fs/select.c

   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/syscalls.h>
  19 #include <linux/module.h>
  20 #include <linux/slab.h>
  21 #include <linux/poll.h>
  22 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
  23 #include <linux/file.h>
  24 #include <linux/fdtable.h>
  25 #include <linux/fs.h>
  26 #include <linux/rcupdate.h>
  27 #include <linux/hrtimer.h>
  28
  29 #include <asm/uaccess.h>
  30
  31
  32 /*
  33  * Estimate expected accuracy in ns from a timeval.
  34  *
  35  * After quite a bit of churning around, we've settled on
  36  * a simple thing of taking 0.1% of the timeout as the
  37  * slack, with a cap of 100 msec.
  38  * "nice" tasks get a 0.5% slack instead.
  39  *
  40  * Consider this comment an open invitation to come up with even
  41  * better solutions..
  42  */
  43
  44 static long __estimate_accuracy(struct timespec *tv)
  45 {
  46         long slack;
  47         int divfactor = 1000;
  48
  49         if (task_nice(current) > 0)
  50                 divfactor = divfactor / 5;
  51
  52         slack = tv->tv_nsec / divfactor;
  53         slack += tv->tv_sec * (NSEC_PER_SEC/divfactor);
  54
  55         if (slack > 100 * NSEC_PER_MSEC)
  56                 slack =  100 * NSEC_PER_MSEC;
  57
  58         if (slack < 0)
  59                 slack = 0;
  60         return slack;
  61 }
  62
  63 static long estimate_accuracy(struct timespec *tv)
  64 {
  65         unsigned long ret;
  66         struct timespec now;
  67
  68         /*
  69          * Realtime tasks get a slack of 0 for obvious reasons.
  70          */
  71
  72         if (rt_task(current))
  73                 return 0;
  74
  75         ktime_get_ts(&now);
  76         now = timespec_sub(*tv, now);
  77         ret = __estimate_accuracy(&now);
  78         if (ret < current->timer_slack_ns)
  79                 return current->timer_slack_ns;
  80         return ret;
  81 }
  82
  83
  84
  85 struct poll_table_page {
  86         struct poll_table_page * next;
  87         struct poll_table_entry * entry;
  88         struct poll_table_entry entries[0];
  89 };
  90
  91 #define POLL_TABLE_FULL(table) \
  92         ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
  93
  94 /*
  95  * Ok, Peter made a complicated, but straightforward multiple_wait() function.
  96  * I have rewritten this, taking some shortcuts: This code may not be easy to
  97  * follow, but it should be free of race-conditions, and it's practical. If you
  98  * understand what I'm doing here, then you understand how the linux
  99  * sleep/wakeup mechanism works.
 100  *
 101  * Two very simple procedures, poll_wait() and poll_freewait() make all the
 102  * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
 103  * as all select/poll functions have to call it to add an entry to the
 104  * poll table.
 105  */
 106 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 107                        poll_table *p);
 108
 109 void poll_initwait(struct poll_wqueues *pwq)
 110 {
 111         init_poll_funcptr(&pwq->pt, __pollwait);
 112         pwq->polling_task = current;
 113         pwq->error = 0;
 114         pwq->table = NULL;
 115         pwq->inline_index = 0;
 116 }
 117 EXPORT_SYMBOL(poll_initwait);
 118
 119 static void free_poll_entry(struct poll_table_entry *entry)
 120 {
 121         remove_wait_queue(entry->wait_address, &entry->wait);
 122         fput(entry->filp);
 123 }
 124
 125 void poll_freewait(struct poll_wqueues *pwq)
 126 {
 127         struct poll_table_page * p = pwq->table;
 128         int i;
 129         for (i = 0; i < pwq->inline_index; i++)
 130                 free_poll_entry(pwq->inline_entries + i);
 131         while (p) {
 132                 struct poll_table_entry * entry;
 133                 struct poll_table_page *old;
 134
 135                 entry = p->entry;
 136                 do {
 137                         entry--;
 138                         free_poll_entry(entry);
 139                 } while (entry > p->entries);
 140                 old = p;
 141                 p = p->next;
 142                 free_page((unsigned long) old);
 143         }
 144 }
 145 EXPORT_SYMBOL(poll_freewait);
 146
 147 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p)
 148 {
 149         struct poll_table_page *table = p->table;
 150
 151         if (p->inline_index < N_INLINE_POLL_ENTRIES)
 152                 return p->inline_entries + p->inline_index++;
 153
 154         if (!table || POLL_TABLE_FULL(table)) {
 155                 struct poll_table_page *new_table;
 156
 157                 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 158                 if (!new_table) {
 159                         p->error = -ENOMEM;
 160                         return NULL;
 161                 }
 162                 new_table->entry = new_table->entries;
 163                 new_table->next = table;
 164                 p->table = new_table;
 165                 table = new_table;
 166         }
 167
 168         return table->entry++;
 169 }
 170
 171 static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key)
 172 {
 173         struct poll_wqueues *pwq = wait->private;
 174         DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task);
 175
 176         /*
 177          * Although this function is called under waitqueue lock, LOCK
 178          * doesn't imply write barrier and the users expect write
 179          * barrier semantics on wakeup functions.  The following
 180          * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up()
 181          * and is paired with set_mb() in poll_schedule_timeout.
 182          */
 183         smp_wmb();
 184         pwq->triggered = 1;
 185
 186         /*
 187          * Perform the default wake up operation using a dummy
 188          * waitqueue.
 189          *
 190          * TODO: This is hacky but there currently is no interface to
 191          * pass in @sync.  @sync is scheduled to be removed and once
 192          * that happens, wake_up_process() can be used directly.
 193          */
 194         return default_wake_function(&dummy_wait, mode, sync, key);
 195 }
 196
 197 /* Add a new entry */
 198 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 199                                 poll_table *p)
 200 {
 201         struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt);
 202         struct poll_table_entry *entry = poll_get_entry(pwq);
 203         if (!entry)
 204                 return;
 205         get_file(filp);
 206         entry->filp = filp;
 207         entry->wait_address = wait_address;
 208         init_waitqueue_func_entry(&entry->wait, pollwake);
 209         entry->wait.private = pwq;
 210         add_wait_queue(wait_address, &entry->wait);
 211 }
 212
 213 int poll_schedule_timeout(struct poll_wqueues *pwq, int state,
 214                           ktime_t *expires, unsigned long slack)
 215 {
 216         int rc = -EINTR;
 217
 218         set_current_state(state);
 219         if (!pwq->triggered)
 220                 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS);
 221         __set_current_state(TASK_RUNNING);
 222
 223         /*
 224          * Prepare for the next iteration.
 225          *
 226          * The following set_mb() serves two purposes.  First, it's
 227          * the counterpart rmb of the wmb in pollwake() such that data
 228          * written before wake up is always visible after wake up.
 229          * Second, the full barrier guarantees that triggered clearing
 230          * doesn't pass event check of the next iteration.  Note that
 231          * this problem doesn't exist for the first iteration as
 232          * add_wait_queue() has full barrier semantics.
 233          */
 234         set_mb(pwq->triggered, 0);
 235
 236         return rc;
 237 }
 238 EXPORT_SYMBOL(poll_schedule_timeout);
 239
 240 /**
 241  * poll_select_set_timeout - helper function to setup the timeout value
 242  * @to:         pointer to timespec variable for the final timeout
 243  * @sec:        seconds (from user space)
 244  * @nsec:       nanoseconds (from user space)
 245  *
 246  * Note, we do not use a timespec for the user space value here, That
 247  * way we can use the function for timeval and compat interfaces as well.
 248  *
 249  * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0.
 250  */
 251 int poll_select_set_timeout(struct timespec *to, long sec, long nsec)
 252 {
 253         struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec};
 254
 255         if (!timespec_valid(&ts))
 256                 return -EINVAL;
 257
 258         /* Optimize for the zero timeout value here */
 259         if (!sec && !nsec) {
 260                 to->tv_sec = to->tv_nsec = 0;
 261         } else {
 262                 ktime_get_ts(to);
 263                 *to = timespec_add_safe(*to, ts);
 264         }
 265         return 0;
 266 }
 267
 268 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p,
 269                                       int timeval, int ret)
 270 {
 271         struct timespec rts;
 272         struct timeval rtv;
 273
 274         if (!p)
 275                 return ret;
 276
 277         if (current->personality & STICKY_TIMEOUTS)
 278                 goto sticky;
 279
 280         /* No update for zero timeout */
 281         if (!end_time->tv_sec && !end_time->tv_nsec)
 282                 return ret;
 283
 284         ktime_get_ts(&rts);
 285         rts = timespec_sub(*end_time, rts);
 286         if (rts.tv_sec < 0)
 287                 rts.tv_sec = rts.tv_nsec = 0;
 288
 289         if (timeval) {
 290                 rtv.tv_sec = rts.tv_sec;
 291                 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC;
 292
 293                 if (!copy_to_user(p, &rtv, sizeof(rtv)))
 294                         return ret;
 295
 296         } else if (!copy_to_user(p, &rts, sizeof(rts)))
 297                 return ret;
 298
 299         /*
 300          * If an application puts its timeval in read-only memory, we
 301          * don't want the Linux-specific update to the timeval to
 302          * cause a fault after the select has completed
 303          * successfully. However, because we're not updating the
 304          * timeval, we can't restart the system call.
 305          */
 306
 307 sticky:
 308         if (ret == -ERESTARTNOHAND)
 309                 ret = -EINTR;
 310         return ret;
 311 }
 312
 313 #define FDS_IN(fds, n)          (fds->in + n)
 314 #define FDS_OUT(fds, n)         (fds->out + n)
 315 #define FDS_EX(fds, n)          (fds->ex + n)
 316
 317 #define BITS(fds, n)    (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 318
 319 static int max_select_fd(unsigned long n, fd_set_bits *fds)
 320 {
 321         unsigned long *open_fds;
 322         unsigned long set;
 323         int max;
 324         struct fdtable *fdt;
 325
 326         /* handle last in-complete long-word first */
 327         set = ~(~0UL << (n & (__NFDBITS-1)));
 328         n /= __NFDBITS;
 329         fdt = files_fdtable(current->files);
 330         open_fds = fdt->open_fds->fds_bits+n;
 331         max = 0;
 332         if (set) {
 333                 set &= BITS(fds, n);
 334                 if (set) {
 335                         if (!(set & ~*open_fds))
 336                                 goto get_max;
 337                         return -EBADF;
 338                 }
 339         }
 340         while (n) {
 341                 open_fds--;
 342                 n--;
 343                 set = BITS(fds, n);
 344                 if (!set)
 345                         continue;
 346                 if (set & ~*open_fds)
 347                         return -EBADF;
 348                 if (max)
 349                         continue;
 350 get_max:
 351                 do {
 352                         max++;
 353                         set >>= 1;
 354                 } while (set);
 355                 max += n * __NFDBITS;
 356         }
 357
 358         return max;
 359 }
 360
 361 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
 362 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
 363 #define POLLEX_SET (POLLPRI)
 364
 365 int do_select(int n, fd_set_bits *fds, struct timespec *end_time)
 366 {
 367         ktime_t expire, *to = NULL;
 368         struct poll_wqueues table;
 369         poll_table *wait;
 370         int retval, i, timed_out = 0;
 371         unsigned long slack = 0;
 372
 373         rcu_read_lock();
 374         retval = max_select_fd(n, fds);
 375         rcu_read_unlock();
 376
 377         if (retval < 0)
 378                 return retval;
 379         n = retval;
 380
 381         poll_initwait(&table);
 382         wait = &table.pt;
 383         if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 384                 wait = NULL;
 385                 timed_out = 1;
 386         }
 387
 388         if (end_time && !timed_out)
 389                 slack = estimate_accuracy(end_time);
 390
 391         retval = 0;
 392         for (;;) {
 393                 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 394
 395                 inp = fds->in; outp = fds->out; exp = fds->ex;
 396                 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 397
 398                 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 399                         unsigned long in, out, ex, all_bits, bit = 1, mask, j;
 400                         unsigned long res_in = 0, res_out = 0, res_ex = 0;
 401                         const struct file_operations *f_op = NULL;
 402                         struct file *file = NULL;
 403
 404                         in = *inp++; out = *outp++; ex = *exp++;
 405                         all_bits = in | out | ex;
 406                         if (all_bits == 0) {
 407                                 i += __NFDBITS;
 408                                 continue;
 409                         }
 410
 411                         for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
 412                                 int fput_needed;
 413                                 if (i >= n)
 414                                         break;
 415                                 if (!(bit & all_bits))
 416                                         continue;
 417                                 file = fget_light(i, &fput_needed);
 418                                 if (file) {
 419                                         f_op = file->f_op;
 420                                         mask = DEFAULT_POLLMASK;
 421                                         if (f_op && f_op->poll)
 422                                                 mask = (*f_op->poll)(file, retval ? NULL : wait);
 423                                         fput_light(file, fput_needed);
 424                                         if ((mask & POLLIN_SET) && (in & bit)) {
 425                                                 res_in |= bit;
 426                                                 retval++;
 427                                         }
 428                                         if ((mask & POLLOUT_SET) && (out & bit)) {
 429                                                 res_out |= bit;
 430                                                 retval++;
 431                                         }
 432                                         if ((mask & POLLEX_SET) && (ex & bit)) {
 433                                                 res_ex |= bit;
 434                                                 retval++;
 435                                         }
 436                                 }
 437                         }
 438                         if (res_in)
 439                                 *rinp = res_in;
 440                         if (res_out)
 441                                 *routp = res_out;
 442                         if (res_ex)
 443                                 *rexp = res_ex;
 444                         cond_resched();
 445                 }
 446                 wait = NULL;
 447                 if (retval || timed_out || signal_pending(current))
 448                         break;
 449                 if (table.error) {
 450                         retval = table.error;
 451                         break;
 452                 }
 453
 454                 /*
 455                  * If this is the first loop and we have a timeout
 456                  * given, then we convert to ktime_t and set the to
 457                  * pointer to the expiry value.
 458                  */
 459                 if (end_time && !to) {
 460                         expire = timespec_to_ktime(*end_time);
 461                         to = &expire;
 462                 }
 463
 464                 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE,
 465                                            to, slack))
 466                         timed_out = 1;
 467         }
 468
 469         poll_freewait(&table);
 470
 471         return retval;
 472 }
 473
 474 /*
 475  * We can actually return ERESTARTSYS instead of EINTR, but I'd
 476  * like to be certain this leads to no problems. So I return
 477  * EINTR just for safety.
 478  *
 479  * Update: ERESTARTSYS breaks at least the xview clock binary, so
 480  * I'm trying ERESTARTNOHAND which restart only when you want to.
 481  */
 482 #define MAX_SELECT_SECONDS \
 483         ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
 484
 485 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 486                            fd_set __user *exp, struct timespec *end_time)
 487 {
 488         fd_set_bits fds;
 489         void *bits;
 490         int ret, max_fds;
 491         unsigned int size;
 492         struct fdtable *fdt;
 493         /* Allocate small arguments on the stack to save memory and be faster */
 494         long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 495
 496         ret = -EINVAL;
 497         if (n < 0)
 498                 goto out_nofds;
 499
 500         /* max_fds can increase, so grab it once to avoid race */
 501         rcu_read_lock();
 502         fdt = files_fdtable(current->files);
 503         max_fds = fdt->max_fds;
 504         rcu_read_unlock();
 505         if (n > max_fds)
 506                 n = max_fds;
 507
 508         /*
 509          * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
 510          * since we used fdset we need to allocate memory in units of
 511          * long-words.
 512          */
 513         size = FDS_BYTES(n);
 514         bits = stack_fds;
 515         if (size > sizeof(stack_fds) / 6) {
 516                 /* Not enough space in on-stack array; must use kmalloc */
 517                 ret = -ENOMEM;
 518                 bits = kmalloc(6 * size, GFP_KERNEL);
 519                 if (!bits)
 520                         goto out_nofds;
 521         }
 522         fds.in      = bits;
 523         fds.out     = bits +   size;
 524         fds.ex      = bits + 2*size;
 525         fds.res_in  = bits + 3*size;
 526         fds.res_out = bits + 4*size;
 527         fds.res_ex  = bits + 5*size;
 528
 529         if ((ret = get_fd_set(n, inp, fds.in)) ||
 530             (ret = get_fd_set(n, outp, fds.out)) ||
 531             (ret = get_fd_set(n, exp, fds.ex)))
 532                 goto out;
 533         zero_fd_set(n, fds.res_in);
 534         zero_fd_set(n, fds.res_out);
 535         zero_fd_set(n, fds.res_ex);
 536
 537         ret = do_select(n, &fds, end_time);
 538
 539         if (ret < 0)
 540                 goto out;
 541         if (!ret) {
 542                 ret = -ERESTARTNOHAND;
 543                 if (signal_pending(current))
 544                         goto out;
 545                 ret = 0;
 546         }
 547
 548         if (set_fd_set(n, inp, fds.res_in) ||
 549             set_fd_set(n, outp, fds.res_out) ||
 550             set_fd_set(n, exp, fds.res_ex))
 551                 ret = -EFAULT;
 552
 553 out:
 554         if (bits != stack_fds)
 555                 kfree(bits);
 556 out_nofds:
 557         return ret;
 558 }
 559
 560 asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 561                         fd_set __user *exp, struct timeval __user *tvp)
 562 {
 563         struct timespec end_time, *to = NULL;
 564         struct timeval tv;
 565         int ret;
 566
 567         if (tvp) {
 568                 if (copy_from_user(&tv, tvp, sizeof(tv)))
 569                         return -EFAULT;
 570
 571                 to = &end_time;
 572                 if (poll_select_set_timeout(to,
 573                                 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC),
 574                                 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC))
 575                         return -EINVAL;
 576         }
 577
 578         ret = core_sys_select(n, inp, outp, exp, to);
 579         ret = poll_select_copy_remaining(&end_time, tvp, 1, ret);
 580
 581         return ret;
 582 }
 583
 584 #ifdef HAVE_SET_RESTORE_SIGMASK
 585 asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
 586                 fd_set __user *exp, struct timespec __user *tsp,
 587                 const sigset_t __user *sigmask, size_t sigsetsize)
 588 {
 589         sigset_t ksigmask, sigsaved;
 590         struct timespec ts, end_time, *to = NULL;
 591         int ret;
 592
 593         if (tsp) {
 594                 if (copy_from_user(&ts, tsp, sizeof(ts)))
 595                         return -EFAULT;
 596
 597                 to = &end_time;
 598                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 599                         return -EINVAL;
 600         }
 601
 602         if (sigmask) {
 603                 /* XXX: Don't preclude handling different sized sigset_t's.  */
 604                 if (sigsetsize != sizeof(sigset_t))
 605                         return -EINVAL;
 606                 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 607                         return -EFAULT;
 608
 609                 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 610                 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 611         }
 612
 613         ret = core_sys_select(n, inp, outp, exp, &end_time);
 614         ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
 615
 616         if (ret == -ERESTARTNOHAND) {
 617                 /*
 618                  * Don't restore the signal mask yet. Let do_signal() deliver
 619                  * the signal on the way back to userspace, before the signal
 620                  * mask is restored.
 621                  */
 622                 if (sigmask) {
 623                         memcpy(&current->saved_sigmask, &sigsaved,
 624                                         sizeof(sigsaved));
 625                         set_restore_sigmask();
 626                 }
 627         } else if (sigmask)
 628                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 629
 630         return ret;
 631 }
 632
 633 /*
 634  * Most architectures can't handle 7-argument syscalls. So we provide a
 635  * 6-argument version where the sixth argument is a pointer to a structure
 636  * which has a pointer to the sigset_t itself followed by a size_t containing
 637  * the sigset size.
 638  */
 639 asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
 640         fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
 641 {
 642         size_t sigsetsize = 0;
 643         sigset_t __user *up = NULL;
 644
 645         if (sig) {
 646                 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
 647                     || __get_user(up, (sigset_t __user * __user *)sig)
 648                     || __get_user(sigsetsize,
 649                                 (size_t __user *)(sig+sizeof(void *))))
 650                         return -EFAULT;
 651         }
 652
 653         return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
 654 }
 655 #endif /* HAVE_SET_RESTORE_SIGMASK */
 656
 657 struct poll_list {
 658         struct poll_list *next;
 659         int len;
 660         struct pollfd entries[0];
 661 };
 662
 663 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 664
 665 /*
 666  * Fish for pollable events on the pollfd->fd file descriptor. We're only
 667  * interested in events matching the pollfd->events mask, and the result
 668  * matching that mask is both recorded in pollfd->revents and returned. The
 669  * pwait poll_table will be used by the fd-provided poll handler for waiting,
 670  * if non-NULL.
 671  */
 672 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
 673 {
 674         unsigned int mask;
 675         int fd;
 676
 677         mask = 0;
 678         fd = pollfd->fd;
 679         if (fd >= 0) {
 680                 int fput_needed;
 681                 struct file * file;
 682
 683                 file = fget_light(fd, &fput_needed);
 684                 mask = POLLNVAL;
 685                 if (file != NULL) {
 686                         mask = DEFAULT_POLLMASK;
 687                         if (file->f_op && file->f_op->poll)
 688                                 mask = file->f_op->poll(file, pwait);
 689                         /* Mask out unneeded events. */
 690                         mask &= pollfd->events | POLLERR | POLLHUP;
 691                         fput_light(file, fput_needed);
 692                 }
 693         }
 694         pollfd->revents = mask;
 695
 696         return mask;
 697 }
 698
 699 static int do_poll(unsigned int nfds,  struct poll_list *list,
 700                    struct poll_wqueues *wait, struct timespec *end_time)
 701 {
 702         poll_table* pt = &wait->pt;
 703         ktime_t expire, *to = NULL;
 704         int timed_out = 0, count = 0;
 705         unsigned long slack = 0;
 706
 707         /* Optimise the no-wait case */
 708         if (end_time && !end_time->tv_sec && !end_time->tv_nsec) {
 709                 pt = NULL;
 710                 timed_out = 1;
 711         }
 712
 713         if (end_time && !timed_out)
 714                 slack = estimate_accuracy(end_time);
 715
 716         for (;;) {
 717                 struct poll_list *walk;
 718
 719                 for (walk = list; walk != NULL; walk = walk->next) {
 720                         struct pollfd * pfd, * pfd_end;
 721
 722                         pfd = walk->entries;
 723                         pfd_end = pfd + walk->len;
 724                         for (; pfd != pfd_end; pfd++) {
 725                                 /*
 726                                  * Fish for events. If we found one, record it
 727                                  * and kill the poll_table, so we don't
 728                                  * needlessly register any other waiters after
 729                                  * this. They'll get immediately deregistered
 730                                  * when we break out and return.
 731                                  */
 732                                 if (do_pollfd(pfd, pt)) {
 733                                         count++;
 734                                         pt = NULL;
 735                                 }
 736                         }
 737                 }
 738                 /*
 739                  * All waiters have already been registered, so don't provide
 740                  * a poll_table to them on the next loop iteration.
 741                  */
 742                 pt = NULL;
 743                 if (!count) {
 744                         count = wait->error;
 745                         if (signal_pending(current))
 746                                 count = -EINTR;
 747                 }
 748                 if (count || timed_out)
 749                         break;
 750
 751                 /*
 752                  * If this is the first loop and we have a timeout
 753                  * given, then we convert to ktime_t and set the to
 754                  * pointer to the expiry value.
 755                  */
 756                 if (end_time && !to) {
 757                         expire = timespec_to_ktime(*end_time);
 758                         to = &expire;
 759                 }
 760
 761                 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack))
 762                         timed_out = 1;
 763         }
 764         return count;
 765 }
 766
 767 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 768                         sizeof(struct pollfd))
 769
 770 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 771                 struct timespec *end_time)
 772 {
 773         struct poll_wqueues table;
 774         int err = -EFAULT, fdcount, len, size;
 775         /* Allocate small arguments on the stack to save memory and be
 776            faster - use long to make sure the buffer is aligned properly
 777            on 64 bit archs to avoid unaligned access */
 778         long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 779         struct poll_list *const head = (struct poll_list *)stack_pps;
 780         struct poll_list *walk = head;
 781         unsigned long todo = nfds;
 782
 783         if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 784                 return -EINVAL;
 785
 786         len = min_t(unsigned int, nfds, N_STACK_PPS);
 787         for (;;) {
 788                 walk->next = NULL;
 789                 walk->len = len;
 790                 if (!len)
 791                         break;
 792
 793                 if (copy_from_user(walk->entries, ufds + nfds-todo,
 794                                         sizeof(struct pollfd) * walk->len))
 795                         goto out_fds;
 796
 797                 todo -= walk->len;
 798                 if (!todo)
 799                         break;
 800
 801                 len = min(todo, POLLFD_PER_PAGE);
 802                 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len;
 803                 walk = walk->next = kmalloc(size, GFP_KERNEL);
 804                 if (!walk) {
 805                         err = -ENOMEM;
 806                         goto out_fds;
 807                 }
 808         }
 809
 810         poll_initwait(&table);
 811         fdcount = do_poll(nfds, head, &table, end_time);
 812         poll_freewait(&table);
 813
 814         for (walk = head; walk; walk = walk->next) {
 815                 struct pollfd *fds = walk->entries;
 816                 int j;
 817
 818                 for (j = 0; j < walk->len; j++, ufds++)
 819                         if (__put_user(fds[j].revents, &ufds->revents))
 820                                 goto out_fds;
 821         }
 822
 823         err = fdcount;
 824 out_fds:
 825         walk = head->next;
 826         while (walk) {
 827                 struct poll_list *pos = walk;
 828                 walk = walk->next;
 829                 kfree(pos);
 830         }
 831
 832         return err;
 833 }
 834
 835 static long do_restart_poll(struct restart_block *restart_block)
 836 {
 837         struct pollfd __user *ufds = restart_block->poll.ufds;
 838         int nfds = restart_block->poll.nfds;
 839         struct timespec *to = NULL, end_time;
 840         int ret;
 841
 842         if (restart_block->poll.has_timeout) {
 843                 end_time.tv_sec = restart_block->poll.tv_sec;
 844                 end_time.tv_nsec = restart_block->poll.tv_nsec;
 845                 to = &end_time;
 846         }
 847
 848         ret = do_sys_poll(ufds, nfds, to);
 849
 850         if (ret == -EINTR) {
 851                 restart_block->fn = do_restart_poll;
 852                 ret = -ERESTART_RESTARTBLOCK;
 853         }
 854         return ret;
 855 }
 856
 857 asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 858                         long timeout_msecs)
 859 {
 860         struct timespec end_time, *to = NULL;
 861         int ret;
 862
 863         if (timeout_msecs >= 0) {
 864                 to = &end_time;
 865                 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC,
 866                         NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC));
 867         }
 868
 869         ret = do_sys_poll(ufds, nfds, to);
 870
 871         if (ret == -EINTR) {
 872                 struct restart_block *restart_block;
 873
 874                 restart_block = &current_thread_info()->restart_block;
 875                 restart_block->fn = do_restart_poll;
 876                 restart_block->poll.ufds = ufds;
 877                 restart_block->poll.nfds = nfds;
 878
 879                 if (timeout_msecs >= 0) {
 880                         restart_block->poll.tv_sec = end_time.tv_sec;
 881                         restart_block->poll.tv_nsec = end_time.tv_nsec;
 882                         restart_block->poll.has_timeout = 1;
 883                 } else
 884                         restart_block->poll.has_timeout = 0;
 885
 886                 ret = -ERESTART_RESTARTBLOCK;
 887         }
 888         return ret;
 889 }
 890
 891 #ifdef HAVE_SET_RESTORE_SIGMASK
 892 asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
 893         struct timespec __user *tsp, const sigset_t __user *sigmask,
 894         size_t sigsetsize)
 895 {
 896         sigset_t ksigmask, sigsaved;
 897         struct timespec ts, end_time, *to = NULL;
 898         int ret;
 899
 900         if (tsp) {
 901                 if (copy_from_user(&ts, tsp, sizeof(ts)))
 902                         return -EFAULT;
 903
 904                 to = &end_time;
 905                 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec))
 906                         return -EINVAL;
 907         }
 908
 909         if (sigmask) {
 910                 /* XXX: Don't preclude handling different sized sigset_t's.  */
 911                 if (sigsetsize != sizeof(sigset_t))
 912                         return -EINVAL;
 913                 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 914                         return -EFAULT;
 915
 916                 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 917                 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 918         }
 919
 920         ret = do_sys_poll(ufds, nfds, to);
 921
 922         /* We can restart this syscall, usually */
 923         if (ret == -EINTR) {
 924                 /*
 925                  * Don't restore the signal mask yet. Let do_signal() deliver
 926                  * the signal on the way back to userspace, before the signal
 927                  * mask is restored.
 928                  */
 929                 if (sigmask) {
 930                         memcpy(&current->saved_sigmask, &sigsaved,
 931                                         sizeof(sigsaved));
 932                         set_restore_sigmask();
 933                 }
 934                 ret = -ERESTARTNOHAND;
 935         } else if (sigmask)
 936                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 937
 938         ret = poll_select_copy_remaining(&end_time, tsp, 0, ret);
 939
 940         return ret;
 941 }
 942 #endif /* HAVE_SET_RESTORE_SIGMASK */