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@redhat.com>. (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
  51 #include <net/protocol.h>
  52 #include <linux/skbuff.h>
  53
  54 #include <net/checksum.h>
  55 #include <net/sock.h>
  56 #include <net/tcp_states.h>
  57
  58 /*
  59  *      Is a socket 'connection oriented' ?
  60  */
  61 static inline int connection_based(struct sock *sk)
  62 {
  63         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
  64 }
  65
  66 /*
  67  * Wait for a packet..
  68  */
  69 static int wait_for_packet(struct sock *sk, int *err, long *timeo_p)
  70 {
  71         int error;
  72         DEFINE_WAIT(wait);
  73
  74         prepare_to_wait_exclusive(sk->sk_sleep, &wait, TASK_INTERRUPTIBLE);
  75
  76         /* Socket errors? */
  77         error = sock_error(sk);
  78         if (error)
  79                 goto out_err;
  80
  81         if (!skb_queue_empty(&sk->sk_receive_queue))
  82                 goto out;
  83
  84         /* Socket shut down? */
  85         if (sk->sk_shutdown & RCV_SHUTDOWN)
  86                 goto out_noerr;
  87
  88         /* Sequenced packets can come disconnected.
  89          * If so we report the problem
  90          */
  91         error = -ENOTCONN;
  92         if (connection_based(sk) &&
  93             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
  94                 goto out_err;
  95
  96         /* handle signals */
  97         if (signal_pending(current))
  98                 goto interrupted;
  99
 100         error = 0;
 101         *timeo_p = schedule_timeout(*timeo_p);
 102 out:
 103         finish_wait(sk->sk_sleep, &wait);
 104         return error;
 105 interrupted:
 106         error = sock_intr_errno(*timeo_p);
 107 out_err:
 108         *err = error;
 109         goto out;
 110 out_noerr:
 111         *err = 0;
 112         error = 1;
 113         goto out;
 114 }
 115
 116 /**
 117  *      skb_recv_datagram - Receive a datagram skbuff
 118  *      @sk: socket
 119  *      @flags: MSG_ flags
 120  *      @noblock: blocking operation?
 121  *      @err: error code returned
 122  *
 123  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
 124  *      and possible races. This replaces identical code in packet, raw and
 125  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
 126  *      the long standing peek and read race for datagram sockets. If you
 127  *      alter this routine remember it must be re-entrant.
 128  *
 129  *      This function will lock the socket if a skb is returned, so the caller
 130  *      needs to unlock the socket in that case (usually by calling
 131  *      skb_free_datagram)
 132  *
 133  *      * It does not lock socket since today. This function is
 134  *      * free of race conditions. This measure should/can improve
 135  *      * significantly datagram socket latencies at high loads,
 136  *      * when data copying to user space takes lots of time.
 137  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
 138  *      *  8) Great win.)
 139  *      *                                           --ANK (980729)
 140  *
 141  *      The order of the tests when we find no data waiting are specified
 142  *      quite explicitly by POSIX 1003.1g, don't change them without having
 143  *      the standard around please.
 144  */
 145 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
 146                                   int noblock, int *err)
 147 {
 148         struct sk_buff *skb;
 149         long timeo;
 150         /*
 151          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
 152          */
 153         int error = sock_error(sk);
 154
 155         if (error)
 156                 goto no_packet;
 157
 158         timeo = sock_rcvtimeo(sk, noblock);
 159
 160         do {
 161                 /* Again only user level code calls this function, so nothing
 162                  * interrupt level will suddenly eat the receive_queue.
 163                  *
 164                  * Look at current nfs client by the way...
 165                  * However, this function was corrent in any case. 8)
 166                  */
 167                 if (flags & MSG_PEEK) {
 168                         unsigned long cpu_flags;
 169
 170                         spin_lock_irqsave(&sk->sk_receive_queue.lock,
 171                                           cpu_flags);
 172                         skb = skb_peek(&sk->sk_receive_queue);
 173                         if (skb)
 174                                 atomic_inc(&skb->users);
 175                         spin_unlock_irqrestore(&sk->sk_receive_queue.lock,
 176                                                cpu_flags);
 177                 } else
 178                         skb = skb_dequeue(&sk->sk_receive_queue);
 179
 180                 if (skb)
 181                         return skb;
 182
 183                 /* User doesn't want to wait */
 184                 error = -EAGAIN;
 185                 if (!timeo)
 186                         goto no_packet;
 187
 188         } while (!wait_for_packet(sk, err, &timeo));
 189
 190         return NULL;
 191
 192 no_packet:
 193         *err = error;
 194         return NULL;
 195 }
 196
 197 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
 198 {
 199         kfree_skb(skb);
 200 }
 201
 202 /**
 203  *      skb_copy_datagram_iovec - Copy a datagram to an iovec.
 204  *      @skb: buffer to copy
 205  *      @offset: offset in the buffer to start copying from
 206  *      @to: io vector to copy to
 207  *      @len: amount of data to copy from buffer to iovec
 208  *
 209  *      Note: the iovec is modified during the copy.
 210  */
 211 int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
 212                             struct iovec *to, int len)
 213 {
 214         int i, err, fraglen, end = 0;
 215         struct sk_buff *next = skb_shinfo(skb)->frag_list;
 216 next_skb:
 217         fraglen = skb_headlen(skb);
 218         i = -1;
 219
 220         while (1) {
 221                 int start = end;
 222
 223                 if ((end += fraglen) > offset) {
 224                         int copy = end - offset, o = offset - start;
 225
 226                         if (copy > len)
 227                                 copy = len;
 228                         if (i == -1)
 229                                 err = memcpy_toiovec(to, skb->data + o, copy);
 230                         else {
 231                                 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 232                                 struct page *page = frag->page;
 233                                 void *p = kmap(page) + frag->page_offset + o;
 234                                 err = memcpy_toiovec(to, p, copy);
 235                                 kunmap(page);
 236                         }
 237                         if (err)
 238                                 goto fault;
 239                         if (!(len -= copy))
 240                                 return 0;
 241                         offset += copy;
 242                 }
 243                 if (++i >= skb_shinfo(skb)->nr_frags)
 244                         break;
 245                 fraglen = skb_shinfo(skb)->frags[i].size;
 246         }
 247         if (next) {
 248                 skb = next;
 249                 BUG_ON(skb_shinfo(skb)->frag_list);
 250                 next = skb->next;
 251                 goto next_skb;
 252         }
 253 fault:
 254         return -EFAULT;
 255 }
 256
 257 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
 258                                       u8 __user *to, int len,
 259                                       unsigned int *csump)
 260 {
 261         int start = skb_headlen(skb);
 262         int pos = 0;
 263         int i, copy = start - offset;
 264
 265         /* Copy header. */
 266         if (copy > 0) {
 267                 int err = 0;
 268                 if (copy > len)
 269                         copy = len;
 270                 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
 271                                                *csump, &err);
 272                 if (err)
 273                         goto fault;
 274                 if ((len -= copy) == 0)
 275                         return 0;
 276                 offset += copy;
 277                 to += copy;
 278                 pos = copy;
 279         }
 280
 281         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
 282                 int end;
 283
 284                 BUG_TRAP(start <= offset + len);
 285
 286                 end = start + skb_shinfo(skb)->frags[i].size;
 287                 if ((copy = end - offset) > 0) {
 288                         unsigned int csum2;
 289                         int err = 0;
 290                         u8  *vaddr;
 291                         skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 292                         struct page *page = frag->page;
 293
 294                         if (copy > len)
 295                                 copy = len;
 296                         vaddr = kmap(page);
 297                         csum2 = csum_and_copy_to_user(vaddr +
 298                                                         frag->page_offset +
 299                                                         offset - start,
 300                                                       to, copy, 0, &err);
 301                         kunmap(page);
 302                         if (err)
 303                                 goto fault;
 304                         *csump = csum_block_add(*csump, csum2, pos);
 305                         if (!(len -= copy))
 306                                 return 0;
 307                         offset += copy;
 308                         to += copy;
 309                         pos += copy;
 310                 }
 311                 start = end;
 312         }
 313
 314         if (skb_shinfo(skb)->frag_list) {
 315                 struct sk_buff *list = skb_shinfo(skb)->frag_list;
 316
 317                 for (; list; list=list->next) {
 318                         int end;
 319
 320                         BUG_TRAP(start <= offset + len);
 321
 322                         end = start + list->len;
 323                         if ((copy = end - offset) > 0) {
 324                                 unsigned int csum2 = 0;
 325                                 if (copy > len)
 326                                         copy = len;
 327                                 if (skb_copy_and_csum_datagram(list,
 328                                                                offset - start,
 329                                                                to, copy,
 330                                                                &csum2))
 331                                         goto fault;
 332                                 *csump = csum_block_add(*csump, csum2, pos);
 333                                 if ((len -= copy) == 0)
 334                                         return 0;
 335                                 offset += copy;
 336                                 to += copy;
 337                                 pos += copy;
 338                         }
 339                         start = end;
 340                 }
 341         }
 342         if (!len)
 343                 return 0;
 344
 345 fault:
 346         return -EFAULT;
 347 }
 348
 349 /**
 350  *      skb_copy_and_csum_datagram_iovec - Copy and checkum skb to user iovec.
 351  *      @skb: skbuff
 352  *      @hlen: hardware length
 353  *      @iov: io vector
 354  *
 355  *      Caller _must_ check that skb will fit to this iovec.
 356  *
 357  *      Returns: 0       - success.
 358  *               -EINVAL - checksum failure.
 359  *               -EFAULT - fault during copy. Beware, in this case iovec
 360  *                         can be modified!
 361  */
 362 int skb_copy_and_csum_datagram_iovec(const struct sk_buff *skb,
 363                                      int hlen, struct iovec *iov)
 364 {
 365         unsigned int csum;
 366         int chunk = skb->len - hlen;
 367
 368         /* Skip filled elements.
 369          * Pretty silly, look at memcpy_toiovec, though 8)
 370          */
 371         while (!iov->iov_len)
 372                 iov++;
 373
 374         if (iov->iov_len < chunk) {
 375                 if ((unsigned short)csum_fold(skb_checksum(skb, 0, chunk + hlen,
 376                                                            skb->csum)))
 377                         goto csum_error;
 378                 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
 379                         goto fault;
 380         } else {
 381                 csum = csum_partial(skb->data, hlen, skb->csum);
 382                 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
 383                                                chunk, &csum))
 384                         goto fault;
 385                 if ((unsigned short)csum_fold(csum))
 386                         goto csum_error;
 387                 iov->iov_len -= chunk;
 388                 iov->iov_base += chunk;
 389         }
 390         return 0;
 391 csum_error:
 392         return -EINVAL;
 393 fault:
 394         return -EFAULT;
 395 }
 396
 397 /**
 398  *      datagram_poll - generic datagram poll
 399  *      @file: file struct
 400  *      @sock: socket
 401  *      @wait: poll table
 402  *
 403  *      Datagram poll: Again totally generic. This also handles
 404  *      sequenced packet sockets providing the socket receive queue
 405  *      is only ever holding data ready to receive.
 406  *
 407  *      Note: when you _don't_ use this routine for this protocol,
 408  *      and you use a different write policy from sock_writeable()
 409  *      then please supply your own write_space callback.
 410  */
 411 unsigned int datagram_poll(struct file *file, struct socket *sock,
 412                            poll_table *wait)
 413 {
 414         struct sock *sk = sock->sk;
 415         unsigned int mask;
 416
 417         poll_wait(file, sk->sk_sleep, wait);
 418         mask = 0;
 419
 420         /* exceptional events? */
 421         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
 422                 mask |= POLLERR;
 423         if (sk->sk_shutdown == SHUTDOWN_MASK)
 424                 mask |= POLLHUP;
 425
 426         /* readable? */
 427         if (!skb_queue_empty(&sk->sk_receive_queue) ||
 428             (sk->sk_shutdown & RCV_SHUTDOWN))
 429                 mask |= POLLIN | POLLRDNORM;
 430
 431         /* Connection-based need to check for termination and startup */
 432         if (connection_based(sk)) {
 433                 if (sk->sk_state == TCP_CLOSE)
 434                         mask |= POLLHUP;
 435                 /* connection hasn't started yet? */
 436                 if (sk->sk_state == TCP_SYN_SENT)
 437                         return mask;
 438         }
 439
 440         /* writable? */
 441         if (sock_writeable(sk))
 442                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
 443         else
 444                 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
 445
 446         return mask;
 447 }
 448
 449 EXPORT_SYMBOL(datagram_poll);
 450 EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
 451 EXPORT_SYMBOL(skb_copy_datagram_iovec);
 452 EXPORT_SYMBOL(skb_free_datagram);
 453 EXPORT_SYMBOL(skb_recv_datagram);