1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * SUCS NET3: 4 * 5 * Generic datagram handling routines. These are generic for all 6 * protocols. Possibly a generic IP version on top of these would 7 * make sense. Not tonight however 8-). 8 * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and 9 * NetROM layer all have identical poll code and mostly 10 * identical recvmsg() code. So we share it here. The poll was 11 * shared before but buried in udp.c so I moved it. 12 * 13 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old 14 * udp.c code) 15 * 16 * Fixes: 17 * Alan Cox : NULL return from skb_peek_copy() 18 * understood 19 * Alan Cox : Rewrote skb_read_datagram to avoid the 20 * skb_peek_copy stuff. 21 * Alan Cox : Added support for SOCK_SEQPACKET. 22 * IPX can no longer use the SO_TYPE hack 23 * but AX.25 now works right, and SPX is 24 * feasible. 25 * Alan Cox : Fixed write poll of non IP protocol 26 * crash. 27 * Florian La Roche: Changed for my new skbuff handling. 28 * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. 29 * Linus Torvalds : BSD semantic fixes. 30 * Alan Cox : Datagram iovec handling 31 * Darryl Miles : Fixed non-blocking SOCK_STREAM. 32 * Alan Cox : POSIXisms 33 * Pete Wyckoff : Unconnected accept() fix. 34 * 35 */ 36 37 #include <linux/module.h> 38 #include <linux/types.h> 39 #include <linux/kernel.h> 40 #include <linux/uaccess.h> 41 #include <linux/mm.h> 42 #include <linux/interrupt.h> 43 #include <linux/errno.h> 44 #include <linux/sched.h> 45 #include <linux/inet.h> 46 #include <linux/netdevice.h> 47 #include <linux/rtnetlink.h> 48 #include <linux/poll.h> 49 #include <linux/highmem.h> 50 #include <linux/spinlock.h> 51 #include <linux/slab.h> 52 #include <linux/pagemap.h> 53 #include <linux/uio.h> 54 #include <linux/indirect_call_wrapper.h> 55 56 #include <net/protocol.h> 57 #include <linux/skbuff.h> 58 59 #include <net/checksum.h> 60 #include <net/sock.h> 61 #include <net/tcp_states.h> 62 #include <trace/events/skb.h> 63 #include <net/busy_poll.h> 64 65 /* 66 * Is a socket 'connection oriented' ? 67 */ 68 static inline int connection_based(struct sock *sk) 69 { 70 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; 71 } 72 73 static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync, 74 void *key) 75 { 76 /* 77 * Avoid a wakeup if event not interesting for us 78 */ 79 if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR))) 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, struct sk_buff_head *queue, 87 int *err, long *timeo_p, 88 const struct sk_buff *skb) 89 { 90 int error; 91 DEFINE_WAIT_FUNC(wait, receiver_wake_function); 92 93 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); 94 95 /* Socket errors? */ 96 error = sock_error(sk); 97 if (error) 98 goto out_err; 99 100 if (READ_ONCE(queue->prev) != skb) 101 goto out; 102 103 /* Socket shut down? */ 104 if (sk->sk_shutdown & RCV_SHUTDOWN) 105 goto out_noerr; 106 107 /* Sequenced packets can come disconnected. 108 * If so we report the problem 109 */ 110 error = -ENOTCONN; 111 if (connection_based(sk) && 112 !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) 113 goto out_err; 114 115 /* handle signals */ 116 if (signal_pending(current)) 117 goto interrupted; 118 119 error = 0; 120 *timeo_p = schedule_timeout(*timeo_p); 121 out: 122 finish_wait(sk_sleep(sk), &wait); 123 return error; 124 interrupted: 125 error = sock_intr_errno(*timeo_p); 126 out_err: 127 *err = error; 128 goto out; 129 out_noerr: 130 *err = 0; 131 error = 1; 132 goto out; 133 } 134 EXPORT_SYMBOL(__skb_wait_for_more_packets); 135 136 static struct sk_buff *skb_set_peeked(struct sk_buff *skb) 137 { 138 struct sk_buff *nskb; 139 140 if (skb->peeked) 141 return skb; 142 143 /* We have to unshare an skb before modifying it. */ 144 if (!skb_shared(skb)) 145 goto done; 146 147 nskb = skb_clone(skb, GFP_ATOMIC); 148 if (!nskb) 149 return ERR_PTR(-ENOMEM); 150 151 skb->prev->next = nskb; 152 skb->next->prev = nskb; 153 nskb->prev = skb->prev; 154 nskb->next = skb->next; 155 156 consume_skb(skb); 157 skb = nskb; 158 159 done: 160 skb->peeked = 1; 161 162 return skb; 163 } 164 165 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 166 struct sk_buff_head *queue, 167 unsigned int flags, 168 int *off, int *err, 169 struct sk_buff **last) 170 { 171 bool peek_at_off = false; 172 struct sk_buff *skb; 173 int _off = 0; 174 175 if (unlikely(flags & MSG_PEEK && *off >= 0)) { 176 peek_at_off = true; 177 _off = *off; 178 } 179 180 *last = queue->prev; 181 skb_queue_walk(queue, skb) { 182 if (flags & MSG_PEEK) { 183 if (peek_at_off && _off >= skb->len && 184 (_off || skb->peeked)) { 185 _off -= skb->len; 186 continue; 187 } 188 if (!skb->len) { 189 skb = skb_set_peeked(skb); 190 if (IS_ERR(skb)) { 191 *err = PTR_ERR(skb); 192 return NULL; 193 } 194 } 195 refcount_inc(&skb->users); 196 } else { 197 __skb_unlink(skb, queue); 198 } 199 *off = _off; 200 return skb; 201 } 202 return NULL; 203 } 204 205 /** 206 * __skb_try_recv_datagram - Receive a datagram skbuff 207 * @sk: socket 208 * @queue: socket queue from which to receive 209 * @flags: MSG\_ flags 210 * @off: an offset in bytes to peek skb from. Returns an offset 211 * within an skb where data actually starts 212 * @err: error code returned 213 * @last: set to last peeked message to inform the wait function 214 * what to look for when peeking 215 * 216 * Get a datagram skbuff, understands the peeking, nonblocking wakeups 217 * and possible races. This replaces identical code in packet, raw and 218 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes 219 * the long standing peek and read race for datagram sockets. If you 220 * alter this routine remember it must be re-entrant. 221 * 222 * This function will lock the socket if a skb is returned, so 223 * the caller needs to unlock the socket in that case (usually by 224 * calling skb_free_datagram). Returns NULL with @err set to 225 * -EAGAIN if no data was available or to some other value if an 226 * error was detected. 227 * 228 * * It does not lock socket since today. This function is 229 * * free of race conditions. This measure should/can improve 230 * * significantly datagram socket latencies at high loads, 231 * * when data copying to user space takes lots of time. 232 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet 233 * * 8) Great win.) 234 * * --ANK (980729) 235 * 236 * The order of the tests when we find no data waiting are specified 237 * quite explicitly by POSIX 1003.1g, don't change them without having 238 * the standard around please. 239 */ 240 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 241 struct sk_buff_head *queue, 242 unsigned int flags, int *off, int *err, 243 struct sk_buff **last) 244 { 245 struct sk_buff *skb; 246 unsigned long cpu_flags; 247 /* 248 * Caller is allowed not to check sk->sk_err before skb_recv_datagram() 249 */ 250 int error = sock_error(sk); 251 252 if (error) 253 goto no_packet; 254 255 do { 256 /* Again only user level code calls this function, so nothing 257 * interrupt level will suddenly eat the receive_queue. 258 * 259 * Look at current nfs client by the way... 260 * However, this function was correct in any case. 8) 261 */ 262 spin_lock_irqsave(&queue->lock, cpu_flags); 263 skb = __skb_try_recv_from_queue(sk, queue, flags, off, &error, 264 last); 265 spin_unlock_irqrestore(&queue->lock, cpu_flags); 266 if (error) 267 goto no_packet; 268 if (skb) 269 return skb; 270 271 if (!sk_can_busy_loop(sk)) 272 break; 273 274 sk_busy_loop(sk, flags & MSG_DONTWAIT); 275 } while (READ_ONCE(queue->prev) != *last); 276 277 error = -EAGAIN; 278 279 no_packet: 280 *err = error; 281 return NULL; 282 } 283 EXPORT_SYMBOL(__skb_try_recv_datagram); 284 285 struct sk_buff *__skb_recv_datagram(struct sock *sk, 286 struct sk_buff_head *sk_queue, 287 unsigned int flags, int *off, int *err) 288 { 289 struct sk_buff *skb, *last; 290 long timeo; 291 292 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); 293 294 do { 295 skb = __skb_try_recv_datagram(sk, sk_queue, flags, off, err, 296 &last); 297 if (skb) 298 return skb; 299 300 if (*err != -EAGAIN) 301 break; 302 } while (timeo && 303 !__skb_wait_for_more_packets(sk, sk_queue, err, 304 &timeo, last)); 305 306 return NULL; 307 } 308 EXPORT_SYMBOL(__skb_recv_datagram); 309 310 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, 311 int *err) 312 { 313 int off = 0; 314 315 return __skb_recv_datagram(sk, &sk->sk_receive_queue, flags, 316 &off, err); 317 } 318 EXPORT_SYMBOL(skb_recv_datagram); 319 320 void skb_free_datagram(struct sock *sk, struct sk_buff *skb) 321 { 322 consume_skb(skb); 323 sk_mem_reclaim_partial(sk); 324 } 325 EXPORT_SYMBOL(skb_free_datagram); 326 327 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len) 328 { 329 bool slow; 330 331 if (!skb_unref(skb)) { 332 sk_peek_offset_bwd(sk, len); 333 return; 334 } 335 336 slow = lock_sock_fast(sk); 337 sk_peek_offset_bwd(sk, len); 338 skb_orphan(skb); 339 sk_mem_reclaim_partial(sk); 340 unlock_sock_fast(sk, slow); 341 342 /* skb is now orphaned, can be freed outside of locked section */ 343 __kfree_skb(skb); 344 } 345 EXPORT_SYMBOL(__skb_free_datagram_locked); 346 347 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, 348 struct sk_buff *skb, unsigned int flags, 349 void (*destructor)(struct sock *sk, 350 struct sk_buff *skb)) 351 { 352 int err = 0; 353 354 if (flags & MSG_PEEK) { 355 err = -ENOENT; 356 spin_lock_bh(&sk_queue->lock); 357 if (skb->next) { 358 __skb_unlink(skb, sk_queue); 359 refcount_dec(&skb->users); 360 if (destructor) 361 destructor(sk, skb); 362 err = 0; 363 } 364 spin_unlock_bh(&sk_queue->lock); 365 } 366 367 atomic_inc(&sk->sk_drops); 368 return err; 369 } 370 EXPORT_SYMBOL(__sk_queue_drop_skb); 371 372 /** 373 * skb_kill_datagram - Free a datagram skbuff forcibly 374 * @sk: socket 375 * @skb: datagram skbuff 376 * @flags: MSG\_ flags 377 * 378 * This function frees a datagram skbuff that was received by 379 * skb_recv_datagram. The flags argument must match the one 380 * used for skb_recv_datagram. 381 * 382 * If the MSG_PEEK flag is set, and the packet is still on the 383 * receive queue of the socket, it will be taken off the queue 384 * before it is freed. 385 * 386 * This function currently only disables BH when acquiring the 387 * sk_receive_queue lock. Therefore it must not be used in a 388 * context where that lock is acquired in an IRQ context. 389 * 390 * It returns 0 if the packet was removed by us. 391 */ 392 393 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) 394 { 395 int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags, 396 NULL); 397 398 kfree_skb(skb); 399 sk_mem_reclaim_partial(sk); 400 return err; 401 } 402 EXPORT_SYMBOL(skb_kill_datagram); 403 404 INDIRECT_CALLABLE_DECLARE(static size_t simple_copy_to_iter(const void *addr, 405 size_t bytes, 406 void *data __always_unused, 407 struct iov_iter *i)); 408 409 static int __skb_datagram_iter(const struct sk_buff *skb, int offset, 410 struct iov_iter *to, int len, bool fault_short, 411 size_t (*cb)(const void *, size_t, void *, 412 struct iov_iter *), void *data) 413 { 414 int start = skb_headlen(skb); 415 int i, copy = start - offset, start_off = offset, n; 416 struct sk_buff *frag_iter; 417 418 /* Copy header. */ 419 if (copy > 0) { 420 if (copy > len) 421 copy = len; 422 n = INDIRECT_CALL_1(cb, simple_copy_to_iter, 423 skb->data + offset, copy, data, to); 424 offset += n; 425 if (n != copy) 426 goto short_copy; 427 if ((len -= copy) == 0) 428 return 0; 429 } 430 431 /* Copy paged appendix. Hmm... why does this look so complicated? */ 432 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 433 int end; 434 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 435 436 WARN_ON(start > offset + len); 437 438 end = start + skb_frag_size(frag); 439 if ((copy = end - offset) > 0) { 440 struct page *page = skb_frag_page(frag); 441 u8 *vaddr = kmap(page); 442 443 if (copy > len) 444 copy = len; 445 n = INDIRECT_CALL_1(cb, simple_copy_to_iter, 446 vaddr + skb_frag_off(frag) + offset - start, 447 copy, data, to); 448 kunmap(page); 449 offset += n; 450 if (n != copy) 451 goto short_copy; 452 if (!(len -= copy)) 453 return 0; 454 } 455 start = end; 456 } 457 458 skb_walk_frags(skb, frag_iter) { 459 int end; 460 461 WARN_ON(start > offset + len); 462 463 end = start + frag_iter->len; 464 if ((copy = end - offset) > 0) { 465 if (copy > len) 466 copy = len; 467 if (__skb_datagram_iter(frag_iter, offset - start, 468 to, copy, fault_short, cb, data)) 469 goto fault; 470 if ((len -= copy) == 0) 471 return 0; 472 offset += copy; 473 } 474 start = end; 475 } 476 if (!len) 477 return 0; 478 479 /* This is not really a user copy fault, but rather someone 480 * gave us a bogus length on the skb. We should probably 481 * print a warning here as it may indicate a kernel bug. 482 */ 483 484 fault: 485 iov_iter_revert(to, offset - start_off); 486 return -EFAULT; 487 488 short_copy: 489 if (fault_short || iov_iter_count(to)) 490 goto fault; 491 492 return 0; 493 } 494 495 /** 496 * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator 497 * and update a hash. 498 * @skb: buffer to copy 499 * @offset: offset in the buffer to start copying from 500 * @to: iovec iterator to copy to 501 * @len: amount of data to copy from buffer to iovec 502 * @hash: hash request to update 503 */ 504 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 505 struct iov_iter *to, int len, 506 struct ahash_request *hash) 507 { 508 return __skb_datagram_iter(skb, offset, to, len, true, 509 hash_and_copy_to_iter, hash); 510 } 511 EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter); 512 513 static size_t simple_copy_to_iter(const void *addr, size_t bytes, 514 void *data __always_unused, struct iov_iter *i) 515 { 516 return copy_to_iter(addr, bytes, i); 517 } 518 519 /** 520 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. 521 * @skb: buffer to copy 522 * @offset: offset in the buffer to start copying from 523 * @to: iovec iterator to copy to 524 * @len: amount of data to copy from buffer to iovec 525 */ 526 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, 527 struct iov_iter *to, int len) 528 { 529 trace_skb_copy_datagram_iovec(skb, len); 530 return __skb_datagram_iter(skb, offset, to, len, false, 531 simple_copy_to_iter, NULL); 532 } 533 EXPORT_SYMBOL(skb_copy_datagram_iter); 534 535 /** 536 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. 537 * @skb: buffer to copy 538 * @offset: offset in the buffer to start copying to 539 * @from: the copy source 540 * @len: amount of data to copy to buffer from iovec 541 * 542 * Returns 0 or -EFAULT. 543 */ 544 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 545 struct iov_iter *from, 546 int len) 547 { 548 int start = skb_headlen(skb); 549 int i, copy = start - offset; 550 struct sk_buff *frag_iter; 551 552 /* Copy header. */ 553 if (copy > 0) { 554 if (copy > len) 555 copy = len; 556 if (copy_from_iter(skb->data + offset, copy, from) != copy) 557 goto fault; 558 if ((len -= copy) == 0) 559 return 0; 560 offset += copy; 561 } 562 563 /* Copy paged appendix. Hmm... why does this look so complicated? */ 564 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 565 int end; 566 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 567 568 WARN_ON(start > offset + len); 569 570 end = start + skb_frag_size(frag); 571 if ((copy = end - offset) > 0) { 572 size_t copied; 573 574 if (copy > len) 575 copy = len; 576 copied = copy_page_from_iter(skb_frag_page(frag), 577 skb_frag_off(frag) + offset - start, 578 copy, from); 579 if (copied != copy) 580 goto fault; 581 582 if (!(len -= copy)) 583 return 0; 584 offset += copy; 585 } 586 start = end; 587 } 588 589 skb_walk_frags(skb, frag_iter) { 590 int end; 591 592 WARN_ON(start > offset + len); 593 594 end = start + frag_iter->len; 595 if ((copy = end - offset) > 0) { 596 if (copy > len) 597 copy = len; 598 if (skb_copy_datagram_from_iter(frag_iter, 599 offset - start, 600 from, copy)) 601 goto fault; 602 if ((len -= copy) == 0) 603 return 0; 604 offset += copy; 605 } 606 start = end; 607 } 608 if (!len) 609 return 0; 610 611 fault: 612 return -EFAULT; 613 } 614 EXPORT_SYMBOL(skb_copy_datagram_from_iter); 615 616 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, 617 struct iov_iter *from, size_t length) 618 { 619 int frag = skb_shinfo(skb)->nr_frags; 620 621 while (length && iov_iter_count(from)) { 622 struct page *pages[MAX_SKB_FRAGS]; 623 struct page *last_head = NULL; 624 size_t start; 625 ssize_t copied; 626 unsigned long truesize; 627 int refs, n = 0; 628 629 if (frag == MAX_SKB_FRAGS) 630 return -EMSGSIZE; 631 632 copied = iov_iter_get_pages(from, pages, length, 633 MAX_SKB_FRAGS - frag, &start); 634 if (copied < 0) 635 return -EFAULT; 636 637 iov_iter_advance(from, copied); 638 length -= copied; 639 640 truesize = PAGE_ALIGN(copied + start); 641 skb->data_len += copied; 642 skb->len += copied; 643 skb->truesize += truesize; 644 if (sk && sk->sk_type == SOCK_STREAM) { 645 sk_wmem_queued_add(sk, truesize); 646 if (!skb_zcopy_pure(skb)) 647 sk_mem_charge(sk, truesize); 648 } else { 649 refcount_add(truesize, &skb->sk->sk_wmem_alloc); 650 } 651 for (refs = 0; copied != 0; start = 0) { 652 int size = min_t(int, copied, PAGE_SIZE - start); 653 struct page *head = compound_head(pages[n]); 654 655 start += (pages[n] - head) << PAGE_SHIFT; 656 copied -= size; 657 n++; 658 if (frag) { 659 skb_frag_t *last = &skb_shinfo(skb)->frags[frag - 1]; 660 661 if (head == skb_frag_page(last) && 662 start == skb_frag_off(last) + skb_frag_size(last)) { 663 skb_frag_size_add(last, size); 664 /* We combined this page, we need to release 665 * a reference. Since compound pages refcount 666 * is shared among many pages, batch the refcount 667 * adjustments to limit false sharing. 668 */ 669 last_head = head; 670 refs++; 671 continue; 672 } 673 } 674 if (refs) { 675 page_ref_sub(last_head, refs); 676 refs = 0; 677 } 678 skb_fill_page_desc(skb, frag++, head, start, size); 679 } 680 if (refs) 681 page_ref_sub(last_head, refs); 682 } 683 return 0; 684 } 685 EXPORT_SYMBOL(__zerocopy_sg_from_iter); 686 687 /** 688 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter 689 * @skb: buffer to copy 690 * @from: the source to copy from 691 * 692 * The function will first copy up to headlen, and then pin the userspace 693 * pages and build frags through them. 694 * 695 * Returns 0, -EFAULT or -EMSGSIZE. 696 */ 697 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) 698 { 699 int copy = min_t(int, skb_headlen(skb), iov_iter_count(from)); 700 701 /* copy up to skb headlen */ 702 if (skb_copy_datagram_from_iter(skb, 0, from, copy)) 703 return -EFAULT; 704 705 return __zerocopy_sg_from_iter(NULL, skb, from, ~0U); 706 } 707 EXPORT_SYMBOL(zerocopy_sg_from_iter); 708 709 /** 710 * skb_copy_and_csum_datagram - Copy datagram to an iovec iterator 711 * and update a checksum. 712 * @skb: buffer to copy 713 * @offset: offset in the buffer to start copying from 714 * @to: iovec iterator to copy to 715 * @len: amount of data to copy from buffer to iovec 716 * @csump: checksum pointer 717 */ 718 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, 719 struct iov_iter *to, int len, 720 __wsum *csump) 721 { 722 struct csum_state csdata = { .csum = *csump }; 723 int ret; 724 725 ret = __skb_datagram_iter(skb, offset, to, len, true, 726 csum_and_copy_to_iter, &csdata); 727 if (ret) 728 return ret; 729 730 *csump = csdata.csum; 731 return 0; 732 } 733 734 /** 735 * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. 736 * @skb: skbuff 737 * @hlen: hardware length 738 * @msg: destination 739 * 740 * Caller _must_ check that skb will fit to this iovec. 741 * 742 * Returns: 0 - success. 743 * -EINVAL - checksum failure. 744 * -EFAULT - fault during copy. 745 */ 746 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, 747 int hlen, struct msghdr *msg) 748 { 749 __wsum csum; 750 int chunk = skb->len - hlen; 751 752 if (!chunk) 753 return 0; 754 755 if (msg_data_left(msg) < chunk) { 756 if (__skb_checksum_complete(skb)) 757 return -EINVAL; 758 if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) 759 goto fault; 760 } else { 761 csum = csum_partial(skb->data, hlen, skb->csum); 762 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, 763 chunk, &csum)) 764 goto fault; 765 766 if (csum_fold(csum)) { 767 iov_iter_revert(&msg->msg_iter, chunk); 768 return -EINVAL; 769 } 770 771 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && 772 !skb->csum_complete_sw) 773 netdev_rx_csum_fault(NULL, skb); 774 } 775 return 0; 776 fault: 777 return -EFAULT; 778 } 779 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); 780 781 /** 782 * datagram_poll - generic datagram poll 783 * @file: file struct 784 * @sock: socket 785 * @wait: poll table 786 * 787 * Datagram poll: Again totally generic. This also handles 788 * sequenced packet sockets providing the socket receive queue 789 * is only ever holding data ready to receive. 790 * 791 * Note: when you *don't* use this routine for this protocol, 792 * and you use a different write policy from sock_writeable() 793 * then please supply your own write_space callback. 794 */ 795 __poll_t datagram_poll(struct file *file, struct socket *sock, 796 poll_table *wait) 797 { 798 struct sock *sk = sock->sk; 799 __poll_t mask; 800 801 sock_poll_wait(file, sock, wait); 802 mask = 0; 803 804 /* exceptional events? */ 805 if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue)) 806 mask |= EPOLLERR | 807 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0); 808 809 if (sk->sk_shutdown & RCV_SHUTDOWN) 810 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM; 811 if (sk->sk_shutdown == SHUTDOWN_MASK) 812 mask |= EPOLLHUP; 813 814 /* readable? */ 815 if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) 816 mask |= EPOLLIN | EPOLLRDNORM; 817 818 /* Connection-based need to check for termination and startup */ 819 if (connection_based(sk)) { 820 if (sk->sk_state == TCP_CLOSE) 821 mask |= EPOLLHUP; 822 /* connection hasn't started yet? */ 823 if (sk->sk_state == TCP_SYN_SENT) 824 return mask; 825 } 826 827 /* writable? */ 828 if (sock_writeable(sk)) 829 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; 830 else 831 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 832 833 return mask; 834 } 835 EXPORT_SYMBOL(datagram_poll); 836