1 /* 2 * SUCS NET3: 3 * 4 * Generic datagram handling routines. These are generic for all 5 * protocols. Possibly a generic IP version on top of these would 6 * make sense. Not tonight however 8-). 7 * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and 8 * NetROM layer all have identical poll code and mostly 9 * identical recvmsg() code. So we share it here. The poll was 10 * shared before but buried in udp.c so I moved it. 11 * 12 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old 13 * udp.c code) 14 * 15 * Fixes: 16 * Alan Cox : NULL return from skb_peek_copy() 17 * understood 18 * Alan Cox : Rewrote skb_read_datagram to avoid the 19 * skb_peek_copy stuff. 20 * Alan Cox : Added support for SOCK_SEQPACKET. 21 * IPX can no longer use the SO_TYPE hack 22 * but AX.25 now works right, and SPX is 23 * feasible. 24 * Alan Cox : Fixed write poll of non IP protocol 25 * crash. 26 * Florian La Roche: Changed for my new skbuff handling. 27 * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. 28 * Linus Torvalds : BSD semantic fixes. 29 * Alan Cox : Datagram iovec handling 30 * Darryl Miles : Fixed non-blocking SOCK_STREAM. 31 * Alan Cox : POSIXisms 32 * Pete Wyckoff : Unconnected accept() fix. 33 * 34 */ 35 36 #include <linux/module.h> 37 #include <linux/types.h> 38 #include <linux/kernel.h> 39 #include <asm/uaccess.h> 40 #include <linux/mm.h> 41 #include <linux/interrupt.h> 42 #include <linux/errno.h> 43 #include <linux/sched.h> 44 #include <linux/inet.h> 45 #include <linux/netdevice.h> 46 #include <linux/rtnetlink.h> 47 #include <linux/poll.h> 48 #include <linux/highmem.h> 49 #include <linux/spinlock.h> 50 #include <linux/slab.h> 51 #include <linux/pagemap.h> 52 #include <linux/uio.h> 53 54 #include <net/protocol.h> 55 #include <linux/skbuff.h> 56 57 #include <net/checksum.h> 58 #include <net/sock.h> 59 #include <net/tcp_states.h> 60 #include <trace/events/skb.h> 61 #include <net/busy_poll.h> 62 63 /* 64 * Is a socket 'connection oriented' ? 65 */ 66 static inline int connection_based(struct sock *sk) 67 { 68 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; 69 } 70 71 static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync, 72 void *key) 73 { 74 unsigned long bits = (unsigned long)key; 75 76 /* 77 * Avoid a wakeup if event not interesting for us 78 */ 79 if (bits && !(bits & (POLLIN | POLLERR))) 80 return 0; 81 return autoremove_wake_function(wait, mode, sync, key); 82 } 83 /* 84 * Wait for the last received packet to be different from skb 85 */ 86 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 87 const struct sk_buff *skb) 88 { 89 int error; 90 DEFINE_WAIT_FUNC(wait, receiver_wake_function); 91 92 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 93 94 /* Socket errors? */ 95 error = sock_error(sk); 96 if (error) 97 goto out_err; 98 99 if (sk->sk_receive_queue.prev != skb) 100 goto out; 101 102 /* Socket shut down? */ 103 if (sk->sk_shutdown & RCV_SHUTDOWN) 104 goto out_noerr; 105 106 /* Sequenced packets can come disconnected. 107 * If so we report the problem 108 */ 109 error = -ENOTCONN; 110 if (connection_based(sk) && 111 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) 112 goto out_err; 113 114 /* handle signals */ 115 if (signal_pending(current)) 116 goto interrupted; 117 118 error = 0; 119 *timeo_p = schedule_timeout(*timeo_p); 120 out: 121 finish_wait(sk_sleep(sk), &wait); 122 return error; 123 interrupted: 124 error = sock_intr_errno(*timeo_p); 125 out_err: 126 *err = error; 127 goto out; 128 out_noerr: 129 *err = 0; 130 error = 1; 131 goto out; 132 } 133 EXPORT_SYMBOL(__skb_wait_for_more_packets); 134 135 static struct sk_buff *skb_set_peeked(struct sk_buff *skb) 136 { 137 struct sk_buff *nskb; 138 139 if (skb->peeked) 140 return skb; 141 142 /* We have to unshare an skb before modifying it. */ 143 if (!skb_shared(skb)) 144 goto done; 145 146 nskb = skb_clone(skb, GFP_ATOMIC); 147 if (!nskb) 148 return ERR_PTR(-ENOMEM); 149 150 skb->prev->next = nskb; 151 skb->next->prev = nskb; 152 nskb->prev = skb->prev; 153 nskb->next = skb->next; 154 155 consume_skb(skb); 156 skb = nskb; 157 158 done: 159 skb->peeked = 1; 160 161 return skb; 162 } 163 164 /** 165 * __skb_try_recv_datagram - Receive a datagram skbuff 166 * @sk: socket 167 * @flags: MSG_ flags 168 * @peeked: returns non-zero if this packet has been seen before 169 * @off: an offset in bytes to peek skb from. Returns an offset 170 * within an skb where data actually starts 171 * @err: error code returned 172 * @last: set to last peeked message to inform the wait function 173 * what to look for when peeking 174 * 175 * Get a datagram skbuff, understands the peeking, nonblocking wakeups 176 * and possible races. This replaces identical code in packet, raw and 177 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes 178 * the long standing peek and read race for datagram sockets. If you 179 * alter this routine remember it must be re-entrant. 180 * 181 * This function will lock the socket if a skb is returned, so 182 * the caller needs to unlock the socket in that case (usually by 183 * calling skb_free_datagram). Returns NULL with *err set to 184 * -EAGAIN if no data was available or to some other value if an 185 * error was detected. 186 * 187 * * It does not lock socket since today. This function is 188 * * free of race conditions. This measure should/can improve 189 * * significantly datagram socket latencies at high loads, 190 * * when data copying to user space takes lots of time. 191 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet 192 * * 8) Great win.) 193 * * --ANK (980729) 194 * 195 * The order of the tests when we find no data waiting are specified 196 * quite explicitly by POSIX 1003.1g, don't change them without having 197 * the standard around please. 198 */ 199 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags, 200 int *peeked, int *off, int *err, 201 struct sk_buff **last) 202 { 203 struct sk_buff_head *queue = &sk->sk_receive_queue; 204 struct sk_buff *skb; 205 unsigned long cpu_flags; 206 /* 207 * Caller is allowed not to check sk->sk_err before skb_recv_datagram() 208 */ 209 int error = sock_error(sk); 210 211 if (error) 212 goto no_packet; 213 214 do { 215 /* Again only user level code calls this function, so nothing 216 * interrupt level will suddenly eat the receive_queue. 217 * 218 * Look at current nfs client by the way... 219 * However, this function was correct in any case. 8) 220 */ 221 int _off = *off; 222 223 *last = (struct sk_buff *)queue; 224 spin_lock_irqsave(&queue->lock, cpu_flags); 225 skb_queue_walk(queue, skb) { 226 *last = skb; 227 *peeked = skb->peeked; 228 if (flags & MSG_PEEK) { 229 if (_off >= skb->len && (skb->len || _off || 230 skb->peeked)) { 231 _off -= skb->len; 232 continue; 233 } 234 235 skb = skb_set_peeked(skb); 236 error = PTR_ERR(skb); 237 if (IS_ERR(skb)) { 238 spin_unlock_irqrestore(&queue->lock, 239 cpu_flags); 240 goto no_packet; 241 } 242 243 atomic_inc(&skb->users); 244 } else 245 __skb_unlink(skb, queue); 246 247 spin_unlock_irqrestore(&queue->lock, cpu_flags); 248 *off = _off; 249 return skb; 250 } 251 252 spin_unlock_irqrestore(&queue->lock, cpu_flags); 253 } while (sk_can_busy_loop(sk) && 254 sk_busy_loop(sk, flags & MSG_DONTWAIT)); 255 256 error = -EAGAIN; 257 258 no_packet: 259 *err = error; 260 return NULL; 261 } 262 EXPORT_SYMBOL(__skb_try_recv_datagram); 263 264 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags, 265 int *peeked, int *off, int *err) 266 { 267 struct sk_buff *skb, *last; 268 long timeo; 269 270 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 271 272 do { 273 skb = __skb_try_recv_datagram(sk, flags, peeked, off, err, 274 &last); 275 if (skb) 276 return skb; 277 278 if (*err != -EAGAIN) 279 break; 280 } while (timeo && 281 !__skb_wait_for_more_packets(sk, err, &timeo, last)); 282 283 return NULL; 284 } 285 EXPORT_SYMBOL(__skb_recv_datagram); 286 287 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, 288 int noblock, int *err) 289 { 290 int peeked, off = 0; 291 292 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0), 293 &peeked, &off, err); 294 } 295 EXPORT_SYMBOL(skb_recv_datagram); 296 297 void skb_free_datagram(struct sock *sk, struct sk_buff *skb) 298 { 299 consume_skb(skb); 300 sk_mem_reclaim_partial(sk); 301 } 302 EXPORT_SYMBOL(skb_free_datagram); 303 304 void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb) 305 { 306 bool slow; 307 308 if (likely(atomic_read(&skb->users) == 1)) 309 smp_rmb(); 310 else if (likely(!atomic_dec_and_test(&skb->users))) 311 return; 312 313 slow = lock_sock_fast(sk); 314 skb_orphan(skb); 315 sk_mem_reclaim_partial(sk); 316 unlock_sock_fast(sk, slow); 317 318 /* skb is now orphaned, can be freed outside of locked section */ 319 __kfree_skb(skb); 320 } 321 EXPORT_SYMBOL(skb_free_datagram_locked); 322 323 /** 324 * skb_kill_datagram - Free a datagram skbuff forcibly 325 * @sk: socket 326 * @skb: datagram skbuff 327 * @flags: MSG_ flags 328 * 329 * This function frees a datagram skbuff that was received by 330 * skb_recv_datagram. The flags argument must match the one 331 * used for skb_recv_datagram. 332 * 333 * If the MSG_PEEK flag is set, and the packet is still on the 334 * receive queue of the socket, it will be taken off the queue 335 * before it is freed. 336 * 337 * This function currently only disables BH when acquiring the 338 * sk_receive_queue lock. Therefore it must not be used in a 339 * context where that lock is acquired in an IRQ context. 340 * 341 * It returns 0 if the packet was removed by us. 342 */ 343 344 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) 345 { 346 int err = 0; 347 348 if (flags & MSG_PEEK) { 349 err = -ENOENT; 350 spin_lock_bh(&sk->sk_receive_queue.lock); 351 if (skb == skb_peek(&sk->sk_receive_queue)) { 352 __skb_unlink(skb, &sk->sk_receive_queue); 353 atomic_dec(&skb->users); 354 err = 0; 355 } 356 spin_unlock_bh(&sk->sk_receive_queue.lock); 357 } 358 359 kfree_skb(skb); 360 atomic_inc(&sk->sk_drops); 361 sk_mem_reclaim_partial(sk); 362 363 return err; 364 } 365 EXPORT_SYMBOL(skb_kill_datagram); 366 367 /** 368 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. 369 * @skb: buffer to copy 370 * @offset: offset in the buffer to start copying from 371 * @to: iovec iterator to copy to 372 * @len: amount of data to copy from buffer to iovec 373 */ 374 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, 375 struct iov_iter *to, int len) 376 { 377 int start = skb_headlen(skb); 378 int i, copy = start - offset; 379 struct sk_buff *frag_iter; 380 381 trace_skb_copy_datagram_iovec(skb, len); 382 383 /* Copy header. */ 384 if (copy > 0) { 385 if (copy > len) 386 copy = len; 387 if (copy_to_iter(skb->data + offset, copy, to) != copy) 388 goto short_copy; 389 if ((len -= copy) == 0) 390 return 0; 391 offset += copy; 392 } 393 394 /* Copy paged appendix. Hmm... why does this look so complicated? */ 395 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 396 int end; 397 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 398 399 WARN_ON(start > offset + len); 400 401 end = start + skb_frag_size(frag); 402 if ((copy = end - offset) > 0) { 403 if (copy > len) 404 copy = len; 405 if (copy_page_to_iter(skb_frag_page(frag), 406 frag->page_offset + offset - 407 start, copy, to) != copy) 408 goto short_copy; 409 if (!(len -= copy)) 410 return 0; 411 offset += copy; 412 } 413 start = end; 414 } 415 416 skb_walk_frags(skb, frag_iter) { 417 int end; 418 419 WARN_ON(start > offset + len); 420 421 end = start + frag_iter->len; 422 if ((copy = end - offset) > 0) { 423 if (copy > len) 424 copy = len; 425 if (skb_copy_datagram_iter(frag_iter, offset - start, 426 to, copy)) 427 goto fault; 428 if ((len -= copy) == 0) 429 return 0; 430 offset += copy; 431 } 432 start = end; 433 } 434 if (!len) 435 return 0; 436 437 /* This is not really a user copy fault, but rather someone 438 * gave us a bogus length on the skb. We should probably 439 * print a warning here as it may indicate a kernel bug. 440 */ 441 442 fault: 443 return -EFAULT; 444 445 short_copy: 446 if (iov_iter_count(to)) 447 goto fault; 448 449 return 0; 450 } 451 EXPORT_SYMBOL(skb_copy_datagram_iter); 452 453 /** 454 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. 455 * @skb: buffer to copy 456 * @offset: offset in the buffer to start copying to 457 * @from: the copy source 458 * @len: amount of data to copy to buffer from iovec 459 * 460 * Returns 0 or -EFAULT. 461 */ 462 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 463 struct iov_iter *from, 464 int len) 465 { 466 int start = skb_headlen(skb); 467 int i, copy = start - offset; 468 struct sk_buff *frag_iter; 469 470 /* Copy header. */ 471 if (copy > 0) { 472 if (copy > len) 473 copy = len; 474 if (copy_from_iter(skb->data + offset, copy, from) != copy) 475 goto fault; 476 if ((len -= copy) == 0) 477 return 0; 478 offset += copy; 479 } 480 481 /* Copy paged appendix. Hmm... why does this look so complicated? */ 482 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 483 int end; 484 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 485 486 WARN_ON(start > offset + len); 487 488 end = start + skb_frag_size(frag); 489 if ((copy = end - offset) > 0) { 490 size_t copied; 491 492 if (copy > len) 493 copy = len; 494 copied = copy_page_from_iter(skb_frag_page(frag), 495 frag->page_offset + offset - start, 496 copy, from); 497 if (copied != copy) 498 goto fault; 499 500 if (!(len -= copy)) 501 return 0; 502 offset += copy; 503 } 504 start = end; 505 } 506 507 skb_walk_frags(skb, frag_iter) { 508 int end; 509 510 WARN_ON(start > offset + len); 511 512 end = start + frag_iter->len; 513 if ((copy = end - offset) > 0) { 514 if (copy > len) 515 copy = len; 516 if (skb_copy_datagram_from_iter(frag_iter, 517 offset - start, 518 from, copy)) 519 goto fault; 520 if ((len -= copy) == 0) 521 return 0; 522 offset += copy; 523 } 524 start = end; 525 } 526 if (!len) 527 return 0; 528 529 fault: 530 return -EFAULT; 531 } 532 EXPORT_SYMBOL(skb_copy_datagram_from_iter); 533 534 /** 535 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter 536 * @skb: buffer to copy 537 * @from: the source to copy from 538 * 539 * The function will first copy up to headlen, and then pin the userspace 540 * pages and build frags through them. 541 * 542 * Returns 0, -EFAULT or -EMSGSIZE. 543 */ 544 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) 545 { 546 int len = iov_iter_count(from); 547 int copy = min_t(int, skb_headlen(skb), len); 548 int frag = 0; 549 550 /* copy up to skb headlen */ 551 if (skb_copy_datagram_from_iter(skb, 0, from, copy)) 552 return -EFAULT; 553 554 while (iov_iter_count(from)) { 555 struct page *pages[MAX_SKB_FRAGS]; 556 size_t start; 557 ssize_t copied; 558 unsigned long truesize; 559 int n = 0; 560 561 if (frag == MAX_SKB_FRAGS) 562 return -EMSGSIZE; 563 564 copied = iov_iter_get_pages(from, pages, ~0U, 565 MAX_SKB_FRAGS - frag, &start); 566 if (copied < 0) 567 return -EFAULT; 568 569 iov_iter_advance(from, copied); 570 571 truesize = PAGE_ALIGN(copied + start); 572 skb->data_len += copied; 573 skb->len += copied; 574 skb->truesize += truesize; 575 atomic_add(truesize, &skb->sk->sk_wmem_alloc); 576 while (copied) { 577 int size = min_t(int, copied, PAGE_SIZE - start); 578 skb_fill_page_desc(skb, frag++, pages[n], start, size); 579 start = 0; 580 copied -= size; 581 n++; 582 } 583 } 584 return 0; 585 } 586 EXPORT_SYMBOL(zerocopy_sg_from_iter); 587 588 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, 589 struct iov_iter *to, int len, 590 __wsum *csump) 591 { 592 int start = skb_headlen(skb); 593 int i, copy = start - offset; 594 struct sk_buff *frag_iter; 595 int pos = 0; 596 int n; 597 598 /* Copy header. */ 599 if (copy > 0) { 600 if (copy > len) 601 copy = len; 602 n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to); 603 if (n != copy) 604 goto fault; 605 if ((len -= copy) == 0) 606 return 0; 607 offset += copy; 608 pos = copy; 609 } 610 611 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 612 int end; 613 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 614 615 WARN_ON(start > offset + len); 616 617 end = start + skb_frag_size(frag); 618 if ((copy = end - offset) > 0) { 619 __wsum csum2 = 0; 620 struct page *page = skb_frag_page(frag); 621 u8 *vaddr = kmap(page); 622 623 if (copy > len) 624 copy = len; 625 n = csum_and_copy_to_iter(vaddr + frag->page_offset + 626 offset - start, copy, 627 &csum2, to); 628 kunmap(page); 629 if (n != copy) 630 goto fault; 631 *csump = csum_block_add(*csump, csum2, pos); 632 if (!(len -= copy)) 633 return 0; 634 offset += copy; 635 pos += copy; 636 } 637 start = end; 638 } 639 640 skb_walk_frags(skb, frag_iter) { 641 int end; 642 643 WARN_ON(start > offset + len); 644 645 end = start + frag_iter->len; 646 if ((copy = end - offset) > 0) { 647 __wsum csum2 = 0; 648 if (copy > len) 649 copy = len; 650 if (skb_copy_and_csum_datagram(frag_iter, 651 offset - start, 652 to, copy, 653 &csum2)) 654 goto fault; 655 *csump = csum_block_add(*csump, csum2, pos); 656 if ((len -= copy) == 0) 657 return 0; 658 offset += copy; 659 pos += copy; 660 } 661 start = end; 662 } 663 if (!len) 664 return 0; 665 666 fault: 667 return -EFAULT; 668 } 669 670 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len) 671 { 672 __sum16 sum; 673 674 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum)); 675 if (likely(!sum)) { 676 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 677 !skb->csum_complete_sw) 678 netdev_rx_csum_fault(skb->dev); 679 } 680 if (!skb_shared(skb)) 681 skb->csum_valid = !sum; 682 return sum; 683 } 684 EXPORT_SYMBOL(__skb_checksum_complete_head); 685 686 __sum16 __skb_checksum_complete(struct sk_buff *skb) 687 { 688 __wsum csum; 689 __sum16 sum; 690 691 csum = skb_checksum(skb, 0, skb->len, 0); 692 693 /* skb->csum holds pseudo checksum */ 694 sum = csum_fold(csum_add(skb->csum, csum)); 695 if (likely(!sum)) { 696 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 697 !skb->csum_complete_sw) 698 netdev_rx_csum_fault(skb->dev); 699 } 700 701 if (!skb_shared(skb)) { 702 /* Save full packet checksum */ 703 skb->csum = csum; 704 skb->ip_summed = CHECKSUM_COMPLETE; 705 skb->csum_complete_sw = 1; 706 skb->csum_valid = !sum; 707 } 708 709 return sum; 710 } 711 EXPORT_SYMBOL(__skb_checksum_complete); 712 713 /** 714 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. 715 * @skb: skbuff 716 * @hlen: hardware length 717 * @msg: destination 718 * 719 * Caller _must_ check that skb will fit to this iovec. 720 * 721 * Returns: 0 - success. 722 * -EINVAL - checksum failure. 723 * -EFAULT - fault during copy. 724 */ 725 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, 726 int hlen, struct msghdr *msg) 727 { 728 __wsum csum; 729 int chunk = skb->len - hlen; 730 731 if (!chunk) 732 return 0; 733 734 if (msg_data_left(msg) < chunk) { 735 if (__skb_checksum_complete(skb)) 736 goto csum_error; 737 if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) 738 goto fault; 739 } else { 740 csum = csum_partial(skb->data, hlen, skb->csum); 741 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, 742 chunk, &csum)) 743 goto fault; 744 if (csum_fold(csum)) 745 goto csum_error; 746 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE)) 747 netdev_rx_csum_fault(skb->dev); 748 } 749 return 0; 750 csum_error: 751 return -EINVAL; 752 fault: 753 return -EFAULT; 754 } 755 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); 756 757 /** 758 * datagram_poll - generic datagram poll 759 * @file: file struct 760 * @sock: socket 761 * @wait: poll table 762 * 763 * Datagram poll: Again totally generic. This also handles 764 * sequenced packet sockets providing the socket receive queue 765 * is only ever holding data ready to receive. 766 * 767 * Note: when you _don't_ use this routine for this protocol, 768 * and you use a different write policy from sock_writeable() 769 * then please supply your own write_space callback. 770 */ 771 unsigned int datagram_poll(struct file *file, struct socket *sock, 772 poll_table *wait) 773 { 774 struct sock *sk = sock->sk; 775 unsigned int mask; 776 777 sock_poll_wait(file, sk_sleep(sk), wait); 778 mask = 0; 779 780 /* exceptional events? */ 781 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue)) 782 mask |= POLLERR | 783 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0); 784 785 if (sk->sk_shutdown & RCV_SHUTDOWN) 786 mask |= POLLRDHUP | POLLIN | POLLRDNORM; 787 if (sk->sk_shutdown == SHUTDOWN_MASK) 788 mask |= POLLHUP; 789 790 /* readable? */ 791 if (!skb_queue_empty(&sk->sk_receive_queue)) 792 mask |= POLLIN | POLLRDNORM; 793 794 /* Connection-based need to check for termination and startup */ 795 if (connection_based(sk)) { 796 if (sk->sk_state == TCP_CLOSE) 797 mask |= POLLHUP; 798 /* connection hasn't started yet? */ 799 if (sk->sk_state == TCP_SYN_SENT) 800 return mask; 801 } 802 803 /* writable? */ 804 if (sock_writeable(sk)) 805 mask |= POLLOUT | POLLWRNORM | POLLWRBAND; 806 else 807 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 808 809 return mask; 810 } 811 EXPORT_SYMBOL(datagram_poll); 812