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/syscalls.h>
  18 #include <linux/module.h>
  19 #include <linux/slab.h>
  20 #include <linux/smp_lock.h>
  21 #include <linux/poll.h>
  22 #include <linux/personality.h> /* for STICKY_TIMEOUTS */
  23 #include <linux/file.h>
  24 #include <linux/fs.h>
  25 #include <linux/rcupdate.h>
  26
  27 #include <asm/uaccess.h>
  28
  29 #define ROUND_UP(x,y) (((x)+(y)-1)/(y))
  30 #define DEFAULT_POLLMASK (POLLIN | POLLOUT | POLLRDNORM | POLLWRNORM)
  31
  32 struct poll_table_page {
  33         struct poll_table_page * next;
  34         struct poll_table_entry * entry;
  35         struct poll_table_entry entries[0];
  36 };
  37
  38 #define POLL_TABLE_FULL(table) \
  39         ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table))
  40
  41 /*
  42  * Ok, Peter made a complicated, but straightforward multiple_wait() function.
  43  * I have rewritten this, taking some shortcuts: This code may not be easy to
  44  * follow, but it should be free of race-conditions, and it's practical. If you
  45  * understand what I'm doing here, then you understand how the linux
  46  * sleep/wakeup mechanism works.
  47  *
  48  * Two very simple procedures, poll_wait() and poll_freewait() make all the
  49  * work.  poll_wait() is an inline-function defined in <linux/poll.h>,
  50  * as all select/poll functions have to call it to add an entry to the
  51  * poll table.
  52  */
  53 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
  54                        poll_table *p);
  55
  56 void poll_initwait(struct poll_wqueues *pwq)
  57 {
  58         init_poll_funcptr(&pwq->pt, __pollwait);
  59         pwq->error = 0;
  60         pwq->table = NULL;
  61         pwq->inline_index = 0;
  62 }
  63
  64 EXPORT_SYMBOL(poll_initwait);
  65
  66 static void free_poll_entry(struct poll_table_entry *entry)
  67 {
  68         remove_wait_queue(entry->wait_address,&entry->wait);
  69         fput(entry->filp);
  70 }
  71
  72 void poll_freewait(struct poll_wqueues *pwq)
  73 {
  74         struct poll_table_page * p = pwq->table;
  75         int i;
  76         for (i = 0; i < pwq->inline_index; i++)
  77                 free_poll_entry(pwq->inline_entries + i);
  78         while (p) {
  79                 struct poll_table_entry * entry;
  80                 struct poll_table_page *old;
  81
  82                 entry = p->entry;
  83                 do {
  84                         entry--;
  85                         free_poll_entry(entry);
  86                 } while (entry > p->entries);
  87                 old = p;
  88                 p = p->next;
  89                 free_page((unsigned long) old);
  90         }
  91 }
  92
  93 EXPORT_SYMBOL(poll_freewait);
  94
  95 static struct poll_table_entry *poll_get_entry(poll_table *_p)
  96 {
  97         struct poll_wqueues *p = container_of(_p, struct poll_wqueues, pt);
  98         struct poll_table_page *table = p->table;
  99
 100         if (p->inline_index < N_INLINE_POLL_ENTRIES)
 101                 return p->inline_entries + p->inline_index++;
 102
 103         if (!table || POLL_TABLE_FULL(table)) {
 104                 struct poll_table_page *new_table;
 105
 106                 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL);
 107                 if (!new_table) {
 108                         p->error = -ENOMEM;
 109                         __set_current_state(TASK_RUNNING);
 110                         return NULL;
 111                 }
 112                 new_table->entry = new_table->entries;
 113                 new_table->next = table;
 114                 p->table = new_table;
 115                 table = new_table;
 116         }
 117
 118         return table->entry++;
 119 }
 120
 121 /* Add a new entry */
 122 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address,
 123                                 poll_table *p)
 124 {
 125         struct poll_table_entry *entry = poll_get_entry(p);
 126         if (!entry)
 127                 return;
 128         get_file(filp);
 129         entry->filp = filp;
 130         entry->wait_address = wait_address;
 131         init_waitqueue_entry(&entry->wait, current);
 132         add_wait_queue(wait_address,&entry->wait);
 133 }
 134
 135 #define FDS_IN(fds, n)          (fds->in + n)
 136 #define FDS_OUT(fds, n)         (fds->out + n)
 137 #define FDS_EX(fds, n)          (fds->ex + n)
 138
 139 #define BITS(fds, n)    (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n))
 140
 141 static int max_select_fd(unsigned long n, fd_set_bits *fds)
 142 {
 143         unsigned long *open_fds;
 144         unsigned long set;
 145         int max;
 146         struct fdtable *fdt;
 147
 148         /* handle last in-complete long-word first */
 149         set = ~(~0UL << (n & (__NFDBITS-1)));
 150         n /= __NFDBITS;
 151         fdt = files_fdtable(current->files);
 152         open_fds = fdt->open_fds->fds_bits+n;
 153         max = 0;
 154         if (set) {
 155                 set &= BITS(fds, n);
 156                 if (set) {
 157                         if (!(set & ~*open_fds))
 158                                 goto get_max;
 159                         return -EBADF;
 160                 }
 161         }
 162         while (n) {
 163                 open_fds--;
 164                 n--;
 165                 set = BITS(fds, n);
 166                 if (!set)
 167                         continue;
 168                 if (set & ~*open_fds)
 169                         return -EBADF;
 170                 if (max)
 171                         continue;
 172 get_max:
 173                 do {
 174                         max++;
 175                         set >>= 1;
 176                 } while (set);
 177                 max += n * __NFDBITS;
 178         }
 179
 180         return max;
 181 }
 182
 183 #define BIT(i)          (1UL << ((i)&(__NFDBITS-1)))
 184 #define MEM(i,m)        ((m)+(unsigned)(i)/__NFDBITS)
 185 #define ISSET(i,m)      (((i)&*(m)) != 0)
 186 #define SET(i,m)        (*(m) |= (i))
 187
 188 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR)
 189 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
 190 #define POLLEX_SET (POLLPRI)
 191
 192 int do_select(int n, fd_set_bits *fds, s64 *timeout)
 193 {
 194         struct poll_wqueues table;
 195         poll_table *wait;
 196         int retval, i;
 197
 198         rcu_read_lock();
 199         retval = max_select_fd(n, fds);
 200         rcu_read_unlock();
 201
 202         if (retval < 0)
 203                 return retval;
 204         n = retval;
 205
 206         poll_initwait(&table);
 207         wait = &table.pt;
 208         if (!*timeout)
 209                 wait = NULL;
 210         retval = 0;
 211         for (;;) {
 212                 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
 213                 long __timeout;
 214
 215                 set_current_state(TASK_INTERRUPTIBLE);
 216
 217                 inp = fds->in; outp = fds->out; exp = fds->ex;
 218                 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex;
 219
 220                 for (i = 0; i < n; ++rinp, ++routp, ++rexp) {
 221                         unsigned long in, out, ex, all_bits, bit = 1, mask, j;
 222                         unsigned long res_in = 0, res_out = 0, res_ex = 0;
 223                         const struct file_operations *f_op = NULL;
 224                         struct file *file = NULL;
 225
 226                         in = *inp++; out = *outp++; ex = *exp++;
 227                         all_bits = in | out | ex;
 228                         if (all_bits == 0) {
 229                                 i += __NFDBITS;
 230                                 continue;
 231                         }
 232
 233                         for (j = 0; j < __NFDBITS; ++j, ++i, bit <<= 1) {
 234                                 int fput_needed;
 235                                 if (i >= n)
 236                                         break;
 237                                 if (!(bit & all_bits))
 238                                         continue;
 239                                 file = fget_light(i, &fput_needed);
 240                                 if (file) {
 241                                         f_op = file->f_op;
 242                                         mask = DEFAULT_POLLMASK;
 243                                         if (f_op && f_op->poll)
 244                                                 mask = (*f_op->poll)(file, retval ? NULL : wait);
 245                                         fput_light(file, fput_needed);
 246                                         if ((mask & POLLIN_SET) && (in & bit)) {
 247                                                 res_in |= bit;
 248                                                 retval++;
 249                                         }
 250                                         if ((mask & POLLOUT_SET) && (out & bit)) {
 251                                                 res_out |= bit;
 252                                                 retval++;
 253                                         }
 254                                         if ((mask & POLLEX_SET) && (ex & bit)) {
 255                                                 res_ex |= bit;
 256                                                 retval++;
 257                                         }
 258                                 }
 259                                 cond_resched();
 260                         }
 261                         if (res_in)
 262                                 *rinp = res_in;
 263                         if (res_out)
 264                                 *routp = res_out;
 265                         if (res_ex)
 266                                 *rexp = res_ex;
 267                 }
 268                 wait = NULL;
 269                 if (retval || !*timeout || signal_pending(current))
 270                         break;
 271                 if(table.error) {
 272                         retval = table.error;
 273                         break;
 274                 }
 275
 276                 if (*timeout < 0) {
 277                         /* Wait indefinitely */
 278                         __timeout = MAX_SCHEDULE_TIMEOUT;
 279                 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) {
 280                         /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */
 281                         __timeout = MAX_SCHEDULE_TIMEOUT - 1;
 282                         *timeout -= __timeout;
 283                 } else {
 284                         __timeout = *timeout;
 285                         *timeout = 0;
 286                 }
 287                 __timeout = schedule_timeout(__timeout);
 288                 if (*timeout >= 0)
 289                         *timeout += __timeout;
 290         }
 291         __set_current_state(TASK_RUNNING);
 292
 293         poll_freewait(&table);
 294
 295         return retval;
 296 }
 297
 298 /*
 299  * We can actually return ERESTARTSYS instead of EINTR, but I'd
 300  * like to be certain this leads to no problems. So I return
 301  * EINTR just for safety.
 302  *
 303  * Update: ERESTARTSYS breaks at least the xview clock binary, so
 304  * I'm trying ERESTARTNOHAND which restart only when you want to.
 305  */
 306 #define MAX_SELECT_SECONDS \
 307         ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
 308
 309 static int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 310                            fd_set __user *exp, s64 *timeout)
 311 {
 312         fd_set_bits fds;
 313         void *bits;
 314         int ret, max_fdset;
 315         unsigned int size;
 316         struct fdtable *fdt;
 317         /* Allocate small arguments on the stack to save memory and be faster */
 318         long stack_fds[SELECT_STACK_ALLOC/sizeof(long)];
 319
 320         ret = -EINVAL;
 321         if (n < 0)
 322                 goto out_nofds;
 323
 324         /* max_fdset can increase, so grab it once to avoid race */
 325         rcu_read_lock();
 326         fdt = files_fdtable(current->files);
 327         max_fdset = fdt->max_fdset;
 328         rcu_read_unlock();
 329         if (n > max_fdset)
 330                 n = max_fdset;
 331
 332         /*
 333          * We need 6 bitmaps (in/out/ex for both incoming and outgoing),
 334          * since we used fdset we need to allocate memory in units of
 335          * long-words.
 336          */
 337         size = FDS_BYTES(n);
 338         bits = stack_fds;
 339         if (size > sizeof(stack_fds) / 6) {
 340                 /* Not enough space in on-stack array; must use kmalloc */
 341                 ret = -ENOMEM;
 342                 bits = kmalloc(6 * size, GFP_KERNEL);
 343                 if (!bits)
 344                         goto out_nofds;
 345         }
 346         fds.in      = bits;
 347         fds.out     = bits +   size;
 348         fds.ex      = bits + 2*size;
 349         fds.res_in  = bits + 3*size;
 350         fds.res_out = bits + 4*size;
 351         fds.res_ex  = bits + 5*size;
 352
 353         if ((ret = get_fd_set(n, inp, fds.in)) ||
 354             (ret = get_fd_set(n, outp, fds.out)) ||
 355             (ret = get_fd_set(n, exp, fds.ex)))
 356                 goto out;
 357         zero_fd_set(n, fds.res_in);
 358         zero_fd_set(n, fds.res_out);
 359         zero_fd_set(n, fds.res_ex);
 360
 361         ret = do_select(n, &fds, timeout);
 362
 363         if (ret < 0)
 364                 goto out;
 365         if (!ret) {
 366                 ret = -ERESTARTNOHAND;
 367                 if (signal_pending(current))
 368                         goto out;
 369                 ret = 0;
 370         }
 371
 372         if (set_fd_set(n, inp, fds.res_in) ||
 373             set_fd_set(n, outp, fds.res_out) ||
 374             set_fd_set(n, exp, fds.res_ex))
 375                 ret = -EFAULT;
 376
 377 out:
 378         if (bits != stack_fds)
 379                 kfree(bits);
 380 out_nofds:
 381         return ret;
 382 }
 383
 384 asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
 385                         fd_set __user *exp, struct timeval __user *tvp)
 386 {
 387         s64 timeout = -1;
 388         struct timeval tv;
 389         int ret;
 390
 391         if (tvp) {
 392                 if (copy_from_user(&tv, tvp, sizeof(tv)))
 393                         return -EFAULT;
 394
 395                 if (tv.tv_sec < 0 || tv.tv_usec < 0)
 396                         return -EINVAL;
 397
 398                 /* Cast to u64 to make GCC stop complaining */
 399                 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
 400                         timeout = -1;   /* infinite */
 401                 else {
 402                         timeout = ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ);
 403                         timeout += tv.tv_sec * HZ;
 404                 }
 405         }
 406
 407         ret = core_sys_select(n, inp, outp, exp, &timeout);
 408
 409         if (tvp) {
 410                 struct timeval rtv;
 411
 412                 if (current->personality & STICKY_TIMEOUTS)
 413                         goto sticky;
 414                 rtv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
 415                 rtv.tv_sec = timeout;
 416                 if (timeval_compare(&rtv, &tv) >= 0)
 417                         rtv = tv;
 418                 if (copy_to_user(tvp, &rtv, sizeof(rtv))) {
 419 sticky:
 420                         /*
 421                          * If an application puts its timeval in read-only
 422                          * memory, we don't want the Linux-specific update to
 423                          * the timeval to cause a fault after the select has
 424                          * completed successfully. However, because we're not
 425                          * updating the timeval, we can't restart the system
 426                          * call.
 427                          */
 428                         if (ret == -ERESTARTNOHAND)
 429                                 ret = -EINTR;
 430                 }
 431         }
 432
 433         return ret;
 434 }
 435
 436 #ifdef TIF_RESTORE_SIGMASK
 437 asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
 438                 fd_set __user *exp, struct timespec __user *tsp,
 439                 const sigset_t __user *sigmask, size_t sigsetsize)
 440 {
 441         s64 timeout = MAX_SCHEDULE_TIMEOUT;
 442         sigset_t ksigmask, sigsaved;
 443         struct timespec ts;
 444         int ret;
 445
 446         if (tsp) {
 447                 if (copy_from_user(&ts, tsp, sizeof(ts)))
 448                         return -EFAULT;
 449
 450                 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
 451                         return -EINVAL;
 452
 453                 /* Cast to u64 to make GCC stop complaining */
 454                 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
 455                         timeout = -1;   /* infinite */
 456                 else {
 457                         timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
 458                         timeout += ts.tv_sec * HZ;
 459                 }
 460         }
 461
 462         if (sigmask) {
 463                 /* XXX: Don't preclude handling different sized sigset_t's.  */
 464                 if (sigsetsize != sizeof(sigset_t))
 465                         return -EINVAL;
 466                 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 467                         return -EFAULT;
 468
 469                 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 470                 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 471         }
 472
 473         ret = core_sys_select(n, inp, outp, exp, &timeout);
 474
 475         if (tsp) {
 476                 struct timespec rts;
 477
 478                 if (current->personality & STICKY_TIMEOUTS)
 479                         goto sticky;
 480                 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
 481                                                 1000;
 482                 rts.tv_sec = timeout;
 483                 if (timespec_compare(&rts, &ts) >= 0)
 484                         rts = ts;
 485                 if (copy_to_user(tsp, &rts, sizeof(rts))) {
 486 sticky:
 487                         /*
 488                          * If an application puts its timeval in read-only
 489                          * memory, we don't want the Linux-specific update to
 490                          * the timeval to cause a fault after the select has
 491                          * completed successfully. However, because we're not
 492                          * updating the timeval, we can't restart the system
 493                          * call.
 494                          */
 495                         if (ret == -ERESTARTNOHAND)
 496                                 ret = -EINTR;
 497                 }
 498         }
 499
 500         if (ret == -ERESTARTNOHAND) {
 501                 /*
 502                  * Don't restore the signal mask yet. Let do_signal() deliver
 503                  * the signal on the way back to userspace, before the signal
 504                  * mask is restored.
 505                  */
 506                 if (sigmask) {
 507                         memcpy(&current->saved_sigmask, &sigsaved,
 508                                         sizeof(sigsaved));
 509                         set_thread_flag(TIF_RESTORE_SIGMASK);
 510                 }
 511         } else if (sigmask)
 512                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 513
 514         return ret;
 515 }
 516
 517 /*
 518  * Most architectures can't handle 7-argument syscalls. So we provide a
 519  * 6-argument version where the sixth argument is a pointer to a structure
 520  * which has a pointer to the sigset_t itself followed by a size_t containing
 521  * the sigset size.
 522  */
 523 asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
 524         fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
 525 {
 526         size_t sigsetsize = 0;
 527         sigset_t __user *up = NULL;
 528
 529         if (sig) {
 530                 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
 531                     || __get_user(up, (sigset_t __user * __user *)sig)
 532                     || __get_user(sigsetsize,
 533                                 (size_t __user *)(sig+sizeof(void *))))
 534                         return -EFAULT;
 535         }
 536
 537         return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
 538 }
 539 #endif /* TIF_RESTORE_SIGMASK */
 540
 541 struct poll_list {
 542         struct poll_list *next;
 543         int len;
 544         struct pollfd entries[0];
 545 };
 546
 547 #define POLLFD_PER_PAGE  ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd))
 548
 549 /*
 550  * Fish for pollable events on the pollfd->fd file descriptor. We're only
 551  * interested in events matching the pollfd->events mask, and the result
 552  * matching that mask is both recorded in pollfd->revents and returned. The
 553  * pwait poll_table will be used by the fd-provided poll handler for waiting,
 554  * if non-NULL.
 555  */
 556 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait)
 557 {
 558         unsigned int mask;
 559         int fd;
 560
 561         mask = 0;
 562         fd = pollfd->fd;
 563         if (fd >= 0) {
 564                 int fput_needed;
 565                 struct file * file;
 566
 567                 file = fget_light(fd, &fput_needed);
 568                 mask = POLLNVAL;
 569                 if (file != NULL) {
 570                         mask = DEFAULT_POLLMASK;
 571                         if (file->f_op && file->f_op->poll)
 572                                 mask = file->f_op->poll(file, pwait);
 573                         /* Mask out unneeded events. */
 574                         mask &= pollfd->events | POLLERR | POLLHUP;
 575                         fput_light(file, fput_needed);
 576                 }
 577         }
 578         pollfd->revents = mask;
 579
 580         return mask;
 581 }
 582
 583 static int do_poll(unsigned int nfds,  struct poll_list *list,
 584                    struct poll_wqueues *wait, s64 *timeout)
 585 {
 586         int count = 0;
 587         poll_table* pt = &wait->pt;
 588
 589         /* Optimise the no-wait case */
 590         if (!(*timeout))
 591                 pt = NULL;
 592
 593         for (;;) {
 594                 struct poll_list *walk;
 595                 long __timeout;
 596
 597                 set_current_state(TASK_INTERRUPTIBLE);
 598                 for (walk = list; walk != NULL; walk = walk->next) {
 599                         struct pollfd * pfd, * pfd_end;
 600
 601                         pfd = walk->entries;
 602                         pfd_end = pfd + walk->len;
 603                         for (; pfd != pfd_end; pfd++) {
 604                                 /*
 605                                  * Fish for events. If we found one, record it
 606                                  * and kill the poll_table, so we don't
 607                                  * needlessly register any other waiters after
 608                                  * this. They'll get immediately deregistered
 609                                  * when we break out and return.
 610                                  */
 611                                 if (do_pollfd(pfd, pt)) {
 612                                         count++;
 613                                         pt = NULL;
 614                                 }
 615                         }
 616                 }
 617                 /*
 618                  * All waiters have already been registered, so don't provide
 619                  * a poll_table to them on the next loop iteration.
 620                  */
 621                 pt = NULL;
 622                 if (count || !*timeout || signal_pending(current))
 623                         break;
 624                 count = wait->error;
 625                 if (count)
 626                         break;
 627
 628                 if (*timeout < 0) {
 629                         /* Wait indefinitely */
 630                         __timeout = MAX_SCHEDULE_TIMEOUT;
 631                 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) {
 632                         /*
 633                          * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in
 634                          * a loop
 635                          */
 636                         __timeout = MAX_SCHEDULE_TIMEOUT - 1;
 637                         *timeout -= __timeout;
 638                 } else {
 639                         __timeout = *timeout;
 640                         *timeout = 0;
 641                 }
 642
 643                 __timeout = schedule_timeout(__timeout);
 644                 if (*timeout >= 0)
 645                         *timeout += __timeout;
 646         }
 647         __set_current_state(TASK_RUNNING);
 648         return count;
 649 }
 650
 651 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list))  / \
 652                         sizeof(struct pollfd))
 653
 654 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout)
 655 {
 656         struct poll_wqueues table;
 657         int fdcount, err;
 658         unsigned int i;
 659         struct poll_list *head;
 660         struct poll_list *walk;
 661         /* Allocate small arguments on the stack to save memory and be
 662            faster - use long to make sure the buffer is aligned properly
 663            on 64 bit archs to avoid unaligned access */
 664         long stack_pps[POLL_STACK_ALLOC/sizeof(long)];
 665         struct poll_list *stack_pp = NULL;
 666
 667         /* Do a sanity check on nfds ... */
 668         if (nfds > current->signal->rlim[RLIMIT_NOFILE].rlim_cur)
 669                 return -EINVAL;
 670
 671         poll_initwait(&table);
 672
 673         head = NULL;
 674         walk = NULL;
 675         i = nfds;
 676         err = -ENOMEM;
 677         while(i!=0) {
 678                 struct poll_list *pp;
 679                 int num, size;
 680                 if (stack_pp == NULL)
 681                         num = N_STACK_PPS;
 682                 else
 683                         num = POLLFD_PER_PAGE;
 684                 if (num > i)
 685                         num = i;
 686                 size = sizeof(struct poll_list) + sizeof(struct pollfd)*num;
 687                 if (!stack_pp)
 688                         stack_pp = pp = (struct poll_list *)stack_pps;
 689                 else {
 690                         pp = kmalloc(size, GFP_KERNEL);
 691                         if (!pp)
 692                                 goto out_fds;
 693                 }
 694                 pp->next=NULL;
 695                 pp->len = num;
 696                 if (head == NULL)
 697                         head = pp;
 698                 else
 699                         walk->next = pp;
 700
 701                 walk = pp;
 702                 if (copy_from_user(pp->entries, ufds + nfds-i,
 703                                 sizeof(struct pollfd)*num)) {
 704                         err = -EFAULT;
 705                         goto out_fds;
 706                 }
 707                 i -= pp->len;
 708         }
 709
 710         fdcount = do_poll(nfds, head, &table, timeout);
 711
 712         /* OK, now copy the revents fields back to user space. */
 713         walk = head;
 714         err = -EFAULT;
 715         while(walk != NULL) {
 716                 struct pollfd *fds = walk->entries;
 717                 int j;
 718
 719                 for (j=0; j < walk->len; j++, ufds++) {
 720                         if(__put_user(fds[j].revents, &ufds->revents))
 721                                 goto out_fds;
 722                 }
 723                 walk = walk->next;
 724         }
 725         err = fdcount;
 726         if (!fdcount && signal_pending(current))
 727                 err = -EINTR;
 728 out_fds:
 729         walk = head;
 730         while(walk!=NULL) {
 731                 struct poll_list *pp = walk->next;
 732                 if (walk != stack_pp)
 733                         kfree(walk);
 734                 walk = pp;
 735         }
 736         poll_freewait(&table);
 737         return err;
 738 }
 739
 740 asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
 741                         long timeout_msecs)
 742 {
 743         s64 timeout_jiffies;
 744
 745         if (timeout_msecs > 0) {
 746 #if HZ > 1000
 747                 /* We can only overflow if HZ > 1000 */
 748                 if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ)
 749                         timeout_jiffies = -1;
 750                 else
 751 #endif
 752                         timeout_jiffies = msecs_to_jiffies(timeout_msecs);
 753         } else {
 754                 /* Infinite (< 0) or no (0) timeout */
 755                 timeout_jiffies = timeout_msecs;
 756         }
 757
 758         return do_sys_poll(ufds, nfds, &timeout_jiffies);
 759 }
 760
 761 #ifdef TIF_RESTORE_SIGMASK
 762 asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
 763         struct timespec __user *tsp, const sigset_t __user *sigmask,
 764         size_t sigsetsize)
 765 {
 766         sigset_t ksigmask, sigsaved;
 767         struct timespec ts;
 768         s64 timeout = -1;
 769         int ret;
 770
 771         if (tsp) {
 772                 if (copy_from_user(&ts, tsp, sizeof(ts)))
 773                         return -EFAULT;
 774
 775                 /* Cast to u64 to make GCC stop complaining */
 776                 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
 777                         timeout = -1;   /* infinite */
 778                 else {
 779                         timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
 780                         timeout += ts.tv_sec * HZ;
 781                 }
 782         }
 783
 784         if (sigmask) {
 785                 /* XXX: Don't preclude handling different sized sigset_t's.  */
 786                 if (sigsetsize != sizeof(sigset_t))
 787                         return -EINVAL;
 788                 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
 789                         return -EFAULT;
 790
 791                 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
 792                 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
 793         }
 794
 795         ret = do_sys_poll(ufds, nfds, &timeout);
 796
 797         /* We can restart this syscall, usually */
 798         if (ret == -EINTR) {
 799                 /*
 800                  * Don't restore the signal mask yet. Let do_signal() deliver
 801                  * the signal on the way back to userspace, before the signal
 802                  * mask is restored.
 803                  */
 804                 if (sigmask) {
 805                         memcpy(&current->saved_sigmask, &sigsaved,
 806                                         sizeof(sigsaved));
 807                         set_thread_flag(TIF_RESTORE_SIGMASK);
 808                 }
 809                 ret = -ERESTARTNOHAND;
 810         } else if (sigmask)
 811                 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
 812
 813         if (tsp && timeout >= 0) {
 814                 struct timespec rts;
 815
 816                 if (current->personality & STICKY_TIMEOUTS)
 817                         goto sticky;
 818                 /* Yes, we know it's actually an s64, but it's also positive. */
 819                 rts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) *
 820                                                 1000;
 821                 rts.tv_sec = timeout;
 822                 if (timespec_compare(&rts, &ts) >= 0)
 823                         rts = ts;
 824                 if (copy_to_user(tsp, &rts, sizeof(rts))) {
 825                 sticky:
 826                         /*
 827                          * If an application puts its timeval in read-only
 828                          * memory, we don't want the Linux-specific update to
 829                          * the timeval to cause a fault after the select has
 830                          * completed successfully. However, because we're not
 831                          * updating the timeval, we can't restart the system
 832                          * call.
 833                          */
 834                         if (ret == -ERESTARTNOHAND && timeout >= 0)
 835                                 ret = -EINTR;
 836                 }
 837         }
 838
 839         return ret;
 840 }
 841 #endif /* TIF_RESTORE_SIGMASK */