net/core/datagram.c

   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 <asm/system.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
  52 #include <net/protocol.h>
  53 #include <linux/skbuff.h>
  54
  55 #include <net/checksum.h>
  56 #include <net/sock.h>
  57 #include <net/tcp_states.h>
  58
  59 /*
  60  *      Is a socket 'connection oriented' ?
  61  */
  62 static inline int connection_based(struct sock *sk)
  63 {
  64         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
  65 }
  66
  67 /*
  68  * Wait for a packet..
  69  */
  70 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
  71 {
  72         int error;
  73         DEFINE_WAIT(wait);
  74
  75         prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
  76
  77         /* Socket errors? */
  78         error = sock_error(sk);
  79         if (error)
  80                 goto out_err;
  81
  82         if (!skb_queue_empty(&sk->sk_receive_queue))
  83                 goto out;
  84
  85         /* Socket shut down? */
  86         if (sk->sk_shutdown & RCV_SHUTDOWN)
  87                 goto out_noerr;
  88
  89         /* Sequenced packets can come disconnected.
  90          * If so we report the problem
  91          */
  92         error = -ENOTCONN;
  93         if (connection_based(sk) &&
  94             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
  95                 goto out_err;
  96
  97         /* handle signals */
  98         if (signal_pending(current))
  99                 goto interrupted;
 100
 101         error = 0;
 102         *timeo_p = schedule_timeout(*timeo_p);
 103 out:
 104         finish_wait(sk->sk_sleep, &wait);
 105         return error;
 106 interrupted:
 107         error = sock_intr_errno(*timeo_p);
 108 out_err:
 109         *err = error;
 110         goto out;
 111 out_noerr:
 112         *err = 0;
 113         error = 1;
 114         goto out;
 115 }
 116
 117 /**
 118  *      __skb_recv_datagram - Receive a datagram skbuff
 119  *      @sk: socket
 120  *      @flags: MSG_ flags
 121  *      @peeked: returns non-zero if this packet has been seen before
 122  *      @err: error code returned
 123  *
 124  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
 125  *      and possible races. This replaces identical code in packet, raw and
 126  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
 127  *      the long standing peek and read race for datagram sockets. If you
 128  *      alter this routine remember it must be re-entrant.
 129  *
 130  *      This function will lock the socket if a skb is returned, so the caller
 131  *      needs to unlock the socket in that case (usually by calling
 132  *      skb_free_datagram)
 133  *
 134  *      * It does not lock socket since today. This function is
 135  *      * free of race conditions. This measure should/can improve
 136  *      * significantly datagram socket latencies at high loads,
 137  *      * when data copying to user space takes lots of time.
 138  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
 139  *      *  8) Great win.)
 140  *      *                                           --ANK (980729)
 141  *
 142  *      The order of the tests when we find no data waiting are specified
 143  *      quite explicitly by POSIX 1003.1g, don't change them without having
 144  *      the standard around please.
 145  */
 146 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
 147                                     int *peeked, int *err)
 148 {
 149         struct sk_buff *skb;
 150         long timeo;
 151         /*
 152          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
 153          */
 154         int error = sock_error(sk);
 155
 156         if (error)
 157                 goto no_packet;
 158
 159         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
 160
 161         do {
 162                 /* Again only user level code calls this function, so nothing
 163                  * interrupt level will suddenly eat the receive_queue.
 164                  *
 165                  * Look at current nfs client by the way...
 166                  * However, this function was corrent in any case. 8)
 167                  */
 168                 unsigned long cpu_flags;
 169
 170                 spin_lock_irqsave(&sk->sk_receive_queue.lock, cpu_flags);
 171                 skb = skb_peek(&sk->sk_receive_queue);
 172                 if (skb) {
 173                         *peeked = skb->peeked;
 174                         if (flags & MSG_PEEK) {
 175                                 skb->peeked = 1;
 176                                 atomic_inc(&skb->users);
 177                         } else
 178                                 __skb_unlink(skb, &sk->sk_receive_queue);
 179                 }
 180                 spin_unlock_irqrestore(&sk->sk_receive_queue.lock, cpu_flags);
 181
 182                 if (skb)
 183                         return skb;
 184
 185                 /* User doesn't want to wait */
 186                 error = -EAGAIN;
 187                 if (!timeo)
 188                         goto no_packet;
 189
 190         } while (!wait_for_packet(sk, err, &timeo));
 191
 192         return NULL;
 193
 194 no_packet:
 195         *err = error;
 196         return NULL;
 197 }
 198 EXPORT_SYMBOL(__skb_recv_datagram);
 199
 200 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
 201                                   int noblock, int *err)
 202 {
 203         int peeked;
 204
 205         return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
 206                                    &peeked, err);
 207 }
 208
 209 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
 210 {
 211         kfree_skb(skb);
 212         sk_mem_reclaim_partial(sk);
 213 }
 214
 215 /**
 216  *      skb_kill_datagram - Free a datagram skbuff forcibly
 217  *      @sk: socket
 218  *      @skb: datagram skbuff
 219  *      @flags: MSG_ flags
 220  *
 221  *      This function frees a datagram skbuff that was received by
 222  *      skb_recv_datagram.  The flags argument must match the one
 223  *      used for skb_recv_datagram.
 224  *
 225  *      If the MSG_PEEK flag is set, and the packet is still on the
 226  *      receive queue of the socket, it will be taken off the queue
 227  *      before it is freed.
 228  *
 229  *      This function currently only disables BH when acquiring the
 230  *      sk_receive_queue lock.  Therefore it must not be used in a
 231  *      context where that lock is acquired in an IRQ context.
 232  *
 233  *      It returns 0 if the packet was removed by us.
 234  */
 235
 236 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
 237 {
 238         int err = 0;
 239
 240         if (flags & MSG_PEEK) {
 241                 err = -ENOENT;
 242                 spin_lock_bh(&sk->sk_receive_queue.lock);
 243                 if (skb == skb_peek(&sk->sk_receive_queue)) {
 244                         __skb_unlink(skb, &sk->sk_receive_queue);
 245                         atomic_dec(&skb->users);
 246                         err = 0;
 247                 }
 248                 spin_unlock_bh(&sk->sk_receive_queue.lock);
 249         }
 250
 251         skb_free_datagram(sk, skb);
 252         return err;
 253 }
 254
 255 EXPORT_SYMBOL(skb_kill_datagram);
 256
 257 /**
 258  *      skb_copy_datagram_iovec - Copy a datagram to an iovec.
 259  *      @skb: buffer to copy
 260  *      @offset: offset in the buffer to start copying from
 261  *      @to: io vector to copy to
 262  *      @len: amount of data to copy from buffer to iovec
 263  *
 264  *      Note: the iovec is modified during the copy.
 265  */
 266 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
 267                             struct iovec *to, int len)
 268 {
 269         int start = skb_headlen(skb);
 270         int i, copy = start - offset;
 271
 272         /* Copy header. */
 273         if (copy > 0) {
 274                 if (copy > len)
 275                         copy = len;
 276                 if (memcpy_toiovec(to, skb->data + offset, copy))
 277                         goto fault;
 278                 if ((len -= copy) == 0)
 279                         return 0;
 280                 offset += copy;
 281         }
 282
 283         /* Copy paged appendix. Hmm... why does this look so complicated? */
 284         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 285                 int end;
 286
 287                 WARN_ON(start > offset + len);
 288
 289                 end = start + skb_shinfo(skb)->frags[i].size;
 290                 if ((copy = end - offset) > 0) {
 291                         int err;
 292                         u8  *vaddr;
 293                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 294                         struct page *page = frag->page;
 295
 296                         if (copy > len)
 297                                 copy = len;
 298                         vaddr = kmap(page);
 299                         err = memcpy_toiovec(to, vaddr + frag->page_offset +
 300                                              offset - start, copy);
 301                         kunmap(page);
 302                         if (err)
 303                                 goto fault;
 304                         if (!(len -= copy))
 305                                 return 0;
 306                         offset += copy;
 307                 }
 308                 start = end;
 309         }
 310
 311         if (skb_shinfo(skb)->frag_list) {
 312                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
 313
 314                 for (; list; list = list->next) {
 315                         int end;
 316
 317                         WARN_ON(start > offset + len);
 318
 319                         end = start + list->len;
 320                         if ((copy = end - offset) > 0) {
 321                                 if (copy > len)
 322                                         copy = len;
 323                                 if (skb_copy_datagram_iovec(list,
 324                                                             offset - start,
 325                                                             to, copy))
 326                                         goto fault;
 327                                 if ((len -= copy) == 0)
 328                                         return 0;
 329                                 offset += copy;
 330                         }
 331                         start = end;
 332                 }
 333         }
 334         if (!len)
 335                 return 0;
 336
 337 fault:
 338         return -EFAULT;
 339 }
 340
 341 /**
 342  *      skb_copy_datagram_from_iovec - Copy a datagram from an iovec.
 343  *      @skb: buffer to copy
 344  *      @offset: offset in the buffer to start copying to
 345  *      @from: io vector to copy to
 346  *      @len: amount of data to copy to buffer from iovec
 347  *
 348  *      Returns 0 or -EFAULT.
 349  *      Note: the iovec is modified during the copy.
 350  */
 351 int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
 352                                  struct iovec *from, int len)
 353 {
 354         int start = skb_headlen(skb);
 355         int i, copy = start - offset;
 356
 357         /* Copy header. */
 358         if (copy > 0) {
 359                 if (copy > len)
 360                         copy = len;
 361                 if (memcpy_fromiovec(skb->data + offset, from, copy))
 362                         goto fault;
 363                 if ((len -= copy) == 0)
 364                         return 0;
 365                 offset += copy;
 366         }
 367
 368         /* Copy paged appendix. Hmm... why does this look so complicated? */
 369         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 370                 int end;
 371
 372                 WARN_ON(start > offset + len);
 373
 374                 end = start + skb_shinfo(skb)->frags[i].size;
 375                 if ((copy = end - offset) > 0) {
 376                         int err;
 377                         u8  *vaddr;
 378                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 379                         struct page *page = frag->page;
 380
 381                         if (copy > len)
 382                                 copy = len;
 383                         vaddr = kmap(page);
 384                         err = memcpy_fromiovec(vaddr + frag->page_offset +
 385                                                offset - start, from, copy);
 386                         kunmap(page);
 387                         if (err)
 388                                 goto fault;
 389
 390                         if (!(len -= copy))
 391                                 return 0;
 392                         offset += copy;
 393                 }
 394                 start = end;
 395         }
 396
 397         if (skb_shinfo(skb)->frag_list) {
 398                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
 399
 400                 for (; list; list = list->next) {
 401                         int end;
 402
 403                         WARN_ON(start > offset + len);
 404
 405                         end = start + list->len;
 406                         if ((copy = end - offset) > 0) {
 407                                 if (copy > len)
 408                                         copy = len;
 409                                 if (skb_copy_datagram_from_iovec(list,
 410                                                                  offset - start,
 411                                                                  from, copy))
 412                                         goto fault;
 413                                 if ((len -= copy) == 0)
 414                                         return 0;
 415                                 offset += copy;
 416                         }
 417                         start = end;
 418                 }
 419         }
 420         if (!len)
 421                 return 0;
 422
 423 fault:
 424         return -EFAULT;
 425 }
 426 EXPORT_SYMBOL(skb_copy_datagram_from_iovec);
 427
 428 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 429                                       u8 __user *to, int len,
 430                                       __wsum *csump)
 431 {
 432         int start = skb_headlen(skb);
 433         int pos = 0;
 434         int i, copy = start - offset;
 435
 436         /* Copy header. */
 437         if (copy > 0) {
 438                 int err = 0;
 439                 if (copy > len)
 440                         copy = len;
 441                 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
 442                                                *csump, &err);
 443                 if (err)
 444                         goto fault;
 445                 if ((len -= copy) == 0)
 446                         return 0;
 447                 offset += copy;
 448                 to += copy;
 449                 pos = copy;
 450         }
 451
 452         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 453                 int end;
 454
 455                 WARN_ON(start > offset + len);
 456
 457                 end = start + skb_shinfo(skb)->frags[i].size;
 458                 if ((copy = end - offset) > 0) {
 459                         __wsum csum2;
 460                         int err = 0;
 461                         u8  *vaddr;
 462                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 463                         struct page *page = frag->page;
 464
 465                         if (copy > len)
 466                                 copy = len;
 467                         vaddr = kmap(page);
 468                         csum2 = csum_and_copy_to_user(vaddr +
 469                                                         frag->page_offset +
 470                                                         offset - start,
 471                                                       to, copy, 0, &err);
 472                         kunmap(page);
 473                         if (err)
 474                                 goto fault;
 475                         *csump = csum_block_add(*csump, csum2, pos);
 476                         if (!(len -= copy))
 477                                 return 0;
 478                         offset += copy;
 479                         to += copy;
 480                         pos += copy;
 481                 }
 482                 start = end;
 483         }
 484
 485         if (skb_shinfo(skb)->frag_list) {
 486                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
 487
 488                 for (; list; list=list->next) {
 489                         int end;
 490
 491                         WARN_ON(start > offset + len);
 492
 493                         end = start + list->len;
 494                         if ((copy = end - offset) > 0) {
 495                                 __wsum csum2 = 0;
 496                                 if (copy > len)
 497                                         copy = len;
 498                                 if (skb_copy_and_csum_datagram(list,
 499                                                                offset - start,
 500                                                                to, copy,
 501                                                                &csum2))
 502                                         goto fault;
 503                                 *csump = csum_block_add(*csump, csum2, pos);
 504                                 if ((len -= copy) == 0)
 505                                         return 0;
 506                                 offset += copy;
 507                                 to += copy;
 508                                 pos += copy;
 509                         }
 510                         start = end;
 511                 }
 512         }
 513         if (!len)
 514                 return 0;
 515
 516 fault:
 517         return -EFAULT;
 518 }
 519
 520 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
 521 {
 522         __sum16 sum;
 523
 524         sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
 525         if (likely(!sum)) {
 526                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 527                         netdev_rx_csum_fault(skb->dev);
 528                 skb->ip_summed = CHECKSUM_UNNECESSARY;
 529         }
 530         return sum;
 531 }
 532 EXPORT_SYMBOL(__skb_checksum_complete_head);
 533
 534 __sum16 __skb_checksum_complete(struct sk_buff *skb)
 535 {
 536         return __skb_checksum_complete_head(skb, skb->len);
 537 }
 538 EXPORT_SYMBOL(__skb_checksum_complete);
 539
 540 /**
 541  *      skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
 542  *      @skb: skbuff
 543  *      @hlen: hardware length
 544  *      @iov: io vector
 545  *
 546  *      Caller _must_ check that skb will fit to this iovec.
 547  *
 548  *      Returns: 0       - success.
 549  *               -EINVAL - checksum failure.
 550  *               -EFAULT - fault during copy. Beware, in this case iovec
 551  *                         can be modified!
 552  */
 553 int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
 554                                      int hlen, struct iovec *iov)
 555 {
 556         __wsum csum;
 557         int chunk = skb->len - hlen;
 558
 559         if (!chunk)
 560                 return 0;
 561
 562         /* Skip filled elements.
 563          * Pretty silly, look at memcpy_toiovec, though 8)
 564          */
 565         while (!iov->iov_len)
 566                 iov++;
 567
 568         if (iov->iov_len < chunk) {
 569                 if (__skb_checksum_complete(skb))
 570                         goto csum_error;
 571                 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
 572                         goto fault;
 573         } else {
 574                 csum = csum_partial(skb->data, hlen, skb->csum);
 575                 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
 576                                                chunk, &csum))
 577                         goto fault;
 578                 if (csum_fold(csum))
 579                         goto csum_error;
 580                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
 581                         netdev_rx_csum_fault(skb->dev);
 582                 iov->iov_len -= chunk;
 583                 iov->iov_base += chunk;
 584         }
 585         return 0;
 586 csum_error:
 587         return -EINVAL;
 588 fault:
 589         return -EFAULT;
 590 }
 591
 592 /**
 593  *      datagram_poll - generic datagram poll
 594  *      @file: file struct
 595  *      @sock: socket
 596  *      @wait: poll table
 597  *
 598  *      Datagram poll: Again totally generic. This also handles
 599  *      sequenced packet sockets providing the socket receive queue
 600  *      is only ever holding data ready to receive.
 601  *
 602  *      Note: when you _don't_ use this routine for this protocol,
 603  *      and you use a different write policy from sock_writeable()
 604  *      then please supply your own write_space callback.
 605  */
 606 unsigned int datagram_poll(struct file *file, struct socket *sock,
 607                            poll_table *wait)
 608 {
 609         struct sock *sk = sock->sk;
 610         unsigned int mask;
 611
 612         poll_wait(file, sk->sk_sleep, wait);
 613         mask = 0;
 614
 615         /* exceptional events? */
 616         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 617                 mask |= POLLERR;
 618         if (sk->sk_shutdown & RCV_SHUTDOWN)
 619                 mask |= POLLRDHUP;
 620         if (sk->sk_shutdown == SHUTDOWN_MASK)
 621                 mask |= POLLHUP;
 622
 623         /* readable? */
 624         if (!skb_queue_empty(&sk->sk_receive_queue) ||
 625             (sk->sk_shutdown & RCV_SHUTDOWN))
 626                 mask |= POLLIN | POLLRDNORM;
 627
 628         /* Connection-based need to check for termination and startup */
 629         if (connection_based(sk)) {
 630                 if (sk->sk_state == TCP_CLOSE)
 631                         mask |= POLLHUP;
 632                 /* connection hasn't started yet? */
 633                 if (sk->sk_state == TCP_SYN_SENT)
 634                         return mask;
 635         }
 636
 637         /* writable? */
 638         if (sock_writeable(sk))
 639                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 640         else
 641                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 642
 643         return mask;
 644 }
 645
 646 EXPORT_SYMBOL(datagram_poll);
 647 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
 648 EXPORT_SYMBOL(skb_copy_datagram_iovec);
 649 EXPORT_SYMBOL(skb_free_datagram);
 650 EXPORT_SYMBOL(skb_recv_datagram);