aboutsummaryrefslogtreecommitdiffstats
path: root/fs
diff options
context:
space:
mode:
Diffstat (limited to 'fs')
-rw-r--r--fs/compat.c260
-rw-r--r--fs/select.c348
2 files changed, 514 insertions, 94 deletions
diff --git a/fs/compat.c b/fs/compat.c
index c6ba9deabada..18b21b4c9e3a 100644
--- a/fs/compat.c
+++ b/fs/compat.c
@@ -53,6 +53,8 @@
53#include <asm/mmu_context.h> 53#include <asm/mmu_context.h>
54#include <asm/ioctls.h> 54#include <asm/ioctls.h>
55 55
56extern void sigset_from_compat(sigset_t *set, compat_sigset_t *compat);
57
56/* 58/*
57 * Not all architectures have sys_utime, so implement this in terms 59 * Not all architectures have sys_utime, so implement this in terms
58 * of sys_utimes. 60 * of sys_utimes.
@@ -1657,36 +1659,14 @@ static void select_bits_free(void *bits, int size)
1657#define MAX_SELECT_SECONDS \ 1659#define MAX_SELECT_SECONDS \
1658 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) 1660 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
1659 1661
1660asmlinkage long 1662int compat_core_sys_select(int n, compat_ulong_t __user *inp,
1661compat_sys_select(int n, compat_ulong_t __user *inp, compat_ulong_t __user *outp, 1663 compat_ulong_t __user *outp, compat_ulong_t __user *exp, s64 *timeout)
1662 compat_ulong_t __user *exp, struct compat_timeval __user *tvp)
1663{ 1664{
1664 fd_set_bits fds; 1665 fd_set_bits fds;
1665 char *bits; 1666 char *bits;
1666 long timeout;
1667 int size, max_fdset, ret = -EINVAL; 1667 int size, max_fdset, ret = -EINVAL;
1668 struct fdtable *fdt; 1668 struct fdtable *fdt;
1669 1669
1670 timeout = MAX_SCHEDULE_TIMEOUT;
1671 if (tvp) {
1672 time_t sec, usec;
1673
1674 if (!access_ok(VERIFY_READ, tvp, sizeof(*tvp))
1675 || __get_user(sec, &tvp->tv_sec)
1676 || __get_user(usec, &tvp->tv_usec)) {
1677 ret = -EFAULT;
1678 goto out_nofds;
1679 }
1680
1681 if (sec < 0 || usec < 0)
1682 goto out_nofds;
1683
1684 if ((unsigned long) sec < MAX_SELECT_SECONDS) {
1685 timeout = ROUND_UP(usec, 1000000/HZ);
1686 timeout += sec * (unsigned long) HZ;
1687 }
1688 }
1689
1690 if (n < 0) 1670 if (n < 0)
1691 goto out_nofds; 1671 goto out_nofds;
1692 1672
@@ -1723,19 +1703,7 @@ compat_sys_select(int n, compat_ulong_t __user *inp, compat_ulong_t __user *outp
1723 zero_fd_set(n, fds.res_out); 1703 zero_fd_set(n, fds.res_out);
1724 zero_fd_set(n, fds.res_ex); 1704 zero_fd_set(n, fds.res_ex);
1725 1705
1726 ret = do_select(n, &fds, &timeout); 1706 ret = do_select(n, &fds, timeout);
1727
1728 if (tvp && !(current->personality & STICKY_TIMEOUTS)) {
1729 time_t sec = 0, usec = 0;
1730 if (timeout) {
1731 sec = timeout / HZ;
1732 usec = timeout % HZ;
1733 usec *= (1000000/HZ);
1734 }
1735 if (put_user(sec, &tvp->tv_sec) ||
1736 put_user(usec, &tvp->tv_usec))
1737 ret = -EFAULT;
1738 }
1739 1707
1740 if (ret < 0) 1708 if (ret < 0)
1741 goto out; 1709 goto out;
@@ -1756,6 +1724,224 @@ out_nofds:
1756 return ret; 1724 return ret;
1757} 1725}
1758 1726
1727asmlinkage long compat_sys_select(int n, compat_ulong_t __user *inp,
1728 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1729 struct compat_timeval __user *tvp)
1730{
1731 s64 timeout = -1;
1732 struct compat_timeval tv;
1733 int ret;
1734
1735 if (tvp) {
1736 if (copy_from_user(&tv, tvp, sizeof(tv)))
1737 return -EFAULT;
1738
1739 if (tv.tv_sec < 0 || tv.tv_usec < 0)
1740 return -EINVAL;
1741
1742 /* Cast to u64 to make GCC stop complaining */
1743 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
1744 timeout = -1; /* infinite */
1745 else {
1746 timeout = ROUND_UP(tv.tv_sec, 1000000/HZ);
1747 timeout += tv.tv_sec * HZ;
1748 }
1749 }
1750
1751 ret = compat_core_sys_select(n, inp, outp, exp, &timeout);
1752
1753 if (tvp) {
1754 if (current->personality & STICKY_TIMEOUTS)
1755 goto sticky;
1756 tv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
1757 tv.tv_sec = timeout;
1758 if (copy_to_user(tvp, &tv, sizeof(tv))) {
1759sticky:
1760 /*
1761 * If an application puts its timeval in read-only
1762 * memory, we don't want the Linux-specific update to
1763 * the timeval to cause a fault after the select has
1764 * completed successfully. However, because we're not
1765 * updating the timeval, we can't restart the system
1766 * call.
1767 */
1768 if (ret == -ERESTARTNOHAND)
1769 ret = -EINTR;
1770 }
1771 }
1772
1773 return ret;
1774}
1775
1776#ifdef TIF_RESTORE_SIGMASK
1777asmlinkage long compat_sys_pselect7(int n, compat_ulong_t __user *inp,
1778 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1779 struct compat_timespec __user *tsp, compat_sigset_t __user *sigmask,
1780 compat_size_t sigsetsize)
1781{
1782 compat_sigset_t ss32;
1783 sigset_t ksigmask, sigsaved;
1784 long timeout = MAX_SCHEDULE_TIMEOUT;
1785 struct compat_timespec ts;
1786 int ret;
1787
1788 if (tsp) {
1789 if (copy_from_user(&ts, tsp, sizeof(ts)))
1790 return -EFAULT;
1791
1792 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
1793 return -EINVAL;
1794 }
1795
1796 if (sigmask) {
1797 if (sigsetsize != sizeof(compat_sigset_t))
1798 return -EINVAL;
1799 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1800 return -EFAULT;
1801 sigset_from_compat(&ksigmask, &ss32);
1802
1803 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1804 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1805 }
1806
1807 do {
1808 if (tsp) {
1809 if ((unsigned long)ts.tv_sec < MAX_SELECT_SECONDS) {
1810 timeout = ROUND_UP(ts.tv_nsec, 1000000000/HZ);
1811 timeout += ts.tv_sec * (unsigned long)HZ;
1812 ts.tv_sec = 0;
1813 ts.tv_nsec = 0;
1814 } else {
1815 ts.tv_sec -= MAX_SELECT_SECONDS;
1816 timeout = MAX_SELECT_SECONDS * HZ;
1817 }
1818 }
1819
1820 ret = compat_core_sys_select(n, inp, outp, exp, &timeout);
1821
1822 } while (!ret && !timeout && tsp && (ts.tv_sec || ts.tv_nsec));
1823
1824 if (tsp && !(current->personality & STICKY_TIMEOUTS)) {
1825 ts.tv_sec += timeout / HZ;
1826 ts.tv_nsec += (timeout % HZ) * (1000000000/HZ);
1827 if (ts.tv_nsec >= 1000000000) {
1828 ts.tv_sec++;
1829 ts.tv_nsec -= 1000000000;
1830 }
1831 (void)copy_to_user(tsp, &ts, sizeof(ts));
1832 }
1833
1834 if (ret == -ERESTARTNOHAND) {
1835 /*
1836 * Don't restore the signal mask yet. Let do_signal() deliver
1837 * the signal on the way back to userspace, before the signal
1838 * mask is restored.
1839 */
1840 if (sigmask) {
1841 memcpy(&current->saved_sigmask, &sigsaved,
1842 sizeof(sigsaved));
1843 set_thread_flag(TIF_RESTORE_SIGMASK);
1844 }
1845 } else if (sigmask)
1846 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1847
1848 return ret;
1849}
1850
1851asmlinkage long compat_sys_pselect6(int n, compat_ulong_t __user *inp,
1852 compat_ulong_t __user *outp, compat_ulong_t __user *exp,
1853 struct compat_timespec __user *tsp, void __user *sig)
1854{
1855 compat_size_t sigsetsize = 0;
1856 compat_uptr_t up = 0;
1857
1858 if (sig) {
1859 if (!access_ok(VERIFY_READ, sig,
1860 sizeof(compat_uptr_t)+sizeof(compat_size_t)) ||
1861 __get_user(up, (compat_uptr_t __user *)sig) ||
1862 __get_user(sigsetsize,
1863 (compat_size_t __user *)(sig+sizeof(up))))
1864 return -EFAULT;
1865 }
1866 return compat_sys_pselect7(n, inp, outp, exp, tsp, compat_ptr(up),
1867 sigsetsize);
1868}
1869
1870asmlinkage long compat_sys_ppoll(struct pollfd __user *ufds,
1871 unsigned int nfds, struct compat_timespec __user *tsp,
1872 const compat_sigset_t __user *sigmask, compat_size_t sigsetsize)
1873{
1874 compat_sigset_t ss32;
1875 sigset_t ksigmask, sigsaved;
1876 struct compat_timespec ts;
1877 s64 timeout = -1;
1878 int ret;
1879
1880 if (tsp) {
1881 if (copy_from_user(&ts, tsp, sizeof(ts)))
1882 return -EFAULT;
1883
1884 /* We assume that ts.tv_sec is always lower than
1885 the number of seconds that can be expressed in
1886 an s64. Otherwise the compiler bitches at us */
1887 timeout = ROUND_UP(ts.tv_sec, 1000000000/HZ);
1888 timeout += ts.tv_sec * HZ;
1889 }
1890
1891 if (sigmask) {
1892 if (sigsetsize |= sizeof(compat_sigset_t))
1893 return -EINVAL;
1894 if (copy_from_user(&ss32, sigmask, sizeof(ss32)))
1895 return -EFAULT;
1896 sigset_from_compat(&ksigmask, &ss32);
1897
1898 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
1899 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
1900 }
1901
1902 ret = do_sys_poll(ufds, nfds, &timeout);
1903
1904 /* We can restart this syscall, usually */
1905 if (ret == -EINTR) {
1906 /*
1907 * Don't restore the signal mask yet. Let do_signal() deliver
1908 * the signal on the way back to userspace, before the signal
1909 * mask is restored.
1910 */
1911 if (sigmask) {
1912 memcpy(&current->saved_sigmask, &sigsaved,
1913 sizeof(sigsaved));
1914 set_thread_flag(TIF_RESTORE_SIGMASK);
1915 }
1916 ret = -ERESTARTNOHAND;
1917 } else if (sigmask)
1918 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
1919
1920 if (tsp && timeout >= 0) {
1921 if (current->personality & STICKY_TIMEOUTS)
1922 goto sticky;
1923 /* Yes, we know it's actually an s64, but it's also positive. */
1924 ts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) * 1000;
1925 ts.tv_sec = timeout;
1926 if (copy_to_user(tsp, &ts, sizeof(ts))) {
1927sticky:
1928 /*
1929 * If an application puts its timeval in read-only
1930 * memory, we don't want the Linux-specific update to
1931 * the timeval to cause a fault after the select has
1932 * completed successfully. However, because we're not
1933 * updating the timeval, we can't restart the system
1934 * call.
1935 */
1936 if (ret == -ERESTARTNOHAND && timeout >= 0)
1937 ret = -EINTR;
1938 }
1939 }
1940
1941 return ret;
1942}
1943#endif /* TIF_RESTORE_SIGMASK */
1944
1759#if defined(CONFIG_NFSD) || defined(CONFIG_NFSD_MODULE) 1945#if defined(CONFIG_NFSD) || defined(CONFIG_NFSD_MODULE)
1760/* Stuff for NFS server syscalls... */ 1946/* Stuff for NFS server syscalls... */
1761struct compat_nfsctl_svc { 1947struct compat_nfsctl_svc {
diff --git a/fs/select.c b/fs/select.c
index f10a10317d54..c0f02d36c60e 100644
--- a/fs/select.c
+++ b/fs/select.c
@@ -179,12 +179,11 @@ get_max:
179#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) 179#define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR)
180#define POLLEX_SET (POLLPRI) 180#define POLLEX_SET (POLLPRI)
181 181
182int do_select(int n, fd_set_bits *fds, long *timeout) 182int do_select(int n, fd_set_bits *fds, s64 *timeout)
183{ 183{
184 struct poll_wqueues table; 184 struct poll_wqueues table;
185 poll_table *wait; 185 poll_table *wait;
186 int retval, i; 186 int retval, i;
187 long __timeout = *timeout;
188 187
189 rcu_read_lock(); 188 rcu_read_lock();
190 retval = max_select_fd(n, fds); 189 retval = max_select_fd(n, fds);
@@ -196,11 +195,12 @@ int do_select(int n, fd_set_bits *fds, long *timeout)
196 195
197 poll_initwait(&table); 196 poll_initwait(&table);
198 wait = &table.pt; 197 wait = &table.pt;
199 if (!__timeout) 198 if (!*timeout)
200 wait = NULL; 199 wait = NULL;
201 retval = 0; 200 retval = 0;
202 for (;;) { 201 for (;;) {
203 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; 202 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp;
203 long __timeout;
204 204
205 set_current_state(TASK_INTERRUPTIBLE); 205 set_current_state(TASK_INTERRUPTIBLE);
206 206
@@ -255,22 +255,32 @@ int do_select(int n, fd_set_bits *fds, long *timeout)
255 *rexp = res_ex; 255 *rexp = res_ex;
256 } 256 }
257 wait = NULL; 257 wait = NULL;
258 if (retval || !__timeout || signal_pending(current)) 258 if (retval || !*timeout || signal_pending(current))
259 break; 259 break;
260 if(table.error) { 260 if(table.error) {
261 retval = table.error; 261 retval = table.error;
262 break; 262 break;
263 } 263 }
264
265 if (*timeout < 0) {
266 /* Wait indefinitely */
267 __timeout = MAX_SCHEDULE_TIMEOUT;
268 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT - 1)) {
269 /* Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in a loop */
270 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
271 *timeout -= __timeout;
272 } else {
273 __timeout = *timeout;
274 *timeout = 0;
275 }
264 __timeout = schedule_timeout(__timeout); 276 __timeout = schedule_timeout(__timeout);
277 if (*timeout >= 0)
278 *timeout += __timeout;
265 } 279 }
266 __set_current_state(TASK_RUNNING); 280 __set_current_state(TASK_RUNNING);
267 281
268 poll_freewait(&table); 282 poll_freewait(&table);
269 283
270 /*
271 * Up-to-date the caller timeout.
272 */
273 *timeout = __timeout;
274 return retval; 284 return retval;
275} 285}
276 286
@@ -295,36 +305,14 @@ static void select_bits_free(void *bits, int size)
295#define MAX_SELECT_SECONDS \ 305#define MAX_SELECT_SECONDS \
296 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1) 306 ((unsigned long) (MAX_SCHEDULE_TIMEOUT / HZ)-1)
297 307
298asmlinkage long 308static int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp,
299sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, struct timeval __user *tvp) 309 fd_set __user *exp, s64 *timeout)
300{ 310{
301 fd_set_bits fds; 311 fd_set_bits fds;
302 char *bits; 312 char *bits;
303 long timeout;
304 int ret, size, max_fdset; 313 int ret, size, max_fdset;
305 struct fdtable *fdt; 314 struct fdtable *fdt;
306 315
307 timeout = MAX_SCHEDULE_TIMEOUT;
308 if (tvp) {
309 time_t sec, usec;
310
311 if (!access_ok(VERIFY_READ, tvp, sizeof(*tvp))
312 || __get_user(sec, &tvp->tv_sec)
313 || __get_user(usec, &tvp->tv_usec)) {
314 ret = -EFAULT;
315 goto out_nofds;
316 }
317
318 ret = -EINVAL;
319 if (sec < 0 || usec < 0)
320 goto out_nofds;
321
322 if ((unsigned long) sec < MAX_SELECT_SECONDS) {
323 timeout = ROUND_UP(usec, 1000000/HZ);
324 timeout += sec * (unsigned long) HZ;
325 }
326 }
327
328 ret = -EINVAL; 316 ret = -EINVAL;
329 if (n < 0) 317 if (n < 0)
330 goto out_nofds; 318 goto out_nofds;
@@ -362,18 +350,7 @@ sys_select(int n, fd_set __user *inp, fd_set __user *outp, fd_set __user *exp, s
362 zero_fd_set(n, fds.res_out); 350 zero_fd_set(n, fds.res_out);
363 zero_fd_set(n, fds.res_ex); 351 zero_fd_set(n, fds.res_ex);
364 352
365 ret = do_select(n, &fds, &timeout); 353 ret = do_select(n, &fds, timeout);
366
367 if (tvp && !(current->personality & STICKY_TIMEOUTS)) {
368 time_t sec = 0, usec = 0;
369 if (timeout) {
370 sec = timeout / HZ;
371 usec = timeout % HZ;
372 usec *= (1000000/HZ);
373 }
374 put_user(sec, &tvp->tv_sec);
375 put_user(usec, &tvp->tv_usec);
376 }
377 354
378 if (ret < 0) 355 if (ret < 0)
379 goto out; 356 goto out;
@@ -395,6 +372,154 @@ out_nofds:
395 return ret; 372 return ret;
396} 373}
397 374
375asmlinkage long sys_select(int n, fd_set __user *inp, fd_set __user *outp,
376 fd_set __user *exp, struct timeval __user *tvp)
377{
378 s64 timeout = -1;
379 struct timeval tv;
380 int ret;
381
382 if (tvp) {
383 if (copy_from_user(&tv, tvp, sizeof(tv)))
384 return -EFAULT;
385
386 if (tv.tv_sec < 0 || tv.tv_usec < 0)
387 return -EINVAL;
388
389 /* Cast to u64 to make GCC stop complaining */
390 if ((u64)tv.tv_sec >= (u64)MAX_INT64_SECONDS)
391 timeout = -1; /* infinite */
392 else {
393 timeout = ROUND_UP(tv.tv_usec, USEC_PER_SEC/HZ);
394 timeout += tv.tv_sec * HZ;
395 }
396 }
397
398 ret = core_sys_select(n, inp, outp, exp, &timeout);
399
400 if (tvp) {
401 if (current->personality & STICKY_TIMEOUTS)
402 goto sticky;
403 tv.tv_usec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ));
404 tv.tv_sec = timeout;
405 if (copy_to_user(tvp, &tv, sizeof(tv))) {
406sticky:
407 /*
408 * If an application puts its timeval in read-only
409 * memory, we don't want the Linux-specific update to
410 * the timeval to cause a fault after the select has
411 * completed successfully. However, because we're not
412 * updating the timeval, we can't restart the system
413 * call.
414 */
415 if (ret == -ERESTARTNOHAND)
416 ret = -EINTR;
417 }
418 }
419
420 return ret;
421}
422
423#ifdef TIF_RESTORE_SIGMASK
424asmlinkage long sys_pselect7(int n, fd_set __user *inp, fd_set __user *outp,
425 fd_set __user *exp, struct timespec __user *tsp,
426 const sigset_t __user *sigmask, size_t sigsetsize)
427{
428 s64 timeout = MAX_SCHEDULE_TIMEOUT;
429 sigset_t ksigmask, sigsaved;
430 struct timespec ts;
431 int ret;
432
433 if (tsp) {
434 if (copy_from_user(&ts, tsp, sizeof(ts)))
435 return -EFAULT;
436
437 if (ts.tv_sec < 0 || ts.tv_nsec < 0)
438 return -EINVAL;
439
440 /* Cast to u64 to make GCC stop complaining */
441 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
442 timeout = -1; /* infinite */
443 else {
444 timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
445 timeout += ts.tv_sec * HZ;
446 }
447 }
448
449 if (sigmask) {
450 /* XXX: Don't preclude handling different sized sigset_t's. */
451 if (sigsetsize != sizeof(sigset_t))
452 return -EINVAL;
453 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
454 return -EFAULT;
455
456 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
457 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
458 }
459
460 ret = core_sys_select(n, inp, outp, exp, &timeout);
461
462 if (tsp) {
463 if (current->personality & STICKY_TIMEOUTS)
464 goto sticky;
465 ts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) * 1000;
466 ts.tv_sec = timeout;
467 if (copy_to_user(tsp, &ts, sizeof(ts))) {
468sticky:
469 /*
470 * If an application puts its timeval in read-only
471 * memory, we don't want the Linux-specific update to
472 * the timeval to cause a fault after the select has
473 * completed successfully. However, because we're not
474 * updating the timeval, we can't restart the system
475 * call.
476 */
477 if (ret == -ERESTARTNOHAND)
478 ret = -EINTR;
479 }
480 }
481
482 if (ret == -ERESTARTNOHAND) {
483 /*
484 * Don't restore the signal mask yet. Let do_signal() deliver
485 * the signal on the way back to userspace, before the signal
486 * mask is restored.
487 */
488 if (sigmask) {
489 memcpy(&current->saved_sigmask, &sigsaved,
490 sizeof(sigsaved));
491 set_thread_flag(TIF_RESTORE_SIGMASK);
492 }
493 } else if (sigmask)
494 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
495
496 return ret;
497}
498
499/*
500 * Most architectures can't handle 7-argument syscalls. So we provide a
501 * 6-argument version where the sixth argument is a pointer to a structure
502 * which has a pointer to the sigset_t itself followed by a size_t containing
503 * the sigset size.
504 */
505asmlinkage long sys_pselect6(int n, fd_set __user *inp, fd_set __user *outp,
506 fd_set __user *exp, struct timespec __user *tsp, void __user *sig)
507{
508 size_t sigsetsize = 0;
509 sigset_t __user *up = NULL;
510
511 if (sig) {
512 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t))
513 || __get_user(up, (sigset_t * __user *)sig)
514 || __get_user(sigsetsize,
515 (size_t * __user)(sig+sizeof(void *))))
516 return -EFAULT;
517 }
518
519 return sys_pselect7(n, inp, outp, exp, tsp, up, sigsetsize);
520}
521#endif /* TIF_RESTORE_SIGMASK */
522
398struct poll_list { 523struct poll_list {
399 struct poll_list *next; 524 struct poll_list *next;
400 int len; 525 int len;
@@ -436,16 +561,19 @@ static void do_pollfd(unsigned int num, struct pollfd * fdpage,
436} 561}
437 562
438static int do_poll(unsigned int nfds, struct poll_list *list, 563static int do_poll(unsigned int nfds, struct poll_list *list,
439 struct poll_wqueues *wait, long timeout) 564 struct poll_wqueues *wait, s64 *timeout)
440{ 565{
441 int count = 0; 566 int count = 0;
442 poll_table* pt = &wait->pt; 567 poll_table* pt = &wait->pt;
443 568
444 if (!timeout) 569 /* Optimise the no-wait case */
570 if (!(*timeout))
445 pt = NULL; 571 pt = NULL;
446 572
447 for (;;) { 573 for (;;) {
448 struct poll_list *walk; 574 struct poll_list *walk;
575 long __timeout;
576
449 set_current_state(TASK_INTERRUPTIBLE); 577 set_current_state(TASK_INTERRUPTIBLE);
450 walk = list; 578 walk = list;
451 while(walk != NULL) { 579 while(walk != NULL) {
@@ -453,18 +581,36 @@ static int do_poll(unsigned int nfds, struct poll_list *list,
453 walk = walk->next; 581 walk = walk->next;
454 } 582 }
455 pt = NULL; 583 pt = NULL;
456 if (count || !timeout || signal_pending(current)) 584 if (count || !*timeout || signal_pending(current))
457 break; 585 break;
458 count = wait->error; 586 count = wait->error;
459 if (count) 587 if (count)
460 break; 588 break;
461 timeout = schedule_timeout(timeout); 589
590 if (*timeout < 0) {
591 /* Wait indefinitely */
592 __timeout = MAX_SCHEDULE_TIMEOUT;
593 } else if (unlikely(*timeout >= (s64)MAX_SCHEDULE_TIMEOUT-1)) {
594 /*
595 * Wait for longer than MAX_SCHEDULE_TIMEOUT. Do it in
596 * a loop
597 */
598 __timeout = MAX_SCHEDULE_TIMEOUT - 1;
599 *timeout -= __timeout;
600 } else {
601 __timeout = *timeout;
602 *timeout = 0;
603 }
604
605 __timeout = schedule_timeout(__timeout);
606 if (*timeout >= 0)
607 *timeout += __timeout;
462 } 608 }
463 __set_current_state(TASK_RUNNING); 609 __set_current_state(TASK_RUNNING);
464 return count; 610 return count;
465} 611}
466 612
467asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long timeout) 613int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, s64 *timeout)
468{ 614{
469 struct poll_wqueues table; 615 struct poll_wqueues table;
470 int fdcount, err; 616 int fdcount, err;
@@ -482,14 +628,6 @@ asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long ti
482 if (nfds > max_fdset && nfds > OPEN_MAX) 628 if (nfds > max_fdset && nfds > OPEN_MAX)
483 return -EINVAL; 629 return -EINVAL;
484 630
485 if (timeout) {
486 /* Careful about overflow in the intermediate values */
487 if ((unsigned long) timeout < MAX_SCHEDULE_TIMEOUT / HZ)
488 timeout = (unsigned long)(timeout*HZ+999)/1000+1;
489 else /* Negative or overflow */
490 timeout = MAX_SCHEDULE_TIMEOUT;
491 }
492
493 poll_initwait(&table); 631 poll_initwait(&table);
494 632
495 head = NULL; 633 head = NULL;
@@ -519,6 +657,7 @@ asmlinkage long sys_poll(struct pollfd __user * ufds, unsigned int nfds, long ti
519 } 657 }
520 i -= pp->len; 658 i -= pp->len;
521 } 659 }
660
522 fdcount = do_poll(nfds, head, &table, timeout); 661 fdcount = do_poll(nfds, head, &table, timeout);
523 662
524 /* OK, now copy the revents fields back to user space. */ 663 /* OK, now copy the revents fields back to user space. */
@@ -547,3 +686,98 @@ out_fds:
547 poll_freewait(&table); 686 poll_freewait(&table);
548 return err; 687 return err;
549} 688}
689
690asmlinkage long sys_poll(struct pollfd __user *ufds, unsigned int nfds,
691 long timeout_msecs)
692{
693 s64 timeout_jiffies = 0;
694
695 if (timeout_msecs) {
696#if HZ > 1000
697 /* We can only overflow if HZ > 1000 */
698 if (timeout_msecs / 1000 > (s64)0x7fffffffffffffffULL / (s64)HZ)
699 timeout_jiffies = -1;
700 else
701#endif
702 timeout_jiffies = msecs_to_jiffies(timeout_msecs);
703 }
704
705 return do_sys_poll(ufds, nfds, &timeout_jiffies);
706}
707
708#ifdef TIF_RESTORE_SIGMASK
709asmlinkage long sys_ppoll(struct pollfd __user *ufds, unsigned int nfds,
710 struct timespec __user *tsp, const sigset_t __user *sigmask,
711 size_t sigsetsize)
712{
713 sigset_t ksigmask, sigsaved;
714 struct timespec ts;
715 s64 timeout = -1;
716 int ret;
717
718 if (tsp) {
719 if (copy_from_user(&ts, tsp, sizeof(ts)))
720 return -EFAULT;
721
722 /* Cast to u64 to make GCC stop complaining */
723 if ((u64)ts.tv_sec >= (u64)MAX_INT64_SECONDS)
724 timeout = -1; /* infinite */
725 else {
726 timeout = ROUND_UP(ts.tv_nsec, NSEC_PER_SEC/HZ);
727 timeout += ts.tv_sec * HZ;
728 }
729 }
730
731 if (sigmask) {
732 /* XXX: Don't preclude handling different sized sigset_t's. */
733 if (sigsetsize != sizeof(sigset_t))
734 return -EINVAL;
735 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask)))
736 return -EFAULT;
737
738 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP));
739 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved);
740 }
741
742 ret = do_sys_poll(ufds, nfds, &timeout);
743
744 /* We can restart this syscall, usually */
745 if (ret == -EINTR) {
746 /*
747 * Don't restore the signal mask yet. Let do_signal() deliver
748 * the signal on the way back to userspace, before the signal
749 * mask is restored.
750 */
751 if (sigmask) {
752 memcpy(&current->saved_sigmask, &sigsaved,
753 sizeof(sigsaved));
754 set_thread_flag(TIF_RESTORE_SIGMASK);
755 }
756 ret = -ERESTARTNOHAND;
757 } else if (sigmask)
758 sigprocmask(SIG_SETMASK, &sigsaved, NULL);
759
760 if (tsp && timeout >= 0) {
761 if (current->personality & STICKY_TIMEOUTS)
762 goto sticky;
763 /* Yes, we know it's actually an s64, but it's also positive. */
764 ts.tv_nsec = jiffies_to_usecs(do_div((*(u64*)&timeout), HZ)) * 1000;
765 ts.tv_sec = timeout;
766 if (copy_to_user(tsp, &ts, sizeof(ts))) {
767 sticky:
768 /*
769 * If an application puts its timeval in read-only
770 * memory, we don't want the Linux-specific update to
771 * the timeval to cause a fault after the select has
772 * completed successfully. However, because we're not
773 * updating the timeval, we can't restart the system
774 * call.
775 */
776 if (ret == -ERESTARTNOHAND && timeout >= 0)
777 ret = -EINTR;
778 }
779 }
780
781 return ret;
782}
783#endif /* TIF_RESTORE_SIGMASK */