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