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/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/*
  53 * Management arrays for POSIX timers. Timers are now kept in static hash table
  54 * with 512 entries.
  55 * Timer ids are allocated by local routine, which selects proper hash head by
  56 * key, constructed from current->signal address and per signal struct counter.
  57 * This keeps timer ids unique per process, but now they can intersect between
  58 * processes.
  59 */
  60
  61/*
  62 * Lets keep our timers in a slab cache :-)
  63 */
  64static struct kmem_cache *posix_timers_cache;
  65
  66static DEFINE_HASHTABLE(posix_timers_hashtable, 9);
  67static DEFINE_SPINLOCK(hash_lock);
  68
  69/*
  70 * we assume that the new SIGEV_THREAD_ID shares no bits with the other
  71 * SIGEV values.  Here we put out an error if this assumption fails.
  72 */
  73#if SIGEV_THREAD_ID != (SIGEV_THREAD_ID & \
  74                       ~(SIGEV_SIGNAL | SIGEV_NONE | SIGEV_THREAD))
  75#error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
  76#endif
  77
  78/*
  79 * parisc wants ENOTSUP instead of EOPNOTSUPP
  80 */
  81#ifndef ENOTSUP
  82# define ENANOSLEEP_NOTSUP EOPNOTSUPP
  83#else
  84# define ENANOSLEEP_NOTSUP ENOTSUP
  85#endif
  86
  87/*
  88 * The timer ID is turned into a timer address by idr_find().
  89 * Verifying a valid ID consists of:
  90 *
  91 * a) checking that idr_find() returns other than -1.
  92 * b) checking that the timer id matches the one in the timer itself.
  93 * c) that the timer owner is in the callers thread group.
  94 */
  95
  96/*
  97 * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
  98 *          to implement others.  This structure defines the various
  99 *          clocks.
 100 *
 101 * RESOLUTION: Clock resolution is used to round up timer and interval
 102 *          times, NOT to report clock times, which are reported with as
 103 *          much resolution as the system can muster.  In some cases this
 104 *          resolution may depend on the underlying clock hardware and
 105 *          may not be quantifiable until run time, and only then is the
 106 *          necessary code is written.  The standard says we should say
 107 *          something about this issue in the documentation...
 108 *
 109 * FUNCTIONS: The CLOCKs structure defines possible functions to
 110 *          handle various clock functions.
 111 *
 112 *          The standard POSIX timer management code assumes the
 113 *          following: 1.) The k_itimer struct (sched.h) is used for
 114 *          the timer.  2.) The list, it_lock, it_clock, it_id and
 115 *          it_pid fields are not modified by timer code.
 116 *
 117 * Permissions: It is assumed that the clock_settime() function defined
 118 *          for each clock will take care of permission checks.  Some
 119 *          clocks may be set able by any user (i.e. local process
 120 *          clocks) others not.  Currently the only set able clock we
 121 *          have is CLOCK_REALTIME and its high res counter part, both of
 122 *          which we beg off on and pass to do_sys_settimeofday().
 123 */
 124
 125static struct k_clock posix_clocks[MAX_CLOCKS];
 126
 127/*
 128 * These ones are defined below.
 129 */
 130static int common_nsleep(const clockid_t, int flags, struct timespec *t,
 131                         struct timespec __user *rmtp);
 132static int common_timer_create(struct k_itimer *new_timer);
 133static void common_timer_get(struct k_itimer *, struct itimerspec *);
 134static int common_timer_set(struct k_itimer *, int,
 135                            struct itimerspec *, struct itimerspec *);
 136static int common_timer_del(struct k_itimer *timer);
 137
 138static enum hrtimer_restart posix_timer_fn(struct hrtimer *data);
 139
 140static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
 141
 142#define lock_timer(tid, flags)                                             \
 143({      struct k_itimer *__timr;                                           \
 144        __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
 145        __timr;                                                            \
 146})
 147
 148static int hash(struct signal_struct *sig, unsigned int nr)
 149{
 150        return hash_32(hash32_ptr(sig) ^ nr, HASH_BITS(posix_timers_hashtable));
 151}
 152
 153static struct k_itimer *__posix_timers_find(struct hlist_head *head,
 154                                            struct signal_struct *sig,
 155                                            timer_t id)
 156{
 157        struct k_itimer *timer;
 158
 159        hlist_for_each_entry_rcu(timer, head, t_hash) {
 160                if ((timer->it_signal == sig) && (timer->it_id == id))
 161                        return timer;
 162        }
 163        return NULL;
 164}
 165
 166static struct k_itimer *posix_timer_by_id(timer_t id)
 167{
 168        struct signal_struct *sig = current->signal;
 169        struct hlist_head *head = &posix_timers_hashtable[hash(sig, id)];
 170
 171        return __posix_timers_find(head, sig, id);
 172}
 173
 174static int posix_timer_add(struct k_itimer *timer)
 175{
 176        struct signal_struct *sig = current->signal;
 177        int first_free_id = sig->posix_timer_id;
 178        struct hlist_head *head;
 179        int ret = -ENOENT;
 180
 181        do {
 182                spin_lock(&hash_lock);
 183                head = &posix_timers_hashtable[hash(sig, sig->posix_timer_id)];
 184                if (!__posix_timers_find(head, sig, sig->posix_timer_id)) {
 185                        hlist_add_head_rcu(&timer->t_hash, head);
 186                        ret = sig->posix_timer_id;
 187                }
 188                if (++sig->posix_timer_id < 0)
 189                        sig->posix_timer_id = 0;
 190                if ((sig->posix_timer_id == first_free_id) && (ret == -ENOENT))
 191                        /* Loop over all possible ids completed */
 192                        ret = -EAGAIN;
 193                spin_unlock(&hash_lock);
 194        } while (ret == -ENOENT);
 195        return ret;
 196}
 197
 198static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
 199{
 200        spin_unlock_irqrestore(&timr->it_lock, flags);
 201}
 202
 203/* Get clock_realtime */
 204static int posix_clock_realtime_get(clockid_t which_clock, struct timespec *tp)
 205{
 206        ktime_get_real_ts(tp);
 207        return 0;
 208}
 209
 210/* Set clock_realtime */
 211static int posix_clock_realtime_set(const clockid_t which_clock,
 212                                    const struct timespec *tp)
 213{
 214        return do_sys_settimeofday(tp, NULL);
 215}
 216
 217static int posix_clock_realtime_adj(const clockid_t which_clock,
 218                                    struct timex *t)
 219{
 220        return do_adjtimex(t);
 221}
 222
 223/*
 224 * Get monotonic time for posix timers
 225 */
 226static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp)
 227{
 228        ktime_get_ts(tp);
 229        return 0;
 230}
 231
 232/*
 233 * Get monotonic-raw time for posix timers
 234 */
 235static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp)
 236{
 237        getrawmonotonic(tp);
 238        return 0;
 239}
 240
 241
 242static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp)
 243{
 244        *tp = current_kernel_time();
 245        return 0;
 246}
 247
 248static int posix_get_monotonic_coarse(clockid_t which_clock,
 249                                                struct timespec *tp)
 250{
 251        *tp = get_monotonic_coarse();
 252        return 0;
 253}
 254
 255static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp)
 256{
 257        *tp = ktime_to_timespec(KTIME_LOW_RES);
 258        return 0;
 259}
 260
 261static int posix_get_boottime(const clockid_t which_clock, struct timespec *tp)
 262{
 263        get_monotonic_boottime(tp);
 264        return 0;
 265}
 266
 267static int posix_get_tai(clockid_t which_clock, struct timespec *tp)
 268{
 269        timekeeping_clocktai(tp);
 270        return 0;
 271}
 272
 273/*
 274 * Initialize everything, well, just everything in Posix clocks/timers ;)
 275 */
 276static __init int init_posix_timers(void)
 277{
 278        struct k_clock clock_realtime = {
 279                .clock_getres   = hrtimer_get_res,
 280                .clock_get      = posix_clock_realtime_get,
 281                .clock_set      = posix_clock_realtime_set,
 282                .clock_adj      = posix_clock_realtime_adj,
 283                .nsleep         = common_nsleep,
 284                .nsleep_restart = hrtimer_nanosleep_restart,
 285                .timer_create   = common_timer_create,
 286                .timer_set      = common_timer_set,
 287                .timer_get      = common_timer_get,
 288                .timer_del      = common_timer_del,
 289        };
 290        struct k_clock clock_monotonic = {
 291                .clock_getres   = hrtimer_get_res,
 292                .clock_get      = posix_ktime_get_ts,
 293                .nsleep         = common_nsleep,
 294                .nsleep_restart = hrtimer_nanosleep_restart,
 295                .timer_create   = common_timer_create,
 296                .timer_set      = common_timer_set,
 297                .timer_get      = common_timer_get,
 298                .timer_del      = common_timer_del,
 299        };
 300        struct k_clock clock_monotonic_raw = {
 301                .clock_getres   = hrtimer_get_res,
 302                .clock_get      = posix_get_monotonic_raw,
 303        };
 304        struct k_clock clock_realtime_coarse = {
 305                .clock_getres   = posix_get_coarse_res,
 306                .clock_get      = posix_get_realtime_coarse,
 307        };
 308        struct k_clock clock_monotonic_coarse = {
 309                .clock_getres   = posix_get_coarse_res,
 310                .clock_get      = posix_get_monotonic_coarse,
 311        };
 312        struct k_clock clock_tai = {
 313                .clock_getres   = hrtimer_get_res,
 314                .clock_get      = posix_get_tai,
 315                .nsleep         = common_nsleep,
 316                .nsleep_restart = hrtimer_nanosleep_restart,
 317                .timer_create   = common_timer_create,
 318                .timer_set      = common_timer_set,
 319                .timer_get      = common_timer_get,
 320                .timer_del      = common_timer_del,
 321        };
 322        struct k_clock clock_boottime = {
 323                .clock_getres   = hrtimer_get_res,
 324                .clock_get      = posix_get_boottime,
 325                .nsleep         = common_nsleep,
 326                .nsleep_restart = hrtimer_nanosleep_restart,
 327                .timer_create   = common_timer_create,
 328                .timer_set      = common_timer_set,
 329                .timer_get      = common_timer_get,
 330                .timer_del      = common_timer_del,
 331        };
 332
 333        posix_timers_register_clock(CLOCK_REALTIME, &clock_realtime);
 334        posix_timers_register_clock(CLOCK_MONOTONIC, &clock_monotonic);
 335        posix_timers_register_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw);
 336        posix_timers_register_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse);
 337        posix_timers_register_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse);
 338        posix_timers_register_clock(CLOCK_BOOTTIME, &clock_boottime);
 339        posix_timers_register_clock(CLOCK_TAI, &clock_tai);
 340
 341        posix_timers_cache = kmem_cache_create("posix_timers_cache",
 342                                        sizeof (struct k_itimer), 0, SLAB_PANIC,
 343                                        NULL);
 344        return 0;
 345}
 346
 347__initcall(init_posix_timers);
 348
 349static void schedule_next_timer(struct k_itimer *timr)
 350{
 351        struct hrtimer *timer = &timr->it.real.timer;
 352
 353        if (timr->it.real.interval.tv64 == 0)
 354                return;
 355
 356        timr->it_overrun += (unsigned int) hrtimer_forward(timer,
 357                                                timer->base->get_time(),
 358                                                timr->it.real.interval);
 359
 360        timr->it_overrun_last = timr->it_overrun;
 361        timr->it_overrun = -1;
 362        ++timr->it_requeue_pending;
 363        hrtimer_restart(timer);
 364}
 365
 366/*
 367 * This function is exported for use by the signal deliver code.  It is
 368 * called just prior to the info block being released and passes that
 369 * block to us.  It's function is to update the overrun entry AND to
 370 * restart the timer.  It should only be called if the timer is to be
 371 * restarted (i.e. we have flagged this in the sys_private entry of the
 372 * info block).
 373 *
 374 * To protect against the timer going away while the interrupt is queued,
 375 * we require that the it_requeue_pending flag be set.
 376 */
 377void do_schedule_next_timer(struct siginfo *info)
 378{
 379        struct k_itimer *timr;
 380        unsigned long flags;
 381
 382        timr = lock_timer(info->si_tid, &flags);
 383
 384        if (timr && timr->it_requeue_pending == info->si_sys_private) {
 385                if (timr->it_clock < 0)
 386                        posix_cpu_timer_schedule(timr);
 387                else
 388                        schedule_next_timer(timr);
 389
 390                info->si_overrun += timr->it_overrun_last;
 391        }
 392
 393        if (timr)
 394                unlock_timer(timr, flags);
 395}
 396
 397int posix_timer_event(struct k_itimer *timr, int si_private)
 398{
 399        struct task_struct *task;
 400        int shared, ret = -1;
 401        /*
 402         * FIXME: if ->sigq is queued we can race with
 403         * dequeue_signal()->do_schedule_next_timer().
 404         *
 405         * If dequeue_signal() sees the "right" value of
 406         * si_sys_private it calls do_schedule_next_timer().
 407         * We re-queue ->sigq and drop ->it_lock().
 408         * do_schedule_next_timer() locks the timer
 409         * and re-schedules it while ->sigq is pending.
 410         * Not really bad, but not that we want.
 411         */
 412        timr->sigq->info.si_sys_private = si_private;
 413
 414        rcu_read_lock();
 415        task = pid_task(timr->it_pid, PIDTYPE_PID);
 416        if (task) {
 417                shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID);
 418                ret = send_sigqueue(timr->sigq, task, shared);
 419        }
 420        rcu_read_unlock();
 421        /* If we failed to send the signal the timer stops. */
 422        return ret > 0;
 423}
 424EXPORT_SYMBOL_GPL(posix_timer_event);
 425
 426/*
 427 * This function gets called when a POSIX.1b interval timer expires.  It
 428 * is used as a callback from the kernel internal timer.  The
 429 * run_timer_list code ALWAYS calls with interrupts on.
 430
 431 * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
 432 */
 433static enum hrtimer_restart posix_timer_fn(struct hrtimer *timer)
 434{
 435        struct k_itimer *timr;
 436        unsigned long flags;
 437        int si_private = 0;
 438        enum hrtimer_restart ret = HRTIMER_NORESTART;
 439
 440        timr = container_of(timer, struct k_itimer, it.real.timer);
 441        spin_lock_irqsave(&timr->it_lock, flags);
 442
 443        if (timr->it.real.interval.tv64 != 0)
 444                si_private = ++timr->it_requeue_pending;
 445
 446        if (posix_timer_event(timr, si_private)) {
 447                /*
 448                 * signal was not sent because of sig_ignor
 449                 * we will not get a call back to restart it AND
 450                 * it should be restarted.
 451                 */
 452                if (timr->it.real.interval.tv64 != 0) {
 453                        ktime_t now = hrtimer_cb_get_time(timer);
 454
 455                        /*
 456                         * FIXME: What we really want, is to stop this
 457                         * timer completely and restart it in case the
 458                         * SIG_IGN is removed. This is a non trivial
 459                         * change which involves sighand locking
 460                         * (sigh !), which we don't want to do late in
 461                         * the release cycle.
 462                         *
 463                         * For now we just let timers with an interval
 464                         * less than a jiffie expire every jiffie to
 465                         * avoid softirq starvation in case of SIG_IGN
 466                         * and a very small interval, which would put
 467                         * the timer right back on the softirq pending
 468                         * list. By moving now ahead of time we trick
 469                         * hrtimer_forward() to expire the timer
 470                         * later, while we still maintain the overrun
 471                         * accuracy, but have some inconsistency in
 472                         * the timer_gettime() case. This is at least
 473                         * better than a starved softirq. A more
 474                         * complex fix which solves also another related
 475                         * inconsistency is already in the pipeline.
 476                         */
 477#ifdef CONFIG_HIGH_RES_TIMERS
 478                        {
 479                                ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ);
 480
 481                                if (timr->it.real.interval.tv64 < kj.tv64)
 482                                        now = ktime_add(now, kj);
 483                        }
 484#endif
 485                        timr->it_overrun += (unsigned int)
 486                                hrtimer_forward(timer, now,
 487                                                timr->it.real.interval);
 488                        ret = HRTIMER_RESTART;
 489                        ++timr->it_requeue_pending;
 490                }
 491        }
 492
 493        unlock_timer(timr, flags);
 494        return ret;
 495}
 496
 497static struct pid *good_sigevent(sigevent_t * event)
 498{
 499        struct task_struct *rtn = current->group_leader;
 500
 501        if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
 502                (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
 503                 !same_thread_group(rtn, current) ||
 504                 (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL))
 505                return NULL;
 506
 507        if (((event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE) &&
 508            ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX)))
 509                return NULL;
 510
 511        return task_pid(rtn);
 512}
 513
 514void posix_timers_register_clock(const clockid_t clock_id,
 515                                 struct k_clock *new_clock)
 516{
 517        if ((unsigned) clock_id >= MAX_CLOCKS) {
 518                printk(KERN_WARNING "POSIX clock register failed for clock_id %d\n",
 519                       clock_id);
 520                return;
 521        }
 522
 523        if (!new_clock->clock_get) {
 524                printk(KERN_WARNING "POSIX clock id %d lacks clock_get()\n",
 525                       clock_id);
 526                return;
 527        }
 528        if (!new_clock->clock_getres) {
 529                printk(KERN_WARNING "POSIX clock id %d lacks clock_getres()\n",
 530                       clock_id);
 531                return;
 532        }
 533
 534        posix_clocks[clock_id] = *new_clock;
 535}
 536EXPORT_SYMBOL_GPL(posix_timers_register_clock);
 537
 538static struct k_itimer * alloc_posix_timer(void)
 539{
 540        struct k_itimer *tmr;
 541        tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
 542        if (!tmr)
 543                return tmr;
 544        if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
 545                kmem_cache_free(posix_timers_cache, tmr);
 546                return NULL;
 547        }
 548        memset(&tmr->sigq->info, 0, sizeof(siginfo_t));
 549        return tmr;
 550}
 551
 552static void k_itimer_rcu_free(struct rcu_head *head)
 553{
 554        struct k_itimer *tmr = container_of(head, struct k_itimer, it.rcu);
 555
 556        kmem_cache_free(posix_timers_cache, tmr);
 557}
 558
 559#define IT_ID_SET       1
 560#define IT_ID_NOT_SET   0
 561static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
 562{
 563        if (it_id_set) {
 564                unsigned long flags;
 565                spin_lock_irqsave(&hash_lock, flags);
 566                hlist_del_rcu(&tmr->t_hash);
 567                spin_unlock_irqrestore(&hash_lock, flags);
 568        }
 569        put_pid(tmr->it_pid);
 570        sigqueue_free(tmr->sigq);
 571        call_rcu(&tmr->it.rcu, k_itimer_rcu_free);
 572}
 573
 574static struct k_clock *clockid_to_kclock(const clockid_t id)
 575{
 576        if (id < 0)
 577                return (id & CLOCKFD_MASK) == CLOCKFD ?
 578                        &clock_posix_dynamic : &clock_posix_cpu;
 579
 580        if (id >= MAX_CLOCKS || !posix_clocks[id].clock_getres)
 581                return NULL;
 582        return &posix_clocks[id];
 583}
 584
 585static int common_timer_create(struct k_itimer *new_timer)
 586{
 587        hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
 588        return 0;
 589}
 590
 591/* Create a POSIX.1b interval timer. */
 592
 593SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
 594                struct sigevent __user *, timer_event_spec,
 595                timer_t __user *, created_timer_id)
 596{
 597        struct k_clock *kc = clockid_to_kclock(which_clock);
 598        struct k_itimer *new_timer;
 599        int error, new_timer_id;
 600        sigevent_t event;
 601        int it_id_set = IT_ID_NOT_SET;
 602
 603        if (!kc)
 604                return -EINVAL;
 605        if (!kc->timer_create)
 606                return -EOPNOTSUPP;
 607
 608        new_timer = alloc_posix_timer();
 609        if (unlikely(!new_timer))
 610                return -EAGAIN;
 611
 612        spin_lock_init(&new_timer->it_lock);
 613        new_timer_id = posix_timer_add(new_timer);
 614        if (new_timer_id < 0) {
 615                error = new_timer_id;
 616                goto out;
 617        }
 618
 619        it_id_set = IT_ID_SET;
 620        new_timer->it_id = (timer_t) new_timer_id;
 621        new_timer->it_clock = which_clock;
 622        new_timer->it_overrun = -1;
 623
 624        if (timer_event_spec) {
 625                if (copy_from_user(&event, timer_event_spec, sizeof (event))) {
 626                        error = -EFAULT;
 627                        goto out;
 628                }
 629                rcu_read_lock();
 630                new_timer->it_pid = get_pid(good_sigevent(&event));
 631                rcu_read_unlock();
 632                if (!new_timer->it_pid) {
 633                        error = -EINVAL;
 634                        goto out;
 635                }
 636        } else {
 637                event.sigev_notify = SIGEV_SIGNAL;
 638                event.sigev_signo = SIGALRM;
 639                event.sigev_value.sival_int = new_timer->it_id;
 640                new_timer->it_pid = get_pid(task_tgid(current));
 641        }
 642
 643        new_timer->it_sigev_notify     = event.sigev_notify;
 644        new_timer->sigq->info.si_signo = event.sigev_signo;
 645        new_timer->sigq->info.si_value = event.sigev_value;
 646        new_timer->sigq->info.si_tid   = new_timer->it_id;
 647        new_timer->sigq->info.si_code  = SI_TIMER;
 648
 649        if (copy_to_user(created_timer_id,
 650                         &new_timer_id, sizeof (new_timer_id))) {
 651                error = -EFAULT;
 652                goto out;
 653        }
 654
 655        error = kc->timer_create(new_timer);
 656        if (error)
 657                goto out;
 658
 659        spin_lock_irq(&current->sighand->siglock);
 660        new_timer->it_signal = current->signal;
 661        list_add(&new_timer->list, &current->signal->posix_timers);
 662        spin_unlock_irq(&current->sighand->siglock);
 663
 664        return 0;
 665        /*
 666         * In the case of the timer belonging to another task, after
 667         * the task is unlocked, the timer is owned by the other task
 668         * and may cease to exist at any time.  Don't use or modify
 669         * new_timer after the unlock call.
 670         */
 671out:
 672        release_posix_timer(new_timer, it_id_set);
 673        return error;
 674}
 675
 676/*
 677 * Locking issues: We need to protect the result of the id look up until
 678 * we get the timer locked down so it is not deleted under us.  The
 679 * removal is done under the idr spinlock so we use that here to bridge
 680 * the find to the timer lock.  To avoid a dead lock, the timer id MUST
 681 * be release with out holding the timer lock.
 682 */
 683static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 684{
 685        struct k_itimer *timr;
 686
 687        /*
 688         * timer_t could be any type >= int and we want to make sure any
 689         * @timer_id outside positive int range fails lookup.
 690         */
 691        if ((unsigned long long)timer_id > INT_MAX)
 692                return NULL;
 693
 694        rcu_read_lock();
 695        timr = posix_timer_by_id(timer_id);
 696        if (timr) {
 697                spin_lock_irqsave(&timr->it_lock, *flags);
 698                if (timr->it_signal == current->signal) {
 699                        rcu_read_unlock();
 700                        return timr;
 701                }
 702                spin_unlock_irqrestore(&timr->it_lock, *flags);
 703        }
 704        rcu_read_unlock();
 705
 706        return NULL;
 707}
 708
 709/*
 710 * Get the time remaining on a POSIX.1b interval timer.  This function
 711 * is ALWAYS called with spin_lock_irq on the timer, thus it must not
 712 * mess with irq.
 713 *
 714 * We have a couple of messes to clean up here.  First there is the case
 715 * of a timer that has a requeue pending.  These timers should appear to
 716 * be in the timer list with an expiry as if we were to requeue them
 717 * now.
 718 *
 719 * The second issue is the SIGEV_NONE timer which may be active but is
 720 * not really ever put in the timer list (to save system resources).
 721 * This timer may be expired, and if so, we will do it here.  Otherwise
 722 * it is the same as a requeue pending timer WRT to what we should
 723 * report.
 724 */
 725static void
 726common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
 727{
 728        ktime_t now, remaining, iv;
 729        struct hrtimer *timer = &timr->it.real.timer;
 730
 731        memset(cur_setting, 0, sizeof(struct itimerspec));
 732
 733        iv = timr->it.real.interval;
 734
 735        /* interval timer ? */
 736        if (iv.tv64)
 737                cur_setting->it_interval = ktime_to_timespec(iv);
 738        else if (!hrtimer_active(timer) &&
 739                 (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
 740                return;
 741
 742        now = timer->base->get_time();
 743
 744        /*
 745         * When a requeue is pending or this is a SIGEV_NONE
 746         * timer move the expiry time forward by intervals, so
 747         * expiry is > now.
 748         */
 749        if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING ||
 750            (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE))
 751                timr->it_overrun += (unsigned int) hrtimer_forward(timer, now, iv);
 752
 753        remaining = ktime_sub(hrtimer_get_expires(timer), now);
 754        /* Return 0 only, when the timer is expired and not pending */
 755        if (remaining.tv64 <= 0) {
 756                /*
 757                 * A single shot SIGEV_NONE timer must return 0, when
 758                 * it is expired !
 759                 */
 760                if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
 761                        cur_setting->it_value.tv_nsec = 1;
 762        } else
 763                cur_setting->it_value = ktime_to_timespec(remaining);
 764}
 765
 766/* Get the time remaining on a POSIX.1b interval timer. */
 767SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
 768                struct itimerspec __user *, setting)
 769{
 770        struct itimerspec cur_setting;
 771        struct k_itimer *timr;
 772        struct k_clock *kc;
 773        unsigned long flags;
 774        int ret = 0;
 775
 776        timr = lock_timer(timer_id, &flags);
 777        if (!timr)
 778                return -EINVAL;
 779
 780        kc = clockid_to_kclock(timr->it_clock);
 781        if (WARN_ON_ONCE(!kc || !kc->timer_get))
 782                ret = -EINVAL;
 783        else
 784                kc->timer_get(timr, &cur_setting);
 785
 786        unlock_timer(timr, flags);
 787
 788        if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
 789                return -EFAULT;
 790
 791        return ret;
 792}
 793
 794/*
 795 * Get the number of overruns of a POSIX.1b interval timer.  This is to
 796 * be the overrun of the timer last delivered.  At the same time we are
 797 * accumulating overruns on the next timer.  The overrun is frozen when
 798 * the signal is delivered, either at the notify time (if the info block
 799 * is not queued) or at the actual delivery time (as we are informed by
 800 * the call back to do_schedule_next_timer().  So all we need to do is
 801 * to pick up the frozen overrun.
 802 */
 803SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 804{
 805        struct k_itimer *timr;
 806        int overrun;
 807        unsigned long flags;
 808
 809        timr = lock_timer(timer_id, &flags);
 810        if (!timr)
 811                return -EINVAL;
 812
 813        overrun = timr->it_overrun_last;
 814        unlock_timer(timr, flags);
 815
 816        return overrun;
 817}
 818
 819/* Set a POSIX.1b interval timer. */
 820/* timr->it_lock is taken. */
 821static int
 822common_timer_set(struct k_itimer *timr, int flags,
 823                 struct itimerspec *new_setting, struct itimerspec *old_setting)
 824{
 825        struct hrtimer *timer = &timr->it.real.timer;
 826        enum hrtimer_mode mode;
 827
 828        if (old_setting)
 829                common_timer_get(timr, old_setting);
 830
 831        /* disable the timer */
 832        timr->it.real.interval.tv64 = 0;
 833        /*
 834         * careful here.  If smp we could be in the "fire" routine which will
 835         * be spinning as we hold the lock.  But this is ONLY an SMP issue.
 836         */
 837        if (hrtimer_try_to_cancel(timer) < 0)
 838                return TIMER_RETRY;
 839
 840        timr->it_requeue_pending = (timr->it_requeue_pending + 2) & 
 841                ~REQUEUE_PENDING;
 842        timr->it_overrun_last = 0;
 843
 844        /* switch off the timer when it_value is zero */
 845        if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
 846                return 0;
 847
 848        mode = flags & TIMER_ABSTIME ? HRTIMER_MODE_ABS : HRTIMER_MODE_REL;
 849        hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
 850        timr->it.real.timer.function = posix_timer_fn;
 851
 852        hrtimer_set_expires(timer, timespec_to_ktime(new_setting->it_value));
 853
 854        /* Convert interval */
 855        timr->it.real.interval = timespec_to_ktime(new_setting->it_interval);
 856
 857        /* SIGEV_NONE timers are not queued ! See common_timer_get */
 858        if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) {
 859                /* Setup correct expiry time for relative timers */
 860                if (mode == HRTIMER_MODE_REL) {
 861                        hrtimer_add_expires(timer, timer->base->get_time());
 862                }
 863                return 0;
 864        }
 865
 866        hrtimer_start_expires(timer, mode);
 867        return 0;
 868}
 869
 870/* Set a POSIX.1b interval timer */
 871SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
 872                const struct itimerspec __user *, new_setting,
 873                struct itimerspec __user *, old_setting)
 874{
 875        struct k_itimer *timr;
 876        struct itimerspec new_spec, old_spec;
 877        int error = 0;
 878        unsigned long flag;
 879        struct itimerspec *rtn = old_setting ? &old_spec : NULL;
 880        struct k_clock *kc;
 881
 882        if (!new_setting)
 883                return -EINVAL;
 884
 885        if (copy_from_user(&new_spec, new_setting, sizeof (new_spec)))
 886                return -EFAULT;
 887
 888        if (!timespec_valid(&new_spec.it_interval) ||
 889            !timespec_valid(&new_spec.it_value))
 890                return -EINVAL;
 891retry:
 892        timr = lock_timer(timer_id, &flag);
 893        if (!timr)
 894                return -EINVAL;
 895
 896        kc = clockid_to_kclock(timr->it_clock);
 897        if (WARN_ON_ONCE(!kc || !kc->timer_set))
 898                error = -EINVAL;
 899        else
 900                error = kc->timer_set(timr, flags, &new_spec, rtn);
 901
 902        unlock_timer(timr, flag);
 903        if (error == TIMER_RETRY) {
 904                rtn = NULL;     // We already got the old time...
 905                goto retry;
 906        }
 907
 908        if (old_setting && !error &&
 909            copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
 910                error = -EFAULT;
 911
 912        return error;
 913}
 914
 915static int common_timer_del(struct k_itimer *timer)
 916{
 917        timer->it.real.interval.tv64 = 0;
 918
 919        if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0)
 920                return TIMER_RETRY;
 921        return 0;
 922}
 923
 924static inline int timer_delete_hook(struct k_itimer *timer)
 925{
 926        struct k_clock *kc = clockid_to_kclock(timer->it_clock);
 927
 928        if (WARN_ON_ONCE(!kc || !kc->timer_del))
 929                return -EINVAL;
 930        return kc->timer_del(timer);
 931}
 932
 933/* Delete a POSIX.1b interval timer. */
 934SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 935{
 936        struct k_itimer *timer;
 937        unsigned long flags;
 938
 939retry_delete:
 940        timer = lock_timer(timer_id, &flags);
 941        if (!timer)
 942                return -EINVAL;
 943
 944        if (timer_delete_hook(timer) == TIMER_RETRY) {
 945                unlock_timer(timer, flags);
 946                goto retry_delete;
 947        }
 948
 949        spin_lock(&current->sighand->siglock);
 950        list_del(&timer->list);
 951        spin_unlock(&current->sighand->siglock);
 952        /*
 953         * This keeps any tasks waiting on the spin lock from thinking
 954         * they got something (see the lock code above).
 955         */
 956        timer->it_signal = NULL;
 957
 958        unlock_timer(timer, flags);
 959        release_posix_timer(timer, IT_ID_SET);
 960        return 0;
 961}
 962
 963/*
 964 * return timer owned by the process, used by exit_itimers
 965 */
 966static void itimer_delete(struct k_itimer *timer)
 967{
 968        unsigned long flags;
 969
 970retry_delete:
 971        spin_lock_irqsave(&timer->it_lock, flags);
 972
 973        if (timer_delete_hook(timer) == TIMER_RETRY) {
 974                unlock_timer(timer, flags);
 975                goto retry_delete;
 976        }
 977        list_del(&timer->list);
 978        /*
 979         * This keeps any tasks waiting on the spin lock from thinking
 980         * they got something (see the lock code above).
 981         */
 982        timer->it_signal = NULL;
 983
 984        unlock_timer(timer, flags);
 985        release_posix_timer(timer, IT_ID_SET);
 986}
 987
 988/*
 989 * This is called by do_exit or de_thread, only when there are no more
 990 * references to the shared signal_struct.
 991 */
 992void exit_itimers(struct signal_struct *sig)
 993{
 994        struct k_itimer *tmr;
 995
 996        while (!list_empty(&sig->posix_timers)) {
 997                tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
 998                itimer_delete(tmr);
 999        }
1000}
1001
1002SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
1003                const struct timespec __user *, tp)
1004{
1005        struct k_clock *kc = clockid_to_kclock(which_clock);
1006        struct timespec new_tp;
1007
1008        if (!kc || !kc->clock_set)
1009                return -EINVAL;
1010
1011        if (copy_from_user(&new_tp, tp, sizeof (*tp)))
1012                return -EFAULT;
1013
1014        return kc->clock_set(which_clock, &new_tp);
1015}
1016
1017SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
1018                struct timespec __user *,tp)
1019{
1020        struct k_clock *kc = clockid_to_kclock(which_clock);
1021        struct timespec kernel_tp;
1022        int error;
1023
1024        if (!kc)
1025                return -EINVAL;
1026
1027        error = kc->clock_get(which_clock, &kernel_tp);
1028
1029        if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
1030                error = -EFAULT;
1031
1032        return error;
1033}
1034
1035SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
1036                struct timex __user *, utx)
1037{
1038        struct k_clock *kc = clockid_to_kclock(which_clock);
1039        struct timex ktx;
1040        int err;
1041
1042        if (!kc)
1043                return -EINVAL;
1044        if (!kc->clock_adj)
1045                return -EOPNOTSUPP;
1046
1047        if (copy_from_user(&ktx, utx, sizeof(ktx)))
1048                return -EFAULT;
1049
1050        err = kc->clock_adj(which_clock, &ktx);
1051
1052        if (err >= 0 && copy_to_user(utx, &ktx, sizeof(ktx)))
1053                return -EFAULT;
1054
1055        return err;
1056}
1057
1058SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
1059                struct timespec __user *, tp)
1060{
1061        struct k_clock *kc = clockid_to_kclock(which_clock);
1062        struct timespec rtn_tp;
1063        int error;
1064
1065        if (!kc)
1066                return -EINVAL;
1067
1068        error = kc->clock_getres(which_clock, &rtn_tp);
1069
1070        if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp)))
1071                error = -EFAULT;
1072
1073        return error;
1074}
1075
1076/*
1077 * nanosleep for monotonic and realtime clocks
1078 */
1079static int common_nsleep(const clockid_t which_clock, int flags,
1080                         struct timespec *tsave, struct timespec __user *rmtp)
1081{
1082        return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
1083                                 HRTIMER_MODE_ABS : HRTIMER_MODE_REL,
1084                                 which_clock);
1085}
1086
1087SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
1088                const struct timespec __user *, rqtp,
1089                struct timespec __user *, rmtp)
1090{
1091        struct k_clock *kc = clockid_to_kclock(which_clock);
1092        struct timespec t;
1093
1094        if (!kc)
1095                return -EINVAL;
1096        if (!kc->nsleep)
1097                return -ENANOSLEEP_NOTSUP;
1098
1099        if (copy_from_user(&t, rqtp, sizeof (struct timespec)))
1100                return -EFAULT;
1101
1102        if (!timespec_valid(&t))
1103                return -EINVAL;
1104
1105        return kc->nsleep(which_clock, flags, &t, rmtp);
1106}
1107
1108/*
1109 * This will restart clock_nanosleep. This is required only by
1110 * compat_clock_nanosleep_restart for now.
1111 */
1112long clock_nanosleep_restart(struct restart_block *restart_block)
1113{
1114        clockid_t which_clock = restart_block->nanosleep.clockid;
1115        struct k_clock *kc = clockid_to_kclock(which_clock);
1116
1117        if (WARN_ON_ONCE(!kc || !kc->nsleep_restart))
1118                return -EINVAL;
1119
1120        return kc->nsleep_restart(restart_block);
1121}
1122
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.