blob: b6e303b0f01fa818e507ef442c87fc4c5803dbd5 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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 *
Alan Cox113aa832008-10-13 19:01:08 -070012 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 * 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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/mm.h>
41#include <linux/interrupt.h>
42#include <linux/errno.h>
43#include <linux/sched.h>
44#include <linux/inet.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070045#include <linux/netdevice.h>
46#include <linux/rtnetlink.h>
47#include <linux/poll.h>
48#include <linux/highmem.h>
Herbert Xu3305b802005-12-13 23:16:37 -080049#include <linux/spinlock.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090050#include <linux/slab.h>
Jason Wang04335472013-08-06 17:45:07 +080051#include <linux/pagemap.h>
Herbert Xua8f820aa2014-11-07 21:22:22 +080052#include <linux/uio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#include <net/protocol.h>
55#include <linux/skbuff.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Arnaldo Carvalho de Meloc752f072005-08-09 20:08:28 -070057#include <net/checksum.h>
58#include <net/sock.h>
59#include <net/tcp_states.h>
Neil Hormane9b3cc12009-08-13 05:19:44 +000060#include <trace/events/skb.h>
Eliezer Tamir076bb0c2013-07-10 17:13:17 +030061#include <net/busy_poll.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63/*
64 * Is a socket 'connection oriented' ?
65 */
66static inline int connection_based(struct sock *sk)
67{
68 return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM;
69}
70
Eric Dumazet95c96172012-04-15 05:58:06 +000071static int receiver_wake_function(wait_queue_t *wait, unsigned int mode, int sync,
Eric Dumazetbf368e42009-04-28 02:24:21 -070072 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}
Linus Torvalds1da177e2005-04-16 15:20:36 -070083/*
Benjamin Poirier39cc8612013-04-29 11:42:13 +000084 * Wait for the last received packet to be different from skb
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 */
Benjamin Poirier39cc8612013-04-29 11:42:13 +000086static int wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
87 const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -070088{
89 int error;
Eric Dumazetbf368e42009-04-28 02:24:21 -070090 DEFINE_WAIT_FUNC(wait, receiver_wake_function);
Linus Torvalds1da177e2005-04-16 15:20:36 -070091
Eric Dumazetaa395142010-04-20 13:03:51 +000092 prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093
94 /* Socket errors? */
95 error = sock_error(sk);
96 if (error)
97 goto out_err;
98
Benjamin Poirier39cc8612013-04-29 11:42:13 +000099 if (sk->sk_receive_queue.prev != skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100 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);
120out:
Eric Dumazetaa395142010-04-20 13:03:51 +0000121 finish_wait(sk_sleep(sk), &wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122 return error;
123interrupted:
124 error = sock_intr_errno(*timeo_p);
125out_err:
126 *err = error;
127 goto out;
128out_noerr:
129 *err = 0;
130 error = 1;
131 goto out;
132}
133
134/**
Herbert Xua59322b2007-12-05 01:53:40 -0800135 * __skb_recv_datagram - Receive a datagram skbuff
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700136 * @sk: socket
137 * @flags: MSG_ flags
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000138 * @peeked: returns non-zero if this packet has been seen before
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000139 * @off: an offset in bytes to peek skb from. Returns an offset
140 * within an skb where data actually starts
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700141 * @err: error code returned
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 *
143 * Get a datagram skbuff, understands the peeking, nonblocking wakeups
144 * and possible races. This replaces identical code in packet, raw and
145 * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes
146 * the long standing peek and read race for datagram sockets. If you
147 * alter this routine remember it must be re-entrant.
148 *
149 * This function will lock the socket if a skb is returned, so the caller
150 * needs to unlock the socket in that case (usually by calling
151 * skb_free_datagram)
152 *
153 * * It does not lock socket since today. This function is
154 * * free of race conditions. This measure should/can improve
155 * * significantly datagram socket latencies at high loads,
156 * * when data copying to user space takes lots of time.
157 * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet
158 * * 8) Great win.)
159 * * --ANK (980729)
160 *
161 * The order of the tests when we find no data waiting are specified
162 * quite explicitly by POSIX 1003.1g, don't change them without having
163 * the standard around please.
164 */
Eric Dumazet95c96172012-04-15 05:58:06 +0000165struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned int flags,
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000166 int *peeked, int *off, int *err)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167{
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000168 struct sk_buff *skb, *last;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 long timeo;
170 /*
171 * Caller is allowed not to check sk->sk_err before skb_recv_datagram()
172 */
173 int error = sock_error(sk);
174
175 if (error)
176 goto no_packet;
177
Herbert Xua59322b2007-12-05 01:53:40 -0800178 timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179
180 do {
181 /* Again only user level code calls this function, so nothing
182 * interrupt level will suddenly eat the receive_queue.
183 *
184 * Look at current nfs client by the way...
David Shwatrz8917a3c2010-12-02 09:01:55 +0000185 * However, this function was correct in any case. 8)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 */
Herbert Xua59322b2007-12-05 01:53:40 -0800187 unsigned long cpu_flags;
Pavel Emelyanov4934b032012-02-21 07:30:33 +0000188 struct sk_buff_head *queue = &sk->sk_receive_queue;
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000189 int _off = *off;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000191 last = (struct sk_buff *)queue;
Pavel Emelyanov4934b032012-02-21 07:30:33 +0000192 spin_lock_irqsave(&queue->lock, cpu_flags);
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000193 skb_queue_walk(queue, skb) {
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000194 last = skb;
Herbert Xua59322b2007-12-05 01:53:40 -0800195 *peeked = skb->peeked;
196 if (flags & MSG_PEEK) {
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000197 if (_off >= skb->len && (skb->len || _off ||
Benjamin Poirieradd05ad2013-04-29 11:42:12 +0000198 skb->peeked)) {
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000199 _off -= skb->len;
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000200 continue;
201 }
Herbert Xua59322b2007-12-05 01:53:40 -0800202 skb->peeked = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 atomic_inc(&skb->users);
Herbert Xua59322b2007-12-05 01:53:40 -0800204 } else
Pavel Emelyanov4934b032012-02-21 07:30:33 +0000205 __skb_unlink(skb, queue);
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000206
207 spin_unlock_irqrestore(&queue->lock, cpu_flags);
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000208 *off = _off;
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000209 return skb;
Herbert Xua59322b2007-12-05 01:53:40 -0800210 }
Pavel Emelyanov4934b032012-02-21 07:30:33 +0000211 spin_unlock_irqrestore(&queue->lock, cpu_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212
Eliezer Tamircbf55002013-07-08 16:20:34 +0300213 if (sk_can_busy_loop(sk) &&
214 sk_busy_loop(sk, flags & MSG_DONTWAIT))
Eliezer Tamira5b50472013-06-10 11:40:00 +0300215 continue;
216
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 /* User doesn't want to wait */
218 error = -EAGAIN;
219 if (!timeo)
220 goto no_packet;
221
Benjamin Poirier39cc8612013-04-29 11:42:13 +0000222 } while (!wait_for_more_packets(sk, err, &timeo, last));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700223
224 return NULL;
225
226no_packet:
227 *err = error;
228 return NULL;
229}
Herbert Xua59322b2007-12-05 01:53:40 -0800230EXPORT_SYMBOL(__skb_recv_datagram);
231
Eric Dumazet95c96172012-04-15 05:58:06 +0000232struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags,
Herbert Xua59322b2007-12-05 01:53:40 -0800233 int noblock, int *err)
234{
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000235 int peeked, off = 0;
Herbert Xua59322b2007-12-05 01:53:40 -0800236
237 return __skb_recv_datagram(sk, flags | (noblock ? MSG_DONTWAIT : 0),
Pavel Emelyanov3f518bf2012-02-21 07:30:58 +0000238 &peeked, &off, err);
Herbert Xua59322b2007-12-05 01:53:40 -0800239}
Eric Dumazet9e34a5b2010-07-09 21:22:04 +0000240EXPORT_SYMBOL(skb_recv_datagram);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241
242void skb_free_datagram(struct sock *sk, struct sk_buff *skb)
243{
Neil Hormanead2ceb2009-03-11 09:49:55 +0000244 consume_skb(skb);
Eric Dumazet270acef2008-11-05 01:38:06 -0800245 sk_mem_reclaim_partial(sk);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246}
Eric Dumazet9d410c72009-10-30 05:03:53 +0000247EXPORT_SYMBOL(skb_free_datagram);
248
249void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb)
250{
Eric Dumazet8a74ad62010-05-26 19:20:18 +0000251 bool slow;
252
Eric Dumazet93bb64e2010-05-03 23:18:14 -0700253 if (likely(atomic_read(&skb->users) == 1))
254 smp_rmb();
255 else if (likely(!atomic_dec_and_test(&skb->users)))
256 return;
257
Eric Dumazet8a74ad62010-05-26 19:20:18 +0000258 slow = lock_sock_fast(sk);
Eric Dumazet4b0b72f2010-04-28 14:35:48 -0700259 skb_orphan(skb);
260 sk_mem_reclaim_partial(sk);
Eric Dumazet8a74ad62010-05-26 19:20:18 +0000261 unlock_sock_fast(sk, slow);
Eric Dumazet4b0b72f2010-04-28 14:35:48 -0700262
Eric Dumazet93bb64e2010-05-03 23:18:14 -0700263 /* skb is now orphaned, can be freed outside of locked section */
264 __kfree_skb(skb);
Eric Dumazet9d410c72009-10-30 05:03:53 +0000265}
266EXPORT_SYMBOL(skb_free_datagram_locked);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267
268/**
Herbert Xu3305b802005-12-13 23:16:37 -0800269 * skb_kill_datagram - Free a datagram skbuff forcibly
270 * @sk: socket
271 * @skb: datagram skbuff
272 * @flags: MSG_ flags
273 *
274 * This function frees a datagram skbuff that was received by
275 * skb_recv_datagram. The flags argument must match the one
276 * used for skb_recv_datagram.
277 *
278 * If the MSG_PEEK flag is set, and the packet is still on the
279 * receive queue of the socket, it will be taken off the queue
280 * before it is freed.
281 *
282 * This function currently only disables BH when acquiring the
283 * sk_receive_queue lock. Therefore it must not be used in a
284 * context where that lock is acquired in an IRQ context.
Herbert Xu27ab2562007-12-05 01:51:58 -0800285 *
286 * It returns 0 if the packet was removed by us.
Herbert Xu3305b802005-12-13 23:16:37 -0800287 */
288
Herbert Xu27ab2562007-12-05 01:51:58 -0800289int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags)
Herbert Xu3305b802005-12-13 23:16:37 -0800290{
Herbert Xu27ab2562007-12-05 01:51:58 -0800291 int err = 0;
292
Herbert Xu3305b802005-12-13 23:16:37 -0800293 if (flags & MSG_PEEK) {
Herbert Xu27ab2562007-12-05 01:51:58 -0800294 err = -ENOENT;
Herbert Xu3305b802005-12-13 23:16:37 -0800295 spin_lock_bh(&sk->sk_receive_queue.lock);
296 if (skb == skb_peek(&sk->sk_receive_queue)) {
297 __skb_unlink(skb, &sk->sk_receive_queue);
298 atomic_dec(&skb->users);
Herbert Xu27ab2562007-12-05 01:51:58 -0800299 err = 0;
Herbert Xu3305b802005-12-13 23:16:37 -0800300 }
301 spin_unlock_bh(&sk->sk_receive_queue.lock);
302 }
303
John Dykstra61de71c2009-05-08 14:57:01 -0700304 kfree_skb(skb);
Eric Dumazet8edf19c2009-10-15 00:12:40 +0000305 atomic_inc(&sk->sk_drops);
John Dykstra61de71c2009-05-08 14:57:01 -0700306 sk_mem_reclaim_partial(sk);
307
Herbert Xu27ab2562007-12-05 01:51:58 -0800308 return err;
Herbert Xu3305b802005-12-13 23:16:37 -0800309}
Herbert Xu3305b802005-12-13 23:16:37 -0800310EXPORT_SYMBOL(skb_kill_datagram);
311
312/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313 * skb_copy_datagram_iovec - Copy a datagram to an iovec.
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700314 * @skb: buffer to copy
315 * @offset: offset in the buffer to start copying from
Martin Waitz67be2dd2005-05-01 08:59:26 -0700316 * @to: io vector to copy to
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700317 * @len: amount of data to copy from buffer to iovec
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318 *
319 * Note: the iovec is modified during the copy.
320 */
321int skb_copy_datagram_iovec(const struct sk_buff *skb, int offset,
322 struct iovec *to, int len)
323{
David S. Miller1a028e52007-04-27 15:21:23 -0700324 int start = skb_headlen(skb);
325 int i, copy = start - offset;
David S. Miller5b1a0022009-06-09 00:18:15 -0700326 struct sk_buff *frag_iter;
Herbert Xuc75d7212005-11-02 18:55:00 +1100327
Neil Hormane9b3cc12009-08-13 05:19:44 +0000328 trace_skb_copy_datagram_iovec(skb, len);
329
David S. Millerb4d9eda2006-02-13 16:06:10 -0800330 /* Copy header. */
331 if (copy > 0) {
332 if (copy > len)
333 copy = len;
334 if (memcpy_toiovec(to, skb->data + offset, copy))
335 goto fault;
336 if ((len -= copy) == 0)
337 return 0;
338 offset += copy;
339 }
Herbert Xuc75d7212005-11-02 18:55:00 +1100340
David S. Millerb4d9eda2006-02-13 16:06:10 -0800341 /* Copy paged appendix. Hmm... why does this look so complicated? */
342 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -0700343 int end;
Eric Dumazet9e903e02011-10-18 21:00:24 +0000344 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700346 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -0700347
Eric Dumazet9e903e02011-10-18 21:00:24 +0000348 end = start + skb_frag_size(frag);
David S. Millerb4d9eda2006-02-13 16:06:10 -0800349 if ((copy = end - offset) > 0) {
350 int err;
351 u8 *vaddr;
Ian Campbellea2ab692011-08-22 23:44:58 +0000352 struct page *page = skb_frag_page(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353
354 if (copy > len)
355 copy = len;
David S. Millerb4d9eda2006-02-13 16:06:10 -0800356 vaddr = kmap(page);
David S. Miller1a028e52007-04-27 15:21:23 -0700357 err = memcpy_toiovec(to, vaddr + frag->page_offset +
358 offset - start, copy);
David S. Millerb4d9eda2006-02-13 16:06:10 -0800359 kunmap(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 if (err)
361 goto fault;
362 if (!(len -= copy))
363 return 0;
364 offset += copy;
365 }
David S. Miller1a028e52007-04-27 15:21:23 -0700366 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
David S. Millerb4d9eda2006-02-13 16:06:10 -0800368
David S. Miller5b1a0022009-06-09 00:18:15 -0700369 skb_walk_frags(skb, frag_iter) {
370 int end;
David S. Millerb4d9eda2006-02-13 16:06:10 -0800371
David S. Miller5b1a0022009-06-09 00:18:15 -0700372 WARN_ON(start > offset + len);
David S. Millerb4d9eda2006-02-13 16:06:10 -0800373
David S. Miller5b1a0022009-06-09 00:18:15 -0700374 end = start + frag_iter->len;
375 if ((copy = end - offset) > 0) {
376 if (copy > len)
377 copy = len;
378 if (skb_copy_datagram_iovec(frag_iter,
379 offset - start,
380 to, copy))
381 goto fault;
382 if ((len -= copy) == 0)
383 return 0;
384 offset += copy;
David S. Millerb4d9eda2006-02-13 16:06:10 -0800385 }
David S. Miller5b1a0022009-06-09 00:18:15 -0700386 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700387 }
David S. Millerb4d9eda2006-02-13 16:06:10 -0800388 if (!len)
389 return 0;
390
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391fault:
392 return -EFAULT;
393}
Eric Dumazet9e34a5b2010-07-09 21:22:04 +0000394EXPORT_SYMBOL(skb_copy_datagram_iovec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700395
Rusty Russelldb543c12008-08-15 15:13:53 -0700396/**
Herbert Xua8f820aa2014-11-07 21:22:22 +0800397 * skb_copy_datagram_iter - Copy a datagram to an iovec iterator.
398 * @skb: buffer to copy
399 * @offset: offset in the buffer to start copying from
400 * @to: iovec iterator to copy to
401 * @len: amount of data to copy from buffer to iovec
402 */
403int skb_copy_datagram_iter(const struct sk_buff *skb, int offset,
404 struct iov_iter *to, int len)
405{
406 int start = skb_headlen(skb);
407 int i, copy = start - offset;
408 struct sk_buff *frag_iter;
409
410 trace_skb_copy_datagram_iovec(skb, len);
411
412 /* Copy header. */
413 if (copy > 0) {
414 if (copy > len)
415 copy = len;
416 if (copy_to_iter(skb->data + offset, copy, to) != copy)
417 goto short_copy;
418 if ((len -= copy) == 0)
419 return 0;
420 offset += copy;
421 }
422
423 /* Copy paged appendix. Hmm... why does this look so complicated? */
424 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
425 int end;
426 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
427
428 WARN_ON(start > offset + len);
429
430 end = start + skb_frag_size(frag);
431 if ((copy = end - offset) > 0) {
432 if (copy > len)
433 copy = len;
434 if (copy_page_to_iter(skb_frag_page(frag),
435 frag->page_offset + offset -
436 start, copy, to) != copy)
437 goto short_copy;
438 if (!(len -= copy))
439 return 0;
440 offset += copy;
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
471fault:
472 return -EFAULT;
473
474short_copy:
475 if (iov_iter_count(to))
476 goto fault;
477
478 return 0;
479}
480EXPORT_SYMBOL(skb_copy_datagram_iter);
481
482/**
Al Viro8feb2fb2014-11-06 01:10:59 -0500483 * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter.
Rusty Russelldb543c12008-08-15 15:13:53 -0700484 * @skb: buffer to copy
485 * @offset: offset in the buffer to start copying to
Al Viro8feb2fb2014-11-06 01:10:59 -0500486 * @from: the copy source
Rusty Russelldb543c12008-08-15 15:13:53 -0700487 * @len: amount of data to copy to buffer from iovec
488 *
489 * Returns 0 or -EFAULT.
Rusty Russelldb543c12008-08-15 15:13:53 -0700490 */
Al Viro3a654f92014-06-19 14:15:22 -0400491int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
492 struct iov_iter *from,
493 int len)
494{
495 int start = skb_headlen(skb);
496 int i, copy = start - offset;
497 struct sk_buff *frag_iter;
498
499 /* Copy header. */
500 if (copy > 0) {
501 if (copy > len)
502 copy = len;
503 if (copy_from_iter(skb->data + offset, copy, from) != copy)
504 goto fault;
505 if ((len -= copy) == 0)
506 return 0;
507 offset += copy;
508 }
509
510 /* Copy paged appendix. Hmm... why does this look so complicated? */
511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
512 int end;
513 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
514
515 WARN_ON(start > offset + len);
516
517 end = start + skb_frag_size(frag);
518 if ((copy = end - offset) > 0) {
519 size_t copied;
520
521 if (copy > len)
522 copy = len;
523 copied = copy_page_from_iter(skb_frag_page(frag),
524 frag->page_offset + offset - start,
525 copy, from);
526 if (copied != copy)
527 goto fault;
528
529 if (!(len -= copy))
530 return 0;
531 offset += copy;
532 }
533 start = end;
534 }
535
536 skb_walk_frags(skb, frag_iter) {
537 int end;
538
539 WARN_ON(start > offset + len);
540
541 end = start + frag_iter->len;
542 if ((copy = end - offset) > 0) {
543 if (copy > len)
544 copy = len;
545 if (skb_copy_datagram_from_iter(frag_iter,
546 offset - start,
547 from, copy))
548 goto fault;
549 if ((len -= copy) == 0)
550 return 0;
551 offset += copy;
552 }
553 start = end;
554 }
555 if (!len)
556 return 0;
557
558fault:
559 return -EFAULT;
560}
561EXPORT_SYMBOL(skb_copy_datagram_from_iter);
562
Jason Wangc3bdeb52013-08-06 17:45:04 +0800563/**
Al Viro195e9522014-11-06 00:56:48 -0500564 * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter
Jason Wangc3bdeb52013-08-06 17:45:04 +0800565 * @skb: buffer to copy
Al Viro195e9522014-11-06 00:56:48 -0500566 * @from: the source to copy from
Jason Wangc3bdeb52013-08-06 17:45:04 +0800567 *
568 * The function will first copy up to headlen, and then pin the userspace
569 * pages and build frags through them.
570 *
571 * Returns 0, -EFAULT or -EMSGSIZE.
Jason Wangc3bdeb52013-08-06 17:45:04 +0800572 */
Al Viro3a654f92014-06-19 14:15:22 -0400573int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from)
574{
575 int len = iov_iter_count(from);
576 int copy = min_t(int, skb_headlen(skb), len);
577 int frag = 0;
578
579 /* copy up to skb headlen */
580 if (skb_copy_datagram_from_iter(skb, 0, from, copy))
581 return -EFAULT;
582
583 while (iov_iter_count(from)) {
584 struct page *pages[MAX_SKB_FRAGS];
585 size_t start;
586 ssize_t copied;
587 unsigned long truesize;
588 int n = 0;
589
590 if (frag == MAX_SKB_FRAGS)
591 return -EMSGSIZE;
592
593 copied = iov_iter_get_pages(from, pages, ~0U,
594 MAX_SKB_FRAGS - frag, &start);
595 if (copied < 0)
596 return -EFAULT;
597
598 iov_iter_advance(from, copied);
599
600 truesize = PAGE_ALIGN(copied + start);
601 skb->data_len += copied;
602 skb->len += copied;
603 skb->truesize += truesize;
604 atomic_add(truesize, &skb->sk->sk_wmem_alloc);
605 while (copied) {
606 int size = min_t(int, copied, PAGE_SIZE - start);
607 skb_fill_page_desc(skb, frag++, pages[n], start, size);
608 start = 0;
609 copied -= size;
610 n++;
611 }
612 }
613 return 0;
614}
615EXPORT_SYMBOL(zerocopy_sg_from_iter);
616
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset,
618 u8 __user *to, int len,
Al Viro50842052006-11-14 21:36:34 -0800619 __wsum *csump)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620{
David S. Miller1a028e52007-04-27 15:21:23 -0700621 int start = skb_headlen(skb);
David S. Miller1a028e52007-04-27 15:21:23 -0700622 int i, copy = start - offset;
David S. Miller5b1a0022009-06-09 00:18:15 -0700623 struct sk_buff *frag_iter;
624 int pos = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625
626 /* Copy header. */
627 if (copy > 0) {
628 int err = 0;
629 if (copy > len)
630 copy = len;
631 *csump = csum_and_copy_to_user(skb->data + offset, to, copy,
632 *csump, &err);
633 if (err)
634 goto fault;
635 if ((len -= copy) == 0)
636 return 0;
637 offset += copy;
638 to += copy;
639 pos = copy;
640 }
641
642 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
David S. Miller1a028e52007-04-27 15:21:23 -0700643 int end;
Eric Dumazet9e903e02011-10-18 21:00:24 +0000644 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645
Ilpo Järvinen547b7922008-07-25 21:43:18 -0700646 WARN_ON(start > offset + len);
David S. Miller1a028e52007-04-27 15:21:23 -0700647
Eric Dumazet9e903e02011-10-18 21:00:24 +0000648 end = start + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 if ((copy = end - offset) > 0) {
Al Viro50842052006-11-14 21:36:34 -0800650 __wsum csum2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651 int err = 0;
652 u8 *vaddr;
Ian Campbellea2ab692011-08-22 23:44:58 +0000653 struct page *page = skb_frag_page(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
655 if (copy > len)
656 copy = len;
657 vaddr = kmap(page);
658 csum2 = csum_and_copy_to_user(vaddr +
David S. Miller1a028e52007-04-27 15:21:23 -0700659 frag->page_offset +
660 offset - start,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 to, copy, 0, &err);
662 kunmap(page);
663 if (err)
664 goto fault;
665 *csump = csum_block_add(*csump, csum2, pos);
666 if (!(len -= copy))
667 return 0;
668 offset += copy;
669 to += copy;
670 pos += copy;
671 }
David S. Miller1a028e52007-04-27 15:21:23 -0700672 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673 }
674
David S. Miller5b1a0022009-06-09 00:18:15 -0700675 skb_walk_frags(skb, frag_iter) {
676 int end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
David S. Miller5b1a0022009-06-09 00:18:15 -0700678 WARN_ON(start > offset + len);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
David S. Miller5b1a0022009-06-09 00:18:15 -0700680 end = start + frag_iter->len;
681 if ((copy = end - offset) > 0) {
682 __wsum csum2 = 0;
683 if (copy > len)
684 copy = len;
685 if (skb_copy_and_csum_datagram(frag_iter,
686 offset - start,
687 to, copy,
688 &csum2))
689 goto fault;
690 *csump = csum_block_add(*csump, csum2, pos);
691 if ((len -= copy) == 0)
692 return 0;
693 offset += copy;
694 to += copy;
695 pos += copy;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696 }
David S. Miller5b1a0022009-06-09 00:18:15 -0700697 start = end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 }
699 if (!len)
700 return 0;
701
702fault:
703 return -EFAULT;
704}
705
Herbert Xu759e5d02007-03-25 20:10:56 -0700706__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
Herbert Xufb286bb2005-11-10 13:01:24 -0800707{
Al Virod3bc23e2006-11-14 21:24:49 -0800708 __sum16 sum;
Herbert Xufb286bb2005-11-10 13:01:24 -0800709
Herbert Xu759e5d02007-03-25 20:10:56 -0700710 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
Tom Herbert46fb51e2014-06-14 23:24:03 -0700711 if (likely(!sum)) {
712 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
713 !skb->csum_complete_sw)
714 netdev_rx_csum_fault(skb->dev);
715 }
716 skb->csum_valid = !sum;
Herbert Xufb286bb2005-11-10 13:01:24 -0800717 return sum;
718}
Herbert Xu759e5d02007-03-25 20:10:56 -0700719EXPORT_SYMBOL(__skb_checksum_complete_head);
720
721__sum16 __skb_checksum_complete(struct sk_buff *skb)
722{
Tom Herbert46fb51e2014-06-14 23:24:03 -0700723 __wsum csum;
724 __sum16 sum;
725
726 csum = skb_checksum(skb, 0, skb->len, 0);
727
728 /* skb->csum holds pseudo checksum */
729 sum = csum_fold(csum_add(skb->csum, csum));
730 if (likely(!sum)) {
731 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
732 !skb->csum_complete_sw)
733 netdev_rx_csum_fault(skb->dev);
734 }
735
736 /* Save full packet checksum */
737 skb->csum = csum;
738 skb->ip_summed = CHECKSUM_COMPLETE;
739 skb->csum_complete_sw = 1;
740 skb->csum_valid = !sum;
741
742 return sum;
Herbert Xu759e5d02007-03-25 20:10:56 -0700743}
Herbert Xufb286bb2005-11-10 13:01:24 -0800744EXPORT_SYMBOL(__skb_checksum_complete);
745
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746/**
Masanari Iidae793c0f2014-09-04 23:44:36 +0900747 * skb_copy_and_csum_datagram_iovec - Copy and checksum skb to user iovec.
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700748 * @skb: skbuff
749 * @hlen: hardware length
Martin Waitz67be2dd2005-05-01 08:59:26 -0700750 * @iov: io vector
YOSHIFUJI Hideaki4ec93ed2007-02-09 23:24:36 +0900751 *
Linus Torvalds1da177e2005-04-16 15:20:36 -0700752 * Caller _must_ check that skb will fit to this iovec.
753 *
754 * Returns: 0 - success.
755 * -EINVAL - checksum failure.
756 * -EFAULT - fault during copy. Beware, in this case iovec
757 * can be modified!
758 */
Herbert Xufb286bb2005-11-10 13:01:24 -0800759int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 int hlen, struct iovec *iov)
761{
Al Virod3bc23e2006-11-14 21:24:49 -0800762 __wsum csum;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 int chunk = skb->len - hlen;
764
Herbert Xuef8aef52007-09-06 14:06:35 +0100765 if (!chunk)
766 return 0;
767
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 /* Skip filled elements.
769 * Pretty silly, look at memcpy_toiovec, though 8)
770 */
771 while (!iov->iov_len)
772 iov++;
773
774 if (iov->iov_len < chunk) {
Herbert Xufb286bb2005-11-10 13:01:24 -0800775 if (__skb_checksum_complete(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 goto csum_error;
777 if (skb_copy_datagram_iovec(skb, hlen, iov, chunk))
778 goto fault;
779 } else {
780 csum = csum_partial(skb->data, hlen, skb->csum);
781 if (skb_copy_and_csum_datagram(skb, hlen, iov->iov_base,
782 chunk, &csum))
783 goto fault;
Al Virod3bc23e2006-11-14 21:24:49 -0800784 if (csum_fold(csum))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 goto csum_error;
Patrick McHardy84fa7932006-08-29 16:44:56 -0700786 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE))
Herbert Xufb286bb2005-11-10 13:01:24 -0800787 netdev_rx_csum_fault(skb->dev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 iov->iov_len -= chunk;
789 iov->iov_base += chunk;
790 }
791 return 0;
792csum_error:
793 return -EINVAL;
794fault:
795 return -EFAULT;
796}
Eric Dumazet9e34a5b2010-07-09 21:22:04 +0000797EXPORT_SYMBOL(skb_copy_and_csum_datagram_iovec);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
799/**
800 * datagram_poll - generic datagram poll
Pavel Pisa4dc3b162005-05-01 08:59:25 -0700801 * @file: file struct
802 * @sock: socket
803 * @wait: poll table
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 *
805 * Datagram poll: Again totally generic. This also handles
806 * sequenced packet sockets providing the socket receive queue
807 * is only ever holding data ready to receive.
808 *
809 * Note: when you _don't_ use this routine for this protocol,
810 * and you use a different write policy from sock_writeable()
811 * then please supply your own write_space callback.
812 */
813unsigned int datagram_poll(struct file *file, struct socket *sock,
814 poll_table *wait)
815{
816 struct sock *sk = sock->sk;
817 unsigned int mask;
818
Eric Dumazetaa395142010-04-20 13:03:51 +0000819 sock_poll_wait(file, sk_sleep(sk), wait);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700820 mask = 0;
821
822 /* exceptional events? */
823 if (sk->sk_err || !skb_queue_empty(&sk->sk_error_queue))
Keller, Jacob E7d4c04f2013-03-28 11:19:25 +0000824 mask |= POLLERR |
Jacob Keller8facd5f2013-04-02 13:55:40 -0700825 (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? POLLPRI : 0);
Keller, Jacob E7d4c04f2013-03-28 11:19:25 +0000826
Davide Libenzif348d702006-03-25 03:07:39 -0800827 if (sk->sk_shutdown & RCV_SHUTDOWN)
Eric Dumazetdb409802010-09-06 11:13:50 +0000828 mask |= POLLRDHUP | POLLIN | POLLRDNORM;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700829 if (sk->sk_shutdown == SHUTDOWN_MASK)
830 mask |= POLLHUP;
831
832 /* readable? */
Eric Dumazetdb409802010-09-06 11:13:50 +0000833 if (!skb_queue_empty(&sk->sk_receive_queue))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 mask |= POLLIN | POLLRDNORM;
835
836 /* Connection-based need to check for termination and startup */
837 if (connection_based(sk)) {
838 if (sk->sk_state == TCP_CLOSE)
839 mask |= POLLHUP;
840 /* connection hasn't started yet? */
841 if (sk->sk_state == TCP_SYN_SENT)
842 return mask;
843 }
844
845 /* writable? */
846 if (sock_writeable(sk))
847 mask |= POLLOUT | POLLWRNORM | POLLWRBAND;
848 else
849 set_bit(SOCK_ASYNC_NOSPACE, &sk->sk_socket->flags);
850
851 return mask;
852}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853EXPORT_SYMBOL(datagram_poll);