1 /* 2 * This file contains the procedures for the handling of select and poll 3 * 4 * Created for Linux based loosely upon Mathius Lattner's minix 5 * patches by Peter MacDonald. Heavily edited by Linus. 6 * 7 * 4 February 1994 8 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS 9 * flag set in its personality we do *not* modify the given timeout 10 * parameter to reflect time remaining. 11 * 12 * 24 January 2000 13 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 14 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). 15 */ 16 17 #include <linux/kernel.h> 18 #include <linux/sched.h> 19 #include <linux/syscalls.h> 20 #include <linux/export.h> 21 #include <linux/slab.h> 22 #include <linux/poll.h> 23 #include <linux/personality.h> /* for STICKY_TIMEOUTS */ 24 #include <linux/file.h> 25 #include <linux/fdtable.h> 26 #include <linux/fs.h> 27 #include <linux/rcupdate.h> 28 #include <linux/hrtimer.h> 29 #include <linux/sched/rt.h> 30 #include <linux/freezer.h> 31 32 #include <asm/uaccess.h> 33 34 35 /* 36 * Estimate expected accuracy in ns from a timeval. 37 * 38 * After quite a bit of churning around, we've settled on 39 * a simple thing of taking 0.1% of the timeout as the 40 * slack, with a cap of 100 msec. 41 * "nice" tasks get a 0.5% slack instead. 42 * 43 * Consider this comment an open invitation to come up with even 44 * better solutions.. 45 */ 46 47 #define MAX_SLACK (100 * NSEC_PER_MSEC) 48 49 static long __estimate_accuracy(struct timespec *tv) 50 { 51 long slack; 52 int divfactor = 1000; 53 54 if (tv->tv_sec < 0) 55 return 0; 56 57 if (task_nice(current) > 0) 58 divfactor = divfactor / 5; 59 60 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) 61 return MAX_SLACK; 62 63 slack = tv->tv_nsec / divfactor; 64 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); 65 66 if (slack > MAX_SLACK) 67 return MAX_SLACK; 68 69 return slack; 70 } 71 72 long select_estimate_accuracy(struct timespec *tv) 73 { 74 unsigned long ret; 75 struct timespec now; 76 77 /* 78 * Realtime tasks get a slack of 0 for obvious reasons. 79 */ 80 81 if (rt_task(current)) 82 return 0; 83 84 ktime_get_ts(&now); 85 now = timespec_sub(*tv, now); 86 ret = __estimate_accuracy(&now); 87 if (ret < current->timer_slack_ns) 88 return current->timer_slack_ns; 89 return ret; 90 } 91 92 93 94 struct poll_table_page { 95 struct poll_table_page * next; 96 struct poll_table_entry * entry; 97 struct poll_table_entry entries[0]; 98 }; 99 100 #define POLL_TABLE_FULL(table) \ 101 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) 102 103 /* 104 * Ok, Peter made a complicated, but straightforward multiple_wait() function. 105 * I have rewritten this, taking some shortcuts: This code may not be easy to 106 * follow, but it should be free of race-conditions, and it's practical. If you 107 * understand what I'm doing here, then you understand how the linux 108 * sleep/wakeup mechanism works. 109 * 110 * Two very simple procedures, poll_wait() and poll_freewait() make all the 111 * work. poll_wait() is an inline-function defined in <linux/poll.h>, 112 * as all select/poll functions have to call it to add an entry to the 113 * poll table. 114 */ 115 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, 116 poll_table *p); 117 118 void poll_initwait(struct poll_wqueues *pwq) 119 { 120 init_poll_funcptr(&pwq->pt, __pollwait); 121 pwq->polling_task = current; 122 pwq->triggered = 0; 123 pwq->error = 0; 124 pwq->table = NULL; 125 pwq->inline_index = 0; 126 } 127 EXPORT_SYMBOL(poll_initwait); 128 129 static void free_poll_entry(struct poll_table_entry *entry) 130 { 131 remove_wait_queue(entry->wait_address, &entry->wait); 132 fput(entry->filp); 133 } 134 135 void poll_freewait(struct poll_wqueues *pwq) 136 { 137 struct poll_table_page * p = pwq->table; 138 int i; 139 for (i = 0; i < pwq->inline_index; i++) 140 free_poll_entry(pwq->inline_entries + i); 141 while (p) { 142 struct poll_table_entry * entry; 143 struct poll_table_page *old; 144 145 entry = p->entry; 146 do { 147 entry--; 148 free_poll_entry(entry); 149 } while (entry > p->entries); 150 old = p; 151 p = p->next; 152 free_page((unsigned long) old); 153 } 154 } 155 EXPORT_SYMBOL(poll_freewait); 156 157 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) 158 { 159 struct poll_table_page *table = p->table; 160 161 if (p->inline_index < N_INLINE_POLL_ENTRIES) 162 return p->inline_entries + p->inline_index++; 163 164 if (!table || POLL_TABLE_FULL(table)) { 165 struct poll_table_page *new_table; 166 167 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); 168 if (!new_table) { 169 p->error = -ENOMEM; 170 return NULL; 171 } 172 new_table->entry = new_table->entries; 173 new_table->next = table; 174 p->table = new_table; 175 table = new_table; 176 } 177 178 return table->entry++; 179 } 180 181 static int __pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) 182 { 183 struct poll_wqueues *pwq = wait->private; 184 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); 185 186 /* 187 * Although this function is called under waitqueue lock, LOCK 188 * doesn't imply write barrier and the users expect write 189 * barrier semantics on wakeup functions. The following 190 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() 191 * and is paired with set_mb() in poll_schedule_timeout. 192 */ 193 smp_wmb(); 194 pwq->triggered = 1; 195 196 /* 197 * Perform the default wake up operation using a dummy 198 * waitqueue. 199 * 200 * TODO: This is hacky but there currently is no interface to 201 * pass in @sync. @sync is scheduled to be removed and once 202 * that happens, wake_up_process() can be used directly. 203 */ 204 return default_wake_function(&dummy_wait, mode, sync, key); 205 } 206 207 static int pollwake(wait_queue_t *wait, unsigned mode, int sync, void *key) 208 { 209 struct poll_table_entry *entry; 210 211 entry = container_of(wait, struct poll_table_entry, wait); 212 if (key && !((unsigned long)key & entry->key)) 213 return 0; 214 return __pollwake(wait, mode, sync, key); 215 } 216 217 /* Add a new entry */ 218 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, 219 poll_table *p) 220 { 221 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); 222 struct poll_table_entry *entry = poll_get_entry(pwq); 223 if (!entry) 224 return; 225 entry->filp = get_file(filp); 226 entry->wait_address = wait_address; 227 entry->key = p->_key; 228 init_waitqueue_func_entry(&entry->wait, pollwake); 229 entry->wait.private = pwq; 230 add_wait_queue(wait_address, &entry->wait); 231 } 232 233 int poll_schedule_timeout(struct poll_wqueues *pwq, int state, 234 ktime_t *expires, unsigned long slack) 235 { 236 int rc = -EINTR; 237 238 set_current_state(state); 239 if (!pwq->triggered) 240 rc = freezable_schedule_hrtimeout_range(expires, slack, 241 HRTIMER_MODE_ABS); 242 __set_current_state(TASK_RUNNING); 243 244 /* 245 * Prepare for the next iteration. 246 * 247 * The following set_mb() serves two purposes. First, it's 248 * the counterpart rmb of the wmb in pollwake() such that data 249 * written before wake up is always visible after wake up. 250 * Second, the full barrier guarantees that triggered clearing 251 * doesn't pass event check of the next iteration. Note that 252 * this problem doesn't exist for the first iteration as 253 * add_wait_queue() has full barrier semantics. 254 */ 255 set_mb(pwq->triggered, 0); 256 257 return rc; 258 } 259 EXPORT_SYMBOL(poll_schedule_timeout); 260 261 /** 262 * poll_select_set_timeout - helper function to setup the timeout value 263 * @to: pointer to timespec variable for the final timeout 264 * @sec: seconds (from user space) 265 * @nsec: nanoseconds (from user space) 266 * 267 * Note, we do not use a timespec for the user space value here, That 268 * way we can use the function for timeval and compat interfaces as well. 269 * 270 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. 271 */ 272 int poll_select_set_timeout(struct timespec *to, long sec, long nsec) 273 { 274 struct timespec ts = {.tv_sec = sec, .tv_nsec = nsec}; 275 276 if (!timespec_valid(&ts)) 277 return -EINVAL; 278 279 /* Optimize for the zero timeout value here */ 280 if (!sec && !nsec) { 281 to->tv_sec = to->tv_nsec = 0; 282 } else { 283 ktime_get_ts(to); 284 *to = timespec_add_safe(*to, ts); 285 } 286 return 0; 287 } 288 289 static int poll_select_copy_remaining(struct timespec *end_time, void __user *p, 290 int timeval, int ret) 291 { 292 struct timespec rts; 293 struct timeval rtv; 294 295 if (!p) 296 return ret; 297 298 if (current->personality & STICKY_TIMEOUTS) 299 goto sticky; 300 301 /* No update for zero timeout */ 302 if (!end_time->tv_sec && !end_time->tv_nsec) 303 return ret; 304 305 ktime_get_ts(&rts); 306 rts = timespec_sub(*end_time, rts); 307 if (rts.tv_sec < 0) 308 rts.tv_sec = rts.tv_nsec = 0; 309 310 if (timeval) { 311 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) 312 memset(&rtv, 0, sizeof(rtv)); 313 rtv.tv_sec = rts.tv_sec; 314 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; 315 316 if (!copy_to_user(p, &rtv, sizeof(rtv))) 317 return ret; 318 319 } else if (!copy_to_user(p, &rts, sizeof(rts))) 320 return ret; 321 322 /* 323 * If an application puts its timeval in read-only memory, we 324 * don't want the Linux-specific update to the timeval to 325 * cause a fault after the select has completed 326 * successfully. However, because we're not updating the 327 * timeval, we can't restart the system call. 328 */ 329 330 sticky: 331 if (ret == -ERESTARTNOHAND) 332 ret = -EINTR; 333 return ret; 334 } 335 336 #define FDS_IN(fds, n) (fds->in + n) 337 #define FDS_OUT(fds, n) (fds->out + n) 338 #define FDS_EX(fds, n) (fds->ex + n) 339 340 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) 341 342 static int max_select_fd(unsigned long n, fd_set_bits *fds) 343 { 344 unsigned long *open_fds; 345 unsigned long set; 346 int max; 347 struct fdtable *fdt; 348 349 /* handle last in-complete long-word first */ 350 set = ~(~0UL << (n & (BITS_PER_LONG-1))); 351 n /= BITS_PER_LONG; 352 fdt = files_fdtable(current->files); 353 open_fds = fdt->open_fds + n; 354 max = 0; 355 if (set) { 356 set &= BITS(fds, n); 357 if (set) { 358 if (!(set & ~*open_fds)) 359 goto get_max; 360 return -EBADF; 361 } 362 } 363 while (n) { 364 open_fds--; 365 n--; 366 set = BITS(fds, n); 367 if (!set) 368 continue; 369 if (set & ~*open_fds) 370 return -EBADF; 371 if (max) 372 continue; 373 get_max: 374 do { 375 max++; 376 set >>= 1; 377 } while (set); 378 max += n * BITS_PER_LONG; 379 } 380 381 return max; 382 } 383 384 #define POLLIN_SET (POLLRDNORM | POLLRDBAND | POLLIN | POLLHUP | POLLERR) 385 #define POLLOUT_SET (POLLWRBAND | POLLWRNORM | POLLOUT | POLLERR) 386 #define POLLEX_SET (POLLPRI) 387 388 static inline void wait_key_set(poll_table *wait, unsigned long in, 389 unsigned long out, unsigned long bit) 390 { 391 wait->_key = POLLEX_SET; 392 if (in & bit) 393 wait->_key |= POLLIN_SET; 394 if (out & bit) 395 wait->_key |= POLLOUT_SET; 396 } 397 398 int do_select(int n, fd_set_bits *fds, struct timespec *end_time) 399 { 400 ktime_t expire, *to = NULL; 401 struct poll_wqueues table; 402 poll_table *wait; 403 int retval, i, timed_out = 0; 404 unsigned long slack = 0; 405 406 rcu_read_lock(); 407 retval = max_select_fd(n, fds); 408 rcu_read_unlock(); 409 410 if (retval < 0) 411 return retval; 412 n = retval; 413 414 poll_initwait(&table); 415 wait = &table.pt; 416 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { 417 wait->_qproc = NULL; 418 timed_out = 1; 419 } 420 421 if (end_time && !timed_out) 422 slack = select_estimate_accuracy(end_time); 423 424 retval = 0; 425 for (;;) { 426 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; 427 428 inp = fds->in; outp = fds->out; exp = fds->ex; 429 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; 430 431 for (i = 0; i < n; ++rinp, ++routp, ++rexp) { 432 unsigned long in, out, ex, all_bits, bit = 1, mask, j; 433 unsigned long res_in = 0, res_out = 0, res_ex = 0; 434 435 in = *inp++; out = *outp++; ex = *exp++; 436 all_bits = in | out | ex; 437 if (all_bits == 0) { 438 i += BITS_PER_LONG; 439 continue; 440 } 441 442 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { 443 struct fd f; 444 if (i >= n) 445 break; 446 if (!(bit & all_bits)) 447 continue; 448 f = fdget(i); 449 if (f.file) { 450 const struct file_operations *f_op; 451 f_op = f.file->f_op; 452 mask = DEFAULT_POLLMASK; 453 if (f_op && f_op->poll) { 454 wait_key_set(wait, in, out, bit); 455 mask = (*f_op->poll)(f.file, wait); 456 } 457 fdput(f); 458 if ((mask & POLLIN_SET) && (in & bit)) { 459 res_in |= bit; 460 retval++; 461 wait->_qproc = NULL; 462 } 463 if ((mask & POLLOUT_SET) && (out & bit)) { 464 res_out |= bit; 465 retval++; 466 wait->_qproc = NULL; 467 } 468 if ((mask & POLLEX_SET) && (ex & bit)) { 469 res_ex |= bit; 470 retval++; 471 wait->_qproc = NULL; 472 } 473 } 474 } 475 if (res_in) 476 *rinp = res_in; 477 if (res_out) 478 *routp = res_out; 479 if (res_ex) 480 *rexp = res_ex; 481 cond_resched(); 482 } 483 wait->_qproc = NULL; 484 if (retval || timed_out || signal_pending(current)) 485 break; 486 if (table.error) { 487 retval = table.error; 488 break; 489 } 490 491 /* 492 * If this is the first loop and we have a timeout 493 * given, then we convert to ktime_t and set the to 494 * pointer to the expiry value. 495 */ 496 if (end_time && !to) { 497 expire = timespec_to_ktime(*end_time); 498 to = &expire; 499 } 500 501 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, 502 to, slack)) 503 timed_out = 1; 504 } 505 506 poll_freewait(&table); 507 508 return retval; 509 } 510 511 /* 512 * We can actually return ERESTARTSYS instead of EINTR, but I'd 513 * like to be certain this leads to no problems. So I return 514 * EINTR just for safety. 515 * 516 * Update: ERESTARTSYS breaks at least the xview clock binary, so 517 * I'm trying ERESTARTNOHAND which restart only when you want to. 518 */ 519 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, 520 fd_set __user *exp, struct timespec *end_time) 521 { 522 fd_set_bits fds; 523 void *bits; 524 int ret, max_fds; 525 unsigned int size; 526 struct fdtable *fdt; 527 /* Allocate small arguments on the stack to save memory and be faster */ 528 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; 529 530 ret = -EINVAL; 531 if (n < 0) 532 goto out_nofds; 533 534 /* max_fds can increase, so grab it once to avoid race */ 535 rcu_read_lock(); 536 fdt = files_fdtable(current->files); 537 max_fds = fdt->max_fds; 538 rcu_read_unlock(); 539 if (n > max_fds) 540 n = max_fds; 541 542 /* 543 * We need 6 bitmaps (in/out/ex for both incoming and outgoing), 544 * since we used fdset we need to allocate memory in units of 545 * long-words. 546 */ 547 size = FDS_BYTES(n); 548 bits = stack_fds; 549 if (size > sizeof(stack_fds) / 6) { 550 /* Not enough space in on-stack array; must use kmalloc */ 551 ret = -ENOMEM; 552 bits = kmalloc(6 * size, GFP_KERNEL); 553 if (!bits) 554 goto out_nofds; 555 } 556 fds.in = bits; 557 fds.out = bits + size; 558 fds.ex = bits + 2*size; 559 fds.res_in = bits + 3*size; 560 fds.res_out = bits + 4*size; 561 fds.res_ex = bits + 5*size; 562 563 if ((ret = get_fd_set(n, inp, fds.in)) || 564 (ret = get_fd_set(n, outp, fds.out)) || 565 (ret = get_fd_set(n, exp, fds.ex))) 566 goto out; 567 zero_fd_set(n, fds.res_in); 568 zero_fd_set(n, fds.res_out); 569 zero_fd_set(n, fds.res_ex); 570 571 ret = do_select(n, &fds, end_time); 572 573 if (ret < 0) 574 goto out; 575 if (!ret) { 576 ret = -ERESTARTNOHAND; 577 if (signal_pending(current)) 578 goto out; 579 ret = 0; 580 } 581 582 if (set_fd_set(n, inp, fds.res_in) || 583 set_fd_set(n, outp, fds.res_out) || 584 set_fd_set(n, exp, fds.res_ex)) 585 ret = -EFAULT; 586 587 out: 588 if (bits != stack_fds) 589 kfree(bits); 590 out_nofds: 591 return ret; 592 } 593 594 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, 595 fd_set __user *, exp, struct timeval __user *, tvp) 596 { 597 struct timespec end_time, *to = NULL; 598 struct timeval tv; 599 int ret; 600 601 if (tvp) { 602 if (copy_from_user(&tv, tvp, sizeof(tv))) 603 return -EFAULT; 604 605 to = &end_time; 606 if (poll_select_set_timeout(to, 607 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), 608 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) 609 return -EINVAL; 610 } 611 612 ret = core_sys_select(n, inp, outp, exp, to); 613 ret = poll_select_copy_remaining(&end_time, tvp, 1, ret); 614 615 return ret; 616 } 617 618 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, 619 fd_set __user *exp, struct timespec __user *tsp, 620 const sigset_t __user *sigmask, size_t sigsetsize) 621 { 622 sigset_t ksigmask, sigsaved; 623 struct timespec ts, end_time, *to = NULL; 624 int ret; 625 626 if (tsp) { 627 if (copy_from_user(&ts, tsp, sizeof(ts))) 628 return -EFAULT; 629 630 to = &end_time; 631 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 632 return -EINVAL; 633 } 634 635 if (sigmask) { 636 /* XXX: Don't preclude handling different sized sigset_t's. */ 637 if (sigsetsize != sizeof(sigset_t)) 638 return -EINVAL; 639 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) 640 return -EFAULT; 641 642 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); 643 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); 644 } 645 646 ret = core_sys_select(n, inp, outp, exp, to); 647 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); 648 649 if (ret == -ERESTARTNOHAND) { 650 /* 651 * Don't restore the signal mask yet. Let do_signal() deliver 652 * the signal on the way back to userspace, before the signal 653 * mask is restored. 654 */ 655 if (sigmask) { 656 memcpy(¤t->saved_sigmask, &sigsaved, 657 sizeof(sigsaved)); 658 set_restore_sigmask(); 659 } 660 } else if (sigmask) 661 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 662 663 return ret; 664 } 665 666 /* 667 * Most architectures can't handle 7-argument syscalls. So we provide a 668 * 6-argument version where the sixth argument is a pointer to a structure 669 * which has a pointer to the sigset_t itself followed by a size_t containing 670 * the sigset size. 671 */ 672 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, 673 fd_set __user *, exp, struct timespec __user *, tsp, 674 void __user *, sig) 675 { 676 size_t sigsetsize = 0; 677 sigset_t __user *up = NULL; 678 679 if (sig) { 680 if (!access_ok(VERIFY_READ, sig, sizeof(void *)+sizeof(size_t)) 681 || __get_user(up, (sigset_t __user * __user *)sig) 682 || __get_user(sigsetsize, 683 (size_t __user *)(sig+sizeof(void *)))) 684 return -EFAULT; 685 } 686 687 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize); 688 } 689 690 #ifdef __ARCH_WANT_SYS_OLD_SELECT 691 struct sel_arg_struct { 692 unsigned long n; 693 fd_set __user *inp, *outp, *exp; 694 struct timeval __user *tvp; 695 }; 696 697 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) 698 { 699 struct sel_arg_struct a; 700 701 if (copy_from_user(&a, arg, sizeof(a))) 702 return -EFAULT; 703 return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp); 704 } 705 #endif 706 707 struct poll_list { 708 struct poll_list *next; 709 int len; 710 struct pollfd entries[0]; 711 }; 712 713 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) 714 715 /* 716 * Fish for pollable events on the pollfd->fd file descriptor. We're only 717 * interested in events matching the pollfd->events mask, and the result 718 * matching that mask is both recorded in pollfd->revents and returned. The 719 * pwait poll_table will be used by the fd-provided poll handler for waiting, 720 * if pwait->_qproc is non-NULL. 721 */ 722 static inline unsigned int do_pollfd(struct pollfd *pollfd, poll_table *pwait) 723 { 724 unsigned int mask; 725 int fd; 726 727 mask = 0; 728 fd = pollfd->fd; 729 if (fd >= 0) { 730 struct fd f = fdget(fd); 731 mask = POLLNVAL; 732 if (f.file) { 733 mask = DEFAULT_POLLMASK; 734 if (f.file->f_op && f.file->f_op->poll) { 735 pwait->_key = pollfd->events|POLLERR|POLLHUP; 736 mask = f.file->f_op->poll(f.file, pwait); 737 } 738 /* Mask out unneeded events. */ 739 mask &= pollfd->events | POLLERR | POLLHUP; 740 fdput(f); 741 } 742 } 743 pollfd->revents = mask; 744 745 return mask; 746 } 747 748 static int do_poll(unsigned int nfds, struct poll_list *list, 749 struct poll_wqueues *wait, struct timespec *end_time) 750 { 751 poll_table* pt = &wait->pt; 752 ktime_t expire, *to = NULL; 753 int timed_out = 0, count = 0; 754 unsigned long slack = 0; 755 756 /* Optimise the no-wait case */ 757 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { 758 pt->_qproc = NULL; 759 timed_out = 1; 760 } 761 762 if (end_time && !timed_out) 763 slack = select_estimate_accuracy(end_time); 764 765 for (;;) { 766 struct poll_list *walk; 767 768 for (walk = list; walk != NULL; walk = walk->next) { 769 struct pollfd * pfd, * pfd_end; 770 771 pfd = walk->entries; 772 pfd_end = pfd + walk->len; 773 for (; pfd != pfd_end; pfd++) { 774 /* 775 * Fish for events. If we found one, record it 776 * and kill poll_table->_qproc, so we don't 777 * needlessly register any other waiters after 778 * this. They'll get immediately deregistered 779 * when we break out and return. 780 */ 781 if (do_pollfd(pfd, pt)) { 782 count++; 783 pt->_qproc = NULL; 784 } 785 } 786 } 787 /* 788 * All waiters have already been registered, so don't provide 789 * a poll_table->_qproc to them on the next loop iteration. 790 */ 791 pt->_qproc = NULL; 792 if (!count) { 793 count = wait->error; 794 if (signal_pending(current)) 795 count = -EINTR; 796 } 797 if (count || timed_out) 798 break; 799 800 /* 801 * If this is the first loop and we have a timeout 802 * given, then we convert to ktime_t and set the to 803 * pointer to the expiry value. 804 */ 805 if (end_time && !to) { 806 expire = timespec_to_ktime(*end_time); 807 to = &expire; 808 } 809 810 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) 811 timed_out = 1; 812 } 813 return count; 814 } 815 816 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ 817 sizeof(struct pollfd)) 818 819 int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, 820 struct timespec *end_time) 821 { 822 struct poll_wqueues table; 823 int err = -EFAULT, fdcount, len, size; 824 /* Allocate small arguments on the stack to save memory and be 825 faster - use long to make sure the buffer is aligned properly 826 on 64 bit archs to avoid unaligned access */ 827 long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; 828 struct poll_list *const head = (struct poll_list *)stack_pps; 829 struct poll_list *walk = head; 830 unsigned long todo = nfds; 831 832 if (nfds > rlimit(RLIMIT_NOFILE)) 833 return -EINVAL; 834 835 len = min_t(unsigned int, nfds, N_STACK_PPS); 836 for (;;) { 837 walk->next = NULL; 838 walk->len = len; 839 if (!len) 840 break; 841 842 if (copy_from_user(walk->entries, ufds + nfds-todo, 843 sizeof(struct pollfd) * walk->len)) 844 goto out_fds; 845 846 todo -= walk->len; 847 if (!todo) 848 break; 849 850 len = min(todo, POLLFD_PER_PAGE); 851 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; 852 walk = walk->next = kmalloc(size, GFP_KERNEL); 853 if (!walk) { 854 err = -ENOMEM; 855 goto out_fds; 856 } 857 } 858 859 poll_initwait(&table); 860 fdcount = do_poll(nfds, head, &table, end_time); 861 poll_freewait(&table); 862 863 for (walk = head; walk; walk = walk->next) { 864 struct pollfd *fds = walk->entries; 865 int j; 866 867 for (j = 0; j < walk->len; j++, ufds++) 868 if (__put_user(fds[j].revents, &ufds->revents)) 869 goto out_fds; 870 } 871 872 err = fdcount; 873 out_fds: 874 walk = head->next; 875 while (walk) { 876 struct poll_list *pos = walk; 877 walk = walk->next; 878 kfree(pos); 879 } 880 881 return err; 882 } 883 884 static long do_restart_poll(struct restart_block *restart_block) 885 { 886 struct pollfd __user *ufds = restart_block->poll.ufds; 887 int nfds = restart_block->poll.nfds; 888 struct timespec *to = NULL, end_time; 889 int ret; 890 891 if (restart_block->poll.has_timeout) { 892 end_time.tv_sec = restart_block->poll.tv_sec; 893 end_time.tv_nsec = restart_block->poll.tv_nsec; 894 to = &end_time; 895 } 896 897 ret = do_sys_poll(ufds, nfds, to); 898 899 if (ret == -EINTR) { 900 restart_block->fn = do_restart_poll; 901 ret = -ERESTART_RESTARTBLOCK; 902 } 903 return ret; 904 } 905 906 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, 907 int, timeout_msecs) 908 { 909 struct timespec end_time, *to = NULL; 910 int ret; 911 912 if (timeout_msecs >= 0) { 913 to = &end_time; 914 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, 915 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); 916 } 917 918 ret = do_sys_poll(ufds, nfds, to); 919 920 if (ret == -EINTR) { 921 struct restart_block *restart_block; 922 923 restart_block = ¤t_thread_info()->restart_block; 924 restart_block->fn = do_restart_poll; 925 restart_block->poll.ufds = ufds; 926 restart_block->poll.nfds = nfds; 927 928 if (timeout_msecs >= 0) { 929 restart_block->poll.tv_sec = end_time.tv_sec; 930 restart_block->poll.tv_nsec = end_time.tv_nsec; 931 restart_block->poll.has_timeout = 1; 932 } else 933 restart_block->poll.has_timeout = 0; 934 935 ret = -ERESTART_RESTARTBLOCK; 936 } 937 return ret; 938 } 939 940 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, 941 struct timespec __user *, tsp, const sigset_t __user *, sigmask, 942 size_t, sigsetsize) 943 { 944 sigset_t ksigmask, sigsaved; 945 struct timespec ts, end_time, *to = NULL; 946 int ret; 947 948 if (tsp) { 949 if (copy_from_user(&ts, tsp, sizeof(ts))) 950 return -EFAULT; 951 952 to = &end_time; 953 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 954 return -EINVAL; 955 } 956 957 if (sigmask) { 958 /* XXX: Don't preclude handling different sized sigset_t's. */ 959 if (sigsetsize != sizeof(sigset_t)) 960 return -EINVAL; 961 if (copy_from_user(&ksigmask, sigmask, sizeof(ksigmask))) 962 return -EFAULT; 963 964 sigdelsetmask(&ksigmask, sigmask(SIGKILL)|sigmask(SIGSTOP)); 965 sigprocmask(SIG_SETMASK, &ksigmask, &sigsaved); 966 } 967 968 ret = do_sys_poll(ufds, nfds, to); 969 970 /* We can restart this syscall, usually */ 971 if (ret == -EINTR) { 972 /* 973 * Don't restore the signal mask yet. Let do_signal() deliver 974 * the signal on the way back to userspace, before the signal 975 * mask is restored. 976 */ 977 if (sigmask) { 978 memcpy(¤t->saved_sigmask, &sigsaved, 979 sizeof(sigsaved)); 980 set_restore_sigmask(); 981 } 982 ret = -ERESTARTNOHAND; 983 } else if (sigmask) 984 sigprocmask(SIG_SETMASK, &sigsaved, NULL); 985 986 ret = poll_select_copy_remaining(&end_time, tsp, 0, ret); 987 988 return ret; 989 } 990