linux/kernel/posix-timers.c
<<
>>
Prefs
   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/export.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 */
  72static struct kmem_cache *posix_timers_cache;
  73static struct idr posix_timers_id;
  74static 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
 132static struct k_clock posix_clocks[MAX_CLOCKS];
 133
 134/*
 135 * These ones are defined below.
 136 */
 137static int common_nsleep(const clockid_t, int flags, struct timespec *t,
 138                         struct timespec __user *rmtp);
 139static int common_timer_create(struct k_itimer *new_timer);
 140static void common_timer_get(struct k_itimer *, struct itimerspec *);
 141static int common_timer_set(struct k_itimer *, int,
 142                            struct itimerspec *, struct itimerspec *);
 143static int common_timer_del(struct k_itimer *timer);
 144
 145static enum hrtimer_restart posix_timer_fn(struct hrtimer *data);
 146
 147static 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
 155static 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 */
 161static 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 */
 168static 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
 174static 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 */
 183static 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 */
 192static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp)
 193{
 194        getrawmonotonic(tp);
 195        return 0;
 196}
 197
 198
 199static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp)
 200{
 201        *tp = current_kernel_time();
 202        return 0;
 203}
 204
 205static int posix_get_monotonic_coarse(clockid_t which_clock,
 206                                                struct timespec *tp)
 207{
 208        *tp = get_monotonic_coarse();
 209        return 0;
 210}
 211
 212static 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
 218static 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 */
 228static __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
 291static 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 */
 319void 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
 339int 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}
 366EXPORT_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 */
 375static 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
 439static 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
 456void 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}
 478EXPORT_SYMBOL_GPL(posix_timers_register_clock);
 479
 480static 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
 494static 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
 503static 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
 516static 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
 527static 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
 535SYSCALL_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
 556        idr_preload(GFP_KERNEL);
 557        spin_lock_irq(&idr_lock);
 558        error = idr_alloc(&posix_timers_id, new_timer, 0, 0, GFP_NOWAIT);
 559        spin_unlock_irq(&idr_lock);
 560        idr_preload_end();
 561        if (error < 0) {
 562                /*
 563                 * Weird looking, but we return EAGAIN if the IDR is
 564                 * full (proper POSIX return value for this)
 565                 */
 566                if (error == -ENOSPC)
 567                        error = -EAGAIN;
 568                goto out;
 569        }
 570        new_timer_id = error;
 571
 572        it_id_set = IT_ID_SET;
 573        new_timer->it_id = (timer_t) new_timer_id;
 574        new_timer->it_clock = which_clock;
 575        new_timer->it_overrun = -1;
 576
 577        if (timer_event_spec) {
 578                if (copy_from_user(&event, timer_event_spec, sizeof (event))) {
 579                        error = -EFAULT;
 580                        goto out;
 581                }
 582                rcu_read_lock();
 583                new_timer->it_pid = get_pid(good_sigevent(&event));
 584                rcu_read_unlock();
 585                if (!new_timer->it_pid) {
 586                        error = -EINVAL;
 587                        goto out;
 588                }
 589        } else {
 590                event.sigev_notify = SIGEV_SIGNAL;
 591                event.sigev_signo = SIGALRM;
 592                event.sigev_value.sival_int = new_timer->it_id;
 593                new_timer->it_pid = get_pid(task_tgid(current));
 594        }
 595
 596        new_timer->it_sigev_notify     = event.sigev_notify;
 597        new_timer->sigq->info.si_signo = event.sigev_signo;
 598        new_timer->sigq->info.si_value = event.sigev_value;
 599        new_timer->sigq->info.si_tid   = new_timer->it_id;
 600        new_timer->sigq->info.si_code  = SI_TIMER;
 601
 602        if (copy_to_user(created_timer_id,
 603                         &new_timer_id, sizeof (new_timer_id))) {
 604                error = -EFAULT;
 605                goto out;
 606        }
 607
 608        error = kc->timer_create(new_timer);
 609        if (error)
 610                goto out;
 611
 612        spin_lock_irq(&current->sighand->siglock);
 613        new_timer->it_signal = current->signal;
 614        list_add(&new_timer->list, &current->signal->posix_timers);
 615        spin_unlock_irq(&current->sighand->siglock);
 616
 617        return 0;
 618        /*
 619         * In the case of the timer belonging to another task, after
 620         * the task is unlocked, the timer is owned by the other task
 621         * and may cease to exist at any time.  Don't use or modify
 622         * new_timer after the unlock call.
 623         */
 624out:
 625        release_posix_timer(new_timer, it_id_set);
 626        return error;
 627}
 628
 629/*
 630 * Locking issues: We need to protect the result of the id look up until
 631 * we get the timer locked down so it is not deleted under us.  The
 632 * removal is done under the idr spinlock so we use that here to bridge
 633 * the find to the timer lock.  To avoid a dead lock, the timer id MUST
 634 * be release with out holding the timer lock.
 635 */
 636static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 637{
 638        struct k_itimer *timr;
 639
 640        /*
 641         * timer_t could be any type >= int and we want to make sure any
 642         * @timer_id outside positive int range fails lookup.
 643         */
 644        if ((unsigned long long)timer_id > INT_MAX)
 645                return NULL;
 646
 647        rcu_read_lock();
 648        timr = idr_find(&posix_timers_id, (int)timer_id);
 649        if (timr) {
 650                spin_lock_irqsave(&timr->it_lock, *flags);
 651                if (timr->it_signal == current->signal) {
 652                        rcu_read_unlock();
 653                        return timr;
 654                }
 655                spin_unlock_irqrestore(&timr->it_lock, *flags);
 656        }
 657        rcu_read_unlock();
 658
 659        return NULL;
 660}
 661
 662/*
 663 * Get the time remaining on a POSIX.1b interval timer.  This function
 664 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
 665 * mess with irq.
 666 *
 667 * We have a couple of messes to clean up here.  First there is the case
 668 * of a timer that has a requeue pending.  These timers should appear to
 669 * be in the timer list with an expiry as if we were to requeue them
 670 * now.
 671 *
 672 * The second issue is the SIGEV_NONE timer which may be active but is
 673 * not really ever put in the timer list (to save system resources).
 674 * This timer may be expired, and if so, we will do it here.  Otherwise
 675 * it is the same as a requeue pending timer WRT to what we should
 676 * report.
 677 */
 678static void
 679common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
 680{
 681        ktime_t now, remaining, iv;
 682        struct hrtimer *timer = &timr->it.real.timer;
 683
 684        memset(cur_setting, 0, sizeof(struct itimerspec));
 685
 686        iv = timr->it.real.interval;
 687
 688        /* interval timer ? */
 689        if (iv.tv64)
 690                cur_setting->it_interval = ktime_to_timespec(iv);
 691        else if (!hrtimer_active(timer) &&
 692                 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
 693                return;
 694
 695        now = timer->base->get_time();
 696
 697        /*
 698         * When a requeue is pending or this is a SIGEV_NONE
 699         * timer move the expiry time forward by intervals, so
 700         * expiry is > now.
 701         */
 702        if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING ||
 703            (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE))
 704                timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv);
 705
 706        remaining = ktime_sub(hrtimer_get_expires(timer), now);
 707        /* Return 0 only, when the timer is expired and not pending */
 708        if (remaining.tv64 <= 0) {
 709                /*
 710                 * A single shot SIGEV_NONE timer must return 0, when
 711                 * it is expired !
 712                 */
 713                if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
 714                        cur_setting->it_value.tv_nsec = 1;
 715        } else
 716                cur_setting->it_value = ktime_to_timespec(remaining);
 717}
 718
 719/* Get the time remaining on a POSIX.1b interval timer. */
 720SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 721                struct itimerspec __user *, setting)
 722{
 723        struct itimerspec cur_setting;
 724        struct k_itimer *timr;
 725        struct k_clock *kc;
 726        unsigned long flags;
 727        int ret = 0;
 728
 729        timr = lock_timer(timer_id, &flags);
 730        if (!timr)
 731                return -EINVAL;
 732
 733        kc = clockid_to_kclock(timr->it_clock);
 734        if (WARN_ON_ONCE(!kc || !kc->timer_get))
 735                ret = -EINVAL;
 736        else
 737                kc->timer_get(timr, &cur_setting);
 738
 739        unlock_timer(timr, flags);
 740
 741        if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
 742                return -EFAULT;
 743
 744        return ret;
 745}
 746
 747/*
 748 * Get the number of overruns of a POSIX.1b interval timer.  This is to
 749 * be the overrun of the timer last delivered.  At the same time we are
 750 * accumulating overruns on the next timer.  The overrun is frozen when
 751 * the signal is delivered, either at the notify time (if the info block
 752 * is not queued) or at the actual delivery time (as we are informed by
 753 * the call back to do_schedule_next_timer().  So all we need to do is
 754 * to pick up the frozen overrun.
 755 */
 756SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 757{
 758        struct k_itimer *timr;
 759        int overrun;
 760        unsigned long flags;
 761
 762        timr = lock_timer(timer_id, &flags);
 763        if (!timr)
 764                return -EINVAL;
 765
 766        overrun = timr->it_overrun_last;
 767        unlock_timer(timr, flags);
 768
 769        return overrun;
 770}
 771
 772/* Set a POSIX.1b interval timer. */
 773/* timr->it_lock is taken. */
 774static int
 775common_timer_set(struct k_itimer *timr, int flags,
 776                 struct itimerspec *new_setting, struct itimerspec *old_setting)
 777{
 778        struct hrtimer *timer = &timr->it.real.timer;
 779        enum hrtimer_mode mode;
 780
 781        if (old_setting)
 782                common_timer_get(timr, old_setting);
 783
 784        /* disable the timer */
 785        timr->it.real.interval.tv64 = 0;
 786        /*
 787         * careful here.  If smp we could be in the "fire" routine which will
 788         * be spinning as we hold the lock.  But this is ONLY an SMP issue.
 789         */
 790        if (hrtimer_try_to_cancel(timer) < 0)
 791                return TIMER_RETRY;
 792
 793        timr->it_requeue_pending = (timr->it_requeue_pending + 2) & 
 794                ~REQUEUE_PENDING;
 795        timr->it_overrun_last = 0;
 796
 797        /* switch off the timer when it_value is zero */
 798        if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 799                return 0;
 800
 801        mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 802        hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 803        timr->it.real.timer.function = posix_timer_fn;
 804
 805        hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value));
 806
 807        /* Convert interval */
 808        timr->it.real.interval = timespec_to_ktime(new_setting->it_interval);
 809
 810        /* SIGEV_NONE timers are not queued ! See common_timer_get */
 811        if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) {
 812                /* Setup correct expiry time for relative timers */
 813                if (mode == HRTIMER_MODE_REL) {
 814                        hrtimer_add_expires(timer, timer->base->get_time());
 815                }
 816                return 0;
 817        }
 818
 819        hrtimer_start_expires(timer, mode);
 820        return 0;
 821}
 822
 823/* Set a POSIX.1b interval timer */
 824SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 825                const struct itimerspec __user *, new_setting,
 826                struct itimerspec __user *, old_setting)
 827{
 828        struct k_itimer *timr;
 829        struct itimerspec new_spec, old_spec;
 830        int error = 0;
 831        unsigned long flag;
 832        struct itimerspec *rtn = old_setting ? &old_spec : NULL;
 833        struct k_clock *kc;
 834
 835        if (!new_setting)
 836                return -EINVAL;
 837
 838        if (copy_from_user(&new_spec, new_setting, sizeof (new_spec)))
 839                return -EFAULT;
 840
 841        if (!timespec_valid(&new_spec.it_interval) ||
 842            !timespec_valid(&new_spec.it_value))
 843                return -EINVAL;
 844retry:
 845        timr = lock_timer(timer_id, &flag);
 846        if (!timr)
 847                return -EINVAL;
 848
 849        kc = clockid_to_kclock(timr->it_clock);
 850        if (WARN_ON_ONCE(!kc || !kc->timer_set))
 851                error = -EINVAL;
 852        else
 853                error = kc->timer_set(timr, flags, &new_spec, rtn);
 854
 855        unlock_timer(timr, flag);
 856        if (error == TIMER_RETRY) {
 857                rtn = NULL;     // We already got the old time...
 858                goto retry;
 859        }
 860
 861        if (old_setting && !error &&
 862            copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
 863                error = -EFAULT;
 864
 865        return error;
 866}
 867
 868static int common_timer_del(struct k_itimer *timer)
 869{
 870        timer->it.real.interval.tv64 = 0;
 871
 872        if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0)
 873                return TIMER_RETRY;
 874        return 0;
 875}
 876
 877static inline int timer_delete_hook(struct k_itimer *timer)
 878{
 879        struct k_clock *kc = clockid_to_kclock(timer->it_clock);
 880
 881        if (WARN_ON_ONCE(!kc || !kc->timer_del))
 882                return -EINVAL;
 883        return kc->timer_del(timer);
 884}
 885
 886/* Delete a POSIX.1b interval timer. */
 887SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 888{
 889        struct k_itimer *timer;
 890        unsigned long flags;
 891
 892retry_delete:
 893        timer = lock_timer(timer_id, &flags);
 894        if (!timer)
 895                return -EINVAL;
 896
 897        if (timer_delete_hook(timer) == TIMER_RETRY) {
 898                unlock_timer(timer, flags);
 899                goto retry_delete;
 900        }
 901
 902        spin_lock(&current->sighand->siglock);
 903        list_del(&timer->list);
 904        spin_unlock(&current->sighand->siglock);
 905        /*
 906         * This keeps any tasks waiting on the spin lock from thinking
 907         * they got something (see the lock code above).
 908         */
 909        timer->it_signal = NULL;
 910
 911        unlock_timer(timer, flags);
 912        release_posix_timer(timer, IT_ID_SET);
 913        return 0;
 914}
 915
 916/*
 917 * return timer owned by the process, used by exit_itimers
 918 */
 919static void itimer_delete(struct k_itimer *timer)
 920{
 921        unsigned long flags;
 922
 923retry_delete:
 924        spin_lock_irqsave(&timer->it_lock, flags);
 925
 926        if (timer_delete_hook(timer) == TIMER_RETRY) {
 927                unlock_timer(timer, flags);
 928                goto retry_delete;
 929        }
 930        list_del(&timer->list);
 931        /*
 932         * This keeps any tasks waiting on the spin lock from thinking
 933         * they got something (see the lock code above).
 934         */
 935        timer->it_signal = NULL;
 936
 937        unlock_timer(timer, flags);
 938        release_posix_timer(timer, IT_ID_SET);
 939}
 940
 941/*
 942 * This is called by do_exit or de_thread, only when there are no more
 943 * references to the shared signal_struct.
 944 */
 945void exit_itimers(struct signal_struct *sig)
 946{
 947        struct k_itimer *tmr;
 948
 949        while (!list_empty(&sig->posix_timers)) {
 950                tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
 951                itimer_delete(tmr);
 952        }
 953}
 954
 955SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
 956                const struct timespec __user *, tp)
 957{
 958        struct k_clock *kc = clockid_to_kclock(which_clock);
 959        struct timespec new_tp;
 960
 961        if (!kc || !kc->clock_set)
 962                return -EINVAL;
 963
 964        if (copy_from_user(&new_tp, tp, sizeof (*tp)))
 965                return -EFAULT;
 966
 967        return kc->clock_set(which_clock, &new_tp);
 968}
 969
 970SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
 971                struct timespec __user *,tp)
 972{
 973        struct k_clock *kc = clockid_to_kclock(which_clock);
 974        struct timespec kernel_tp;
 975        int error;
 976
 977        if (!kc)
 978                return -EINVAL;
 979
 980        error = kc->clock_get(which_clock, &kernel_tp);
 981
 982        if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
 983                error = -EFAULT;
 984
 985        return error;
 986}
 987
 988SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
 989                struct timex __user *, utx)
 990{
 991        struct k_clock *kc = clockid_to_kclock(which_clock);
 992        struct timex ktx;
 993        int err;
 994
 995        if (!kc)
 996                return -EINVAL;
 997        if (!kc->clock_adj)
 998                return -EOPNOTSUPP;
 999
1000        if (copy_from_user(&ktx, utx, sizeof(ktx)))
1001                return -EFAULT;
1002
1003        err = kc->clock_adj(which_clock, &ktx);
1004
1005        if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1006                return -EFAULT;
1007
1008        return err;
1009}
1010
1011SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1012                struct timespec __user *, tp)
1013{
1014        struct k_clock *kc = clockid_to_kclock(which_clock);
1015        struct timespec rtn_tp;
1016        int error;
1017
1018        if (!kc)
1019                return -EINVAL;
1020
1021        error = kc->clock_getres(which_clock, &rtn_tp);
1022
1023        if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp)))
1024                error = -EFAULT;
1025
1026        return error;
1027}
1028
1029/*
1030 * nanosleep for monotonic and realtime clocks
1031 */
1032static int common_nsleep(const clockid_t which_clock, int flags,
1033                         struct timespec *tsave, struct timespec __user *rmtp)
1034{
1035        return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
1036                                 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1037                                 which_clock);
1038}
1039
1040SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1041                const struct timespec __user *, rqtp,
1042                struct timespec __user *, rmtp)
1043{
1044        struct k_clock *kc = clockid_to_kclock(which_clock);
1045        struct timespec t;
1046
1047        if (!kc)
1048                return -EINVAL;
1049        if (!kc->nsleep)
1050                return -ENANOSLEEP_NOTSUP;
1051
1052        if (copy_from_user(&t, rqtp, sizeof (struct timespec)))
1053                return -EFAULT;
1054
1055        if (!timespec_valid(&t))
1056                return -EINVAL;
1057
1058        return kc->nsleep(which_clock, flags, &t, rmtp);
1059}
1060
1061/*
1062 * This will restart clock_nanosleep. This is required only by
1063 * compat_clock_nanosleep_restart for now.
1064 */
1065long clock_nanosleep_restart(struct restart_block *restart_block)
1066{
1067        clockid_t which_clock = restart_block->nanosleep.clockid;
1068        struct k_clock *kc = clockid_to_kclock(which_clock);
1069
1070        if (WARN_ON_ONCE(!kc || !kc->nsleep_restart))
1071                return -EINVAL;
1072
1073        return kc->nsleep_restart(restart_block);
1074}
1075
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.