root/net/core/datagram.c

DEFINITIONS

1. connection_based
2. receiver_wake_function
3. __skb_wait_for_more_packets
4. skb_set_peeked
5. __skb_try_recv_from_queue
6. __skb_try_recv_datagram
7. __skb_recv_datagram
8. skb_recv_datagram
9. skb_free_datagram
10. __skb_free_datagram_locked
11. __sk_queue_drop_skb
12. skb_kill_datagram
13. __skb_datagram_iter
14. skb_copy_and_hash_datagram_iter
15. simple_copy_to_iter
16. skb_copy_datagram_iter
17. skb_copy_datagram_from_iter
18. __zerocopy_sg_from_iter
19. zerocopy_sg_from_iter
20. skb_copy_and_csum_datagram
21. skb_copy_and_csum_datagram_msg
22. datagram_poll

   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 #include "datagram.h"
  66 
  67 /*
  68  *      Is a socket 'connection oriented' ?
  69  */
  70 static inline int connection_based(struct sock *sk)
  71 {
  72         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
  73 }
  74 
  75 static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync,
  76                                   void *key)
  77 {
  78         /*
  79          * Avoid a wakeup if event not interesting for us
  80          */
  81         if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR)))
  82                 return 0;
  83         return autoremove_wake_function(wait, mode, sync, key);
  84 }
  85 /*
  86  * Wait for the last received packet to be different from skb
  87  */
  88 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
  89                                 const struct sk_buff *skb)
  90 {
  91         int error;
  92         DEFINE_WAIT_FUNC(wait, receiver_wake_function);
  93 
  94         prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
  95 
  96         /* Socket errors? */
  97         error = sock_error(sk);
  98         if (error)
  99                 goto out_err;
 100 
 101         if (READ_ONCE(sk->sk_receive_queue.prev) != skb)
 102                 goto out;
 103 
 104         /* Socket shut down? */
 105         if (sk->sk_shutdown & RCV_SHUTDOWN)
 106                 goto out_noerr;
 107 
 108         /* Sequenced packets can come disconnected.
 109          * If so we report the problem
 110          */
 111         error = -ENOTCONN;
 112         if (connection_based(sk) &&
 113             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
 114                 goto out_err;
 115 
 116         /* handle signals */
 117         if (signal_pending(current))
 118                 goto interrupted;
 119 
 120         error = 0;
 121         *timeo_p = schedule_timeout(*timeo_p);
 122 out:
 123         finish_wait(sk_sleep(sk), &wait);
 124         return error;
 125 interrupted:
 126         error = sock_intr_errno(*timeo_p);
 127 out_err:
 128         *err = error;
 129         goto out;
 130 out_noerr:
 131         *err = 0;
 132         error = 1;
 133         goto out;
 134 }
 135 EXPORT_SYMBOL(__skb_wait_for_more_packets);
 136 
 137 static struct sk_buff *skb_set_peeked(struct sk_buff *skb)
 138 {
 139         struct sk_buff *nskb;
 140 
 141         if (skb->peeked)
 142                 return skb;
 143 
 144         /* We have to unshare an skb before modifying it. */
 145         if (!skb_shared(skb))
 146                 goto done;
 147 
 148         nskb = skb_clone(skb, GFP_ATOMIC);
 149         if (!nskb)
 150                 return ERR_PTR(-ENOMEM);
 151 
 152         skb->prev->next = nskb;
 153         skb->next->prev = nskb;
 154         nskb->prev = skb->prev;
 155         nskb->next = skb->next;
 156 
 157         consume_skb(skb);
 158         skb = nskb;
 159 
 160 done:
 161         skb->peeked = 1;
 162 
 163         return skb;
 164 }
 165 
 166 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
 167                                           struct sk_buff_head *queue,
 168                                           unsigned int flags,
 169                                           void (*destructor)(struct sock *sk,
 170                                                            struct sk_buff *skb),
 171                                           int *off, int *err,
 172                                           struct sk_buff **last)
 173 {
 174         bool peek_at_off = false;
 175         struct sk_buff *skb;
 176         int _off = 0;
 177 
 178         if (unlikely(flags & MSG_PEEK && *off >= 0)) {
 179                 peek_at_off = true;
 180                 _off = *off;
 181         }
 182 
 183         *last = queue->prev;
 184         skb_queue_walk(queue, skb) {
 185                 if (flags & MSG_PEEK) {
 186                         if (peek_at_off && _off >= skb->len &&
 187                             (_off || skb->peeked)) {
 188                                 _off -= skb->len;
 189                                 continue;
 190                         }
 191                         if (!skb->len) {
 192                                 skb = skb_set_peeked(skb);
 193                                 if (IS_ERR(skb)) {
 194                                         *err = PTR_ERR(skb);
 195                                         return NULL;
 196                                 }
 197                         }
 198                         refcount_inc(&skb->users);
 199                 } else {
 200                         __skb_unlink(skb, queue);
 201                         if (destructor)
 202                                 destructor(sk, skb);
 203                 }
 204                 *off = _off;
 205                 return skb;
 206         }
 207         return NULL;
 208 }
 209 
 210 /**
 211  *      __skb_try_recv_datagram - Receive a datagram skbuff
 212  *      @sk: socket
 213  *      @flags: MSG\_ flags
 214  *      @destructor: invoked under the receive lock on successful dequeue
 215  *      @off: an offset in bytes to peek skb from. Returns an offset
 216  *            within an skb where data actually starts
 217  *      @err: error code returned
 218  *      @last: set to last peeked message to inform the wait function
 219  *             what to look for when peeking
 220  *
 221  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
 222  *      and possible races. This replaces identical code in packet, raw and
 223  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
 224  *      the long standing peek and read race for datagram sockets. If you
 225  *      alter this routine remember it must be re-entrant.
 226  *
 227  *      This function will lock the socket if a skb is returned, so
 228  *      the caller needs to unlock the socket in that case (usually by
 229  *      calling skb_free_datagram). Returns NULL with @err set to
 230  *      -EAGAIN if no data was available or to some other value if an
 231  *      error was detected.
 232  *
 233  *      * It does not lock socket since today. This function is
 234  *      * free of race conditions. This measure should/can improve
 235  *      * significantly datagram socket latencies at high loads,
 236  *      * when data copying to user space takes lots of time.
 237  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
 238  *      *  8) Great win.)
 239  *      *                                           --ANK (980729)
 240  *
 241  *      The order of the tests when we find no data waiting are specified
 242  *      quite explicitly by POSIX 1003.1g, don't change them without having
 243  *      the standard around please.
 244  */
 245 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags,
 246                                         void (*destructor)(struct sock *sk,
 247                                                            struct sk_buff *skb),
 248                                         int *off, int *err,
 249                                         struct sk_buff **last)
 250 {
 251         struct sk_buff_head *queue = &sk->sk_receive_queue;
 252         struct sk_buff *skb;
 253         unsigned long cpu_flags;
 254         /*
 255          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
 256          */
 257         int error = sock_error(sk);
 258 
 259         if (error)
 260                 goto no_packet;
 261 
 262         do {
 263                 /* Again only user level code calls this function, so nothing
 264                  * interrupt level will suddenly eat the receive_queue.
 265                  *
 266                  * Look at current nfs client by the way...
 267                  * However, this function was correct in any case. 8)
 268                  */
 269                 spin_lock_irqsave(&queue->lock, cpu_flags);
 270                 skb = __skb_try_recv_from_queue(sk, queue, flags, destructor,
 271                                                 off, &error, last);
 272                 spin_unlock_irqrestore(&queue->lock, cpu_flags);
 273                 if (error)
 274                         goto no_packet;
 275                 if (skb)
 276                         return skb;
 277 
 278                 if (!sk_can_busy_loop(sk))
 279                         break;
 280 
 281                 sk_busy_loop(sk, flags & MSG_DONTWAIT);
 282         } while (READ_ONCE(sk->sk_receive_queue.prev) != *last);
 283 
 284         error = -EAGAIN;
 285 
 286 no_packet:
 287         *err = error;
 288         return NULL;
 289 }
 290 EXPORT_SYMBOL(__skb_try_recv_datagram);
 291 
 292 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
 293                                     void (*destructor)(struct sock *sk,
 294                                                        struct sk_buff *skb),
 295                                     int *off, int *err)
 296 {
 297         struct sk_buff *skb, *last;
 298         long timeo;
 299 
 300         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 301 
 302         do {
 303                 skb = __skb_try_recv_datagram(sk, flags, destructor, off, err,
 304                                               &last);
 305                 if (skb)
 306                         return skb;
 307 
 308                 if (*err != -EAGAIN)
 309                         break;
 310         } while (timeo &&
 311                 !__skb_wait_for_more_packets(sk, err, &timeo, last));
 312 
 313         return NULL;
 314 }
 315 EXPORT_SYMBOL(__skb_recv_datagram);
 316 
 317 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
 318                                   int noblock, int *err)
 319 {
 320         int off = 0;
 321 
 322         return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 323                                    NULL, &off, err);
 324 }
 325 EXPORT_SYMBOL(skb_recv_datagram);
 326 
 327 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
 328 {
 329         consume_skb(skb);
 330         sk_mem_reclaim_partial(sk);
 331 }
 332 EXPORT_SYMBOL(skb_free_datagram);
 333 
 334 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len)
 335 {
 336         bool slow;
 337 
 338         if (!skb_unref(skb)) {
 339                 sk_peek_offset_bwd(sk, len);
 340                 return;
 341         }
 342 
 343         slow = lock_sock_fast(sk);
 344         sk_peek_offset_bwd(sk, len);
 345         skb_orphan(skb);
 346         sk_mem_reclaim_partial(sk);
 347         unlock_sock_fast(sk, slow);
 348 
 349         /* skb is now orphaned, can be freed outside of locked section */
 350         __kfree_skb(skb);
 351 }
 352 EXPORT_SYMBOL(__skb_free_datagram_locked);
 353 
 354 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue,
 355                         struct sk_buff *skb, unsigned int flags,
 356                         void (*destructor)(struct sock *sk,
 357                                            struct sk_buff *skb))
 358 {
 359         int err = 0;
 360 
 361         if (flags & MSG_PEEK) {
 362                 err = -ENOENT;
 363                 spin_lock_bh(&sk_queue->lock);
 364                 if (skb->next) {
 365                         __skb_unlink(skb, sk_queue);
 366                         refcount_dec(&skb->users);
 367                         if (destructor)
 368                                 destructor(sk, skb);
 369                         err = 0;
 370                 }
 371                 spin_unlock_bh(&sk_queue->lock);
 372         }
 373 
 374         atomic_inc(&sk->sk_drops);
 375         return err;
 376 }
 377 EXPORT_SYMBOL(__sk_queue_drop_skb);
 378 
 379 /**
 380  *      skb_kill_datagram - Free a datagram skbuff forcibly
 381  *      @sk: socket
 382  *      @skb: datagram skbuff
 383  *      @flags: MSG\_ flags
 384  *
 385  *      This function frees a datagram skbuff that was received by
 386  *      skb_recv_datagram.  The flags argument must match the one
 387  *      used for skb_recv_datagram.
 388  *
 389  *      If the MSG_PEEK flag is set, and the packet is still on the
 390  *      receive queue of the socket, it will be taken off the queue
 391  *      before it is freed.
 392  *
 393  *      This function currently only disables BH when acquiring the
 394  *      sk_receive_queue lock.  Therefore it must not be used in a
 395  *      context where that lock is acquired in an IRQ context.
 396  *
 397  *      It returns 0 if the packet was removed by us.
 398  */
 399 
 400 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
 401 {
 402         int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags,
 403                                       NULL);
 404 
 405         kfree_skb(skb);
 406         sk_mem_reclaim_partial(sk);
 407         return err;
 408 }
 409 EXPORT_SYMBOL(skb_kill_datagram);
 410 
 411 INDIRECT_CALLABLE_DECLARE(static size_t simple_copy_to_iter(const void *addr,
 412                                                 size_t bytes,
 413                                                 void *data __always_unused,
 414                                                 struct iov_iter *i));
 415 
 416 static int __skb_datagram_iter(const struct sk_buff *skb, int offset,
 417                                struct iov_iter *to, int len, bool fault_short,
 418                                size_t (*cb)(const void *, size_t, void *,
 419                                             struct iov_iter *), void *data)
 420 {
 421         int start = skb_headlen(skb);
 422         int i, copy = start - offset, start_off = offset, n;
 423         struct sk_buff *frag_iter;
 424 
 425         /* Copy header. */
 426         if (copy > 0) {
 427                 if (copy > len)
 428                         copy = len;
 429                 n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
 430                                     skb->data + offset, copy, data, to);
 431                 offset += n;
 432                 if (n != copy)
 433                         goto short_copy;
 434                 if ((len -= copy) == 0)
 435                         return 0;
 436         }
 437 
 438         /* Copy paged appendix. Hmm... why does this look so complicated? */
 439         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 440                 int end;
 441                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 442 
 443                 WARN_ON(start > offset + len);
 444 
 445                 end = start + skb_frag_size(frag);
 446                 if ((copy = end - offset) > 0) {
 447                         struct page *page = skb_frag_page(frag);
 448                         u8 *vaddr = kmap(page);
 449 
 450                         if (copy > len)
 451                                 copy = len;
 452                         n = INDIRECT_CALL_1(cb, simple_copy_to_iter,
 453                                         vaddr + skb_frag_off(frag) + offset - start,
 454                                         copy, data, to);
 455                         kunmap(page);
 456                         offset += n;
 457                         if (n != copy)
 458                                 goto short_copy;
 459                         if (!(len -= copy))
 460                                 return 0;
 461                 }
 462                 start = end;
 463         }
 464 
 465         skb_walk_frags(skb, frag_iter) {
 466                 int end;
 467 
 468                 WARN_ON(start > offset + len);
 469 
 470                 end = start + frag_iter->len;
 471                 if ((copy = end - offset) > 0) {
 472                         if (copy > len)
 473                                 copy = len;
 474                         if (__skb_datagram_iter(frag_iter, offset - start,
 475                                                 to, copy, fault_short, cb, data))
 476                                 goto fault;
 477                         if ((len -= copy) == 0)
 478                                 return 0;
 479                         offset += copy;
 480                 }
 481                 start = end;
 482         }
 483         if (!len)
 484                 return 0;
 485 
 486         /* This is not really a user copy fault, but rather someone
 487          * gave us a bogus length on the skb.  We should probably
 488          * print a warning here as it may indicate a kernel bug.
 489          */
 490 
 491 fault:
 492         iov_iter_revert(to, offset - start_off);
 493         return -EFAULT;
 494 
 495 short_copy:
 496         if (fault_short || iov_iter_count(to))
 497                 goto fault;
 498 
 499         return 0;
 500 }
 501 
 502 /**
 503  *      skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator
 504  *          and update a hash.
 505  *      @skb: buffer to copy
 506  *      @offset: offset in the buffer to start copying from
 507  *      @to: iovec iterator to copy to
 508  *      @len: amount of data to copy from buffer to iovec
 509  *      @hash: hash request to update
 510  */
 511 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
 512                            struct iov_iter *to, int len,
 513                            struct ahash_request *hash)
 514 {
 515         return __skb_datagram_iter(skb, offset, to, len, true,
 516                         hash_and_copy_to_iter, hash);
 517 }
 518 EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter);
 519 
 520 static size_t simple_copy_to_iter(const void *addr, size_t bytes,
 521                 void *data __always_unused, struct iov_iter *i)
 522 {
 523         return copy_to_iter(addr, bytes, i);
 524 }
 525 
 526 /**
 527  *      skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
 528  *      @skb: buffer to copy
 529  *      @offset: offset in the buffer to start copying from
 530  *      @to: iovec iterator to copy to
 531  *      @len: amount of data to copy from buffer to iovec
 532  */
 533 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
 534                            struct iov_iter *to, int len)
 535 {
 536         trace_skb_copy_datagram_iovec(skb, len);
 537         return __skb_datagram_iter(skb, offset, to, len, false,
 538                         simple_copy_to_iter, NULL);
 539 }
 540 EXPORT_SYMBOL(skb_copy_datagram_iter);
 541 
 542 /**
 543  *      skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
 544  *      @skb: buffer to copy
 545  *      @offset: offset in the buffer to start copying to
 546  *      @from: the copy source
 547  *      @len: amount of data to copy to buffer from iovec
 548  *
 549  *      Returns 0 or -EFAULT.
 550  */
 551 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
 552                                  struct iov_iter *from,
 553                                  int len)
 554 {
 555         int start = skb_headlen(skb);
 556         int i, copy = start - offset;
 557         struct sk_buff *frag_iter;
 558 
 559         /* Copy header. */
 560         if (copy > 0) {
 561                 if (copy > len)
 562                         copy = len;
 563                 if (copy_from_iter(skb->data + offset, copy, from) != copy)
 564                         goto fault;
 565                 if ((len -= copy) == 0)
 566                         return 0;
 567                 offset += copy;
 568         }
 569 
 570         /* Copy paged appendix. Hmm... why does this look so complicated? */
 571         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 572                 int end;
 573                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 574 
 575                 WARN_ON(start > offset + len);
 576 
 577                 end = start + skb_frag_size(frag);
 578                 if ((copy = end - offset) > 0) {
 579                         size_t copied;
 580 
 581                         if (copy > len)
 582                                 copy = len;
 583                         copied = copy_page_from_iter(skb_frag_page(frag),
 584                                           skb_frag_off(frag) + offset - start,
 585                                           copy, from);
 586                         if (copied != copy)
 587                                 goto fault;
 588 
 589                         if (!(len -= copy))
 590                                 return 0;
 591                         offset += copy;
 592                 }
 593                 start = end;
 594         }
 595 
 596         skb_walk_frags(skb, frag_iter) {
 597                 int end;
 598 
 599                 WARN_ON(start > offset + len);
 600 
 601                 end = start + frag_iter->len;
 602                 if ((copy = end - offset) > 0) {
 603                         if (copy > len)
 604                                 copy = len;
 605                         if (skb_copy_datagram_from_iter(frag_iter,
 606                                                         offset - start,
 607                                                         from, copy))
 608                                 goto fault;
 609                         if ((len -= copy) == 0)
 610                                 return 0;
 611                         offset += copy;
 612                 }
 613                 start = end;
 614         }
 615         if (!len)
 616                 return 0;
 617 
 618 fault:
 619         return -EFAULT;
 620 }
 621 EXPORT_SYMBOL(skb_copy_datagram_from_iter);
 622 
 623 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
 624                             struct iov_iter *from, size_t length)
 625 {
 626         int frag = skb_shinfo(skb)->nr_frags;
 627 
 628         while (length && iov_iter_count(from)) {
 629                 struct page *pages[MAX_SKB_FRAGS];
 630                 size_t start;
 631                 ssize_t copied;
 632                 unsigned long truesize;
 633                 int n = 0;
 634 
 635                 if (frag == MAX_SKB_FRAGS)
 636                         return -EMSGSIZE;
 637 
 638                 copied = iov_iter_get_pages(from, pages, length,
 639                                             MAX_SKB_FRAGS - frag, &start);
 640                 if (copied < 0)
 641                         return -EFAULT;
 642 
 643                 iov_iter_advance(from, copied);
 644                 length -= copied;
 645 
 646                 truesize = PAGE_ALIGN(copied + start);
 647                 skb->data_len += copied;
 648                 skb->len += copied;
 649                 skb->truesize += truesize;
 650                 if (sk && sk->sk_type == SOCK_STREAM) {
 651                         sk_wmem_queued_add(sk, truesize);
 652                         sk_mem_charge(sk, truesize);
 653                 } else {
 654                         refcount_add(truesize, &skb->sk->sk_wmem_alloc);
 655                 }
 656                 while (copied) {
 657                         int size = min_t(int, copied, PAGE_SIZE - start);
 658                         skb_fill_page_desc(skb, frag++, pages[n], start, size);
 659                         start = 0;
 660                         copied -= size;
 661                         n++;
 662                 }
 663         }
 664         return 0;
 665 }
 666 EXPORT_SYMBOL(__zerocopy_sg_from_iter);
 667 
 668 /**
 669  *      zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
 670  *      @skb: buffer to copy
 671  *      @from: the source to copy from
 672  *
 673  *      The function will first copy up to headlen, and then pin the userspace
 674  *      pages and build frags through them.
 675  *
 676  *      Returns 0, -EFAULT or -EMSGSIZE.
 677  */
 678 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
 679 {
 680         int copy = min_t(int, skb_headlen(skb), iov_iter_count(from));
 681 
 682         /* copy up to skb headlen */
 683         if (skb_copy_datagram_from_iter(skb, 0, from, copy))
 684                 return -EFAULT;
 685 
 686         return __zerocopy_sg_from_iter(NULL, skb, from, ~0U);
 687 }
 688 EXPORT_SYMBOL(zerocopy_sg_from_iter);
 689 
 690 /**
 691  *      skb_copy_and_csum_datagram_iter - Copy datagram to an iovec iterator
 692  *          and update a checksum.
 693  *      @skb: buffer to copy
 694  *      @offset: offset in the buffer to start copying from
 695  *      @to: iovec iterator to copy to
 696  *      @len: amount of data to copy from buffer to iovec
 697  *      @csump: checksum pointer
 698  */
 699 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 700                                       struct iov_iter *to, int len,
 701                                       __wsum *csump)
 702 {
 703         return __skb_datagram_iter(skb, offset, to, len, true,
 704                         csum_and_copy_to_iter, csump);
 705 }
 706 
 707 /**
 708  *      skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
 709  *      @skb: skbuff
 710  *      @hlen: hardware length
 711  *      @msg: destination
 712  *
 713  *      Caller _must_ check that skb will fit to this iovec.
 714  *
 715  *      Returns: 0       - success.
 716  *               -EINVAL - checksum failure.
 717  *               -EFAULT - fault during copy.
 718  */
 719 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb,
 720                                    int hlen, struct msghdr *msg)
 721 {
 722         __wsum csum;
 723         int chunk = skb->len - hlen;
 724 
 725         if (!chunk)
 726                 return 0;
 727 
 728         if (msg_data_left(msg) < chunk) {
 729                 if (__skb_checksum_complete(skb))
 730                         return -EINVAL;
 731                 if (skb_copy_datagram_msg(skb, hlen, msg, chunk))
 732                         goto fault;
 733         } else {
 734                 csum = csum_partial(skb->data, hlen, skb->csum);
 735                 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter,
 736                                                chunk, &csum))
 737                         goto fault;
 738 
 739                 if (csum_fold(csum)) {
 740                         iov_iter_revert(&msg->msg_iter, chunk);
 741                         return -EINVAL;
 742                 }
 743 
 744                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
 745                     !skb->csum_complete_sw)
 746                         netdev_rx_csum_fault(NULL, skb);
 747         }
 748         return 0;
 749 fault:
 750         return -EFAULT;
 751 }
 752 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg);
 753 
 754 /**
 755  *      datagram_poll - generic datagram poll
 756  *      @file: file struct
 757  *      @sock: socket
 758  *      @wait: poll table
 759  *
 760  *      Datagram poll: Again totally generic. This also handles
 761  *      sequenced packet sockets providing the socket receive queue
 762  *      is only ever holding data ready to receive.
 763  *
 764  *      Note: when you *don't* use this routine for this protocol,
 765  *      and you use a different write policy from sock_writeable()
 766  *      then please supply your own write_space callback.
 767  */
 768 __poll_t datagram_poll(struct file *file, struct socket *sock,
 769                            poll_table *wait)
 770 {
 771         struct sock *sk = sock->sk;
 772         __poll_t mask;
 773 
 774         sock_poll_wait(file, sock, wait);
 775         mask = 0;
 776 
 777         /* exceptional events? */
 778         if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue))
 779                 mask |= EPOLLERR |
 780                         (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0);
 781 
 782         if (sk->sk_shutdown & RCV_SHUTDOWN)
 783                 mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM;
 784         if (sk->sk_shutdown == SHUTDOWN_MASK)
 785                 mask |= EPOLLHUP;
 786 
 787         /* readable? */
 788         if (!skb_queue_empty_lockless(&sk->sk_receive_queue))
 789                 mask |= EPOLLIN | EPOLLRDNORM;
 790 
 791         /* Connection-based need to check for termination and startup */
 792         if (connection_based(sk)) {
 793                 if (sk->sk_state == TCP_CLOSE)
 794                         mask |= EPOLLHUP;
 795                 /* connection hasn't started yet? */
 796                 if (sk->sk_state == TCP_SYN_SENT)
 797                         return mask;
 798         }
 799 
 800         /* writable? */
 801         if (sock_writeable(sk))
 802                 mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND;
 803         else
 804                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
 805 
 806         return mask;
 807 }
 808 EXPORT_SYMBOL(datagram_poll);