Commit | Line | Data |
---|---|---|
b2441318 | 1 | // SPDX-License-Identifier: GPL-2.0 |
1da177e4 LT |
2 | /* |
3 | * SUCS NET3: | |
4 | * | |
5 | * Generic datagram handling routines. These are generic for all | |
6 | * protocols. Possibly a generic IP version on top of these would | |
7 | * make sense. Not tonight however 8-). | |
8 | * This is used because UDP, RAW, PACKET, DDP, IPX, AX.25 and | |
9 | * NetROM layer all have identical poll code and mostly | |
10 | * identical recvmsg() code. So we share it here. The poll was | |
11 | * shared before but buried in udp.c so I moved it. | |
12 | * | |
113aa838 | 13 | * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>. (datagram_poll() from old |
1da177e4 LT |
14 | * udp.c code) |
15 | * | |
16 | * Fixes: | |
17 | * Alan Cox : NULL return from skb_peek_copy() | |
18 | * understood | |
19 | * Alan Cox : Rewrote skb_read_datagram to avoid the | |
20 | * skb_peek_copy stuff. | |
21 | * Alan Cox : Added support for SOCK_SEQPACKET. | |
22 | * IPX can no longer use the SO_TYPE hack | |
23 | * but AX.25 now works right, and SPX is | |
24 | * feasible. | |
25 | * Alan Cox : Fixed write poll of non IP protocol | |
26 | * crash. | |
27 | * Florian La Roche: Changed for my new skbuff handling. | |
28 | * Darryl Miles : Fixed non-blocking SOCK_SEQPACKET. | |
29 | * Linus Torvalds : BSD semantic fixes. | |
30 | * Alan Cox : Datagram iovec handling | |
31 | * Darryl Miles : Fixed non-blocking SOCK_STREAM. | |
32 | * Alan Cox : POSIXisms | |
33 | * Pete Wyckoff : Unconnected accept() fix. | |
34 | * | |
35 | */ | |
36 | ||
37 | #include <linux/module.h> | |
38 | #include <linux/types.h> | |
39 | #include <linux/kernel.h> | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 LT |
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> | |
1da177e4 LT |
46 | #include <linux/netdevice.h> |
47 | #include <linux/rtnetlink.h> | |
48 | #include <linux/poll.h> | |
49 | #include <linux/highmem.h> | |
3305b80c | 50 | #include <linux/spinlock.h> |
5a0e3ad6 | 51 | #include <linux/slab.h> |
0433547a | 52 | #include <linux/pagemap.h> |
a8f820aa | 53 | #include <linux/uio.h> |
29f3490b | 54 | #include <linux/indirect_call_wrapper.h> |
1da177e4 LT |
55 | |
56 | #include <net/protocol.h> | |
57 | #include <linux/skbuff.h> | |
1da177e4 | 58 | |
c752f073 ACM |
59 | #include <net/checksum.h> |
60 | #include <net/sock.h> | |
61 | #include <net/tcp_states.h> | |
e9b3cc1b | 62 | #include <trace/events/skb.h> |
076bb0c8 | 63 | #include <net/busy_poll.h> |
1da177e4 | 64 | |
7b7ed885 BVA |
65 | #include "datagram.h" |
66 | ||
1da177e4 LT |
67 | /* |
68 | * Is a socket 'connection oriented' ? | |
69 | */ | |
70 | static inline int connection_based(struct sock *sk) | |
71 | { | |
72 | return sk->sk_type == SOCK_SEQPACKET || sk->sk_type == SOCK_STREAM; | |
73 | } | |
74 | ||
ac6424b9 | 75 | static int receiver_wake_function(wait_queue_entry_t *wait, unsigned int mode, int sync, |
bf368e4e ED |
76 | void *key) |
77 | { | |
bf368e4e ED |
78 | /* |
79 | * Avoid a wakeup if event not interesting for us | |
80 | */ | |
a9a08845 | 81 | if (key && !(key_to_poll(key) & (EPOLLIN | EPOLLERR))) |
bf368e4e ED |
82 | return 0; |
83 | return autoremove_wake_function(wait, mode, sync, key); | |
84 | } | |
1da177e4 | 85 | /* |
39cc8613 | 86 | * Wait for the last received packet to be different from skb |
1da177e4 | 87 | */ |
b50b0580 SD |
88 | int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, |
89 | int *err, long *timeo_p, | |
ea3793ee | 90 | const struct sk_buff *skb) |
1da177e4 LT |
91 | { |
92 | int error; | |
bf368e4e | 93 | DEFINE_WAIT_FUNC(wait, receiver_wake_function); |
1da177e4 | 94 | |
aa395145 | 95 | prepare_to_wait_exclusive(sk_sleep(sk), &wait, TASK_INTERRUPTIBLE); |
1da177e4 LT |
96 | |
97 | /* Socket errors? */ | |
98 | error = sock_error(sk); | |
99 | if (error) | |
100 | goto out_err; | |
101 | ||
b50b0580 | 102 | if (READ_ONCE(queue->prev) != skb) |
1da177e4 LT |
103 | goto out; |
104 | ||
105 | /* Socket shut down? */ | |
106 | if (sk->sk_shutdown & RCV_SHUTDOWN) | |
107 | goto out_noerr; | |
108 | ||
109 | /* Sequenced packets can come disconnected. | |
110 | * If so we report the problem | |
111 | */ | |
112 | error = -ENOTCONN; | |
113 | if (connection_based(sk) && | |
114 | !(sk->sk_state == TCP_ESTABLISHED || sk->sk_state == TCP_LISTEN)) | |
115 | goto out_err; | |
116 | ||
117 | /* handle signals */ | |
118 | if (signal_pending(current)) | |
119 | goto interrupted; | |
120 | ||
121 | error = 0; | |
122 | *timeo_p = schedule_timeout(*timeo_p); | |
123 | out: | |
aa395145 | 124 | finish_wait(sk_sleep(sk), &wait); |
1da177e4 LT |
125 | return error; |
126 | interrupted: | |
127 | error = sock_intr_errno(*timeo_p); | |
128 | out_err: | |
129 | *err = error; | |
130 | goto out; | |
131 | out_noerr: | |
132 | *err = 0; | |
133 | error = 1; | |
134 | goto out; | |
135 | } | |
ea3793ee | 136 | EXPORT_SYMBOL(__skb_wait_for_more_packets); |
1da177e4 | 137 | |
a0a2a660 | 138 | static struct sk_buff *skb_set_peeked(struct sk_buff *skb) |
738ac1eb HX |
139 | { |
140 | struct sk_buff *nskb; | |
141 | ||
142 | if (skb->peeked) | |
a0a2a660 | 143 | return skb; |
738ac1eb HX |
144 | |
145 | /* We have to unshare an skb before modifying it. */ | |
146 | if (!skb_shared(skb)) | |
147 | goto done; | |
148 | ||
149 | nskb = skb_clone(skb, GFP_ATOMIC); | |
150 | if (!nskb) | |
a0a2a660 | 151 | return ERR_PTR(-ENOMEM); |
738ac1eb HX |
152 | |
153 | skb->prev->next = nskb; | |
154 | skb->next->prev = nskb; | |
155 | nskb->prev = skb->prev; | |
156 | nskb->next = skb->next; | |
157 | ||
158 | consume_skb(skb); | |
159 | skb = nskb; | |
160 | ||
161 | done: | |
162 | skb->peeked = 1; | |
163 | ||
a0a2a660 | 164 | return skb; |
738ac1eb HX |
165 | } |
166 | ||
65101aec PA |
167 | struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, |
168 | struct sk_buff_head *queue, | |
169 | unsigned int flags, | |
fd69c399 | 170 | int *off, int *err, |
65101aec PA |
171 | struct sk_buff **last) |
172 | { | |
a0917e0b | 173 | bool peek_at_off = false; |
65101aec | 174 | struct sk_buff *skb; |
a0917e0b MD |
175 | int _off = 0; |
176 | ||
177 | if (unlikely(flags & MSG_PEEK && *off >= 0)) { | |
178 | peek_at_off = true; | |
179 | _off = *off; | |
180 | } | |
65101aec PA |
181 | |
182 | *last = queue->prev; | |
183 | skb_queue_walk(queue, skb) { | |
184 | if (flags & MSG_PEEK) { | |
a0917e0b MD |
185 | if (peek_at_off && _off >= skb->len && |
186 | (_off || skb->peeked)) { | |
de321ed3 | 187 | _off -= skb->len; |
65101aec PA |
188 | continue; |
189 | } | |
190 | if (!skb->len) { | |
191 | skb = skb_set_peeked(skb); | |
98e4fcff | 192 | if (IS_ERR(skb)) { |
65101aec | 193 | *err = PTR_ERR(skb); |
de321ed3 | 194 | return NULL; |
65101aec PA |
195 | } |
196 | } | |
63354797 | 197 | refcount_inc(&skb->users); |
65101aec PA |
198 | } else { |
199 | __skb_unlink(skb, queue); | |
65101aec | 200 | } |
de321ed3 | 201 | *off = _off; |
65101aec PA |
202 | return skb; |
203 | } | |
204 | return NULL; | |
205 | } | |
206 | ||
1da177e4 | 207 | /** |
ea3793ee | 208 | * __skb_try_recv_datagram - Receive a datagram skbuff |
4dc3b16b | 209 | * @sk: socket |
b50b0580 | 210 | * @queue: socket queue from which to receive |
d3f6cd9e | 211 | * @flags: MSG\_ flags |
3f518bf7 PE |
212 | * @off: an offset in bytes to peek skb from. Returns an offset |
213 | * within an skb where data actually starts | |
4dc3b16b | 214 | * @err: error code returned |
ea3793ee RW |
215 | * @last: set to last peeked message to inform the wait function |
216 | * what to look for when peeking | |
1da177e4 LT |
217 | * |
218 | * Get a datagram skbuff, understands the peeking, nonblocking wakeups | |
219 | * and possible races. This replaces identical code in packet, raw and | |
220 | * udp, as well as the IPX AX.25 and Appletalk. It also finally fixes | |
221 | * the long standing peek and read race for datagram sockets. If you | |
222 | * alter this routine remember it must be re-entrant. | |
223 | * | |
ea3793ee RW |
224 | * This function will lock the socket if a skb is returned, so |
225 | * the caller needs to unlock the socket in that case (usually by | |
d651983d | 226 | * calling skb_free_datagram). Returns NULL with @err set to |
ea3793ee RW |
227 | * -EAGAIN if no data was available or to some other value if an |
228 | * error was detected. | |
1da177e4 LT |
229 | * |
230 | * * It does not lock socket since today. This function is | |
231 | * * free of race conditions. This measure should/can improve | |
232 | * * significantly datagram socket latencies at high loads, | |
233 | * * when data copying to user space takes lots of time. | |
234 | * * (BTW I've just killed the last cli() in IP/IPv6/core/netlink/packet | |
235 | * * 8) Great win.) | |
236 | * * --ANK (980729) | |
237 | * | |
238 | * The order of the tests when we find no data waiting are specified | |
239 | * quite explicitly by POSIX 1003.1g, don't change them without having | |
240 | * the standard around please. | |
241 | */ | |
b50b0580 SD |
242 | struct sk_buff *__skb_try_recv_datagram(struct sock *sk, |
243 | struct sk_buff_head *queue, | |
e427cad6 | 244 | unsigned int flags, int *off, int *err, |
ea3793ee | 245 | struct sk_buff **last) |
1da177e4 | 246 | { |
ea3793ee | 247 | struct sk_buff *skb; |
738ac1eb | 248 | unsigned long cpu_flags; |
1da177e4 LT |
249 | /* |
250 | * Caller is allowed not to check sk->sk_err before skb_recv_datagram() | |
251 | */ | |
252 | int error = sock_error(sk); | |
253 | ||
254 | if (error) | |
255 | goto no_packet; | |
256 | ||
1da177e4 LT |
257 | do { |
258 | /* Again only user level code calls this function, so nothing | |
259 | * interrupt level will suddenly eat the receive_queue. | |
260 | * | |
261 | * Look at current nfs client by the way... | |
8917a3c0 | 262 | * However, this function was correct in any case. 8) |
1da177e4 | 263 | */ |
4934b032 | 264 | spin_lock_irqsave(&queue->lock, cpu_flags); |
e427cad6 PA |
265 | skb = __skb_try_recv_from_queue(sk, queue, flags, off, &error, |
266 | last); | |
3f518bf7 | 267 | spin_unlock_irqrestore(&queue->lock, cpu_flags); |
de321ed3 AV |
268 | if (error) |
269 | goto no_packet; | |
65101aec PA |
270 | if (skb) |
271 | return skb; | |
2b5cd0df AD |
272 | |
273 | if (!sk_can_busy_loop(sk)) | |
274 | break; | |
275 | ||
276 | sk_busy_loop(sk, flags & MSG_DONTWAIT); | |
b50b0580 | 277 | } while (READ_ONCE(queue->prev) != *last); |
1da177e4 | 278 | |
ea3793ee | 279 | error = -EAGAIN; |
a5b50476 | 280 | |
ea3793ee RW |
281 | no_packet: |
282 | *err = error; | |
283 | return NULL; | |
284 | } | |
285 | EXPORT_SYMBOL(__skb_try_recv_datagram); | |
1da177e4 | 286 | |
b50b0580 SD |
287 | struct sk_buff *__skb_recv_datagram(struct sock *sk, |
288 | struct sk_buff_head *sk_queue, | |
e427cad6 | 289 | unsigned int flags, int *off, int *err) |
ea3793ee RW |
290 | { |
291 | struct sk_buff *skb, *last; | |
292 | long timeo; | |
1da177e4 | 293 | |
ea3793ee RW |
294 | timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT); |
295 | ||
296 | do { | |
e427cad6 PA |
297 | skb = __skb_try_recv_datagram(sk, sk_queue, flags, off, err, |
298 | &last); | |
ea3793ee RW |
299 | if (skb) |
300 | return skb; | |
301 | ||
760a4322 | 302 | if (*err != -EAGAIN) |
ea3793ee RW |
303 | break; |
304 | } while (timeo && | |
b50b0580 SD |
305 | !__skb_wait_for_more_packets(sk, sk_queue, err, |
306 | &timeo, last)); | |
1da177e4 | 307 | |
1da177e4 LT |
308 | return NULL; |
309 | } | |
a59322be HX |
310 | EXPORT_SYMBOL(__skb_recv_datagram); |
311 | ||
95c96174 | 312 | struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, |
a59322be HX |
313 | int noblock, int *err) |
314 | { | |
fd69c399 | 315 | int off = 0; |
a59322be | 316 | |
b50b0580 SD |
317 | return __skb_recv_datagram(sk, &sk->sk_receive_queue, |
318 | flags | (noblock ? MSG_DONTWAIT : 0), | |
e427cad6 | 319 | &off, err); |
a59322be | 320 | } |
9e34a5b5 | 321 | EXPORT_SYMBOL(skb_recv_datagram); |
1da177e4 LT |
322 | |
323 | void skb_free_datagram(struct sock *sk, struct sk_buff *skb) | |
324 | { | |
ead2ceb0 | 325 | consume_skb(skb); |
270acefa | 326 | sk_mem_reclaim_partial(sk); |
1da177e4 | 327 | } |
9d410c79 ED |
328 | EXPORT_SYMBOL(skb_free_datagram); |
329 | ||
627d2d6b | 330 | void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len) |
9d410c79 | 331 | { |
8a74ad60 ED |
332 | bool slow; |
333 | ||
3889a803 | 334 | if (!skb_unref(skb)) { |
627d2d6b | 335 | sk_peek_offset_bwd(sk, len); |
93bb64ea | 336 | return; |
627d2d6b | 337 | } |
93bb64ea | 338 | |
8a74ad60 | 339 | slow = lock_sock_fast(sk); |
627d2d6b | 340 | sk_peek_offset_bwd(sk, len); |
4b0b72f7 ED |
341 | skb_orphan(skb); |
342 | sk_mem_reclaim_partial(sk); | |
8a74ad60 | 343 | unlock_sock_fast(sk, slow); |
4b0b72f7 | 344 | |
93bb64ea ED |
345 | /* skb is now orphaned, can be freed outside of locked section */ |
346 | __kfree_skb(skb); | |
9d410c79 | 347 | } |
627d2d6b | 348 | EXPORT_SYMBOL(__skb_free_datagram_locked); |
1da177e4 | 349 | |
65101aec PA |
350 | int __sk_queue_drop_skb(struct sock *sk, struct sk_buff_head *sk_queue, |
351 | struct sk_buff *skb, unsigned int flags, | |
69629464 ED |
352 | void (*destructor)(struct sock *sk, |
353 | struct sk_buff *skb)) | |
f8c3bf00 PA |
354 | { |
355 | int err = 0; | |
356 | ||
357 | if (flags & MSG_PEEK) { | |
358 | err = -ENOENT; | |
65101aec | 359 | spin_lock_bh(&sk_queue->lock); |
fd6055a8 | 360 | if (skb->next) { |
65101aec | 361 | __skb_unlink(skb, sk_queue); |
63354797 | 362 | refcount_dec(&skb->users); |
69629464 ED |
363 | if (destructor) |
364 | destructor(sk, skb); | |
f8c3bf00 PA |
365 | err = 0; |
366 | } | |
65101aec | 367 | spin_unlock_bh(&sk_queue->lock); |
f8c3bf00 PA |
368 | } |
369 | ||
370 | atomic_inc(&sk->sk_drops); | |
371 | return err; | |
372 | } | |
373 | EXPORT_SYMBOL(__sk_queue_drop_skb); | |
374 | ||
3305b80c HX |
375 | /** |
376 | * skb_kill_datagram - Free a datagram skbuff forcibly | |
377 | * @sk: socket | |
378 | * @skb: datagram skbuff | |
d3f6cd9e | 379 | * @flags: MSG\_ flags |
3305b80c HX |
380 | * |
381 | * This function frees a datagram skbuff that was received by | |
382 | * skb_recv_datagram. The flags argument must match the one | |
383 | * used for skb_recv_datagram. | |
384 | * | |
385 | * If the MSG_PEEK flag is set, and the packet is still on the | |
386 | * receive queue of the socket, it will be taken off the queue | |
387 | * before it is freed. | |
388 | * | |
389 | * This function currently only disables BH when acquiring the | |
390 | * sk_receive_queue lock. Therefore it must not be used in a | |
391 | * context where that lock is acquired in an IRQ context. | |
27ab2568 HX |
392 | * |
393 | * It returns 0 if the packet was removed by us. | |
3305b80c HX |
394 | */ |
395 | ||
27ab2568 | 396 | int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags) |
3305b80c | 397 | { |
65101aec PA |
398 | int err = __sk_queue_drop_skb(sk, &sk->sk_receive_queue, skb, flags, |
399 | NULL); | |
3305b80c | 400 | |
61de71c6 JD |
401 | kfree_skb(skb); |
402 | sk_mem_reclaim_partial(sk); | |
27ab2568 | 403 | return err; |
3305b80c | 404 | } |
3305b80c HX |
405 | EXPORT_SYMBOL(skb_kill_datagram); |
406 | ||
29f3490b ED |
407 | INDIRECT_CALLABLE_DECLARE(static size_t simple_copy_to_iter(const void *addr, |
408 | size_t bytes, | |
409 | void *data __always_unused, | |
410 | struct iov_iter *i)); | |
411 | ||
56dc6d63 Y |
412 | static int __skb_datagram_iter(const struct sk_buff *skb, int offset, |
413 | struct iov_iter *to, int len, bool fault_short, | |
414 | size_t (*cb)(const void *, size_t, void *, | |
415 | struct iov_iter *), void *data) | |
a8f820aa HX |
416 | { |
417 | int start = skb_headlen(skb); | |
32786821 | 418 | int i, copy = start - offset, start_off = offset, n; |
a8f820aa HX |
419 | struct sk_buff *frag_iter; |
420 | ||
a8f820aa HX |
421 | /* Copy header. */ |
422 | if (copy > 0) { | |
423 | if (copy > len) | |
424 | copy = len; | |
29f3490b ED |
425 | n = INDIRECT_CALL_1(cb, simple_copy_to_iter, |
426 | skb->data + offset, copy, data, to); | |
32786821 AV |
427 | offset += n; |
428 | if (n != copy) | |
a8f820aa HX |
429 | goto short_copy; |
430 | if ((len -= copy) == 0) | |
431 | return 0; | |
a8f820aa HX |
432 | } |
433 | ||
434 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
435 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
436 | int end; | |
437 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
438 | ||
439 | WARN_ON(start > offset + len); | |
440 | ||
441 | end = start + skb_frag_size(frag); | |
442 | if ((copy = end - offset) > 0) { | |
0fc07791 SG |
443 | struct page *page = skb_frag_page(frag); |
444 | u8 *vaddr = kmap(page); | |
445 | ||
a8f820aa HX |
446 | if (copy > len) |
447 | copy = len; | |
29f3490b ED |
448 | n = INDIRECT_CALL_1(cb, simple_copy_to_iter, |
449 | vaddr + skb_frag_off(frag) + offset - start, | |
450 | copy, data, to); | |
0fc07791 | 451 | kunmap(page); |
32786821 AV |
452 | offset += n; |
453 | if (n != copy) | |
a8f820aa HX |
454 | goto short_copy; |
455 | if (!(len -= copy)) | |
456 | return 0; | |
a8f820aa HX |
457 | } |
458 | start = end; | |
459 | } | |
460 | ||
461 | skb_walk_frags(skb, frag_iter) { | |
462 | int end; | |
463 | ||
464 | WARN_ON(start > offset + len); | |
465 | ||
466 | end = start + frag_iter->len; | |
467 | if ((copy = end - offset) > 0) { | |
468 | if (copy > len) | |
469 | copy = len; | |
950fcaec | 470 | if (__skb_datagram_iter(frag_iter, offset - start, |
65d69e25 | 471 | to, copy, fault_short, cb, data)) |
a8f820aa HX |
472 | goto fault; |
473 | if ((len -= copy) == 0) | |
474 | return 0; | |
475 | offset += copy; | |
476 | } | |
477 | start = end; | |
478 | } | |
479 | if (!len) | |
480 | return 0; | |
481 | ||
482 | /* This is not really a user copy fault, but rather someone | |
483 | * gave us a bogus length on the skb. We should probably | |
484 | * print a warning here as it may indicate a kernel bug. | |
485 | */ | |
486 | ||
487 | fault: | |
32786821 | 488 | iov_iter_revert(to, offset - start_off); |
a8f820aa HX |
489 | return -EFAULT; |
490 | ||
491 | short_copy: | |
950fcaec | 492 | if (fault_short || iov_iter_count(to)) |
a8f820aa HX |
493 | goto fault; |
494 | ||
495 | return 0; | |
496 | } | |
950fcaec | 497 | |
65d69e25 SG |
498 | /** |
499 | * skb_copy_and_hash_datagram_iter - Copy datagram to an iovec iterator | |
500 | * and update a hash. | |
501 | * @skb: buffer to copy | |
502 | * @offset: offset in the buffer to start copying from | |
503 | * @to: iovec iterator to copy to | |
504 | * @len: amount of data to copy from buffer to iovec | |
505 | * @hash: hash request to update | |
506 | */ | |
507 | int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, | |
508 | struct iov_iter *to, int len, | |
509 | struct ahash_request *hash) | |
510 | { | |
511 | return __skb_datagram_iter(skb, offset, to, len, true, | |
512 | hash_and_copy_to_iter, hash); | |
513 | } | |
514 | EXPORT_SYMBOL(skb_copy_and_hash_datagram_iter); | |
515 | ||
950fcaec SG |
516 | static size_t simple_copy_to_iter(const void *addr, size_t bytes, |
517 | void *data __always_unused, struct iov_iter *i) | |
518 | { | |
519 | return copy_to_iter(addr, bytes, i); | |
520 | } | |
521 | ||
522 | /** | |
523 | * skb_copy_datagram_iter - Copy a datagram to an iovec iterator. | |
524 | * @skb: buffer to copy | |
525 | * @offset: offset in the buffer to start copying from | |
526 | * @to: iovec iterator to copy to | |
527 | * @len: amount of data to copy from buffer to iovec | |
528 | */ | |
529 | int skb_copy_datagram_iter(const struct sk_buff *skb, int offset, | |
530 | struct iov_iter *to, int len) | |
531 | { | |
532 | trace_skb_copy_datagram_iovec(skb, len); | |
533 | return __skb_datagram_iter(skb, offset, to, len, false, | |
534 | simple_copy_to_iter, NULL); | |
535 | } | |
a8f820aa HX |
536 | EXPORT_SYMBOL(skb_copy_datagram_iter); |
537 | ||
db543c1f | 538 | /** |
8feb2fb2 | 539 | * skb_copy_datagram_from_iter - Copy a datagram from an iov_iter. |
db543c1f RR |
540 | * @skb: buffer to copy |
541 | * @offset: offset in the buffer to start copying to | |
8feb2fb2 | 542 | * @from: the copy source |
db543c1f RR |
543 | * @len: amount of data to copy to buffer from iovec |
544 | * | |
545 | * Returns 0 or -EFAULT. | |
db543c1f | 546 | */ |
3a654f97 AV |
547 | int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, |
548 | struct iov_iter *from, | |
549 | int len) | |
550 | { | |
551 | int start = skb_headlen(skb); | |
552 | int i, copy = start - offset; | |
553 | struct sk_buff *frag_iter; | |
554 | ||
555 | /* Copy header. */ | |
556 | if (copy > 0) { | |
557 | if (copy > len) | |
558 | copy = len; | |
559 | if (copy_from_iter(skb->data + offset, copy, from) != copy) | |
560 | goto fault; | |
561 | if ((len -= copy) == 0) | |
562 | return 0; | |
563 | offset += copy; | |
564 | } | |
565 | ||
566 | /* Copy paged appendix. Hmm... why does this look so complicated? */ | |
567 | for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { | |
568 | int end; | |
569 | const skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; | |
570 | ||
571 | WARN_ON(start > offset + len); | |
572 | ||
573 | end = start + skb_frag_size(frag); | |
574 | if ((copy = end - offset) > 0) { | |
575 | size_t copied; | |
576 | ||
577 | if (copy > len) | |
578 | copy = len; | |
579 | copied = copy_page_from_iter(skb_frag_page(frag), | |
b54c9d5b | 580 | skb_frag_off(frag) + offset - start, |
3a654f97 AV |
581 | copy, from); |
582 | if (copied != copy) | |
583 | goto fault; | |
584 | ||
585 | if (!(len -= copy)) | |
586 | return 0; | |
587 | offset += copy; | |
588 | } | |
589 | start = end; | |
590 | } | |
591 | ||
592 | skb_walk_frags(skb, frag_iter) { | |
593 | int end; | |
594 | ||
595 | WARN_ON(start > offset + len); | |
596 | ||
597 | end = start + frag_iter->len; | |
598 | if ((copy = end - offset) > 0) { | |
599 | if (copy > len) | |
600 | copy = len; | |
601 | if (skb_copy_datagram_from_iter(frag_iter, | |
602 | offset - start, | |
603 | from, copy)) | |
604 | goto fault; | |
605 | if ((len -= copy) == 0) | |
606 | return 0; | |
607 | offset += copy; | |
608 | } | |
609 | start = end; | |
610 | } | |
611 | if (!len) | |
612 | return 0; | |
613 | ||
614 | fault: | |
615 | return -EFAULT; | |
616 | } | |
617 | EXPORT_SYMBOL(skb_copy_datagram_from_iter); | |
618 | ||
52267790 WB |
619 | int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, |
620 | struct iov_iter *from, size_t length) | |
3a654f97 | 621 | { |
52267790 | 622 | int frag = skb_shinfo(skb)->nr_frags; |
3a654f97 | 623 | |
52267790 | 624 | while (length && iov_iter_count(from)) { |
3a654f97 | 625 | struct page *pages[MAX_SKB_FRAGS]; |
394fcd8a | 626 | struct page *last_head = NULL; |
3a654f97 AV |
627 | size_t start; |
628 | ssize_t copied; | |
629 | unsigned long truesize; | |
394fcd8a | 630 | int refs, n = 0; |
3a654f97 AV |
631 | |
632 | if (frag == MAX_SKB_FRAGS) | |
633 | return -EMSGSIZE; | |
634 | ||
52267790 | 635 | copied = iov_iter_get_pages(from, pages, length, |
3a654f97 AV |
636 | MAX_SKB_FRAGS - frag, &start); |
637 | if (copied < 0) | |
638 | return -EFAULT; | |
639 | ||
640 | iov_iter_advance(from, copied); | |
52267790 | 641 | length -= copied; |
3a654f97 AV |
642 | |
643 | truesize = PAGE_ALIGN(copied + start); | |
644 | skb->data_len += copied; | |
645 | skb->len += copied; | |
646 | skb->truesize += truesize; | |
52267790 | 647 | if (sk && sk->sk_type == SOCK_STREAM) { |
ab4e846a | 648 | sk_wmem_queued_add(sk, truesize); |
52267790 WB |
649 | sk_mem_charge(sk, truesize); |
650 | } else { | |
651 | refcount_add(truesize, &skb->sk->sk_wmem_alloc); | |
652 | } | |
394fcd8a | 653 | for (refs = 0; copied != 0; start = 0) { |
3a654f97 | 654 | int size = min_t(int, copied, PAGE_SIZE - start); |
394fcd8a ED |
655 | struct page *head = compound_head(pages[n]); |
656 | ||
657 | start += (pages[n] - head) << PAGE_SHIFT; | |
3a654f97 AV |
658 | copied -= size; |
659 | n++; | |
394fcd8a ED |
660 | if (frag) { |
661 | skb_frag_t *last = &skb_shinfo(skb)->frags[frag - 1]; | |
662 | ||
663 | if (head == skb_frag_page(last) && | |
664 | start == skb_frag_off(last) + skb_frag_size(last)) { | |
665 | skb_frag_size_add(last, size); | |
666 | /* We combined this page, we need to release | |
667 | * a reference. Since compound pages refcount | |
668 | * is shared among many pages, batch the refcount | |
669 | * adjustments to limit false sharing. | |
670 | */ | |
671 | last_head = head; | |
672 | refs++; | |
673 | continue; | |
674 | } | |
675 | } | |
676 | if (refs) { | |
677 | page_ref_sub(last_head, refs); | |
678 | refs = 0; | |
679 | } | |
680 | skb_fill_page_desc(skb, frag++, head, start, size); | |
3a654f97 | 681 | } |
394fcd8a ED |
682 | if (refs) |
683 | page_ref_sub(last_head, refs); | |
3a654f97 AV |
684 | } |
685 | return 0; | |
686 | } | |
52267790 WB |
687 | EXPORT_SYMBOL(__zerocopy_sg_from_iter); |
688 | ||
689 | /** | |
690 | * zerocopy_sg_from_iter - Build a zerocopy datagram from an iov_iter | |
691 | * @skb: buffer to copy | |
692 | * @from: the source to copy from | |
693 | * | |
694 | * The function will first copy up to headlen, and then pin the userspace | |
695 | * pages and build frags through them. | |
696 | * | |
697 | * Returns 0, -EFAULT or -EMSGSIZE. | |
698 | */ | |
699 | int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *from) | |
700 | { | |
701 | int copy = min_t(int, skb_headlen(skb), iov_iter_count(from)); | |
702 | ||
703 | /* copy up to skb headlen */ | |
704 | if (skb_copy_datagram_from_iter(skb, 0, from, copy)) | |
705 | return -EFAULT; | |
706 | ||
707 | return __zerocopy_sg_from_iter(NULL, skb, from, ~0U); | |
708 | } | |
3a654f97 AV |
709 | EXPORT_SYMBOL(zerocopy_sg_from_iter); |
710 | ||
950fcaec | 711 | /** |
c1639be9 | 712 | * skb_copy_and_csum_datagram - Copy datagram to an iovec iterator |
950fcaec SG |
713 | * and update a checksum. |
714 | * @skb: buffer to copy | |
715 | * @offset: offset in the buffer to start copying from | |
716 | * @to: iovec iterator to copy to | |
717 | * @len: amount of data to copy from buffer to iovec | |
718 | * @csump: checksum pointer | |
719 | */ | |
1da177e4 | 720 | static int skb_copy_and_csum_datagram(const struct sk_buff *skb, int offset, |
e5a4b0bb | 721 | struct iov_iter *to, int len, |
5084205f | 722 | __wsum *csump) |
1da177e4 | 723 | { |
52cbd23a WB |
724 | struct csum_state csdata = { .csum = *csump }; |
725 | int ret; | |
726 | ||
727 | ret = __skb_datagram_iter(skb, offset, to, len, true, | |
728 | csum_and_copy_to_iter, &csdata); | |
729 | if (ret) | |
730 | return ret; | |
731 | ||
732 | *csump = csdata.csum; | |
733 | return 0; | |
1da177e4 LT |
734 | } |
735 | ||
736 | /** | |
e5a4b0bb | 737 | * skb_copy_and_csum_datagram_msg - Copy and checksum skb to user iovec. |
4dc3b16b PP |
738 | * @skb: skbuff |
739 | * @hlen: hardware length | |
e5a4b0bb | 740 | * @msg: destination |
4ec93edb | 741 | * |
1da177e4 LT |
742 | * Caller _must_ check that skb will fit to this iovec. |
743 | * | |
744 | * Returns: 0 - success. | |
745 | * -EINVAL - checksum failure. | |
e5a4b0bb | 746 | * -EFAULT - fault during copy. |
1da177e4 | 747 | */ |
e5a4b0bb AV |
748 | int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, |
749 | int hlen, struct msghdr *msg) | |
1da177e4 | 750 | { |
d3bc23e7 | 751 | __wsum csum; |
1da177e4 LT |
752 | int chunk = skb->len - hlen; |
753 | ||
ef8aef55 HX |
754 | if (!chunk) |
755 | return 0; | |
756 | ||
01e97e65 | 757 | if (msg_data_left(msg) < chunk) { |
fb286bb2 | 758 | if (__skb_checksum_complete(skb)) |
a6a59932 | 759 | return -EINVAL; |
e5a4b0bb | 760 | if (skb_copy_datagram_msg(skb, hlen, msg, chunk)) |
1da177e4 LT |
761 | goto fault; |
762 | } else { | |
763 | csum = csum_partial(skb->data, hlen, skb->csum); | |
e5a4b0bb | 764 | if (skb_copy_and_csum_datagram(skb, hlen, &msg->msg_iter, |
1da177e4 LT |
765 | chunk, &csum)) |
766 | goto fault; | |
a6a59932 DT |
767 | |
768 | if (csum_fold(csum)) { | |
769 | iov_iter_revert(&msg->msg_iter, chunk); | |
770 | return -EINVAL; | |
771 | } | |
772 | ||
db4f1be3 ST |
773 | if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) && |
774 | !skb->csum_complete_sw) | |
7fe50ac8 | 775 | netdev_rx_csum_fault(NULL, skb); |
1da177e4 LT |
776 | } |
777 | return 0; | |
1da177e4 LT |
778 | fault: |
779 | return -EFAULT; | |
780 | } | |
e5a4b0bb | 781 | EXPORT_SYMBOL(skb_copy_and_csum_datagram_msg); |
1da177e4 LT |
782 | |
783 | /** | |
784 | * datagram_poll - generic datagram poll | |
a11e1d43 | 785 | * @file: file struct |
4dc3b16b | 786 | * @sock: socket |
a11e1d43 | 787 | * @wait: poll table |
1da177e4 LT |
788 | * |
789 | * Datagram poll: Again totally generic. This also handles | |
790 | * sequenced packet sockets providing the socket receive queue | |
791 | * is only ever holding data ready to receive. | |
792 | * | |
d3f6cd9e | 793 | * Note: when you *don't* use this routine for this protocol, |
1da177e4 LT |
794 | * and you use a different write policy from sock_writeable() |
795 | * then please supply your own write_space callback. | |
796 | */ | |
a11e1d43 LT |
797 | __poll_t datagram_poll(struct file *file, struct socket *sock, |
798 | poll_table *wait) | |
1da177e4 LT |
799 | { |
800 | struct sock *sk = sock->sk; | |
a11e1d43 LT |
801 | __poll_t mask; |
802 | ||
89ab066d | 803 | sock_poll_wait(file, sock, wait); |
a11e1d43 | 804 | mask = 0; |
1da177e4 LT |
805 | |
806 | /* exceptional events? */ | |
3ef7cf57 | 807 | if (sk->sk_err || !skb_queue_empty_lockless(&sk->sk_error_queue)) |
a9a08845 LT |
808 | mask |= EPOLLERR | |
809 | (sock_flag(sk, SOCK_SELECT_ERR_QUEUE) ? EPOLLPRI : 0); | |
7d4c04fc | 810 | |
f348d70a | 811 | if (sk->sk_shutdown & RCV_SHUTDOWN) |
a9a08845 | 812 | mask |= EPOLLRDHUP | EPOLLIN | EPOLLRDNORM; |
1da177e4 | 813 | if (sk->sk_shutdown == SHUTDOWN_MASK) |
a9a08845 | 814 | mask |= EPOLLHUP; |
1da177e4 LT |
815 | |
816 | /* readable? */ | |
3ef7cf57 | 817 | if (!skb_queue_empty_lockless(&sk->sk_receive_queue)) |
a9a08845 | 818 | mask |= EPOLLIN | EPOLLRDNORM; |
1da177e4 LT |
819 | |
820 | /* Connection-based need to check for termination and startup */ | |
821 | if (connection_based(sk)) { | |
822 | if (sk->sk_state == TCP_CLOSE) | |
a9a08845 | 823 | mask |= EPOLLHUP; |
1da177e4 LT |
824 | /* connection hasn't started yet? */ |
825 | if (sk->sk_state == TCP_SYN_SENT) | |
826 | return mask; | |
827 | } | |
828 | ||
829 | /* writable? */ | |
830 | if (sock_writeable(sk)) | |
a9a08845 | 831 | mask |= EPOLLOUT | EPOLLWRNORM | EPOLLWRBAND; |
1da177e4 | 832 | else |
9cd3e072 | 833 | sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); |
1da177e4 LT |
834 | |
835 | return mask; | |
836 | } | |
a11e1d43 | 837 | EXPORT_SYMBOL(datagram_poll); |