1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * This file contains the procedures for the handling of select and poll 4 * 5 * Created for Linux based loosely upon Mathius Lattner's minix 6 * patches by Peter MacDonald. Heavily edited by Linus. 7 * 8 * 4 February 1994 9 * COFF/ELF binary emulation. If the process has the STICKY_TIMEOUTS 10 * flag set in its personality we do *not* modify the given timeout 11 * parameter to reflect time remaining. 12 * 13 * 24 January 2000 14 * Changed sys_poll()/do_poll() to use PAGE_SIZE chunk-based allocation 15 * of fds to overcome nfds < 16390 descriptors limit (Tigran Aivazian). 16 */ 17 18 #include <linux/kernel.h> 19 #include <linux/sched/signal.h> 20 #include <linux/sched/rt.h> 21 #include <linux/syscalls.h> 22 #include <linux/export.h> 23 #include <linux/slab.h> 24 #include <linux/poll.h> 25 #include <linux/personality.h> /* for STICKY_TIMEOUTS */ 26 #include <linux/file.h> 27 #include <linux/fdtable.h> 28 #include <linux/fs.h> 29 #include <linux/rcupdate.h> 30 #include <linux/hrtimer.h> 31 #include <linux/freezer.h> 32 #include <net/busy_poll.h> 33 #include <linux/vmalloc.h> 34 35 #include <linux/uaccess.h> 36 37 38 /* 39 * Estimate expected accuracy in ns from a timeval. 40 * 41 * After quite a bit of churning around, we've settled on 42 * a simple thing of taking 0.1% of the timeout as the 43 * slack, with a cap of 100 msec. 44 * "nice" tasks get a 0.5% slack instead. 45 * 46 * Consider this comment an open invitation to come up with even 47 * better solutions.. 48 */ 49 50 #define MAX_SLACK (100 * NSEC_PER_MSEC) 51 52 static long __estimate_accuracy(struct timespec64 *tv) 53 { 54 long slack; 55 int divfactor = 1000; 56 57 if (tv->tv_sec < 0) 58 return 0; 59 60 if (task_nice(current) > 0) 61 divfactor = divfactor / 5; 62 63 if (tv->tv_sec > MAX_SLACK / (NSEC_PER_SEC/divfactor)) 64 return MAX_SLACK; 65 66 slack = tv->tv_nsec / divfactor; 67 slack += tv->tv_sec * (NSEC_PER_SEC/divfactor); 68 69 if (slack > MAX_SLACK) 70 return MAX_SLACK; 71 72 return slack; 73 } 74 75 u64 select_estimate_accuracy(struct timespec64 *tv) 76 { 77 u64 ret; 78 struct timespec64 now; 79 80 /* 81 * Realtime tasks get a slack of 0 for obvious reasons. 82 */ 83 84 if (rt_task(current)) 85 return 0; 86 87 ktime_get_ts64(&now); 88 now = timespec64_sub(*tv, now); 89 ret = __estimate_accuracy(&now); 90 if (ret < current->timer_slack_ns) 91 return current->timer_slack_ns; 92 return ret; 93 } 94 95 96 97 struct poll_table_page { 98 struct poll_table_page * next; 99 struct poll_table_entry * entry; 100 struct poll_table_entry entries[0]; 101 }; 102 103 #define POLL_TABLE_FULL(table) \ 104 ((unsigned long)((table)->entry+1) > PAGE_SIZE + (unsigned long)(table)) 105 106 /* 107 * Ok, Peter made a complicated, but straightforward multiple_wait() function. 108 * I have rewritten this, taking some shortcuts: This code may not be easy to 109 * follow, but it should be free of race-conditions, and it's practical. If you 110 * understand what I'm doing here, then you understand how the linux 111 * sleep/wakeup mechanism works. 112 * 113 * Two very simple procedures, poll_wait() and poll_freewait() make all the 114 * work. poll_wait() is an inline-function defined in <linux/poll.h>, 115 * as all select/poll functions have to call it to add an entry to the 116 * poll table. 117 */ 118 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, 119 poll_table *p); 120 121 void poll_initwait(struct poll_wqueues *pwq) 122 { 123 init_poll_funcptr(&pwq->pt, __pollwait); 124 pwq->polling_task = current; 125 pwq->triggered = 0; 126 pwq->error = 0; 127 pwq->table = NULL; 128 pwq->inline_index = 0; 129 } 130 EXPORT_SYMBOL(poll_initwait); 131 132 static void free_poll_entry(struct poll_table_entry *entry) 133 { 134 remove_wait_queue(entry->wait_address, &entry->wait); 135 fput(entry->filp); 136 } 137 138 void poll_freewait(struct poll_wqueues *pwq) 139 { 140 struct poll_table_page * p = pwq->table; 141 int i; 142 for (i = 0; i < pwq->inline_index; i++) 143 free_poll_entry(pwq->inline_entries + i); 144 while (p) { 145 struct poll_table_entry * entry; 146 struct poll_table_page *old; 147 148 entry = p->entry; 149 do { 150 entry--; 151 free_poll_entry(entry); 152 } while (entry > p->entries); 153 old = p; 154 p = p->next; 155 free_page((unsigned long) old); 156 } 157 } 158 EXPORT_SYMBOL(poll_freewait); 159 160 static struct poll_table_entry *poll_get_entry(struct poll_wqueues *p) 161 { 162 struct poll_table_page *table = p->table; 163 164 if (p->inline_index < N_INLINE_POLL_ENTRIES) 165 return p->inline_entries + p->inline_index++; 166 167 if (!table || POLL_TABLE_FULL(table)) { 168 struct poll_table_page *new_table; 169 170 new_table = (struct poll_table_page *) __get_free_page(GFP_KERNEL); 171 if (!new_table) { 172 p->error = -ENOMEM; 173 return NULL; 174 } 175 new_table->entry = new_table->entries; 176 new_table->next = table; 177 p->table = new_table; 178 table = new_table; 179 } 180 181 return table->entry++; 182 } 183 184 static int __pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) 185 { 186 struct poll_wqueues *pwq = wait->private; 187 DECLARE_WAITQUEUE(dummy_wait, pwq->polling_task); 188 189 /* 190 * Although this function is called under waitqueue lock, LOCK 191 * doesn't imply write barrier and the users expect write 192 * barrier semantics on wakeup functions. The following 193 * smp_wmb() is equivalent to smp_wmb() in try_to_wake_up() 194 * and is paired with smp_store_mb() in poll_schedule_timeout. 195 */ 196 smp_wmb(); 197 pwq->triggered = 1; 198 199 /* 200 * Perform the default wake up operation using a dummy 201 * waitqueue. 202 * 203 * TODO: This is hacky but there currently is no interface to 204 * pass in @sync. @sync is scheduled to be removed and once 205 * that happens, wake_up_process() can be used directly. 206 */ 207 return default_wake_function(&dummy_wait, mode, sync, key); 208 } 209 210 static int pollwake(wait_queue_entry_t *wait, unsigned mode, int sync, void *key) 211 { 212 struct poll_table_entry *entry; 213 214 entry = container_of(wait, struct poll_table_entry, wait); 215 if (key && !(key_to_poll(key) & entry->key)) 216 return 0; 217 return __pollwake(wait, mode, sync, key); 218 } 219 220 /* Add a new entry */ 221 static void __pollwait(struct file *filp, wait_queue_head_t *wait_address, 222 poll_table *p) 223 { 224 struct poll_wqueues *pwq = container_of(p, struct poll_wqueues, pt); 225 struct poll_table_entry *entry = poll_get_entry(pwq); 226 if (!entry) 227 return; 228 entry->filp = get_file(filp); 229 entry->wait_address = wait_address; 230 entry->key = p->_key; 231 init_waitqueue_func_entry(&entry->wait, pollwake); 232 entry->wait.private = pwq; 233 add_wait_queue(wait_address, &entry->wait); 234 } 235 236 static int poll_schedule_timeout(struct poll_wqueues *pwq, int state, 237 ktime_t *expires, unsigned long slack) 238 { 239 int rc = -EINTR; 240 241 set_current_state(state); 242 if (!pwq->triggered) 243 rc = schedule_hrtimeout_range(expires, slack, HRTIMER_MODE_ABS); 244 __set_current_state(TASK_RUNNING); 245 246 /* 247 * Prepare for the next iteration. 248 * 249 * The following smp_store_mb() serves two purposes. First, it's 250 * the counterpart rmb of the wmb in pollwake() such that data 251 * written before wake up is always visible after wake up. 252 * Second, the full barrier guarantees that triggered clearing 253 * doesn't pass event check of the next iteration. Note that 254 * this problem doesn't exist for the first iteration as 255 * add_wait_queue() has full barrier semantics. 256 */ 257 smp_store_mb(pwq->triggered, 0); 258 259 return rc; 260 } 261 262 /** 263 * poll_select_set_timeout - helper function to setup the timeout value 264 * @to: pointer to timespec64 variable for the final timeout 265 * @sec: seconds (from user space) 266 * @nsec: nanoseconds (from user space) 267 * 268 * Note, we do not use a timespec for the user space value here, That 269 * way we can use the function for timeval and compat interfaces as well. 270 * 271 * Returns -EINVAL if sec/nsec are not normalized. Otherwise 0. 272 */ 273 int poll_select_set_timeout(struct timespec64 *to, time64_t sec, long nsec) 274 { 275 struct timespec64 ts = {.tv_sec = sec, .tv_nsec = nsec}; 276 277 if (!timespec64_valid(&ts)) 278 return -EINVAL; 279 280 /* Optimize for the zero timeout value here */ 281 if (!sec && !nsec) { 282 to->tv_sec = to->tv_nsec = 0; 283 } else { 284 ktime_get_ts64(to); 285 *to = timespec64_add_safe(*to, ts); 286 } 287 return 0; 288 } 289 290 enum poll_time_type { 291 PT_TIMEVAL = 0, 292 PT_OLD_TIMEVAL = 1, 293 PT_TIMESPEC = 2, 294 PT_OLD_TIMESPEC = 3, 295 }; 296 297 static int poll_select_copy_remaining(struct timespec64 *end_time, 298 void __user *p, 299 enum poll_time_type pt_type, int ret) 300 { 301 struct timespec64 rts; 302 303 if (!p) 304 return ret; 305 306 if (current->personality & STICKY_TIMEOUTS) 307 goto sticky; 308 309 /* No update for zero timeout */ 310 if (!end_time->tv_sec && !end_time->tv_nsec) 311 return ret; 312 313 ktime_get_ts64(&rts); 314 rts = timespec64_sub(*end_time, rts); 315 if (rts.tv_sec < 0) 316 rts.tv_sec = rts.tv_nsec = 0; 317 318 319 switch (pt_type) { 320 case PT_TIMEVAL: 321 { 322 struct timeval rtv; 323 324 if (sizeof(rtv) > sizeof(rtv.tv_sec) + sizeof(rtv.tv_usec)) 325 memset(&rtv, 0, sizeof(rtv)); 326 rtv.tv_sec = rts.tv_sec; 327 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; 328 if (!copy_to_user(p, &rtv, sizeof(rtv))) 329 return ret; 330 } 331 break; 332 case PT_OLD_TIMEVAL: 333 { 334 struct old_timeval32 rtv; 335 336 rtv.tv_sec = rts.tv_sec; 337 rtv.tv_usec = rts.tv_nsec / NSEC_PER_USEC; 338 if (!copy_to_user(p, &rtv, sizeof(rtv))) 339 return ret; 340 } 341 break; 342 case PT_TIMESPEC: 343 if (!put_timespec64(&rts, p)) 344 return ret; 345 break; 346 case PT_OLD_TIMESPEC: 347 if (!put_old_timespec32(&rts, p)) 348 return ret; 349 break; 350 default: 351 BUG(); 352 } 353 /* 354 * If an application puts its timeval in read-only memory, we 355 * don't want the Linux-specific update to the timeval to 356 * cause a fault after the select has completed 357 * successfully. However, because we're not updating the 358 * timeval, we can't restart the system call. 359 */ 360 361 sticky: 362 if (ret == -ERESTARTNOHAND) 363 ret = -EINTR; 364 return ret; 365 } 366 367 /* 368 * Scalable version of the fd_set. 369 */ 370 371 typedef struct { 372 unsigned long *in, *out, *ex; 373 unsigned long *res_in, *res_out, *res_ex; 374 } fd_set_bits; 375 376 /* 377 * How many longwords for "nr" bits? 378 */ 379 #define FDS_BITPERLONG (8*sizeof(long)) 380 #define FDS_LONGS(nr) (((nr)+FDS_BITPERLONG-1)/FDS_BITPERLONG) 381 #define FDS_BYTES(nr) (FDS_LONGS(nr)*sizeof(long)) 382 383 /* 384 * Use "unsigned long" accesses to let user-mode fd_set's be long-aligned. 385 */ 386 static inline 387 int get_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) 388 { 389 nr = FDS_BYTES(nr); 390 if (ufdset) 391 return copy_from_user(fdset, ufdset, nr) ? -EFAULT : 0; 392 393 memset(fdset, 0, nr); 394 return 0; 395 } 396 397 static inline unsigned long __must_check 398 set_fd_set(unsigned long nr, void __user *ufdset, unsigned long *fdset) 399 { 400 if (ufdset) 401 return __copy_to_user(ufdset, fdset, FDS_BYTES(nr)); 402 return 0; 403 } 404 405 static inline 406 void zero_fd_set(unsigned long nr, unsigned long *fdset) 407 { 408 memset(fdset, 0, FDS_BYTES(nr)); 409 } 410 411 #define FDS_IN(fds, n) (fds->in + n) 412 #define FDS_OUT(fds, n) (fds->out + n) 413 #define FDS_EX(fds, n) (fds->ex + n) 414 415 #define BITS(fds, n) (*FDS_IN(fds, n)|*FDS_OUT(fds, n)|*FDS_EX(fds, n)) 416 417 static int max_select_fd(unsigned long n, fd_set_bits *fds) 418 { 419 unsigned long *open_fds; 420 unsigned long set; 421 int max; 422 struct fdtable *fdt; 423 424 /* handle last in-complete long-word first */ 425 set = ~(~0UL << (n & (BITS_PER_LONG-1))); 426 n /= BITS_PER_LONG; 427 fdt = files_fdtable(current->files); 428 open_fds = fdt->open_fds + n; 429 max = 0; 430 if (set) { 431 set &= BITS(fds, n); 432 if (set) { 433 if (!(set & ~*open_fds)) 434 goto get_max; 435 return -EBADF; 436 } 437 } 438 while (n) { 439 open_fds--; 440 n--; 441 set = BITS(fds, n); 442 if (!set) 443 continue; 444 if (set & ~*open_fds) 445 return -EBADF; 446 if (max) 447 continue; 448 get_max: 449 do { 450 max++; 451 set >>= 1; 452 } while (set); 453 max += n * BITS_PER_LONG; 454 } 455 456 return max; 457 } 458 459 #define POLLIN_SET (EPOLLRDNORM | EPOLLRDBAND | EPOLLIN | EPOLLHUP | EPOLLERR) 460 #define POLLOUT_SET (EPOLLWRBAND | EPOLLWRNORM | EPOLLOUT | EPOLLERR) 461 #define POLLEX_SET (EPOLLPRI) 462 463 static inline void wait_key_set(poll_table *wait, unsigned long in, 464 unsigned long out, unsigned long bit, 465 __poll_t ll_flag) 466 { 467 wait->_key = POLLEX_SET | ll_flag; 468 if (in & bit) 469 wait->_key |= POLLIN_SET; 470 if (out & bit) 471 wait->_key |= POLLOUT_SET; 472 } 473 474 static int do_select(int n, fd_set_bits *fds, struct timespec64 *end_time) 475 { 476 ktime_t expire, *to = NULL; 477 struct poll_wqueues table; 478 poll_table *wait; 479 int retval, i, timed_out = 0; 480 u64 slack = 0; 481 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; 482 unsigned long busy_start = 0; 483 484 rcu_read_lock(); 485 retval = max_select_fd(n, fds); 486 rcu_read_unlock(); 487 488 if (retval < 0) 489 return retval; 490 n = retval; 491 492 poll_initwait(&table); 493 wait = &table.pt; 494 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { 495 wait->_qproc = NULL; 496 timed_out = 1; 497 } 498 499 if (end_time && !timed_out) 500 slack = select_estimate_accuracy(end_time); 501 502 retval = 0; 503 for (;;) { 504 unsigned long *rinp, *routp, *rexp, *inp, *outp, *exp; 505 bool can_busy_loop = false; 506 507 inp = fds->in; outp = fds->out; exp = fds->ex; 508 rinp = fds->res_in; routp = fds->res_out; rexp = fds->res_ex; 509 510 for (i = 0; i < n; ++rinp, ++routp, ++rexp) { 511 unsigned long in, out, ex, all_bits, bit = 1, j; 512 unsigned long res_in = 0, res_out = 0, res_ex = 0; 513 __poll_t mask; 514 515 in = *inp++; out = *outp++; ex = *exp++; 516 all_bits = in | out | ex; 517 if (all_bits == 0) { 518 i += BITS_PER_LONG; 519 continue; 520 } 521 522 for (j = 0; j < BITS_PER_LONG; ++j, ++i, bit <<= 1) { 523 struct fd f; 524 if (i >= n) 525 break; 526 if (!(bit & all_bits)) 527 continue; 528 f = fdget(i); 529 if (f.file) { 530 wait_key_set(wait, in, out, bit, 531 busy_flag); 532 mask = vfs_poll(f.file, wait); 533 534 fdput(f); 535 if ((mask & POLLIN_SET) && (in & bit)) { 536 res_in |= bit; 537 retval++; 538 wait->_qproc = NULL; 539 } 540 if ((mask & POLLOUT_SET) && (out & bit)) { 541 res_out |= bit; 542 retval++; 543 wait->_qproc = NULL; 544 } 545 if ((mask & POLLEX_SET) && (ex & bit)) { 546 res_ex |= bit; 547 retval++; 548 wait->_qproc = NULL; 549 } 550 /* got something, stop busy polling */ 551 if (retval) { 552 can_busy_loop = false; 553 busy_flag = 0; 554 555 /* 556 * only remember a returned 557 * POLL_BUSY_LOOP if we asked for it 558 */ 559 } else if (busy_flag & mask) 560 can_busy_loop = true; 561 562 } 563 } 564 if (res_in) 565 *rinp = res_in; 566 if (res_out) 567 *routp = res_out; 568 if (res_ex) 569 *rexp = res_ex; 570 cond_resched(); 571 } 572 wait->_qproc = NULL; 573 if (retval || timed_out || signal_pending(current)) 574 break; 575 if (table.error) { 576 retval = table.error; 577 break; 578 } 579 580 /* only if found POLL_BUSY_LOOP sockets && not out of time */ 581 if (can_busy_loop && !need_resched()) { 582 if (!busy_start) { 583 busy_start = busy_loop_current_time(); 584 continue; 585 } 586 if (!busy_loop_timeout(busy_start)) 587 continue; 588 } 589 busy_flag = 0; 590 591 /* 592 * If this is the first loop and we have a timeout 593 * given, then we convert to ktime_t and set the to 594 * pointer to the expiry value. 595 */ 596 if (end_time && !to) { 597 expire = timespec64_to_ktime(*end_time); 598 to = &expire; 599 } 600 601 if (!poll_schedule_timeout(&table, TASK_INTERRUPTIBLE, 602 to, slack)) 603 timed_out = 1; 604 } 605 606 poll_freewait(&table); 607 608 return retval; 609 } 610 611 /* 612 * We can actually return ERESTARTSYS instead of EINTR, but I'd 613 * like to be certain this leads to no problems. So I return 614 * EINTR just for safety. 615 * 616 * Update: ERESTARTSYS breaks at least the xview clock binary, so 617 * I'm trying ERESTARTNOHAND which restart only when you want to. 618 */ 619 int core_sys_select(int n, fd_set __user *inp, fd_set __user *outp, 620 fd_set __user *exp, struct timespec64 *end_time) 621 { 622 fd_set_bits fds; 623 void *bits; 624 int ret, max_fds; 625 size_t size, alloc_size; 626 struct fdtable *fdt; 627 /* Allocate small arguments on the stack to save memory and be faster */ 628 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; 629 630 ret = -EINVAL; 631 if (n < 0) 632 goto out_nofds; 633 634 /* max_fds can increase, so grab it once to avoid race */ 635 rcu_read_lock(); 636 fdt = files_fdtable(current->files); 637 max_fds = fdt->max_fds; 638 rcu_read_unlock(); 639 if (n > max_fds) 640 n = max_fds; 641 642 /* 643 * We need 6 bitmaps (in/out/ex for both incoming and outgoing), 644 * since we used fdset we need to allocate memory in units of 645 * long-words. 646 */ 647 size = FDS_BYTES(n); 648 bits = stack_fds; 649 if (size > sizeof(stack_fds) / 6) { 650 /* Not enough space in on-stack array; must use kmalloc */ 651 ret = -ENOMEM; 652 if (size > (SIZE_MAX / 6)) 653 goto out_nofds; 654 655 alloc_size = 6 * size; 656 bits = kvmalloc(alloc_size, GFP_KERNEL); 657 if (!bits) 658 goto out_nofds; 659 } 660 fds.in = bits; 661 fds.out = bits + size; 662 fds.ex = bits + 2*size; 663 fds.res_in = bits + 3*size; 664 fds.res_out = bits + 4*size; 665 fds.res_ex = bits + 5*size; 666 667 if ((ret = get_fd_set(n, inp, fds.in)) || 668 (ret = get_fd_set(n, outp, fds.out)) || 669 (ret = get_fd_set(n, exp, fds.ex))) 670 goto out; 671 zero_fd_set(n, fds.res_in); 672 zero_fd_set(n, fds.res_out); 673 zero_fd_set(n, fds.res_ex); 674 675 ret = do_select(n, &fds, end_time); 676 677 if (ret < 0) 678 goto out; 679 if (!ret) { 680 ret = -ERESTARTNOHAND; 681 if (signal_pending(current)) 682 goto out; 683 ret = 0; 684 } 685 686 if (set_fd_set(n, inp, fds.res_in) || 687 set_fd_set(n, outp, fds.res_out) || 688 set_fd_set(n, exp, fds.res_ex)) 689 ret = -EFAULT; 690 691 out: 692 if (bits != stack_fds) 693 kvfree(bits); 694 out_nofds: 695 return ret; 696 } 697 698 static int kern_select(int n, fd_set __user *inp, fd_set __user *outp, 699 fd_set __user *exp, struct timeval __user *tvp) 700 { 701 struct timespec64 end_time, *to = NULL; 702 struct timeval tv; 703 int ret; 704 705 if (tvp) { 706 if (copy_from_user(&tv, tvp, sizeof(tv))) 707 return -EFAULT; 708 709 to = &end_time; 710 if (poll_select_set_timeout(to, 711 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), 712 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) 713 return -EINVAL; 714 } 715 716 ret = core_sys_select(n, inp, outp, exp, to); 717 ret = poll_select_copy_remaining(&end_time, tvp, PT_TIMEVAL, ret); 718 719 return ret; 720 } 721 722 SYSCALL_DEFINE5(select, int, n, fd_set __user *, inp, fd_set __user *, outp, 723 fd_set __user *, exp, struct timeval __user *, tvp) 724 { 725 return kern_select(n, inp, outp, exp, tvp); 726 } 727 728 static long do_pselect(int n, fd_set __user *inp, fd_set __user *outp, 729 fd_set __user *exp, void __user *tsp, 730 const sigset_t __user *sigmask, size_t sigsetsize, 731 enum poll_time_type type) 732 { 733 sigset_t ksigmask, sigsaved; 734 struct timespec64 ts, end_time, *to = NULL; 735 int ret; 736 737 if (tsp) { 738 switch (type) { 739 case PT_TIMESPEC: 740 if (get_timespec64(&ts, tsp)) 741 return -EFAULT; 742 break; 743 case PT_OLD_TIMESPEC: 744 if (get_old_timespec32(&ts, tsp)) 745 return -EFAULT; 746 break; 747 default: 748 BUG(); 749 } 750 751 to = &end_time; 752 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 753 return -EINVAL; 754 } 755 756 ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 757 if (ret) 758 return ret; 759 760 ret = core_sys_select(n, inp, outp, exp, to); 761 ret = poll_select_copy_remaining(&end_time, tsp, type, ret); 762 763 restore_user_sigmask(sigmask, &sigsaved); 764 765 return ret; 766 } 767 768 /* 769 * Most architectures can't handle 7-argument syscalls. So we provide a 770 * 6-argument version where the sixth argument is a pointer to a structure 771 * which has a pointer to the sigset_t itself followed by a size_t containing 772 * the sigset size. 773 */ 774 SYSCALL_DEFINE6(pselect6, int, n, fd_set __user *, inp, fd_set __user *, outp, 775 fd_set __user *, exp, struct __kernel_timespec __user *, tsp, 776 void __user *, sig) 777 { 778 size_t sigsetsize = 0; 779 sigset_t __user *up = NULL; 780 781 if (sig) { 782 if (!access_ok(sig, sizeof(void *)+sizeof(size_t)) 783 || __get_user(up, (sigset_t __user * __user *)sig) 784 || __get_user(sigsetsize, 785 (size_t __user *)(sig+sizeof(void *)))) 786 return -EFAULT; 787 } 788 789 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_TIMESPEC); 790 } 791 792 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) 793 794 SYSCALL_DEFINE6(pselect6_time32, int, n, fd_set __user *, inp, fd_set __user *, outp, 795 fd_set __user *, exp, struct old_timespec32 __user *, tsp, 796 void __user *, sig) 797 { 798 size_t sigsetsize = 0; 799 sigset_t __user *up = NULL; 800 801 if (sig) { 802 if (!access_ok(sig, sizeof(void *)+sizeof(size_t)) 803 || __get_user(up, (sigset_t __user * __user *)sig) 804 || __get_user(sigsetsize, 805 (size_t __user *)(sig+sizeof(void *)))) 806 return -EFAULT; 807 } 808 809 return do_pselect(n, inp, outp, exp, tsp, up, sigsetsize, PT_OLD_TIMESPEC); 810 } 811 812 #endif 813 814 #ifdef __ARCH_WANT_SYS_OLD_SELECT 815 struct sel_arg_struct { 816 unsigned long n; 817 fd_set __user *inp, *outp, *exp; 818 struct timeval __user *tvp; 819 }; 820 821 SYSCALL_DEFINE1(old_select, struct sel_arg_struct __user *, arg) 822 { 823 struct sel_arg_struct a; 824 825 if (copy_from_user(&a, arg, sizeof(a))) 826 return -EFAULT; 827 return kern_select(a.n, a.inp, a.outp, a.exp, a.tvp); 828 } 829 #endif 830 831 struct poll_list { 832 struct poll_list *next; 833 int len; 834 struct pollfd entries[0]; 835 }; 836 837 #define POLLFD_PER_PAGE ((PAGE_SIZE-sizeof(struct poll_list)) / sizeof(struct pollfd)) 838 839 /* 840 * Fish for pollable events on the pollfd->fd file descriptor. We're only 841 * interested in events matching the pollfd->events mask, and the result 842 * matching that mask is both recorded in pollfd->revents and returned. The 843 * pwait poll_table will be used by the fd-provided poll handler for waiting, 844 * if pwait->_qproc is non-NULL. 845 */ 846 static inline __poll_t do_pollfd(struct pollfd *pollfd, poll_table *pwait, 847 bool *can_busy_poll, 848 __poll_t busy_flag) 849 { 850 int fd = pollfd->fd; 851 __poll_t mask = 0, filter; 852 struct fd f; 853 854 if (fd < 0) 855 goto out; 856 mask = EPOLLNVAL; 857 f = fdget(fd); 858 if (!f.file) 859 goto out; 860 861 /* userland u16 ->events contains POLL... bitmap */ 862 filter = demangle_poll(pollfd->events) | EPOLLERR | EPOLLHUP; 863 pwait->_key = filter | busy_flag; 864 mask = vfs_poll(f.file, pwait); 865 if (mask & busy_flag) 866 *can_busy_poll = true; 867 mask &= filter; /* Mask out unneeded events. */ 868 fdput(f); 869 870 out: 871 /* ... and so does ->revents */ 872 pollfd->revents = mangle_poll(mask); 873 return mask; 874 } 875 876 static int do_poll(struct poll_list *list, struct poll_wqueues *wait, 877 struct timespec64 *end_time) 878 { 879 poll_table* pt = &wait->pt; 880 ktime_t expire, *to = NULL; 881 int timed_out = 0, count = 0; 882 u64 slack = 0; 883 __poll_t busy_flag = net_busy_loop_on() ? POLL_BUSY_LOOP : 0; 884 unsigned long busy_start = 0; 885 886 /* Optimise the no-wait case */ 887 if (end_time && !end_time->tv_sec && !end_time->tv_nsec) { 888 pt->_qproc = NULL; 889 timed_out = 1; 890 } 891 892 if (end_time && !timed_out) 893 slack = select_estimate_accuracy(end_time); 894 895 for (;;) { 896 struct poll_list *walk; 897 bool can_busy_loop = false; 898 899 for (walk = list; walk != NULL; walk = walk->next) { 900 struct pollfd * pfd, * pfd_end; 901 902 pfd = walk->entries; 903 pfd_end = pfd + walk->len; 904 for (; pfd != pfd_end; pfd++) { 905 /* 906 * Fish for events. If we found one, record it 907 * and kill poll_table->_qproc, so we don't 908 * needlessly register any other waiters after 909 * this. They'll get immediately deregistered 910 * when we break out and return. 911 */ 912 if (do_pollfd(pfd, pt, &can_busy_loop, 913 busy_flag)) { 914 count++; 915 pt->_qproc = NULL; 916 /* found something, stop busy polling */ 917 busy_flag = 0; 918 can_busy_loop = false; 919 } 920 } 921 } 922 /* 923 * All waiters have already been registered, so don't provide 924 * a poll_table->_qproc to them on the next loop iteration. 925 */ 926 pt->_qproc = NULL; 927 if (!count) { 928 count = wait->error; 929 if (signal_pending(current)) 930 count = -EINTR; 931 } 932 if (count || timed_out) 933 break; 934 935 /* only if found POLL_BUSY_LOOP sockets && not out of time */ 936 if (can_busy_loop && !need_resched()) { 937 if (!busy_start) { 938 busy_start = busy_loop_current_time(); 939 continue; 940 } 941 if (!busy_loop_timeout(busy_start)) 942 continue; 943 } 944 busy_flag = 0; 945 946 /* 947 * If this is the first loop and we have a timeout 948 * given, then we convert to ktime_t and set the to 949 * pointer to the expiry value. 950 */ 951 if (end_time && !to) { 952 expire = timespec64_to_ktime(*end_time); 953 to = &expire; 954 } 955 956 if (!poll_schedule_timeout(wait, TASK_INTERRUPTIBLE, to, slack)) 957 timed_out = 1; 958 } 959 return count; 960 } 961 962 #define N_STACK_PPS ((sizeof(stack_pps) - sizeof(struct poll_list)) / \ 963 sizeof(struct pollfd)) 964 965 static int do_sys_poll(struct pollfd __user *ufds, unsigned int nfds, 966 struct timespec64 *end_time) 967 { 968 struct poll_wqueues table; 969 int err = -EFAULT, fdcount, len, size; 970 /* Allocate small arguments on the stack to save memory and be 971 faster - use long to make sure the buffer is aligned properly 972 on 64 bit archs to avoid unaligned access */ 973 long stack_pps[POLL_STACK_ALLOC/sizeof(long)]; 974 struct poll_list *const head = (struct poll_list *)stack_pps; 975 struct poll_list *walk = head; 976 unsigned long todo = nfds; 977 978 if (nfds > rlimit(RLIMIT_NOFILE)) 979 return -EINVAL; 980 981 len = min_t(unsigned int, nfds, N_STACK_PPS); 982 for (;;) { 983 walk->next = NULL; 984 walk->len = len; 985 if (!len) 986 break; 987 988 if (copy_from_user(walk->entries, ufds + nfds-todo, 989 sizeof(struct pollfd) * walk->len)) 990 goto out_fds; 991 992 todo -= walk->len; 993 if (!todo) 994 break; 995 996 len = min(todo, POLLFD_PER_PAGE); 997 size = sizeof(struct poll_list) + sizeof(struct pollfd) * len; 998 walk = walk->next = kmalloc(size, GFP_KERNEL); 999 if (!walk) { 1000 err = -ENOMEM; 1001 goto out_fds; 1002 } 1003 } 1004 1005 poll_initwait(&table); 1006 fdcount = do_poll(head, &table, end_time); 1007 poll_freewait(&table); 1008 1009 for (walk = head; walk; walk = walk->next) { 1010 struct pollfd *fds = walk->entries; 1011 int j; 1012 1013 for (j = 0; j < walk->len; j++, ufds++) 1014 if (__put_user(fds[j].revents, &ufds->revents)) 1015 goto out_fds; 1016 } 1017 1018 err = fdcount; 1019 out_fds: 1020 walk = head->next; 1021 while (walk) { 1022 struct poll_list *pos = walk; 1023 walk = walk->next; 1024 kfree(pos); 1025 } 1026 1027 return err; 1028 } 1029 1030 static long do_restart_poll(struct restart_block *restart_block) 1031 { 1032 struct pollfd __user *ufds = restart_block->poll.ufds; 1033 int nfds = restart_block->poll.nfds; 1034 struct timespec64 *to = NULL, end_time; 1035 int ret; 1036 1037 if (restart_block->poll.has_timeout) { 1038 end_time.tv_sec = restart_block->poll.tv_sec; 1039 end_time.tv_nsec = restart_block->poll.tv_nsec; 1040 to = &end_time; 1041 } 1042 1043 ret = do_sys_poll(ufds, nfds, to); 1044 1045 if (ret == -EINTR) { 1046 restart_block->fn = do_restart_poll; 1047 ret = -ERESTART_RESTARTBLOCK; 1048 } 1049 return ret; 1050 } 1051 1052 SYSCALL_DEFINE3(poll, struct pollfd __user *, ufds, unsigned int, nfds, 1053 int, timeout_msecs) 1054 { 1055 struct timespec64 end_time, *to = NULL; 1056 int ret; 1057 1058 if (timeout_msecs >= 0) { 1059 to = &end_time; 1060 poll_select_set_timeout(to, timeout_msecs / MSEC_PER_SEC, 1061 NSEC_PER_MSEC * (timeout_msecs % MSEC_PER_SEC)); 1062 } 1063 1064 ret = do_sys_poll(ufds, nfds, to); 1065 1066 if (ret == -EINTR) { 1067 struct restart_block *restart_block; 1068 1069 restart_block = ¤t->restart_block; 1070 restart_block->fn = do_restart_poll; 1071 restart_block->poll.ufds = ufds; 1072 restart_block->poll.nfds = nfds; 1073 1074 if (timeout_msecs >= 0) { 1075 restart_block->poll.tv_sec = end_time.tv_sec; 1076 restart_block->poll.tv_nsec = end_time.tv_nsec; 1077 restart_block->poll.has_timeout = 1; 1078 } else 1079 restart_block->poll.has_timeout = 0; 1080 1081 ret = -ERESTART_RESTARTBLOCK; 1082 } 1083 return ret; 1084 } 1085 1086 SYSCALL_DEFINE5(ppoll, struct pollfd __user *, ufds, unsigned int, nfds, 1087 struct __kernel_timespec __user *, tsp, const sigset_t __user *, sigmask, 1088 size_t, sigsetsize) 1089 { 1090 sigset_t ksigmask, sigsaved; 1091 struct timespec64 ts, end_time, *to = NULL; 1092 int ret; 1093 1094 if (tsp) { 1095 if (get_timespec64(&ts, tsp)) 1096 return -EFAULT; 1097 1098 to = &end_time; 1099 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 1100 return -EINVAL; 1101 } 1102 1103 ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 1104 if (ret) 1105 return ret; 1106 1107 ret = do_sys_poll(ufds, nfds, to); 1108 1109 restore_user_sigmask(sigmask, &sigsaved); 1110 1111 /* We can restart this syscall, usually */ 1112 if (ret == -EINTR) 1113 ret = -ERESTARTNOHAND; 1114 1115 ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret); 1116 1117 return ret; 1118 } 1119 1120 #if defined(CONFIG_COMPAT_32BIT_TIME) && !defined(CONFIG_64BIT) 1121 1122 SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, unsigned int, nfds, 1123 struct old_timespec32 __user *, tsp, const sigset_t __user *, sigmask, 1124 size_t, sigsetsize) 1125 { 1126 sigset_t ksigmask, sigsaved; 1127 struct timespec64 ts, end_time, *to = NULL; 1128 int ret; 1129 1130 if (tsp) { 1131 if (get_old_timespec32(&ts, tsp)) 1132 return -EFAULT; 1133 1134 to = &end_time; 1135 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 1136 return -EINVAL; 1137 } 1138 1139 ret = set_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 1140 if (ret) 1141 return ret; 1142 1143 ret = do_sys_poll(ufds, nfds, to); 1144 1145 restore_user_sigmask(sigmask, &sigsaved); 1146 1147 /* We can restart this syscall, usually */ 1148 if (ret == -EINTR) 1149 ret = -ERESTARTNOHAND; 1150 1151 ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret); 1152 1153 return ret; 1154 } 1155 #endif 1156 1157 #ifdef CONFIG_COMPAT 1158 #define __COMPAT_NFDBITS (8 * sizeof(compat_ulong_t)) 1159 1160 /* 1161 * Ooo, nasty. We need here to frob 32-bit unsigned longs to 1162 * 64-bit unsigned longs. 1163 */ 1164 static 1165 int compat_get_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, 1166 unsigned long *fdset) 1167 { 1168 if (ufdset) { 1169 return compat_get_bitmap(fdset, ufdset, nr); 1170 } else { 1171 zero_fd_set(nr, fdset); 1172 return 0; 1173 } 1174 } 1175 1176 static 1177 int compat_set_fd_set(unsigned long nr, compat_ulong_t __user *ufdset, 1178 unsigned long *fdset) 1179 { 1180 if (!ufdset) 1181 return 0; 1182 return compat_put_bitmap(ufdset, fdset, nr); 1183 } 1184 1185 1186 /* 1187 * This is a virtual copy of sys_select from fs/select.c and probably 1188 * should be compared to it from time to time 1189 */ 1190 1191 /* 1192 * We can actually return ERESTARTSYS instead of EINTR, but I'd 1193 * like to be certain this leads to no problems. So I return 1194 * EINTR just for safety. 1195 * 1196 * Update: ERESTARTSYS breaks at least the xview clock binary, so 1197 * I'm trying ERESTARTNOHAND which restart only when you want to. 1198 */ 1199 static int compat_core_sys_select(int n, compat_ulong_t __user *inp, 1200 compat_ulong_t __user *outp, compat_ulong_t __user *exp, 1201 struct timespec64 *end_time) 1202 { 1203 fd_set_bits fds; 1204 void *bits; 1205 int size, max_fds, ret = -EINVAL; 1206 struct fdtable *fdt; 1207 long stack_fds[SELECT_STACK_ALLOC/sizeof(long)]; 1208 1209 if (n < 0) 1210 goto out_nofds; 1211 1212 /* max_fds can increase, so grab it once to avoid race */ 1213 rcu_read_lock(); 1214 fdt = files_fdtable(current->files); 1215 max_fds = fdt->max_fds; 1216 rcu_read_unlock(); 1217 if (n > max_fds) 1218 n = max_fds; 1219 1220 /* 1221 * We need 6 bitmaps (in/out/ex for both incoming and outgoing), 1222 * since we used fdset we need to allocate memory in units of 1223 * long-words. 1224 */ 1225 size = FDS_BYTES(n); 1226 bits = stack_fds; 1227 if (size > sizeof(stack_fds) / 6) { 1228 bits = kmalloc_array(6, size, GFP_KERNEL); 1229 ret = -ENOMEM; 1230 if (!bits) 1231 goto out_nofds; 1232 } 1233 fds.in = (unsigned long *) bits; 1234 fds.out = (unsigned long *) (bits + size); 1235 fds.ex = (unsigned long *) (bits + 2*size); 1236 fds.res_in = (unsigned long *) (bits + 3*size); 1237 fds.res_out = (unsigned long *) (bits + 4*size); 1238 fds.res_ex = (unsigned long *) (bits + 5*size); 1239 1240 if ((ret = compat_get_fd_set(n, inp, fds.in)) || 1241 (ret = compat_get_fd_set(n, outp, fds.out)) || 1242 (ret = compat_get_fd_set(n, exp, fds.ex))) 1243 goto out; 1244 zero_fd_set(n, fds.res_in); 1245 zero_fd_set(n, fds.res_out); 1246 zero_fd_set(n, fds.res_ex); 1247 1248 ret = do_select(n, &fds, end_time); 1249 1250 if (ret < 0) 1251 goto out; 1252 if (!ret) { 1253 ret = -ERESTARTNOHAND; 1254 if (signal_pending(current)) 1255 goto out; 1256 ret = 0; 1257 } 1258 1259 if (compat_set_fd_set(n, inp, fds.res_in) || 1260 compat_set_fd_set(n, outp, fds.res_out) || 1261 compat_set_fd_set(n, exp, fds.res_ex)) 1262 ret = -EFAULT; 1263 out: 1264 if (bits != stack_fds) 1265 kfree(bits); 1266 out_nofds: 1267 return ret; 1268 } 1269 1270 static int do_compat_select(int n, compat_ulong_t __user *inp, 1271 compat_ulong_t __user *outp, compat_ulong_t __user *exp, 1272 struct old_timeval32 __user *tvp) 1273 { 1274 struct timespec64 end_time, *to = NULL; 1275 struct old_timeval32 tv; 1276 int ret; 1277 1278 if (tvp) { 1279 if (copy_from_user(&tv, tvp, sizeof(tv))) 1280 return -EFAULT; 1281 1282 to = &end_time; 1283 if (poll_select_set_timeout(to, 1284 tv.tv_sec + (tv.tv_usec / USEC_PER_SEC), 1285 (tv.tv_usec % USEC_PER_SEC) * NSEC_PER_USEC)) 1286 return -EINVAL; 1287 } 1288 1289 ret = compat_core_sys_select(n, inp, outp, exp, to); 1290 ret = poll_select_copy_remaining(&end_time, tvp, PT_OLD_TIMEVAL, ret); 1291 1292 return ret; 1293 } 1294 1295 COMPAT_SYSCALL_DEFINE5(select, int, n, compat_ulong_t __user *, inp, 1296 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, 1297 struct old_timeval32 __user *, tvp) 1298 { 1299 return do_compat_select(n, inp, outp, exp, tvp); 1300 } 1301 1302 struct compat_sel_arg_struct { 1303 compat_ulong_t n; 1304 compat_uptr_t inp; 1305 compat_uptr_t outp; 1306 compat_uptr_t exp; 1307 compat_uptr_t tvp; 1308 }; 1309 1310 COMPAT_SYSCALL_DEFINE1(old_select, struct compat_sel_arg_struct __user *, arg) 1311 { 1312 struct compat_sel_arg_struct a; 1313 1314 if (copy_from_user(&a, arg, sizeof(a))) 1315 return -EFAULT; 1316 return do_compat_select(a.n, compat_ptr(a.inp), compat_ptr(a.outp), 1317 compat_ptr(a.exp), compat_ptr(a.tvp)); 1318 } 1319 1320 static long do_compat_pselect(int n, compat_ulong_t __user *inp, 1321 compat_ulong_t __user *outp, compat_ulong_t __user *exp, 1322 void __user *tsp, compat_sigset_t __user *sigmask, 1323 compat_size_t sigsetsize, enum poll_time_type type) 1324 { 1325 sigset_t ksigmask, sigsaved; 1326 struct timespec64 ts, end_time, *to = NULL; 1327 int ret; 1328 1329 if (tsp) { 1330 switch (type) { 1331 case PT_OLD_TIMESPEC: 1332 if (get_old_timespec32(&ts, tsp)) 1333 return -EFAULT; 1334 break; 1335 case PT_TIMESPEC: 1336 if (get_timespec64(&ts, tsp)) 1337 return -EFAULT; 1338 break; 1339 default: 1340 BUG(); 1341 } 1342 1343 to = &end_time; 1344 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 1345 return -EINVAL; 1346 } 1347 1348 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 1349 if (ret) 1350 return ret; 1351 1352 ret = compat_core_sys_select(n, inp, outp, exp, to); 1353 ret = poll_select_copy_remaining(&end_time, tsp, type, ret); 1354 1355 restore_user_sigmask(sigmask, &sigsaved); 1356 1357 return ret; 1358 } 1359 1360 COMPAT_SYSCALL_DEFINE6(pselect6_time64, int, n, compat_ulong_t __user *, inp, 1361 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, 1362 struct __kernel_timespec __user *, tsp, void __user *, sig) 1363 { 1364 compat_size_t sigsetsize = 0; 1365 compat_uptr_t up = 0; 1366 1367 if (sig) { 1368 if (!access_ok(sig, 1369 sizeof(compat_uptr_t)+sizeof(compat_size_t)) || 1370 __get_user(up, (compat_uptr_t __user *)sig) || 1371 __get_user(sigsetsize, 1372 (compat_size_t __user *)(sig+sizeof(up)))) 1373 return -EFAULT; 1374 } 1375 1376 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up), 1377 sigsetsize, PT_TIMESPEC); 1378 } 1379 1380 #if defined(CONFIG_COMPAT_32BIT_TIME) 1381 1382 COMPAT_SYSCALL_DEFINE6(pselect6_time32, int, n, compat_ulong_t __user *, inp, 1383 compat_ulong_t __user *, outp, compat_ulong_t __user *, exp, 1384 struct old_timespec32 __user *, tsp, void __user *, sig) 1385 { 1386 compat_size_t sigsetsize = 0; 1387 compat_uptr_t up = 0; 1388 1389 if (sig) { 1390 if (!access_ok(sig, 1391 sizeof(compat_uptr_t)+sizeof(compat_size_t)) || 1392 __get_user(up, (compat_uptr_t __user *)sig) || 1393 __get_user(sigsetsize, 1394 (compat_size_t __user *)(sig+sizeof(up)))) 1395 return -EFAULT; 1396 } 1397 1398 return do_compat_pselect(n, inp, outp, exp, tsp, compat_ptr(up), 1399 sigsetsize, PT_OLD_TIMESPEC); 1400 } 1401 1402 #endif 1403 1404 #if defined(CONFIG_COMPAT_32BIT_TIME) 1405 COMPAT_SYSCALL_DEFINE5(ppoll_time32, struct pollfd __user *, ufds, 1406 unsigned int, nfds, struct old_timespec32 __user *, tsp, 1407 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) 1408 { 1409 sigset_t ksigmask, sigsaved; 1410 struct timespec64 ts, end_time, *to = NULL; 1411 int ret; 1412 1413 if (tsp) { 1414 if (get_old_timespec32(&ts, tsp)) 1415 return -EFAULT; 1416 1417 to = &end_time; 1418 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 1419 return -EINVAL; 1420 } 1421 1422 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 1423 if (ret) 1424 return ret; 1425 1426 ret = do_sys_poll(ufds, nfds, to); 1427 1428 restore_user_sigmask(sigmask, &sigsaved); 1429 1430 /* We can restart this syscall, usually */ 1431 if (ret == -EINTR) 1432 ret = -ERESTARTNOHAND; 1433 1434 ret = poll_select_copy_remaining(&end_time, tsp, PT_OLD_TIMESPEC, ret); 1435 1436 return ret; 1437 } 1438 #endif 1439 1440 /* New compat syscall for 64 bit time_t*/ 1441 COMPAT_SYSCALL_DEFINE5(ppoll_time64, struct pollfd __user *, ufds, 1442 unsigned int, nfds, struct __kernel_timespec __user *, tsp, 1443 const compat_sigset_t __user *, sigmask, compat_size_t, sigsetsize) 1444 { 1445 sigset_t ksigmask, sigsaved; 1446 struct timespec64 ts, end_time, *to = NULL; 1447 int ret; 1448 1449 if (tsp) { 1450 if (get_timespec64(&ts, tsp)) 1451 return -EFAULT; 1452 1453 to = &end_time; 1454 if (poll_select_set_timeout(to, ts.tv_sec, ts.tv_nsec)) 1455 return -EINVAL; 1456 } 1457 1458 ret = set_compat_user_sigmask(sigmask, &ksigmask, &sigsaved, sigsetsize); 1459 if (ret) 1460 return ret; 1461 1462 ret = do_sys_poll(ufds, nfds, to); 1463 1464 restore_user_sigmask(sigmask, &sigsaved); 1465 1466 /* We can restart this syscall, usually */ 1467 if (ret == -EINTR) 1468 ret = -ERESTARTNOHAND; 1469 1470 ret = poll_select_copy_remaining(&end_time, tsp, PT_TIMESPEC, ret); 1471 1472 return ret; 1473 } 1474 1475 #endif 1476