_RETRY) { 955 unlock_timer(timer, flags); 956 goto retry_delete; 957 } 958 959 spin_lock(¤t->sighand->siglock); 960 list_del(&timer->list); 961 spin_unlock(¤t->sighand->siglock); 962 /* 963 * This keeps any tasks waiting on the spin lock from thinking 964 * they got something (see the lock code above). 965 */ 966 timer->it_signal = NULL; 967 968 unlock_timer(timer, flags); 969 release_posix_timer(timer, IT_ID_SET); 970 return 0; 971 } 972 973 /* 974 * return timer owned by the process, used by exit_itimers 975 */ 976 static void itimer_delete(struct k_itimer *timer) 977 { 978 unsigned long flags; 979 980 retry_delete: 981 spin_lock_irqsave(&timer->it_lock, flags); 982 983 if (timer_delete_hook(timer) == TIMER_RETRY) { 984 unlock_timer(timer, flags); 985 goto retry_delete; 986 } 987 list_del(&timer->list); 988 /* 989 * This keeps any tasks waiting on the spin lock from thinking 990 * they got something (see the lock code above). 991 */ 992 timer->it_signal = NULL; 993 994 unlock_timer(timer, flags); 995 release_posix_timer(timer, IT_ID_SET); 996 } 997 998 /* 999 * This is called by do_exit or de_thread, only when there are no more 1000 * references to the shared signal_struct. 1001 */ 1002 void exit_itimers(struct signal_struct *sig) 1003 { 1004 struct k_itimer *tmr; 1005 1006 while (!list_empty(&sig->posix_timers)) { 1007 tmr = list_entry(sig->posix_timers.next, struct k_itimer, list); 1008 itimer_delete(tmr); 1009 } 1010 } 1011 1012 SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock, 1013 const struct timespec __user *, tp) 1014 { 1015 struct k_clock *kc = clockid_to_kclock(which_clock); 1016 struct timespec new_tp; 1017 1018 if (!kc || !kc->clock_set) 1019 return -EINVAL; 1020 1021 if (copy_from_user(&new_tp, tp, sizeof (*tp))) 1022 return -EFAULT; 1023 1024 return kc->clock_set(which_clock, &new_tp); 1025 } 1026 1027 SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock, 1028 struct timespec __user *,tp) 1029 { 1030 struct k_clock *kc = clockid_to_kclock(which_clock); 1031 struct timespec kernel_tp; 1032 int error; 1033 1034 if (!kc) 1035 return -EINVAL; 1036 1037 error = kc->clock_get(which_clock, &kernel_tp); 1038 1039 if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp))) 1040 error = -EFAULT; 1041 1042 return error; 1043 } 1044 1045 SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock, 1046 struct timex __user *, utx) 1047 { 1048 struct k_clock *kc = clockid_to_kclock(which_clock); 1049 struct timex ktx; 1050 int err; 1051 1052 if (!kc) 1053 return -EINVAL; 1054 if (!kc->clock_adj) 1055 return -EOPNOTSUPP; 1056 1057 if (copy_from_user(&ktx, utx, sizeof(ktx))) 1058 return -EFAULT; 1059 1060 err = kc->clock_adj(which_clock, &ktx); 1061 1062 if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx))) 1063 return -EFAULT; 1064 1065 return err; 1066 } 1067 1068 SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock, 1069 struct timespec __user *, tp) 1070 { 1071 struct k_clock *kc = clockid_to_kclock(which_clock); 1072 struct timespec rtn_tp; 1073 int error; 1074 1075 if (!kc) 1076 return -EINVAL; 1077 1078 error = kc->clock_getres(which_clock, &rtn_tp); 1079 1080 if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) 1081 error = -EFAULT; 1082 1083 return error; 1084 } 1085 1086 /* 1087 * nanosleep for monotonic and realtime clocks 1088 */ 1089 static int common_nsleep(const clockid_t which_clock, int flags, 1090 struct timespec *tsave, struct timespec __user *rmtp) 1091 { 1092 return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ? 1093 HRTIMER_MODE_ABS : HRTIMER_MODE_REL, 1094 which_clock); 1095 } 1096 1097 SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags, 1098 const struct timespec __user *, rqtp, 1099 struct timespec __user *, rmtp) 1100 { 1101 struct k_clock *kc = clockid_to_kclock(which_clock); 1102 struct timespec t; 1103 1104 if (!kc) 1105 return -EINVAL; 1106 if (!kc->nsleep) 1107 return -ENANOSLEEP_NOTSUP; 1108 1109 if (copy_from_user(&t, rqtp, sizeof (struct timespec))) 1110 return -EFAULT; 1111 1112 if (!timespec_valid(&t)) 1113 return -EINVAL; 1114 1115 return kc->nsleep(which_clock, flags, &t, rmtp); 1116 } 1117 1118 /* 1119 * This will restart clock_nanosleep. This is required only by 1120 * compat_clock_nanosleep_restart for now. 1121 */ 1122 long clock_nanosleep_restart(struct restart_block *restart_block) 1123 { 1124 clockid_t which_clock = restart_block->nanosleep.clockid; 1125 struct k_clock *kc = clockid_to_kclock(which_clock); 1126 1127 if (WARN_ON_ONCE(!kc || !kc->nsleep_restart)) 1128 return -EINVAL; 1129 1130 return kc->nsleep_restart(restart_block); 1131 } 1132