linux/fs/select.c/
<<
>>
Prefs
   1/*
   2 * This file contains the procedures for the handling of select and poll
   3 *
   4 * Created for Linux based loosely upon Mathius Lattner's minix
   5 * patches by Peter MacDonald. Heavily edited by Linus.
   6 *
   7 *  4 February 1994
   8 *     COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS
   9 *     flag set in its personality we do *not* modify the given timeout
  10 *     parameter to reflect time remaining.
  11 *
  12 *  24 January 2000
  13 *     Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 
  14 *     of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian).
  15 */
  16
  17#include <linux/kernel.h>
  18#include <linux/sched.h>
  19#include <linux/syscalls.h>
  20#include <linux/export.h>
  21#include <linux/slab.h>
  22#include <linux/poll.h>
  23#include <linux/personality.h> /* for STICKY_TIMEOUTS */
  24#include <linux/file.h>
  25#include <linux/fdtable.h>
  26#include <linux/fs.h>
  27#include <linux/rcupdate.h>
  28#include <linux/hrtimer.h>
  29
  30#include <asm/uaccess.h>
  31
  32
  33/*
  34 * Estimate expected accuracy in ns from a timeval.
  35 *
  36 * After quite a bit of churning around, we've settled on
  37 * a simple thing of taking 0.1% of the timeout as the
  38 * slack, with a cap of 100 msec.
  39 * "nice" tasks get a 0.5% slack instead.
  40 *
  41 * Consider this comment an open invitation to come up with even
  42 * better solutions..
  43 */
  44
  45#define MAX_SLACK       (100 * NSEC_PER_MSEC)
  46
  47static long __estimate_accuracy(struct timespec *tv)
  48{
  49        long slack;
  50        int divfactor = 1000;
  51
  52        if (tv->tv_sec < 0)
  53                return 0;
  54
  55        if (task_nice(current) > 0)
  56                divfactor = divfactor / 5;
  57
  58        if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor))
  59                return MAX_SLACK;
  60
  61        slack = tv->tv_nsec / divfactor;
  62        slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
  63
  64        if (slack > MAX_SLACK)
  65                return MAX_SLACK;
  66
  67        return slack;
  68}
  69
  70long select_estimate_accuracy(struct timespec *tv)
  71{
  72        unsigned long ret;
  73        struct timespec now;
  74
  75        /*
  76         * Realtime tasks get a slack of 0 for obvious reasons.
  77         */
  78
  79        if (rt_task(current))
  80                return 0;
  81
  82        ktime_get_ts(&now);
  83        now = timespec_sub(*tv, now);
  84        ret = __estimate_accuracy(&now);
  85        if (ret < current->timer_slack_ns)
  86                return current->timer_slack_ns;
  87        return ret;
  88}
  89
  90
  91
  92struct poll_table_page {
  93        struct poll_table_page * next;
  94        struct poll_table_entry * entry;
  95        struct poll_table_entry entries[0];
  96};
  97
  98#define POLL_TABLE_FULL(table) \
  99        ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
 100
 101/*
 102 * Ok, Peter made a complicated, but straightforward multiple_wait() function.
 103 * I have rewritten this, taking some shortcuts: This code may not be easy to
 104 * follow, but it should be free of race-conditions, and it's practical. If you
 105 * understand what I'm doing here, then you understand how the linux
 106 * sleep/wakeup mechanism works.
 107 *
 108 * Two very simple procedures, poll_wait() and poll_freewait() make all the
 109 * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
 110 * as all select/poll functions have to call it to add an entry to the
 111 * poll table.
 112 */
 113static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 114                       poll_table *p);
 115
 116void poll_initwait(struct poll_wqueues *pwq)
 117{
 118        init_poll_funcptr(&pwq->pt, __pollwait);
 119        pwq->polling_task = current;
 120        pwq->triggered = 0;
 121        pwq->error = 0;
 122        pwq->table = NULL;
 123        pwq->inline_index = 0;
 124}
 125EXPORT_SYMBOL(poll_initwait);
 126
 127static void free_poll_entry(struct poll_table_entry *entry)
 128{
 129        remove_wait_queue(entry->wait_address, &entry->wait);
 130        fput(entry->filp);
 131}
 132
 133void poll_freewait(struct poll_wqueues *pwq)
 134{
 135        struct poll_table_page * p = pwq->table;
 136        int i;
 137        for (i = 0; i < pwq->inline_index; i++)
 138                free_poll_entry(pwq->inline_entries + i);
 139        while (p) {
 140                struct poll_table_entry * entry;
 141                struct poll_table_page *old;
 142
 143                entry = p->entry;
 144                do {
 145                        entry--;
 146                        free_poll_entry(entry);
 147                } while (entry > p->entries);
 148                old = p;
 149                p = p->next;
 150                free_page((unsigned long) old);
 151        }
 152}
 153EXPORT_SYMBOL(poll_freewait);
 154
 155static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
 156{
 157        struct poll_table_page *table = p->table;
 158
 159        if (p->inline_index < N_INLINE_POLL_ENTRIES)
 160                return p->inline_entries + p->inline_index++;
 161
 162        if (!table || POLL_TABLE_FULL(table)) {
 163                struct poll_table_page *new_table;
 164
 165                new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 166                if (!new_table) {
 167                        p->error = -ENOMEM;
 168                        return NULL;
 169                }
 170                new_table->entry = new_table->entries;
 171                new_table->next = table;
 172                p->table = new_table;
 173                table = new_table;
 174        }
 175
 176        return table->entry++;
 177}
 178
 179static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
 180{
 181        struct poll_wqueues *pwq = wait->private;
 182        DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
 183
 184        /*
 185         * Although this function is called under waitqueue lock, LOCK
 186         * doesn't imply write barrier and the users expect write
 187         * barrier semantics on wakeup functions.  The following
 188         * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
 189         * and is paired with set_mb() in poll_schedule_timeout.
 190         */
 191        smp_wmb();
 192        pwq->triggered = 1;
 193
 194        /*
 195         * Perform the default wake up operation using a dummy
 196         * waitqueue.
 197         *
 198         * TODO: This is hacky but there currently is no interface to
 199         * pass in @sync.  @sync is scheduled to be removed and once
 200         * that happens, wake_up_process() can be used directly.
 201         */
 202        return default_wake_function(&dummy_wait, mode, sync, key);
 203}
 204
 205static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
 206{
 207        struct poll_table_entry *entry;
 208
 209        entry = container_of(wait, struct poll_table_entry, wait);
 210        if (key && !((unsigned long)key & entry->key))
 211                return 0;
 212        return __pollwake(wait, mode, sync, key);
 213}
 214
 215/* Add a new entry */
 216static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 217                                poll_table *p)
 218{
 219        struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
 220        struct poll_table_entry *entry = poll_get_entry(pwq);
 221        if (!entry)
 222                return;
 223        entry->filp = get_file(filp);
 224        entry->wait_address = wait_address;
 225        entry->key = p->_key;
 226        init_waitqueue_func_entry(&entry->wait, pollwake);
 227        entry->wait.private = pwq;
 228        add_wait_queue(wait_address, &entry->wait);
 229}
 230
 231int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
 232                          ktime_t *expires, unsigned long slack)
 233{
 234        int rc = -EINTR;
 235
 236        set_current_state(state);
 237        if (!pwq->triggered)
 238                rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
 239        __set_current_state(TASK_RUNNING);
 240
 241        /*
 242         * Prepare for the next iteration.
 243         *
 244         * The following set_mb() serves two purposes.  First, it's
 245         * the counterpart rmb of the wmb in pollwake() such that data
 246         * written before wake up is always visible after wake up.
 247         * Second, the full barrier guarantees that triggered clearing
 248         * doesn't pass event check of the next iteration.  Note that
 249         * this problem doesn't exist for the first iteration as
 250         * add_wait_queue() has full barrier semantics.
 251         */
 252        set_mb(pwq->triggered, 0);
 253
 254        return rc;
 255}
 256EXPORT_SYMBOL(poll_schedule_timeout);
 257
 258/**
 259 * poll_select_set_timeout - helper function to setup the timeout value
 260 * @to:         pointer to timespec variable for the final timeout
 261 * @sec:        seconds (from user space)
 262 * @nsec:       nanoseconds (from user space)
 263 *
 264 * Note, we do not use a timespec for the user space value here, That
 265 * way we can use the function for timeval and compat interfaces as well.
 266 *
 267 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
 268 */
 269int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
 270{
 271        struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
 272
 273        if (!timespec_valid(&ts))
 274                return -EINVAL;
 275
 276        /* Optimize for the zero timeout value here */
 277        if (!sec && !nsec) {
 278                to->tv_sec = to->tv_nsec = 0;
 279        } else {
 280                ktime_get_ts(to);
 281                *to = timespec_add_safe(*to, ts);
 282        }
 283        return 0;
 284}
 285
 286static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
 287                                      int timeval, int ret)
 288{
 289        struct timespec rts;
 290        struct timeval rtv;
 291
 292        if (!p)
 293                return ret;
 294
 295        if (current->personality & STICKY_TIMEOUTS)
 296                goto sticky;
 297
 298        /* No update for zero timeout */
 299        if (!end_time->tv_sec && !end_time->tv_nsec)
 300                return ret;
 301
 302        ktime_get_ts(&rts);
 303        rts = timespec_sub(*end_time, rts);
 304        if (rts.tv_sec < 0)
 305                rts.tv_sec = rts.tv_nsec = 0;
 306
 307        if (timeval) {
 308                if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec))
 309                        memset(&rtv, 0, sizeof(rtv));
 310                rtv.tv_sec = rts.tv_sec;
 311                rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 312
 313                if (!copy_to_user(p, &rtv, sizeof(rtv)))
 314                        return ret;
 315
 316        } else if (!copy_to_user(p, &rts, sizeof(rts)))
 317                return ret;
 318
 319        /*
 320         * If an application puts its timeval in read-only memory, we
 321         * don't want the Linux-specific update to the timeval to
 322         * cause a fault after the select has completed
 323         * successfully. However, because we're not updating the
 324         * timeval, we can't restart the system call.
 325         */
 326
 327sticky:
 328        if (ret == -ERESTARTNOHAND)
 329                ret = -EINTR;
 330        return ret;
 331}
 332
 333#define FDS_IN(fds, n)          (fds->in + n)
 334#define FDS_OUT(fds, n)         (fds->out + n)
 335#define FDS_EX(fds, n)          (fds->ex + n)
 336
 337#define BITS(fds, n)    (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 338
 339static int max_select_fd(unsigned long n, fd_set_bits *fds)
 340{
 341        unsigned long *open_fds;
 342        unsigned long set;
 343        int max;
 344        struct fdtable *fdt;
 345
 346        /* handle last in-complete long-word first */
 347        set = ~(~0UL << (n & (BITS_PER_LONG-1)));
 348        n /= BITS_PER_LONG;
 349        fdt = files_fdtable(current->files);
 350        open_fds = fdt->open_fds + n;
 351        max = 0;
 352        if (set) {
 353                set &= BITS(fds, n);
 354                if (set) {
 355                        if (!(set & ~*open_fds))
 356                                goto get_max;
 357                        return -EBADF;
 358                }
 359        }
 360        while (n) {
 361                open_fds--;
 362                n--;
 363                set = BITS(fds, n);
 364                if (!set)
 365                        continue;
 366                if (set & ~*open_fds)
 367                        return -EBADF;
 368                if (max)
 369                        continue;
 370get_max:
 371                do {
 372                        max++;
 373                        set >>= 1;
 374                } while (set);
 375                max += n * BITS_PER_LONG;
 376        }
 377
 378        return max;
 379}
 380
 381#define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
 382#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
 383#define POLLEX_SET (POLLPRI)
 384
 385static inline void wait_key_set(poll_table *wait, unsigned long in,
 386                                unsigned long out, unsigned long bit)
 387{
 388        wait->_key = POLLEX_SET;
 389        if (in & bit)
 390                wait->_key |= POLLIN_SET;
 391        if (out & bit)
 392                wait->_key |= POLLOUT_SET;
 393}
 394
 395int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
 396{
 397        ktime_t expire, *to = NULL;
 398        struct poll_wqueues table;
 399        poll_table *wait;
 400        int retval, i, timed_out = 0;
 401        unsigned long slack = 0;
 402
 403        rcu_read_lock();
 404        retval = max_select_fd(n, fds);
 405        rcu_read_unlock();
 406
 407        if (retval < 0)
 408                return retval;
 409        n = retval;
 410
 411        poll_initwait(&table);
 412        wait = &table.pt;
 413        if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 414                wait->_qproc = NULL;
 415                timed_out = 1;
 416        }
 417
 418        if (end_time && !timed_out)
 419                slack = select_estimate_accuracy(end_time);
 420
 421        retval = 0;
 422        for (;;) {
 423                unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 424
 425                inp = fds->in; outp = fds->out; exp = fds->ex;
 426                rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 427
 428                for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 429                        unsigned long in, out, ex, all_bits, bit = 1, mask, j;
 430                        unsigned long res_in = 0, res_out = 0, res_ex = 0;
 431
 432                        in = *inp++; out = *outp++; ex = *exp++;
 433                        all_bits = in | out | ex;
 434                        if (all_bits == 0) {
 435                                i += BITS_PER_LONG;
 436                                continue;
 437                        }
 438
 439                        for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) {
 440                                struct fd f;
 441                                if (i >= n)
 442                                        break;
 443                                if (!(bit & all_bits))
 444                                        continue;
 445                                f = fdget(i);
 446                                if (f.file) {
 447                                        const struct file_operations *f_op;
 448                                        f_op = f.file->f_op;
 449                                        mask = DEFAULT_POLLMASK;
 450                                        if (f_op && f_op->poll) {
 451                                                wait_key_set(wait, in, out, bit);
 452                                                mask = (*f_op->poll)(f.file, wait);
 453                                        }
 454                                        fdput(f);
 455                                        if ((mask & POLLIN_SET) && (in & bit)) {
 456                                                res_in |= bit;
 457                                                retval++;
 458                                                wait->_qproc = NULL;
 459                                        }
 460                                        if ((mask & POLLOUT_SET) && (out & bit)) {
 461                                                res_out |= bit;
 462                                                retval++;
 463                                                wait->_qproc = NULL;
 464                                        }
 465                                        if ((mask & POLLEX_SET) && (ex & bit)) {
 466                                                res_ex |= bit;
 467                                                retval++;
 468                                                wait->_qproc = NULL;
 469                                        }
 470                                }
 471                        }
 472                        if (res_in)
 473                                *rinp = res_in;
 474                        if (res_out)
 475                                *routp = res_out;
 476                        if (res_ex)
 477                                *rexp = res_ex;
 478                        cond_resched();
 479                }
 480                wait->_qproc = NULL;
 481                if (retval || timed_out || signal_pending(current))
 482                        break;
 483                if (table.error) {
 484                        retval = table.error;
 485                        break;
 486                }
 487
 488                /*
 489                 * If this is the first loop and we have a timeout
 490                 * given, then we convert to ktime_t and set the to
 491                 * pointer to the expiry value.
 492                 */
 493                if (end_time && !to) {
 494                        expire = timespec_to_ktime(*end_time);
 495                        to = &expire;
 496                }
 497
 498                if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
 499                                           to, slack))
 500                        timed_out = 1;
 501        }
 502
 503        poll_freewait(&table);
 504
 505        return retval;
 506}
 507
 508/*
 509 * We can actually return ERESTARTSYS instead of EINTR, but I'd
 510 * like to be certain this leads to no problems. So I return
 511 * EINTR just for safety.
 512 *
 513 * Update: ERESTARTSYS breaks at least the xview clock binary, so
 514 * I'm trying ERESTARTNOHAND which restart only when you want to.
 515 */
 516int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 517                           fd_set __user *exp, struct timespec *end_time)
 518{
 519        fd_set_bits fds;
 520        void *bits;
 521        int ret, max_fds;
 522        unsigned int size;
 523        struct fdtable *fdt;
 524        /* Allocate small arguments on the stack to save memory and be faster */
 525        long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 526
 527        ret = -EINVAL;
 528        if (n < 0)
 529                goto out_nofds;
 530
 531        /* max_fds can increase, so grab it once to avoid race */
 532        rcu_read_lock();
 533        fdt = files_fdtable(current->files);
 534        max_fds = fdt->max_fds;
 535        rcu_read_unlock();
 536        if (n > max_fds)
 537                n = max_fds;
 538
 539        /*
 540         * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
 541         * since we used fdset we need to allocate memory in units of
 542         * long-words. 
 543         */
 544        size = FDS_BYTES(n);
 545        bits = stack_fds;
 546        if (size > sizeof(stack_fds) / 6) {
 547                /* Not enough space in on-stack array; must use kmalloc */
 548                ret = -ENOMEM;
 549                bits = kmalloc(6 * size, GFP_KERNEL);
 550                if (!bits)
 551                        goto out_nofds;
 552        }
 553        fds.in      = bits;
 554        fds.out     = bits +   size;
 555        fds.ex      = bits + 2*size;
 556        fds.res_in  = bits + 3*size;
 557        fds.res_out = bits + 4*size;
 558        fds.res_ex  = bits + 5*size;
 559
 560        if ((ret = get_fd_set(n, inp, fds.in)) ||
 561            (ret = get_fd_set(n, outp, fds.out)) ||
 562            (ret = get_fd_set(n, exp, fds.ex)))
 563                goto out;
 564        zero_fd_set(n, fds.res_in);
 565        zero_fd_set(n, fds.res_out);
 566        zero_fd_set(n, fds.res_ex);
 567
 568        ret = do_select(n, &fds, end_time);
 569
 570        if (ret < 0)
 571                goto out;
 572        if (!ret) {
 573                ret = -ERESTARTNOHAND;
 574                if (signal_pending(current))
 575                        goto out;
 576                ret = 0;
 577        }
 578
 579        if (set_fd_set(n, inp, fds.res_in) ||
 580            set_fd_set(n, outp, fds.res_out) ||
 581            set_fd_set(n, exp, fds.res_ex))
 582                ret = -EFAULT;
 583
 584out:
 585        if (bits != stack_fds)
 586                kfree(bits);
 587out_nofds:
 588        return ret;
 589}
 590
 591SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp,
 592                fd_set __user *, exp, struct timeval __user *, tvp)
 593{
 594        struct timespec end_time, *to = NULL;
 595        struct timeval tv;
 596        int ret;
 597
 598        if (tvp) {
 599                if (copy_from_user(&tv, tvp, sizeof(tv)))
 600                        return -EFAULT;
 601
 602                to = &end_time;
 603                if (poll_select_set_timeout(to,
 604                                tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
 605                                (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
 606                        return -EINVAL;
 607        }
 608
 609        ret = core_sys_select(n, inp, outp, exp, to);
 610        ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
 611
 612        return ret;
 613}
 614
 615static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp,
 616                       fd_set __user *exp, struct timespec __user *tsp,
 617                       const sigset_t __user *sigmask, size_t sigsetsize)
 618{
 619        sigset_t ksigmask, sigsaved;
 620        struct timespec ts, end_time, *to = NULL;
 621        int ret;
 622
 623        if (tsp) {
 624                if (copy_from_user(&ts, tsp, sizeof(ts)))
 625                        return -EFAULT;
 626
 627                to = &end_time;
 628                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 629                        return -EINVAL;
 630        }
 631
 632        if (sigmask) {
 633                /* XXX: Don't preclude handling different sized sigset_t's.  */
 634                if (sigsetsize != sizeof(sigset_t))
 635                        return -EINVAL;
 636                if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 637                        return -EFAULT;
 638
 639                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 640                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 641        }
 642
 643        ret = core_sys_select(n, inp, outp, exp, to);
 644        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
 645
 646        if (ret == -ERESTARTNOHAND) {
 647                /*
 648                 * Don't restore the signal mask yet. Let do_signal() deliver
 649                 * the signal on the way back to userspace, before the signal
 650                 * mask is restored.
 651                 */
 652                if (sigmask) {
 653                        memcpy(&current->saved_sigmask, &sigsaved,
 654                                        sizeof(sigsaved));
 655                        set_restore_sigmask();
 656                }
 657        } else if (sigmask)
 658                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 659
 660        return ret;
 661}
 662
 663/*
 664 * Most architectures can't handle 7-argument syscalls. So we provide a
 665 * 6-argument version where the sixth argument is a pointer to a structure
 666 * which has a pointer to the sigset_t itself followed by a size_t containing
 667 * the sigset size.
 668 */
 669SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp,
 670                fd_set __user *, exp, struct timespec __user *, tsp,
 671                void __user *, sig)
 672{
 673        size_t sigsetsize = 0;
 674        sigset_t __user *up = NULL;
 675
 676        if (sig) {
 677                if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
 678                    || __get_user(up, (sigset_t __user * __user *)sig)
 679                    || __get_user(sigsetsize,
 680                                (size_t __user *)(sig+sizeof(void *))))
 681                        return -EFAULT;
 682        }
 683
 684        return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize);
 685}
 686
 687#ifdef __ARCH_WANT_SYS_OLD_SELECT
 688struct sel_arg_struct {
 689        unsigned long n;
 690        fd_set __user *inp, *outp, *exp;
 691        struct timeval __user *tvp;
 692};
 693
 694SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg)
 695{
 696        struct sel_arg_struct a;
 697
 698        if (copy_from_user(&a, arg, sizeof(a)))
 699                return -EFAULT;
 700        return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
 701}
 702#endif
 703
 704struct poll_list {
 705        struct poll_list *next;
 706        int len;
 707        struct pollfd entries[0];
 708};
 709
 710#define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 711
 712/*
 713 * Fish for pollable events on the pollfd->fd file descriptor. We're only
 714 * interested in events matching the pollfd->events mask, and the result
 715 * matching that mask is both recorded in pollfd->revents and returned. The
 716 * pwait poll_table will be used by the fd-provided poll handler for waiting,
 717 * if pwait->_qproc is non-NULL.
 718 */
 719static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
 720{
 721        unsigned int mask;
 722        int fd;
 723
 724        mask = 0;
 725        fd = pollfd->fd;
 726        if (fd >= 0) {
 727                struct fd f = fdget(fd);
 728                mask = POLLNVAL;
 729                if (f.file) {
 730                        mask = DEFAULT_POLLMASK;
 731                        if (f.file->f_op && f.file->f_op->poll) {
 732                                pwait->_key = pollfd->events|POLLERR|POLLHUP;
 733                                mask = f.file->f_op->poll(f.file, pwait);
 734                        }
 735                        /* Mask out unneeded events. */
 736                        mask &= pollfd->events | POLLERR | POLLHUP;
 737                        fdput(f);
 738                }
 739        }
 740        pollfd->revents = mask;
 741
 742        return mask;
 743}
 744
 745static int do_poll(unsigned int nfds,  struct poll_list *list,
 746                   struct poll_wqueues *wait, struct timespec *end_time)
 747{
 748        poll_table* pt = &wait->pt;
 749        ktime_t expire, *to = NULL;
 750        int timed_out = 0, count = 0;
 751        unsigned long slack = 0;
 752
 753        /* Optimise the no-wait case */
 754        if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 755                pt->_qproc = NULL;
 756                timed_out = 1;
 757        }
 758
 759        if (end_time && !timed_out)
 760                slack = select_estimate_accuracy(end_time);
 761
 762        for (;;) {
 763                struct poll_list *walk;
 764
 765                for (walk = list; walk != NULL; walk = walk->next) {
 766                        struct pollfd * pfd, * pfd_end;
 767
 768                        pfd = walk->entries;
 769                        pfd_end = pfd + walk->len;
 770                        for (; pfd != pfd_end; pfd++) {
 771                                /*
 772                                 * Fish for events. If we found one, record it
 773                                 * and kill poll_table->_qproc, so we don't
 774                                 * needlessly register any other waiters after
 775                                 * this. They'll get immediately deregistered
 776                                 * when we break out and return.
 777                                 */
 778                                if (do_pollfd(pfd, pt)) {
 779                                        count++;
 780                                        pt->_qproc = NULL;
 781                                }
 782                        }
 783                }
 784                /*
 785                 * All waiters have already been registered, so don't provide
 786                 * a poll_table->_qproc to them on the next loop iteration.
 787                 */
 788                pt->_qproc = NULL;
 789                if (!count) {
 790                        count = wait->error;
 791                        if (signal_pending(current))
 792                                count = -EINTR;
 793                }
 794                if (count || timed_out)
 795                        break;
 796
 797                /*
 798                 * If this is the first loop and we have a timeout
 799                 * given, then we convert to ktime_t and set the to
 800                 * pointer to the expiry value.
 801                 */
 802                if (end_time && !to) {
 803                        expire = timespec_to_ktime(*end_time);
 804                        to = &expire;
 805                }
 806
 807                if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
 808                        timed_out = 1;
 809        }
 810        return count;
 811}
 812
 813#define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 814                        sizeof(struct pollfd))
 815
 816int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 817                struct timespec *end_time)
 818{
 819        struct poll_wqueues table;
 820        int err = -EFAULT, fdcount, len, size;
 821        /* Allocate small arguments on the stack to save memory and be
 822           faster - use long to make sure the buffer is aligned properly
 823           on 64 bit archs to avoid unaligned access */
 824        long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 825        struct poll_list *const head = (struct poll_list *)stack_pps;
 826        struct poll_list *walk = head;
 827        unsigned long todo = nfds;
 828
 829        if (nfds > rlimit(RLIMIT_NOFILE))
 830                return -EINVAL;
 831
 832        len = min_t(unsigned int, nfds, N_STACK_PPS);
 833        for (;;) {
 834                walk->next = NULL;
 835                walk->len = len;
 836                if (!len)
 837                        break;
 838
 839                if (copy_from_user(walk->entries, ufds + nfds-todo,
 840                                        sizeof(struct pollfd) * walk->len))
 841                        goto out_fds;
 842
 843                todo -= walk->len;
 844                if (!todo)
 845                        break;
 846
 847                len = min(todo, POLLFD_PER_PAGE);
 848                size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
 849                walk = walk->next = kmalloc(size, GFP_KERNEL);
 850                if (!walk) {
 851                        err = -ENOMEM;
 852                        goto out_fds;
 853                }
 854        }
 855
 856        poll_initwait(&table);
 857        fdcount = do_poll(nfds, head, &table, end_time);
 858        poll_freewait(&table);
 859
 860        for (walk = head; walk; walk = walk->next) {
 861                struct pollfd *fds = walk->entries;
 862                int j;
 863
 864                for (j = 0; j < walk->len; j++, ufds++)
 865                        if (__put_user(fds[j].revents, &ufds->revents))
 866                                goto out_fds;
 867        }
 868
 869        err = fdcount;
 870out_fds:
 871        walk = head->next;
 872        while (walk) {
 873                struct poll_list *pos = walk;
 874                walk = walk->next;
 875                kfree(pos);
 876        }
 877
 878        return err;
 879}
 880
 881static long do_restart_poll(struct restart_block *restart_block)
 882{
 883        struct pollfd __user *ufds = restart_block->poll.ufds;
 884        int nfds = restart_block->poll.nfds;
 885        struct timespec *to = NULL, end_time;
 886        int ret;
 887
 888        if (restart_block->poll.has_timeout) {
 889                end_time.tv_sec = restart_block->poll.tv_sec;
 890                end_time.tv_nsec = restart_block->poll.tv_nsec;
 891                to = &end_time;
 892        }
 893
 894        ret = do_sys_poll(ufds, nfds, to);
 895
 896        if (ret == -EINTR) {
 897                restart_block->fn = do_restart_poll;
 898                ret = -ERESTART_RESTARTBLOCK;
 899        }
 900        return ret;
 901}
 902
 903SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds,
 904                int, timeout_msecs)
 905{
 906        struct timespec end_time, *to = NULL;
 907        int ret;
 908
 909        if (timeout_msecs >= 0) {
 910                to = &end_time;
 911                poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
 912                        NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
 913        }
 914
 915        ret = do_sys_poll(ufds, nfds, to);
 916
 917        if (ret == -EINTR) {
 918                struct restart_block *restart_block;
 919
 920                restart_block = &current_thread_info()->restart_block;
 921                restart_block->fn = do_restart_poll;
 922                restart_block->poll.ufds = ufds;
 923                restart_block->poll.nfds = nfds;
 924
 925                if (timeout_msecs >= 0) {
 926                        restart_block->poll.tv_sec = end_time.tv_sec;
 927                        restart_block->poll.tv_nsec = end_time.tv_nsec;
 928                        restart_block->poll.has_timeout = 1;
 929                } else
 930                        restart_block->poll.has_timeout = 0;
 931
 932                ret = -ERESTART_RESTARTBLOCK;
 933        }
 934        return ret;
 935}
 936
 937SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds,
 938                struct timespec __user *, tsp, const sigset_t __user *, sigmask,
 939                size_t, sigsetsize)
 940{
 941        sigset_t ksigmask, sigsaved;
 942        struct timespec ts, end_time, *to = NULL;
 943        int ret;
 944
 945        if (tsp) {
 946                if (copy_from_user(&ts, tsp, sizeof(ts)))
 947                        return -EFAULT;
 948
 949                to = &end_time;
 950                if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 951                        return -EINVAL;
 952        }
 953
 954        if (sigmask) {
 955                /* XXX: Don't preclude handling different sized sigset_t's.  */
 956                if (sigsetsize != sizeof(sigset_t))
 957                        return -EINVAL;
 958                if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 959                        return -EFAULT;
 960
 961                sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 962                sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 963        }
 964
 965        ret = do_sys_poll(ufds, nfds, to);
 966
 967        /* We can restart this syscall, usually */
 968        if (ret == -EINTR) {
 969                /*
 970                 * Don't restore the signal mask yet. Let do_signal() deliver
 971                 * the signal on the way back to userspace, before the signal
 972                 * mask is restored.
 973                 */
 974                if (sigmask) {
 975                        memcpy(&current->saved_sigmask, &sigsaved,
 976                                        sizeof(sigsaved));
 977                        set_restore_sigmask();
 978                }
 979                ret = -ERESTARTNOHAND;
 980        } else if (sigmask)
 981                sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 982
 983        ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
 984
 985        return ret;
 986}
 987