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