~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/net/core/datagram.c

Version: ~ [ linux-5.8-rc3 ] ~ [ linux-5.7.5 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.48 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.129 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.185 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.228 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.228 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 <linux/uaccess.h>
 40 #include <linux/mm.h>
 41 #include <linux/interrupt.h>
 42 #include <linux/errno.h>
 43 #include <linux/sched.h>
 44 #include <linux/inet.h>
 45 #include <linux/netdevice.h>
 46 #include <linux/rtnetlink.h>
 47 #include <linux/poll.h>
 48 #include <linux/highmem.h>
 49 #include <linux/spinlock.h>
 50 #include <linux/slab.h>
 51 #include <linux/pagemap.h>
 52 #include <linux/uio.h>
 53 
 54 #include <net/protocol.h>
 55 #include <linux/skbuff.h>
 56 
 57 #include <net/checksum.h>
 58 #include <net/sock.h>
 59 #include <net/tcp_states.h>
 60 #include <trace/events/skb.h>
 61 #include <net/busy_poll.h>
 62 
 63 /*
 64  *      Is a socket 'connection oriented' ?
 65  */
 66 static inline int connection_based(struct sock *sk)
 67 {
 68         return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
 69 }
 70 
 71 static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync,
 72                                   void *key)
 73 {
 74         unsigned long bits = (unsigned long)key;
 75 
 76         /*
 77          * Avoid a wakeup if event not interesting for us
 78          */
 79         if (bits && !(bits & (POLLIN | POLLERR)))
 80                 return 0;
 81         return autoremove_wake_function(wait, mode, sync, key);
 82 }
 83 /*
 84  * Wait for the last received packet to be different from skb
 85  */
 86 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
 87                                 const struct sk_buff *skb)
 88 {
 89         int error;
 90         DEFINE_WAIT_FUNC(wait, receiver_wake_function);
 91 
 92         prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
 93 
 94         /* Socket errors? */
 95         error = sock_error(sk);
 96         if (error)
 97                 goto out_err;
 98 
 99         if (sk->sk_receive_queue.prev != skb)
100                 goto out;
101 
102         /* Socket shut down? */
103         if (sk->sk_shutdown & RCV_SHUTDOWN)
104                 goto out_noerr;
105 
106         /* Sequenced packets can come disconnected.
107          * If so we report the problem
108          */
109         error = -ENOTCONN;
110         if (connection_based(sk) &&
111             !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN))
112                 goto out_err;
113 
114         /* handle signals */
115         if (signal_pending(current))
116                 goto interrupted;
117 
118         error = 0;
119         *timeo_p = schedule_timeout(*timeo_p);
120 out:
121         finish_wait(sk_sleep(sk), &wait);
122         return error;
123 interrupted:
124         error = sock_intr_errno(*timeo_p);
125 out_err:
126         *err = error;
127         goto out;
128 out_noerr:
129         *err = 0;
130         error = 1;
131         goto out;
132 }
133 EXPORT_SYMBOL(__skb_wait_for_more_packets);
134 
135 static struct sk_buff *skb_set_peeked(struct sk_buff *skb)
136 {
137         struct sk_buff *nskb;
138 
139         if (skb->peeked)
140                 return skb;
141 
142         /* We have to unshare an skb before modifying it. */
143         if (!skb_shared(skb))
144                 goto done;
145 
146         nskb = skb_clone(skb, GFP_ATOMIC);
147         if (!nskb)
148                 return ERR_PTR(-ENOMEM);
149 
150         skb->prev->next = nskb;
151         skb->next->prev = nskb;
152         nskb->prev = skb->prev;
153         nskb->next = skb->next;
154 
155         consume_skb(skb);
156         skb = nskb;
157 
158 done:
159         skb->peeked = 1;
160 
161         return skb;
162 }
163 
164 /**
165  *      __skb_try_recv_datagram - Receive a datagram skbuff
166  *      @sk: socket
167  *      @flags: MSG_ flags
168  *      @destructor: invoked under the receive lock on successful dequeue
169  *      @peeked: returns non-zero if this packet has been seen before
170  *      @off: an offset in bytes to peek skb from. Returns an offset
171  *            within an skb where data actually starts
172  *      @err: error code returned
173  *      @last: set to last peeked message to inform the wait function
174  *             what to look for when peeking
175  *
176  *      Get a datagram skbuff, understands the peeking, nonblocking wakeups
177  *      and possible races. This replaces identical code in packet, raw and
178  *      udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
179  *      the long standing peek and read race for datagram sockets. If you
180  *      alter this routine remember it must be re-entrant.
181  *
182  *      This function will lock the socket if a skb is returned, so
183  *      the caller needs to unlock the socket in that case (usually by
184  *      calling skb_free_datagram). Returns NULL with *err set to
185  *      -EAGAIN if no data was available or to some other value if an
186  *      error was detected.
187  *
188  *      * It does not lock socket since today. This function is
189  *      * free of race conditions. This measure should/can improve
190  *      * significantly datagram socket latencies at high loads,
191  *      * when data copying to user space takes lots of time.
192  *      * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
193  *      *  8) Great win.)
194  *      *                                           --ANK (980729)
195  *
196  *      The order of the tests when we find no data waiting are specified
197  *      quite explicitly by POSIX 1003.1g, don't change them without having
198  *      the standard around please.
199  */
200 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned int flags,
201                                         void (*destructor)(struct sock *sk,
202                                                            struct sk_buff *skb),
203                                         int *peeked, int *off, int *err,
204                                         struct sk_buff **last)
205 {
206         struct sk_buff_head *queue = &sk->sk_receive_queue;
207         struct sk_buff *skb;
208         unsigned long cpu_flags;
209         /*
210          * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
211          */
212         int error = sock_error(sk);
213 
214         if (error)
215                 goto no_packet;
216 
217         *peeked = 0;
218         do {
219                 /* Again only user level code calls this function, so nothing
220                  * interrupt level will suddenly eat the receive_queue.
221                  *
222                  * Look at current nfs client by the way...
223                  * However, this function was correct in any case. 8)
224                  */
225                 int _off = *off;
226 
227                 *last = (struct sk_buff *)queue;
228                 spin_lock_irqsave(&queue->lock, cpu_flags);
229                 skb_queue_walk(queue, skb) {
230                         *last = skb;
231                         if (flags & MSG_PEEK) {
232                                 if (_off >= skb->len && (skb->len || _off ||
233                                                          skb->peeked)) {
234                                         _off -= skb->len;
235                                         continue;
236                                 }
237                                 if (!skb->len) {
238                                         skb = skb_set_peeked(skb);
239                                         if (IS_ERR(skb)) {
240                                                 error = PTR_ERR(skb);
241                                                 spin_unlock_irqrestore(&queue->lock,
242                                                                        cpu_flags);
243                                                 goto no_packet;
244                                         }
245                                 }
246                                 *peeked = 1;
247                                 atomic_inc(&skb->users);
248                         } else {
249                                 __skb_unlink(skb, queue);
250                                 if (destructor)
251                                         destructor(sk, skb);
252                         }
253                         spin_unlock_irqrestore(&queue->lock, cpu_flags);
254                         *off = _off;
255                         return skb;
256                 }
257 
258                 spin_unlock_irqrestore(&queue->lock, cpu_flags);
259 
260                 if (!sk_can_busy_loop(sk))
261                         break;
262 
263                 sk_busy_loop(sk, flags & MSG_DONTWAIT);
264         } while (!skb_queue_empty(&sk->sk_receive_queue));
265 
266         error = -EAGAIN;
267 
268 no_packet:
269         *err = error;
270         return NULL;
271 }
272 EXPORT_SYMBOL(__skb_try_recv_datagram);
273 
274 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
275                                     void (*destructor)(struct sock *sk,
276                                                        struct sk_buff *skb),
277                                     int *peeked, int *off, int *err)
278 {
279         struct sk_buff *skb, *last;
280         long timeo;
281 
282         timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
283 
284         do {
285                 skb = __skb_try_recv_datagram(sk, flags, destructor, peeked,
286                                               off, err, &last);
287                 if (skb)
288                         return skb;
289 
290                 if (*err != -EAGAIN)
291                         break;
292         } while (timeo &&
293                 !__skb_wait_for_more_packets(sk, err, &timeo, last));
294 
295         return NULL;
296 }
297 EXPORT_SYMBOL(__skb_recv_datagram);
298 
299 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
300                                   int noblock, int *err)
301 {
302         int peeked, off = 0;
303 
304         return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
305                                    NULL, &peeked, &off, err);
306 }
307 EXPORT_SYMBOL(skb_recv_datagram);
308 
309 void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
310 {
311         consume_skb(skb);
312         sk_mem_reclaim_partial(sk);
313 }
314 EXPORT_SYMBOL(skb_free_datagram);
315 
316 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len)
317 {
318         bool slow;
319 
320         if (likely(atomic_read(&skb->users) == 1))
321                 smp_rmb();
322         else if (likely(!atomic_dec_and_test(&skb->users))) {
323                 sk_peek_offset_bwd(sk, len);
324                 return;
325         }
326 
327         slow = lock_sock_fast(sk);
328         sk_peek_offset_bwd(sk, len);
329         skb_orphan(skb);
330         sk_mem_reclaim_partial(sk);
331         unlock_sock_fast(sk, slow);
332 
333         /* skb is now orphaned, can be freed outside of locked section */
334         __kfree_skb(skb);
335 }
336 EXPORT_SYMBOL(__skb_free_datagram_locked);
337 
338 int __sk_queue_drop_skb(struct sock *sk, struct sk_buff *skb,
339                         unsigned int flags,
340                         void (*destructor)(struct sock *sk,
341                                            struct sk_buff *skb))
342 {
343         int err = 0;
344 
345         if (flags & MSG_PEEK) {
346                 err = -ENOENT;
347                 spin_lock_bh(&sk->sk_receive_queue.lock);
348                 if (skb->next) {
349                         __skb_unlink(skb, &sk->sk_receive_queue);
350                         atomic_dec(&skb->users);
351                         if (destructor)
352                                 destructor(sk, skb);
353                         err = 0;
354                 }
355                 spin_unlock_bh(&sk->sk_receive_queue.lock);
356         }
357 
358         atomic_inc(&sk->sk_drops);
359         return err;
360 }
361 EXPORT_SYMBOL(__sk_queue_drop_skb);
362 
363 /**
364  *      skb_kill_datagram - Free a datagram skbuff forcibly
365  *      @sk: socket
366  *      @skb: datagram skbuff
367  *      @flags: MSG_ flags
368  *
369  *      This function frees a datagram skbuff that was received by
370  *      skb_recv_datagram.  The flags argument must match the one
371  *      used for skb_recv_datagram.
372  *
373  *      If the MSG_PEEK flag is set, and the packet is still on the
374  *      receive queue of the socket, it will be taken off the queue
375  *      before it is freed.
376  *
377  *      This function currently only disables BH when acquiring the
378  *      sk_receive_queue lock.  Therefore it must not be used in a
379  *      context where that lock is acquired in an IRQ context.
380  *
381  *      It returns 0 if the packet was removed by us.
382  */
383 
384 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
385 {
386         int err = __sk_queue_drop_skb(sk, skb, flags, NULL);
387 
388         kfree_skb(skb);
389         sk_mem_reclaim_partial(sk);
390         return err;
391 }
392 EXPORT_SYMBOL(skb_kill_datagram);
393 
394 /**
395  *      skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
396  *      @skb: buffer to copy
397  *      @offset: offset in the buffer to start copying from
398  *      @to: iovec iterator to copy to
399  *      @len: amount of data to copy from buffer to iovec
400  */
401 int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
402                            struct iov_iter *to, int len)
403 {
404         int start = skb_headlen(skb);
405         int i, copy = start - offset, start_off = offset, n;
406         struct sk_buff *frag_iter;
407 
408         trace_skb_copy_datagram_iovec(skb, len);
409 
410         /* Copy header. */
411         if (copy > 0) {
412                 if (copy > len)
413                         copy = len;
414                 n = copy_to_iter(skb->data + offset, copy, to);
415                 offset += n;
416                 if (n != copy)
417                         goto short_copy;
418                 if ((len -= copy) == 0)
419                         return 0;
420         }
421 
422         /* Copy paged appendix. Hmm... why does this look so complicated? */
423         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
424                 int end;
425                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
426 
427                 WARN_ON(start > offset + len);
428 
429                 end = start + skb_frag_size(frag);
430                 if ((copy = end - offset) > 0) {
431                         if (copy > len)
432                                 copy = len;
433                         n = copy_page_to_iter(skb_frag_page(frag),
434                                               frag->page_offset + offset -
435                                               start, copy, to);
436                         offset += n;
437                         if (n != copy)
438                                 goto short_copy;
439                         if (!(len -= copy))
440                                 return 0;
441                 }
442                 start = end;
443         }
444 
445         skb_walk_frags(skb, frag_iter) {
446                 int end;
447 
448                 WARN_ON(start > offset + len);
449 
450                 end = start + frag_iter->len;
451                 if ((copy = end - offset) > 0) {
452                         if (copy > len)
453                                 copy = len;
454                         if (skb_copy_datagram_iter(frag_iter, offset - start,
455                                                    to, copy))
456                                 goto fault;
457                         if ((len -= copy) == 0)
458                                 return 0;
459                         offset += copy;
460                 }
461                 start = end;
462         }
463         if (!len)
464                 return 0;
465 
466         /* This is not really a user copy fault, but rather someone
467          * gave us a bogus length on the skb.  We should probably
468          * print a warning here as it may indicate a kernel bug.
469          */
470 
471 fault:
472         iov_iter_revert(to, offset - start_off);
473         return -EFAULT;
474 
475 short_copy:
476         if (iov_iter_count(to))
477                 goto fault;
478 
479         return 0;
480 }
481 EXPORT_SYMBOL(skb_copy_datagram_iter);
482 
483 /**
484  *      skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
485  *      @skb: buffer to copy
486  *      @offset: offset in the buffer to start copying to
487  *      @from: the copy source
488  *      @len: amount of data to copy to buffer from iovec
489  *
490  *      Returns 0 or -EFAULT.
491  */
492 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
493                                  struct iov_iter *from,
494                                  int len)
495 {
496         int start = skb_headlen(skb);
497         int i, copy = start - offset;
498         struct sk_buff *frag_iter;
499 
500         /* Copy header. */
501         if (copy > 0) {
502                 if (copy > len)
503                         copy = len;
504                 if (copy_from_iter(skb->data + offset, copy, from) != copy)
505                         goto fault;
506                 if ((len -= copy) == 0)
507                         return 0;
508                 offset += copy;
509         }
510 
511         /* Copy paged appendix. Hmm... why does this look so complicated? */
512         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
513                 int end;
514                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
515 
516                 WARN_ON(start > offset + len);
517 
518                 end = start + skb_frag_size(frag);
519                 if ((copy = end - offset) > 0) {
520                         size_t copied;
521 
522                         if (copy > len)
523                                 copy = len;
524                         copied = copy_page_from_iter(skb_frag_page(frag),
525                                           frag->page_offset + offset - start,
526                                           copy, from);
527                         if (copied != copy)
528                                 goto fault;
529 
530                         if (!(len -= copy))
531                                 return 0;
532                         offset += copy;
533                 }
534                 start = end;
535         }
536 
537         skb_walk_frags(skb, frag_iter) {
538                 int end;
539 
540                 WARN_ON(start > offset + len);
541 
542                 end = start + frag_iter->len;
543                 if ((copy = end - offset) > 0) {
544                         if (copy > len)
545                                 copy = len;
546                         if (skb_copy_datagram_from_iter(frag_iter,
547                                                         offset - start,
548                                                         from, copy))
549                                 goto fault;
550                         if ((len -= copy) == 0)
551                                 return 0;
552                         offset += copy;
553                 }
554                 start = end;
555         }
556         if (!len)
557                 return 0;
558 
559 fault:
560         return -EFAULT;
561 }
562 EXPORT_SYMBOL(skb_copy_datagram_from_iter);
563 
564 /**
565  *      zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
566  *      @skb: buffer to copy
567  *      @from: the source to copy from
568  *
569  *      The function will first copy up to headlen, and then pin the userspace
570  *      pages and build frags through them.
571  *
572  *      Returns 0, -EFAULT or -EMSGSIZE.
573  */
574 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
575 {
576         int len = iov_iter_count(from);
577         int copy = min_t(int, skb_headlen(skb), len);
578         int frag = 0;
579 
580         /* copy up to skb headlen */
581         if (skb_copy_datagram_from_iter(skb, 0, from, copy))
582                 return -EFAULT;
583 
584         while (iov_iter_count(from)) {
585                 struct page *pages[MAX_SKB_FRAGS];
586                 size_t start;
587                 ssize_t copied;
588                 unsigned long truesize;
589                 int n = 0;
590 
591                 if (frag == MAX_SKB_FRAGS)
592                         return -EMSGSIZE;
593 
594                 copied = iov_iter_get_pages(from, pages, ~0U,
595                                             MAX_SKB_FRAGS - frag, &start);
596                 if (copied < 0)
597                         return -EFAULT;
598 
599                 iov_iter_advance(from, copied);
600 
601                 truesize = PAGE_ALIGN(copied + start);
602                 skb->data_len += copied;
603                 skb->len += copied;
604                 skb->truesize += truesize;
605                 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
606                 while (copied) {
607                         int size = min_t(int, copied, PAGE_SIZE - start);
608                         skb_fill_page_desc(skb, frag++, pages[n], start, size);
609                         start = 0;
610                         copied -= size;
611                         n++;
612                 }
613         }
614         return 0;
615 }
616 EXPORT_SYMBOL(zerocopy_sg_from_iter);
617 
618 static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
619                                       struct iov_iter *to, int len,
620                                       __wsum *csump)
621 {
622         int start = skb_headlen(skb);
623         int i, copy = start - offset, start_off = offset;
624         struct sk_buff *frag_iter;
625         int pos = 0;
626         int n;
627 
628         /* Copy header. */
629         if (copy > 0) {
630                 if (copy > len)
631                         copy = len;
632                 n = csum_and_copy_to_iter(skb->data + offset, copy, csump, to);
633                 offset += n;
634                 if (n != copy)
635                         goto fault;
636                 if ((len -= copy) == 0)
637                         return 0;
638                 pos = copy;
639         }
640 
641         for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
642                 int end;
643                 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
644 
645                 WARN_ON(start > offset + len);
646 
647                 end = start + skb_frag_size(frag);
648                 if ((copy = end - offset) > 0) {
649                         __wsum csum2 = 0;
650                         struct page *page = skb_frag_page(frag);
651                         u8  *vaddr = kmap(page);
652 
653                         if (copy > len)
654                                 copy = len;
655                         n = csum_and_copy_to_iter(vaddr + frag->page_offset +
656                                                   offset - start, copy,
657                                                   &csum2, to);
658                         kunmap(page);
659                         offset += n;
660                         if (n != copy)
661                                 goto fault;
662                         *csump = csum_block_add(*csump, csum2, pos);
663                         if (!(len -= copy))
664                                 return 0;
665                         pos += copy;
666                 }
667                 start = end;
668         }
669 
670         skb_walk_frags(skb, frag_iter) {
671                 int end;
672 
673                 WARN_ON(start > offset + len);
674 
675                 end = start + frag_iter->len;
676                 if ((copy = end - offset) > 0) {
677                         __wsum csum2 = 0;
678                         if (copy > len)
679                                 copy = len;
680                         if (skb_copy_and_csum_datagram(frag_iter,
681                                                        offset - start,
682                                                        to, copy,
683                                                        &csum2))
684                                 goto fault;
685                         *csump = csum_block_add(*csump, csum2, pos);
686                         if ((len -= copy) == 0)
687                                 return 0;
688                         offset += copy;
689                         pos += copy;
690                 }
691                 start = end;
692         }
693         if (!len)
694                 return 0;
695 
696 fault:
697         iov_iter_revert(to, offset - start_off);
698         return -EFAULT;
699 }
700 
701 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
702 {
703         __sum16 sum;
704 
705         sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
706         if (likely(!sum)) {
707                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
708                     !skb->csum_complete_sw)
709                         netdev_rx_csum_fault(skb->dev);
710         }
711         if (!skb_shared(skb))
712                 skb->csum_valid = !sum;
713         return sum;
714 }
715 EXPORT_SYMBOL(__skb_checksum_complete_head);
716 
717 __sum16 __skb_checksum_complete(struct sk_buff *skb)
718 {
719         __wsum csum;
720         __sum16 sum;
721 
722         csum = skb_checksum(skb, 0, skb->len, 0);
723 
724         /* skb->csum holds pseudo checksum */
725         sum = csum_fold(csum_add(skb->csum, csum));
726         if (likely(!sum)) {
727                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
728                     !skb->csum_complete_sw)
729                         netdev_rx_csum_fault(skb->dev);
730         }
731 
732         if (!skb_shared(skb)) {
733                 /* Save full packet checksum */
734                 skb->csum = csum;
735                 skb->ip_summed = CHECKSUM_COMPLETE;
736                 skb->csum_complete_sw = 1;
737                 skb->csum_valid = !sum;
738         }
739 
740         return sum;
741 }
742 EXPORT_SYMBOL(__skb_checksum_complete);
743 
744 /**
745  *      skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec.
746  *      @skb: skbuff
747  *      @hlen: hardware length
748  *      @msg: destination
749  *
750  *      Caller _must_ check that skb will fit to this iovec.
751  *
752  *      Returns: 0       - success.
753  *               -EINVAL - checksum failure.
754  *               -EFAULT - fault during copy.
755  */
756 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb,
757                                    int hlen, struct msghdr *msg)
758 {
759         __wsum csum;
760         int chunk = skb->len - hlen;
761 
762         if (!chunk)
763                 return 0;
764 
765         if (msg_data_left(msg) < chunk) {
766                 if (__skb_checksum_complete(skb))
767                         return -EINVAL;
768                 if (skb_copy_datagram_msg(skb, hlen, msg, chunk))
769                         goto fault;
770         } else {
771                 csum = csum_partial(skb->data, hlen, skb->csum);
772                 if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter,
773                                                chunk, &csum))
774                         goto fault;
775 
776                 if (csum_fold(csum)) {
777                         iov_iter_revert(&msg->msg_iter, chunk);
778                         return -EINVAL;
779                 }
780 
781                 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
782                         netdev_rx_csum_fault(skb->dev);
783         }
784         return 0;
785 fault:
786         return -EFAULT;
787 }
788 EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg);
789 
790 /**
791  *      datagram_poll - generic datagram poll
792  *      @file: file struct
793  *      @sock: socket
794  *      @wait: poll table
795  *
796  *      Datagram poll: Again totally generic. This also handles
797  *      sequenced packet sockets providing the socket receive queue
798  *      is only ever holding data ready to receive.
799  *
800  *      Note: when you _don't_ use this routine for this protocol,
801  *      and you use a different write policy from sock_writeable()
802  *      then please supply your own write_space callback.
803  */
804 unsigned int datagram_poll(struct file *file, struct socket *sock,
805                            poll_table *wait)
806 {
807         struct sock *sk = sock->sk;
808         unsigned int mask;
809 
810         sock_poll_wait(file, sk_sleep(sk), wait);
811         mask = 0;
812 
813         /* exceptional events? */
814         if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
815                 mask |= POLLERR |
816                         (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
817 
818         if (sk->sk_shutdown & RCV_SHUTDOWN)
819                 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
820         if (sk->sk_shutdown == SHUTDOWN_MASK)
821                 mask |= POLLHUP;
822 
823         /* readable? */
824         if (!skb_queue_empty(&sk->sk_receive_queue))
825                 mask |= POLLIN | POLLRDNORM;
826 
827         /* Connection-based need to check for termination and startup */
828         if (connection_based(sk)) {
829                 if (sk->sk_state == TCP_CLOSE)
830                         mask |= POLLHUP;
831                 /* connection hasn't started yet? */
832                 if (sk->sk_state == TCP_SYN_SENT)
833                         return mask;
834         }
835 
836         /* writable? */
837         if (sock_writeable(sk))
838                 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
839         else
840                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
841 
842         return mask;
843 }
844 EXPORT_SYMBOL(datagram_poll);
845 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp