Loading...
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#include <linux/module.h>
40#include <linux/types.h>
41#include <linux/kernel.h>
42#include <linux/kmemcheck.h>
43#include <linux/mm.h>
44#include <linux/interrupt.h>
45#include <linux/in.h>
46#include <linux/inet.h>
47#include <linux/slab.h>
48#include <linux/netdevice.h>
49#ifdef CONFIG_NET_CLS_ACT
50#include <net/pkt_sched.h>
51#endif
52#include <linux/string.h>
53#include <linux/skbuff.h>
54#include <linux/splice.h>
55#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
58#include <linux/scatterlist.h>
59#include <linux/errqueue.h>
60#include <linux/prefetch.h>
61
62#include <net/protocol.h>
63#include <net/dst.h>
64#include <net/sock.h>
65#include <net/checksum.h>
66#include <net/xfrm.h>
67
68#include <asm/uaccess.h>
69#include <asm/system.h>
70#include <trace/events/skb.h>
71
72#include "kmap_skb.h"
73
74static struct kmem_cache *skbuff_head_cache __read_mostly;
75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
76
77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
79{
80 put_page(buf->page);
81}
82
83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
85{
86 get_page(buf->page);
87}
88
89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
91{
92 return 1;
93}
94
95
96/* Pipe buffer operations for a socket. */
97static const struct pipe_buf_operations sock_pipe_buf_ops = {
98 .can_merge = 0,
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
105};
106
107/*
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
110 * reliable.
111 */
112
113/**
114 * skb_over_panic - private function
115 * @skb: buffer
116 * @sz: size
117 * @here: address
118 *
119 * Out of line support code for skb_put(). Not user callable.
120 */
121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
122{
123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here, skb->len, sz, skb->head, skb->data,
126 (unsigned long)skb->tail, (unsigned long)skb->end,
127 skb->dev ? skb->dev->name : "<NULL>");
128 BUG();
129}
130
131/**
132 * skb_under_panic - private function
133 * @skb: buffer
134 * @sz: size
135 * @here: address
136 *
137 * Out of line support code for skb_push(). Not user callable.
138 */
139
140static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141{
142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here, skb->len, sz, skb->head, skb->data,
145 (unsigned long)skb->tail, (unsigned long)skb->end,
146 skb->dev ? skb->dev->name : "<NULL>");
147 BUG();
148}
149
150/* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
152 * [BEEP] leaks.
153 *
154 */
155
156/**
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
163 *
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
167 *
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 * %GFP_ATOMIC.
170 */
171struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
173{
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
176 struct sk_buff *skb;
177 u8 *data;
178
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
180
181 /* Get the HEAD */
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
183 if (!skb)
184 goto out;
185 prefetchw(skb);
186
187 size = SKB_DATA_ALIGN(size);
188 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
189 gfp_mask, node);
190 if (!data)
191 goto nodata;
192 prefetchw(data + size);
193
194 /*
195 * Only clear those fields we need to clear, not those that we will
196 * actually initialise below. Hence, don't put any more fields after
197 * the tail pointer in struct sk_buff!
198 */
199 memset(skb, 0, offsetof(struct sk_buff, tail));
200 skb->truesize = size + sizeof(struct sk_buff);
201 atomic_set(&skb->users, 1);
202 skb->head = data;
203 skb->data = data;
204 skb_reset_tail_pointer(skb);
205 skb->end = skb->tail + size;
206#ifdef NET_SKBUFF_DATA_USES_OFFSET
207 skb->mac_header = ~0U;
208#endif
209
210 /* make sure we initialize shinfo sequentially */
211 shinfo = skb_shinfo(skb);
212 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
213 atomic_set(&shinfo->dataref, 1);
214 kmemcheck_annotate_variable(shinfo->destructor_arg);
215
216 if (fclone) {
217 struct sk_buff *child = skb + 1;
218 atomic_t *fclone_ref = (atomic_t *) (child + 1);
219
220 kmemcheck_annotate_bitfield(child, flags1);
221 kmemcheck_annotate_bitfield(child, flags2);
222 skb->fclone = SKB_FCLONE_ORIG;
223 atomic_set(fclone_ref, 1);
224
225 child->fclone = SKB_FCLONE_UNAVAILABLE;
226 }
227out:
228 return skb;
229nodata:
230 kmem_cache_free(cache, skb);
231 skb = NULL;
232 goto out;
233}
234EXPORT_SYMBOL(__alloc_skb);
235
236/**
237 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
238 * @dev: network device to receive on
239 * @length: length to allocate
240 * @gfp_mask: get_free_pages mask, passed to alloc_skb
241 *
242 * Allocate a new &sk_buff and assign it a usage count of one. The
243 * buffer has unspecified headroom built in. Users should allocate
244 * the headroom they think they need without accounting for the
245 * built in space. The built in space is used for optimisations.
246 *
247 * %NULL is returned if there is no free memory.
248 */
249struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
250 unsigned int length, gfp_t gfp_mask)
251{
252 struct sk_buff *skb;
253
254 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
255 if (likely(skb)) {
256 skb_reserve(skb, NET_SKB_PAD);
257 skb->dev = dev;
258 }
259 return skb;
260}
261EXPORT_SYMBOL(__netdev_alloc_skb);
262
263void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
264 int size)
265{
266 skb_fill_page_desc(skb, i, page, off, size);
267 skb->len += size;
268 skb->data_len += size;
269 skb->truesize += size;
270}
271EXPORT_SYMBOL(skb_add_rx_frag);
272
273/**
274 * dev_alloc_skb - allocate an skbuff for receiving
275 * @length: length to allocate
276 *
277 * Allocate a new &sk_buff and assign it a usage count of one. The
278 * buffer has unspecified headroom built in. Users should allocate
279 * the headroom they think they need without accounting for the
280 * built in space. The built in space is used for optimisations.
281 *
282 * %NULL is returned if there is no free memory. Although this function
283 * allocates memory it can be called from an interrupt.
284 */
285struct sk_buff *dev_alloc_skb(unsigned int length)
286{
287 /*
288 * There is more code here than it seems:
289 * __dev_alloc_skb is an inline
290 */
291 return __dev_alloc_skb(length, GFP_ATOMIC);
292}
293EXPORT_SYMBOL(dev_alloc_skb);
294
295static void skb_drop_list(struct sk_buff **listp)
296{
297 struct sk_buff *list = *listp;
298
299 *listp = NULL;
300
301 do {
302 struct sk_buff *this = list;
303 list = list->next;
304 kfree_skb(this);
305 } while (list);
306}
307
308static inline void skb_drop_fraglist(struct sk_buff *skb)
309{
310 skb_drop_list(&skb_shinfo(skb)->frag_list);
311}
312
313static void skb_clone_fraglist(struct sk_buff *skb)
314{
315 struct sk_buff *list;
316
317 skb_walk_frags(skb, list)
318 skb_get(list);
319}
320
321static void skb_release_data(struct sk_buff *skb)
322{
323 if (!skb->cloned ||
324 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
325 &skb_shinfo(skb)->dataref)) {
326 if (skb_shinfo(skb)->nr_frags) {
327 int i;
328 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
329 put_page(skb_shinfo(skb)->frags[i].page);
330 }
331
332 /*
333 * If skb buf is from userspace, we need to notify the caller
334 * the lower device DMA has done;
335 */
336 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
337 struct ubuf_info *uarg;
338
339 uarg = skb_shinfo(skb)->destructor_arg;
340 if (uarg->callback)
341 uarg->callback(uarg);
342 }
343
344 if (skb_has_frag_list(skb))
345 skb_drop_fraglist(skb);
346
347 kfree(skb->head);
348 }
349}
350
351/*
352 * Free an skbuff by memory without cleaning the state.
353 */
354static void kfree_skbmem(struct sk_buff *skb)
355{
356 struct sk_buff *other;
357 atomic_t *fclone_ref;
358
359 switch (skb->fclone) {
360 case SKB_FCLONE_UNAVAILABLE:
361 kmem_cache_free(skbuff_head_cache, skb);
362 break;
363
364 case SKB_FCLONE_ORIG:
365 fclone_ref = (atomic_t *) (skb + 2);
366 if (atomic_dec_and_test(fclone_ref))
367 kmem_cache_free(skbuff_fclone_cache, skb);
368 break;
369
370 case SKB_FCLONE_CLONE:
371 fclone_ref = (atomic_t *) (skb + 1);
372 other = skb - 1;
373
374 /* The clone portion is available for
375 * fast-cloning again.
376 */
377 skb->fclone = SKB_FCLONE_UNAVAILABLE;
378
379 if (atomic_dec_and_test(fclone_ref))
380 kmem_cache_free(skbuff_fclone_cache, other);
381 break;
382 }
383}
384
385static void skb_release_head_state(struct sk_buff *skb)
386{
387 skb_dst_drop(skb);
388#ifdef CONFIG_XFRM
389 secpath_put(skb->sp);
390#endif
391 if (skb->destructor) {
392 WARN_ON(in_irq());
393 skb->destructor(skb);
394 }
395#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
396 nf_conntrack_put(skb->nfct);
397#endif
398#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
399 nf_conntrack_put_reasm(skb->nfct_reasm);
400#endif
401#ifdef CONFIG_BRIDGE_NETFILTER
402 nf_bridge_put(skb->nf_bridge);
403#endif
404/* XXX: IS this still necessary? - JHS */
405#ifdef CONFIG_NET_SCHED
406 skb->tc_index = 0;
407#ifdef CONFIG_NET_CLS_ACT
408 skb->tc_verd = 0;
409#endif
410#endif
411}
412
413/* Free everything but the sk_buff shell. */
414static void skb_release_all(struct sk_buff *skb)
415{
416 skb_release_head_state(skb);
417 skb_release_data(skb);
418}
419
420/**
421 * __kfree_skb - private function
422 * @skb: buffer
423 *
424 * Free an sk_buff. Release anything attached to the buffer.
425 * Clean the state. This is an internal helper function. Users should
426 * always call kfree_skb
427 */
428
429void __kfree_skb(struct sk_buff *skb)
430{
431 skb_release_all(skb);
432 kfree_skbmem(skb);
433}
434EXPORT_SYMBOL(__kfree_skb);
435
436/**
437 * kfree_skb - free an sk_buff
438 * @skb: buffer to free
439 *
440 * Drop a reference to the buffer and free it if the usage count has
441 * hit zero.
442 */
443void kfree_skb(struct sk_buff *skb)
444{
445 if (unlikely(!skb))
446 return;
447 if (likely(atomic_read(&skb->users) == 1))
448 smp_rmb();
449 else if (likely(!atomic_dec_and_test(&skb->users)))
450 return;
451 trace_kfree_skb(skb, __builtin_return_address(0));
452 __kfree_skb(skb);
453}
454EXPORT_SYMBOL(kfree_skb);
455
456/**
457 * consume_skb - free an skbuff
458 * @skb: buffer to free
459 *
460 * Drop a ref to the buffer and free it if the usage count has hit zero
461 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
462 * is being dropped after a failure and notes that
463 */
464void consume_skb(struct sk_buff *skb)
465{
466 if (unlikely(!skb))
467 return;
468 if (likely(atomic_read(&skb->users) == 1))
469 smp_rmb();
470 else if (likely(!atomic_dec_and_test(&skb->users)))
471 return;
472 trace_consume_skb(skb);
473 __kfree_skb(skb);
474}
475EXPORT_SYMBOL(consume_skb);
476
477/**
478 * skb_recycle_check - check if skb can be reused for receive
479 * @skb: buffer
480 * @skb_size: minimum receive buffer size
481 *
482 * Checks that the skb passed in is not shared or cloned, and
483 * that it is linear and its head portion at least as large as
484 * skb_size so that it can be recycled as a receive buffer.
485 * If these conditions are met, this function does any necessary
486 * reference count dropping and cleans up the skbuff as if it
487 * just came from __alloc_skb().
488 */
489bool skb_recycle_check(struct sk_buff *skb, int skb_size)
490{
491 struct skb_shared_info *shinfo;
492
493 if (irqs_disabled())
494 return false;
495
496 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
497 return false;
498
499 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
500 return false;
501
502 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
503 if (skb_end_pointer(skb) - skb->head < skb_size)
504 return false;
505
506 if (skb_shared(skb) || skb_cloned(skb))
507 return false;
508
509 skb_release_head_state(skb);
510
511 shinfo = skb_shinfo(skb);
512 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
513 atomic_set(&shinfo->dataref, 1);
514
515 memset(skb, 0, offsetof(struct sk_buff, tail));
516 skb->data = skb->head + NET_SKB_PAD;
517 skb_reset_tail_pointer(skb);
518
519 return true;
520}
521EXPORT_SYMBOL(skb_recycle_check);
522
523static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
524{
525 new->tstamp = old->tstamp;
526 new->dev = old->dev;
527 new->transport_header = old->transport_header;
528 new->network_header = old->network_header;
529 new->mac_header = old->mac_header;
530 skb_dst_copy(new, old);
531 new->rxhash = old->rxhash;
532#ifdef CONFIG_XFRM
533 new->sp = secpath_get(old->sp);
534#endif
535 memcpy(new->cb, old->cb, sizeof(old->cb));
536 new->csum = old->csum;
537 new->local_df = old->local_df;
538 new->pkt_type = old->pkt_type;
539 new->ip_summed = old->ip_summed;
540 skb_copy_queue_mapping(new, old);
541 new->priority = old->priority;
542#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
543 new->ipvs_property = old->ipvs_property;
544#endif
545 new->protocol = old->protocol;
546 new->mark = old->mark;
547 new->skb_iif = old->skb_iif;
548 __nf_copy(new, old);
549#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
550 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
551 new->nf_trace = old->nf_trace;
552#endif
553#ifdef CONFIG_NET_SCHED
554 new->tc_index = old->tc_index;
555#ifdef CONFIG_NET_CLS_ACT
556 new->tc_verd = old->tc_verd;
557#endif
558#endif
559 new->vlan_tci = old->vlan_tci;
560
561 skb_copy_secmark(new, old);
562}
563
564/*
565 * You should not add any new code to this function. Add it to
566 * __copy_skb_header above instead.
567 */
568static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
569{
570#define C(x) n->x = skb->x
571
572 n->next = n->prev = NULL;
573 n->sk = NULL;
574 __copy_skb_header(n, skb);
575
576 C(len);
577 C(data_len);
578 C(mac_len);
579 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
580 n->cloned = 1;
581 n->nohdr = 0;
582 n->destructor = NULL;
583 C(tail);
584 C(end);
585 C(head);
586 C(data);
587 C(truesize);
588 atomic_set(&n->users, 1);
589
590 atomic_inc(&(skb_shinfo(skb)->dataref));
591 skb->cloned = 1;
592
593 return n;
594#undef C
595}
596
597/**
598 * skb_morph - morph one skb into another
599 * @dst: the skb to receive the contents
600 * @src: the skb to supply the contents
601 *
602 * This is identical to skb_clone except that the target skb is
603 * supplied by the user.
604 *
605 * The target skb is returned upon exit.
606 */
607struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
608{
609 skb_release_all(dst);
610 return __skb_clone(dst, src);
611}
612EXPORT_SYMBOL_GPL(skb_morph);
613
614/* skb_copy_ubufs - copy userspace skb frags buffers to kernel
615 * @skb: the skb to modify
616 * @gfp_mask: allocation priority
617 *
618 * This must be called on SKBTX_DEV_ZEROCOPY skb.
619 * It will copy all frags into kernel and drop the reference
620 * to userspace pages.
621 *
622 * If this function is called from an interrupt gfp_mask() must be
623 * %GFP_ATOMIC.
624 *
625 * Returns 0 on success or a negative error code on failure
626 * to allocate kernel memory to copy to.
627 */
628int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
629{
630 int i;
631 int num_frags = skb_shinfo(skb)->nr_frags;
632 struct page *page, *head = NULL;
633 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
634
635 for (i = 0; i < num_frags; i++) {
636 u8 *vaddr;
637 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
638
639 page = alloc_page(GFP_ATOMIC);
640 if (!page) {
641 while (head) {
642 struct page *next = (struct page *)head->private;
643 put_page(head);
644 head = next;
645 }
646 return -ENOMEM;
647 }
648 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
649 memcpy(page_address(page),
650 vaddr + f->page_offset, f->size);
651 kunmap_skb_frag(vaddr);
652 page->private = (unsigned long)head;
653 head = page;
654 }
655
656 /* skb frags release userspace buffers */
657 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
658 put_page(skb_shinfo(skb)->frags[i].page);
659
660 uarg->callback(uarg);
661
662 /* skb frags point to kernel buffers */
663 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
664 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
665 skb_shinfo(skb)->frags[i - 1].page = head;
666 head = (struct page *)head->private;
667 }
668
669 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
670 return 0;
671}
672
673
674/**
675 * skb_clone - duplicate an sk_buff
676 * @skb: buffer to clone
677 * @gfp_mask: allocation priority
678 *
679 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
680 * copies share the same packet data but not structure. The new
681 * buffer has a reference count of 1. If the allocation fails the
682 * function returns %NULL otherwise the new buffer is returned.
683 *
684 * If this function is called from an interrupt gfp_mask() must be
685 * %GFP_ATOMIC.
686 */
687
688struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
689{
690 struct sk_buff *n;
691
692 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
693 if (skb_copy_ubufs(skb, gfp_mask))
694 return NULL;
695 }
696
697 n = skb + 1;
698 if (skb->fclone == SKB_FCLONE_ORIG &&
699 n->fclone == SKB_FCLONE_UNAVAILABLE) {
700 atomic_t *fclone_ref = (atomic_t *) (n + 1);
701 n->fclone = SKB_FCLONE_CLONE;
702 atomic_inc(fclone_ref);
703 } else {
704 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
705 if (!n)
706 return NULL;
707
708 kmemcheck_annotate_bitfield(n, flags1);
709 kmemcheck_annotate_bitfield(n, flags2);
710 n->fclone = SKB_FCLONE_UNAVAILABLE;
711 }
712
713 return __skb_clone(n, skb);
714}
715EXPORT_SYMBOL(skb_clone);
716
717static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
718{
719#ifndef NET_SKBUFF_DATA_USES_OFFSET
720 /*
721 * Shift between the two data areas in bytes
722 */
723 unsigned long offset = new->data - old->data;
724#endif
725
726 __copy_skb_header(new, old);
727
728#ifndef NET_SKBUFF_DATA_USES_OFFSET
729 /* {transport,network,mac}_header are relative to skb->head */
730 new->transport_header += offset;
731 new->network_header += offset;
732 if (skb_mac_header_was_set(new))
733 new->mac_header += offset;
734#endif
735 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
736 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
737 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
738}
739
740/**
741 * skb_copy - create private copy of an sk_buff
742 * @skb: buffer to copy
743 * @gfp_mask: allocation priority
744 *
745 * Make a copy of both an &sk_buff and its data. This is used when the
746 * caller wishes to modify the data and needs a private copy of the
747 * data to alter. Returns %NULL on failure or the pointer to the buffer
748 * on success. The returned buffer has a reference count of 1.
749 *
750 * As by-product this function converts non-linear &sk_buff to linear
751 * one, so that &sk_buff becomes completely private and caller is allowed
752 * to modify all the data of returned buffer. This means that this
753 * function is not recommended for use in circumstances when only
754 * header is going to be modified. Use pskb_copy() instead.
755 */
756
757struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
758{
759 int headerlen = skb_headroom(skb);
760 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
761 struct sk_buff *n = alloc_skb(size, gfp_mask);
762
763 if (!n)
764 return NULL;
765
766 /* Set the data pointer */
767 skb_reserve(n, headerlen);
768 /* Set the tail pointer and length */
769 skb_put(n, skb->len);
770
771 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
772 BUG();
773
774 copy_skb_header(n, skb);
775 return n;
776}
777EXPORT_SYMBOL(skb_copy);
778
779/**
780 * pskb_copy - create copy of an sk_buff with private head.
781 * @skb: buffer to copy
782 * @gfp_mask: allocation priority
783 *
784 * Make a copy of both an &sk_buff and part of its data, located
785 * in header. Fragmented data remain shared. This is used when
786 * the caller wishes to modify only header of &sk_buff and needs
787 * private copy of the header to alter. Returns %NULL on failure
788 * or the pointer to the buffer on success.
789 * The returned buffer has a reference count of 1.
790 */
791
792struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
793{
794 unsigned int size = skb_end_pointer(skb) - skb->head;
795 struct sk_buff *n = alloc_skb(size, gfp_mask);
796
797 if (!n)
798 goto out;
799
800 /* Set the data pointer */
801 skb_reserve(n, skb_headroom(skb));
802 /* Set the tail pointer and length */
803 skb_put(n, skb_headlen(skb));
804 /* Copy the bytes */
805 skb_copy_from_linear_data(skb, n->data, n->len);
806
807 n->truesize += skb->data_len;
808 n->data_len = skb->data_len;
809 n->len = skb->len;
810
811 if (skb_shinfo(skb)->nr_frags) {
812 int i;
813
814 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
815 if (skb_copy_ubufs(skb, gfp_mask)) {
816 kfree_skb(n);
817 n = NULL;
818 goto out;
819 }
820 }
821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
822 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
823 get_page(skb_shinfo(n)->frags[i].page);
824 }
825 skb_shinfo(n)->nr_frags = i;
826 }
827
828 if (skb_has_frag_list(skb)) {
829 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
830 skb_clone_fraglist(n);
831 }
832
833 copy_skb_header(n, skb);
834out:
835 return n;
836}
837EXPORT_SYMBOL(pskb_copy);
838
839/**
840 * pskb_expand_head - reallocate header of &sk_buff
841 * @skb: buffer to reallocate
842 * @nhead: room to add at head
843 * @ntail: room to add at tail
844 * @gfp_mask: allocation priority
845 *
846 * Expands (or creates identical copy, if &nhead and &ntail are zero)
847 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
848 * reference count of 1. Returns zero in the case of success or error,
849 * if expansion failed. In the last case, &sk_buff is not changed.
850 *
851 * All the pointers pointing into skb header may change and must be
852 * reloaded after call to this function.
853 */
854
855int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
856 gfp_t gfp_mask)
857{
858 int i;
859 u8 *data;
860 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
861 long off;
862 bool fastpath;
863
864 BUG_ON(nhead < 0);
865
866 if (skb_shared(skb))
867 BUG();
868
869 size = SKB_DATA_ALIGN(size);
870
871 /* Check if we can avoid taking references on fragments if we own
872 * the last reference on skb->head. (see skb_release_data())
873 */
874 if (!skb->cloned)
875 fastpath = true;
876 else {
877 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
878 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
879 }
880
881 if (fastpath &&
882 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
883 memmove(skb->head + size, skb_shinfo(skb),
884 offsetof(struct skb_shared_info,
885 frags[skb_shinfo(skb)->nr_frags]));
886 memmove(skb->head + nhead, skb->head,
887 skb_tail_pointer(skb) - skb->head);
888 off = nhead;
889 goto adjust_others;
890 }
891
892 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
893 if (!data)
894 goto nodata;
895
896 /* Copy only real data... and, alas, header. This should be
897 * optimized for the cases when header is void.
898 */
899 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
900
901 memcpy((struct skb_shared_info *)(data + size),
902 skb_shinfo(skb),
903 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
904
905 if (fastpath) {
906 kfree(skb->head);
907 } else {
908 /* copy this zero copy skb frags */
909 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
910 if (skb_copy_ubufs(skb, gfp_mask))
911 goto nofrags;
912 }
913 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
914 get_page(skb_shinfo(skb)->frags[i].page);
915
916 if (skb_has_frag_list(skb))
917 skb_clone_fraglist(skb);
918
919 skb_release_data(skb);
920 }
921 off = (data + nhead) - skb->head;
922
923 skb->head = data;
924adjust_others:
925 skb->data += off;
926#ifdef NET_SKBUFF_DATA_USES_OFFSET
927 skb->end = size;
928 off = nhead;
929#else
930 skb->end = skb->head + size;
931#endif
932 /* {transport,network,mac}_header and tail are relative to skb->head */
933 skb->tail += off;
934 skb->transport_header += off;
935 skb->network_header += off;
936 if (skb_mac_header_was_set(skb))
937 skb->mac_header += off;
938 /* Only adjust this if it actually is csum_start rather than csum */
939 if (skb->ip_summed == CHECKSUM_PARTIAL)
940 skb->csum_start += nhead;
941 skb->cloned = 0;
942 skb->hdr_len = 0;
943 skb->nohdr = 0;
944 atomic_set(&skb_shinfo(skb)->dataref, 1);
945 return 0;
946
947nofrags:
948 kfree(data);
949nodata:
950 return -ENOMEM;
951}
952EXPORT_SYMBOL(pskb_expand_head);
953
954/* Make private copy of skb with writable head and some headroom */
955
956struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
957{
958 struct sk_buff *skb2;
959 int delta = headroom - skb_headroom(skb);
960
961 if (delta <= 0)
962 skb2 = pskb_copy(skb, GFP_ATOMIC);
963 else {
964 skb2 = skb_clone(skb, GFP_ATOMIC);
965 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
966 GFP_ATOMIC)) {
967 kfree_skb(skb2);
968 skb2 = NULL;
969 }
970 }
971 return skb2;
972}
973EXPORT_SYMBOL(skb_realloc_headroom);
974
975/**
976 * skb_copy_expand - copy and expand sk_buff
977 * @skb: buffer to copy
978 * @newheadroom: new free bytes at head
979 * @newtailroom: new free bytes at tail
980 * @gfp_mask: allocation priority
981 *
982 * Make a copy of both an &sk_buff and its data and while doing so
983 * allocate additional space.
984 *
985 * This is used when the caller wishes to modify the data and needs a
986 * private copy of the data to alter as well as more space for new fields.
987 * Returns %NULL on failure or the pointer to the buffer
988 * on success. The returned buffer has a reference count of 1.
989 *
990 * You must pass %GFP_ATOMIC as the allocation priority if this function
991 * is called from an interrupt.
992 */
993struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
994 int newheadroom, int newtailroom,
995 gfp_t gfp_mask)
996{
997 /*
998 * Allocate the copy buffer
999 */
1000 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1001 gfp_mask);
1002 int oldheadroom = skb_headroom(skb);
1003 int head_copy_len, head_copy_off;
1004 int off;
1005
1006 if (!n)
1007 return NULL;
1008
1009 skb_reserve(n, newheadroom);
1010
1011 /* Set the tail pointer and length */
1012 skb_put(n, skb->len);
1013
1014 head_copy_len = oldheadroom;
1015 head_copy_off = 0;
1016 if (newheadroom <= head_copy_len)
1017 head_copy_len = newheadroom;
1018 else
1019 head_copy_off = newheadroom - head_copy_len;
1020
1021 /* Copy the linear header and data. */
1022 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1023 skb->len + head_copy_len))
1024 BUG();
1025
1026 copy_skb_header(n, skb);
1027
1028 off = newheadroom - oldheadroom;
1029 if (n->ip_summed == CHECKSUM_PARTIAL)
1030 n->csum_start += off;
1031#ifdef NET_SKBUFF_DATA_USES_OFFSET
1032 n->transport_header += off;
1033 n->network_header += off;
1034 if (skb_mac_header_was_set(skb))
1035 n->mac_header += off;
1036#endif
1037
1038 return n;
1039}
1040EXPORT_SYMBOL(skb_copy_expand);
1041
1042/**
1043 * skb_pad - zero pad the tail of an skb
1044 * @skb: buffer to pad
1045 * @pad: space to pad
1046 *
1047 * Ensure that a buffer is followed by a padding area that is zero
1048 * filled. Used by network drivers which may DMA or transfer data
1049 * beyond the buffer end onto the wire.
1050 *
1051 * May return error in out of memory cases. The skb is freed on error.
1052 */
1053
1054int skb_pad(struct sk_buff *skb, int pad)
1055{
1056 int err;
1057 int ntail;
1058
1059 /* If the skbuff is non linear tailroom is always zero.. */
1060 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1061 memset(skb->data+skb->len, 0, pad);
1062 return 0;
1063 }
1064
1065 ntail = skb->data_len + pad - (skb->end - skb->tail);
1066 if (likely(skb_cloned(skb) || ntail > 0)) {
1067 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1068 if (unlikely(err))
1069 goto free_skb;
1070 }
1071
1072 /* FIXME: The use of this function with non-linear skb's really needs
1073 * to be audited.
1074 */
1075 err = skb_linearize(skb);
1076 if (unlikely(err))
1077 goto free_skb;
1078
1079 memset(skb->data + skb->len, 0, pad);
1080 return 0;
1081
1082free_skb:
1083 kfree_skb(skb);
1084 return err;
1085}
1086EXPORT_SYMBOL(skb_pad);
1087
1088/**
1089 * skb_put - add data to a buffer
1090 * @skb: buffer to use
1091 * @len: amount of data to add
1092 *
1093 * This function extends the used data area of the buffer. If this would
1094 * exceed the total buffer size the kernel will panic. A pointer to the
1095 * first byte of the extra data is returned.
1096 */
1097unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1098{
1099 unsigned char *tmp = skb_tail_pointer(skb);
1100 SKB_LINEAR_ASSERT(skb);
1101 skb->tail += len;
1102 skb->len += len;
1103 if (unlikely(skb->tail > skb->end))
1104 skb_over_panic(skb, len, __builtin_return_address(0));
1105 return tmp;
1106}
1107EXPORT_SYMBOL(skb_put);
1108
1109/**
1110 * skb_push - add data to the start of a buffer
1111 * @skb: buffer to use
1112 * @len: amount of data to add
1113 *
1114 * This function extends the used data area of the buffer at the buffer
1115 * start. If this would exceed the total buffer headroom the kernel will
1116 * panic. A pointer to the first byte of the extra data is returned.
1117 */
1118unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1119{
1120 skb->data -= len;
1121 skb->len += len;
1122 if (unlikely(skb->data<skb->head))
1123 skb_under_panic(skb, len, __builtin_return_address(0));
1124 return skb->data;
1125}
1126EXPORT_SYMBOL(skb_push);
1127
1128/**
1129 * skb_pull - remove data from the start of a buffer
1130 * @skb: buffer to use
1131 * @len: amount of data to remove
1132 *
1133 * This function removes data from the start of a buffer, returning
1134 * the memory to the headroom. A pointer to the next data in the buffer
1135 * is returned. Once the data has been pulled future pushes will overwrite
1136 * the old data.
1137 */
1138unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1139{
1140 return skb_pull_inline(skb, len);
1141}
1142EXPORT_SYMBOL(skb_pull);
1143
1144/**
1145 * skb_trim - remove end from a buffer
1146 * @skb: buffer to alter
1147 * @len: new length
1148 *
1149 * Cut the length of a buffer down by removing data from the tail. If
1150 * the buffer is already under the length specified it is not modified.
1151 * The skb must be linear.
1152 */
1153void skb_trim(struct sk_buff *skb, unsigned int len)
1154{
1155 if (skb->len > len)
1156 __skb_trim(skb, len);
1157}
1158EXPORT_SYMBOL(skb_trim);
1159
1160/* Trims skb to length len. It can change skb pointers.
1161 */
1162
1163int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1164{
1165 struct sk_buff **fragp;
1166 struct sk_buff *frag;
1167 int offset = skb_headlen(skb);
1168 int nfrags = skb_shinfo(skb)->nr_frags;
1169 int i;
1170 int err;
1171
1172 if (skb_cloned(skb) &&
1173 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1174 return err;
1175
1176 i = 0;
1177 if (offset >= len)
1178 goto drop_pages;
1179
1180 for (; i < nfrags; i++) {
1181 int end = offset + skb_shinfo(skb)->frags[i].size;
1182
1183 if (end < len) {
1184 offset = end;
1185 continue;
1186 }
1187
1188 skb_shinfo(skb)->frags[i++].size = len - offset;
1189
1190drop_pages:
1191 skb_shinfo(skb)->nr_frags = i;
1192
1193 for (; i < nfrags; i++)
1194 put_page(skb_shinfo(skb)->frags[i].page);
1195
1196 if (skb_has_frag_list(skb))
1197 skb_drop_fraglist(skb);
1198 goto done;
1199 }
1200
1201 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1202 fragp = &frag->next) {
1203 int end = offset + frag->len;
1204
1205 if (skb_shared(frag)) {
1206 struct sk_buff *nfrag;
1207
1208 nfrag = skb_clone(frag, GFP_ATOMIC);
1209 if (unlikely(!nfrag))
1210 return -ENOMEM;
1211
1212 nfrag->next = frag->next;
1213 kfree_skb(frag);
1214 frag = nfrag;
1215 *fragp = frag;
1216 }
1217
1218 if (end < len) {
1219 offset = end;
1220 continue;
1221 }
1222
1223 if (end > len &&
1224 unlikely((err = pskb_trim(frag, len - offset))))
1225 return err;
1226
1227 if (frag->next)
1228 skb_drop_list(&frag->next);
1229 break;
1230 }
1231
1232done:
1233 if (len > skb_headlen(skb)) {
1234 skb->data_len -= skb->len - len;
1235 skb->len = len;
1236 } else {
1237 skb->len = len;
1238 skb->data_len = 0;
1239 skb_set_tail_pointer(skb, len);
1240 }
1241
1242 return 0;
1243}
1244EXPORT_SYMBOL(___pskb_trim);
1245
1246/**
1247 * __pskb_pull_tail - advance tail of skb header
1248 * @skb: buffer to reallocate
1249 * @delta: number of bytes to advance tail
1250 *
1251 * The function makes a sense only on a fragmented &sk_buff,
1252 * it expands header moving its tail forward and copying necessary
1253 * data from fragmented part.
1254 *
1255 * &sk_buff MUST have reference count of 1.
1256 *
1257 * Returns %NULL (and &sk_buff does not change) if pull failed
1258 * or value of new tail of skb in the case of success.
1259 *
1260 * All the pointers pointing into skb header may change and must be
1261 * reloaded after call to this function.
1262 */
1263
1264/* Moves tail of skb head forward, copying data from fragmented part,
1265 * when it is necessary.
1266 * 1. It may fail due to malloc failure.
1267 * 2. It may change skb pointers.
1268 *
1269 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1270 */
1271unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1272{
1273 /* If skb has not enough free space at tail, get new one
1274 * plus 128 bytes for future expansions. If we have enough
1275 * room at tail, reallocate without expansion only if skb is cloned.
1276 */
1277 int i, k, eat = (skb->tail + delta) - skb->end;
1278
1279 if (eat > 0 || skb_cloned(skb)) {
1280 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1281 GFP_ATOMIC))
1282 return NULL;
1283 }
1284
1285 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1286 BUG();
1287
1288 /* Optimization: no fragments, no reasons to preestimate
1289 * size of pulled pages. Superb.
1290 */
1291 if (!skb_has_frag_list(skb))
1292 goto pull_pages;
1293
1294 /* Estimate size of pulled pages. */
1295 eat = delta;
1296 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1297 if (skb_shinfo(skb)->frags[i].size >= eat)
1298 goto pull_pages;
1299 eat -= skb_shinfo(skb)->frags[i].size;
1300 }
1301
1302 /* If we need update frag list, we are in troubles.
1303 * Certainly, it possible to add an offset to skb data,
1304 * but taking into account that pulling is expected to
1305 * be very rare operation, it is worth to fight against
1306 * further bloating skb head and crucify ourselves here instead.
1307 * Pure masohism, indeed. 8)8)
1308 */
1309 if (eat) {
1310 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311 struct sk_buff *clone = NULL;
1312 struct sk_buff *insp = NULL;
1313
1314 do {
1315 BUG_ON(!list);
1316
1317 if (list->len <= eat) {
1318 /* Eaten as whole. */
1319 eat -= list->len;
1320 list = list->next;
1321 insp = list;
1322 } else {
1323 /* Eaten partially. */
1324
1325 if (skb_shared(list)) {
1326 /* Sucks! We need to fork list. :-( */
1327 clone = skb_clone(list, GFP_ATOMIC);
1328 if (!clone)
1329 return NULL;
1330 insp = list->next;
1331 list = clone;
1332 } else {
1333 /* This may be pulled without
1334 * problems. */
1335 insp = list;
1336 }
1337 if (!pskb_pull(list, eat)) {
1338 kfree_skb(clone);
1339 return NULL;
1340 }
1341 break;
1342 }
1343 } while (eat);
1344
1345 /* Free pulled out fragments. */
1346 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1347 skb_shinfo(skb)->frag_list = list->next;
1348 kfree_skb(list);
1349 }
1350 /* And insert new clone at head. */
1351 if (clone) {
1352 clone->next = list;
1353 skb_shinfo(skb)->frag_list = clone;
1354 }
1355 }
1356 /* Success! Now we may commit changes to skb data. */
1357
1358pull_pages:
1359 eat = delta;
1360 k = 0;
1361 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362 if (skb_shinfo(skb)->frags[i].size <= eat) {
1363 put_page(skb_shinfo(skb)->frags[i].page);
1364 eat -= skb_shinfo(skb)->frags[i].size;
1365 } else {
1366 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1367 if (eat) {
1368 skb_shinfo(skb)->frags[k].page_offset += eat;
1369 skb_shinfo(skb)->frags[k].size -= eat;
1370 eat = 0;
1371 }
1372 k++;
1373 }
1374 }
1375 skb_shinfo(skb)->nr_frags = k;
1376
1377 skb->tail += delta;
1378 skb->data_len -= delta;
1379
1380 return skb_tail_pointer(skb);
1381}
1382EXPORT_SYMBOL(__pskb_pull_tail);
1383
1384/**
1385 * skb_copy_bits - copy bits from skb to kernel buffer
1386 * @skb: source skb
1387 * @offset: offset in source
1388 * @to: destination buffer
1389 * @len: number of bytes to copy
1390 *
1391 * Copy the specified number of bytes from the source skb to the
1392 * destination buffer.
1393 *
1394 * CAUTION ! :
1395 * If its prototype is ever changed,
1396 * check arch/{*}/net/{*}.S files,
1397 * since it is called from BPF assembly code.
1398 */
1399int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1400{
1401 int start = skb_headlen(skb);
1402 struct sk_buff *frag_iter;
1403 int i, copy;
1404
1405 if (offset > (int)skb->len - len)
1406 goto fault;
1407
1408 /* Copy header. */
1409 if ((copy = start - offset) > 0) {
1410 if (copy > len)
1411 copy = len;
1412 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1413 if ((len -= copy) == 0)
1414 return 0;
1415 offset += copy;
1416 to += copy;
1417 }
1418
1419 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1420 int end;
1421
1422 WARN_ON(start > offset + len);
1423
1424 end = start + skb_shinfo(skb)->frags[i].size;
1425 if ((copy = end - offset) > 0) {
1426 u8 *vaddr;
1427
1428 if (copy > len)
1429 copy = len;
1430
1431 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1432 memcpy(to,
1433 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1434 offset - start, copy);
1435 kunmap_skb_frag(vaddr);
1436
1437 if ((len -= copy) == 0)
1438 return 0;
1439 offset += copy;
1440 to += copy;
1441 }
1442 start = end;
1443 }
1444
1445 skb_walk_frags(skb, frag_iter) {
1446 int end;
1447
1448 WARN_ON(start > offset + len);
1449
1450 end = start + frag_iter->len;
1451 if ((copy = end - offset) > 0) {
1452 if (copy > len)
1453 copy = len;
1454 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1455 goto fault;
1456 if ((len -= copy) == 0)
1457 return 0;
1458 offset += copy;
1459 to += copy;
1460 }
1461 start = end;
1462 }
1463
1464 if (!len)
1465 return 0;
1466
1467fault:
1468 return -EFAULT;
1469}
1470EXPORT_SYMBOL(skb_copy_bits);
1471
1472/*
1473 * Callback from splice_to_pipe(), if we need to release some pages
1474 * at the end of the spd in case we error'ed out in filling the pipe.
1475 */
1476static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1477{
1478 put_page(spd->pages[i]);
1479}
1480
1481static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1482 unsigned int *offset,
1483 struct sk_buff *skb, struct sock *sk)
1484{
1485 struct page *p = sk->sk_sndmsg_page;
1486 unsigned int off;
1487
1488 if (!p) {
1489new_page:
1490 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1491 if (!p)
1492 return NULL;
1493
1494 off = sk->sk_sndmsg_off = 0;
1495 /* hold one ref to this page until it's full */
1496 } else {
1497 unsigned int mlen;
1498
1499 off = sk->sk_sndmsg_off;
1500 mlen = PAGE_SIZE - off;
1501 if (mlen < 64 && mlen < *len) {
1502 put_page(p);
1503 goto new_page;
1504 }
1505
1506 *len = min_t(unsigned int, *len, mlen);
1507 }
1508
1509 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1510 sk->sk_sndmsg_off += *len;
1511 *offset = off;
1512 get_page(p);
1513
1514 return p;
1515}
1516
1517/*
1518 * Fill page/offset/length into spd, if it can hold more pages.
1519 */
1520static inline int spd_fill_page(struct splice_pipe_desc *spd,
1521 struct pipe_inode_info *pipe, struct page *page,
1522 unsigned int *len, unsigned int offset,
1523 struct sk_buff *skb, int linear,
1524 struct sock *sk)
1525{
1526 if (unlikely(spd->nr_pages == pipe->buffers))
1527 return 1;
1528
1529 if (linear) {
1530 page = linear_to_page(page, len, &offset, skb, sk);
1531 if (!page)
1532 return 1;
1533 } else
1534 get_page(page);
1535
1536 spd->pages[spd->nr_pages] = page;
1537 spd->partial[spd->nr_pages].len = *len;
1538 spd->partial[spd->nr_pages].offset = offset;
1539 spd->nr_pages++;
1540
1541 return 0;
1542}
1543
1544static inline void __segment_seek(struct page **page, unsigned int *poff,
1545 unsigned int *plen, unsigned int off)
1546{
1547 unsigned long n;
1548
1549 *poff += off;
1550 n = *poff / PAGE_SIZE;
1551 if (n)
1552 *page = nth_page(*page, n);
1553
1554 *poff = *poff % PAGE_SIZE;
1555 *plen -= off;
1556}
1557
1558static inline int __splice_segment(struct page *page, unsigned int poff,
1559 unsigned int plen, unsigned int *off,
1560 unsigned int *len, struct sk_buff *skb,
1561 struct splice_pipe_desc *spd, int linear,
1562 struct sock *sk,
1563 struct pipe_inode_info *pipe)
1564{
1565 if (!*len)
1566 return 1;
1567
1568 /* skip this segment if already processed */
1569 if (*off >= plen) {
1570 *off -= plen;
1571 return 0;
1572 }
1573
1574 /* ignore any bits we already processed */
1575 if (*off) {
1576 __segment_seek(&page, &poff, &plen, *off);
1577 *off = 0;
1578 }
1579
1580 do {
1581 unsigned int flen = min(*len, plen);
1582
1583 /* the linear region may spread across several pages */
1584 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1585
1586 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1587 return 1;
1588
1589 __segment_seek(&page, &poff, &plen, flen);
1590 *len -= flen;
1591
1592 } while (*len && plen);
1593
1594 return 0;
1595}
1596
1597/*
1598 * Map linear and fragment data from the skb to spd. It reports failure if the
1599 * pipe is full or if we already spliced the requested length.
1600 */
1601static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1602 unsigned int *offset, unsigned int *len,
1603 struct splice_pipe_desc *spd, struct sock *sk)
1604{
1605 int seg;
1606
1607 /*
1608 * map the linear part
1609 */
1610 if (__splice_segment(virt_to_page(skb->data),
1611 (unsigned long) skb->data & (PAGE_SIZE - 1),
1612 skb_headlen(skb),
1613 offset, len, skb, spd, 1, sk, pipe))
1614 return 1;
1615
1616 /*
1617 * then map the fragments
1618 */
1619 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1620 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1621
1622 if (__splice_segment(f->page, f->page_offset, f->size,
1623 offset, len, skb, spd, 0, sk, pipe))
1624 return 1;
1625 }
1626
1627 return 0;
1628}
1629
1630/*
1631 * Map data from the skb to a pipe. Should handle both the linear part,
1632 * the fragments, and the frag list. It does NOT handle frag lists within
1633 * the frag list, if such a thing exists. We'd probably need to recurse to
1634 * handle that cleanly.
1635 */
1636int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1637 struct pipe_inode_info *pipe, unsigned int tlen,
1638 unsigned int flags)
1639{
1640 struct partial_page partial[PIPE_DEF_BUFFERS];
1641 struct page *pages[PIPE_DEF_BUFFERS];
1642 struct splice_pipe_desc spd = {
1643 .pages = pages,
1644 .partial = partial,
1645 .flags = flags,
1646 .ops = &sock_pipe_buf_ops,
1647 .spd_release = sock_spd_release,
1648 };
1649 struct sk_buff *frag_iter;
1650 struct sock *sk = skb->sk;
1651 int ret = 0;
1652
1653 if (splice_grow_spd(pipe, &spd))
1654 return -ENOMEM;
1655
1656 /*
1657 * __skb_splice_bits() only fails if the output has no room left,
1658 * so no point in going over the frag_list for the error case.
1659 */
1660 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1661 goto done;
1662 else if (!tlen)
1663 goto done;
1664
1665 /*
1666 * now see if we have a frag_list to map
1667 */
1668 skb_walk_frags(skb, frag_iter) {
1669 if (!tlen)
1670 break;
1671 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1672 break;
1673 }
1674
1675done:
1676 if (spd.nr_pages) {
1677 /*
1678 * Drop the socket lock, otherwise we have reverse
1679 * locking dependencies between sk_lock and i_mutex
1680 * here as compared to sendfile(). We enter here
1681 * with the socket lock held, and splice_to_pipe() will
1682 * grab the pipe inode lock. For sendfile() emulation,
1683 * we call into ->sendpage() with the i_mutex lock held
1684 * and networking will grab the socket lock.
1685 */
1686 release_sock(sk);
1687 ret = splice_to_pipe(pipe, &spd);
1688 lock_sock(sk);
1689 }
1690
1691 splice_shrink_spd(pipe, &spd);
1692 return ret;
1693}
1694
1695/**
1696 * skb_store_bits - store bits from kernel buffer to skb
1697 * @skb: destination buffer
1698 * @offset: offset in destination
1699 * @from: source buffer
1700 * @len: number of bytes to copy
1701 *
1702 * Copy the specified number of bytes from the source buffer to the
1703 * destination skb. This function handles all the messy bits of
1704 * traversing fragment lists and such.
1705 */
1706
1707int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1708{
1709 int start = skb_headlen(skb);
1710 struct sk_buff *frag_iter;
1711 int i, copy;
1712
1713 if (offset > (int)skb->len - len)
1714 goto fault;
1715
1716 if ((copy = start - offset) > 0) {
1717 if (copy > len)
1718 copy = len;
1719 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1720 if ((len -= copy) == 0)
1721 return 0;
1722 offset += copy;
1723 from += copy;
1724 }
1725
1726 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1728 int end;
1729
1730 WARN_ON(start > offset + len);
1731
1732 end = start + frag->size;
1733 if ((copy = end - offset) > 0) {
1734 u8 *vaddr;
1735
1736 if (copy > len)
1737 copy = len;
1738
1739 vaddr = kmap_skb_frag(frag);
1740 memcpy(vaddr + frag->page_offset + offset - start,
1741 from, copy);
1742 kunmap_skb_frag(vaddr);
1743
1744 if ((len -= copy) == 0)
1745 return 0;
1746 offset += copy;
1747 from += copy;
1748 }
1749 start = end;
1750 }
1751
1752 skb_walk_frags(skb, frag_iter) {
1753 int end;
1754
1755 WARN_ON(start > offset + len);
1756
1757 end = start + frag_iter->len;
1758 if ((copy = end - offset) > 0) {
1759 if (copy > len)
1760 copy = len;
1761 if (skb_store_bits(frag_iter, offset - start,
1762 from, copy))
1763 goto fault;
1764 if ((len -= copy) == 0)
1765 return 0;
1766 offset += copy;
1767 from += copy;
1768 }
1769 start = end;
1770 }
1771 if (!len)
1772 return 0;
1773
1774fault:
1775 return -EFAULT;
1776}
1777EXPORT_SYMBOL(skb_store_bits);
1778
1779/* Checksum skb data. */
1780
1781__wsum skb_checksum(const struct sk_buff *skb, int offset,
1782 int len, __wsum csum)
1783{
1784 int start = skb_headlen(skb);
1785 int i, copy = start - offset;
1786 struct sk_buff *frag_iter;
1787 int pos = 0;
1788
1789 /* Checksum header. */
1790 if (copy > 0) {
1791 if (copy > len)
1792 copy = len;
1793 csum = csum_partial(skb->data + offset, copy, csum);
1794 if ((len -= copy) == 0)
1795 return csum;
1796 offset += copy;
1797 pos = copy;
1798 }
1799
1800 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1801 int end;
1802
1803 WARN_ON(start > offset + len);
1804
1805 end = start + skb_shinfo(skb)->frags[i].size;
1806 if ((copy = end - offset) > 0) {
1807 __wsum csum2;
1808 u8 *vaddr;
1809 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1810
1811 if (copy > len)
1812 copy = len;
1813 vaddr = kmap_skb_frag(frag);
1814 csum2 = csum_partial(vaddr + frag->page_offset +
1815 offset - start, copy, 0);
1816 kunmap_skb_frag(vaddr);
1817 csum = csum_block_add(csum, csum2, pos);
1818 if (!(len -= copy))
1819 return csum;
1820 offset += copy;
1821 pos += copy;
1822 }
1823 start = end;
1824 }
1825
1826 skb_walk_frags(skb, frag_iter) {
1827 int end;
1828
1829 WARN_ON(start > offset + len);
1830
1831 end = start + frag_iter->len;
1832 if ((copy = end - offset) > 0) {
1833 __wsum csum2;
1834 if (copy > len)
1835 copy = len;
1836 csum2 = skb_checksum(frag_iter, offset - start,
1837 copy, 0);
1838 csum = csum_block_add(csum, csum2, pos);
1839 if ((len -= copy) == 0)
1840 return csum;
1841 offset += copy;
1842 pos += copy;
1843 }
1844 start = end;
1845 }
1846 BUG_ON(len);
1847
1848 return csum;
1849}
1850EXPORT_SYMBOL(skb_checksum);
1851
1852/* Both of above in one bottle. */
1853
1854__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1855 u8 *to, int len, __wsum csum)
1856{
1857 int start = skb_headlen(skb);
1858 int i, copy = start - offset;
1859 struct sk_buff *frag_iter;
1860 int pos = 0;
1861
1862 /* Copy header. */
1863 if (copy > 0) {
1864 if (copy > len)
1865 copy = len;
1866 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1867 copy, csum);
1868 if ((len -= copy) == 0)
1869 return csum;
1870 offset += copy;
1871 to += copy;
1872 pos = copy;
1873 }
1874
1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876 int end;
1877
1878 WARN_ON(start > offset + len);
1879
1880 end = start + skb_shinfo(skb)->frags[i].size;
1881 if ((copy = end - offset) > 0) {
1882 __wsum csum2;
1883 u8 *vaddr;
1884 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1885
1886 if (copy > len)
1887 copy = len;
1888 vaddr = kmap_skb_frag(frag);
1889 csum2 = csum_partial_copy_nocheck(vaddr +
1890 frag->page_offset +
1891 offset - start, to,
1892 copy, 0);
1893 kunmap_skb_frag(vaddr);
1894 csum = csum_block_add(csum, csum2, pos);
1895 if (!(len -= copy))
1896 return csum;
1897 offset += copy;
1898 to += copy;
1899 pos += copy;
1900 }
1901 start = end;
1902 }
1903
1904 skb_walk_frags(skb, frag_iter) {
1905 __wsum csum2;
1906 int end;
1907
1908 WARN_ON(start > offset + len);
1909
1910 end = start + frag_iter->len;
1911 if ((copy = end - offset) > 0) {
1912 if (copy > len)
1913 copy = len;
1914 csum2 = skb_copy_and_csum_bits(frag_iter,
1915 offset - start,
1916 to, copy, 0);
1917 csum = csum_block_add(csum, csum2, pos);
1918 if ((len -= copy) == 0)
1919 return csum;
1920 offset += copy;
1921 to += copy;
1922 pos += copy;
1923 }
1924 start = end;
1925 }
1926 BUG_ON(len);
1927 return csum;
1928}
1929EXPORT_SYMBOL(skb_copy_and_csum_bits);
1930
1931void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1932{
1933 __wsum csum;
1934 long csstart;
1935
1936 if (skb->ip_summed == CHECKSUM_PARTIAL)
1937 csstart = skb_checksum_start_offset(skb);
1938 else
1939 csstart = skb_headlen(skb);
1940
1941 BUG_ON(csstart > skb_headlen(skb));
1942
1943 skb_copy_from_linear_data(skb, to, csstart);
1944
1945 csum = 0;
1946 if (csstart != skb->len)
1947 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1948 skb->len - csstart, 0);
1949
1950 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1951 long csstuff = csstart + skb->csum_offset;
1952
1953 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1954 }
1955}
1956EXPORT_SYMBOL(skb_copy_and_csum_dev);
1957
1958/**
1959 * skb_dequeue - remove from the head of the queue
1960 * @list: list to dequeue from
1961 *
1962 * Remove the head of the list. The list lock is taken so the function
1963 * may be used safely with other locking list functions. The head item is
1964 * returned or %NULL if the list is empty.
1965 */
1966
1967struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1968{
1969 unsigned long flags;
1970 struct sk_buff *result;
1971
1972 spin_lock_irqsave(&list->lock, flags);
1973 result = __skb_dequeue(list);
1974 spin_unlock_irqrestore(&list->lock, flags);
1975 return result;
1976}
1977EXPORT_SYMBOL(skb_dequeue);
1978
1979/**
1980 * skb_dequeue_tail - remove from the tail of the queue
1981 * @list: list to dequeue from
1982 *
1983 * Remove the tail of the list. The list lock is taken so the function
1984 * may be used safely with other locking list functions. The tail item is
1985 * returned or %NULL if the list is empty.
1986 */
1987struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1988{
1989 unsigned long flags;
1990 struct sk_buff *result;
1991
1992 spin_lock_irqsave(&list->lock, flags);
1993 result = __skb_dequeue_tail(list);
1994 spin_unlock_irqrestore(&list->lock, flags);
1995 return result;
1996}
1997EXPORT_SYMBOL(skb_dequeue_tail);
1998
1999/**
2000 * skb_queue_purge - empty a list
2001 * @list: list to empty
2002 *
2003 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 * the list and one reference dropped. This function takes the list
2005 * lock and is atomic with respect to other list locking functions.
2006 */
2007void skb_queue_purge(struct sk_buff_head *list)
2008{
2009 struct sk_buff *skb;
2010 while ((skb = skb_dequeue(list)) != NULL)
2011 kfree_skb(skb);
2012}
2013EXPORT_SYMBOL(skb_queue_purge);
2014
2015/**
2016 * skb_queue_head - queue a buffer at the list head
2017 * @list: list to use
2018 * @newsk: buffer to queue
2019 *
2020 * Queue a buffer at the start of the list. This function takes the
2021 * list lock and can be used safely with other locking &sk_buff functions
2022 * safely.
2023 *
2024 * A buffer cannot be placed on two lists at the same time.
2025 */
2026void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2027{
2028 unsigned long flags;
2029
2030 spin_lock_irqsave(&list->lock, flags);
2031 __skb_queue_head(list, newsk);
2032 spin_unlock_irqrestore(&list->lock, flags);
2033}
2034EXPORT_SYMBOL(skb_queue_head);
2035
2036/**
2037 * skb_queue_tail - queue a buffer at the list tail
2038 * @list: list to use
2039 * @newsk: buffer to queue
2040 *
2041 * Queue a buffer at the tail of the list. This function takes the
2042 * list lock and can be used safely with other locking &sk_buff functions
2043 * safely.
2044 *
2045 * A buffer cannot be placed on two lists at the same time.
2046 */
2047void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2048{
2049 unsigned long flags;
2050
2051 spin_lock_irqsave(&list->lock, flags);
2052 __skb_queue_tail(list, newsk);
2053 spin_unlock_irqrestore(&list->lock, flags);
2054}
2055EXPORT_SYMBOL(skb_queue_tail);
2056
2057/**
2058 * skb_unlink - remove a buffer from a list
2059 * @skb: buffer to remove
2060 * @list: list to use
2061 *
2062 * Remove a packet from a list. The list locks are taken and this
2063 * function is atomic with respect to other list locked calls
2064 *
2065 * You must know what list the SKB is on.
2066 */
2067void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2068{
2069 unsigned long flags;
2070
2071 spin_lock_irqsave(&list->lock, flags);
2072 __skb_unlink(skb, list);
2073 spin_unlock_irqrestore(&list->lock, flags);
2074}
2075EXPORT_SYMBOL(skb_unlink);
2076
2077/**
2078 * skb_append - append a buffer
2079 * @old: buffer to insert after
2080 * @newsk: buffer to insert
2081 * @list: list to use
2082 *
2083 * Place a packet after a given packet in a list. The list locks are taken
2084 * and this function is atomic with respect to other list locked calls.
2085 * A buffer cannot be placed on two lists at the same time.
2086 */
2087void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2088{
2089 unsigned long flags;
2090
2091 spin_lock_irqsave(&list->lock, flags);
2092 __skb_queue_after(list, old, newsk);
2093 spin_unlock_irqrestore(&list->lock, flags);
2094}
2095EXPORT_SYMBOL(skb_append);
2096
2097/**
2098 * skb_insert - insert a buffer
2099 * @old: buffer to insert before
2100 * @newsk: buffer to insert
2101 * @list: list to use
2102 *
2103 * Place a packet before a given packet in a list. The list locks are
2104 * taken and this function is atomic with respect to other list locked
2105 * calls.
2106 *
2107 * A buffer cannot be placed on two lists at the same time.
2108 */
2109void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2110{
2111 unsigned long flags;
2112
2113 spin_lock_irqsave(&list->lock, flags);
2114 __skb_insert(newsk, old->prev, old, list);
2115 spin_unlock_irqrestore(&list->lock, flags);
2116}
2117EXPORT_SYMBOL(skb_insert);
2118
2119static inline void skb_split_inside_header(struct sk_buff *skb,
2120 struct sk_buff* skb1,
2121 const u32 len, const int pos)
2122{
2123 int i;
2124
2125 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2126 pos - len);
2127 /* And move data appendix as is. */
2128 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2129 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2130
2131 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2132 skb_shinfo(skb)->nr_frags = 0;
2133 skb1->data_len = skb->data_len;
2134 skb1->len += skb1->data_len;
2135 skb->data_len = 0;
2136 skb->len = len;
2137 skb_set_tail_pointer(skb, len);
2138}
2139
2140static inline void skb_split_no_header(struct sk_buff *skb,
2141 struct sk_buff* skb1,
2142 const u32 len, int pos)
2143{
2144 int i, k = 0;
2145 const int nfrags = skb_shinfo(skb)->nr_frags;
2146
2147 skb_shinfo(skb)->nr_frags = 0;
2148 skb1->len = skb1->data_len = skb->len - len;
2149 skb->len = len;
2150 skb->data_len = len - pos;
2151
2152 for (i = 0; i < nfrags; i++) {
2153 int size = skb_shinfo(skb)->frags[i].size;
2154
2155 if (pos + size > len) {
2156 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2157
2158 if (pos < len) {
2159 /* Split frag.
2160 * We have two variants in this case:
2161 * 1. Move all the frag to the second
2162 * part, if it is possible. F.e.
2163 * this approach is mandatory for TUX,
2164 * where splitting is expensive.
2165 * 2. Split is accurately. We make this.
2166 */
2167 get_page(skb_shinfo(skb)->frags[i].page);
2168 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2169 skb_shinfo(skb1)->frags[0].size -= len - pos;
2170 skb_shinfo(skb)->frags[i].size = len - pos;
2171 skb_shinfo(skb)->nr_frags++;
2172 }
2173 k++;
2174 } else
2175 skb_shinfo(skb)->nr_frags++;
2176 pos += size;
2177 }
2178 skb_shinfo(skb1)->nr_frags = k;
2179}
2180
2181/**
2182 * skb_split - Split fragmented skb to two parts at length len.
2183 * @skb: the buffer to split
2184 * @skb1: the buffer to receive the second part
2185 * @len: new length for skb
2186 */
2187void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2188{
2189 int pos = skb_headlen(skb);
2190
2191 if (len < pos) /* Split line is inside header. */
2192 skb_split_inside_header(skb, skb1, len, pos);
2193 else /* Second chunk has no header, nothing to copy. */
2194 skb_split_no_header(skb, skb1, len, pos);
2195}
2196EXPORT_SYMBOL(skb_split);
2197
2198/* Shifting from/to a cloned skb is a no-go.
2199 *
2200 * Caller cannot keep skb_shinfo related pointers past calling here!
2201 */
2202static int skb_prepare_for_shift(struct sk_buff *skb)
2203{
2204 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2205}
2206
2207/**
2208 * skb_shift - Shifts paged data partially from skb to another
2209 * @tgt: buffer into which tail data gets added
2210 * @skb: buffer from which the paged data comes from
2211 * @shiftlen: shift up to this many bytes
2212 *
2213 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2214 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2215 * It's up to caller to free skb if everything was shifted.
2216 *
2217 * If @tgt runs out of frags, the whole operation is aborted.
2218 *
2219 * Skb cannot include anything else but paged data while tgt is allowed
2220 * to have non-paged data as well.
2221 *
2222 * TODO: full sized shift could be optimized but that would need
2223 * specialized skb free'er to handle frags without up-to-date nr_frags.
2224 */
2225int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2226{
2227 int from, to, merge, todo;
2228 struct skb_frag_struct *fragfrom, *fragto;
2229
2230 BUG_ON(shiftlen > skb->len);
2231 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2232
2233 todo = shiftlen;
2234 from = 0;
2235 to = skb_shinfo(tgt)->nr_frags;
2236 fragfrom = &skb_shinfo(skb)->frags[from];
2237
2238 /* Actual merge is delayed until the point when we know we can
2239 * commit all, so that we don't have to undo partial changes
2240 */
2241 if (!to ||
2242 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2243 merge = -1;
2244 } else {
2245 merge = to - 1;
2246
2247 todo -= fragfrom->size;
2248 if (todo < 0) {
2249 if (skb_prepare_for_shift(skb) ||
2250 skb_prepare_for_shift(tgt))
2251 return 0;
2252
2253 /* All previous frag pointers might be stale! */
2254 fragfrom = &skb_shinfo(skb)->frags[from];
2255 fragto = &skb_shinfo(tgt)->frags[merge];
2256
2257 fragto->size += shiftlen;
2258 fragfrom->size -= shiftlen;
2259 fragfrom->page_offset += shiftlen;
2260
2261 goto onlymerged;
2262 }
2263
2264 from++;
2265 }
2266
2267 /* Skip full, not-fitting skb to avoid expensive operations */
2268 if ((shiftlen == skb->len) &&
2269 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2270 return 0;
2271
2272 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2273 return 0;
2274
2275 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2276 if (to == MAX_SKB_FRAGS)
2277 return 0;
2278
2279 fragfrom = &skb_shinfo(skb)->frags[from];
2280 fragto = &skb_shinfo(tgt)->frags[to];
2281
2282 if (todo >= fragfrom->size) {
2283 *fragto = *fragfrom;
2284 todo -= fragfrom->size;
2285 from++;
2286 to++;
2287
2288 } else {
2289 get_page(fragfrom->page);
2290 fragto->page = fragfrom->page;
2291 fragto->page_offset = fragfrom->page_offset;
2292 fragto->size = todo;
2293
2294 fragfrom->page_offset += todo;
2295 fragfrom->size -= todo;
2296 todo = 0;
2297
2298 to++;
2299 break;
2300 }
2301 }
2302
2303 /* Ready to "commit" this state change to tgt */
2304 skb_shinfo(tgt)->nr_frags = to;
2305
2306 if (merge >= 0) {
2307 fragfrom = &skb_shinfo(skb)->frags[0];
2308 fragto = &skb_shinfo(tgt)->frags[merge];
2309
2310 fragto->size += fragfrom->size;
2311 put_page(fragfrom->page);
2312 }
2313
2314 /* Reposition in the original skb */
2315 to = 0;
2316 while (from < skb_shinfo(skb)->nr_frags)
2317 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2318 skb_shinfo(skb)->nr_frags = to;
2319
2320 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2321
2322onlymerged:
2323 /* Most likely the tgt won't ever need its checksum anymore, skb on
2324 * the other hand might need it if it needs to be resent
2325 */
2326 tgt->ip_summed = CHECKSUM_PARTIAL;
2327 skb->ip_summed = CHECKSUM_PARTIAL;
2328
2329 /* Yak, is it really working this way? Some helper please? */
2330 skb->len -= shiftlen;
2331 skb->data_len -= shiftlen;
2332 skb->truesize -= shiftlen;
2333 tgt->len += shiftlen;
2334 tgt->data_len += shiftlen;
2335 tgt->truesize += shiftlen;
2336
2337 return shiftlen;
2338}
2339
2340/**
2341 * skb_prepare_seq_read - Prepare a sequential read of skb data
2342 * @skb: the buffer to read
2343 * @from: lower offset of data to be read
2344 * @to: upper offset of data to be read
2345 * @st: state variable
2346 *
2347 * Initializes the specified state variable. Must be called before
2348 * invoking skb_seq_read() for the first time.
2349 */
2350void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2351 unsigned int to, struct skb_seq_state *st)
2352{
2353 st->lower_offset = from;
2354 st->upper_offset = to;
2355 st->root_skb = st->cur_skb = skb;
2356 st->frag_idx = st->stepped_offset = 0;
2357 st->frag_data = NULL;
2358}
2359EXPORT_SYMBOL(skb_prepare_seq_read);
2360
2361/**
2362 * skb_seq_read - Sequentially read skb data
2363 * @consumed: number of bytes consumed by the caller so far
2364 * @data: destination pointer for data to be returned
2365 * @st: state variable
2366 *
2367 * Reads a block of skb data at &consumed relative to the
2368 * lower offset specified to skb_prepare_seq_read(). Assigns
2369 * the head of the data block to &data and returns the length
2370 * of the block or 0 if the end of the skb data or the upper
2371 * offset has been reached.
2372 *
2373 * The caller is not required to consume all of the data
2374 * returned, i.e. &consumed is typically set to the number
2375 * of bytes already consumed and the next call to
2376 * skb_seq_read() will return the remaining part of the block.
2377 *
2378 * Note 1: The size of each block of data returned can be arbitrary,
2379 * this limitation is the cost for zerocopy seqeuental
2380 * reads of potentially non linear data.
2381 *
2382 * Note 2: Fragment lists within fragments are not implemented
2383 * at the moment, state->root_skb could be replaced with
2384 * a stack for this purpose.
2385 */
2386unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2387 struct skb_seq_state *st)
2388{
2389 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2390 skb_frag_t *frag;
2391
2392 if (unlikely(abs_offset >= st->upper_offset))
2393 return 0;
2394
2395next_skb:
2396 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2397
2398 if (abs_offset < block_limit && !st->frag_data) {
2399 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2400 return block_limit - abs_offset;
2401 }
2402
2403 if (st->frag_idx == 0 && !st->frag_data)
2404 st->stepped_offset += skb_headlen(st->cur_skb);
2405
2406 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2407 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2408 block_limit = frag->size + st->stepped_offset;
2409
2410 if (abs_offset < block_limit) {
2411 if (!st->frag_data)
2412 st->frag_data = kmap_skb_frag(frag);
2413
2414 *data = (u8 *) st->frag_data + frag->page_offset +
2415 (abs_offset - st->stepped_offset);
2416
2417 return block_limit - abs_offset;
2418 }
2419
2420 if (st->frag_data) {
2421 kunmap_skb_frag(st->frag_data);
2422 st->frag_data = NULL;
2423 }
2424
2425 st->frag_idx++;
2426 st->stepped_offset += frag->size;
2427 }
2428
2429 if (st->frag_data) {
2430 kunmap_skb_frag(st->frag_data);
2431 st->frag_data = NULL;
2432 }
2433
2434 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2435 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2436 st->frag_idx = 0;
2437 goto next_skb;
2438 } else if (st->cur_skb->next) {
2439 st->cur_skb = st->cur_skb->next;
2440 st->frag_idx = 0;
2441 goto next_skb;
2442 }
2443
2444 return 0;
2445}
2446EXPORT_SYMBOL(skb_seq_read);
2447
2448/**
2449 * skb_abort_seq_read - Abort a sequential read of skb data
2450 * @st: state variable
2451 *
2452 * Must be called if skb_seq_read() was not called until it
2453 * returned 0.
2454 */
2455void skb_abort_seq_read(struct skb_seq_state *st)
2456{
2457 if (st->frag_data)
2458 kunmap_skb_frag(st->frag_data);
2459}
2460EXPORT_SYMBOL(skb_abort_seq_read);
2461
2462#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2463
2464static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2465 struct ts_config *conf,
2466 struct ts_state *state)
2467{
2468 return skb_seq_read(offset, text, TS_SKB_CB(state));
2469}
2470
2471static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2472{
2473 skb_abort_seq_read(TS_SKB_CB(state));
2474}
2475
2476/**
2477 * skb_find_text - Find a text pattern in skb data
2478 * @skb: the buffer to look in
2479 * @from: search offset
2480 * @to: search limit
2481 * @config: textsearch configuration
2482 * @state: uninitialized textsearch state variable
2483 *
2484 * Finds a pattern in the skb data according to the specified
2485 * textsearch configuration. Use textsearch_next() to retrieve
2486 * subsequent occurrences of the pattern. Returns the offset
2487 * to the first occurrence or UINT_MAX if no match was found.
2488 */
2489unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2490 unsigned int to, struct ts_config *config,
2491 struct ts_state *state)
2492{
2493 unsigned int ret;
2494
2495 config->get_next_block = skb_ts_get_next_block;
2496 config->finish = skb_ts_finish;
2497
2498 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2499
2500 ret = textsearch_find(config, state);
2501 return (ret <= to - from ? ret : UINT_MAX);
2502}
2503EXPORT_SYMBOL(skb_find_text);
2504
2505/**
2506 * skb_append_datato_frags: - append the user data to a skb
2507 * @sk: sock structure
2508 * @skb: skb structure to be appened with user data.
2509 * @getfrag: call back function to be used for getting the user data
2510 * @from: pointer to user message iov
2511 * @length: length of the iov message
2512 *
2513 * Description: This procedure append the user data in the fragment part
2514 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2515 */
2516int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2517 int (*getfrag)(void *from, char *to, int offset,
2518 int len, int odd, struct sk_buff *skb),
2519 void *from, int length)
2520{
2521 int frg_cnt = 0;
2522 skb_frag_t *frag = NULL;
2523 struct page *page = NULL;
2524 int copy, left;
2525 int offset = 0;
2526 int ret;
2527
2528 do {
2529 /* Return error if we don't have space for new frag */
2530 frg_cnt = skb_shinfo(skb)->nr_frags;
2531 if (frg_cnt >= MAX_SKB_FRAGS)
2532 return -EFAULT;
2533
2534 /* allocate a new page for next frag */
2535 page = alloc_pages(sk->sk_allocation, 0);
2536
2537 /* If alloc_page fails just return failure and caller will
2538 * free previous allocated pages by doing kfree_skb()
2539 */
2540 if (page == NULL)
2541 return -ENOMEM;
2542
2543 /* initialize the next frag */
2544 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2545 skb->truesize += PAGE_SIZE;
2546 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2547
2548 /* get the new initialized frag */
2549 frg_cnt = skb_shinfo(skb)->nr_frags;
2550 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2551
2552 /* copy the user data to page */
2553 left = PAGE_SIZE - frag->page_offset;
2554 copy = (length > left)? left : length;
2555
2556 ret = getfrag(from, (page_address(frag->page) +
2557 frag->page_offset + frag->size),
2558 offset, copy, 0, skb);
2559 if (ret < 0)
2560 return -EFAULT;
2561
2562 /* copy was successful so update the size parameters */
2563 frag->size += copy;
2564 skb->len += copy;
2565 skb->data_len += copy;
2566 offset += copy;
2567 length -= copy;
2568
2569 } while (length > 0);
2570
2571 return 0;
2572}
2573EXPORT_SYMBOL(skb_append_datato_frags);
2574
2575/**
2576 * skb_pull_rcsum - pull skb and update receive checksum
2577 * @skb: buffer to update
2578 * @len: length of data pulled
2579 *
2580 * This function performs an skb_pull on the packet and updates
2581 * the CHECKSUM_COMPLETE checksum. It should be used on
2582 * receive path processing instead of skb_pull unless you know
2583 * that the checksum difference is zero (e.g., a valid IP header)
2584 * or you are setting ip_summed to CHECKSUM_NONE.
2585 */
2586unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2587{
2588 BUG_ON(len > skb->len);
2589 skb->len -= len;
2590 BUG_ON(skb->len < skb->data_len);
2591 skb_postpull_rcsum(skb, skb->data, len);
2592 return skb->data += len;
2593}
2594EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2595
2596/**
2597 * skb_segment - Perform protocol segmentation on skb.
2598 * @skb: buffer to segment
2599 * @features: features for the output path (see dev->features)
2600 *
2601 * This function performs segmentation on the given skb. It returns
2602 * a pointer to the first in a list of new skbs for the segments.
2603 * In case of error it returns ERR_PTR(err).
2604 */
2605struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2606{
2607 struct sk_buff *segs = NULL;
2608 struct sk_buff *tail = NULL;
2609 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2610 unsigned int mss = skb_shinfo(skb)->gso_size;
2611 unsigned int doffset = skb->data - skb_mac_header(skb);
2612 unsigned int offset = doffset;
2613 unsigned int headroom;
2614 unsigned int len;
2615 int sg = !!(features & NETIF_F_SG);
2616 int nfrags = skb_shinfo(skb)->nr_frags;
2617 int err = -ENOMEM;
2618 int i = 0;
2619 int pos;
2620
2621 __skb_push(skb, doffset);
2622 headroom = skb_headroom(skb);
2623 pos = skb_headlen(skb);
2624
2625 do {
2626 struct sk_buff *nskb;
2627 skb_frag_t *frag;
2628 int hsize;
2629 int size;
2630
2631 len = skb->len - offset;
2632 if (len > mss)
2633 len = mss;
2634
2635 hsize = skb_headlen(skb) - offset;
2636 if (hsize < 0)
2637 hsize = 0;
2638 if (hsize > len || !sg)
2639 hsize = len;
2640
2641 if (!hsize && i >= nfrags) {
2642 BUG_ON(fskb->len != len);
2643
2644 pos += len;
2645 nskb = skb_clone(fskb, GFP_ATOMIC);
2646 fskb = fskb->next;
2647
2648 if (unlikely(!nskb))
2649 goto err;
2650
2651 hsize = skb_end_pointer(nskb) - nskb->head;
2652 if (skb_cow_head(nskb, doffset + headroom)) {
2653 kfree_skb(nskb);
2654 goto err;
2655 }
2656
2657 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2658 hsize;
2659 skb_release_head_state(nskb);
2660 __skb_push(nskb, doffset);
2661 } else {
2662 nskb = alloc_skb(hsize + doffset + headroom,
2663 GFP_ATOMIC);
2664
2665 if (unlikely(!nskb))
2666 goto err;
2667
2668 skb_reserve(nskb, headroom);
2669 __skb_put(nskb, doffset);
2670 }
2671
2672 if (segs)
2673 tail->next = nskb;
2674 else
2675 segs = nskb;
2676 tail = nskb;
2677
2678 __copy_skb_header(nskb, skb);
2679 nskb->mac_len = skb->mac_len;
2680
2681 /* nskb and skb might have different headroom */
2682 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2683 nskb->csum_start += skb_headroom(nskb) - headroom;
2684
2685 skb_reset_mac_header(nskb);
2686 skb_set_network_header(nskb, skb->mac_len);
2687 nskb->transport_header = (nskb->network_header +
2688 skb_network_header_len(skb));
2689 skb_copy_from_linear_data(skb, nskb->data, doffset);
2690
2691 if (fskb != skb_shinfo(skb)->frag_list)
2692 continue;
2693
2694 if (!sg) {
2695 nskb->ip_summed = CHECKSUM_NONE;
2696 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2697 skb_put(nskb, len),
2698 len, 0);
2699 continue;
2700 }
2701
2702 frag = skb_shinfo(nskb)->frags;
2703
2704 skb_copy_from_linear_data_offset(skb, offset,
2705 skb_put(nskb, hsize), hsize);
2706
2707 while (pos < offset + len && i < nfrags) {
2708 *frag = skb_shinfo(skb)->frags[i];
2709 get_page(frag->page);
2710 size = frag->size;
2711
2712 if (pos < offset) {
2713 frag->page_offset += offset - pos;
2714 frag->size -= offset - pos;
2715 }
2716
2717 skb_shinfo(nskb)->nr_frags++;
2718
2719 if (pos + size <= offset + len) {
2720 i++;
2721 pos += size;
2722 } else {
2723 frag->size -= pos + size - (offset + len);
2724 goto skip_fraglist;
2725 }
2726
2727 frag++;
2728 }
2729
2730 if (pos < offset + len) {
2731 struct sk_buff *fskb2 = fskb;
2732
2733 BUG_ON(pos + fskb->len != offset + len);
2734
2735 pos += fskb->len;
2736 fskb = fskb->next;
2737
2738 if (fskb2->next) {
2739 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2740 if (!fskb2)
2741 goto err;
2742 } else
2743 skb_get(fskb2);
2744
2745 SKB_FRAG_ASSERT(nskb);
2746 skb_shinfo(nskb)->frag_list = fskb2;
2747 }
2748
2749skip_fraglist:
2750 nskb->data_len = len - hsize;
2751 nskb->len += nskb->data_len;
2752 nskb->truesize += nskb->data_len;
2753 } while ((offset += len) < skb->len);
2754
2755 return segs;
2756
2757err:
2758 while ((skb = segs)) {
2759 segs = skb->next;
2760 kfree_skb(skb);
2761 }
2762 return ERR_PTR(err);
2763}
2764EXPORT_SYMBOL_GPL(skb_segment);
2765
2766int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2767{
2768 struct sk_buff *p = *head;
2769 struct sk_buff *nskb;
2770 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2771 struct skb_shared_info *pinfo = skb_shinfo(p);
2772 unsigned int headroom;
2773 unsigned int len = skb_gro_len(skb);
2774 unsigned int offset = skb_gro_offset(skb);
2775 unsigned int headlen = skb_headlen(skb);
2776
2777 if (p->len + len >= 65536)
2778 return -E2BIG;
2779
2780 if (pinfo->frag_list)
2781 goto merge;
2782 else if (headlen <= offset) {
2783 skb_frag_t *frag;
2784 skb_frag_t *frag2;
2785 int i = skbinfo->nr_frags;
2786 int nr_frags = pinfo->nr_frags + i;
2787
2788 offset -= headlen;
2789
2790 if (nr_frags > MAX_SKB_FRAGS)
2791 return -E2BIG;
2792
2793 pinfo->nr_frags = nr_frags;
2794 skbinfo->nr_frags = 0;
2795
2796 frag = pinfo->frags + nr_frags;
2797 frag2 = skbinfo->frags + i;
2798 do {
2799 *--frag = *--frag2;
2800 } while (--i);
2801
2802 frag->page_offset += offset;
2803 frag->size -= offset;
2804
2805 skb->truesize -= skb->data_len;
2806 skb->len -= skb->data_len;
2807 skb->data_len = 0;
2808
2809 NAPI_GRO_CB(skb)->free = 1;
2810 goto done;
2811 } else if (skb_gro_len(p) != pinfo->gso_size)
2812 return -E2BIG;
2813
2814 headroom = skb_headroom(p);
2815 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2816 if (unlikely(!nskb))
2817 return -ENOMEM;
2818
2819 __copy_skb_header(nskb, p);
2820 nskb->mac_len = p->mac_len;
2821
2822 skb_reserve(nskb, headroom);
2823 __skb_put(nskb, skb_gro_offset(p));
2824
2825 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2826 skb_set_network_header(nskb, skb_network_offset(p));
2827 skb_set_transport_header(nskb, skb_transport_offset(p));
2828
2829 __skb_pull(p, skb_gro_offset(p));
2830 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2831 p->data - skb_mac_header(p));
2832
2833 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2834 skb_shinfo(nskb)->frag_list = p;
2835 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2836 pinfo->gso_size = 0;
2837 skb_header_release(p);
2838 nskb->prev = p;
2839
2840 nskb->data_len += p->len;
2841 nskb->truesize += p->len;
2842 nskb->len += p->len;
2843
2844 *head = nskb;
2845 nskb->next = p->next;
2846 p->next = NULL;
2847
2848 p = nskb;
2849
2850merge:
2851 if (offset > headlen) {
2852 unsigned int eat = offset - headlen;
2853
2854 skbinfo->frags[0].page_offset += eat;
2855 skbinfo->frags[0].size -= eat;
2856 skb->data_len -= eat;
2857 skb->len -= eat;
2858 offset = headlen;
2859 }
2860
2861 __skb_pull(skb, offset);
2862
2863 p->prev->next = skb;
2864 p->prev = skb;
2865 skb_header_release(skb);
2866
2867done:
2868 NAPI_GRO_CB(p)->count++;
2869 p->data_len += len;
2870 p->truesize += len;
2871 p->len += len;
2872
2873 NAPI_GRO_CB(skb)->same_flow = 1;
2874 return 0;
2875}
2876EXPORT_SYMBOL_GPL(skb_gro_receive);
2877
2878void __init skb_init(void)
2879{
2880 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2881 sizeof(struct sk_buff),
2882 0,
2883 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2884 NULL);
2885 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2886 (2*sizeof(struct sk_buff)) +
2887 sizeof(atomic_t),
2888 0,
2889 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2890 NULL);
2891}
2892
2893/**
2894 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2895 * @skb: Socket buffer containing the buffers to be mapped
2896 * @sg: The scatter-gather list to map into
2897 * @offset: The offset into the buffer's contents to start mapping
2898 * @len: Length of buffer space to be mapped
2899 *
2900 * Fill the specified scatter-gather list with mappings/pointers into a
2901 * region of the buffer space attached to a socket buffer.
2902 */
2903static int
2904__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2905{
2906 int start = skb_headlen(skb);
2907 int i, copy = start - offset;
2908 struct sk_buff *frag_iter;
2909 int elt = 0;
2910
2911 if (copy > 0) {
2912 if (copy > len)
2913 copy = len;
2914 sg_set_buf(sg, skb->data + offset, copy);
2915 elt++;
2916 if ((len -= copy) == 0)
2917 return elt;
2918 offset += copy;
2919 }
2920
2921 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2922 int end;
2923
2924 WARN_ON(start > offset + len);
2925
2926 end = start + skb_shinfo(skb)->frags[i].size;
2927 if ((copy = end - offset) > 0) {
2928 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2929
2930 if (copy > len)
2931 copy = len;
2932 sg_set_page(&sg[elt], frag->page, copy,
2933 frag->page_offset+offset-start);
2934 elt++;
2935 if (!(len -= copy))
2936 return elt;
2937 offset += copy;
2938 }
2939 start = end;
2940 }
2941
2942 skb_walk_frags(skb, frag_iter) {
2943 int end;
2944
2945 WARN_ON(start > offset + len);
2946
2947 end = start + frag_iter->len;
2948 if ((copy = end - offset) > 0) {
2949 if (copy > len)
2950 copy = len;
2951 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2952 copy);
2953 if ((len -= copy) == 0)
2954 return elt;
2955 offset += copy;
2956 }
2957 start = end;
2958 }
2959 BUG_ON(len);
2960 return elt;
2961}
2962
2963int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2964{
2965 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2966
2967 sg_mark_end(&sg[nsg - 1]);
2968
2969 return nsg;
2970}
2971EXPORT_SYMBOL_GPL(skb_to_sgvec);
2972
2973/**
2974 * skb_cow_data - Check that a socket buffer's data buffers are writable
2975 * @skb: The socket buffer to check.
2976 * @tailbits: Amount of trailing space to be added
2977 * @trailer: Returned pointer to the skb where the @tailbits space begins
2978 *
2979 * Make sure that the data buffers attached to a socket buffer are
2980 * writable. If they are not, private copies are made of the data buffers
2981 * and the socket buffer is set to use these instead.
2982 *
2983 * If @tailbits is given, make sure that there is space to write @tailbits
2984 * bytes of data beyond current end of socket buffer. @trailer will be
2985 * set to point to the skb in which this space begins.
2986 *
2987 * The number of scatterlist elements required to completely map the
2988 * COW'd and extended socket buffer will be returned.
2989 */
2990int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2991{
2992 int copyflag;
2993 int elt;
2994 struct sk_buff *skb1, **skb_p;
2995
2996 /* If skb is cloned or its head is paged, reallocate
2997 * head pulling out all the pages (pages are considered not writable
2998 * at the moment even if they are anonymous).
2999 */
3000 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3001 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3002 return -ENOMEM;
3003
3004 /* Easy case. Most of packets will go this way. */
3005 if (!skb_has_frag_list(skb)) {
3006 /* A little of trouble, not enough of space for trailer.
3007 * This should not happen, when stack is tuned to generate
3008 * good frames. OK, on miss we reallocate and reserve even more
3009 * space, 128 bytes is fair. */
3010
3011 if (skb_tailroom(skb) < tailbits &&
3012 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3013 return -ENOMEM;
3014
3015 /* Voila! */
3016 *trailer = skb;
3017 return 1;
3018 }
3019
3020 /* Misery. We are in troubles, going to mincer fragments... */
3021
3022 elt = 1;
3023 skb_p = &skb_shinfo(skb)->frag_list;
3024 copyflag = 0;
3025
3026 while ((skb1 = *skb_p) != NULL) {
3027 int ntail = 0;
3028
3029 /* The fragment is partially pulled by someone,
3030 * this can happen on input. Copy it and everything
3031 * after it. */
3032
3033 if (skb_shared(skb1))
3034 copyflag = 1;
3035
3036 /* If the skb is the last, worry about trailer. */
3037
3038 if (skb1->next == NULL && tailbits) {
3039 if (skb_shinfo(skb1)->nr_frags ||
3040 skb_has_frag_list(skb1) ||
3041 skb_tailroom(skb1) < tailbits)
3042 ntail = tailbits + 128;
3043 }
3044
3045 if (copyflag ||
3046 skb_cloned(skb1) ||
3047 ntail ||
3048 skb_shinfo(skb1)->nr_frags ||
3049 skb_has_frag_list(skb1)) {
3050 struct sk_buff *skb2;
3051
3052 /* Fuck, we are miserable poor guys... */
3053 if (ntail == 0)
3054 skb2 = skb_copy(skb1, GFP_ATOMIC);
3055 else
3056 skb2 = skb_copy_expand(skb1,
3057 skb_headroom(skb1),
3058 ntail,
3059 GFP_ATOMIC);
3060 if (unlikely(skb2 == NULL))
3061 return -ENOMEM;
3062
3063 if (skb1->sk)
3064 skb_set_owner_w(skb2, skb1->sk);
3065
3066 /* Looking around. Are we still alive?
3067 * OK, link new skb, drop old one */
3068
3069 skb2->next = skb1->next;
3070 *skb_p = skb2;
3071 kfree_skb(skb1);
3072 skb1 = skb2;
3073 }
3074 elt++;
3075 *trailer = skb1;
3076 skb_p = &skb1->next;
3077 }
3078
3079 return elt;
3080}
3081EXPORT_SYMBOL_GPL(skb_cow_data);
3082
3083static void sock_rmem_free(struct sk_buff *skb)
3084{
3085 struct sock *sk = skb->sk;
3086
3087 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3088}
3089
3090/*
3091 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3092 */
3093int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3094{
3095 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3096 (unsigned)sk->sk_rcvbuf)
3097 return -ENOMEM;
3098
3099 skb_orphan(skb);
3100 skb->sk = sk;
3101 skb->destructor = sock_rmem_free;
3102 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3103
3104 /* before exiting rcu section, make sure dst is refcounted */
3105 skb_dst_force(skb);
3106
3107 skb_queue_tail(&sk->sk_error_queue, skb);
3108 if (!sock_flag(sk, SOCK_DEAD))
3109 sk->sk_data_ready(sk, skb->len);
3110 return 0;
3111}
3112EXPORT_SYMBOL(sock_queue_err_skb);
3113
3114void skb_tstamp_tx(struct sk_buff *orig_skb,
3115 struct skb_shared_hwtstamps *hwtstamps)
3116{
3117 struct sock *sk = orig_skb->sk;
3118 struct sock_exterr_skb *serr;
3119 struct sk_buff *skb;
3120 int err;
3121
3122 if (!sk)
3123 return;
3124
3125 skb = skb_clone(orig_skb, GFP_ATOMIC);
3126 if (!skb)
3127 return;
3128
3129 if (hwtstamps) {
3130 *skb_hwtstamps(skb) =
3131 *hwtstamps;
3132 } else {
3133 /*
3134 * no hardware time stamps available,
3135 * so keep the shared tx_flags and only
3136 * store software time stamp
3137 */
3138 skb->tstamp = ktime_get_real();
3139 }
3140
3141 serr = SKB_EXT_ERR(skb);
3142 memset(serr, 0, sizeof(*serr));
3143 serr->ee.ee_errno = ENOMSG;
3144 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3145
3146 err = sock_queue_err_skb(sk, skb);
3147
3148 if (err)
3149 kfree_skb(skb);
3150}
3151EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3152
3153
3154/**
3155 * skb_partial_csum_set - set up and verify partial csum values for packet
3156 * @skb: the skb to set
3157 * @start: the number of bytes after skb->data to start checksumming.
3158 * @off: the offset from start to place the checksum.
3159 *
3160 * For untrusted partially-checksummed packets, we need to make sure the values
3161 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3162 *
3163 * This function checks and sets those values and skb->ip_summed: if this
3164 * returns false you should drop the packet.
3165 */
3166bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3167{
3168 if (unlikely(start > skb_headlen(skb)) ||
3169 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3170 if (net_ratelimit())
3171 printk(KERN_WARNING
3172 "bad partial csum: csum=%u/%u len=%u\n",
3173 start, off, skb_headlen(skb));
3174 return false;
3175 }
3176 skb->ip_summed = CHECKSUM_PARTIAL;
3177 skb->csum_start = skb_headroom(skb) + start;
3178 skb->csum_offset = off;
3179 return true;
3180}
3181EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3182
3183void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3184{
3185 if (net_ratelimit())
3186 pr_warning("%s: received packets cannot be forwarded"
3187 " while LRO is enabled\n", skb->dev->name);
3188}
3189EXPORT_SYMBOL(__skb_warn_lro_forwarding);
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Routines having to do with the 'struct sk_buff' memory handlers.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 *
8 * Fixes:
9 * Alan Cox : Fixed the worst of the load
10 * balancer bugs.
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
23 *
24 * NOTE:
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
29 */
30
31/*
32 * The functions in this file will not compile correctly with gcc 2.4.x
33 */
34
35#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36
37#include <linux/module.h>
38#include <linux/types.h>
39#include <linux/kernel.h>
40#include <linux/mm.h>
41#include <linux/interrupt.h>
42#include <linux/in.h>
43#include <linux/inet.h>
44#include <linux/slab.h>
45#include <linux/tcp.h>
46#include <linux/udp.h>
47#include <linux/sctp.h>
48#include <linux/netdevice.h>
49#ifdef CONFIG_NET_CLS_ACT
50#include <net/pkt_sched.h>
51#endif
52#include <linux/string.h>
53#include <linux/skbuff.h>
54#include <linux/splice.h>
55#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
58#include <linux/scatterlist.h>
59#include <linux/errqueue.h>
60#include <linux/prefetch.h>
61#include <linux/if_vlan.h>
62#include <linux/mpls.h>
63
64#include <net/protocol.h>
65#include <net/dst.h>
66#include <net/sock.h>
67#include <net/checksum.h>
68#include <net/ip6_checksum.h>
69#include <net/xfrm.h>
70#include <net/mpls.h>
71
72#include <linux/uaccess.h>
73#include <trace/events/skb.h>
74#include <linux/highmem.h>
75#include <linux/capability.h>
76#include <linux/user_namespace.h>
77#include <linux/indirect_call_wrapper.h>
78
79#include "datagram.h"
80
81struct kmem_cache *skbuff_head_cache __ro_after_init;
82static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
83#ifdef CONFIG_SKB_EXTENSIONS
84static struct kmem_cache *skbuff_ext_cache __ro_after_init;
85#endif
86int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
87EXPORT_SYMBOL(sysctl_max_skb_frags);
88
89/**
90 * skb_panic - private function for out-of-line support
91 * @skb: buffer
92 * @sz: size
93 * @addr: address
94 * @msg: skb_over_panic or skb_under_panic
95 *
96 * Out-of-line support for skb_put() and skb_push().
97 * Called via the wrapper skb_over_panic() or skb_under_panic().
98 * Keep out of line to prevent kernel bloat.
99 * __builtin_return_address is not used because it is not always reliable.
100 */
101static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
102 const char msg[])
103{
104 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
105 msg, addr, skb->len, sz, skb->head, skb->data,
106 (unsigned long)skb->tail, (unsigned long)skb->end,
107 skb->dev ? skb->dev->name : "<NULL>");
108 BUG();
109}
110
111static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
112{
113 skb_panic(skb, sz, addr, __func__);
114}
115
116static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
117{
118 skb_panic(skb, sz, addr, __func__);
119}
120
121/*
122 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
123 * the caller if emergency pfmemalloc reserves are being used. If it is and
124 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
125 * may be used. Otherwise, the packet data may be discarded until enough
126 * memory is free
127 */
128#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
129 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
130
131static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
132 unsigned long ip, bool *pfmemalloc)
133{
134 void *obj;
135 bool ret_pfmemalloc = false;
136
137 /*
138 * Try a regular allocation, when that fails and we're not entitled
139 * to the reserves, fail.
140 */
141 obj = kmalloc_node_track_caller(size,
142 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
143 node);
144 if (obj || !(gfp_pfmemalloc_allowed(flags)))
145 goto out;
146
147 /* Try again but now we are using pfmemalloc reserves */
148 ret_pfmemalloc = true;
149 obj = kmalloc_node_track_caller(size, flags, node);
150
151out:
152 if (pfmemalloc)
153 *pfmemalloc = ret_pfmemalloc;
154
155 return obj;
156}
157
158/* Allocate a new skbuff. We do this ourselves so we can fill in a few
159 * 'private' fields and also do memory statistics to find all the
160 * [BEEP] leaks.
161 *
162 */
163
164/**
165 * __alloc_skb - allocate a network buffer
166 * @size: size to allocate
167 * @gfp_mask: allocation mask
168 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
169 * instead of head cache and allocate a cloned (child) skb.
170 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
171 * allocations in case the data is required for writeback
172 * @node: numa node to allocate memory on
173 *
174 * Allocate a new &sk_buff. The returned buffer has no headroom and a
175 * tail room of at least size bytes. The object has a reference count
176 * of one. The return is the buffer. On a failure the return is %NULL.
177 *
178 * Buffers may only be allocated from interrupts using a @gfp_mask of
179 * %GFP_ATOMIC.
180 */
181struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
182 int flags, int node)
183{
184 struct kmem_cache *cache;
185 struct skb_shared_info *shinfo;
186 struct sk_buff *skb;
187 u8 *data;
188 bool pfmemalloc;
189
190 cache = (flags & SKB_ALLOC_FCLONE)
191 ? skbuff_fclone_cache : skbuff_head_cache;
192
193 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
194 gfp_mask |= __GFP_MEMALLOC;
195
196 /* Get the HEAD */
197 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
198 if (!skb)
199 goto out;
200 prefetchw(skb);
201
202 /* We do our best to align skb_shared_info on a separate cache
203 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
204 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
205 * Both skb->head and skb_shared_info are cache line aligned.
206 */
207 size = SKB_DATA_ALIGN(size);
208 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
209 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
210 if (!data)
211 goto nodata;
212 /* kmalloc(size) might give us more room than requested.
213 * Put skb_shared_info exactly at the end of allocated zone,
214 * to allow max possible filling before reallocation.
215 */
216 size = SKB_WITH_OVERHEAD(ksize(data));
217 prefetchw(data + size);
218
219 /*
220 * Only clear those fields we need to clear, not those that we will
221 * actually initialise below. Hence, don't put any more fields after
222 * the tail pointer in struct sk_buff!
223 */
224 memset(skb, 0, offsetof(struct sk_buff, tail));
225 /* Account for allocated memory : skb + skb->head */
226 skb->truesize = SKB_TRUESIZE(size);
227 skb->pfmemalloc = pfmemalloc;
228 refcount_set(&skb->users, 1);
229 skb->head = data;
230 skb->data = data;
231 skb_reset_tail_pointer(skb);
232 skb->end = skb->tail + size;
233 skb->mac_header = (typeof(skb->mac_header))~0U;
234 skb->transport_header = (typeof(skb->transport_header))~0U;
235
236 /* make sure we initialize shinfo sequentially */
237 shinfo = skb_shinfo(skb);
238 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
239 atomic_set(&shinfo->dataref, 1);
240
241 if (flags & SKB_ALLOC_FCLONE) {
242 struct sk_buff_fclones *fclones;
243
244 fclones = container_of(skb, struct sk_buff_fclones, skb1);
245
246 skb->fclone = SKB_FCLONE_ORIG;
247 refcount_set(&fclones->fclone_ref, 1);
248
249 fclones->skb2.fclone = SKB_FCLONE_CLONE;
250 }
251out:
252 return skb;
253nodata:
254 kmem_cache_free(cache, skb);
255 skb = NULL;
256 goto out;
257}
258EXPORT_SYMBOL(__alloc_skb);
259
260/* Caller must provide SKB that is memset cleared */
261static struct sk_buff *__build_skb_around(struct sk_buff *skb,
262 void *data, unsigned int frag_size)
263{
264 struct skb_shared_info *shinfo;
265 unsigned int size = frag_size ? : ksize(data);
266
267 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
268
269 /* Assumes caller memset cleared SKB */
270 skb->truesize = SKB_TRUESIZE(size);
271 refcount_set(&skb->users, 1);
272 skb->head = data;
273 skb->data = data;
274 skb_reset_tail_pointer(skb);
275 skb->end = skb->tail + size;
276 skb->mac_header = (typeof(skb->mac_header))~0U;
277 skb->transport_header = (typeof(skb->transport_header))~0U;
278
279 /* make sure we initialize shinfo sequentially */
280 shinfo = skb_shinfo(skb);
281 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
282 atomic_set(&shinfo->dataref, 1);
283
284 return skb;
285}
286
287/**
288 * __build_skb - build a network buffer
289 * @data: data buffer provided by caller
290 * @frag_size: size of data, or 0 if head was kmalloced
291 *
292 * Allocate a new &sk_buff. Caller provides space holding head and
293 * skb_shared_info. @data must have been allocated by kmalloc() only if
294 * @frag_size is 0, otherwise data should come from the page allocator
295 * or vmalloc()
296 * The return is the new skb buffer.
297 * On a failure the return is %NULL, and @data is not freed.
298 * Notes :
299 * Before IO, driver allocates only data buffer where NIC put incoming frame
300 * Driver should add room at head (NET_SKB_PAD) and
301 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
302 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
303 * before giving packet to stack.
304 * RX rings only contains data buffers, not full skbs.
305 */
306struct sk_buff *__build_skb(void *data, unsigned int frag_size)
307{
308 struct sk_buff *skb;
309
310 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
311 if (unlikely(!skb))
312 return NULL;
313
314 memset(skb, 0, offsetof(struct sk_buff, tail));
315
316 return __build_skb_around(skb, data, frag_size);
317}
318
319/* build_skb() is wrapper over __build_skb(), that specifically
320 * takes care of skb->head and skb->pfmemalloc
321 * This means that if @frag_size is not zero, then @data must be backed
322 * by a page fragment, not kmalloc() or vmalloc()
323 */
324struct sk_buff *build_skb(void *data, unsigned int frag_size)
325{
326 struct sk_buff *skb = __build_skb(data, frag_size);
327
328 if (skb && frag_size) {
329 skb->head_frag = 1;
330 if (page_is_pfmemalloc(virt_to_head_page(data)))
331 skb->pfmemalloc = 1;
332 }
333 return skb;
334}
335EXPORT_SYMBOL(build_skb);
336
337/**
338 * build_skb_around - build a network buffer around provided skb
339 * @skb: sk_buff provide by caller, must be memset cleared
340 * @data: data buffer provided by caller
341 * @frag_size: size of data, or 0 if head was kmalloced
342 */
343struct sk_buff *build_skb_around(struct sk_buff *skb,
344 void *data, unsigned int frag_size)
345{
346 if (unlikely(!skb))
347 return NULL;
348
349 skb = __build_skb_around(skb, data, frag_size);
350
351 if (skb && frag_size) {
352 skb->head_frag = 1;
353 if (page_is_pfmemalloc(virt_to_head_page(data)))
354 skb->pfmemalloc = 1;
355 }
356 return skb;
357}
358EXPORT_SYMBOL(build_skb_around);
359
360#define NAPI_SKB_CACHE_SIZE 64
361
362struct napi_alloc_cache {
363 struct page_frag_cache page;
364 unsigned int skb_count;
365 void *skb_cache[NAPI_SKB_CACHE_SIZE];
366};
367
368static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
369static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
370
371static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
372{
373 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
374
375 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
376}
377
378void *napi_alloc_frag(unsigned int fragsz)
379{
380 fragsz = SKB_DATA_ALIGN(fragsz);
381
382 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
383}
384EXPORT_SYMBOL(napi_alloc_frag);
385
386/**
387 * netdev_alloc_frag - allocate a page fragment
388 * @fragsz: fragment size
389 *
390 * Allocates a frag from a page for receive buffer.
391 * Uses GFP_ATOMIC allocations.
392 */
393void *netdev_alloc_frag(unsigned int fragsz)
394{
395 struct page_frag_cache *nc;
396 void *data;
397
398 fragsz = SKB_DATA_ALIGN(fragsz);
399 if (in_irq() || irqs_disabled()) {
400 nc = this_cpu_ptr(&netdev_alloc_cache);
401 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
402 } else {
403 local_bh_disable();
404 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
405 local_bh_enable();
406 }
407 return data;
408}
409EXPORT_SYMBOL(netdev_alloc_frag);
410
411/**
412 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
413 * @dev: network device to receive on
414 * @len: length to allocate
415 * @gfp_mask: get_free_pages mask, passed to alloc_skb
416 *
417 * Allocate a new &sk_buff and assign it a usage count of one. The
418 * buffer has NET_SKB_PAD headroom built in. Users should allocate
419 * the headroom they think they need without accounting for the
420 * built in space. The built in space is used for optimisations.
421 *
422 * %NULL is returned if there is no free memory.
423 */
424struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
425 gfp_t gfp_mask)
426{
427 struct page_frag_cache *nc;
428 struct sk_buff *skb;
429 bool pfmemalloc;
430 void *data;
431
432 len += NET_SKB_PAD;
433
434 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
435 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
436 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
437 if (!skb)
438 goto skb_fail;
439 goto skb_success;
440 }
441
442 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
443 len = SKB_DATA_ALIGN(len);
444
445 if (sk_memalloc_socks())
446 gfp_mask |= __GFP_MEMALLOC;
447
448 if (in_irq() || irqs_disabled()) {
449 nc = this_cpu_ptr(&netdev_alloc_cache);
450 data = page_frag_alloc(nc, len, gfp_mask);
451 pfmemalloc = nc->pfmemalloc;
452 } else {
453 local_bh_disable();
454 nc = this_cpu_ptr(&napi_alloc_cache.page);
455 data = page_frag_alloc(nc, len, gfp_mask);
456 pfmemalloc = nc->pfmemalloc;
457 local_bh_enable();
458 }
459
460 if (unlikely(!data))
461 return NULL;
462
463 skb = __build_skb(data, len);
464 if (unlikely(!skb)) {
465 skb_free_frag(data);
466 return NULL;
467 }
468
469 /* use OR instead of assignment to avoid clearing of bits in mask */
470 if (pfmemalloc)
471 skb->pfmemalloc = 1;
472 skb->head_frag = 1;
473
474skb_success:
475 skb_reserve(skb, NET_SKB_PAD);
476 skb->dev = dev;
477
478skb_fail:
479 return skb;
480}
481EXPORT_SYMBOL(__netdev_alloc_skb);
482
483/**
484 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
485 * @napi: napi instance this buffer was allocated for
486 * @len: length to allocate
487 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
488 *
489 * Allocate a new sk_buff for use in NAPI receive. This buffer will
490 * attempt to allocate the head from a special reserved region used
491 * only for NAPI Rx allocation. By doing this we can save several
492 * CPU cycles by avoiding having to disable and re-enable IRQs.
493 *
494 * %NULL is returned if there is no free memory.
495 */
496struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
497 gfp_t gfp_mask)
498{
499 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
500 struct sk_buff *skb;
501 void *data;
502
503 len += NET_SKB_PAD + NET_IP_ALIGN;
504
505 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
506 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
507 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
508 if (!skb)
509 goto skb_fail;
510 goto skb_success;
511 }
512
513 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
514 len = SKB_DATA_ALIGN(len);
515
516 if (sk_memalloc_socks())
517 gfp_mask |= __GFP_MEMALLOC;
518
519 data = page_frag_alloc(&nc->page, len, gfp_mask);
520 if (unlikely(!data))
521 return NULL;
522
523 skb = __build_skb(data, len);
524 if (unlikely(!skb)) {
525 skb_free_frag(data);
526 return NULL;
527 }
528
529 /* use OR instead of assignment to avoid clearing of bits in mask */
530 if (nc->page.pfmemalloc)
531 skb->pfmemalloc = 1;
532 skb->head_frag = 1;
533
534skb_success:
535 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
536 skb->dev = napi->dev;
537
538skb_fail:
539 return skb;
540}
541EXPORT_SYMBOL(__napi_alloc_skb);
542
543void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
544 int size, unsigned int truesize)
545{
546 skb_fill_page_desc(skb, i, page, off, size);
547 skb->len += size;
548 skb->data_len += size;
549 skb->truesize += truesize;
550}
551EXPORT_SYMBOL(skb_add_rx_frag);
552
553void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
554 unsigned int truesize)
555{
556 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
557
558 skb_frag_size_add(frag, size);
559 skb->len += size;
560 skb->data_len += size;
561 skb->truesize += truesize;
562}
563EXPORT_SYMBOL(skb_coalesce_rx_frag);
564
565static void skb_drop_list(struct sk_buff **listp)
566{
567 kfree_skb_list(*listp);
568 *listp = NULL;
569}
570
571static inline void skb_drop_fraglist(struct sk_buff *skb)
572{
573 skb_drop_list(&skb_shinfo(skb)->frag_list);
574}
575
576static void skb_clone_fraglist(struct sk_buff *skb)
577{
578 struct sk_buff *list;
579
580 skb_walk_frags(skb, list)
581 skb_get(list);
582}
583
584static void skb_free_head(struct sk_buff *skb)
585{
586 unsigned char *head = skb->head;
587
588 if (skb->head_frag)
589 skb_free_frag(head);
590 else
591 kfree(head);
592}
593
594static void skb_release_data(struct sk_buff *skb)
595{
596 struct skb_shared_info *shinfo = skb_shinfo(skb);
597 int i;
598
599 if (skb->cloned &&
600 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
601 &shinfo->dataref))
602 return;
603
604 for (i = 0; i < shinfo->nr_frags; i++)
605 __skb_frag_unref(&shinfo->frags[i]);
606
607 if (shinfo->frag_list)
608 kfree_skb_list(shinfo->frag_list);
609
610 skb_zcopy_clear(skb, true);
611 skb_free_head(skb);
612}
613
614/*
615 * Free an skbuff by memory without cleaning the state.
616 */
617static void kfree_skbmem(struct sk_buff *skb)
618{
619 struct sk_buff_fclones *fclones;
620
621 switch (skb->fclone) {
622 case SKB_FCLONE_UNAVAILABLE:
623 kmem_cache_free(skbuff_head_cache, skb);
624 return;
625
626 case SKB_FCLONE_ORIG:
627 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628
629 /* We usually free the clone (TX completion) before original skb
630 * This test would have no chance to be true for the clone,
631 * while here, branch prediction will be good.
632 */
633 if (refcount_read(&fclones->fclone_ref) == 1)
634 goto fastpath;
635 break;
636
637 default: /* SKB_FCLONE_CLONE */
638 fclones = container_of(skb, struct sk_buff_fclones, skb2);
639 break;
640 }
641 if (!refcount_dec_and_test(&fclones->fclone_ref))
642 return;
643fastpath:
644 kmem_cache_free(skbuff_fclone_cache, fclones);
645}
646
647void skb_release_head_state(struct sk_buff *skb)
648{
649 skb_dst_drop(skb);
650 if (skb->destructor) {
651 WARN_ON(in_irq());
652 skb->destructor(skb);
653 }
654#if IS_ENABLED(CONFIG_NF_CONNTRACK)
655 nf_conntrack_put(skb_nfct(skb));
656#endif
657 skb_ext_put(skb);
658}
659
660/* Free everything but the sk_buff shell. */
661static void skb_release_all(struct sk_buff *skb)
662{
663 skb_release_head_state(skb);
664 if (likely(skb->head))
665 skb_release_data(skb);
666}
667
668/**
669 * __kfree_skb - private function
670 * @skb: buffer
671 *
672 * Free an sk_buff. Release anything attached to the buffer.
673 * Clean the state. This is an internal helper function. Users should
674 * always call kfree_skb
675 */
676
677void __kfree_skb(struct sk_buff *skb)
678{
679 skb_release_all(skb);
680 kfree_skbmem(skb);
681}
682EXPORT_SYMBOL(__kfree_skb);
683
684/**
685 * kfree_skb - free an sk_buff
686 * @skb: buffer to free
687 *
688 * Drop a reference to the buffer and free it if the usage count has
689 * hit zero.
690 */
691void kfree_skb(struct sk_buff *skb)
692{
693 if (!skb_unref(skb))
694 return;
695
696 trace_kfree_skb(skb, __builtin_return_address(0));
697 __kfree_skb(skb);
698}
699EXPORT_SYMBOL(kfree_skb);
700
701void kfree_skb_list(struct sk_buff *segs)
702{
703 while (segs) {
704 struct sk_buff *next = segs->next;
705
706 kfree_skb(segs);
707 segs = next;
708 }
709}
710EXPORT_SYMBOL(kfree_skb_list);
711
712/* Dump skb information and contents.
713 *
714 * Must only be called from net_ratelimit()-ed paths.
715 *
716 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
717 */
718void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
719{
720 static atomic_t can_dump_full = ATOMIC_INIT(5);
721 struct skb_shared_info *sh = skb_shinfo(skb);
722 struct net_device *dev = skb->dev;
723 struct sock *sk = skb->sk;
724 struct sk_buff *list_skb;
725 bool has_mac, has_trans;
726 int headroom, tailroom;
727 int i, len, seg_len;
728
729 if (full_pkt)
730 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
731
732 if (full_pkt)
733 len = skb->len;
734 else
735 len = min_t(int, skb->len, MAX_HEADER + 128);
736
737 headroom = skb_headroom(skb);
738 tailroom = skb_tailroom(skb);
739
740 has_mac = skb_mac_header_was_set(skb);
741 has_trans = skb_transport_header_was_set(skb);
742
743 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
744 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
745 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
746 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
747 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
748 level, skb->len, headroom, skb_headlen(skb), tailroom,
749 has_mac ? skb->mac_header : -1,
750 has_mac ? skb_mac_header_len(skb) : -1,
751 skb->network_header,
752 has_trans ? skb_network_header_len(skb) : -1,
753 has_trans ? skb->transport_header : -1,
754 sh->tx_flags, sh->nr_frags,
755 sh->gso_size, sh->gso_type, sh->gso_segs,
756 skb->csum, skb->ip_summed, skb->csum_complete_sw,
757 skb->csum_valid, skb->csum_level,
758 skb->hash, skb->sw_hash, skb->l4_hash,
759 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
760
761 if (dev)
762 printk("%sdev name=%s feat=0x%pNF\n",
763 level, dev->name, &dev->features);
764 if (sk)
765 printk("%ssk family=%hu type=%u proto=%u\n",
766 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
767
768 if (full_pkt && headroom)
769 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
770 16, 1, skb->head, headroom, false);
771
772 seg_len = min_t(int, skb_headlen(skb), len);
773 if (seg_len)
774 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
775 16, 1, skb->data, seg_len, false);
776 len -= seg_len;
777
778 if (full_pkt && tailroom)
779 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
780 16, 1, skb_tail_pointer(skb), tailroom, false);
781
782 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
783 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
784 u32 p_off, p_len, copied;
785 struct page *p;
786 u8 *vaddr;
787
788 skb_frag_foreach_page(frag, skb_frag_off(frag),
789 skb_frag_size(frag), p, p_off, p_len,
790 copied) {
791 seg_len = min_t(int, p_len, len);
792 vaddr = kmap_atomic(p);
793 print_hex_dump(level, "skb frag: ",
794 DUMP_PREFIX_OFFSET,
795 16, 1, vaddr + p_off, seg_len, false);
796 kunmap_atomic(vaddr);
797 len -= seg_len;
798 if (!len)
799 break;
800 }
801 }
802
803 if (full_pkt && skb_has_frag_list(skb)) {
804 printk("skb fraglist:\n");
805 skb_walk_frags(skb, list_skb)
806 skb_dump(level, list_skb, true);
807 }
808}
809EXPORT_SYMBOL(skb_dump);
810
811/**
812 * skb_tx_error - report an sk_buff xmit error
813 * @skb: buffer that triggered an error
814 *
815 * Report xmit error if a device callback is tracking this skb.
816 * skb must be freed afterwards.
817 */
818void skb_tx_error(struct sk_buff *skb)
819{
820 skb_zcopy_clear(skb, true);
821}
822EXPORT_SYMBOL(skb_tx_error);
823
824/**
825 * consume_skb - free an skbuff
826 * @skb: buffer to free
827 *
828 * Drop a ref to the buffer and free it if the usage count has hit zero
829 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
830 * is being dropped after a failure and notes that
831 */
832void consume_skb(struct sk_buff *skb)
833{
834 if (!skb_unref(skb))
835 return;
836
837 trace_consume_skb(skb);
838 __kfree_skb(skb);
839}
840EXPORT_SYMBOL(consume_skb);
841
842/**
843 * consume_stateless_skb - free an skbuff, assuming it is stateless
844 * @skb: buffer to free
845 *
846 * Alike consume_skb(), but this variant assumes that this is the last
847 * skb reference and all the head states have been already dropped
848 */
849void __consume_stateless_skb(struct sk_buff *skb)
850{
851 trace_consume_skb(skb);
852 skb_release_data(skb);
853 kfree_skbmem(skb);
854}
855
856void __kfree_skb_flush(void)
857{
858 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
859
860 /* flush skb_cache if containing objects */
861 if (nc->skb_count) {
862 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
863 nc->skb_cache);
864 nc->skb_count = 0;
865 }
866}
867
868static inline void _kfree_skb_defer(struct sk_buff *skb)
869{
870 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
871
872 /* drop skb->head and call any destructors for packet */
873 skb_release_all(skb);
874
875 /* record skb to CPU local list */
876 nc->skb_cache[nc->skb_count++] = skb;
877
878#ifdef CONFIG_SLUB
879 /* SLUB writes into objects when freeing */
880 prefetchw(skb);
881#endif
882
883 /* flush skb_cache if it is filled */
884 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
885 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
886 nc->skb_cache);
887 nc->skb_count = 0;
888 }
889}
890void __kfree_skb_defer(struct sk_buff *skb)
891{
892 _kfree_skb_defer(skb);
893}
894
895void napi_consume_skb(struct sk_buff *skb, int budget)
896{
897 if (unlikely(!skb))
898 return;
899
900 /* Zero budget indicate non-NAPI context called us, like netpoll */
901 if (unlikely(!budget)) {
902 dev_consume_skb_any(skb);
903 return;
904 }
905
906 if (!skb_unref(skb))
907 return;
908
909 /* if reaching here SKB is ready to free */
910 trace_consume_skb(skb);
911
912 /* if SKB is a clone, don't handle this case */
913 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
914 __kfree_skb(skb);
915 return;
916 }
917
918 _kfree_skb_defer(skb);
919}
920EXPORT_SYMBOL(napi_consume_skb);
921
922/* Make sure a field is enclosed inside headers_start/headers_end section */
923#define CHECK_SKB_FIELD(field) \
924 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
925 offsetof(struct sk_buff, headers_start)); \
926 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
927 offsetof(struct sk_buff, headers_end)); \
928
929static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
930{
931 new->tstamp = old->tstamp;
932 /* We do not copy old->sk */
933 new->dev = old->dev;
934 memcpy(new->cb, old->cb, sizeof(old->cb));
935 skb_dst_copy(new, old);
936 __skb_ext_copy(new, old);
937 __nf_copy(new, old, false);
938
939 /* Note : this field could be in headers_start/headers_end section
940 * It is not yet because we do not want to have a 16 bit hole
941 */
942 new->queue_mapping = old->queue_mapping;
943
944 memcpy(&new->headers_start, &old->headers_start,
945 offsetof(struct sk_buff, headers_end) -
946 offsetof(struct sk_buff, headers_start));
947 CHECK_SKB_FIELD(protocol);
948 CHECK_SKB_FIELD(csum);
949 CHECK_SKB_FIELD(hash);
950 CHECK_SKB_FIELD(priority);
951 CHECK_SKB_FIELD(skb_iif);
952 CHECK_SKB_FIELD(vlan_proto);
953 CHECK_SKB_FIELD(vlan_tci);
954 CHECK_SKB_FIELD(transport_header);
955 CHECK_SKB_FIELD(network_header);
956 CHECK_SKB_FIELD(mac_header);
957 CHECK_SKB_FIELD(inner_protocol);
958 CHECK_SKB_FIELD(inner_transport_header);
959 CHECK_SKB_FIELD(inner_network_header);
960 CHECK_SKB_FIELD(inner_mac_header);
961 CHECK_SKB_FIELD(mark);
962#ifdef CONFIG_NETWORK_SECMARK
963 CHECK_SKB_FIELD(secmark);
964#endif
965#ifdef CONFIG_NET_RX_BUSY_POLL
966 CHECK_SKB_FIELD(napi_id);
967#endif
968#ifdef CONFIG_XPS
969 CHECK_SKB_FIELD(sender_cpu);
970#endif
971#ifdef CONFIG_NET_SCHED
972 CHECK_SKB_FIELD(tc_index);
973#endif
974
975}
976
977/*
978 * You should not add any new code to this function. Add it to
979 * __copy_skb_header above instead.
980 */
981static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
982{
983#define C(x) n->x = skb->x
984
985 n->next = n->prev = NULL;
986 n->sk = NULL;
987 __copy_skb_header(n, skb);
988
989 C(len);
990 C(data_len);
991 C(mac_len);
992 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
993 n->cloned = 1;
994 n->nohdr = 0;
995 n->peeked = 0;
996 C(pfmemalloc);
997 n->destructor = NULL;
998 C(tail);
999 C(end);
1000 C(head);
1001 C(head_frag);
1002 C(data);
1003 C(truesize);
1004 refcount_set(&n->users, 1);
1005
1006 atomic_inc(&(skb_shinfo(skb)->dataref));
1007 skb->cloned = 1;
1008
1009 return n;
1010#undef C
1011}
1012
1013/**
1014 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1015 * @first: first sk_buff of the msg
1016 */
1017struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1018{
1019 struct sk_buff *n;
1020
1021 n = alloc_skb(0, GFP_ATOMIC);
1022 if (!n)
1023 return NULL;
1024
1025 n->len = first->len;
1026 n->data_len = first->len;
1027 n->truesize = first->truesize;
1028
1029 skb_shinfo(n)->frag_list = first;
1030
1031 __copy_skb_header(n, first);
1032 n->destructor = NULL;
1033
1034 return n;
1035}
1036EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1037
1038/**
1039 * skb_morph - morph one skb into another
1040 * @dst: the skb to receive the contents
1041 * @src: the skb to supply the contents
1042 *
1043 * This is identical to skb_clone except that the target skb is
1044 * supplied by the user.
1045 *
1046 * The target skb is returned upon exit.
1047 */
1048struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1049{
1050 skb_release_all(dst);
1051 return __skb_clone(dst, src);
1052}
1053EXPORT_SYMBOL_GPL(skb_morph);
1054
1055int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1056{
1057 unsigned long max_pg, num_pg, new_pg, old_pg;
1058 struct user_struct *user;
1059
1060 if (capable(CAP_IPC_LOCK) || !size)
1061 return 0;
1062
1063 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1064 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1065 user = mmp->user ? : current_user();
1066
1067 do {
1068 old_pg = atomic_long_read(&user->locked_vm);
1069 new_pg = old_pg + num_pg;
1070 if (new_pg > max_pg)
1071 return -ENOBUFS;
1072 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1073 old_pg);
1074
1075 if (!mmp->user) {
1076 mmp->user = get_uid(user);
1077 mmp->num_pg = num_pg;
1078 } else {
1079 mmp->num_pg += num_pg;
1080 }
1081
1082 return 0;
1083}
1084EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1085
1086void mm_unaccount_pinned_pages(struct mmpin *mmp)
1087{
1088 if (mmp->user) {
1089 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1090 free_uid(mmp->user);
1091 }
1092}
1093EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1094
1095struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1096{
1097 struct ubuf_info *uarg;
1098 struct sk_buff *skb;
1099
1100 WARN_ON_ONCE(!in_task());
1101
1102 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1103 if (!skb)
1104 return NULL;
1105
1106 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1107 uarg = (void *)skb->cb;
1108 uarg->mmp.user = NULL;
1109
1110 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1111 kfree_skb(skb);
1112 return NULL;
1113 }
1114
1115 uarg->callback = sock_zerocopy_callback;
1116 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1117 uarg->len = 1;
1118 uarg->bytelen = size;
1119 uarg->zerocopy = 1;
1120 refcount_set(&uarg->refcnt, 1);
1121 sock_hold(sk);
1122
1123 return uarg;
1124}
1125EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1126
1127static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1128{
1129 return container_of((void *)uarg, struct sk_buff, cb);
1130}
1131
1132struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1133 struct ubuf_info *uarg)
1134{
1135 if (uarg) {
1136 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1137 u32 bytelen, next;
1138
1139 /* realloc only when socket is locked (TCP, UDP cork),
1140 * so uarg->len and sk_zckey access is serialized
1141 */
1142 if (!sock_owned_by_user(sk)) {
1143 WARN_ON_ONCE(1);
1144 return NULL;
1145 }
1146
1147 bytelen = uarg->bytelen + size;
1148 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1149 /* TCP can create new skb to attach new uarg */
1150 if (sk->sk_type == SOCK_STREAM)
1151 goto new_alloc;
1152 return NULL;
1153 }
1154
1155 next = (u32)atomic_read(&sk->sk_zckey);
1156 if ((u32)(uarg->id + uarg->len) == next) {
1157 if (mm_account_pinned_pages(&uarg->mmp, size))
1158 return NULL;
1159 uarg->len++;
1160 uarg->bytelen = bytelen;
1161 atomic_set(&sk->sk_zckey, ++next);
1162
1163 /* no extra ref when appending to datagram (MSG_MORE) */
1164 if (sk->sk_type == SOCK_STREAM)
1165 sock_zerocopy_get(uarg);
1166
1167 return uarg;
1168 }
1169 }
1170
1171new_alloc:
1172 return sock_zerocopy_alloc(sk, size);
1173}
1174EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1175
1176static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1177{
1178 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1179 u32 old_lo, old_hi;
1180 u64 sum_len;
1181
1182 old_lo = serr->ee.ee_info;
1183 old_hi = serr->ee.ee_data;
1184 sum_len = old_hi - old_lo + 1ULL + len;
1185
1186 if (sum_len >= (1ULL << 32))
1187 return false;
1188
1189 if (lo != old_hi + 1)
1190 return false;
1191
1192 serr->ee.ee_data += len;
1193 return true;
1194}
1195
1196void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1197{
1198 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1199 struct sock_exterr_skb *serr;
1200 struct sock *sk = skb->sk;
1201 struct sk_buff_head *q;
1202 unsigned long flags;
1203 u32 lo, hi;
1204 u16 len;
1205
1206 mm_unaccount_pinned_pages(&uarg->mmp);
1207
1208 /* if !len, there was only 1 call, and it was aborted
1209 * so do not queue a completion notification
1210 */
1211 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1212 goto release;
1213
1214 len = uarg->len;
1215 lo = uarg->id;
1216 hi = uarg->id + len - 1;
1217
1218 serr = SKB_EXT_ERR(skb);
1219 memset(serr, 0, sizeof(*serr));
1220 serr->ee.ee_errno = 0;
1221 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1222 serr->ee.ee_data = hi;
1223 serr->ee.ee_info = lo;
1224 if (!success)
1225 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1226
1227 q = &sk->sk_error_queue;
1228 spin_lock_irqsave(&q->lock, flags);
1229 tail = skb_peek_tail(q);
1230 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1231 !skb_zerocopy_notify_extend(tail, lo, len)) {
1232 __skb_queue_tail(q, skb);
1233 skb = NULL;
1234 }
1235 spin_unlock_irqrestore(&q->lock, flags);
1236
1237 sk->sk_error_report(sk);
1238
1239release:
1240 consume_skb(skb);
1241 sock_put(sk);
1242}
1243EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1244
1245void sock_zerocopy_put(struct ubuf_info *uarg)
1246{
1247 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1248 if (uarg->callback)
1249 uarg->callback(uarg, uarg->zerocopy);
1250 else
1251 consume_skb(skb_from_uarg(uarg));
1252 }
1253}
1254EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1255
1256void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1257{
1258 if (uarg) {
1259 struct sock *sk = skb_from_uarg(uarg)->sk;
1260
1261 atomic_dec(&sk->sk_zckey);
1262 uarg->len--;
1263
1264 if (have_uref)
1265 sock_zerocopy_put(uarg);
1266 }
1267}
1268EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1269
1270int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1271{
1272 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1273}
1274EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1275
1276int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1277 struct msghdr *msg, int len,
1278 struct ubuf_info *uarg)
1279{
1280 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1281 struct iov_iter orig_iter = msg->msg_iter;
1282 int err, orig_len = skb->len;
1283
1284 /* An skb can only point to one uarg. This edge case happens when
1285 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1286 */
1287 if (orig_uarg && uarg != orig_uarg)
1288 return -EEXIST;
1289
1290 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1291 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1292 struct sock *save_sk = skb->sk;
1293
1294 /* Streams do not free skb on error. Reset to prev state. */
1295 msg->msg_iter = orig_iter;
1296 skb->sk = sk;
1297 ___pskb_trim(skb, orig_len);
1298 skb->sk = save_sk;
1299 return err;
1300 }
1301
1302 skb_zcopy_set(skb, uarg, NULL);
1303 return skb->len - orig_len;
1304}
1305EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1306
1307static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1308 gfp_t gfp_mask)
1309{
1310 if (skb_zcopy(orig)) {
1311 if (skb_zcopy(nskb)) {
1312 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1313 if (!gfp_mask) {
1314 WARN_ON_ONCE(1);
1315 return -ENOMEM;
1316 }
1317 if (skb_uarg(nskb) == skb_uarg(orig))
1318 return 0;
1319 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1320 return -EIO;
1321 }
1322 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1323 }
1324 return 0;
1325}
1326
1327/**
1328 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1329 * @skb: the skb to modify
1330 * @gfp_mask: allocation priority
1331 *
1332 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1333 * It will copy all frags into kernel and drop the reference
1334 * to userspace pages.
1335 *
1336 * If this function is called from an interrupt gfp_mask() must be
1337 * %GFP_ATOMIC.
1338 *
1339 * Returns 0 on success or a negative error code on failure
1340 * to allocate kernel memory to copy to.
1341 */
1342int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1343{
1344 int num_frags = skb_shinfo(skb)->nr_frags;
1345 struct page *page, *head = NULL;
1346 int i, new_frags;
1347 u32 d_off;
1348
1349 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1350 return -EINVAL;
1351
1352 if (!num_frags)
1353 goto release;
1354
1355 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1356 for (i = 0; i < new_frags; i++) {
1357 page = alloc_page(gfp_mask);
1358 if (!page) {
1359 while (head) {
1360 struct page *next = (struct page *)page_private(head);
1361 put_page(head);
1362 head = next;
1363 }
1364 return -ENOMEM;
1365 }
1366 set_page_private(page, (unsigned long)head);
1367 head = page;
1368 }
1369
1370 page = head;
1371 d_off = 0;
1372 for (i = 0; i < num_frags; i++) {
1373 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1374 u32 p_off, p_len, copied;
1375 struct page *p;
1376 u8 *vaddr;
1377
1378 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1379 p, p_off, p_len, copied) {
1380 u32 copy, done = 0;
1381 vaddr = kmap_atomic(p);
1382
1383 while (done < p_len) {
1384 if (d_off == PAGE_SIZE) {
1385 d_off = 0;
1386 page = (struct page *)page_private(page);
1387 }
1388 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1389 memcpy(page_address(page) + d_off,
1390 vaddr + p_off + done, copy);
1391 done += copy;
1392 d_off += copy;
1393 }
1394 kunmap_atomic(vaddr);
1395 }
1396 }
1397
1398 /* skb frags release userspace buffers */
1399 for (i = 0; i < num_frags; i++)
1400 skb_frag_unref(skb, i);
1401
1402 /* skb frags point to kernel buffers */
1403 for (i = 0; i < new_frags - 1; i++) {
1404 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1405 head = (struct page *)page_private(head);
1406 }
1407 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1408 skb_shinfo(skb)->nr_frags = new_frags;
1409
1410release:
1411 skb_zcopy_clear(skb, false);
1412 return 0;
1413}
1414EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1415
1416/**
1417 * skb_clone - duplicate an sk_buff
1418 * @skb: buffer to clone
1419 * @gfp_mask: allocation priority
1420 *
1421 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1422 * copies share the same packet data but not structure. The new
1423 * buffer has a reference count of 1. If the allocation fails the
1424 * function returns %NULL otherwise the new buffer is returned.
1425 *
1426 * If this function is called from an interrupt gfp_mask() must be
1427 * %GFP_ATOMIC.
1428 */
1429
1430struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1431{
1432 struct sk_buff_fclones *fclones = container_of(skb,
1433 struct sk_buff_fclones,
1434 skb1);
1435 struct sk_buff *n;
1436
1437 if (skb_orphan_frags(skb, gfp_mask))
1438 return NULL;
1439
1440 if (skb->fclone == SKB_FCLONE_ORIG &&
1441 refcount_read(&fclones->fclone_ref) == 1) {
1442 n = &fclones->skb2;
1443 refcount_set(&fclones->fclone_ref, 2);
1444 } else {
1445 if (skb_pfmemalloc(skb))
1446 gfp_mask |= __GFP_MEMALLOC;
1447
1448 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1449 if (!n)
1450 return NULL;
1451
1452 n->fclone = SKB_FCLONE_UNAVAILABLE;
1453 }
1454
1455 return __skb_clone(n, skb);
1456}
1457EXPORT_SYMBOL(skb_clone);
1458
1459void skb_headers_offset_update(struct sk_buff *skb, int off)
1460{
1461 /* Only adjust this if it actually is csum_start rather than csum */
1462 if (skb->ip_summed == CHECKSUM_PARTIAL)
1463 skb->csum_start += off;
1464 /* {transport,network,mac}_header and tail are relative to skb->head */
1465 skb->transport_header += off;
1466 skb->network_header += off;
1467 if (skb_mac_header_was_set(skb))
1468 skb->mac_header += off;
1469 skb->inner_transport_header += off;
1470 skb->inner_network_header += off;
1471 skb->inner_mac_header += off;
1472}
1473EXPORT_SYMBOL(skb_headers_offset_update);
1474
1475void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1476{
1477 __copy_skb_header(new, old);
1478
1479 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1480 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1481 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1482}
1483EXPORT_SYMBOL(skb_copy_header);
1484
1485static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1486{
1487 if (skb_pfmemalloc(skb))
1488 return SKB_ALLOC_RX;
1489 return 0;
1490}
1491
1492/**
1493 * skb_copy - create private copy of an sk_buff
1494 * @skb: buffer to copy
1495 * @gfp_mask: allocation priority
1496 *
1497 * Make a copy of both an &sk_buff and its data. This is used when the
1498 * caller wishes to modify the data and needs a private copy of the
1499 * data to alter. Returns %NULL on failure or the pointer to the buffer
1500 * on success. The returned buffer has a reference count of 1.
1501 *
1502 * As by-product this function converts non-linear &sk_buff to linear
1503 * one, so that &sk_buff becomes completely private and caller is allowed
1504 * to modify all the data of returned buffer. This means that this
1505 * function is not recommended for use in circumstances when only
1506 * header is going to be modified. Use pskb_copy() instead.
1507 */
1508
1509struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1510{
1511 int headerlen = skb_headroom(skb);
1512 unsigned int size = skb_end_offset(skb) + skb->data_len;
1513 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1514 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1515
1516 if (!n)
1517 return NULL;
1518
1519 /* Set the data pointer */
1520 skb_reserve(n, headerlen);
1521 /* Set the tail pointer and length */
1522 skb_put(n, skb->len);
1523
1524 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1525
1526 skb_copy_header(n, skb);
1527 return n;
1528}
1529EXPORT_SYMBOL(skb_copy);
1530
1531/**
1532 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1533 * @skb: buffer to copy
1534 * @headroom: headroom of new skb
1535 * @gfp_mask: allocation priority
1536 * @fclone: if true allocate the copy of the skb from the fclone
1537 * cache instead of the head cache; it is recommended to set this
1538 * to true for the cases where the copy will likely be cloned
1539 *
1540 * Make a copy of both an &sk_buff and part of its data, located
1541 * in header. Fragmented data remain shared. This is used when
1542 * the caller wishes to modify only header of &sk_buff and needs
1543 * private copy of the header to alter. Returns %NULL on failure
1544 * or the pointer to the buffer on success.
1545 * The returned buffer has a reference count of 1.
1546 */
1547
1548struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1549 gfp_t gfp_mask, bool fclone)
1550{
1551 unsigned int size = skb_headlen(skb) + headroom;
1552 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1553 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1554
1555 if (!n)
1556 goto out;
1557
1558 /* Set the data pointer */
1559 skb_reserve(n, headroom);
1560 /* Set the tail pointer and length */
1561 skb_put(n, skb_headlen(skb));
1562 /* Copy the bytes */
1563 skb_copy_from_linear_data(skb, n->data, n->len);
1564
1565 n->truesize += skb->data_len;
1566 n->data_len = skb->data_len;
1567 n->len = skb->len;
1568
1569 if (skb_shinfo(skb)->nr_frags) {
1570 int i;
1571
1572 if (skb_orphan_frags(skb, gfp_mask) ||
1573 skb_zerocopy_clone(n, skb, gfp_mask)) {
1574 kfree_skb(n);
1575 n = NULL;
1576 goto out;
1577 }
1578 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1579 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1580 skb_frag_ref(skb, i);
1581 }
1582 skb_shinfo(n)->nr_frags = i;
1583 }
1584
1585 if (skb_has_frag_list(skb)) {
1586 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1587 skb_clone_fraglist(n);
1588 }
1589
1590 skb_copy_header(n, skb);
1591out:
1592 return n;
1593}
1594EXPORT_SYMBOL(__pskb_copy_fclone);
1595
1596/**
1597 * pskb_expand_head - reallocate header of &sk_buff
1598 * @skb: buffer to reallocate
1599 * @nhead: room to add at head
1600 * @ntail: room to add at tail
1601 * @gfp_mask: allocation priority
1602 *
1603 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1604 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1605 * reference count of 1. Returns zero in the case of success or error,
1606 * if expansion failed. In the last case, &sk_buff is not changed.
1607 *
1608 * All the pointers pointing into skb header may change and must be
1609 * reloaded after call to this function.
1610 */
1611
1612int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1613 gfp_t gfp_mask)
1614{
1615 int i, osize = skb_end_offset(skb);
1616 int size = osize + nhead + ntail;
1617 long off;
1618 u8 *data;
1619
1620 BUG_ON(nhead < 0);
1621
1622 BUG_ON(skb_shared(skb));
1623
1624 size = SKB_DATA_ALIGN(size);
1625
1626 if (skb_pfmemalloc(skb))
1627 gfp_mask |= __GFP_MEMALLOC;
1628 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1629 gfp_mask, NUMA_NO_NODE, NULL);
1630 if (!data)
1631 goto nodata;
1632 size = SKB_WITH_OVERHEAD(ksize(data));
1633
1634 /* Copy only real data... and, alas, header. This should be
1635 * optimized for the cases when header is void.
1636 */
1637 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1638
1639 memcpy((struct skb_shared_info *)(data + size),
1640 skb_shinfo(skb),
1641 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1642
1643 /*
1644 * if shinfo is shared we must drop the old head gracefully, but if it
1645 * is not we can just drop the old head and let the existing refcount
1646 * be since all we did is relocate the values
1647 */
1648 if (skb_cloned(skb)) {
1649 if (skb_orphan_frags(skb, gfp_mask))
1650 goto nofrags;
1651 if (skb_zcopy(skb))
1652 refcount_inc(&skb_uarg(skb)->refcnt);
1653 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1654 skb_frag_ref(skb, i);
1655
1656 if (skb_has_frag_list(skb))
1657 skb_clone_fraglist(skb);
1658
1659 skb_release_data(skb);
1660 } else {
1661 skb_free_head(skb);
1662 }
1663 off = (data + nhead) - skb->head;
1664
1665 skb->head = data;
1666 skb->head_frag = 0;
1667 skb->data += off;
1668#ifdef NET_SKBUFF_DATA_USES_OFFSET
1669 skb->end = size;
1670 off = nhead;
1671#else
1672 skb->end = skb->head + size;
1673#endif
1674 skb->tail += off;
1675 skb_headers_offset_update(skb, nhead);
1676 skb->cloned = 0;
1677 skb->hdr_len = 0;
1678 skb->nohdr = 0;
1679 atomic_set(&skb_shinfo(skb)->dataref, 1);
1680
1681 skb_metadata_clear(skb);
1682
1683 /* It is not generally safe to change skb->truesize.
1684 * For the moment, we really care of rx path, or
1685 * when skb is orphaned (not attached to a socket).
1686 */
1687 if (!skb->sk || skb->destructor == sock_edemux)
1688 skb->truesize += size - osize;
1689
1690 return 0;
1691
1692nofrags:
1693 kfree(data);
1694nodata:
1695 return -ENOMEM;
1696}
1697EXPORT_SYMBOL(pskb_expand_head);
1698
1699/* Make private copy of skb with writable head and some headroom */
1700
1701struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1702{
1703 struct sk_buff *skb2;
1704 int delta = headroom - skb_headroom(skb);
1705
1706 if (delta <= 0)
1707 skb2 = pskb_copy(skb, GFP_ATOMIC);
1708 else {
1709 skb2 = skb_clone(skb, GFP_ATOMIC);
1710 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1711 GFP_ATOMIC)) {
1712 kfree_skb(skb2);
1713 skb2 = NULL;
1714 }
1715 }
1716 return skb2;
1717}
1718EXPORT_SYMBOL(skb_realloc_headroom);
1719
1720/**
1721 * skb_copy_expand - copy and expand sk_buff
1722 * @skb: buffer to copy
1723 * @newheadroom: new free bytes at head
1724 * @newtailroom: new free bytes at tail
1725 * @gfp_mask: allocation priority
1726 *
1727 * Make a copy of both an &sk_buff and its data and while doing so
1728 * allocate additional space.
1729 *
1730 * This is used when the caller wishes to modify the data and needs a
1731 * private copy of the data to alter as well as more space for new fields.
1732 * Returns %NULL on failure or the pointer to the buffer
1733 * on success. The returned buffer has a reference count of 1.
1734 *
1735 * You must pass %GFP_ATOMIC as the allocation priority if this function
1736 * is called from an interrupt.
1737 */
1738struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1739 int newheadroom, int newtailroom,
1740 gfp_t gfp_mask)
1741{
1742 /*
1743 * Allocate the copy buffer
1744 */
1745 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1746 gfp_mask, skb_alloc_rx_flag(skb),
1747 NUMA_NO_NODE);
1748 int oldheadroom = skb_headroom(skb);
1749 int head_copy_len, head_copy_off;
1750
1751 if (!n)
1752 return NULL;
1753
1754 skb_reserve(n, newheadroom);
1755
1756 /* Set the tail pointer and length */
1757 skb_put(n, skb->len);
1758
1759 head_copy_len = oldheadroom;
1760 head_copy_off = 0;
1761 if (newheadroom <= head_copy_len)
1762 head_copy_len = newheadroom;
1763 else
1764 head_copy_off = newheadroom - head_copy_len;
1765
1766 /* Copy the linear header and data. */
1767 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1768 skb->len + head_copy_len));
1769
1770 skb_copy_header(n, skb);
1771
1772 skb_headers_offset_update(n, newheadroom - oldheadroom);
1773
1774 return n;
1775}
1776EXPORT_SYMBOL(skb_copy_expand);
1777
1778/**
1779 * __skb_pad - zero pad the tail of an skb
1780 * @skb: buffer to pad
1781 * @pad: space to pad
1782 * @free_on_error: free buffer on error
1783 *
1784 * Ensure that a buffer is followed by a padding area that is zero
1785 * filled. Used by network drivers which may DMA or transfer data
1786 * beyond the buffer end onto the wire.
1787 *
1788 * May return error in out of memory cases. The skb is freed on error
1789 * if @free_on_error is true.
1790 */
1791
1792int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1793{
1794 int err;
1795 int ntail;
1796
1797 /* If the skbuff is non linear tailroom is always zero.. */
1798 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1799 memset(skb->data+skb->len, 0, pad);
1800 return 0;
1801 }
1802
1803 ntail = skb->data_len + pad - (skb->end - skb->tail);
1804 if (likely(skb_cloned(skb) || ntail > 0)) {
1805 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1806 if (unlikely(err))
1807 goto free_skb;
1808 }
1809
1810 /* FIXME: The use of this function with non-linear skb's really needs
1811 * to be audited.
1812 */
1813 err = skb_linearize(skb);
1814 if (unlikely(err))
1815 goto free_skb;
1816
1817 memset(skb->data + skb->len, 0, pad);
1818 return 0;
1819
1820free_skb:
1821 if (free_on_error)
1822 kfree_skb(skb);
1823 return err;
1824}
1825EXPORT_SYMBOL(__skb_pad);
1826
1827/**
1828 * pskb_put - add data to the tail of a potentially fragmented buffer
1829 * @skb: start of the buffer to use
1830 * @tail: tail fragment of the buffer to use
1831 * @len: amount of data to add
1832 *
1833 * This function extends the used data area of the potentially
1834 * fragmented buffer. @tail must be the last fragment of @skb -- or
1835 * @skb itself. If this would exceed the total buffer size the kernel
1836 * will panic. A pointer to the first byte of the extra data is
1837 * returned.
1838 */
1839
1840void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1841{
1842 if (tail != skb) {
1843 skb->data_len += len;
1844 skb->len += len;
1845 }
1846 return skb_put(tail, len);
1847}
1848EXPORT_SYMBOL_GPL(pskb_put);
1849
1850/**
1851 * skb_put - add data to a buffer
1852 * @skb: buffer to use
1853 * @len: amount of data to add
1854 *
1855 * This function extends the used data area of the buffer. If this would
1856 * exceed the total buffer size the kernel will panic. A pointer to the
1857 * first byte of the extra data is returned.
1858 */
1859void *skb_put(struct sk_buff *skb, unsigned int len)
1860{
1861 void *tmp = skb_tail_pointer(skb);
1862 SKB_LINEAR_ASSERT(skb);
1863 skb->tail += len;
1864 skb->len += len;
1865 if (unlikely(skb->tail > skb->end))
1866 skb_over_panic(skb, len, __builtin_return_address(0));
1867 return tmp;
1868}
1869EXPORT_SYMBOL(skb_put);
1870
1871/**
1872 * skb_push - add data to the start of a buffer
1873 * @skb: buffer to use
1874 * @len: amount of data to add
1875 *
1876 * This function extends the used data area of the buffer at the buffer
1877 * start. If this would exceed the total buffer headroom the kernel will
1878 * panic. A pointer to the first byte of the extra data is returned.
1879 */
1880void *skb_push(struct sk_buff *skb, unsigned int len)
1881{
1882 skb->data -= len;
1883 skb->len += len;
1884 if (unlikely(skb->data < skb->head))
1885 skb_under_panic(skb, len, __builtin_return_address(0));
1886 return skb->data;
1887}
1888EXPORT_SYMBOL(skb_push);
1889
1890/**
1891 * skb_pull - remove data from the start of a buffer
1892 * @skb: buffer to use
1893 * @len: amount of data to remove
1894 *
1895 * This function removes data from the start of a buffer, returning
1896 * the memory to the headroom. A pointer to the next data in the buffer
1897 * is returned. Once the data has been pulled future pushes will overwrite
1898 * the old data.
1899 */
1900void *skb_pull(struct sk_buff *skb, unsigned int len)
1901{
1902 return skb_pull_inline(skb, len);
1903}
1904EXPORT_SYMBOL(skb_pull);
1905
1906/**
1907 * skb_trim - remove end from a buffer
1908 * @skb: buffer to alter
1909 * @len: new length
1910 *
1911 * Cut the length of a buffer down by removing data from the tail. If
1912 * the buffer is already under the length specified it is not modified.
1913 * The skb must be linear.
1914 */
1915void skb_trim(struct sk_buff *skb, unsigned int len)
1916{
1917 if (skb->len > len)
1918 __skb_trim(skb, len);
1919}
1920EXPORT_SYMBOL(skb_trim);
1921
1922/* Trims skb to length len. It can change skb pointers.
1923 */
1924
1925int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1926{
1927 struct sk_buff **fragp;
1928 struct sk_buff *frag;
1929 int offset = skb_headlen(skb);
1930 int nfrags = skb_shinfo(skb)->nr_frags;
1931 int i;
1932 int err;
1933
1934 if (skb_cloned(skb) &&
1935 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1936 return err;
1937
1938 i = 0;
1939 if (offset >= len)
1940 goto drop_pages;
1941
1942 for (; i < nfrags; i++) {
1943 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1944
1945 if (end < len) {
1946 offset = end;
1947 continue;
1948 }
1949
1950 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1951
1952drop_pages:
1953 skb_shinfo(skb)->nr_frags = i;
1954
1955 for (; i < nfrags; i++)
1956 skb_frag_unref(skb, i);
1957
1958 if (skb_has_frag_list(skb))
1959 skb_drop_fraglist(skb);
1960 goto done;
1961 }
1962
1963 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1964 fragp = &frag->next) {
1965 int end = offset + frag->len;
1966
1967 if (skb_shared(frag)) {
1968 struct sk_buff *nfrag;
1969
1970 nfrag = skb_clone(frag, GFP_ATOMIC);
1971 if (unlikely(!nfrag))
1972 return -ENOMEM;
1973
1974 nfrag->next = frag->next;
1975 consume_skb(frag);
1976 frag = nfrag;
1977 *fragp = frag;
1978 }
1979
1980 if (end < len) {
1981 offset = end;
1982 continue;
1983 }
1984
1985 if (end > len &&
1986 unlikely((err = pskb_trim(frag, len - offset))))
1987 return err;
1988
1989 if (frag->next)
1990 skb_drop_list(&frag->next);
1991 break;
1992 }
1993
1994done:
1995 if (len > skb_headlen(skb)) {
1996 skb->data_len -= skb->len - len;
1997 skb->len = len;
1998 } else {
1999 skb->len = len;
2000 skb->data_len = 0;
2001 skb_set_tail_pointer(skb, len);
2002 }
2003
2004 if (!skb->sk || skb->destructor == sock_edemux)
2005 skb_condense(skb);
2006 return 0;
2007}
2008EXPORT_SYMBOL(___pskb_trim);
2009
2010/* Note : use pskb_trim_rcsum() instead of calling this directly
2011 */
2012int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2013{
2014 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2015 int delta = skb->len - len;
2016
2017 skb->csum = csum_block_sub(skb->csum,
2018 skb_checksum(skb, len, delta, 0),
2019 len);
2020 }
2021 return __pskb_trim(skb, len);
2022}
2023EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2024
2025/**
2026 * __pskb_pull_tail - advance tail of skb header
2027 * @skb: buffer to reallocate
2028 * @delta: number of bytes to advance tail
2029 *
2030 * The function makes a sense only on a fragmented &sk_buff,
2031 * it expands header moving its tail forward and copying necessary
2032 * data from fragmented part.
2033 *
2034 * &sk_buff MUST have reference count of 1.
2035 *
2036 * Returns %NULL (and &sk_buff does not change) if pull failed
2037 * or value of new tail of skb in the case of success.
2038 *
2039 * All the pointers pointing into skb header may change and must be
2040 * reloaded after call to this function.
2041 */
2042
2043/* Moves tail of skb head forward, copying data from fragmented part,
2044 * when it is necessary.
2045 * 1. It may fail due to malloc failure.
2046 * 2. It may change skb pointers.
2047 *
2048 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2049 */
2050void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2051{
2052 /* If skb has not enough free space at tail, get new one
2053 * plus 128 bytes for future expansions. If we have enough
2054 * room at tail, reallocate without expansion only if skb is cloned.
2055 */
2056 int i, k, eat = (skb->tail + delta) - skb->end;
2057
2058 if (eat > 0 || skb_cloned(skb)) {
2059 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2060 GFP_ATOMIC))
2061 return NULL;
2062 }
2063
2064 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2065 skb_tail_pointer(skb), delta));
2066
2067 /* Optimization: no fragments, no reasons to preestimate
2068 * size of pulled pages. Superb.
2069 */
2070 if (!skb_has_frag_list(skb))
2071 goto pull_pages;
2072
2073 /* Estimate size of pulled pages. */
2074 eat = delta;
2075 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2076 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2077
2078 if (size >= eat)
2079 goto pull_pages;
2080 eat -= size;
2081 }
2082
2083 /* If we need update frag list, we are in troubles.
2084 * Certainly, it is possible to add an offset to skb data,
2085 * but taking into account that pulling is expected to
2086 * be very rare operation, it is worth to fight against
2087 * further bloating skb head and crucify ourselves here instead.
2088 * Pure masohism, indeed. 8)8)
2089 */
2090 if (eat) {
2091 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2092 struct sk_buff *clone = NULL;
2093 struct sk_buff *insp = NULL;
2094
2095 do {
2096 if (list->len <= eat) {
2097 /* Eaten as whole. */
2098 eat -= list->len;
2099 list = list->next;
2100 insp = list;
2101 } else {
2102 /* Eaten partially. */
2103
2104 if (skb_shared(list)) {
2105 /* Sucks! We need to fork list. :-( */
2106 clone = skb_clone(list, GFP_ATOMIC);
2107 if (!clone)
2108 return NULL;
2109 insp = list->next;
2110 list = clone;
2111 } else {
2112 /* This may be pulled without
2113 * problems. */
2114 insp = list;
2115 }
2116 if (!pskb_pull(list, eat)) {
2117 kfree_skb(clone);
2118 return NULL;
2119 }
2120 break;
2121 }
2122 } while (eat);
2123
2124 /* Free pulled out fragments. */
2125 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2126 skb_shinfo(skb)->frag_list = list->next;
2127 kfree_skb(list);
2128 }
2129 /* And insert new clone at head. */
2130 if (clone) {
2131 clone->next = list;
2132 skb_shinfo(skb)->frag_list = clone;
2133 }
2134 }
2135 /* Success! Now we may commit changes to skb data. */
2136
2137pull_pages:
2138 eat = delta;
2139 k = 0;
2140 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2141 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2142
2143 if (size <= eat) {
2144 skb_frag_unref(skb, i);
2145 eat -= size;
2146 } else {
2147 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2148
2149 *frag = skb_shinfo(skb)->frags[i];
2150 if (eat) {
2151 skb_frag_off_add(frag, eat);
2152 skb_frag_size_sub(frag, eat);
2153 if (!i)
2154 goto end;
2155 eat = 0;
2156 }
2157 k++;
2158 }
2159 }
2160 skb_shinfo(skb)->nr_frags = k;
2161
2162end:
2163 skb->tail += delta;
2164 skb->data_len -= delta;
2165
2166 if (!skb->data_len)
2167 skb_zcopy_clear(skb, false);
2168
2169 return skb_tail_pointer(skb);
2170}
2171EXPORT_SYMBOL(__pskb_pull_tail);
2172
2173/**
2174 * skb_copy_bits - copy bits from skb to kernel buffer
2175 * @skb: source skb
2176 * @offset: offset in source
2177 * @to: destination buffer
2178 * @len: number of bytes to copy
2179 *
2180 * Copy the specified number of bytes from the source skb to the
2181 * destination buffer.
2182 *
2183 * CAUTION ! :
2184 * If its prototype is ever changed,
2185 * check arch/{*}/net/{*}.S files,
2186 * since it is called from BPF assembly code.
2187 */
2188int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2189{
2190 int start = skb_headlen(skb);
2191 struct sk_buff *frag_iter;
2192 int i, copy;
2193
2194 if (offset > (int)skb->len - len)
2195 goto fault;
2196
2197 /* Copy header. */
2198 if ((copy = start - offset) > 0) {
2199 if (copy > len)
2200 copy = len;
2201 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2202 if ((len -= copy) == 0)
2203 return 0;
2204 offset += copy;
2205 to += copy;
2206 }
2207
2208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209 int end;
2210 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2211
2212 WARN_ON(start > offset + len);
2213
2214 end = start + skb_frag_size(f);
2215 if ((copy = end - offset) > 0) {
2216 u32 p_off, p_len, copied;
2217 struct page *p;
2218 u8 *vaddr;
2219
2220 if (copy > len)
2221 copy = len;
2222
2223 skb_frag_foreach_page(f,
2224 skb_frag_off(f) + offset - start,
2225 copy, p, p_off, p_len, copied) {
2226 vaddr = kmap_atomic(p);
2227 memcpy(to + copied, vaddr + p_off, p_len);
2228 kunmap_atomic(vaddr);
2229 }
2230
2231 if ((len -= copy) == 0)
2232 return 0;
2233 offset += copy;
2234 to += copy;
2235 }
2236 start = end;
2237 }
2238
2239 skb_walk_frags(skb, frag_iter) {
2240 int end;
2241
2242 WARN_ON(start > offset + len);
2243
2244 end = start + frag_iter->len;
2245 if ((copy = end - offset) > 0) {
2246 if (copy > len)
2247 copy = len;
2248 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2249 goto fault;
2250 if ((len -= copy) == 0)
2251 return 0;
2252 offset += copy;
2253 to += copy;
2254 }
2255 start = end;
2256 }
2257
2258 if (!len)
2259 return 0;
2260
2261fault:
2262 return -EFAULT;
2263}
2264EXPORT_SYMBOL(skb_copy_bits);
2265
2266/*
2267 * Callback from splice_to_pipe(), if we need to release some pages
2268 * at the end of the spd in case we error'ed out in filling the pipe.
2269 */
2270static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2271{
2272 put_page(spd->pages[i]);
2273}
2274
2275static struct page *linear_to_page(struct page *page, unsigned int *len,
2276 unsigned int *offset,
2277 struct sock *sk)
2278{
2279 struct page_frag *pfrag = sk_page_frag(sk);
2280
2281 if (!sk_page_frag_refill(sk, pfrag))
2282 return NULL;
2283
2284 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2285
2286 memcpy(page_address(pfrag->page) + pfrag->offset,
2287 page_address(page) + *offset, *len);
2288 *offset = pfrag->offset;
2289 pfrag->offset += *len;
2290
2291 return pfrag->page;
2292}
2293
2294static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2295 struct page *page,
2296 unsigned int offset)
2297{
2298 return spd->nr_pages &&
2299 spd->pages[spd->nr_pages - 1] == page &&
2300 (spd->partial[spd->nr_pages - 1].offset +
2301 spd->partial[spd->nr_pages - 1].len == offset);
2302}
2303
2304/*
2305 * Fill page/offset/length into spd, if it can hold more pages.
2306 */
2307static bool spd_fill_page(struct splice_pipe_desc *spd,
2308 struct pipe_inode_info *pipe, struct page *page,
2309 unsigned int *len, unsigned int offset,
2310 bool linear,
2311 struct sock *sk)
2312{
2313 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2314 return true;
2315
2316 if (linear) {
2317 page = linear_to_page(page, len, &offset, sk);
2318 if (!page)
2319 return true;
2320 }
2321 if (spd_can_coalesce(spd, page, offset)) {
2322 spd->partial[spd->nr_pages - 1].len += *len;
2323 return false;
2324 }
2325 get_page(page);
2326 spd->pages[spd->nr_pages] = page;
2327 spd->partial[spd->nr_pages].len = *len;
2328 spd->partial[spd->nr_pages].offset = offset;
2329 spd->nr_pages++;
2330
2331 return false;
2332}
2333
2334static bool __splice_segment(struct page *page, unsigned int poff,
2335 unsigned int plen, unsigned int *off,
2336 unsigned int *len,
2337 struct splice_pipe_desc *spd, bool linear,
2338 struct sock *sk,
2339 struct pipe_inode_info *pipe)
2340{
2341 if (!*len)
2342 return true;
2343
2344 /* skip this segment if already processed */
2345 if (*off >= plen) {
2346 *off -= plen;
2347 return false;
2348 }
2349
2350 /* ignore any bits we already processed */
2351 poff += *off;
2352 plen -= *off;
2353 *off = 0;
2354
2355 do {
2356 unsigned int flen = min(*len, plen);
2357
2358 if (spd_fill_page(spd, pipe, page, &flen, poff,
2359 linear, sk))
2360 return true;
2361 poff += flen;
2362 plen -= flen;
2363 *len -= flen;
2364 } while (*len && plen);
2365
2366 return false;
2367}
2368
2369/*
2370 * Map linear and fragment data from the skb to spd. It reports true if the
2371 * pipe is full or if we already spliced the requested length.
2372 */
2373static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2374 unsigned int *offset, unsigned int *len,
2375 struct splice_pipe_desc *spd, struct sock *sk)
2376{
2377 int seg;
2378 struct sk_buff *iter;
2379
2380 /* map the linear part :
2381 * If skb->head_frag is set, this 'linear' part is backed by a
2382 * fragment, and if the head is not shared with any clones then
2383 * we can avoid a copy since we own the head portion of this page.
2384 */
2385 if (__splice_segment(virt_to_page(skb->data),
2386 (unsigned long) skb->data & (PAGE_SIZE - 1),
2387 skb_headlen(skb),
2388 offset, len, spd,
2389 skb_head_is_locked(skb),
2390 sk, pipe))
2391 return true;
2392
2393 /*
2394 * then map the fragments
2395 */
2396 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2397 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2398
2399 if (__splice_segment(skb_frag_page(f),
2400 skb_frag_off(f), skb_frag_size(f),
2401 offset, len, spd, false, sk, pipe))
2402 return true;
2403 }
2404
2405 skb_walk_frags(skb, iter) {
2406 if (*offset >= iter->len) {
2407 *offset -= iter->len;
2408 continue;
2409 }
2410 /* __skb_splice_bits() only fails if the output has no room
2411 * left, so no point in going over the frag_list for the error
2412 * case.
2413 */
2414 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2415 return true;
2416 }
2417
2418 return false;
2419}
2420
2421/*
2422 * Map data from the skb to a pipe. Should handle both the linear part,
2423 * the fragments, and the frag list.
2424 */
2425int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2426 struct pipe_inode_info *pipe, unsigned int tlen,
2427 unsigned int flags)
2428{
2429 struct partial_page partial[MAX_SKB_FRAGS];
2430 struct page *pages[MAX_SKB_FRAGS];
2431 struct splice_pipe_desc spd = {
2432 .pages = pages,
2433 .partial = partial,
2434 .nr_pages_max = MAX_SKB_FRAGS,
2435 .ops = &nosteal_pipe_buf_ops,
2436 .spd_release = sock_spd_release,
2437 };
2438 int ret = 0;
2439
2440 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2441
2442 if (spd.nr_pages)
2443 ret = splice_to_pipe(pipe, &spd);
2444
2445 return ret;
2446}
2447EXPORT_SYMBOL_GPL(skb_splice_bits);
2448
2449/* Send skb data on a socket. Socket must be locked. */
2450int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2451 int len)
2452{
2453 unsigned int orig_len = len;
2454 struct sk_buff *head = skb;
2455 unsigned short fragidx;
2456 int slen, ret;
2457
2458do_frag_list:
2459
2460 /* Deal with head data */
2461 while (offset < skb_headlen(skb) && len) {
2462 struct kvec kv;
2463 struct msghdr msg;
2464
2465 slen = min_t(int, len, skb_headlen(skb) - offset);
2466 kv.iov_base = skb->data + offset;
2467 kv.iov_len = slen;
2468 memset(&msg, 0, sizeof(msg));
2469 msg.msg_flags = MSG_DONTWAIT;
2470
2471 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2472 if (ret <= 0)
2473 goto error;
2474
2475 offset += ret;
2476 len -= ret;
2477 }
2478
2479 /* All the data was skb head? */
2480 if (!len)
2481 goto out;
2482
2483 /* Make offset relative to start of frags */
2484 offset -= skb_headlen(skb);
2485
2486 /* Find where we are in frag list */
2487 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2488 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2489
2490 if (offset < skb_frag_size(frag))
2491 break;
2492
2493 offset -= skb_frag_size(frag);
2494 }
2495
2496 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2497 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2498
2499 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2500
2501 while (slen) {
2502 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2503 skb_frag_off(frag) + offset,
2504 slen, MSG_DONTWAIT);
2505 if (ret <= 0)
2506 goto error;
2507
2508 len -= ret;
2509 offset += ret;
2510 slen -= ret;
2511 }
2512
2513 offset = 0;
2514 }
2515
2516 if (len) {
2517 /* Process any frag lists */
2518
2519 if (skb == head) {
2520 if (skb_has_frag_list(skb)) {
2521 skb = skb_shinfo(skb)->frag_list;
2522 goto do_frag_list;
2523 }
2524 } else if (skb->next) {
2525 skb = skb->next;
2526 goto do_frag_list;
2527 }
2528 }
2529
2530out:
2531 return orig_len - len;
2532
2533error:
2534 return orig_len == len ? ret : orig_len - len;
2535}
2536EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2537
2538/**
2539 * skb_store_bits - store bits from kernel buffer to skb
2540 * @skb: destination buffer
2541 * @offset: offset in destination
2542 * @from: source buffer
2543 * @len: number of bytes to copy
2544 *
2545 * Copy the specified number of bytes from the source buffer to the
2546 * destination skb. This function handles all the messy bits of
2547 * traversing fragment lists and such.
2548 */
2549
2550int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2551{
2552 int start = skb_headlen(skb);
2553 struct sk_buff *frag_iter;
2554 int i, copy;
2555
2556 if (offset > (int)skb->len - len)
2557 goto fault;
2558
2559 if ((copy = start - offset) > 0) {
2560 if (copy > len)
2561 copy = len;
2562 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2563 if ((len -= copy) == 0)
2564 return 0;
2565 offset += copy;
2566 from += copy;
2567 }
2568
2569 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2570 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2571 int end;
2572
2573 WARN_ON(start > offset + len);
2574
2575 end = start + skb_frag_size(frag);
2576 if ((copy = end - offset) > 0) {
2577 u32 p_off, p_len, copied;
2578 struct page *p;
2579 u8 *vaddr;
2580
2581 if (copy > len)
2582 copy = len;
2583
2584 skb_frag_foreach_page(frag,
2585 skb_frag_off(frag) + offset - start,
2586 copy, p, p_off, p_len, copied) {
2587 vaddr = kmap_atomic(p);
2588 memcpy(vaddr + p_off, from + copied, p_len);
2589 kunmap_atomic(vaddr);
2590 }
2591
2592 if ((len -= copy) == 0)
2593 return 0;
2594 offset += copy;
2595 from += copy;
2596 }
2597 start = end;
2598 }
2599
2600 skb_walk_frags(skb, frag_iter) {
2601 int end;
2602
2603 WARN_ON(start > offset + len);
2604
2605 end = start + frag_iter->len;
2606 if ((copy = end - offset) > 0) {
2607 if (copy > len)
2608 copy = len;
2609 if (skb_store_bits(frag_iter, offset - start,
2610 from, copy))
2611 goto fault;
2612 if ((len -= copy) == 0)
2613 return 0;
2614 offset += copy;
2615 from += copy;
2616 }
2617 start = end;
2618 }
2619 if (!len)
2620 return 0;
2621
2622fault:
2623 return -EFAULT;
2624}
2625EXPORT_SYMBOL(skb_store_bits);
2626
2627/* Checksum skb data. */
2628__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2629 __wsum csum, const struct skb_checksum_ops *ops)
2630{
2631 int start = skb_headlen(skb);
2632 int i, copy = start - offset;
2633 struct sk_buff *frag_iter;
2634 int pos = 0;
2635
2636 /* Checksum header. */
2637 if (copy > 0) {
2638 if (copy > len)
2639 copy = len;
2640 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2641 skb->data + offset, copy, csum);
2642 if ((len -= copy) == 0)
2643 return csum;
2644 offset += copy;
2645 pos = copy;
2646 }
2647
2648 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2649 int end;
2650 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2651
2652 WARN_ON(start > offset + len);
2653
2654 end = start + skb_frag_size(frag);
2655 if ((copy = end - offset) > 0) {
2656 u32 p_off, p_len, copied;
2657 struct page *p;
2658 __wsum csum2;
2659 u8 *vaddr;
2660
2661 if (copy > len)
2662 copy = len;
2663
2664 skb_frag_foreach_page(frag,
2665 skb_frag_off(frag) + offset - start,
2666 copy, p, p_off, p_len, copied) {
2667 vaddr = kmap_atomic(p);
2668 csum2 = INDIRECT_CALL_1(ops->update,
2669 csum_partial_ext,
2670 vaddr + p_off, p_len, 0);
2671 kunmap_atomic(vaddr);
2672 csum = INDIRECT_CALL_1(ops->combine,
2673 csum_block_add_ext, csum,
2674 csum2, pos, p_len);
2675 pos += p_len;
2676 }
2677
2678 if (!(len -= copy))
2679 return csum;
2680 offset += copy;
2681 }
2682 start = end;
2683 }
2684
2685 skb_walk_frags(skb, frag_iter) {
2686 int end;
2687
2688 WARN_ON(start > offset + len);
2689
2690 end = start + frag_iter->len;
2691 if ((copy = end - offset) > 0) {
2692 __wsum csum2;
2693 if (copy > len)
2694 copy = len;
2695 csum2 = __skb_checksum(frag_iter, offset - start,
2696 copy, 0, ops);
2697 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2698 csum, csum2, pos, copy);
2699 if ((len -= copy) == 0)
2700 return csum;
2701 offset += copy;
2702 pos += copy;
2703 }
2704 start = end;
2705 }
2706 BUG_ON(len);
2707
2708 return csum;
2709}
2710EXPORT_SYMBOL(__skb_checksum);
2711
2712__wsum skb_checksum(const struct sk_buff *skb, int offset,
2713 int len, __wsum csum)
2714{
2715 const struct skb_checksum_ops ops = {
2716 .update = csum_partial_ext,
2717 .combine = csum_block_add_ext,
2718 };
2719
2720 return __skb_checksum(skb, offset, len, csum, &ops);
2721}
2722EXPORT_SYMBOL(skb_checksum);
2723
2724/* Both of above in one bottle. */
2725
2726__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2727 u8 *to, int len, __wsum csum)
2728{
2729 int start = skb_headlen(skb);
2730 int i, copy = start - offset;
2731 struct sk_buff *frag_iter;
2732 int pos = 0;
2733
2734 /* Copy header. */
2735 if (copy > 0) {
2736 if (copy > len)
2737 copy = len;
2738 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2739 copy, csum);
2740 if ((len -= copy) == 0)
2741 return csum;
2742 offset += copy;
2743 to += copy;
2744 pos = copy;
2745 }
2746
2747 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2748 int end;
2749
2750 WARN_ON(start > offset + len);
2751
2752 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2753 if ((copy = end - offset) > 0) {
2754 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2755 u32 p_off, p_len, copied;
2756 struct page *p;
2757 __wsum csum2;
2758 u8 *vaddr;
2759
2760 if (copy > len)
2761 copy = len;
2762
2763 skb_frag_foreach_page(frag,
2764 skb_frag_off(frag) + offset - start,
2765 copy, p, p_off, p_len, copied) {
2766 vaddr = kmap_atomic(p);
2767 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2768 to + copied,
2769 p_len, 0);
2770 kunmap_atomic(vaddr);
2771 csum = csum_block_add(csum, csum2, pos);
2772 pos += p_len;
2773 }
2774
2775 if (!(len -= copy))
2776 return csum;
2777 offset += copy;
2778 to += copy;
2779 }
2780 start = end;
2781 }
2782
2783 skb_walk_frags(skb, frag_iter) {
2784 __wsum csum2;
2785 int end;
2786
2787 WARN_ON(start > offset + len);
2788
2789 end = start + frag_iter->len;
2790 if ((copy = end - offset) > 0) {
2791 if (copy > len)
2792 copy = len;
2793 csum2 = skb_copy_and_csum_bits(frag_iter,
2794 offset - start,
2795 to, copy, 0);
2796 csum = csum_block_add(csum, csum2, pos);
2797 if ((len -= copy) == 0)
2798 return csum;
2799 offset += copy;
2800 to += copy;
2801 pos += copy;
2802 }
2803 start = end;
2804 }
2805 BUG_ON(len);
2806 return csum;
2807}
2808EXPORT_SYMBOL(skb_copy_and_csum_bits);
2809
2810__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2811{
2812 __sum16 sum;
2813
2814 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2815 /* See comments in __skb_checksum_complete(). */
2816 if (likely(!sum)) {
2817 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2818 !skb->csum_complete_sw)
2819 netdev_rx_csum_fault(skb->dev, skb);
2820 }
2821 if (!skb_shared(skb))
2822 skb->csum_valid = !sum;
2823 return sum;
2824}
2825EXPORT_SYMBOL(__skb_checksum_complete_head);
2826
2827/* This function assumes skb->csum already holds pseudo header's checksum,
2828 * which has been changed from the hardware checksum, for example, by
2829 * __skb_checksum_validate_complete(). And, the original skb->csum must
2830 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2831 *
2832 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2833 * zero. The new checksum is stored back into skb->csum unless the skb is
2834 * shared.
2835 */
2836__sum16 __skb_checksum_complete(struct sk_buff *skb)
2837{
2838 __wsum csum;
2839 __sum16 sum;
2840
2841 csum = skb_checksum(skb, 0, skb->len, 0);
2842
2843 sum = csum_fold(csum_add(skb->csum, csum));
2844 /* This check is inverted, because we already knew the hardware
2845 * checksum is invalid before calling this function. So, if the
2846 * re-computed checksum is valid instead, then we have a mismatch
2847 * between the original skb->csum and skb_checksum(). This means either
2848 * the original hardware checksum is incorrect or we screw up skb->csum
2849 * when moving skb->data around.
2850 */
2851 if (likely(!sum)) {
2852 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2853 !skb->csum_complete_sw)
2854 netdev_rx_csum_fault(skb->dev, skb);
2855 }
2856
2857 if (!skb_shared(skb)) {
2858 /* Save full packet checksum */
2859 skb->csum = csum;
2860 skb->ip_summed = CHECKSUM_COMPLETE;
2861 skb->csum_complete_sw = 1;
2862 skb->csum_valid = !sum;
2863 }
2864
2865 return sum;
2866}
2867EXPORT_SYMBOL(__skb_checksum_complete);
2868
2869static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2870{
2871 net_warn_ratelimited(
2872 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2873 __func__);
2874 return 0;
2875}
2876
2877static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2878 int offset, int len)
2879{
2880 net_warn_ratelimited(
2881 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2882 __func__);
2883 return 0;
2884}
2885
2886static const struct skb_checksum_ops default_crc32c_ops = {
2887 .update = warn_crc32c_csum_update,
2888 .combine = warn_crc32c_csum_combine,
2889};
2890
2891const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2892 &default_crc32c_ops;
2893EXPORT_SYMBOL(crc32c_csum_stub);
2894
2895 /**
2896 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2897 * @from: source buffer
2898 *
2899 * Calculates the amount of linear headroom needed in the 'to' skb passed
2900 * into skb_zerocopy().
2901 */
2902unsigned int
2903skb_zerocopy_headlen(const struct sk_buff *from)
2904{
2905 unsigned int hlen = 0;
2906
2907 if (!from->head_frag ||
2908 skb_headlen(from) < L1_CACHE_BYTES ||
2909 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2910 hlen = skb_headlen(from);
2911
2912 if (skb_has_frag_list(from))
2913 hlen = from->len;
2914
2915 return hlen;
2916}
2917EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2918
2919/**
2920 * skb_zerocopy - Zero copy skb to skb
2921 * @to: destination buffer
2922 * @from: source buffer
2923 * @len: number of bytes to copy from source buffer
2924 * @hlen: size of linear headroom in destination buffer
2925 *
2926 * Copies up to `len` bytes from `from` to `to` by creating references
2927 * to the frags in the source buffer.
2928 *
2929 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2930 * headroom in the `to` buffer.
2931 *
2932 * Return value:
2933 * 0: everything is OK
2934 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2935 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2936 */
2937int
2938skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2939{
2940 int i, j = 0;
2941 int plen = 0; /* length of skb->head fragment */
2942 int ret;
2943 struct page *page;
2944 unsigned int offset;
2945
2946 BUG_ON(!from->head_frag && !hlen);
2947
2948 /* dont bother with small payloads */
2949 if (len <= skb_tailroom(to))
2950 return skb_copy_bits(from, 0, skb_put(to, len), len);
2951
2952 if (hlen) {
2953 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2954 if (unlikely(ret))
2955 return ret;
2956 len -= hlen;
2957 } else {
2958 plen = min_t(int, skb_headlen(from), len);
2959 if (plen) {
2960 page = virt_to_head_page(from->head);
2961 offset = from->data - (unsigned char *)page_address(page);
2962 __skb_fill_page_desc(to, 0, page, offset, plen);
2963 get_page(page);
2964 j = 1;
2965 len -= plen;
2966 }
2967 }
2968
2969 to->truesize += len + plen;
2970 to->len += len + plen;
2971 to->data_len += len + plen;
2972
2973 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2974 skb_tx_error(from);
2975 return -ENOMEM;
2976 }
2977 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2978
2979 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2980 int size;
2981
2982 if (!len)
2983 break;
2984 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2985 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2986 len);
2987 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2988 len -= size;
2989 skb_frag_ref(to, j);
2990 j++;
2991 }
2992 skb_shinfo(to)->nr_frags = j;
2993
2994 return 0;
2995}
2996EXPORT_SYMBOL_GPL(skb_zerocopy);
2997
2998void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2999{
3000 __wsum csum;
3001 long csstart;
3002
3003 if (skb->ip_summed == CHECKSUM_PARTIAL)
3004 csstart = skb_checksum_start_offset(skb);
3005 else
3006 csstart = skb_headlen(skb);
3007
3008 BUG_ON(csstart > skb_headlen(skb));
3009
3010 skb_copy_from_linear_data(skb, to, csstart);
3011
3012 csum = 0;
3013 if (csstart != skb->len)
3014 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3015 skb->len - csstart, 0);
3016
3017 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3018 long csstuff = csstart + skb->csum_offset;
3019
3020 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3021 }
3022}
3023EXPORT_SYMBOL(skb_copy_and_csum_dev);
3024
3025/**
3026 * skb_dequeue - remove from the head of the queue
3027 * @list: list to dequeue from
3028 *
3029 * Remove the head of the list. The list lock is taken so the function
3030 * may be used safely with other locking list functions. The head item is
3031 * returned or %NULL if the list is empty.
3032 */
3033
3034struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3035{
3036 unsigned long flags;
3037 struct sk_buff *result;
3038
3039 spin_lock_irqsave(&list->lock, flags);
3040 result = __skb_dequeue(list);
3041 spin_unlock_irqrestore(&list->lock, flags);
3042 return result;
3043}
3044EXPORT_SYMBOL(skb_dequeue);
3045
3046/**
3047 * skb_dequeue_tail - remove from the tail of the queue
3048 * @list: list to dequeue from
3049 *
3050 * Remove the tail of the list. The list lock is taken so the function
3051 * may be used safely with other locking list functions. The tail item is
3052 * returned or %NULL if the list is empty.
3053 */
3054struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3055{
3056 unsigned long flags;
3057 struct sk_buff *result;
3058
3059 spin_lock_irqsave(&list->lock, flags);
3060 result = __skb_dequeue_tail(list);
3061 spin_unlock_irqrestore(&list->lock, flags);
3062 return result;
3063}
3064EXPORT_SYMBOL(skb_dequeue_tail);
3065
3066/**
3067 * skb_queue_purge - empty a list
3068 * @list: list to empty
3069 *
3070 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3071 * the list and one reference dropped. This function takes the list
3072 * lock and is atomic with respect to other list locking functions.
3073 */
3074void skb_queue_purge(struct sk_buff_head *list)
3075{
3076 struct sk_buff *skb;
3077 while ((skb = skb_dequeue(list)) != NULL)
3078 kfree_skb(skb);
3079}
3080EXPORT_SYMBOL(skb_queue_purge);
3081
3082/**
3083 * skb_rbtree_purge - empty a skb rbtree
3084 * @root: root of the rbtree to empty
3085 * Return value: the sum of truesizes of all purged skbs.
3086 *
3087 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3088 * the list and one reference dropped. This function does not take
3089 * any lock. Synchronization should be handled by the caller (e.g., TCP
3090 * out-of-order queue is protected by the socket lock).
3091 */
3092unsigned int skb_rbtree_purge(struct rb_root *root)
3093{
3094 struct rb_node *p = rb_first(root);
3095 unsigned int sum = 0;
3096
3097 while (p) {
3098 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3099
3100 p = rb_next(p);
3101 rb_erase(&skb->rbnode, root);
3102 sum += skb->truesize;
3103 kfree_skb(skb);
3104 }
3105 return sum;
3106}
3107
3108/**
3109 * skb_queue_head - queue a buffer at the list head
3110 * @list: list to use
3111 * @newsk: buffer to queue
3112 *
3113 * Queue a buffer at the start of the list. This function takes the
3114 * list lock and can be used safely with other locking &sk_buff functions
3115 * safely.
3116 *
3117 * A buffer cannot be placed on two lists at the same time.
3118 */
3119void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3120{
3121 unsigned long flags;
3122
3123 spin_lock_irqsave(&list->lock, flags);
3124 __skb_queue_head(list, newsk);
3125 spin_unlock_irqrestore(&list->lock, flags);
3126}
3127EXPORT_SYMBOL(skb_queue_head);
3128
3129/**
3130 * skb_queue_tail - queue a buffer at the list tail
3131 * @list: list to use
3132 * @newsk: buffer to queue
3133 *
3134 * Queue a buffer at the tail of the list. This function takes the
3135 * list lock and can be used safely with other locking &sk_buff functions
3136 * safely.
3137 *
3138 * A buffer cannot be placed on two lists at the same time.
3139 */
3140void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3141{
3142 unsigned long flags;
3143
3144 spin_lock_irqsave(&list->lock, flags);
3145 __skb_queue_tail(list, newsk);
3146 spin_unlock_irqrestore(&list->lock, flags);
3147}
3148EXPORT_SYMBOL(skb_queue_tail);
3149
3150/**
3151 * skb_unlink - remove a buffer from a list
3152 * @skb: buffer to remove
3153 * @list: list to use
3154 *
3155 * Remove a packet from a list. The list locks are taken and this
3156 * function is atomic with respect to other list locked calls
3157 *
3158 * You must know what list the SKB is on.
3159 */
3160void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3161{
3162 unsigned long flags;
3163
3164 spin_lock_irqsave(&list->lock, flags);
3165 __skb_unlink(skb, list);
3166 spin_unlock_irqrestore(&list->lock, flags);
3167}
3168EXPORT_SYMBOL(skb_unlink);
3169
3170/**
3171 * skb_append - append a buffer
3172 * @old: buffer to insert after
3173 * @newsk: buffer to insert
3174 * @list: list to use
3175 *
3176 * Place a packet after a given packet in a list. The list locks are taken
3177 * and this function is atomic with respect to other list locked calls.
3178 * A buffer cannot be placed on two lists at the same time.
3179 */
3180void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3181{
3182 unsigned long flags;
3183
3184 spin_lock_irqsave(&list->lock, flags);
3185 __skb_queue_after(list, old, newsk);
3186 spin_unlock_irqrestore(&list->lock, flags);
3187}
3188EXPORT_SYMBOL(skb_append);
3189
3190static inline void skb_split_inside_header(struct sk_buff *skb,
3191 struct sk_buff* skb1,
3192 const u32 len, const int pos)
3193{
3194 int i;
3195
3196 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3197 pos - len);
3198 /* And move data appendix as is. */
3199 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3200 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3201
3202 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3203 skb_shinfo(skb)->nr_frags = 0;
3204 skb1->data_len = skb->data_len;
3205 skb1->len += skb1->data_len;
3206 skb->data_len = 0;
3207 skb->len = len;
3208 skb_set_tail_pointer(skb, len);
3209}
3210
3211static inline void skb_split_no_header(struct sk_buff *skb,
3212 struct sk_buff* skb1,
3213 const u32 len, int pos)
3214{
3215 int i, k = 0;
3216 const int nfrags = skb_shinfo(skb)->nr_frags;
3217
3218 skb_shinfo(skb)->nr_frags = 0;
3219 skb1->len = skb1->data_len = skb->len - len;
3220 skb->len = len;
3221 skb->data_len = len - pos;
3222
3223 for (i = 0; i < nfrags; i++) {
3224 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3225
3226 if (pos + size > len) {
3227 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3228
3229 if (pos < len) {
3230 /* Split frag.
3231 * We have two variants in this case:
3232 * 1. Move all the frag to the second
3233 * part, if it is possible. F.e.
3234 * this approach is mandatory for TUX,
3235 * where splitting is expensive.
3236 * 2. Split is accurately. We make this.
3237 */
3238 skb_frag_ref(skb, i);
3239 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3240 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3241 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3242 skb_shinfo(skb)->nr_frags++;
3243 }
3244 k++;
3245 } else
3246 skb_shinfo(skb)->nr_frags++;
3247 pos += size;
3248 }
3249 skb_shinfo(skb1)->nr_frags = k;
3250}
3251
3252/**
3253 * skb_split - Split fragmented skb to two parts at length len.
3254 * @skb: the buffer to split
3255 * @skb1: the buffer to receive the second part
3256 * @len: new length for skb
3257 */
3258void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3259{
3260 int pos = skb_headlen(skb);
3261
3262 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3263 SKBTX_SHARED_FRAG;
3264 skb_zerocopy_clone(skb1, skb, 0);
3265 if (len < pos) /* Split line is inside header. */
3266 skb_split_inside_header(skb, skb1, len, pos);
3267 else /* Second chunk has no header, nothing to copy. */
3268 skb_split_no_header(skb, skb1, len, pos);
3269}
3270EXPORT_SYMBOL(skb_split);
3271
3272/* Shifting from/to a cloned skb is a no-go.
3273 *
3274 * Caller cannot keep skb_shinfo related pointers past calling here!
3275 */
3276static int skb_prepare_for_shift(struct sk_buff *skb)
3277{
3278 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3279}
3280
3281/**
3282 * skb_shift - Shifts paged data partially from skb to another
3283 * @tgt: buffer into which tail data gets added
3284 * @skb: buffer from which the paged data comes from
3285 * @shiftlen: shift up to this many bytes
3286 *
3287 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3288 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3289 * It's up to caller to free skb if everything was shifted.
3290 *
3291 * If @tgt runs out of frags, the whole operation is aborted.
3292 *
3293 * Skb cannot include anything else but paged data while tgt is allowed
3294 * to have non-paged data as well.
3295 *
3296 * TODO: full sized shift could be optimized but that would need
3297 * specialized skb free'er to handle frags without up-to-date nr_frags.
3298 */
3299int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3300{
3301 int from, to, merge, todo;
3302 skb_frag_t *fragfrom, *fragto;
3303
3304 BUG_ON(shiftlen > skb->len);
3305
3306 if (skb_headlen(skb))
3307 return 0;
3308 if (skb_zcopy(tgt) || skb_zcopy(skb))
3309 return 0;
3310
3311 todo = shiftlen;
3312 from = 0;
3313 to = skb_shinfo(tgt)->nr_frags;
3314 fragfrom = &skb_shinfo(skb)->frags[from];
3315
3316 /* Actual merge is delayed until the point when we know we can
3317 * commit all, so that we don't have to undo partial changes
3318 */
3319 if (!to ||
3320 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3321 skb_frag_off(fragfrom))) {
3322 merge = -1;
3323 } else {
3324 merge = to - 1;
3325
3326 todo -= skb_frag_size(fragfrom);
3327 if (todo < 0) {
3328 if (skb_prepare_for_shift(skb) ||
3329 skb_prepare_for_shift(tgt))
3330 return 0;
3331
3332 /* All previous frag pointers might be stale! */
3333 fragfrom = &skb_shinfo(skb)->frags[from];
3334 fragto = &skb_shinfo(tgt)->frags[merge];
3335
3336 skb_frag_size_add(fragto, shiftlen);
3337 skb_frag_size_sub(fragfrom, shiftlen);
3338 skb_frag_off_add(fragfrom, shiftlen);
3339
3340 goto onlymerged;
3341 }
3342
3343 from++;
3344 }
3345
3346 /* Skip full, not-fitting skb to avoid expensive operations */
3347 if ((shiftlen == skb->len) &&
3348 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3349 return 0;
3350
3351 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3352 return 0;
3353
3354 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3355 if (to == MAX_SKB_FRAGS)
3356 return 0;
3357
3358 fragfrom = &skb_shinfo(skb)->frags[from];
3359 fragto = &skb_shinfo(tgt)->frags[to];
3360
3361 if (todo >= skb_frag_size(fragfrom)) {
3362 *fragto = *fragfrom;
3363 todo -= skb_frag_size(fragfrom);
3364 from++;
3365 to++;
3366
3367 } else {
3368 __skb_frag_ref(fragfrom);
3369 skb_frag_page_copy(fragto, fragfrom);
3370 skb_frag_off_copy(fragto, fragfrom);
3371 skb_frag_size_set(fragto, todo);
3372
3373 skb_frag_off_add(fragfrom, todo);
3374 skb_frag_size_sub(fragfrom, todo);
3375 todo = 0;
3376
3377 to++;
3378 break;
3379 }
3380 }
3381
3382 /* Ready to "commit" this state change to tgt */
3383 skb_shinfo(tgt)->nr_frags = to;
3384
3385 if (merge >= 0) {
3386 fragfrom = &skb_shinfo(skb)->frags[0];
3387 fragto = &skb_shinfo(tgt)->frags[merge];
3388
3389 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3390 __skb_frag_unref(fragfrom);
3391 }
3392
3393 /* Reposition in the original skb */
3394 to = 0;
3395 while (from < skb_shinfo(skb)->nr_frags)
3396 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3397 skb_shinfo(skb)->nr_frags = to;
3398
3399 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3400
3401onlymerged:
3402 /* Most likely the tgt won't ever need its checksum anymore, skb on
3403 * the other hand might need it if it needs to be resent
3404 */
3405 tgt->ip_summed = CHECKSUM_PARTIAL;
3406 skb->ip_summed = CHECKSUM_PARTIAL;
3407
3408 /* Yak, is it really working this way? Some helper please? */
3409 skb->len -= shiftlen;
3410 skb->data_len -= shiftlen;
3411 skb->truesize -= shiftlen;
3412 tgt->len += shiftlen;
3413 tgt->data_len += shiftlen;
3414 tgt->truesize += shiftlen;
3415
3416 return shiftlen;
3417}
3418
3419/**
3420 * skb_prepare_seq_read - Prepare a sequential read of skb data
3421 * @skb: the buffer to read
3422 * @from: lower offset of data to be read
3423 * @to: upper offset of data to be read
3424 * @st: state variable
3425 *
3426 * Initializes the specified state variable. Must be called before
3427 * invoking skb_seq_read() for the first time.
3428 */
3429void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3430 unsigned int to, struct skb_seq_state *st)
3431{
3432 st->lower_offset = from;
3433 st->upper_offset = to;
3434 st->root_skb = st->cur_skb = skb;
3435 st->frag_idx = st->stepped_offset = 0;
3436 st->frag_data = NULL;
3437}
3438EXPORT_SYMBOL(skb_prepare_seq_read);
3439
3440/**
3441 * skb_seq_read - Sequentially read skb data
3442 * @consumed: number of bytes consumed by the caller so far
3443 * @data: destination pointer for data to be returned
3444 * @st: state variable
3445 *
3446 * Reads a block of skb data at @consumed relative to the
3447 * lower offset specified to skb_prepare_seq_read(). Assigns
3448 * the head of the data block to @data and returns the length
3449 * of the block or 0 if the end of the skb data or the upper
3450 * offset has been reached.
3451 *
3452 * The caller is not required to consume all of the data
3453 * returned, i.e. @consumed is typically set to the number
3454 * of bytes already consumed and the next call to
3455 * skb_seq_read() will return the remaining part of the block.
3456 *
3457 * Note 1: The size of each block of data returned can be arbitrary,
3458 * this limitation is the cost for zerocopy sequential
3459 * reads of potentially non linear data.
3460 *
3461 * Note 2: Fragment lists within fragments are not implemented
3462 * at the moment, state->root_skb could be replaced with
3463 * a stack for this purpose.
3464 */
3465unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3466 struct skb_seq_state *st)
3467{
3468 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3469 skb_frag_t *frag;
3470
3471 if (unlikely(abs_offset >= st->upper_offset)) {
3472 if (st->frag_data) {
3473 kunmap_atomic(st->frag_data);
3474 st->frag_data = NULL;
3475 }
3476 return 0;
3477 }
3478
3479next_skb:
3480 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3481
3482 if (abs_offset < block_limit && !st->frag_data) {
3483 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3484 return block_limit - abs_offset;
3485 }
3486
3487 if (st->frag_idx == 0 && !st->frag_data)
3488 st->stepped_offset += skb_headlen(st->cur_skb);
3489
3490 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3491 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3492 block_limit = skb_frag_size(frag) + st->stepped_offset;
3493
3494 if (abs_offset < block_limit) {
3495 if (!st->frag_data)
3496 st->frag_data = kmap_atomic(skb_frag_page(frag));
3497
3498 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3499 (abs_offset - st->stepped_offset);
3500
3501 return block_limit - abs_offset;
3502 }
3503
3504 if (st->frag_data) {
3505 kunmap_atomic(st->frag_data);
3506 st->frag_data = NULL;
3507 }
3508
3509 st->frag_idx++;
3510 st->stepped_offset += skb_frag_size(frag);
3511 }
3512
3513 if (st->frag_data) {
3514 kunmap_atomic(st->frag_data);
3515 st->frag_data = NULL;
3516 }
3517
3518 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3519 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3520 st->frag_idx = 0;
3521 goto next_skb;
3522 } else if (st->cur_skb->next) {
3523 st->cur_skb = st->cur_skb->next;
3524 st->frag_idx = 0;
3525 goto next_skb;
3526 }
3527
3528 return 0;
3529}
3530EXPORT_SYMBOL(skb_seq_read);
3531
3532/**
3533 * skb_abort_seq_read - Abort a sequential read of skb data
3534 * @st: state variable
3535 *
3536 * Must be called if skb_seq_read() was not called until it
3537 * returned 0.
3538 */
3539void skb_abort_seq_read(struct skb_seq_state *st)
3540{
3541 if (st->frag_data)
3542 kunmap_atomic(st->frag_data);
3543}
3544EXPORT_SYMBOL(skb_abort_seq_read);
3545
3546#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3547
3548static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3549 struct ts_config *conf,
3550 struct ts_state *state)
3551{
3552 return skb_seq_read(offset, text, TS_SKB_CB(state));
3553}
3554
3555static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3556{
3557 skb_abort_seq_read(TS_SKB_CB(state));
3558}
3559
3560/**
3561 * skb_find_text - Find a text pattern in skb data
3562 * @skb: the buffer to look in
3563 * @from: search offset
3564 * @to: search limit
3565 * @config: textsearch configuration
3566 *
3567 * Finds a pattern in the skb data according to the specified
3568 * textsearch configuration. Use textsearch_next() to retrieve
3569 * subsequent occurrences of the pattern. Returns the offset
3570 * to the first occurrence or UINT_MAX if no match was found.
3571 */
3572unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3573 unsigned int to, struct ts_config *config)
3574{
3575 struct ts_state state;
3576 unsigned int ret;
3577
3578 config->get_next_block = skb_ts_get_next_block;
3579 config->finish = skb_ts_finish;
3580
3581 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3582
3583 ret = textsearch_find(config, &state);
3584 return (ret <= to - from ? ret : UINT_MAX);
3585}
3586EXPORT_SYMBOL(skb_find_text);
3587
3588int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3589 int offset, size_t size)
3590{
3591 int i = skb_shinfo(skb)->nr_frags;
3592
3593 if (skb_can_coalesce(skb, i, page, offset)) {
3594 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3595 } else if (i < MAX_SKB_FRAGS) {
3596 get_page(page);
3597 skb_fill_page_desc(skb, i, page, offset, size);
3598 } else {
3599 return -EMSGSIZE;
3600 }
3601
3602 return 0;
3603}
3604EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3605
3606/**
3607 * skb_pull_rcsum - pull skb and update receive checksum
3608 * @skb: buffer to update
3609 * @len: length of data pulled
3610 *
3611 * This function performs an skb_pull on the packet and updates
3612 * the CHECKSUM_COMPLETE checksum. It should be used on
3613 * receive path processing instead of skb_pull unless you know
3614 * that the checksum difference is zero (e.g., a valid IP header)
3615 * or you are setting ip_summed to CHECKSUM_NONE.
3616 */
3617void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3618{
3619 unsigned char *data = skb->data;
3620
3621 BUG_ON(len > skb->len);
3622 __skb_pull(skb, len);
3623 skb_postpull_rcsum(skb, data, len);
3624 return skb->data;
3625}
3626EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3627
3628static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3629{
3630 skb_frag_t head_frag;
3631 struct page *page;
3632
3633 page = virt_to_head_page(frag_skb->head);
3634 __skb_frag_set_page(&head_frag, page);
3635 skb_frag_off_set(&head_frag, frag_skb->data -
3636 (unsigned char *)page_address(page));
3637 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3638 return head_frag;
3639}
3640
3641/**
3642 * skb_segment - Perform protocol segmentation on skb.
3643 * @head_skb: buffer to segment
3644 * @features: features for the output path (see dev->features)
3645 *
3646 * This function performs segmentation on the given skb. It returns
3647 * a pointer to the first in a list of new skbs for the segments.
3648 * In case of error it returns ERR_PTR(err).
3649 */
3650struct sk_buff *skb_segment(struct sk_buff *head_skb,
3651 netdev_features_t features)
3652{
3653 struct sk_buff *segs = NULL;
3654 struct sk_buff *tail = NULL;
3655 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3656 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3657 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3658 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3659 struct sk_buff *frag_skb = head_skb;
3660 unsigned int offset = doffset;
3661 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3662 unsigned int partial_segs = 0;
3663 unsigned int headroom;
3664 unsigned int len = head_skb->len;
3665 __be16 proto;
3666 bool csum, sg;
3667 int nfrags = skb_shinfo(head_skb)->nr_frags;
3668 int err = -ENOMEM;
3669 int i = 0;
3670 int pos;
3671 int dummy;
3672
3673 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3674 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3675 /* gso_size is untrusted, and we have a frag_list with a linear
3676 * non head_frag head.
3677 *
3678 * (we assume checking the first list_skb member suffices;
3679 * i.e if either of the list_skb members have non head_frag
3680 * head, then the first one has too).
3681 *
3682 * If head_skb's headlen does not fit requested gso_size, it
3683 * means that the frag_list members do NOT terminate on exact
3684 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3685 * sharing. Therefore we must fallback to copying the frag_list
3686 * skbs; we do so by disabling SG.
3687 */
3688 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3689 features &= ~NETIF_F_SG;
3690 }
3691
3692 __skb_push(head_skb, doffset);
3693 proto = skb_network_protocol(head_skb, &dummy);
3694 if (unlikely(!proto))
3695 return ERR_PTR(-EINVAL);
3696
3697 sg = !!(features & NETIF_F_SG);
3698 csum = !!can_checksum_protocol(features, proto);
3699
3700 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3701 if (!(features & NETIF_F_GSO_PARTIAL)) {
3702 struct sk_buff *iter;
3703 unsigned int frag_len;
3704
3705 if (!list_skb ||
3706 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3707 goto normal;
3708
3709 /* If we get here then all the required
3710 * GSO features except frag_list are supported.
3711 * Try to split the SKB to multiple GSO SKBs
3712 * with no frag_list.
3713 * Currently we can do that only when the buffers don't
3714 * have a linear part and all the buffers except
3715 * the last are of the same length.
3716 */
3717 frag_len = list_skb->len;
3718 skb_walk_frags(head_skb, iter) {
3719 if (frag_len != iter->len && iter->next)
3720 goto normal;
3721 if (skb_headlen(iter) && !iter->head_frag)
3722 goto normal;
3723
3724 len -= iter->len;
3725 }
3726
3727 if (len != frag_len)
3728 goto normal;
3729 }
3730
3731 /* GSO partial only requires that we trim off any excess that
3732 * doesn't fit into an MSS sized block, so take care of that
3733 * now.
3734 */
3735 partial_segs = len / mss;
3736 if (partial_segs > 1)
3737 mss *= partial_segs;
3738 else
3739 partial_segs = 0;
3740 }
3741
3742normal:
3743 headroom = skb_headroom(head_skb);
3744 pos = skb_headlen(head_skb);
3745
3746 do {
3747 struct sk_buff *nskb;
3748 skb_frag_t *nskb_frag;
3749 int hsize;
3750 int size;
3751
3752 if (unlikely(mss == GSO_BY_FRAGS)) {
3753 len = list_skb->len;
3754 } else {
3755 len = head_skb->len - offset;
3756 if (len > mss)
3757 len = mss;
3758 }
3759
3760 hsize = skb_headlen(head_skb) - offset;
3761 if (hsize < 0)
3762 hsize = 0;
3763 if (hsize > len || !sg)
3764 hsize = len;
3765
3766 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3767 (skb_headlen(list_skb) == len || sg)) {
3768 BUG_ON(skb_headlen(list_skb) > len);
3769
3770 i = 0;
3771 nfrags = skb_shinfo(list_skb)->nr_frags;
3772 frag = skb_shinfo(list_skb)->frags;
3773 frag_skb = list_skb;
3774 pos += skb_headlen(list_skb);
3775
3776 while (pos < offset + len) {
3777 BUG_ON(i >= nfrags);
3778
3779 size = skb_frag_size(frag);
3780 if (pos + size > offset + len)
3781 break;
3782
3783 i++;
3784 pos += size;
3785 frag++;
3786 }
3787
3788 nskb = skb_clone(list_skb, GFP_ATOMIC);
3789 list_skb = list_skb->next;
3790
3791 if (unlikely(!nskb))
3792 goto err;
3793
3794 if (unlikely(pskb_trim(nskb, len))) {
3795 kfree_skb(nskb);
3796 goto err;
3797 }
3798
3799 hsize = skb_end_offset(nskb);
3800 if (skb_cow_head(nskb, doffset + headroom)) {
3801 kfree_skb(nskb);
3802 goto err;
3803 }
3804
3805 nskb->truesize += skb_end_offset(nskb) - hsize;
3806 skb_release_head_state(nskb);
3807 __skb_push(nskb, doffset);
3808 } else {
3809 nskb = __alloc_skb(hsize + doffset + headroom,
3810 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3811 NUMA_NO_NODE);
3812
3813 if (unlikely(!nskb))
3814 goto err;
3815
3816 skb_reserve(nskb, headroom);
3817 __skb_put(nskb, doffset);
3818 }
3819
3820 if (segs)
3821 tail->next = nskb;
3822 else
3823 segs = nskb;
3824 tail = nskb;
3825
3826 __copy_skb_header(nskb, head_skb);
3827
3828 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3829 skb_reset_mac_len(nskb);
3830
3831 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3832 nskb->data - tnl_hlen,
3833 doffset + tnl_hlen);
3834
3835 if (nskb->len == len + doffset)
3836 goto perform_csum_check;
3837
3838 if (!sg) {
3839 if (!nskb->remcsum_offload)
3840 nskb->ip_summed = CHECKSUM_NONE;
3841 SKB_GSO_CB(nskb)->csum =
3842 skb_copy_and_csum_bits(head_skb, offset,
3843 skb_put(nskb, len),
3844 len, 0);
3845 SKB_GSO_CB(nskb)->csum_start =
3846 skb_headroom(nskb) + doffset;
3847 continue;
3848 }
3849
3850 nskb_frag = skb_shinfo(nskb)->frags;
3851
3852 skb_copy_from_linear_data_offset(head_skb, offset,
3853 skb_put(nskb, hsize), hsize);
3854
3855 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3856 SKBTX_SHARED_FRAG;
3857
3858 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3859 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3860 goto err;
3861
3862 while (pos < offset + len) {
3863 if (i >= nfrags) {
3864 i = 0;
3865 nfrags = skb_shinfo(list_skb)->nr_frags;
3866 frag = skb_shinfo(list_skb)->frags;
3867 frag_skb = list_skb;
3868 if (!skb_headlen(list_skb)) {
3869 BUG_ON(!nfrags);
3870 } else {
3871 BUG_ON(!list_skb->head_frag);
3872
3873 /* to make room for head_frag. */
3874 i--;
3875 frag--;
3876 }
3877 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3878 skb_zerocopy_clone(nskb, frag_skb,
3879 GFP_ATOMIC))
3880 goto err;
3881
3882 list_skb = list_skb->next;
3883 }
3884
3885 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3886 MAX_SKB_FRAGS)) {
3887 net_warn_ratelimited(
3888 "skb_segment: too many frags: %u %u\n",
3889 pos, mss);
3890 err = -EINVAL;
3891 goto err;
3892 }
3893
3894 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3895 __skb_frag_ref(nskb_frag);
3896 size = skb_frag_size(nskb_frag);
3897
3898 if (pos < offset) {
3899 skb_frag_off_add(nskb_frag, offset - pos);
3900 skb_frag_size_sub(nskb_frag, offset - pos);
3901 }
3902
3903 skb_shinfo(nskb)->nr_frags++;
3904
3905 if (pos + size <= offset + len) {
3906 i++;
3907 frag++;
3908 pos += size;
3909 } else {
3910 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3911 goto skip_fraglist;
3912 }
3913
3914 nskb_frag++;
3915 }
3916
3917skip_fraglist:
3918 nskb->data_len = len - hsize;
3919 nskb->len += nskb->data_len;
3920 nskb->truesize += nskb->data_len;
3921
3922perform_csum_check:
3923 if (!csum) {
3924 if (skb_has_shared_frag(nskb) &&
3925 __skb_linearize(nskb))
3926 goto err;
3927
3928 if (!nskb->remcsum_offload)
3929 nskb->ip_summed = CHECKSUM_NONE;
3930 SKB_GSO_CB(nskb)->csum =
3931 skb_checksum(nskb, doffset,
3932 nskb->len - doffset, 0);
3933 SKB_GSO_CB(nskb)->csum_start =
3934 skb_headroom(nskb) + doffset;
3935 }
3936 } while ((offset += len) < head_skb->len);
3937
3938 /* Some callers want to get the end of the list.
3939 * Put it in segs->prev to avoid walking the list.
3940 * (see validate_xmit_skb_list() for example)
3941 */
3942 segs->prev = tail;
3943
3944 if (partial_segs) {
3945 struct sk_buff *iter;
3946 int type = skb_shinfo(head_skb)->gso_type;
3947 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3948
3949 /* Update type to add partial and then remove dodgy if set */
3950 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3951 type &= ~SKB_GSO_DODGY;
3952
3953 /* Update GSO info and prepare to start updating headers on
3954 * our way back down the stack of protocols.
3955 */
3956 for (iter = segs; iter; iter = iter->next) {
3957 skb_shinfo(iter)->gso_size = gso_size;
3958 skb_shinfo(iter)->gso_segs = partial_segs;
3959 skb_shinfo(iter)->gso_type = type;
3960 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3961 }
3962
3963 if (tail->len - doffset <= gso_size)
3964 skb_shinfo(tail)->gso_size = 0;
3965 else if (tail != segs)
3966 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3967 }
3968
3969 /* Following permits correct backpressure, for protocols
3970 * using skb_set_owner_w().
3971 * Idea is to tranfert ownership from head_skb to last segment.
3972 */
3973 if (head_skb->destructor == sock_wfree) {
3974 swap(tail->truesize, head_skb->truesize);
3975 swap(tail->destructor, head_skb->destructor);
3976 swap(tail->sk, head_skb->sk);
3977 }
3978 return segs;
3979
3980err:
3981 kfree_skb_list(segs);
3982 return ERR_PTR(err);
3983}
3984EXPORT_SYMBOL_GPL(skb_segment);
3985
3986int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
3987{
3988 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3989 unsigned int offset = skb_gro_offset(skb);
3990 unsigned int headlen = skb_headlen(skb);
3991 unsigned int len = skb_gro_len(skb);
3992 unsigned int delta_truesize;
3993 struct sk_buff *lp;
3994
3995 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
3996 return -E2BIG;
3997
3998 lp = NAPI_GRO_CB(p)->last;
3999 pinfo = skb_shinfo(lp);
4000
4001 if (headlen <= offset) {
4002 skb_frag_t *frag;
4003 skb_frag_t *frag2;
4004 int i = skbinfo->nr_frags;
4005 int nr_frags = pinfo->nr_frags + i;
4006
4007 if (nr_frags > MAX_SKB_FRAGS)
4008 goto merge;
4009
4010 offset -= headlen;
4011 pinfo->nr_frags = nr_frags;
4012 skbinfo->nr_frags = 0;
4013
4014 frag = pinfo->frags + nr_frags;
4015 frag2 = skbinfo->frags + i;
4016 do {
4017 *--frag = *--frag2;
4018 } while (--i);
4019
4020 skb_frag_off_add(frag, offset);
4021 skb_frag_size_sub(frag, offset);
4022
4023 /* all fragments truesize : remove (head size + sk_buff) */
4024 delta_truesize = skb->truesize -
4025 SKB_TRUESIZE(skb_end_offset(skb));
4026
4027 skb->truesize -= skb->data_len;
4028 skb->len -= skb->data_len;
4029 skb->data_len = 0;
4030
4031 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4032 goto done;
4033 } else if (skb->head_frag) {
4034 int nr_frags = pinfo->nr_frags;
4035 skb_frag_t *frag = pinfo->frags + nr_frags;
4036 struct page *page = virt_to_head_page(skb->head);
4037 unsigned int first_size = headlen - offset;
4038 unsigned int first_offset;
4039
4040 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4041 goto merge;
4042
4043 first_offset = skb->data -
4044 (unsigned char *)page_address(page) +
4045 offset;
4046
4047 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4048
4049 __skb_frag_set_page(frag, page);
4050 skb_frag_off_set(frag, first_offset);
4051 skb_frag_size_set(frag, first_size);
4052
4053 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4054 /* We dont need to clear skbinfo->nr_frags here */
4055
4056 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4057 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4058 goto done;
4059 }
4060
4061merge:
4062 delta_truesize = skb->truesize;
4063 if (offset > headlen) {
4064 unsigned int eat = offset - headlen;
4065
4066 skb_frag_off_add(&skbinfo->frags[0], eat);
4067 skb_frag_size_sub(&skbinfo->frags[0], eat);
4068 skb->data_len -= eat;
4069 skb->len -= eat;
4070 offset = headlen;
4071 }
4072
4073 __skb_pull(skb, offset);
4074
4075 if (NAPI_GRO_CB(p)->last == p)
4076 skb_shinfo(p)->frag_list = skb;
4077 else
4078 NAPI_GRO_CB(p)->last->next = skb;
4079 NAPI_GRO_CB(p)->last = skb;
4080 __skb_header_release(skb);
4081 lp = p;
4082
4083done:
4084 NAPI_GRO_CB(p)->count++;
4085 p->data_len += len;
4086 p->truesize += delta_truesize;
4087 p->len += len;
4088 if (lp != p) {
4089 lp->data_len += len;
4090 lp->truesize += delta_truesize;
4091 lp->len += len;
4092 }
4093 NAPI_GRO_CB(skb)->same_flow = 1;
4094 return 0;
4095}
4096EXPORT_SYMBOL_GPL(skb_gro_receive);
4097
4098#ifdef CONFIG_SKB_EXTENSIONS
4099#define SKB_EXT_ALIGN_VALUE 8
4100#define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4101
4102static const u8 skb_ext_type_len[] = {
4103#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4104 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4105#endif
4106#ifdef CONFIG_XFRM
4107 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4108#endif
4109#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4110 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4111#endif
4112};
4113
4114static __always_inline unsigned int skb_ext_total_length(void)
4115{
4116 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4117#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4118 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4119#endif
4120#ifdef CONFIG_XFRM
4121 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4122#endif
4123#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4124 skb_ext_type_len[TC_SKB_EXT] +
4125#endif
4126 0;
4127}
4128
4129static void skb_extensions_init(void)
4130{
4131 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4132 BUILD_BUG_ON(skb_ext_total_length() > 255);
4133
4134 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4135 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4136 0,
4137 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4138 NULL);
4139}
4140#else
4141static void skb_extensions_init(void) {}
4142#endif
4143
4144void __init skb_init(void)
4145{
4146 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4147 sizeof(struct sk_buff),
4148 0,
4149 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4150 offsetof(struct sk_buff, cb),
4151 sizeof_field(struct sk_buff, cb),
4152 NULL);
4153 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4154 sizeof(struct sk_buff_fclones),
4155 0,
4156 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4157 NULL);
4158 skb_extensions_init();
4159}
4160
4161static int
4162__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4163 unsigned int recursion_level)
4164{
4165 int start = skb_headlen(skb);
4166 int i, copy = start - offset;
4167 struct sk_buff *frag_iter;
4168 int elt = 0;
4169
4170 if (unlikely(recursion_level >= 24))
4171 return -EMSGSIZE;
4172
4173 if (copy > 0) {
4174 if (copy > len)
4175 copy = len;
4176 sg_set_buf(sg, skb->data + offset, copy);
4177 elt++;
4178 if ((len -= copy) == 0)
4179 return elt;
4180 offset += copy;
4181 }
4182
4183 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4184 int end;
4185
4186 WARN_ON(start > offset + len);
4187
4188 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4189 if ((copy = end - offset) > 0) {
4190 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4191 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4192 return -EMSGSIZE;
4193
4194 if (copy > len)
4195 copy = len;
4196 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4197 skb_frag_off(frag) + offset - start);
4198 elt++;
4199 if (!(len -= copy))
4200 return elt;
4201 offset += copy;
4202 }
4203 start = end;
4204 }
4205
4206 skb_walk_frags(skb, frag_iter) {
4207 int end, ret;
4208
4209 WARN_ON(start > offset + len);
4210
4211 end = start + frag_iter->len;
4212 if ((copy = end - offset) > 0) {
4213 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4214 return -EMSGSIZE;
4215
4216 if (copy > len)
4217 copy = len;
4218 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4219 copy, recursion_level + 1);
4220 if (unlikely(ret < 0))
4221 return ret;
4222 elt += ret;
4223 if ((len -= copy) == 0)
4224 return elt;
4225 offset += copy;
4226 }
4227 start = end;
4228 }
4229 BUG_ON(len);
4230 return elt;
4231}
4232
4233/**
4234 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4235 * @skb: Socket buffer containing the buffers to be mapped
4236 * @sg: The scatter-gather list to map into
4237 * @offset: The offset into the buffer's contents to start mapping
4238 * @len: Length of buffer space to be mapped
4239 *
4240 * Fill the specified scatter-gather list with mappings/pointers into a
4241 * region of the buffer space attached to a socket buffer. Returns either
4242 * the number of scatterlist items used, or -EMSGSIZE if the contents
4243 * could not fit.
4244 */
4245int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4246{
4247 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4248
4249 if (nsg <= 0)
4250 return nsg;
4251
4252 sg_mark_end(&sg[nsg - 1]);
4253
4254 return nsg;
4255}
4256EXPORT_SYMBOL_GPL(skb_to_sgvec);
4257
4258/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4259 * sglist without mark the sg which contain last skb data as the end.
4260 * So the caller can mannipulate sg list as will when padding new data after
4261 * the first call without calling sg_unmark_end to expend sg list.
4262 *
4263 * Scenario to use skb_to_sgvec_nomark:
4264 * 1. sg_init_table
4265 * 2. skb_to_sgvec_nomark(payload1)
4266 * 3. skb_to_sgvec_nomark(payload2)
4267 *
4268 * This is equivalent to:
4269 * 1. sg_init_table
4270 * 2. skb_to_sgvec(payload1)
4271 * 3. sg_unmark_end
4272 * 4. skb_to_sgvec(payload2)
4273 *
4274 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4275 * is more preferable.
4276 */
4277int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4278 int offset, int len)
4279{
4280 return __skb_to_sgvec(skb, sg, offset, len, 0);
4281}
4282EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4283
4284
4285
4286/**
4287 * skb_cow_data - Check that a socket buffer's data buffers are writable
4288 * @skb: The socket buffer to check.
4289 * @tailbits: Amount of trailing space to be added
4290 * @trailer: Returned pointer to the skb where the @tailbits space begins
4291 *
4292 * Make sure that the data buffers attached to a socket buffer are
4293 * writable. If they are not, private copies are made of the data buffers
4294 * and the socket buffer is set to use these instead.
4295 *
4296 * If @tailbits is given, make sure that there is space to write @tailbits
4297 * bytes of data beyond current end of socket buffer. @trailer will be
4298 * set to point to the skb in which this space begins.
4299 *
4300 * The number of scatterlist elements required to completely map the
4301 * COW'd and extended socket buffer will be returned.
4302 */
4303int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4304{
4305 int copyflag;
4306 int elt;
4307 struct sk_buff *skb1, **skb_p;
4308
4309 /* If skb is cloned or its head is paged, reallocate
4310 * head pulling out all the pages (pages are considered not writable
4311 * at the moment even if they are anonymous).
4312 */
4313 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4314 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4315 return -ENOMEM;
4316
4317 /* Easy case. Most of packets will go this way. */
4318 if (!skb_has_frag_list(skb)) {
4319 /* A little of trouble, not enough of space for trailer.
4320 * This should not happen, when stack is tuned to generate
4321 * good frames. OK, on miss we reallocate and reserve even more
4322 * space, 128 bytes is fair. */
4323
4324 if (skb_tailroom(skb) < tailbits &&
4325 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4326 return -ENOMEM;
4327
4328 /* Voila! */
4329 *trailer = skb;
4330 return 1;
4331 }
4332
4333 /* Misery. We are in troubles, going to mincer fragments... */
4334
4335 elt = 1;
4336 skb_p = &skb_shinfo(skb)->frag_list;
4337 copyflag = 0;
4338
4339 while ((skb1 = *skb_p) != NULL) {
4340 int ntail = 0;
4341
4342 /* The fragment is partially pulled by someone,
4343 * this can happen on input. Copy it and everything
4344 * after it. */
4345
4346 if (skb_shared(skb1))
4347 copyflag = 1;
4348
4349 /* If the skb is the last, worry about trailer. */
4350
4351 if (skb1->next == NULL && tailbits) {
4352 if (skb_shinfo(skb1)->nr_frags ||
4353 skb_has_frag_list(skb1) ||
4354 skb_tailroom(skb1) < tailbits)
4355 ntail = tailbits + 128;
4356 }
4357
4358 if (copyflag ||
4359 skb_cloned(skb1) ||
4360 ntail ||
4361 skb_shinfo(skb1)->nr_frags ||
4362 skb_has_frag_list(skb1)) {
4363 struct sk_buff *skb2;
4364
4365 /* Fuck, we are miserable poor guys... */
4366 if (ntail == 0)
4367 skb2 = skb_copy(skb1, GFP_ATOMIC);
4368 else
4369 skb2 = skb_copy_expand(skb1,
4370 skb_headroom(skb1),
4371 ntail,
4372 GFP_ATOMIC);
4373 if (unlikely(skb2 == NULL))
4374 return -ENOMEM;
4375
4376 if (skb1->sk)
4377 skb_set_owner_w(skb2, skb1->sk);
4378
4379 /* Looking around. Are we still alive?
4380 * OK, link new skb, drop old one */
4381
4382 skb2->next = skb1->next;
4383 *skb_p = skb2;
4384 kfree_skb(skb1);
4385 skb1 = skb2;
4386 }
4387 elt++;
4388 *trailer = skb1;
4389 skb_p = &skb1->next;
4390 }
4391
4392 return elt;
4393}
4394EXPORT_SYMBOL_GPL(skb_cow_data);
4395
4396static void sock_rmem_free(struct sk_buff *skb)
4397{
4398 struct sock *sk = skb->sk;
4399
4400 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4401}
4402
4403static void skb_set_err_queue(struct sk_buff *skb)
4404{
4405 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4406 * So, it is safe to (mis)use it to mark skbs on the error queue.
4407 */
4408 skb->pkt_type = PACKET_OUTGOING;
4409 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4410}
4411
4412/*
4413 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4414 */
4415int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4416{
4417 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4418 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4419 return -ENOMEM;
4420
4421 skb_orphan(skb);
4422 skb->sk = sk;
4423 skb->destructor = sock_rmem_free;
4424 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4425 skb_set_err_queue(skb);
4426
4427 /* before exiting rcu section, make sure dst is refcounted */
4428 skb_dst_force(skb);
4429
4430 skb_queue_tail(&sk->sk_error_queue, skb);
4431 if (!sock_flag(sk, SOCK_DEAD))
4432 sk->sk_error_report(sk);
4433 return 0;
4434}
4435EXPORT_SYMBOL(sock_queue_err_skb);
4436
4437static bool is_icmp_err_skb(const struct sk_buff *skb)
4438{
4439 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4440 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4441}
4442
4443struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4444{
4445 struct sk_buff_head *q = &sk->sk_error_queue;
4446 struct sk_buff *skb, *skb_next = NULL;
4447 bool icmp_next = false;
4448 unsigned long flags;
4449
4450 spin_lock_irqsave(&q->lock, flags);
4451 skb = __skb_dequeue(q);
4452 if (skb && (skb_next = skb_peek(q))) {
4453 icmp_next = is_icmp_err_skb(skb_next);
4454 if (icmp_next)
4455 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4456 }
4457 spin_unlock_irqrestore(&q->lock, flags);
4458
4459 if (is_icmp_err_skb(skb) && !icmp_next)
4460 sk->sk_err = 0;
4461
4462 if (skb_next)
4463 sk->sk_error_report(sk);
4464
4465 return skb;
4466}
4467EXPORT_SYMBOL(sock_dequeue_err_skb);
4468
4469/**
4470 * skb_clone_sk - create clone of skb, and take reference to socket
4471 * @skb: the skb to clone
4472 *
4473 * This function creates a clone of a buffer that holds a reference on
4474 * sk_refcnt. Buffers created via this function are meant to be
4475 * returned using sock_queue_err_skb, or free via kfree_skb.
4476 *
4477 * When passing buffers allocated with this function to sock_queue_err_skb
4478 * it is necessary to wrap the call with sock_hold/sock_put in order to
4479 * prevent the socket from being released prior to being enqueued on
4480 * the sk_error_queue.
4481 */
4482struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4483{
4484 struct sock *sk = skb->sk;
4485 struct sk_buff *clone;
4486
4487 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4488 return NULL;
4489
4490 clone = skb_clone(skb, GFP_ATOMIC);
4491 if (!clone) {
4492 sock_put(sk);
4493 return NULL;
4494 }
4495
4496 clone->sk = sk;
4497 clone->destructor = sock_efree;
4498
4499 return clone;
4500}
4501EXPORT_SYMBOL(skb_clone_sk);
4502
4503static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4504 struct sock *sk,
4505 int tstype,
4506 bool opt_stats)
4507{
4508 struct sock_exterr_skb *serr;
4509 int err;
4510
4511 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4512
4513 serr = SKB_EXT_ERR(skb);
4514 memset(serr, 0, sizeof(*serr));
4515 serr->ee.ee_errno = ENOMSG;
4516 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4517 serr->ee.ee_info = tstype;
4518 serr->opt_stats = opt_stats;
4519 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4520 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4521 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4522 if (sk->sk_protocol == IPPROTO_TCP &&
4523 sk->sk_type == SOCK_STREAM)
4524 serr->ee.ee_data -= sk->sk_tskey;
4525 }
4526
4527 err = sock_queue_err_skb(sk, skb);
4528
4529 if (err)
4530 kfree_skb(skb);
4531}
4532
4533static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4534{
4535 bool ret;
4536
4537 if (likely(sysctl_tstamp_allow_data || tsonly))
4538 return true;
4539
4540 read_lock_bh(&sk->sk_callback_lock);
4541 ret = sk->sk_socket && sk->sk_socket->file &&
4542 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4543 read_unlock_bh(&sk->sk_callback_lock);
4544 return ret;
4545}
4546
4547void skb_complete_tx_timestamp(struct sk_buff *skb,
4548 struct skb_shared_hwtstamps *hwtstamps)
4549{
4550 struct sock *sk = skb->sk;
4551
4552 if (!skb_may_tx_timestamp(sk, false))
4553 goto err;
4554
4555 /* Take a reference to prevent skb_orphan() from freeing the socket,
4556 * but only if the socket refcount is not zero.
4557 */
4558 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4559 *skb_hwtstamps(skb) = *hwtstamps;
4560 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4561 sock_put(sk);
4562 return;
4563 }
4564
4565err:
4566 kfree_skb(skb);
4567}
4568EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4569
4570void __skb_tstamp_tx(struct sk_buff *orig_skb,
4571 struct skb_shared_hwtstamps *hwtstamps,
4572 struct sock *sk, int tstype)
4573{
4574 struct sk_buff *skb;
4575 bool tsonly, opt_stats = false;
4576
4577 if (!sk)
4578 return;
4579
4580 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4581 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4582 return;
4583
4584 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4585 if (!skb_may_tx_timestamp(sk, tsonly))
4586 return;
4587
4588 if (tsonly) {
4589#ifdef CONFIG_INET
4590 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4591 sk->sk_protocol == IPPROTO_TCP &&
4592 sk->sk_type == SOCK_STREAM) {
4593 skb = tcp_get_timestamping_opt_stats(sk);
4594 opt_stats = true;
4595 } else
4596#endif
4597 skb = alloc_skb(0, GFP_ATOMIC);
4598 } else {
4599 skb = skb_clone(orig_skb, GFP_ATOMIC);
4600 }
4601 if (!skb)
4602 return;
4603
4604 if (tsonly) {
4605 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4606 SKBTX_ANY_TSTAMP;
4607 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4608 }
4609
4610 if (hwtstamps)
4611 *skb_hwtstamps(skb) = *hwtstamps;
4612 else
4613 skb->tstamp = ktime_get_real();
4614
4615 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4616}
4617EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4618
4619void skb_tstamp_tx(struct sk_buff *orig_skb,
4620 struct skb_shared_hwtstamps *hwtstamps)
4621{
4622 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4623 SCM_TSTAMP_SND);
4624}
4625EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4626
4627void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4628{
4629 struct sock *sk = skb->sk;
4630 struct sock_exterr_skb *serr;
4631 int err = 1;
4632
4633 skb->wifi_acked_valid = 1;
4634 skb->wifi_acked = acked;
4635
4636 serr = SKB_EXT_ERR(skb);
4637 memset(serr, 0, sizeof(*serr));
4638 serr->ee.ee_errno = ENOMSG;
4639 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4640
4641 /* Take a reference to prevent skb_orphan() from freeing the socket,
4642 * but only if the socket refcount is not zero.
4643 */
4644 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4645 err = sock_queue_err_skb(sk, skb);
4646 sock_put(sk);
4647 }
4648 if (err)
4649 kfree_skb(skb);
4650}
4651EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4652
4653/**
4654 * skb_partial_csum_set - set up and verify partial csum values for packet
4655 * @skb: the skb to set
4656 * @start: the number of bytes after skb->data to start checksumming.
4657 * @off: the offset from start to place the checksum.
4658 *
4659 * For untrusted partially-checksummed packets, we need to make sure the values
4660 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4661 *
4662 * This function checks and sets those values and skb->ip_summed: if this
4663 * returns false you should drop the packet.
4664 */
4665bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4666{
4667 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4668 u32 csum_start = skb_headroom(skb) + (u32)start;
4669
4670 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4671 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4672 start, off, skb_headroom(skb), skb_headlen(skb));
4673 return false;
4674 }
4675 skb->ip_summed = CHECKSUM_PARTIAL;
4676 skb->csum_start = csum_start;
4677 skb->csum_offset = off;
4678 skb_set_transport_header(skb, start);
4679 return true;
4680}
4681EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4682
4683static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4684 unsigned int max)
4685{
4686 if (skb_headlen(skb) >= len)
4687 return 0;
4688
4689 /* If we need to pullup then pullup to the max, so we
4690 * won't need to do it again.
4691 */
4692 if (max > skb->len)
4693 max = skb->len;
4694
4695 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4696 return -ENOMEM;
4697
4698 if (skb_headlen(skb) < len)
4699 return -EPROTO;
4700
4701 return 0;
4702}
4703
4704#define MAX_TCP_HDR_LEN (15 * 4)
4705
4706static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4707 typeof(IPPROTO_IP) proto,
4708 unsigned int off)
4709{
4710 switch (proto) {
4711 int err;
4712
4713 case IPPROTO_TCP:
4714 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4715 off + MAX_TCP_HDR_LEN);
4716 if (!err && !skb_partial_csum_set(skb, off,
4717 offsetof(struct tcphdr,
4718 check)))
4719 err = -EPROTO;
4720 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4721
4722 case IPPROTO_UDP:
4723 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4724 off + sizeof(struct udphdr));
4725 if (!err && !skb_partial_csum_set(skb, off,
4726 offsetof(struct udphdr,
4727 check)))
4728 err = -EPROTO;
4729 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4730 }
4731
4732 return ERR_PTR(-EPROTO);
4733}
4734
4735/* This value should be large enough to cover a tagged ethernet header plus
4736 * maximally sized IP and TCP or UDP headers.
4737 */
4738#define MAX_IP_HDR_LEN 128
4739
4740static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4741{
4742 unsigned int off;
4743 bool fragment;
4744 __sum16 *csum;
4745 int err;
4746
4747 fragment = false;
4748
4749 err = skb_maybe_pull_tail(skb,
4750 sizeof(struct iphdr),
4751 MAX_IP_HDR_LEN);
4752 if (err < 0)
4753 goto out;
4754
4755 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4756 fragment = true;
4757
4758 off = ip_hdrlen(skb);
4759
4760 err = -EPROTO;
4761
4762 if (fragment)
4763 goto out;
4764
4765 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4766 if (IS_ERR(csum))
4767 return PTR_ERR(csum);
4768
4769 if (recalculate)
4770 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4771 ip_hdr(skb)->daddr,
4772 skb->len - off,
4773 ip_hdr(skb)->protocol, 0);
4774 err = 0;
4775
4776out:
4777 return err;
4778}
4779
4780/* This value should be large enough to cover a tagged ethernet header plus
4781 * an IPv6 header, all options, and a maximal TCP or UDP header.
4782 */
4783#define MAX_IPV6_HDR_LEN 256
4784
4785#define OPT_HDR(type, skb, off) \
4786 (type *)(skb_network_header(skb) + (off))
4787
4788static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4789{
4790 int err;
4791 u8 nexthdr;
4792 unsigned int off;
4793 unsigned int len;
4794 bool fragment;
4795 bool done;
4796 __sum16 *csum;
4797
4798 fragment = false;
4799 done = false;
4800
4801 off = sizeof(struct ipv6hdr);
4802
4803 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4804 if (err < 0)
4805 goto out;
4806
4807 nexthdr = ipv6_hdr(skb)->nexthdr;
4808
4809 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4810 while (off <= len && !done) {
4811 switch (nexthdr) {
4812 case IPPROTO_DSTOPTS:
4813 case IPPROTO_HOPOPTS:
4814 case IPPROTO_ROUTING: {
4815 struct ipv6_opt_hdr *hp;
4816
4817 err = skb_maybe_pull_tail(skb,
4818 off +
4819 sizeof(struct ipv6_opt_hdr),
4820 MAX_IPV6_HDR_LEN);
4821 if (err < 0)
4822 goto out;
4823
4824 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4825 nexthdr = hp->nexthdr;
4826 off += ipv6_optlen(hp);
4827 break;
4828 }
4829 case IPPROTO_AH: {
4830 struct ip_auth_hdr *hp;
4831
4832 err = skb_maybe_pull_tail(skb,
4833 off +
4834 sizeof(struct ip_auth_hdr),
4835 MAX_IPV6_HDR_LEN);
4836 if (err < 0)
4837 goto out;
4838
4839 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4840 nexthdr = hp->nexthdr;
4841 off += ipv6_authlen(hp);
4842 break;
4843 }
4844 case IPPROTO_FRAGMENT: {
4845 struct frag_hdr *hp;
4846
4847 err = skb_maybe_pull_tail(skb,
4848 off +
4849 sizeof(struct frag_hdr),
4850 MAX_IPV6_HDR_LEN);
4851 if (err < 0)
4852 goto out;
4853
4854 hp = OPT_HDR(struct frag_hdr, skb, off);
4855
4856 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4857 fragment = true;
4858
4859 nexthdr = hp->nexthdr;
4860 off += sizeof(struct frag_hdr);
4861 break;
4862 }
4863 default:
4864 done = true;
4865 break;
4866 }
4867 }
4868
4869 err = -EPROTO;
4870
4871 if (!done || fragment)
4872 goto out;
4873
4874 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4875 if (IS_ERR(csum))
4876 return PTR_ERR(csum);
4877
4878 if (recalculate)
4879 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4880 &ipv6_hdr(skb)->daddr,
4881 skb->len - off, nexthdr, 0);
4882 err = 0;
4883
4884out:
4885 return err;
4886}
4887
4888/**
4889 * skb_checksum_setup - set up partial checksum offset
4890 * @skb: the skb to set up
4891 * @recalculate: if true the pseudo-header checksum will be recalculated
4892 */
4893int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4894{
4895 int err;
4896
4897 switch (skb->protocol) {
4898 case htons(ETH_P_IP):
4899 err = skb_checksum_setup_ipv4(skb, recalculate);
4900 break;
4901
4902 case htons(ETH_P_IPV6):
4903 err = skb_checksum_setup_ipv6(skb, recalculate);
4904 break;
4905
4906 default:
4907 err = -EPROTO;
4908 break;
4909 }
4910
4911 return err;
4912}
4913EXPORT_SYMBOL(skb_checksum_setup);
4914
4915/**
4916 * skb_checksum_maybe_trim - maybe trims the given skb
4917 * @skb: the skb to check
4918 * @transport_len: the data length beyond the network header
4919 *
4920 * Checks whether the given skb has data beyond the given transport length.
4921 * If so, returns a cloned skb trimmed to this transport length.
4922 * Otherwise returns the provided skb. Returns NULL in error cases
4923 * (e.g. transport_len exceeds skb length or out-of-memory).
4924 *
4925 * Caller needs to set the skb transport header and free any returned skb if it
4926 * differs from the provided skb.
4927 */
4928static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4929 unsigned int transport_len)
4930{
4931 struct sk_buff *skb_chk;
4932 unsigned int len = skb_transport_offset(skb) + transport_len;
4933 int ret;
4934
4935 if (skb->len < len)
4936 return NULL;
4937 else if (skb->len == len)
4938 return skb;
4939
4940 skb_chk = skb_clone(skb, GFP_ATOMIC);
4941 if (!skb_chk)
4942 return NULL;
4943
4944 ret = pskb_trim_rcsum(skb_chk, len);
4945 if (ret) {
4946 kfree_skb(skb_chk);
4947 return NULL;
4948 }
4949
4950 return skb_chk;
4951}
4952
4953/**
4954 * skb_checksum_trimmed - validate checksum of an skb
4955 * @skb: the skb to check
4956 * @transport_len: the data length beyond the network header
4957 * @skb_chkf: checksum function to use
4958 *
4959 * Applies the given checksum function skb_chkf to the provided skb.
4960 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4961 *
4962 * If the skb has data beyond the given transport length, then a
4963 * trimmed & cloned skb is checked and returned.
4964 *
4965 * Caller needs to set the skb transport header and free any returned skb if it
4966 * differs from the provided skb.
4967 */
4968struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4969 unsigned int transport_len,
4970 __sum16(*skb_chkf)(struct sk_buff *skb))
4971{
4972 struct sk_buff *skb_chk;
4973 unsigned int offset = skb_transport_offset(skb);
4974 __sum16 ret;
4975
4976 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4977 if (!skb_chk)
4978 goto err;
4979
4980 if (!pskb_may_pull(skb_chk, offset))
4981 goto err;
4982
4983 skb_pull_rcsum(skb_chk, offset);
4984 ret = skb_chkf(skb_chk);
4985 skb_push_rcsum(skb_chk, offset);
4986
4987 if (ret)
4988 goto err;
4989
4990 return skb_chk;
4991
4992err:
4993 if (skb_chk && skb_chk != skb)
4994 kfree_skb(skb_chk);
4995
4996 return NULL;
4997
4998}
4999EXPORT_SYMBOL(skb_checksum_trimmed);
5000
5001void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5002{
5003 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5004 skb->dev->name);
5005}
5006EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5007
5008void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5009{
5010 if (head_stolen) {
5011 skb_release_head_state(skb);
5012 kmem_cache_free(skbuff_head_cache, skb);
5013 } else {
5014 __kfree_skb(skb);
5015 }
5016}
5017EXPORT_SYMBOL(kfree_skb_partial);
5018
5019/**
5020 * skb_try_coalesce - try to merge skb to prior one
5021 * @to: prior buffer
5022 * @from: buffer to add
5023 * @fragstolen: pointer to boolean
5024 * @delta_truesize: how much more was allocated than was requested
5025 */
5026bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5027 bool *fragstolen, int *delta_truesize)
5028{
5029 struct skb_shared_info *to_shinfo, *from_shinfo;
5030 int i, delta, len = from->len;
5031
5032 *fragstolen = false;
5033
5034 if (skb_cloned(to))
5035 return false;
5036
5037 if (len <= skb_tailroom(to)) {
5038 if (len)
5039 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5040 *delta_truesize = 0;
5041 return true;
5042 }
5043
5044 to_shinfo = skb_shinfo(to);
5045 from_shinfo = skb_shinfo(from);
5046 if (to_shinfo->frag_list || from_shinfo->frag_list)
5047 return false;
5048 if (skb_zcopy(to) || skb_zcopy(from))
5049 return false;
5050
5051 if (skb_headlen(from) != 0) {
5052 struct page *page;
5053 unsigned int offset;
5054
5055 if (to_shinfo->nr_frags +
5056 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5057 return false;
5058
5059 if (skb_head_is_locked(from))
5060 return false;
5061
5062 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5063
5064 page = virt_to_head_page(from->head);
5065 offset = from->data - (unsigned char *)page_address(page);
5066
5067 skb_fill_page_desc(to, to_shinfo->nr_frags,
5068 page, offset, skb_headlen(from));
5069 *fragstolen = true;
5070 } else {
5071 if (to_shinfo->nr_frags +
5072 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5073 return false;
5074
5075 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5076 }
5077
5078 WARN_ON_ONCE(delta < len);
5079
5080 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5081 from_shinfo->frags,
5082 from_shinfo->nr_frags * sizeof(skb_frag_t));
5083 to_shinfo->nr_frags += from_shinfo->nr_frags;
5084
5085 if (!skb_cloned(from))
5086 from_shinfo->nr_frags = 0;
5087
5088 /* if the skb is not cloned this does nothing
5089 * since we set nr_frags to 0.
5090 */
5091 for (i = 0; i < from_shinfo->nr_frags; i++)
5092 __skb_frag_ref(&from_shinfo->frags[i]);
5093
5094 to->truesize += delta;
5095 to->len += len;
5096 to->data_len += len;
5097
5098 *delta_truesize = delta;
5099 return true;
5100}
5101EXPORT_SYMBOL(skb_try_coalesce);
5102
5103/**
5104 * skb_scrub_packet - scrub an skb
5105 *
5106 * @skb: buffer to clean
5107 * @xnet: packet is crossing netns
5108 *
5109 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5110 * into/from a tunnel. Some information have to be cleared during these
5111 * operations.
5112 * skb_scrub_packet can also be used to clean a skb before injecting it in
5113 * another namespace (@xnet == true). We have to clear all information in the
5114 * skb that could impact namespace isolation.
5115 */
5116void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5117{
5118 skb->pkt_type = PACKET_HOST;
5119 skb->skb_iif = 0;
5120 skb->ignore_df = 0;
5121 skb_dst_drop(skb);
5122 skb_ext_reset(skb);
5123 nf_reset_ct(skb);
5124 nf_reset_trace(skb);
5125
5126#ifdef CONFIG_NET_SWITCHDEV
5127 skb->offload_fwd_mark = 0;
5128 skb->offload_l3_fwd_mark = 0;
5129#endif
5130
5131 if (!xnet)
5132 return;
5133
5134 ipvs_reset(skb);
5135 skb->mark = 0;
5136 skb->tstamp = 0;
5137}
5138EXPORT_SYMBOL_GPL(skb_scrub_packet);
5139
5140/**
5141 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5142 *
5143 * @skb: GSO skb
5144 *
5145 * skb_gso_transport_seglen is used to determine the real size of the
5146 * individual segments, including Layer4 headers (TCP/UDP).
5147 *
5148 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5149 */
5150static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5151{
5152 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5153 unsigned int thlen = 0;
5154
5155 if (skb->encapsulation) {
5156 thlen = skb_inner_transport_header(skb) -
5157 skb_transport_header(skb);
5158
5159 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5160 thlen += inner_tcp_hdrlen(skb);
5161 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5162 thlen = tcp_hdrlen(skb);
5163 } else if (unlikely(skb_is_gso_sctp(skb))) {
5164 thlen = sizeof(struct sctphdr);
5165 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5166 thlen = sizeof(struct udphdr);
5167 }
5168 /* UFO sets gso_size to the size of the fragmentation
5169 * payload, i.e. the size of the L4 (UDP) header is already
5170 * accounted for.
5171 */
5172 return thlen + shinfo->gso_size;
5173}
5174
5175/**
5176 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5177 *
5178 * @skb: GSO skb
5179 *
5180 * skb_gso_network_seglen is used to determine the real size of the
5181 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5182 *
5183 * The MAC/L2 header is not accounted for.
5184 */
5185static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5186{
5187 unsigned int hdr_len = skb_transport_header(skb) -
5188 skb_network_header(skb);
5189
5190 return hdr_len + skb_gso_transport_seglen(skb);
5191}
5192
5193/**
5194 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5195 *
5196 * @skb: GSO skb
5197 *
5198 * skb_gso_mac_seglen is used to determine the real size of the
5199 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5200 * headers (TCP/UDP).
5201 */
5202static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5203{
5204 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5205
5206 return hdr_len + skb_gso_transport_seglen(skb);
5207}
5208
5209/**
5210 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5211 *
5212 * There are a couple of instances where we have a GSO skb, and we
5213 * want to determine what size it would be after it is segmented.
5214 *
5215 * We might want to check:
5216 * - L3+L4+payload size (e.g. IP forwarding)
5217 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5218 *
5219 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5220 *
5221 * @skb: GSO skb
5222 *
5223 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5224 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5225 *
5226 * @max_len: The maximum permissible length.
5227 *
5228 * Returns true if the segmented length <= max length.
5229 */
5230static inline bool skb_gso_size_check(const struct sk_buff *skb,
5231 unsigned int seg_len,
5232 unsigned int max_len) {
5233 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5234 const struct sk_buff *iter;
5235
5236 if (shinfo->gso_size != GSO_BY_FRAGS)
5237 return seg_len <= max_len;
5238
5239 /* Undo this so we can re-use header sizes */
5240 seg_len -= GSO_BY_FRAGS;
5241
5242 skb_walk_frags(skb, iter) {
5243 if (seg_len + skb_headlen(iter) > max_len)
5244 return false;
5245 }
5246
5247 return true;
5248}
5249
5250/**
5251 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5252 *
5253 * @skb: GSO skb
5254 * @mtu: MTU to validate against
5255 *
5256 * skb_gso_validate_network_len validates if a given skb will fit a
5257 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5258 * payload.
5259 */
5260bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5261{
5262 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5263}
5264EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5265
5266/**
5267 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5268 *
5269 * @skb: GSO skb
5270 * @len: length to validate against
5271 *
5272 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5273 * length once split, including L2, L3 and L4 headers and the payload.
5274 */
5275bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5276{
5277 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5278}
5279EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5280
5281static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5282{
5283 int mac_len, meta_len;
5284 void *meta;
5285
5286 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5287 kfree_skb(skb);
5288 return NULL;
5289 }
5290
5291 mac_len = skb->data - skb_mac_header(skb);
5292 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5293 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5294 mac_len - VLAN_HLEN - ETH_TLEN);
5295 }
5296
5297 meta_len = skb_metadata_len(skb);
5298 if (meta_len) {
5299 meta = skb_metadata_end(skb) - meta_len;
5300 memmove(meta + VLAN_HLEN, meta, meta_len);
5301 }
5302
5303 skb->mac_header += VLAN_HLEN;
5304 return skb;
5305}
5306
5307struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5308{
5309 struct vlan_hdr *vhdr;
5310 u16 vlan_tci;
5311
5312 if (unlikely(skb_vlan_tag_present(skb))) {
5313 /* vlan_tci is already set-up so leave this for another time */
5314 return skb;
5315 }
5316
5317 skb = skb_share_check(skb, GFP_ATOMIC);
5318 if (unlikely(!skb))
5319 goto err_free;
5320
5321 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5322 goto err_free;
5323
5324 vhdr = (struct vlan_hdr *)skb->data;
5325 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5326 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5327
5328 skb_pull_rcsum(skb, VLAN_HLEN);
5329 vlan_set_encap_proto(skb, vhdr);
5330
5331 skb = skb_reorder_vlan_header(skb);
5332 if (unlikely(!skb))
5333 goto err_free;
5334
5335 skb_reset_network_header(skb);
5336 skb_reset_transport_header(skb);
5337 skb_reset_mac_len(skb);
5338
5339 return skb;
5340
5341err_free:
5342 kfree_skb(skb);
5343 return NULL;
5344}
5345EXPORT_SYMBOL(skb_vlan_untag);
5346
5347int skb_ensure_writable(struct sk_buff *skb, int write_len)
5348{
5349 if (!pskb_may_pull(skb, write_len))
5350 return -ENOMEM;
5351
5352 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5353 return 0;
5354
5355 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5356}
5357EXPORT_SYMBOL(skb_ensure_writable);
5358
5359/* remove VLAN header from packet and update csum accordingly.
5360 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5361 */
5362int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5363{
5364 struct vlan_hdr *vhdr;
5365 int offset = skb->data - skb_mac_header(skb);
5366 int err;
5367
5368 if (WARN_ONCE(offset,
5369 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5370 offset)) {
5371 return -EINVAL;
5372 }
5373
5374 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5375 if (unlikely(err))
5376 return err;
5377
5378 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5379
5380 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5381 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5382
5383 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5384 __skb_pull(skb, VLAN_HLEN);
5385
5386 vlan_set_encap_proto(skb, vhdr);
5387 skb->mac_header += VLAN_HLEN;
5388
5389 if (skb_network_offset(skb) < ETH_HLEN)
5390 skb_set_network_header(skb, ETH_HLEN);
5391
5392 skb_reset_mac_len(skb);
5393
5394 return err;
5395}
5396EXPORT_SYMBOL(__skb_vlan_pop);
5397
5398/* Pop a vlan tag either from hwaccel or from payload.
5399 * Expects skb->data at mac header.
5400 */
5401int skb_vlan_pop(struct sk_buff *skb)
5402{
5403 u16 vlan_tci;
5404 __be16 vlan_proto;
5405 int err;
5406
5407 if (likely(skb_vlan_tag_present(skb))) {
5408 __vlan_hwaccel_clear_tag(skb);
5409 } else {
5410 if (unlikely(!eth_type_vlan(skb->protocol)))
5411 return 0;
5412
5413 err = __skb_vlan_pop(skb, &vlan_tci);
5414 if (err)
5415 return err;
5416 }
5417 /* move next vlan tag to hw accel tag */
5418 if (likely(!eth_type_vlan(skb->protocol)))
5419 return 0;
5420
5421 vlan_proto = skb->protocol;
5422 err = __skb_vlan_pop(skb, &vlan_tci);
5423 if (unlikely(err))
5424 return err;
5425
5426 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5427 return 0;
5428}
5429EXPORT_SYMBOL(skb_vlan_pop);
5430
5431/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5432 * Expects skb->data at mac header.
5433 */
5434int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5435{
5436 if (skb_vlan_tag_present(skb)) {
5437 int offset = skb->data - skb_mac_header(skb);
5438 int err;
5439
5440 if (WARN_ONCE(offset,
5441 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5442 offset)) {
5443 return -EINVAL;
5444 }
5445
5446 err = __vlan_insert_tag(skb, skb->vlan_proto,
5447 skb_vlan_tag_get(skb));
5448 if (err)
5449 return err;
5450
5451 skb->protocol = skb->vlan_proto;
5452 skb->mac_len += VLAN_HLEN;
5453
5454 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5455 }
5456 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5457 return 0;
5458}
5459EXPORT_SYMBOL(skb_vlan_push);
5460
5461/* Update the ethertype of hdr and the skb csum value if required. */
5462static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5463 __be16 ethertype)
5464{
5465 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5466 __be16 diff[] = { ~hdr->h_proto, ethertype };
5467
5468 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5469 }
5470
5471 hdr->h_proto = ethertype;
5472}
5473
5474/**
5475 * skb_mpls_push() - push a new MPLS header after the mac header
5476 *
5477 * @skb: buffer
5478 * @mpls_lse: MPLS label stack entry to push
5479 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5480 * @mac_len: length of the MAC header
5481 *
5482 * Expects skb->data at mac header.
5483 *
5484 * Returns 0 on success, -errno otherwise.
5485 */
5486int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5487 int mac_len)
5488{
5489 struct mpls_shim_hdr *lse;
5490 int err;
5491
5492 if (unlikely(!eth_p_mpls(mpls_proto)))
5493 return -EINVAL;
5494
5495 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5496 if (skb->encapsulation)
5497 return -EINVAL;
5498
5499 err = skb_cow_head(skb, MPLS_HLEN);
5500 if (unlikely(err))
5501 return err;
5502
5503 if (!skb->inner_protocol) {
5504 skb_set_inner_network_header(skb, mac_len);
5505 skb_set_inner_protocol(skb, skb->protocol);
5506 }
5507
5508 skb_push(skb, MPLS_HLEN);
5509 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5510 mac_len);
5511 skb_reset_mac_header(skb);
5512 skb_set_network_header(skb, mac_len);
5513
5514 lse = mpls_hdr(skb);
5515 lse->label_stack_entry = mpls_lse;
5516 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5517
5518 if (skb->dev && skb->dev->type == ARPHRD_ETHER)
5519 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5520 skb->protocol = mpls_proto;
5521
5522 return 0;
5523}
5524EXPORT_SYMBOL_GPL(skb_mpls_push);
5525
5526/**
5527 * skb_mpls_pop() - pop the outermost MPLS header
5528 *
5529 * @skb: buffer
5530 * @next_proto: ethertype of header after popped MPLS header
5531 * @mac_len: length of the MAC header
5532 *
5533 * Expects skb->data at mac header.
5534 *
5535 * Returns 0 on success, -errno otherwise.
5536 */
5537int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len)
5538{
5539 int err;
5540
5541 if (unlikely(!eth_p_mpls(skb->protocol)))
5542 return 0;
5543
5544 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5545 if (unlikely(err))
5546 return err;
5547
5548 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5549 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5550 mac_len);
5551
5552 __skb_pull(skb, MPLS_HLEN);
5553 skb_reset_mac_header(skb);
5554 skb_set_network_header(skb, mac_len);
5555
5556 if (skb->dev && skb->dev->type == ARPHRD_ETHER) {
5557 struct ethhdr *hdr;
5558
5559 /* use mpls_hdr() to get ethertype to account for VLANs. */
5560 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5561 skb_mod_eth_type(skb, hdr, next_proto);
5562 }
5563 skb->protocol = next_proto;
5564
5565 return 0;
5566}
5567EXPORT_SYMBOL_GPL(skb_mpls_pop);
5568
5569/**
5570 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5571 *
5572 * @skb: buffer
5573 * @mpls_lse: new MPLS label stack entry to update to
5574 *
5575 * Expects skb->data at mac header.
5576 *
5577 * Returns 0 on success, -errno otherwise.
5578 */
5579int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5580{
5581 int err;
5582
5583 if (unlikely(!eth_p_mpls(skb->protocol)))
5584 return -EINVAL;
5585
5586 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5587 if (unlikely(err))
5588 return err;
5589
5590 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5591 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5592
5593 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5594 }
5595
5596 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5597
5598 return 0;
5599}
5600EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5601
5602/**
5603 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5604 *
5605 * @skb: buffer
5606 *
5607 * Expects skb->data at mac header.
5608 *
5609 * Returns 0 on success, -errno otherwise.
5610 */
5611int skb_mpls_dec_ttl(struct sk_buff *skb)
5612{
5613 u32 lse;
5614 u8 ttl;
5615
5616 if (unlikely(!eth_p_mpls(skb->protocol)))
5617 return -EINVAL;
5618
5619 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5620 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5621 if (!--ttl)
5622 return -EINVAL;
5623
5624 lse &= ~MPLS_LS_TTL_MASK;
5625 lse |= ttl << MPLS_LS_TTL_SHIFT;
5626
5627 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5628}
5629EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5630
5631/**
5632 * alloc_skb_with_frags - allocate skb with page frags
5633 *
5634 * @header_len: size of linear part
5635 * @data_len: needed length in frags
5636 * @max_page_order: max page order desired.
5637 * @errcode: pointer to error code if any
5638 * @gfp_mask: allocation mask
5639 *
5640 * This can be used to allocate a paged skb, given a maximal order for frags.
5641 */
5642struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5643 unsigned long data_len,
5644 int max_page_order,
5645 int *errcode,
5646 gfp_t gfp_mask)
5647{
5648 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5649 unsigned long chunk;
5650 struct sk_buff *skb;
5651 struct page *page;
5652 int i;
5653
5654 *errcode = -EMSGSIZE;
5655 /* Note this test could be relaxed, if we succeed to allocate
5656 * high order pages...
5657 */
5658 if (npages > MAX_SKB_FRAGS)
5659 return NULL;
5660
5661 *errcode = -ENOBUFS;
5662 skb = alloc_skb(header_len, gfp_mask);
5663 if (!skb)
5664 return NULL;
5665
5666 skb->truesize += npages << PAGE_SHIFT;
5667
5668 for (i = 0; npages > 0; i++) {
5669 int order = max_page_order;
5670
5671 while (order) {
5672 if (npages >= 1 << order) {
5673 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5674 __GFP_COMP |
5675 __GFP_NOWARN,
5676 order);
5677 if (page)
5678 goto fill_page;
5679 /* Do not retry other high order allocations */
5680 order = 1;
5681 max_page_order = 0;
5682 }
5683 order--;
5684 }
5685 page = alloc_page(gfp_mask);
5686 if (!page)
5687 goto failure;
5688fill_page:
5689 chunk = min_t(unsigned long, data_len,
5690 PAGE_SIZE << order);
5691 skb_fill_page_desc(skb, i, page, 0, chunk);
5692 data_len -= chunk;
5693 npages -= 1 << order;
5694 }
5695 return skb;
5696
5697failure:
5698 kfree_skb(skb);
5699 return NULL;
5700}
5701EXPORT_SYMBOL(alloc_skb_with_frags);
5702
5703/* carve out the first off bytes from skb when off < headlen */
5704static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5705 const int headlen, gfp_t gfp_mask)
5706{
5707 int i;
5708 int size = skb_end_offset(skb);
5709 int new_hlen = headlen - off;
5710 u8 *data;
5711
5712 size = SKB_DATA_ALIGN(size);
5713
5714 if (skb_pfmemalloc(skb))
5715 gfp_mask |= __GFP_MEMALLOC;
5716 data = kmalloc_reserve(size +
5717 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5718 gfp_mask, NUMA_NO_NODE, NULL);
5719 if (!data)
5720 return -ENOMEM;
5721
5722 size = SKB_WITH_OVERHEAD(ksize(data));
5723
5724 /* Copy real data, and all frags */
5725 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5726 skb->len -= off;
5727
5728 memcpy((struct skb_shared_info *)(data + size),
5729 skb_shinfo(skb),
5730 offsetof(struct skb_shared_info,
5731 frags[skb_shinfo(skb)->nr_frags]));
5732 if (skb_cloned(skb)) {
5733 /* drop the old head gracefully */
5734 if (skb_orphan_frags(skb, gfp_mask)) {
5735 kfree(data);
5736 return -ENOMEM;
5737 }
5738 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5739 skb_frag_ref(skb, i);
5740 if (skb_has_frag_list(skb))
5741 skb_clone_fraglist(skb);
5742 skb_release_data(skb);
5743 } else {
5744 /* we can reuse existing recount- all we did was
5745 * relocate values
5746 */
5747 skb_free_head(skb);
5748 }
5749
5750 skb->head = data;
5751 skb->data = data;
5752 skb->head_frag = 0;
5753#ifdef NET_SKBUFF_DATA_USES_OFFSET
5754 skb->end = size;
5755#else
5756 skb->end = skb->head + size;
5757#endif
5758 skb_set_tail_pointer(skb, skb_headlen(skb));
5759 skb_headers_offset_update(skb, 0);
5760 skb->cloned = 0;
5761 skb->hdr_len = 0;
5762 skb->nohdr = 0;
5763 atomic_set(&skb_shinfo(skb)->dataref, 1);
5764
5765 return 0;
5766}
5767
5768static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5769
5770/* carve out the first eat bytes from skb's frag_list. May recurse into
5771 * pskb_carve()
5772 */
5773static int pskb_carve_frag_list(struct sk_buff *skb,
5774 struct skb_shared_info *shinfo, int eat,
5775 gfp_t gfp_mask)
5776{
5777 struct sk_buff *list = shinfo->frag_list;
5778 struct sk_buff *clone = NULL;
5779 struct sk_buff *insp = NULL;
5780
5781 do {
5782 if (!list) {
5783 pr_err("Not enough bytes to eat. Want %d\n", eat);
5784 return -EFAULT;
5785 }
5786 if (list->len <= eat) {
5787 /* Eaten as whole. */
5788 eat -= list->len;
5789 list = list->next;
5790 insp = list;
5791 } else {
5792 /* Eaten partially. */
5793 if (skb_shared(list)) {
5794 clone = skb_clone(list, gfp_mask);
5795 if (!clone)
5796 return -ENOMEM;
5797 insp = list->next;
5798 list = clone;
5799 } else {
5800 /* This may be pulled without problems. */
5801 insp = list;
5802 }
5803 if (pskb_carve(list, eat, gfp_mask) < 0) {
5804 kfree_skb(clone);
5805 return -ENOMEM;
5806 }
5807 break;
5808 }
5809 } while (eat);
5810
5811 /* Free pulled out fragments. */
5812 while ((list = shinfo->frag_list) != insp) {
5813 shinfo->frag_list = list->next;
5814 kfree_skb(list);
5815 }
5816 /* And insert new clone at head. */
5817 if (clone) {
5818 clone->next = list;
5819 shinfo->frag_list = clone;
5820 }
5821 return 0;
5822}
5823
5824/* carve off first len bytes from skb. Split line (off) is in the
5825 * non-linear part of skb
5826 */
5827static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5828 int pos, gfp_t gfp_mask)
5829{
5830 int i, k = 0;
5831 int size = skb_end_offset(skb);
5832 u8 *data;
5833 const int nfrags = skb_shinfo(skb)->nr_frags;
5834 struct skb_shared_info *shinfo;
5835
5836 size = SKB_DATA_ALIGN(size);
5837
5838 if (skb_pfmemalloc(skb))
5839 gfp_mask |= __GFP_MEMALLOC;
5840 data = kmalloc_reserve(size +
5841 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5842 gfp_mask, NUMA_NO_NODE, NULL);
5843 if (!data)
5844 return -ENOMEM;
5845
5846 size = SKB_WITH_OVERHEAD(ksize(data));
5847
5848 memcpy((struct skb_shared_info *)(data + size),
5849 skb_shinfo(skb), offsetof(struct skb_shared_info,
5850 frags[skb_shinfo(skb)->nr_frags]));
5851 if (skb_orphan_frags(skb, gfp_mask)) {
5852 kfree(data);
5853 return -ENOMEM;
5854 }
5855 shinfo = (struct skb_shared_info *)(data + size);
5856 for (i = 0; i < nfrags; i++) {
5857 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5858
5859 if (pos + fsize > off) {
5860 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5861
5862 if (pos < off) {
5863 /* Split frag.
5864 * We have two variants in this case:
5865 * 1. Move all the frag to the second
5866 * part, if it is possible. F.e.
5867 * this approach is mandatory for TUX,
5868 * where splitting is expensive.
5869 * 2. Split is accurately. We make this.
5870 */
5871 skb_frag_off_add(&shinfo->frags[0], off - pos);
5872 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5873 }
5874 skb_frag_ref(skb, i);
5875 k++;
5876 }
5877 pos += fsize;
5878 }
5879 shinfo->nr_frags = k;
5880 if (skb_has_frag_list(skb))
5881 skb_clone_fraglist(skb);
5882
5883 if (k == 0) {
5884 /* split line is in frag list */
5885 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5886 }
5887 skb_release_data(skb);
5888
5889 skb->head = data;
5890 skb->head_frag = 0;
5891 skb->data = data;
5892#ifdef NET_SKBUFF_DATA_USES_OFFSET
5893 skb->end = size;
5894#else
5895 skb->end = skb->head + size;
5896#endif
5897 skb_reset_tail_pointer(skb);
5898 skb_headers_offset_update(skb, 0);
5899 skb->cloned = 0;
5900 skb->hdr_len = 0;
5901 skb->nohdr = 0;
5902 skb->len -= off;
5903 skb->data_len = skb->len;
5904 atomic_set(&skb_shinfo(skb)->dataref, 1);
5905 return 0;
5906}
5907
5908/* remove len bytes from the beginning of the skb */
5909static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5910{
5911 int headlen = skb_headlen(skb);
5912
5913 if (len < headlen)
5914 return pskb_carve_inside_header(skb, len, headlen, gfp);
5915 else
5916 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5917}
5918
5919/* Extract to_copy bytes starting at off from skb, and return this in
5920 * a new skb
5921 */
5922struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5923 int to_copy, gfp_t gfp)
5924{
5925 struct sk_buff *clone = skb_clone(skb, gfp);
5926
5927 if (!clone)
5928 return NULL;
5929
5930 if (pskb_carve(clone, off, gfp) < 0 ||
5931 pskb_trim(clone, to_copy)) {
5932 kfree_skb(clone);
5933 return NULL;
5934 }
5935 return clone;
5936}
5937EXPORT_SYMBOL(pskb_extract);
5938
5939/**
5940 * skb_condense - try to get rid of fragments/frag_list if possible
5941 * @skb: buffer
5942 *
5943 * Can be used to save memory before skb is added to a busy queue.
5944 * If packet has bytes in frags and enough tail room in skb->head,
5945 * pull all of them, so that we can free the frags right now and adjust
5946 * truesize.
5947 * Notes:
5948 * We do not reallocate skb->head thus can not fail.
5949 * Caller must re-evaluate skb->truesize if needed.
5950 */
5951void skb_condense(struct sk_buff *skb)
5952{
5953 if (skb->data_len) {
5954 if (skb->data_len > skb->end - skb->tail ||
5955 skb_cloned(skb))
5956 return;
5957
5958 /* Nice, we can free page frag(s) right now */
5959 __pskb_pull_tail(skb, skb->data_len);
5960 }
5961 /* At this point, skb->truesize might be over estimated,
5962 * because skb had a fragment, and fragments do not tell
5963 * their truesize.
5964 * When we pulled its content into skb->head, fragment
5965 * was freed, but __pskb_pull_tail() could not possibly
5966 * adjust skb->truesize, not knowing the frag truesize.
5967 */
5968 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5969}
5970
5971#ifdef CONFIG_SKB_EXTENSIONS
5972static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
5973{
5974 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
5975}
5976
5977static struct skb_ext *skb_ext_alloc(void)
5978{
5979 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5980
5981 if (new) {
5982 memset(new->offset, 0, sizeof(new->offset));
5983 refcount_set(&new->refcnt, 1);
5984 }
5985
5986 return new;
5987}
5988
5989static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
5990 unsigned int old_active)
5991{
5992 struct skb_ext *new;
5993
5994 if (refcount_read(&old->refcnt) == 1)
5995 return old;
5996
5997 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5998 if (!new)
5999 return NULL;
6000
6001 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6002 refcount_set(&new->refcnt, 1);
6003
6004#ifdef CONFIG_XFRM
6005 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6006 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6007 unsigned int i;
6008
6009 for (i = 0; i < sp->len; i++)
6010 xfrm_state_hold(sp->xvec[i]);
6011 }
6012#endif
6013 __skb_ext_put(old);
6014 return new;
6015}
6016
6017/**
6018 * skb_ext_add - allocate space for given extension, COW if needed
6019 * @skb: buffer
6020 * @id: extension to allocate space for
6021 *
6022 * Allocates enough space for the given extension.
6023 * If the extension is already present, a pointer to that extension
6024 * is returned.
6025 *
6026 * If the skb was cloned, COW applies and the returned memory can be
6027 * modified without changing the extension space of clones buffers.
6028 *
6029 * Returns pointer to the extension or NULL on allocation failure.
6030 */
6031void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6032{
6033 struct skb_ext *new, *old = NULL;
6034 unsigned int newlen, newoff;
6035
6036 if (skb->active_extensions) {
6037 old = skb->extensions;
6038
6039 new = skb_ext_maybe_cow(old, skb->active_extensions);
6040 if (!new)
6041 return NULL;
6042
6043 if (__skb_ext_exist(new, id))
6044 goto set_active;
6045
6046 newoff = new->chunks;
6047 } else {
6048 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6049
6050 new = skb_ext_alloc();
6051 if (!new)
6052 return NULL;
6053 }
6054
6055 newlen = newoff + skb_ext_type_len[id];
6056 new->chunks = newlen;
6057 new->offset[id] = newoff;
6058set_active:
6059 skb->extensions = new;
6060 skb->active_extensions |= 1 << id;
6061 return skb_ext_get_ptr(new, id);
6062}
6063EXPORT_SYMBOL(skb_ext_add);
6064
6065#ifdef CONFIG_XFRM
6066static void skb_ext_put_sp(struct sec_path *sp)
6067{
6068 unsigned int i;
6069
6070 for (i = 0; i < sp->len; i++)
6071 xfrm_state_put(sp->xvec[i]);
6072}
6073#endif
6074
6075void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6076{
6077 struct skb_ext *ext = skb->extensions;
6078
6079 skb->active_extensions &= ~(1 << id);
6080 if (skb->active_extensions == 0) {
6081 skb->extensions = NULL;
6082 __skb_ext_put(ext);
6083#ifdef CONFIG_XFRM
6084 } else if (id == SKB_EXT_SEC_PATH &&
6085 refcount_read(&ext->refcnt) == 1) {
6086 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6087
6088 skb_ext_put_sp(sp);
6089 sp->len = 0;
6090#endif
6091 }
6092}
6093EXPORT_SYMBOL(__skb_ext_del);
6094
6095void __skb_ext_put(struct skb_ext *ext)
6096{
6097 /* If this is last clone, nothing can increment
6098 * it after check passes. Avoids one atomic op.
6099 */
6100 if (refcount_read(&ext->refcnt) == 1)
6101 goto free_now;
6102
6103 if (!refcount_dec_and_test(&ext->refcnt))
6104 return;
6105free_now:
6106#ifdef CONFIG_XFRM
6107 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6108 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6109#endif
6110
6111 kmem_cache_free(skbuff_ext_cache, ext);
6112}
6113EXPORT_SYMBOL(__skb_ext_put);
6114#endif /* CONFIG_SKB_EXTENSIONS */