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#include <net/mptcp.h>
72
73#include <linux/uaccess.h>
74#include <trace/events/skb.h>
75#include <linux/highmem.h>
76#include <linux/capability.h>
77#include <linux/user_namespace.h>
78#include <linux/indirect_call_wrapper.h>
79
80#include "datagram.h"
81
82struct kmem_cache *skbuff_head_cache __ro_after_init;
83static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84#ifdef CONFIG_SKB_EXTENSIONS
85static struct kmem_cache *skbuff_ext_cache __ro_after_init;
86#endif
87int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88EXPORT_SYMBOL(sysctl_max_skb_frags);
89
90/**
91 * skb_panic - private function for out-of-line support
92 * @skb: buffer
93 * @sz: size
94 * @addr: address
95 * @msg: skb_over_panic or skb_under_panic
96 *
97 * Out-of-line support for skb_put() and skb_push().
98 * Called via the wrapper skb_over_panic() or skb_under_panic().
99 * Keep out of line to prevent kernel bloat.
100 * __builtin_return_address is not used because it is not always reliable.
101 */
102static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103 const char msg[])
104{
105 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
109 BUG();
110}
111
112static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
118{
119 skb_panic(skb, sz, addr, __func__);
120}
121
122/*
123 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
124 * the caller if emergency pfmemalloc reserves are being used. If it is and
125 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
126 * may be used. Otherwise, the packet data may be discarded until enough
127 * memory is free
128 */
129#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
131
132static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
134{
135 void *obj;
136 bool ret_pfmemalloc = false;
137
138 /*
139 * Try a regular allocation, when that fails and we're not entitled
140 * to the reserves, fail.
141 */
142 obj = kmalloc_node_track_caller(size,
143 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
144 node);
145 if (obj || !(gfp_pfmemalloc_allowed(flags)))
146 goto out;
147
148 /* Try again but now we are using pfmemalloc reserves */
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
151
152out:
153 if (pfmemalloc)
154 *pfmemalloc = ret_pfmemalloc;
155
156 return obj;
157}
158
159/* Allocate a new skbuff. We do this ourselves so we can fill in a few
160 * 'private' fields and also do memory statistics to find all the
161 * [BEEP] leaks.
162 *
163 */
164
165/**
166 * __alloc_skb - allocate a network buffer
167 * @size: size to allocate
168 * @gfp_mask: allocation mask
169 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
170 * instead of head cache and allocate a cloned (child) skb.
171 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
172 * allocations in case the data is required for writeback
173 * @node: numa node to allocate memory on
174 *
175 * Allocate a new &sk_buff. The returned buffer has no headroom and a
176 * tail room of at least size bytes. The object has a reference count
177 * of one. The return is the buffer. On a failure the return is %NULL.
178 *
179 * Buffers may only be allocated from interrupts using a @gfp_mask of
180 * %GFP_ATOMIC.
181 */
182struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183 int flags, int node)
184{
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
187 struct sk_buff *skb;
188 u8 *data;
189 bool pfmemalloc;
190
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
193
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
196
197 /* Get the HEAD */
198 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
199 if (!skb)
200 goto out;
201 prefetchw(skb);
202
203 /* We do our best to align skb_shared_info on a separate cache
204 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
205 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
206 * Both skb->head and skb_shared_info are cache line aligned.
207 */
208 size = SKB_DATA_ALIGN(size);
209 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
211 if (!data)
212 goto nodata;
213 /* kmalloc(size) might give us more room than requested.
214 * Put skb_shared_info exactly at the end of allocated zone,
215 * to allow max possible filling before reallocation.
216 */
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
219
220 /*
221 * Only clear those fields we need to clear, not those that we will
222 * actually initialise below. Hence, don't put any more fields after
223 * the tail pointer in struct sk_buff!
224 */
225 memset(skb, 0, offsetof(struct sk_buff, tail));
226 /* Account for allocated memory : skb + skb->head */
227 skb->truesize = SKB_TRUESIZE(size);
228 skb->pfmemalloc = pfmemalloc;
229 refcount_set(&skb->users, 1);
230 skb->head = data;
231 skb->data = data;
232 skb_reset_tail_pointer(skb);
233 skb->end = skb->tail + size;
234 skb->mac_header = (typeof(skb->mac_header))~0U;
235 skb->transport_header = (typeof(skb->transport_header))~0U;
236
237 /* make sure we initialize shinfo sequentially */
238 shinfo = skb_shinfo(skb);
239 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240 atomic_set(&shinfo->dataref, 1);
241
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
244
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
246
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
249
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
251 }
252out:
253 return skb;
254nodata:
255 kmem_cache_free(cache, skb);
256 skb = NULL;
257 goto out;
258}
259EXPORT_SYMBOL(__alloc_skb);
260
261/* Caller must provide SKB that is memset cleared */
262static struct sk_buff *__build_skb_around(struct sk_buff *skb,
263 void *data, unsigned int frag_size)
264{
265 struct skb_shared_info *shinfo;
266 unsigned int size = frag_size ? : ksize(data);
267
268 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
269
270 /* Assumes caller memset cleared SKB */
271 skb->truesize = SKB_TRUESIZE(size);
272 refcount_set(&skb->users, 1);
273 skb->head = data;
274 skb->data = data;
275 skb_reset_tail_pointer(skb);
276 skb->end = skb->tail + size;
277 skb->mac_header = (typeof(skb->mac_header))~0U;
278 skb->transport_header = (typeof(skb->transport_header))~0U;
279
280 /* make sure we initialize shinfo sequentially */
281 shinfo = skb_shinfo(skb);
282 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
283 atomic_set(&shinfo->dataref, 1);
284
285 return skb;
286}
287
288/**
289 * __build_skb - build a network buffer
290 * @data: data buffer provided by caller
291 * @frag_size: size of data, or 0 if head was kmalloced
292 *
293 * Allocate a new &sk_buff. Caller provides space holding head and
294 * skb_shared_info. @data must have been allocated by kmalloc() only if
295 * @frag_size is 0, otherwise data should come from the page allocator
296 * or vmalloc()
297 * The return is the new skb buffer.
298 * On a failure the return is %NULL, and @data is not freed.
299 * Notes :
300 * Before IO, driver allocates only data buffer where NIC put incoming frame
301 * Driver should add room at head (NET_SKB_PAD) and
302 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
303 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
304 * before giving packet to stack.
305 * RX rings only contains data buffers, not full skbs.
306 */
307struct sk_buff *__build_skb(void *data, unsigned int frag_size)
308{
309 struct sk_buff *skb;
310
311 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
312 if (unlikely(!skb))
313 return NULL;
314
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316
317 return __build_skb_around(skb, data, frag_size);
318}
319
320/* build_skb() is wrapper over __build_skb(), that specifically
321 * takes care of skb->head and skb->pfmemalloc
322 * This means that if @frag_size is not zero, then @data must be backed
323 * by a page fragment, not kmalloc() or vmalloc()
324 */
325struct sk_buff *build_skb(void *data, unsigned int frag_size)
326{
327 struct sk_buff *skb = __build_skb(data, frag_size);
328
329 if (skb && frag_size) {
330 skb->head_frag = 1;
331 if (page_is_pfmemalloc(virt_to_head_page(data)))
332 skb->pfmemalloc = 1;
333 }
334 return skb;
335}
336EXPORT_SYMBOL(build_skb);
337
338/**
339 * build_skb_around - build a network buffer around provided skb
340 * @skb: sk_buff provide by caller, must be memset cleared
341 * @data: data buffer provided by caller
342 * @frag_size: size of data, or 0 if head was kmalloced
343 */
344struct sk_buff *build_skb_around(struct sk_buff *skb,
345 void *data, unsigned int frag_size)
346{
347 if (unlikely(!skb))
348 return NULL;
349
350 skb = __build_skb_around(skb, data, frag_size);
351
352 if (skb && frag_size) {
353 skb->head_frag = 1;
354 if (page_is_pfmemalloc(virt_to_head_page(data)))
355 skb->pfmemalloc = 1;
356 }
357 return skb;
358}
359EXPORT_SYMBOL(build_skb_around);
360
361#define NAPI_SKB_CACHE_SIZE 64
362
363struct napi_alloc_cache {
364 struct page_frag_cache page;
365 unsigned int skb_count;
366 void *skb_cache[NAPI_SKB_CACHE_SIZE];
367};
368
369static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
370static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
371
372static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
373{
374 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
375
376 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
377}
378
379void *napi_alloc_frag(unsigned int fragsz)
380{
381 fragsz = SKB_DATA_ALIGN(fragsz);
382
383 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
384}
385EXPORT_SYMBOL(napi_alloc_frag);
386
387/**
388 * netdev_alloc_frag - allocate a page fragment
389 * @fragsz: fragment size
390 *
391 * Allocates a frag from a page for receive buffer.
392 * Uses GFP_ATOMIC allocations.
393 */
394void *netdev_alloc_frag(unsigned int fragsz)
395{
396 struct page_frag_cache *nc;
397 void *data;
398
399 fragsz = SKB_DATA_ALIGN(fragsz);
400 if (in_irq() || irqs_disabled()) {
401 nc = this_cpu_ptr(&netdev_alloc_cache);
402 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
403 } else {
404 local_bh_disable();
405 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
406 local_bh_enable();
407 }
408 return data;
409}
410EXPORT_SYMBOL(netdev_alloc_frag);
411
412/**
413 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
414 * @dev: network device to receive on
415 * @len: length to allocate
416 * @gfp_mask: get_free_pages mask, passed to alloc_skb
417 *
418 * Allocate a new &sk_buff and assign it a usage count of one. The
419 * buffer has NET_SKB_PAD headroom built in. Users should allocate
420 * the headroom they think they need without accounting for the
421 * built in space. The built in space is used for optimisations.
422 *
423 * %NULL is returned if there is no free memory.
424 */
425struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
426 gfp_t gfp_mask)
427{
428 struct page_frag_cache *nc;
429 struct sk_buff *skb;
430 bool pfmemalloc;
431 void *data;
432
433 len += NET_SKB_PAD;
434
435 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
436 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
437 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
438 if (!skb)
439 goto skb_fail;
440 goto skb_success;
441 }
442
443 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
444 len = SKB_DATA_ALIGN(len);
445
446 if (sk_memalloc_socks())
447 gfp_mask |= __GFP_MEMALLOC;
448
449 if (in_irq() || irqs_disabled()) {
450 nc = this_cpu_ptr(&netdev_alloc_cache);
451 data = page_frag_alloc(nc, len, gfp_mask);
452 pfmemalloc = nc->pfmemalloc;
453 } else {
454 local_bh_disable();
455 nc = this_cpu_ptr(&napi_alloc_cache.page);
456 data = page_frag_alloc(nc, len, gfp_mask);
457 pfmemalloc = nc->pfmemalloc;
458 local_bh_enable();
459 }
460
461 if (unlikely(!data))
462 return NULL;
463
464 skb = __build_skb(data, len);
465 if (unlikely(!skb)) {
466 skb_free_frag(data);
467 return NULL;
468 }
469
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 if (nc->page.pfmemalloc)
530 skb->pfmemalloc = 1;
531 skb->head_frag = 1;
532
533skb_success:
534 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
535 skb->dev = napi->dev;
536
537skb_fail:
538 return skb;
539}
540EXPORT_SYMBOL(__napi_alloc_skb);
541
542void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
543 int size, unsigned int truesize)
544{
545 skb_fill_page_desc(skb, i, page, off, size);
546 skb->len += size;
547 skb->data_len += size;
548 skb->truesize += truesize;
549}
550EXPORT_SYMBOL(skb_add_rx_frag);
551
552void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
553 unsigned int truesize)
554{
555 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
556
557 skb_frag_size_add(frag, size);
558 skb->len += size;
559 skb->data_len += size;
560 skb->truesize += truesize;
561}
562EXPORT_SYMBOL(skb_coalesce_rx_frag);
563
564static void skb_drop_list(struct sk_buff **listp)
565{
566 kfree_skb_list(*listp);
567 *listp = NULL;
568}
569
570static inline void skb_drop_fraglist(struct sk_buff *skb)
571{
572 skb_drop_list(&skb_shinfo(skb)->frag_list);
573}
574
575static void skb_clone_fraglist(struct sk_buff *skb)
576{
577 struct sk_buff *list;
578
579 skb_walk_frags(skb, list)
580 skb_get(list);
581}
582
583static void skb_free_head(struct sk_buff *skb)
584{
585 unsigned char *head = skb->head;
586
587 if (skb->head_frag)
588 skb_free_frag(head);
589 else
590 kfree(head);
591}
592
593static void skb_release_data(struct sk_buff *skb)
594{
595 struct skb_shared_info *shinfo = skb_shinfo(skb);
596 int i;
597
598 if (skb->cloned &&
599 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
600 &shinfo->dataref))
601 return;
602
603 for (i = 0; i < shinfo->nr_frags; i++)
604 __skb_frag_unref(&shinfo->frags[i]);
605
606 if (shinfo->frag_list)
607 kfree_skb_list(shinfo->frag_list);
608
609 skb_zcopy_clear(skb, true);
610 skb_free_head(skb);
611}
612
613/*
614 * Free an skbuff by memory without cleaning the state.
615 */
616static void kfree_skbmem(struct sk_buff *skb)
617{
618 struct sk_buff_fclones *fclones;
619
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
623 return;
624
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
627
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
631 */
632 if (refcount_read(&fclones->fclone_ref) == 1)
633 goto fastpath;
634 break;
635
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
638 break;
639 }
640 if (!refcount_dec_and_test(&fclones->fclone_ref))
641 return;
642fastpath:
643 kmem_cache_free(skbuff_fclone_cache, fclones);
644}
645
646void skb_release_head_state(struct sk_buff *skb)
647{
648 skb_dst_drop(skb);
649 if (skb->destructor) {
650 WARN_ON(in_irq());
651 skb->destructor(skb);
652 }
653#if IS_ENABLED(CONFIG_NF_CONNTRACK)
654 nf_conntrack_put(skb_nfct(skb));
655#endif
656 skb_ext_put(skb);
657}
658
659/* Free everything but the sk_buff shell. */
660static void skb_release_all(struct sk_buff *skb)
661{
662 skb_release_head_state(skb);
663 if (likely(skb->head))
664 skb_release_data(skb);
665}
666
667/**
668 * __kfree_skb - private function
669 * @skb: buffer
670 *
671 * Free an sk_buff. Release anything attached to the buffer.
672 * Clean the state. This is an internal helper function. Users should
673 * always call kfree_skb
674 */
675
676void __kfree_skb(struct sk_buff *skb)
677{
678 skb_release_all(skb);
679 kfree_skbmem(skb);
680}
681EXPORT_SYMBOL(__kfree_skb);
682
683/**
684 * kfree_skb - free an sk_buff
685 * @skb: buffer to free
686 *
687 * Drop a reference to the buffer and free it if the usage count has
688 * hit zero.
689 */
690void kfree_skb(struct sk_buff *skb)
691{
692 if (!skb_unref(skb))
693 return;
694
695 trace_kfree_skb(skb, __builtin_return_address(0));
696 __kfree_skb(skb);
697}
698EXPORT_SYMBOL(kfree_skb);
699
700void kfree_skb_list(struct sk_buff *segs)
701{
702 while (segs) {
703 struct sk_buff *next = segs->next;
704
705 kfree_skb(segs);
706 segs = next;
707 }
708}
709EXPORT_SYMBOL(kfree_skb_list);
710
711/* Dump skb information and contents.
712 *
713 * Must only be called from net_ratelimit()-ed paths.
714 *
715 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
716 */
717void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
718{
719 static atomic_t can_dump_full = ATOMIC_INIT(5);
720 struct skb_shared_info *sh = skb_shinfo(skb);
721 struct net_device *dev = skb->dev;
722 struct sock *sk = skb->sk;
723 struct sk_buff *list_skb;
724 bool has_mac, has_trans;
725 int headroom, tailroom;
726 int i, len, seg_len;
727
728 if (full_pkt)
729 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
730
731 if (full_pkt)
732 len = skb->len;
733 else
734 len = min_t(int, skb->len, MAX_HEADER + 128);
735
736 headroom = skb_headroom(skb);
737 tailroom = skb_tailroom(skb);
738
739 has_mac = skb_mac_header_was_set(skb);
740 has_trans = skb_transport_header_was_set(skb);
741
742 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
743 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
744 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
745 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
746 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
747 level, skb->len, headroom, skb_headlen(skb), tailroom,
748 has_mac ? skb->mac_header : -1,
749 has_mac ? skb_mac_header_len(skb) : -1,
750 skb->network_header,
751 has_trans ? skb_network_header_len(skb) : -1,
752 has_trans ? skb->transport_header : -1,
753 sh->tx_flags, sh->nr_frags,
754 sh->gso_size, sh->gso_type, sh->gso_segs,
755 skb->csum, skb->ip_summed, skb->csum_complete_sw,
756 skb->csum_valid, skb->csum_level,
757 skb->hash, skb->sw_hash, skb->l4_hash,
758 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
759
760 if (dev)
761 printk("%sdev name=%s feat=0x%pNF\n",
762 level, dev->name, &dev->features);
763 if (sk)
764 printk("%ssk family=%hu type=%u proto=%u\n",
765 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
766
767 if (full_pkt && headroom)
768 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
769 16, 1, skb->head, headroom, false);
770
771 seg_len = min_t(int, skb_headlen(skb), len);
772 if (seg_len)
773 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
774 16, 1, skb->data, seg_len, false);
775 len -= seg_len;
776
777 if (full_pkt && tailroom)
778 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb_tail_pointer(skb), tailroom, false);
780
781 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
782 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
783 u32 p_off, p_len, copied;
784 struct page *p;
785 u8 *vaddr;
786
787 skb_frag_foreach_page(frag, skb_frag_off(frag),
788 skb_frag_size(frag), p, p_off, p_len,
789 copied) {
790 seg_len = min_t(int, p_len, len);
791 vaddr = kmap_atomic(p);
792 print_hex_dump(level, "skb frag: ",
793 DUMP_PREFIX_OFFSET,
794 16, 1, vaddr + p_off, seg_len, false);
795 kunmap_atomic(vaddr);
796 len -= seg_len;
797 if (!len)
798 break;
799 }
800 }
801
802 if (full_pkt && skb_has_frag_list(skb)) {
803 printk("skb fraglist:\n");
804 skb_walk_frags(skb, list_skb)
805 skb_dump(level, list_skb, true);
806 }
807}
808EXPORT_SYMBOL(skb_dump);
809
810/**
811 * skb_tx_error - report an sk_buff xmit error
812 * @skb: buffer that triggered an error
813 *
814 * Report xmit error if a device callback is tracking this skb.
815 * skb must be freed afterwards.
816 */
817void skb_tx_error(struct sk_buff *skb)
818{
819 skb_zcopy_clear(skb, true);
820}
821EXPORT_SYMBOL(skb_tx_error);
822
823#ifdef CONFIG_TRACEPOINTS
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#endif
842
843/**
844 * consume_stateless_skb - free an skbuff, assuming it is stateless
845 * @skb: buffer to free
846 *
847 * Alike consume_skb(), but this variant assumes that this is the last
848 * skb reference and all the head states have been already dropped
849 */
850void __consume_stateless_skb(struct sk_buff *skb)
851{
852 trace_consume_skb(skb);
853 skb_release_data(skb);
854 kfree_skbmem(skb);
855}
856
857void __kfree_skb_flush(void)
858{
859 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
860
861 /* flush skb_cache if containing objects */
862 if (nc->skb_count) {
863 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
864 nc->skb_cache);
865 nc->skb_count = 0;
866 }
867}
868
869static inline void _kfree_skb_defer(struct sk_buff *skb)
870{
871 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
872
873 /* drop skb->head and call any destructors for packet */
874 skb_release_all(skb);
875
876 /* record skb to CPU local list */
877 nc->skb_cache[nc->skb_count++] = skb;
878
879#ifdef CONFIG_SLUB
880 /* SLUB writes into objects when freeing */
881 prefetchw(skb);
882#endif
883
884 /* flush skb_cache if it is filled */
885 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
886 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
887 nc->skb_cache);
888 nc->skb_count = 0;
889 }
890}
891void __kfree_skb_defer(struct sk_buff *skb)
892{
893 _kfree_skb_defer(skb);
894}
895
896void napi_consume_skb(struct sk_buff *skb, int budget)
897{
898 if (unlikely(!skb))
899 return;
900
901 /* Zero budget indicate non-NAPI context called us, like netpoll */
902 if (unlikely(!budget)) {
903 dev_consume_skb_any(skb);
904 return;
905 }
906
907 if (!skb_unref(skb))
908 return;
909
910 /* if reaching here SKB is ready to free */
911 trace_consume_skb(skb);
912
913 /* if SKB is a clone, don't handle this case */
914 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
915 __kfree_skb(skb);
916 return;
917 }
918
919 _kfree_skb_defer(skb);
920}
921EXPORT_SYMBOL(napi_consume_skb);
922
923/* Make sure a field is enclosed inside headers_start/headers_end section */
924#define CHECK_SKB_FIELD(field) \
925 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
926 offsetof(struct sk_buff, headers_start)); \
927 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
928 offsetof(struct sk_buff, headers_end)); \
929
930static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
931{
932 new->tstamp = old->tstamp;
933 /* We do not copy old->sk */
934 new->dev = old->dev;
935 memcpy(new->cb, old->cb, sizeof(old->cb));
936 skb_dst_copy(new, old);
937 __skb_ext_copy(new, old);
938 __nf_copy(new, old, false);
939
940 /* Note : this field could be in headers_start/headers_end section
941 * It is not yet because we do not want to have a 16 bit hole
942 */
943 new->queue_mapping = old->queue_mapping;
944
945 memcpy(&new->headers_start, &old->headers_start,
946 offsetof(struct sk_buff, headers_end) -
947 offsetof(struct sk_buff, headers_start));
948 CHECK_SKB_FIELD(protocol);
949 CHECK_SKB_FIELD(csum);
950 CHECK_SKB_FIELD(hash);
951 CHECK_SKB_FIELD(priority);
952 CHECK_SKB_FIELD(skb_iif);
953 CHECK_SKB_FIELD(vlan_proto);
954 CHECK_SKB_FIELD(vlan_tci);
955 CHECK_SKB_FIELD(transport_header);
956 CHECK_SKB_FIELD(network_header);
957 CHECK_SKB_FIELD(mac_header);
958 CHECK_SKB_FIELD(inner_protocol);
959 CHECK_SKB_FIELD(inner_transport_header);
960 CHECK_SKB_FIELD(inner_network_header);
961 CHECK_SKB_FIELD(inner_mac_header);
962 CHECK_SKB_FIELD(mark);
963#ifdef CONFIG_NETWORK_SECMARK
964 CHECK_SKB_FIELD(secmark);
965#endif
966#ifdef CONFIG_NET_RX_BUSY_POLL
967 CHECK_SKB_FIELD(napi_id);
968#endif
969#ifdef CONFIG_XPS
970 CHECK_SKB_FIELD(sender_cpu);
971#endif
972#ifdef CONFIG_NET_SCHED
973 CHECK_SKB_FIELD(tc_index);
974#endif
975
976}
977
978/*
979 * You should not add any new code to this function. Add it to
980 * __copy_skb_header above instead.
981 */
982static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
983{
984#define C(x) n->x = skb->x
985
986 n->next = n->prev = NULL;
987 n->sk = NULL;
988 __copy_skb_header(n, skb);
989
990 C(len);
991 C(data_len);
992 C(mac_len);
993 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
994 n->cloned = 1;
995 n->nohdr = 0;
996 n->peeked = 0;
997 C(pfmemalloc);
998 n->destructor = NULL;
999 C(tail);
1000 C(end);
1001 C(head);
1002 C(head_frag);
1003 C(data);
1004 C(truesize);
1005 refcount_set(&n->users, 1);
1006
1007 atomic_inc(&(skb_shinfo(skb)->dataref));
1008 skb->cloned = 1;
1009
1010 return n;
1011#undef C
1012}
1013
1014/**
1015 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1016 * @first: first sk_buff of the msg
1017 */
1018struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1019{
1020 struct sk_buff *n;
1021
1022 n = alloc_skb(0, GFP_ATOMIC);
1023 if (!n)
1024 return NULL;
1025
1026 n->len = first->len;
1027 n->data_len = first->len;
1028 n->truesize = first->truesize;
1029
1030 skb_shinfo(n)->frag_list = first;
1031
1032 __copy_skb_header(n, first);
1033 n->destructor = NULL;
1034
1035 return n;
1036}
1037EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1038
1039/**
1040 * skb_morph - morph one skb into another
1041 * @dst: the skb to receive the contents
1042 * @src: the skb to supply the contents
1043 *
1044 * This is identical to skb_clone except that the target skb is
1045 * supplied by the user.
1046 *
1047 * The target skb is returned upon exit.
1048 */
1049struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1050{
1051 skb_release_all(dst);
1052 return __skb_clone(dst, src);
1053}
1054EXPORT_SYMBOL_GPL(skb_morph);
1055
1056int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1057{
1058 unsigned long max_pg, num_pg, new_pg, old_pg;
1059 struct user_struct *user;
1060
1061 if (capable(CAP_IPC_LOCK) || !size)
1062 return 0;
1063
1064 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1065 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1066 user = mmp->user ? : current_user();
1067
1068 do {
1069 old_pg = atomic_long_read(&user->locked_vm);
1070 new_pg = old_pg + num_pg;
1071 if (new_pg > max_pg)
1072 return -ENOBUFS;
1073 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1074 old_pg);
1075
1076 if (!mmp->user) {
1077 mmp->user = get_uid(user);
1078 mmp->num_pg = num_pg;
1079 } else {
1080 mmp->num_pg += num_pg;
1081 }
1082
1083 return 0;
1084}
1085EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1086
1087void mm_unaccount_pinned_pages(struct mmpin *mmp)
1088{
1089 if (mmp->user) {
1090 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1091 free_uid(mmp->user);
1092 }
1093}
1094EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1095
1096struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1097{
1098 struct ubuf_info *uarg;
1099 struct sk_buff *skb;
1100
1101 WARN_ON_ONCE(!in_task());
1102
1103 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1104 if (!skb)
1105 return NULL;
1106
1107 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1108 uarg = (void *)skb->cb;
1109 uarg->mmp.user = NULL;
1110
1111 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1112 kfree_skb(skb);
1113 return NULL;
1114 }
1115
1116 uarg->callback = sock_zerocopy_callback;
1117 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1118 uarg->len = 1;
1119 uarg->bytelen = size;
1120 uarg->zerocopy = 1;
1121 refcount_set(&uarg->refcnt, 1);
1122 sock_hold(sk);
1123
1124 return uarg;
1125}
1126EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1127
1128static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1129{
1130 return container_of((void *)uarg, struct sk_buff, cb);
1131}
1132
1133struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1134 struct ubuf_info *uarg)
1135{
1136 if (uarg) {
1137 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1138 u32 bytelen, next;
1139
1140 /* realloc only when socket is locked (TCP, UDP cork),
1141 * so uarg->len and sk_zckey access is serialized
1142 */
1143 if (!sock_owned_by_user(sk)) {
1144 WARN_ON_ONCE(1);
1145 return NULL;
1146 }
1147
1148 bytelen = uarg->bytelen + size;
1149 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1150 /* TCP can create new skb to attach new uarg */
1151 if (sk->sk_type == SOCK_STREAM)
1152 goto new_alloc;
1153 return NULL;
1154 }
1155
1156 next = (u32)atomic_read(&sk->sk_zckey);
1157 if ((u32)(uarg->id + uarg->len) == next) {
1158 if (mm_account_pinned_pages(&uarg->mmp, size))
1159 return NULL;
1160 uarg->len++;
1161 uarg->bytelen = bytelen;
1162 atomic_set(&sk->sk_zckey, ++next);
1163
1164 /* no extra ref when appending to datagram (MSG_MORE) */
1165 if (sk->sk_type == SOCK_STREAM)
1166 sock_zerocopy_get(uarg);
1167
1168 return uarg;
1169 }
1170 }
1171
1172new_alloc:
1173 return sock_zerocopy_alloc(sk, size);
1174}
1175EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1176
1177static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1178{
1179 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1180 u32 old_lo, old_hi;
1181 u64 sum_len;
1182
1183 old_lo = serr->ee.ee_info;
1184 old_hi = serr->ee.ee_data;
1185 sum_len = old_hi - old_lo + 1ULL + len;
1186
1187 if (sum_len >= (1ULL << 32))
1188 return false;
1189
1190 if (lo != old_hi + 1)
1191 return false;
1192
1193 serr->ee.ee_data += len;
1194 return true;
1195}
1196
1197void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1198{
1199 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1200 struct sock_exterr_skb *serr;
1201 struct sock *sk = skb->sk;
1202 struct sk_buff_head *q;
1203 unsigned long flags;
1204 u32 lo, hi;
1205 u16 len;
1206
1207 mm_unaccount_pinned_pages(&uarg->mmp);
1208
1209 /* if !len, there was only 1 call, and it was aborted
1210 * so do not queue a completion notification
1211 */
1212 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1213 goto release;
1214
1215 len = uarg->len;
1216 lo = uarg->id;
1217 hi = uarg->id + len - 1;
1218
1219 serr = SKB_EXT_ERR(skb);
1220 memset(serr, 0, sizeof(*serr));
1221 serr->ee.ee_errno = 0;
1222 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1223 serr->ee.ee_data = hi;
1224 serr->ee.ee_info = lo;
1225 if (!success)
1226 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1227
1228 q = &sk->sk_error_queue;
1229 spin_lock_irqsave(&q->lock, flags);
1230 tail = skb_peek_tail(q);
1231 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1232 !skb_zerocopy_notify_extend(tail, lo, len)) {
1233 __skb_queue_tail(q, skb);
1234 skb = NULL;
1235 }
1236 spin_unlock_irqrestore(&q->lock, flags);
1237
1238 sk->sk_error_report(sk);
1239
1240release:
1241 consume_skb(skb);
1242 sock_put(sk);
1243}
1244EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1245
1246void sock_zerocopy_put(struct ubuf_info *uarg)
1247{
1248 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1249 if (uarg->callback)
1250 uarg->callback(uarg, uarg->zerocopy);
1251 else
1252 consume_skb(skb_from_uarg(uarg));
1253 }
1254}
1255EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1256
1257void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1258{
1259 if (uarg) {
1260 struct sock *sk = skb_from_uarg(uarg)->sk;
1261
1262 atomic_dec(&sk->sk_zckey);
1263 uarg->len--;
1264
1265 if (have_uref)
1266 sock_zerocopy_put(uarg);
1267 }
1268}
1269EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1270
1271int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1272{
1273 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1274}
1275EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1276
1277int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1278 struct msghdr *msg, int len,
1279 struct ubuf_info *uarg)
1280{
1281 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1282 struct iov_iter orig_iter = msg->msg_iter;
1283 int err, orig_len = skb->len;
1284
1285 /* An skb can only point to one uarg. This edge case happens when
1286 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1287 */
1288 if (orig_uarg && uarg != orig_uarg)
1289 return -EEXIST;
1290
1291 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1292 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1293 struct sock *save_sk = skb->sk;
1294
1295 /* Streams do not free skb on error. Reset to prev state. */
1296 msg->msg_iter = orig_iter;
1297 skb->sk = sk;
1298 ___pskb_trim(skb, orig_len);
1299 skb->sk = save_sk;
1300 return err;
1301 }
1302
1303 skb_zcopy_set(skb, uarg, NULL);
1304 return skb->len - orig_len;
1305}
1306EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1307
1308static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1309 gfp_t gfp_mask)
1310{
1311 if (skb_zcopy(orig)) {
1312 if (skb_zcopy(nskb)) {
1313 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1314 if (!gfp_mask) {
1315 WARN_ON_ONCE(1);
1316 return -ENOMEM;
1317 }
1318 if (skb_uarg(nskb) == skb_uarg(orig))
1319 return 0;
1320 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1321 return -EIO;
1322 }
1323 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1324 }
1325 return 0;
1326}
1327
1328/**
1329 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1330 * @skb: the skb to modify
1331 * @gfp_mask: allocation priority
1332 *
1333 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1334 * It will copy all frags into kernel and drop the reference
1335 * to userspace pages.
1336 *
1337 * If this function is called from an interrupt gfp_mask() must be
1338 * %GFP_ATOMIC.
1339 *
1340 * Returns 0 on success or a negative error code on failure
1341 * to allocate kernel memory to copy to.
1342 */
1343int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1344{
1345 int num_frags = skb_shinfo(skb)->nr_frags;
1346 struct page *page, *head = NULL;
1347 int i, new_frags;
1348 u32 d_off;
1349
1350 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1351 return -EINVAL;
1352
1353 if (!num_frags)
1354 goto release;
1355
1356 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1357 for (i = 0; i < new_frags; i++) {
1358 page = alloc_page(gfp_mask);
1359 if (!page) {
1360 while (head) {
1361 struct page *next = (struct page *)page_private(head);
1362 put_page(head);
1363 head = next;
1364 }
1365 return -ENOMEM;
1366 }
1367 set_page_private(page, (unsigned long)head);
1368 head = page;
1369 }
1370
1371 page = head;
1372 d_off = 0;
1373 for (i = 0; i < num_frags; i++) {
1374 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1375 u32 p_off, p_len, copied;
1376 struct page *p;
1377 u8 *vaddr;
1378
1379 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1380 p, p_off, p_len, copied) {
1381 u32 copy, done = 0;
1382 vaddr = kmap_atomic(p);
1383
1384 while (done < p_len) {
1385 if (d_off == PAGE_SIZE) {
1386 d_off = 0;
1387 page = (struct page *)page_private(page);
1388 }
1389 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1390 memcpy(page_address(page) + d_off,
1391 vaddr + p_off + done, copy);
1392 done += copy;
1393 d_off += copy;
1394 }
1395 kunmap_atomic(vaddr);
1396 }
1397 }
1398
1399 /* skb frags release userspace buffers */
1400 for (i = 0; i < num_frags; i++)
1401 skb_frag_unref(skb, i);
1402
1403 /* skb frags point to kernel buffers */
1404 for (i = 0; i < new_frags - 1; i++) {
1405 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1406 head = (struct page *)page_private(head);
1407 }
1408 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1409 skb_shinfo(skb)->nr_frags = new_frags;
1410
1411release:
1412 skb_zcopy_clear(skb, false);
1413 return 0;
1414}
1415EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1416
1417/**
1418 * skb_clone - duplicate an sk_buff
1419 * @skb: buffer to clone
1420 * @gfp_mask: allocation priority
1421 *
1422 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1423 * copies share the same packet data but not structure. The new
1424 * buffer has a reference count of 1. If the allocation fails the
1425 * function returns %NULL otherwise the new buffer is returned.
1426 *
1427 * If this function is called from an interrupt gfp_mask() must be
1428 * %GFP_ATOMIC.
1429 */
1430
1431struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1432{
1433 struct sk_buff_fclones *fclones = container_of(skb,
1434 struct sk_buff_fclones,
1435 skb1);
1436 struct sk_buff *n;
1437
1438 if (skb_orphan_frags(skb, gfp_mask))
1439 return NULL;
1440
1441 if (skb->fclone == SKB_FCLONE_ORIG &&
1442 refcount_read(&fclones->fclone_ref) == 1) {
1443 n = &fclones->skb2;
1444 refcount_set(&fclones->fclone_ref, 2);
1445 } else {
1446 if (skb_pfmemalloc(skb))
1447 gfp_mask |= __GFP_MEMALLOC;
1448
1449 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1450 if (!n)
1451 return NULL;
1452
1453 n->fclone = SKB_FCLONE_UNAVAILABLE;
1454 }
1455
1456 return __skb_clone(n, skb);
1457}
1458EXPORT_SYMBOL(skb_clone);
1459
1460void skb_headers_offset_update(struct sk_buff *skb, int off)
1461{
1462 /* Only adjust this if it actually is csum_start rather than csum */
1463 if (skb->ip_summed == CHECKSUM_PARTIAL)
1464 skb->csum_start += off;
1465 /* {transport,network,mac}_header and tail are relative to skb->head */
1466 skb->transport_header += off;
1467 skb->network_header += off;
1468 if (skb_mac_header_was_set(skb))
1469 skb->mac_header += off;
1470 skb->inner_transport_header += off;
1471 skb->inner_network_header += off;
1472 skb->inner_mac_header += off;
1473}
1474EXPORT_SYMBOL(skb_headers_offset_update);
1475
1476void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1477{
1478 __copy_skb_header(new, old);
1479
1480 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1481 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1482 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1483}
1484EXPORT_SYMBOL(skb_copy_header);
1485
1486static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1487{
1488 if (skb_pfmemalloc(skb))
1489 return SKB_ALLOC_RX;
1490 return 0;
1491}
1492
1493/**
1494 * skb_copy - create private copy of an sk_buff
1495 * @skb: buffer to copy
1496 * @gfp_mask: allocation priority
1497 *
1498 * Make a copy of both an &sk_buff and its data. This is used when the
1499 * caller wishes to modify the data and needs a private copy of the
1500 * data to alter. Returns %NULL on failure or the pointer to the buffer
1501 * on success. The returned buffer has a reference count of 1.
1502 *
1503 * As by-product this function converts non-linear &sk_buff to linear
1504 * one, so that &sk_buff becomes completely private and caller is allowed
1505 * to modify all the data of returned buffer. This means that this
1506 * function is not recommended for use in circumstances when only
1507 * header is going to be modified. Use pskb_copy() instead.
1508 */
1509
1510struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1511{
1512 int headerlen = skb_headroom(skb);
1513 unsigned int size = skb_end_offset(skb) + skb->data_len;
1514 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1515 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1516
1517 if (!n)
1518 return NULL;
1519
1520 /* Set the data pointer */
1521 skb_reserve(n, headerlen);
1522 /* Set the tail pointer and length */
1523 skb_put(n, skb->len);
1524
1525 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1526
1527 skb_copy_header(n, skb);
1528 return n;
1529}
1530EXPORT_SYMBOL(skb_copy);
1531
1532/**
1533 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1534 * @skb: buffer to copy
1535 * @headroom: headroom of new skb
1536 * @gfp_mask: allocation priority
1537 * @fclone: if true allocate the copy of the skb from the fclone
1538 * cache instead of the head cache; it is recommended to set this
1539 * to true for the cases where the copy will likely be cloned
1540 *
1541 * Make a copy of both an &sk_buff and part of its data, located
1542 * in header. Fragmented data remain shared. This is used when
1543 * the caller wishes to modify only header of &sk_buff and needs
1544 * private copy of the header to alter. Returns %NULL on failure
1545 * or the pointer to the buffer on success.
1546 * The returned buffer has a reference count of 1.
1547 */
1548
1549struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1550 gfp_t gfp_mask, bool fclone)
1551{
1552 unsigned int size = skb_headlen(skb) + headroom;
1553 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1554 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1555
1556 if (!n)
1557 goto out;
1558
1559 /* Set the data pointer */
1560 skb_reserve(n, headroom);
1561 /* Set the tail pointer and length */
1562 skb_put(n, skb_headlen(skb));
1563 /* Copy the bytes */
1564 skb_copy_from_linear_data(skb, n->data, n->len);
1565
1566 n->truesize += skb->data_len;
1567 n->data_len = skb->data_len;
1568 n->len = skb->len;
1569
1570 if (skb_shinfo(skb)->nr_frags) {
1571 int i;
1572
1573 if (skb_orphan_frags(skb, gfp_mask) ||
1574 skb_zerocopy_clone(n, skb, gfp_mask)) {
1575 kfree_skb(n);
1576 n = NULL;
1577 goto out;
1578 }
1579 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1580 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1581 skb_frag_ref(skb, i);
1582 }
1583 skb_shinfo(n)->nr_frags = i;
1584 }
1585
1586 if (skb_has_frag_list(skb)) {
1587 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1588 skb_clone_fraglist(n);
1589 }
1590
1591 skb_copy_header(n, skb);
1592out:
1593 return n;
1594}
1595EXPORT_SYMBOL(__pskb_copy_fclone);
1596
1597/**
1598 * pskb_expand_head - reallocate header of &sk_buff
1599 * @skb: buffer to reallocate
1600 * @nhead: room to add at head
1601 * @ntail: room to add at tail
1602 * @gfp_mask: allocation priority
1603 *
1604 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1605 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1606 * reference count of 1. Returns zero in the case of success or error,
1607 * if expansion failed. In the last case, &sk_buff is not changed.
1608 *
1609 * All the pointers pointing into skb header may change and must be
1610 * reloaded after call to this function.
1611 */
1612
1613int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1614 gfp_t gfp_mask)
1615{
1616 int i, osize = skb_end_offset(skb);
1617 int size = osize + nhead + ntail;
1618 long off;
1619 u8 *data;
1620
1621 BUG_ON(nhead < 0);
1622
1623 BUG_ON(skb_shared(skb));
1624
1625 size = SKB_DATA_ALIGN(size);
1626
1627 if (skb_pfmemalloc(skb))
1628 gfp_mask |= __GFP_MEMALLOC;
1629 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1630 gfp_mask, NUMA_NO_NODE, NULL);
1631 if (!data)
1632 goto nodata;
1633 size = SKB_WITH_OVERHEAD(ksize(data));
1634
1635 /* Copy only real data... and, alas, header. This should be
1636 * optimized for the cases when header is void.
1637 */
1638 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1639
1640 memcpy((struct skb_shared_info *)(data + size),
1641 skb_shinfo(skb),
1642 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1643
1644 /*
1645 * if shinfo is shared we must drop the old head gracefully, but if it
1646 * is not we can just drop the old head and let the existing refcount
1647 * be since all we did is relocate the values
1648 */
1649 if (skb_cloned(skb)) {
1650 if (skb_orphan_frags(skb, gfp_mask))
1651 goto nofrags;
1652 if (skb_zcopy(skb))
1653 refcount_inc(&skb_uarg(skb)->refcnt);
1654 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1655 skb_frag_ref(skb, i);
1656
1657 if (skb_has_frag_list(skb))
1658 skb_clone_fraglist(skb);
1659
1660 skb_release_data(skb);
1661 } else {
1662 skb_free_head(skb);
1663 }
1664 off = (data + nhead) - skb->head;
1665
1666 skb->head = data;
1667 skb->head_frag = 0;
1668 skb->data += off;
1669#ifdef NET_SKBUFF_DATA_USES_OFFSET
1670 skb->end = size;
1671 off = nhead;
1672#else
1673 skb->end = skb->head + size;
1674#endif
1675 skb->tail += off;
1676 skb_headers_offset_update(skb, nhead);
1677 skb->cloned = 0;
1678 skb->hdr_len = 0;
1679 skb->nohdr = 0;
1680 atomic_set(&skb_shinfo(skb)->dataref, 1);
1681
1682 skb_metadata_clear(skb);
1683
1684 /* It is not generally safe to change skb->truesize.
1685 * For the moment, we really care of rx path, or
1686 * when skb is orphaned (not attached to a socket).
1687 */
1688 if (!skb->sk || skb->destructor == sock_edemux)
1689 skb->truesize += size - osize;
1690
1691 return 0;
1692
1693nofrags:
1694 kfree(data);
1695nodata:
1696 return -ENOMEM;
1697}
1698EXPORT_SYMBOL(pskb_expand_head);
1699
1700/* Make private copy of skb with writable head and some headroom */
1701
1702struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1703{
1704 struct sk_buff *skb2;
1705 int delta = headroom - skb_headroom(skb);
1706
1707 if (delta <= 0)
1708 skb2 = pskb_copy(skb, GFP_ATOMIC);
1709 else {
1710 skb2 = skb_clone(skb, GFP_ATOMIC);
1711 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1712 GFP_ATOMIC)) {
1713 kfree_skb(skb2);
1714 skb2 = NULL;
1715 }
1716 }
1717 return skb2;
1718}
1719EXPORT_SYMBOL(skb_realloc_headroom);
1720
1721/**
1722 * skb_copy_expand - copy and expand sk_buff
1723 * @skb: buffer to copy
1724 * @newheadroom: new free bytes at head
1725 * @newtailroom: new free bytes at tail
1726 * @gfp_mask: allocation priority
1727 *
1728 * Make a copy of both an &sk_buff and its data and while doing so
1729 * allocate additional space.
1730 *
1731 * This is used when the caller wishes to modify the data and needs a
1732 * private copy of the data to alter as well as more space for new fields.
1733 * Returns %NULL on failure or the pointer to the buffer
1734 * on success. The returned buffer has a reference count of 1.
1735 *
1736 * You must pass %GFP_ATOMIC as the allocation priority if this function
1737 * is called from an interrupt.
1738 */
1739struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1740 int newheadroom, int newtailroom,
1741 gfp_t gfp_mask)
1742{
1743 /*
1744 * Allocate the copy buffer
1745 */
1746 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1747 gfp_mask, skb_alloc_rx_flag(skb),
1748 NUMA_NO_NODE);
1749 int oldheadroom = skb_headroom(skb);
1750 int head_copy_len, head_copy_off;
1751
1752 if (!n)
1753 return NULL;
1754
1755 skb_reserve(n, newheadroom);
1756
1757 /* Set the tail pointer and length */
1758 skb_put(n, skb->len);
1759
1760 head_copy_len = oldheadroom;
1761 head_copy_off = 0;
1762 if (newheadroom <= head_copy_len)
1763 head_copy_len = newheadroom;
1764 else
1765 head_copy_off = newheadroom - head_copy_len;
1766
1767 /* Copy the linear header and data. */
1768 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1769 skb->len + head_copy_len));
1770
1771 skb_copy_header(n, skb);
1772
1773 skb_headers_offset_update(n, newheadroom - oldheadroom);
1774
1775 return n;
1776}
1777EXPORT_SYMBOL(skb_copy_expand);
1778
1779/**
1780 * __skb_pad - zero pad the tail of an skb
1781 * @skb: buffer to pad
1782 * @pad: space to pad
1783 * @free_on_error: free buffer on error
1784 *
1785 * Ensure that a buffer is followed by a padding area that is zero
1786 * filled. Used by network drivers which may DMA or transfer data
1787 * beyond the buffer end onto the wire.
1788 *
1789 * May return error in out of memory cases. The skb is freed on error
1790 * if @free_on_error is true.
1791 */
1792
1793int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1794{
1795 int err;
1796 int ntail;
1797
1798 /* If the skbuff is non linear tailroom is always zero.. */
1799 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1800 memset(skb->data+skb->len, 0, pad);
1801 return 0;
1802 }
1803
1804 ntail = skb->data_len + pad - (skb->end - skb->tail);
1805 if (likely(skb_cloned(skb) || ntail > 0)) {
1806 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1807 if (unlikely(err))
1808 goto free_skb;
1809 }
1810
1811 /* FIXME: The use of this function with non-linear skb's really needs
1812 * to be audited.
1813 */
1814 err = skb_linearize(skb);
1815 if (unlikely(err))
1816 goto free_skb;
1817
1818 memset(skb->data + skb->len, 0, pad);
1819 return 0;
1820
1821free_skb:
1822 if (free_on_error)
1823 kfree_skb(skb);
1824 return err;
1825}
1826EXPORT_SYMBOL(__skb_pad);
1827
1828/**
1829 * pskb_put - add data to the tail of a potentially fragmented buffer
1830 * @skb: start of the buffer to use
1831 * @tail: tail fragment of the buffer to use
1832 * @len: amount of data to add
1833 *
1834 * This function extends the used data area of the potentially
1835 * fragmented buffer. @tail must be the last fragment of @skb -- or
1836 * @skb itself. If this would exceed the total buffer size the kernel
1837 * will panic. A pointer to the first byte of the extra data is
1838 * returned.
1839 */
1840
1841void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1842{
1843 if (tail != skb) {
1844 skb->data_len += len;
1845 skb->len += len;
1846 }
1847 return skb_put(tail, len);
1848}
1849EXPORT_SYMBOL_GPL(pskb_put);
1850
1851/**
1852 * skb_put - add data to a buffer
1853 * @skb: buffer to use
1854 * @len: amount of data to add
1855 *
1856 * This function extends the used data area of the buffer. If this would
1857 * exceed the total buffer size the kernel will panic. A pointer to the
1858 * first byte of the extra data is returned.
1859 */
1860void *skb_put(struct sk_buff *skb, unsigned int len)
1861{
1862 void *tmp = skb_tail_pointer(skb);
1863 SKB_LINEAR_ASSERT(skb);
1864 skb->tail += len;
1865 skb->len += len;
1866 if (unlikely(skb->tail > skb->end))
1867 skb_over_panic(skb, len, __builtin_return_address(0));
1868 return tmp;
1869}
1870EXPORT_SYMBOL(skb_put);
1871
1872/**
1873 * skb_push - add data to the start of a buffer
1874 * @skb: buffer to use
1875 * @len: amount of data to add
1876 *
1877 * This function extends the used data area of the buffer at the buffer
1878 * start. If this would exceed the total buffer headroom the kernel will
1879 * panic. A pointer to the first byte of the extra data is returned.
1880 */
1881void *skb_push(struct sk_buff *skb, unsigned int len)
1882{
1883 skb->data -= len;
1884 skb->len += len;
1885 if (unlikely(skb->data < skb->head))
1886 skb_under_panic(skb, len, __builtin_return_address(0));
1887 return skb->data;
1888}
1889EXPORT_SYMBOL(skb_push);
1890
1891/**
1892 * skb_pull - remove data from the start of a buffer
1893 * @skb: buffer to use
1894 * @len: amount of data to remove
1895 *
1896 * This function removes data from the start of a buffer, returning
1897 * the memory to the headroom. A pointer to the next data in the buffer
1898 * is returned. Once the data has been pulled future pushes will overwrite
1899 * the old data.
1900 */
1901void *skb_pull(struct sk_buff *skb, unsigned int len)
1902{
1903 return skb_pull_inline(skb, len);
1904}
1905EXPORT_SYMBOL(skb_pull);
1906
1907/**
1908 * skb_trim - remove end from a buffer
1909 * @skb: buffer to alter
1910 * @len: new length
1911 *
1912 * Cut the length of a buffer down by removing data from the tail. If
1913 * the buffer is already under the length specified it is not modified.
1914 * The skb must be linear.
1915 */
1916void skb_trim(struct sk_buff *skb, unsigned int len)
1917{
1918 if (skb->len > len)
1919 __skb_trim(skb, len);
1920}
1921EXPORT_SYMBOL(skb_trim);
1922
1923/* Trims skb to length len. It can change skb pointers.
1924 */
1925
1926int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1927{
1928 struct sk_buff **fragp;
1929 struct sk_buff *frag;
1930 int offset = skb_headlen(skb);
1931 int nfrags = skb_shinfo(skb)->nr_frags;
1932 int i;
1933 int err;
1934
1935 if (skb_cloned(skb) &&
1936 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1937 return err;
1938
1939 i = 0;
1940 if (offset >= len)
1941 goto drop_pages;
1942
1943 for (; i < nfrags; i++) {
1944 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1945
1946 if (end < len) {
1947 offset = end;
1948 continue;
1949 }
1950
1951 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1952
1953drop_pages:
1954 skb_shinfo(skb)->nr_frags = i;
1955
1956 for (; i < nfrags; i++)
1957 skb_frag_unref(skb, i);
1958
1959 if (skb_has_frag_list(skb))
1960 skb_drop_fraglist(skb);
1961 goto done;
1962 }
1963
1964 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1965 fragp = &frag->next) {
1966 int end = offset + frag->len;
1967
1968 if (skb_shared(frag)) {
1969 struct sk_buff *nfrag;
1970
1971 nfrag = skb_clone(frag, GFP_ATOMIC);
1972 if (unlikely(!nfrag))
1973 return -ENOMEM;
1974
1975 nfrag->next = frag->next;
1976 consume_skb(frag);
1977 frag = nfrag;
1978 *fragp = frag;
1979 }
1980
1981 if (end < len) {
1982 offset = end;
1983 continue;
1984 }
1985
1986 if (end > len &&
1987 unlikely((err = pskb_trim(frag, len - offset))))
1988 return err;
1989
1990 if (frag->next)
1991 skb_drop_list(&frag->next);
1992 break;
1993 }
1994
1995done:
1996 if (len > skb_headlen(skb)) {
1997 skb->data_len -= skb->len - len;
1998 skb->len = len;
1999 } else {
2000 skb->len = len;
2001 skb->data_len = 0;
2002 skb_set_tail_pointer(skb, len);
2003 }
2004
2005 if (!skb->sk || skb->destructor == sock_edemux)
2006 skb_condense(skb);
2007 return 0;
2008}
2009EXPORT_SYMBOL(___pskb_trim);
2010
2011/* Note : use pskb_trim_rcsum() instead of calling this directly
2012 */
2013int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2014{
2015 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2016 int delta = skb->len - len;
2017
2018 skb->csum = csum_block_sub(skb->csum,
2019 skb_checksum(skb, len, delta, 0),
2020 len);
2021 }
2022 return __pskb_trim(skb, len);
2023}
2024EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2025
2026/**
2027 * __pskb_pull_tail - advance tail of skb header
2028 * @skb: buffer to reallocate
2029 * @delta: number of bytes to advance tail
2030 *
2031 * The function makes a sense only on a fragmented &sk_buff,
2032 * it expands header moving its tail forward and copying necessary
2033 * data from fragmented part.
2034 *
2035 * &sk_buff MUST have reference count of 1.
2036 *
2037 * Returns %NULL (and &sk_buff does not change) if pull failed
2038 * or value of new tail of skb in the case of success.
2039 *
2040 * All the pointers pointing into skb header may change and must be
2041 * reloaded after call to this function.
2042 */
2043
2044/* Moves tail of skb head forward, copying data from fragmented part,
2045 * when it is necessary.
2046 * 1. It may fail due to malloc failure.
2047 * 2. It may change skb pointers.
2048 *
2049 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2050 */
2051void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2052{
2053 /* If skb has not enough free space at tail, get new one
2054 * plus 128 bytes for future expansions. If we have enough
2055 * room at tail, reallocate without expansion only if skb is cloned.
2056 */
2057 int i, k, eat = (skb->tail + delta) - skb->end;
2058
2059 if (eat > 0 || skb_cloned(skb)) {
2060 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2061 GFP_ATOMIC))
2062 return NULL;
2063 }
2064
2065 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2066 skb_tail_pointer(skb), delta));
2067
2068 /* Optimization: no fragments, no reasons to preestimate
2069 * size of pulled pages. Superb.
2070 */
2071 if (!skb_has_frag_list(skb))
2072 goto pull_pages;
2073
2074 /* Estimate size of pulled pages. */
2075 eat = delta;
2076 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2077 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2078
2079 if (size >= eat)
2080 goto pull_pages;
2081 eat -= size;
2082 }
2083
2084 /* If we need update frag list, we are in troubles.
2085 * Certainly, it is possible to add an offset to skb data,
2086 * but taking into account that pulling is expected to
2087 * be very rare operation, it is worth to fight against
2088 * further bloating skb head and crucify ourselves here instead.
2089 * Pure masohism, indeed. 8)8)
2090 */
2091 if (eat) {
2092 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2093 struct sk_buff *clone = NULL;
2094 struct sk_buff *insp = NULL;
2095
2096 do {
2097 if (list->len <= eat) {
2098 /* Eaten as whole. */
2099 eat -= list->len;
2100 list = list->next;
2101 insp = list;
2102 } else {
2103 /* Eaten partially. */
2104
2105 if (skb_shared(list)) {
2106 /* Sucks! We need to fork list. :-( */
2107 clone = skb_clone(list, GFP_ATOMIC);
2108 if (!clone)
2109 return NULL;
2110 insp = list->next;
2111 list = clone;
2112 } else {
2113 /* This may be pulled without
2114 * problems. */
2115 insp = list;
2116 }
2117 if (!pskb_pull(list, eat)) {
2118 kfree_skb(clone);
2119 return NULL;
2120 }
2121 break;
2122 }
2123 } while (eat);
2124
2125 /* Free pulled out fragments. */
2126 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2127 skb_shinfo(skb)->frag_list = list->next;
2128 kfree_skb(list);
2129 }
2130 /* And insert new clone at head. */
2131 if (clone) {
2132 clone->next = list;
2133 skb_shinfo(skb)->frag_list = clone;
2134 }
2135 }
2136 /* Success! Now we may commit changes to skb data. */
2137
2138pull_pages:
2139 eat = delta;
2140 k = 0;
2141 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2142 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2143
2144 if (size <= eat) {
2145 skb_frag_unref(skb, i);
2146 eat -= size;
2147 } else {
2148 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2149
2150 *frag = skb_shinfo(skb)->frags[i];
2151 if (eat) {
2152 skb_frag_off_add(frag, eat);
2153 skb_frag_size_sub(frag, eat);
2154 if (!i)
2155 goto end;
2156 eat = 0;
2157 }
2158 k++;
2159 }
2160 }
2161 skb_shinfo(skb)->nr_frags = k;
2162
2163end:
2164 skb->tail += delta;
2165 skb->data_len -= delta;
2166
2167 if (!skb->data_len)
2168 skb_zcopy_clear(skb, false);
2169
2170 return skb_tail_pointer(skb);
2171}
2172EXPORT_SYMBOL(__pskb_pull_tail);
2173
2174/**
2175 * skb_copy_bits - copy bits from skb to kernel buffer
2176 * @skb: source skb
2177 * @offset: offset in source
2178 * @to: destination buffer
2179 * @len: number of bytes to copy
2180 *
2181 * Copy the specified number of bytes from the source skb to the
2182 * destination buffer.
2183 *
2184 * CAUTION ! :
2185 * If its prototype is ever changed,
2186 * check arch/{*}/net/{*}.S files,
2187 * since it is called from BPF assembly code.
2188 */
2189int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2190{
2191 int start = skb_headlen(skb);
2192 struct sk_buff *frag_iter;
2193 int i, copy;
2194
2195 if (offset > (int)skb->len - len)
2196 goto fault;
2197
2198 /* Copy header. */
2199 if ((copy = start - offset) > 0) {
2200 if (copy > len)
2201 copy = len;
2202 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2203 if ((len -= copy) == 0)
2204 return 0;
2205 offset += copy;
2206 to += copy;
2207 }
2208
2209 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2210 int end;
2211 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2212
2213 WARN_ON(start > offset + len);
2214
2215 end = start + skb_frag_size(f);
2216 if ((copy = end - offset) > 0) {
2217 u32 p_off, p_len, copied;
2218 struct page *p;
2219 u8 *vaddr;
2220
2221 if (copy > len)
2222 copy = len;
2223
2224 skb_frag_foreach_page(f,
2225 skb_frag_off(f) + offset - start,
2226 copy, p, p_off, p_len, copied) {
2227 vaddr = kmap_atomic(p);
2228 memcpy(to + copied, vaddr + p_off, p_len);
2229 kunmap_atomic(vaddr);
2230 }
2231
2232 if ((len -= copy) == 0)
2233 return 0;
2234 offset += copy;
2235 to += copy;
2236 }
2237 start = end;
2238 }
2239
2240 skb_walk_frags(skb, frag_iter) {
2241 int end;
2242
2243 WARN_ON(start > offset + len);
2244
2245 end = start + frag_iter->len;
2246 if ((copy = end - offset) > 0) {
2247 if (copy > len)
2248 copy = len;
2249 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2250 goto fault;
2251 if ((len -= copy) == 0)
2252 return 0;
2253 offset += copy;
2254 to += copy;
2255 }
2256 start = end;
2257 }
2258
2259 if (!len)
2260 return 0;
2261
2262fault:
2263 return -EFAULT;
2264}
2265EXPORT_SYMBOL(skb_copy_bits);
2266
2267/*
2268 * Callback from splice_to_pipe(), if we need to release some pages
2269 * at the end of the spd in case we error'ed out in filling the pipe.
2270 */
2271static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2272{
2273 put_page(spd->pages[i]);
2274}
2275
2276static struct page *linear_to_page(struct page *page, unsigned int *len,
2277 unsigned int *offset,
2278 struct sock *sk)
2279{
2280 struct page_frag *pfrag = sk_page_frag(sk);
2281
2282 if (!sk_page_frag_refill(sk, pfrag))
2283 return NULL;
2284
2285 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2286
2287 memcpy(page_address(pfrag->page) + pfrag->offset,
2288 page_address(page) + *offset, *len);
2289 *offset = pfrag->offset;
2290 pfrag->offset += *len;
2291
2292 return pfrag->page;
2293}
2294
2295static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2296 struct page *page,
2297 unsigned int offset)
2298{
2299 return spd->nr_pages &&
2300 spd->pages[spd->nr_pages - 1] == page &&
2301 (spd->partial[spd->nr_pages - 1].offset +
2302 spd->partial[spd->nr_pages - 1].len == offset);
2303}
2304
2305/*
2306 * Fill page/offset/length into spd, if it can hold more pages.
2307 */
2308static bool spd_fill_page(struct splice_pipe_desc *spd,
2309 struct pipe_inode_info *pipe, struct page *page,
2310 unsigned int *len, unsigned int offset,
2311 bool linear,
2312 struct sock *sk)
2313{
2314 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2315 return true;
2316
2317 if (linear) {
2318 page = linear_to_page(page, len, &offset, sk);
2319 if (!page)
2320 return true;
2321 }
2322 if (spd_can_coalesce(spd, page, offset)) {
2323 spd->partial[spd->nr_pages - 1].len += *len;
2324 return false;
2325 }
2326 get_page(page);
2327 spd->pages[spd->nr_pages] = page;
2328 spd->partial[spd->nr_pages].len = *len;
2329 spd->partial[spd->nr_pages].offset = offset;
2330 spd->nr_pages++;
2331
2332 return false;
2333}
2334
2335static bool __splice_segment(struct page *page, unsigned int poff,
2336 unsigned int plen, unsigned int *off,
2337 unsigned int *len,
2338 struct splice_pipe_desc *spd, bool linear,
2339 struct sock *sk,
2340 struct pipe_inode_info *pipe)
2341{
2342 if (!*len)
2343 return true;
2344
2345 /* skip this segment if already processed */
2346 if (*off >= plen) {
2347 *off -= plen;
2348 return false;
2349 }
2350
2351 /* ignore any bits we already processed */
2352 poff += *off;
2353 plen -= *off;
2354 *off = 0;
2355
2356 do {
2357 unsigned int flen = min(*len, plen);
2358
2359 if (spd_fill_page(spd, pipe, page, &flen, poff,
2360 linear, sk))
2361 return true;
2362 poff += flen;
2363 plen -= flen;
2364 *len -= flen;
2365 } while (*len && plen);
2366
2367 return false;
2368}
2369
2370/*
2371 * Map linear and fragment data from the skb to spd. It reports true if the
2372 * pipe is full or if we already spliced the requested length.
2373 */
2374static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2375 unsigned int *offset, unsigned int *len,
2376 struct splice_pipe_desc *spd, struct sock *sk)
2377{
2378 int seg;
2379 struct sk_buff *iter;
2380
2381 /* map the linear part :
2382 * If skb->head_frag is set, this 'linear' part is backed by a
2383 * fragment, and if the head is not shared with any clones then
2384 * we can avoid a copy since we own the head portion of this page.
2385 */
2386 if (__splice_segment(virt_to_page(skb->data),
2387 (unsigned long) skb->data & (PAGE_SIZE - 1),
2388 skb_headlen(skb),
2389 offset, len, spd,
2390 skb_head_is_locked(skb),
2391 sk, pipe))
2392 return true;
2393
2394 /*
2395 * then map the fragments
2396 */
2397 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2398 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2399
2400 if (__splice_segment(skb_frag_page(f),
2401 skb_frag_off(f), skb_frag_size(f),
2402 offset, len, spd, false, sk, pipe))
2403 return true;
2404 }
2405
2406 skb_walk_frags(skb, iter) {
2407 if (*offset >= iter->len) {
2408 *offset -= iter->len;
2409 continue;
2410 }
2411 /* __skb_splice_bits() only fails if the output has no room
2412 * left, so no point in going over the frag_list for the error
2413 * case.
2414 */
2415 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2416 return true;
2417 }
2418
2419 return false;
2420}
2421
2422/*
2423 * Map data from the skb to a pipe. Should handle both the linear part,
2424 * the fragments, and the frag list.
2425 */
2426int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2427 struct pipe_inode_info *pipe, unsigned int tlen,
2428 unsigned int flags)
2429{
2430 struct partial_page partial[MAX_SKB_FRAGS];
2431 struct page *pages[MAX_SKB_FRAGS];
2432 struct splice_pipe_desc spd = {
2433 .pages = pages,
2434 .partial = partial,
2435 .nr_pages_max = MAX_SKB_FRAGS,
2436 .ops = &nosteal_pipe_buf_ops,
2437 .spd_release = sock_spd_release,
2438 };
2439 int ret = 0;
2440
2441 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2442
2443 if (spd.nr_pages)
2444 ret = splice_to_pipe(pipe, &spd);
2445
2446 return ret;
2447}
2448EXPORT_SYMBOL_GPL(skb_splice_bits);
2449
2450/* Send skb data on a socket. Socket must be locked. */
2451int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2452 int len)
2453{
2454 unsigned int orig_len = len;
2455 struct sk_buff *head = skb;
2456 unsigned short fragidx;
2457 int slen, ret;
2458
2459do_frag_list:
2460
2461 /* Deal with head data */
2462 while (offset < skb_headlen(skb) && len) {
2463 struct kvec kv;
2464 struct msghdr msg;
2465
2466 slen = min_t(int, len, skb_headlen(skb) - offset);
2467 kv.iov_base = skb->data + offset;
2468 kv.iov_len = slen;
2469 memset(&msg, 0, sizeof(msg));
2470 msg.msg_flags = MSG_DONTWAIT;
2471
2472 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2473 if (ret <= 0)
2474 goto error;
2475
2476 offset += ret;
2477 len -= ret;
2478 }
2479
2480 /* All the data was skb head? */
2481 if (!len)
2482 goto out;
2483
2484 /* Make offset relative to start of frags */
2485 offset -= skb_headlen(skb);
2486
2487 /* Find where we are in frag list */
2488 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2489 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2490
2491 if (offset < skb_frag_size(frag))
2492 break;
2493
2494 offset -= skb_frag_size(frag);
2495 }
2496
2497 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2498 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2499
2500 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2501
2502 while (slen) {
2503 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2504 skb_frag_off(frag) + offset,
2505 slen, MSG_DONTWAIT);
2506 if (ret <= 0)
2507 goto error;
2508
2509 len -= ret;
2510 offset += ret;
2511 slen -= ret;
2512 }
2513
2514 offset = 0;
2515 }
2516
2517 if (len) {
2518 /* Process any frag lists */
2519
2520 if (skb == head) {
2521 if (skb_has_frag_list(skb)) {
2522 skb = skb_shinfo(skb)->frag_list;
2523 goto do_frag_list;
2524 }
2525 } else if (skb->next) {
2526 skb = skb->next;
2527 goto do_frag_list;
2528 }
2529 }
2530
2531out:
2532 return orig_len - len;
2533
2534error:
2535 return orig_len == len ? ret : orig_len - len;
2536}
2537EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2538
2539/**
2540 * skb_store_bits - store bits from kernel buffer to skb
2541 * @skb: destination buffer
2542 * @offset: offset in destination
2543 * @from: source buffer
2544 * @len: number of bytes to copy
2545 *
2546 * Copy the specified number of bytes from the source buffer to the
2547 * destination skb. This function handles all the messy bits of
2548 * traversing fragment lists and such.
2549 */
2550
2551int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2552{
2553 int start = skb_headlen(skb);
2554 struct sk_buff *frag_iter;
2555 int i, copy;
2556
2557 if (offset > (int)skb->len - len)
2558 goto fault;
2559
2560 if ((copy = start - offset) > 0) {
2561 if (copy > len)
2562 copy = len;
2563 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2564 if ((len -= copy) == 0)
2565 return 0;
2566 offset += copy;
2567 from += copy;
2568 }
2569
2570 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2571 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2572 int end;
2573
2574 WARN_ON(start > offset + len);
2575
2576 end = start + skb_frag_size(frag);
2577 if ((copy = end - offset) > 0) {
2578 u32 p_off, p_len, copied;
2579 struct page *p;
2580 u8 *vaddr;
2581
2582 if (copy > len)
2583 copy = len;
2584
2585 skb_frag_foreach_page(frag,
2586 skb_frag_off(frag) + offset - start,
2587 copy, p, p_off, p_len, copied) {
2588 vaddr = kmap_atomic(p);
2589 memcpy(vaddr + p_off, from + copied, p_len);
2590 kunmap_atomic(vaddr);
2591 }
2592
2593 if ((len -= copy) == 0)
2594 return 0;
2595 offset += copy;
2596 from += copy;
2597 }
2598 start = end;
2599 }
2600
2601 skb_walk_frags(skb, frag_iter) {
2602 int end;
2603
2604 WARN_ON(start > offset + len);
2605
2606 end = start + frag_iter->len;
2607 if ((copy = end - offset) > 0) {
2608 if (copy > len)
2609 copy = len;
2610 if (skb_store_bits(frag_iter, offset - start,
2611 from, copy))
2612 goto fault;
2613 if ((len -= copy) == 0)
2614 return 0;
2615 offset += copy;
2616 from += copy;
2617 }
2618 start = end;
2619 }
2620 if (!len)
2621 return 0;
2622
2623fault:
2624 return -EFAULT;
2625}
2626EXPORT_SYMBOL(skb_store_bits);
2627
2628/* Checksum skb data. */
2629__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2630 __wsum csum, const struct skb_checksum_ops *ops)
2631{
2632 int start = skb_headlen(skb);
2633 int i, copy = start - offset;
2634 struct sk_buff *frag_iter;
2635 int pos = 0;
2636
2637 /* Checksum header. */
2638 if (copy > 0) {
2639 if (copy > len)
2640 copy = len;
2641 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2642 skb->data + offset, copy, csum);
2643 if ((len -= copy) == 0)
2644 return csum;
2645 offset += copy;
2646 pos = copy;
2647 }
2648
2649 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2650 int end;
2651 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2652
2653 WARN_ON(start > offset + len);
2654
2655 end = start + skb_frag_size(frag);
2656 if ((copy = end - offset) > 0) {
2657 u32 p_off, p_len, copied;
2658 struct page *p;
2659 __wsum csum2;
2660 u8 *vaddr;
2661
2662 if (copy > len)
2663 copy = len;
2664
2665 skb_frag_foreach_page(frag,
2666 skb_frag_off(frag) + offset - start,
2667 copy, p, p_off, p_len, copied) {
2668 vaddr = kmap_atomic(p);
2669 csum2 = INDIRECT_CALL_1(ops->update,
2670 csum_partial_ext,
2671 vaddr + p_off, p_len, 0);
2672 kunmap_atomic(vaddr);
2673 csum = INDIRECT_CALL_1(ops->combine,
2674 csum_block_add_ext, csum,
2675 csum2, pos, p_len);
2676 pos += p_len;
2677 }
2678
2679 if (!(len -= copy))
2680 return csum;
2681 offset += copy;
2682 }
2683 start = end;
2684 }
2685
2686 skb_walk_frags(skb, frag_iter) {
2687 int end;
2688
2689 WARN_ON(start > offset + len);
2690
2691 end = start + frag_iter->len;
2692 if ((copy = end - offset) > 0) {
2693 __wsum csum2;
2694 if (copy > len)
2695 copy = len;
2696 csum2 = __skb_checksum(frag_iter, offset - start,
2697 copy, 0, ops);
2698 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2699 csum, csum2, pos, copy);
2700 if ((len -= copy) == 0)
2701 return csum;
2702 offset += copy;
2703 pos += copy;
2704 }
2705 start = end;
2706 }
2707 BUG_ON(len);
2708
2709 return csum;
2710}
2711EXPORT_SYMBOL(__skb_checksum);
2712
2713__wsum skb_checksum(const struct sk_buff *skb, int offset,
2714 int len, __wsum csum)
2715{
2716 const struct skb_checksum_ops ops = {
2717 .update = csum_partial_ext,
2718 .combine = csum_block_add_ext,
2719 };
2720
2721 return __skb_checksum(skb, offset, len, csum, &ops);
2722}
2723EXPORT_SYMBOL(skb_checksum);
2724
2725/* Both of above in one bottle. */
2726
2727__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2728 u8 *to, int len, __wsum csum)
2729{
2730 int start = skb_headlen(skb);
2731 int i, copy = start - offset;
2732 struct sk_buff *frag_iter;
2733 int pos = 0;
2734
2735 /* Copy header. */
2736 if (copy > 0) {
2737 if (copy > len)
2738 copy = len;
2739 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2740 copy, csum);
2741 if ((len -= copy) == 0)
2742 return csum;
2743 offset += copy;
2744 to += copy;
2745 pos = copy;
2746 }
2747
2748 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2749 int end;
2750
2751 WARN_ON(start > offset + len);
2752
2753 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2754 if ((copy = end - offset) > 0) {
2755 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2756 u32 p_off, p_len, copied;
2757 struct page *p;
2758 __wsum csum2;
2759 u8 *vaddr;
2760
2761 if (copy > len)
2762 copy = len;
2763
2764 skb_frag_foreach_page(frag,
2765 skb_frag_off(frag) + offset - start,
2766 copy, p, p_off, p_len, copied) {
2767 vaddr = kmap_atomic(p);
2768 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2769 to + copied,
2770 p_len, 0);
2771 kunmap_atomic(vaddr);
2772 csum = csum_block_add(csum, csum2, pos);
2773 pos += p_len;
2774 }
2775
2776 if (!(len -= copy))
2777 return csum;
2778 offset += copy;
2779 to += copy;
2780 }
2781 start = end;
2782 }
2783
2784 skb_walk_frags(skb, frag_iter) {
2785 __wsum csum2;
2786 int end;
2787
2788 WARN_ON(start > offset + len);
2789
2790 end = start + frag_iter->len;
2791 if ((copy = end - offset) > 0) {
2792 if (copy > len)
2793 copy = len;
2794 csum2 = skb_copy_and_csum_bits(frag_iter,
2795 offset - start,
2796 to, copy, 0);
2797 csum = csum_block_add(csum, csum2, pos);
2798 if ((len -= copy) == 0)
2799 return csum;
2800 offset += copy;
2801 to += copy;
2802 pos += copy;
2803 }
2804 start = end;
2805 }
2806 BUG_ON(len);
2807 return csum;
2808}
2809EXPORT_SYMBOL(skb_copy_and_csum_bits);
2810
2811__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2812{
2813 __sum16 sum;
2814
2815 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2816 /* See comments in __skb_checksum_complete(). */
2817 if (likely(!sum)) {
2818 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2819 !skb->csum_complete_sw)
2820 netdev_rx_csum_fault(skb->dev, skb);
2821 }
2822 if (!skb_shared(skb))
2823 skb->csum_valid = !sum;
2824 return sum;
2825}
2826EXPORT_SYMBOL(__skb_checksum_complete_head);
2827
2828/* This function assumes skb->csum already holds pseudo header's checksum,
2829 * which has been changed from the hardware checksum, for example, by
2830 * __skb_checksum_validate_complete(). And, the original skb->csum must
2831 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2832 *
2833 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2834 * zero. The new checksum is stored back into skb->csum unless the skb is
2835 * shared.
2836 */
2837__sum16 __skb_checksum_complete(struct sk_buff *skb)
2838{
2839 __wsum csum;
2840 __sum16 sum;
2841
2842 csum = skb_checksum(skb, 0, skb->len, 0);
2843
2844 sum = csum_fold(csum_add(skb->csum, csum));
2845 /* This check is inverted, because we already knew the hardware
2846 * checksum is invalid before calling this function. So, if the
2847 * re-computed checksum is valid instead, then we have a mismatch
2848 * between the original skb->csum and skb_checksum(). This means either
2849 * the original hardware checksum is incorrect or we screw up skb->csum
2850 * when moving skb->data around.
2851 */
2852 if (likely(!sum)) {
2853 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2854 !skb->csum_complete_sw)
2855 netdev_rx_csum_fault(skb->dev, skb);
2856 }
2857
2858 if (!skb_shared(skb)) {
2859 /* Save full packet checksum */
2860 skb->csum = csum;
2861 skb->ip_summed = CHECKSUM_COMPLETE;
2862 skb->csum_complete_sw = 1;
2863 skb->csum_valid = !sum;
2864 }
2865
2866 return sum;
2867}
2868EXPORT_SYMBOL(__skb_checksum_complete);
2869
2870static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2871{
2872 net_warn_ratelimited(
2873 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2874 __func__);
2875 return 0;
2876}
2877
2878static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2879 int offset, int len)
2880{
2881 net_warn_ratelimited(
2882 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2883 __func__);
2884 return 0;
2885}
2886
2887static const struct skb_checksum_ops default_crc32c_ops = {
2888 .update = warn_crc32c_csum_update,
2889 .combine = warn_crc32c_csum_combine,
2890};
2891
2892const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2893 &default_crc32c_ops;
2894EXPORT_SYMBOL(crc32c_csum_stub);
2895
2896 /**
2897 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2898 * @from: source buffer
2899 *
2900 * Calculates the amount of linear headroom needed in the 'to' skb passed
2901 * into skb_zerocopy().
2902 */
2903unsigned int
2904skb_zerocopy_headlen(const struct sk_buff *from)
2905{
2906 unsigned int hlen = 0;
2907
2908 if (!from->head_frag ||
2909 skb_headlen(from) < L1_CACHE_BYTES ||
2910 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2911 hlen = skb_headlen(from);
2912
2913 if (skb_has_frag_list(from))
2914 hlen = from->len;
2915
2916 return hlen;
2917}
2918EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2919
2920/**
2921 * skb_zerocopy - Zero copy skb to skb
2922 * @to: destination buffer
2923 * @from: source buffer
2924 * @len: number of bytes to copy from source buffer
2925 * @hlen: size of linear headroom in destination buffer
2926 *
2927 * Copies up to `len` bytes from `from` to `to` by creating references
2928 * to the frags in the source buffer.
2929 *
2930 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2931 * headroom in the `to` buffer.
2932 *
2933 * Return value:
2934 * 0: everything is OK
2935 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2936 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2937 */
2938int
2939skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2940{
2941 int i, j = 0;
2942 int plen = 0; /* length of skb->head fragment */
2943 int ret;
2944 struct page *page;
2945 unsigned int offset;
2946
2947 BUG_ON(!from->head_frag && !hlen);
2948
2949 /* dont bother with small payloads */
2950 if (len <= skb_tailroom(to))
2951 return skb_copy_bits(from, 0, skb_put(to, len), len);
2952
2953 if (hlen) {
2954 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2955 if (unlikely(ret))
2956 return ret;
2957 len -= hlen;
2958 } else {
2959 plen = min_t(int, skb_headlen(from), len);
2960 if (plen) {
2961 page = virt_to_head_page(from->head);
2962 offset = from->data - (unsigned char *)page_address(page);
2963 __skb_fill_page_desc(to, 0, page, offset, plen);
2964 get_page(page);
2965 j = 1;
2966 len -= plen;
2967 }
2968 }
2969
2970 to->truesize += len + plen;
2971 to->len += len + plen;
2972 to->data_len += len + plen;
2973
2974 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2975 skb_tx_error(from);
2976 return -ENOMEM;
2977 }
2978 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2979
2980 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2981 int size;
2982
2983 if (!len)
2984 break;
2985 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2986 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2987 len);
2988 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2989 len -= size;
2990 skb_frag_ref(to, j);
2991 j++;
2992 }
2993 skb_shinfo(to)->nr_frags = j;
2994
2995 return 0;
2996}
2997EXPORT_SYMBOL_GPL(skb_zerocopy);
2998
2999void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3000{
3001 __wsum csum;
3002 long csstart;
3003
3004 if (skb->ip_summed == CHECKSUM_PARTIAL)
3005 csstart = skb_checksum_start_offset(skb);
3006 else
3007 csstart = skb_headlen(skb);
3008
3009 BUG_ON(csstart > skb_headlen(skb));
3010
3011 skb_copy_from_linear_data(skb, to, csstart);
3012
3013 csum = 0;
3014 if (csstart != skb->len)
3015 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3016 skb->len - csstart, 0);
3017
3018 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3019 long csstuff = csstart + skb->csum_offset;
3020
3021 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3022 }
3023}
3024EXPORT_SYMBOL(skb_copy_and_csum_dev);
3025
3026/**
3027 * skb_dequeue - remove from the head of the queue
3028 * @list: list to dequeue from
3029 *
3030 * Remove the head of the list. The list lock is taken so the function
3031 * may be used safely with other locking list functions. The head item is
3032 * returned or %NULL if the list is empty.
3033 */
3034
3035struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3036{
3037 unsigned long flags;
3038 struct sk_buff *result;
3039
3040 spin_lock_irqsave(&list->lock, flags);
3041 result = __skb_dequeue(list);
3042 spin_unlock_irqrestore(&list->lock, flags);
3043 return result;
3044}
3045EXPORT_SYMBOL(skb_dequeue);
3046
3047/**
3048 * skb_dequeue_tail - remove from the tail of the queue
3049 * @list: list to dequeue from
3050 *
3051 * Remove the tail of the list. The list lock is taken so the function
3052 * may be used safely with other locking list functions. The tail item is
3053 * returned or %NULL if the list is empty.
3054 */
3055struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3056{
3057 unsigned long flags;
3058 struct sk_buff *result;
3059
3060 spin_lock_irqsave(&list->lock, flags);
3061 result = __skb_dequeue_tail(list);
3062 spin_unlock_irqrestore(&list->lock, flags);
3063 return result;
3064}
3065EXPORT_SYMBOL(skb_dequeue_tail);
3066
3067/**
3068 * skb_queue_purge - empty a list
3069 * @list: list to empty
3070 *
3071 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3072 * the list and one reference dropped. This function takes the list
3073 * lock and is atomic with respect to other list locking functions.
3074 */
3075void skb_queue_purge(struct sk_buff_head *list)
3076{
3077 struct sk_buff *skb;
3078 while ((skb = skb_dequeue(list)) != NULL)
3079 kfree_skb(skb);
3080}
3081EXPORT_SYMBOL(skb_queue_purge);
3082
3083/**
3084 * skb_rbtree_purge - empty a skb rbtree
3085 * @root: root of the rbtree to empty
3086 * Return value: the sum of truesizes of all purged skbs.
3087 *
3088 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3089 * the list and one reference dropped. This function does not take
3090 * any lock. Synchronization should be handled by the caller (e.g., TCP
3091 * out-of-order queue is protected by the socket lock).
3092 */
3093unsigned int skb_rbtree_purge(struct rb_root *root)
3094{
3095 struct rb_node *p = rb_first(root);
3096 unsigned int sum = 0;
3097
3098 while (p) {
3099 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3100
3101 p = rb_next(p);
3102 rb_erase(&skb->rbnode, root);
3103 sum += skb->truesize;
3104 kfree_skb(skb);
3105 }
3106 return sum;
3107}
3108
3109/**
3110 * skb_queue_head - queue a buffer at the list head
3111 * @list: list to use
3112 * @newsk: buffer to queue
3113 *
3114 * Queue a buffer at the start of the list. This function takes the
3115 * list lock and can be used safely with other locking &sk_buff functions
3116 * safely.
3117 *
3118 * A buffer cannot be placed on two lists at the same time.
3119 */
3120void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3121{
3122 unsigned long flags;
3123
3124 spin_lock_irqsave(&list->lock, flags);
3125 __skb_queue_head(list, newsk);
3126 spin_unlock_irqrestore(&list->lock, flags);
3127}
3128EXPORT_SYMBOL(skb_queue_head);
3129
3130/**
3131 * skb_queue_tail - queue a buffer at the list tail
3132 * @list: list to use
3133 * @newsk: buffer to queue
3134 *
3135 * Queue a buffer at the tail of the list. This function takes the
3136 * list lock and can be used safely with other locking &sk_buff functions
3137 * safely.
3138 *
3139 * A buffer cannot be placed on two lists at the same time.
3140 */
3141void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3142{
3143 unsigned long flags;
3144
3145 spin_lock_irqsave(&list->lock, flags);
3146 __skb_queue_tail(list, newsk);
3147 spin_unlock_irqrestore(&list->lock, flags);
3148}
3149EXPORT_SYMBOL(skb_queue_tail);
3150
3151/**
3152 * skb_unlink - remove a buffer from a list
3153 * @skb: buffer to remove
3154 * @list: list to use
3155 *
3156 * Remove a packet from a list. The list locks are taken and this
3157 * function is atomic with respect to other list locked calls
3158 *
3159 * You must know what list the SKB is on.
3160 */
3161void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3162{
3163 unsigned long flags;
3164
3165 spin_lock_irqsave(&list->lock, flags);
3166 __skb_unlink(skb, list);
3167 spin_unlock_irqrestore(&list->lock, flags);
3168}
3169EXPORT_SYMBOL(skb_unlink);
3170
3171/**
3172 * skb_append - append a buffer
3173 * @old: buffer to insert after
3174 * @newsk: buffer to insert
3175 * @list: list to use
3176 *
3177 * Place a packet after a given packet in a list. The list locks are taken
3178 * and this function is atomic with respect to other list locked calls.
3179 * A buffer cannot be placed on two lists at the same time.
3180 */
3181void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3182{
3183 unsigned long flags;
3184
3185 spin_lock_irqsave(&list->lock, flags);
3186 __skb_queue_after(list, old, newsk);
3187 spin_unlock_irqrestore(&list->lock, flags);
3188}
3189EXPORT_SYMBOL(skb_append);
3190
3191static inline void skb_split_inside_header(struct sk_buff *skb,
3192 struct sk_buff* skb1,
3193 const u32 len, const int pos)
3194{
3195 int i;
3196
3197 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3198 pos - len);
3199 /* And move data appendix as is. */
3200 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3201 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3202
3203 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3204 skb_shinfo(skb)->nr_frags = 0;
3205 skb1->data_len = skb->data_len;
3206 skb1->len += skb1->data_len;
3207 skb->data_len = 0;
3208 skb->len = len;
3209 skb_set_tail_pointer(skb, len);
3210}
3211
3212static inline void skb_split_no_header(struct sk_buff *skb,
3213 struct sk_buff* skb1,
3214 const u32 len, int pos)
3215{
3216 int i, k = 0;
3217 const int nfrags = skb_shinfo(skb)->nr_frags;
3218
3219 skb_shinfo(skb)->nr_frags = 0;
3220 skb1->len = skb1->data_len = skb->len - len;
3221 skb->len = len;
3222 skb->data_len = len - pos;
3223
3224 for (i = 0; i < nfrags; i++) {
3225 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3226
3227 if (pos + size > len) {
3228 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3229
3230 if (pos < len) {
3231 /* Split frag.
3232 * We have two variants in this case:
3233 * 1. Move all the frag to the second
3234 * part, if it is possible. F.e.
3235 * this approach is mandatory for TUX,
3236 * where splitting is expensive.
3237 * 2. Split is accurately. We make this.
3238 */
3239 skb_frag_ref(skb, i);
3240 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3241 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3242 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3243 skb_shinfo(skb)->nr_frags++;
3244 }
3245 k++;
3246 } else
3247 skb_shinfo(skb)->nr_frags++;
3248 pos += size;
3249 }
3250 skb_shinfo(skb1)->nr_frags = k;
3251}
3252
3253/**
3254 * skb_split - Split fragmented skb to two parts at length len.
3255 * @skb: the buffer to split
3256 * @skb1: the buffer to receive the second part
3257 * @len: new length for skb
3258 */
3259void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3260{
3261 int pos = skb_headlen(skb);
3262
3263 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3264 SKBTX_SHARED_FRAG;
3265 skb_zerocopy_clone(skb1, skb, 0);
3266 if (len < pos) /* Split line is inside header. */
3267 skb_split_inside_header(skb, skb1, len, pos);
3268 else /* Second chunk has no header, nothing to copy. */
3269 skb_split_no_header(skb, skb1, len, pos);
3270}
3271EXPORT_SYMBOL(skb_split);
3272
3273/* Shifting from/to a cloned skb is a no-go.
3274 *
3275 * Caller cannot keep skb_shinfo related pointers past calling here!
3276 */
3277static int skb_prepare_for_shift(struct sk_buff *skb)
3278{
3279 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3280}
3281
3282/**
3283 * skb_shift - Shifts paged data partially from skb to another
3284 * @tgt: buffer into which tail data gets added
3285 * @skb: buffer from which the paged data comes from
3286 * @shiftlen: shift up to this many bytes
3287 *
3288 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3289 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3290 * It's up to caller to free skb if everything was shifted.
3291 *
3292 * If @tgt runs out of frags, the whole operation is aborted.
3293 *
3294 * Skb cannot include anything else but paged data while tgt is allowed
3295 * to have non-paged data as well.
3296 *
3297 * TODO: full sized shift could be optimized but that would need
3298 * specialized skb free'er to handle frags without up-to-date nr_frags.
3299 */
3300int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3301{
3302 int from, to, merge, todo;
3303 skb_frag_t *fragfrom, *fragto;
3304
3305 BUG_ON(shiftlen > skb->len);
3306
3307 if (skb_headlen(skb))
3308 return 0;
3309 if (skb_zcopy(tgt) || skb_zcopy(skb))
3310 return 0;
3311
3312 todo = shiftlen;
3313 from = 0;
3314 to = skb_shinfo(tgt)->nr_frags;
3315 fragfrom = &skb_shinfo(skb)->frags[from];
3316
3317 /* Actual merge is delayed until the point when we know we can
3318 * commit all, so that we don't have to undo partial changes
3319 */
3320 if (!to ||
3321 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3322 skb_frag_off(fragfrom))) {
3323 merge = -1;
3324 } else {
3325 merge = to - 1;
3326
3327 todo -= skb_frag_size(fragfrom);
3328 if (todo < 0) {
3329 if (skb_prepare_for_shift(skb) ||
3330 skb_prepare_for_shift(tgt))
3331 return 0;
3332
3333 /* All previous frag pointers might be stale! */
3334 fragfrom = &skb_shinfo(skb)->frags[from];
3335 fragto = &skb_shinfo(tgt)->frags[merge];
3336
3337 skb_frag_size_add(fragto, shiftlen);
3338 skb_frag_size_sub(fragfrom, shiftlen);
3339 skb_frag_off_add(fragfrom, shiftlen);
3340
3341 goto onlymerged;
3342 }
3343
3344 from++;
3345 }
3346
3347 /* Skip full, not-fitting skb to avoid expensive operations */
3348 if ((shiftlen == skb->len) &&
3349 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3350 return 0;
3351
3352 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3353 return 0;
3354
3355 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3356 if (to == MAX_SKB_FRAGS)
3357 return 0;
3358
3359 fragfrom = &skb_shinfo(skb)->frags[from];
3360 fragto = &skb_shinfo(tgt)->frags[to];
3361
3362 if (todo >= skb_frag_size(fragfrom)) {
3363 *fragto = *fragfrom;
3364 todo -= skb_frag_size(fragfrom);
3365 from++;
3366 to++;
3367
3368 } else {
3369 __skb_frag_ref(fragfrom);
3370 skb_frag_page_copy(fragto, fragfrom);
3371 skb_frag_off_copy(fragto, fragfrom);
3372 skb_frag_size_set(fragto, todo);
3373
3374 skb_frag_off_add(fragfrom, todo);
3375 skb_frag_size_sub(fragfrom, todo);
3376 todo = 0;
3377
3378 to++;
3379 break;
3380 }
3381 }
3382
3383 /* Ready to "commit" this state change to tgt */
3384 skb_shinfo(tgt)->nr_frags = to;
3385
3386 if (merge >= 0) {
3387 fragfrom = &skb_shinfo(skb)->frags[0];
3388 fragto = &skb_shinfo(tgt)->frags[merge];
3389
3390 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3391 __skb_frag_unref(fragfrom);
3392 }
3393
3394 /* Reposition in the original skb */
3395 to = 0;
3396 while (from < skb_shinfo(skb)->nr_frags)
3397 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3398 skb_shinfo(skb)->nr_frags = to;
3399
3400 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3401
3402onlymerged:
3403 /* Most likely the tgt won't ever need its checksum anymore, skb on
3404 * the other hand might need it if it needs to be resent
3405 */
3406 tgt->ip_summed = CHECKSUM_PARTIAL;
3407 skb->ip_summed = CHECKSUM_PARTIAL;
3408
3409 /* Yak, is it really working this way? Some helper please? */
3410 skb->len -= shiftlen;
3411 skb->data_len -= shiftlen;
3412 skb->truesize -= shiftlen;
3413 tgt->len += shiftlen;
3414 tgt->data_len += shiftlen;
3415 tgt->truesize += shiftlen;
3416
3417 return shiftlen;
3418}
3419
3420/**
3421 * skb_prepare_seq_read - Prepare a sequential read of skb data
3422 * @skb: the buffer to read
3423 * @from: lower offset of data to be read
3424 * @to: upper offset of data to be read
3425 * @st: state variable
3426 *
3427 * Initializes the specified state variable. Must be called before
3428 * invoking skb_seq_read() for the first time.
3429 */
3430void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3431 unsigned int to, struct skb_seq_state *st)
3432{
3433 st->lower_offset = from;
3434 st->upper_offset = to;
3435 st->root_skb = st->cur_skb = skb;
3436 st->frag_idx = st->stepped_offset = 0;
3437 st->frag_data = NULL;
3438}
3439EXPORT_SYMBOL(skb_prepare_seq_read);
3440
3441/**
3442 * skb_seq_read - Sequentially read skb data
3443 * @consumed: number of bytes consumed by the caller so far
3444 * @data: destination pointer for data to be returned
3445 * @st: state variable
3446 *
3447 * Reads a block of skb data at @consumed relative to the
3448 * lower offset specified to skb_prepare_seq_read(). Assigns
3449 * the head of the data block to @data and returns the length
3450 * of the block or 0 if the end of the skb data or the upper
3451 * offset has been reached.
3452 *
3453 * The caller is not required to consume all of the data
3454 * returned, i.e. @consumed is typically set to the number
3455 * of bytes already consumed and the next call to
3456 * skb_seq_read() will return the remaining part of the block.
3457 *
3458 * Note 1: The size of each block of data returned can be arbitrary,
3459 * this limitation is the cost for zerocopy sequential
3460 * reads of potentially non linear data.
3461 *
3462 * Note 2: Fragment lists within fragments are not implemented
3463 * at the moment, state->root_skb could be replaced with
3464 * a stack for this purpose.
3465 */
3466unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3467 struct skb_seq_state *st)
3468{
3469 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3470 skb_frag_t *frag;
3471
3472 if (unlikely(abs_offset >= st->upper_offset)) {
3473 if (st->frag_data) {
3474 kunmap_atomic(st->frag_data);
3475 st->frag_data = NULL;
3476 }
3477 return 0;
3478 }
3479
3480next_skb:
3481 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3482
3483 if (abs_offset < block_limit && !st->frag_data) {
3484 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3485 return block_limit - abs_offset;
3486 }
3487
3488 if (st->frag_idx == 0 && !st->frag_data)
3489 st->stepped_offset += skb_headlen(st->cur_skb);
3490
3491 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3492 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3493 block_limit = skb_frag_size(frag) + st->stepped_offset;
3494
3495 if (abs_offset < block_limit) {
3496 if (!st->frag_data)
3497 st->frag_data = kmap_atomic(skb_frag_page(frag));
3498
3499 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3500 (abs_offset - st->stepped_offset);
3501
3502 return block_limit - abs_offset;
3503 }
3504
3505 if (st->frag_data) {
3506 kunmap_atomic(st->frag_data);
3507 st->frag_data = NULL;
3508 }
3509
3510 st->frag_idx++;
3511 st->stepped_offset += skb_frag_size(frag);
3512 }
3513
3514 if (st->frag_data) {
3515 kunmap_atomic(st->frag_data);
3516 st->frag_data = NULL;
3517 }
3518
3519 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3520 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3521 st->frag_idx = 0;
3522 goto next_skb;
3523 } else if (st->cur_skb->next) {
3524 st->cur_skb = st->cur_skb->next;
3525 st->frag_idx = 0;
3526 goto next_skb;
3527 }
3528
3529 return 0;
3530}
3531EXPORT_SYMBOL(skb_seq_read);
3532
3533/**
3534 * skb_abort_seq_read - Abort a sequential read of skb data
3535 * @st: state variable
3536 *
3537 * Must be called if skb_seq_read() was not called until it
3538 * returned 0.
3539 */
3540void skb_abort_seq_read(struct skb_seq_state *st)
3541{
3542 if (st->frag_data)
3543 kunmap_atomic(st->frag_data);
3544}
3545EXPORT_SYMBOL(skb_abort_seq_read);
3546
3547#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3548
3549static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3550 struct ts_config *conf,
3551 struct ts_state *state)
3552{
3553 return skb_seq_read(offset, text, TS_SKB_CB(state));
3554}
3555
3556static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3557{
3558 skb_abort_seq_read(TS_SKB_CB(state));
3559}
3560
3561/**
3562 * skb_find_text - Find a text pattern in skb data
3563 * @skb: the buffer to look in
3564 * @from: search offset
3565 * @to: search limit
3566 * @config: textsearch configuration
3567 *
3568 * Finds a pattern in the skb data according to the specified
3569 * textsearch configuration. Use textsearch_next() to retrieve
3570 * subsequent occurrences of the pattern. Returns the offset
3571 * to the first occurrence or UINT_MAX if no match was found.
3572 */
3573unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3574 unsigned int to, struct ts_config *config)
3575{
3576 struct ts_state state;
3577 unsigned int ret;
3578
3579 config->get_next_block = skb_ts_get_next_block;
3580 config->finish = skb_ts_finish;
3581
3582 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3583
3584 ret = textsearch_find(config, &state);
3585 return (ret <= to - from ? ret : UINT_MAX);
3586}
3587EXPORT_SYMBOL(skb_find_text);
3588
3589int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3590 int offset, size_t size)
3591{
3592 int i = skb_shinfo(skb)->nr_frags;
3593
3594 if (skb_can_coalesce(skb, i, page, offset)) {
3595 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3596 } else if (i < MAX_SKB_FRAGS) {
3597 get_page(page);
3598 skb_fill_page_desc(skb, i, page, offset, size);
3599 } else {
3600 return -EMSGSIZE;
3601 }
3602
3603 return 0;
3604}
3605EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3606
3607/**
3608 * skb_pull_rcsum - pull skb and update receive checksum
3609 * @skb: buffer to update
3610 * @len: length of data pulled
3611 *
3612 * This function performs an skb_pull on the packet and updates
3613 * the CHECKSUM_COMPLETE checksum. It should be used on
3614 * receive path processing instead of skb_pull unless you know
3615 * that the checksum difference is zero (e.g., a valid IP header)
3616 * or you are setting ip_summed to CHECKSUM_NONE.
3617 */
3618void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3619{
3620 unsigned char *data = skb->data;
3621
3622 BUG_ON(len > skb->len);
3623 __skb_pull(skb, len);
3624 skb_postpull_rcsum(skb, data, len);
3625 return skb->data;
3626}
3627EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3628
3629static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3630{
3631 skb_frag_t head_frag;
3632 struct page *page;
3633
3634 page = virt_to_head_page(frag_skb->head);
3635 __skb_frag_set_page(&head_frag, page);
3636 skb_frag_off_set(&head_frag, frag_skb->data -
3637 (unsigned char *)page_address(page));
3638 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3639 return head_frag;
3640}
3641
3642struct sk_buff *skb_segment_list(struct sk_buff *skb,
3643 netdev_features_t features,
3644 unsigned int offset)
3645{
3646 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3647 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3648 unsigned int delta_truesize = 0;
3649 unsigned int delta_len = 0;
3650 struct sk_buff *tail = NULL;
3651 struct sk_buff *nskb;
3652
3653 skb_push(skb, -skb_network_offset(skb) + offset);
3654
3655 skb_shinfo(skb)->frag_list = NULL;
3656
3657 do {
3658 nskb = list_skb;
3659 list_skb = list_skb->next;
3660
3661 if (!tail)
3662 skb->next = nskb;
3663 else
3664 tail->next = nskb;
3665
3666 tail = nskb;
3667
3668 delta_len += nskb->len;
3669 delta_truesize += nskb->truesize;
3670
3671 skb_push(nskb, -skb_network_offset(nskb) + offset);
3672
3673 skb_release_head_state(nskb);
3674 __copy_skb_header(nskb, skb);
3675
3676 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3677 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3678 nskb->data - tnl_hlen,
3679 offset + tnl_hlen);
3680
3681 if (skb_needs_linearize(nskb, features) &&
3682 __skb_linearize(nskb))
3683 goto err_linearize;
3684
3685 } while (list_skb);
3686
3687 skb->truesize = skb->truesize - delta_truesize;
3688 skb->data_len = skb->data_len - delta_len;
3689 skb->len = skb->len - delta_len;
3690
3691 skb_gso_reset(skb);
3692
3693 skb->prev = tail;
3694
3695 if (skb_needs_linearize(skb, features) &&
3696 __skb_linearize(skb))
3697 goto err_linearize;
3698
3699 skb_get(skb);
3700
3701 return skb;
3702
3703err_linearize:
3704 kfree_skb_list(skb->next);
3705 skb->next = NULL;
3706 return ERR_PTR(-ENOMEM);
3707}
3708EXPORT_SYMBOL_GPL(skb_segment_list);
3709
3710int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3711{
3712 if (unlikely(p->len + skb->len >= 65536))
3713 return -E2BIG;
3714
3715 if (NAPI_GRO_CB(p)->last == p)
3716 skb_shinfo(p)->frag_list = skb;
3717 else
3718 NAPI_GRO_CB(p)->last->next = skb;
3719
3720 skb_pull(skb, skb_gro_offset(skb));
3721
3722 NAPI_GRO_CB(p)->last = skb;
3723 NAPI_GRO_CB(p)->count++;
3724 p->data_len += skb->len;
3725 p->truesize += skb->truesize;
3726 p->len += skb->len;
3727
3728 NAPI_GRO_CB(skb)->same_flow = 1;
3729
3730 return 0;
3731}
3732
3733/**
3734 * skb_segment - Perform protocol segmentation on skb.
3735 * @head_skb: buffer to segment
3736 * @features: features for the output path (see dev->features)
3737 *
3738 * This function performs segmentation on the given skb. It returns
3739 * a pointer to the first in a list of new skbs for the segments.
3740 * In case of error it returns ERR_PTR(err).
3741 */
3742struct sk_buff *skb_segment(struct sk_buff *head_skb,
3743 netdev_features_t features)
3744{
3745 struct sk_buff *segs = NULL;
3746 struct sk_buff *tail = NULL;
3747 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3748 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3749 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3750 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3751 struct sk_buff *frag_skb = head_skb;
3752 unsigned int offset = doffset;
3753 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3754 unsigned int partial_segs = 0;
3755 unsigned int headroom;
3756 unsigned int len = head_skb->len;
3757 __be16 proto;
3758 bool csum, sg;
3759 int nfrags = skb_shinfo(head_skb)->nr_frags;
3760 int err = -ENOMEM;
3761 int i = 0;
3762 int pos;
3763
3764 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3765 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3766 /* gso_size is untrusted, and we have a frag_list with a linear
3767 * non head_frag head.
3768 *
3769 * (we assume checking the first list_skb member suffices;
3770 * i.e if either of the list_skb members have non head_frag
3771 * head, then the first one has too).
3772 *
3773 * If head_skb's headlen does not fit requested gso_size, it
3774 * means that the frag_list members do NOT terminate on exact
3775 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3776 * sharing. Therefore we must fallback to copying the frag_list
3777 * skbs; we do so by disabling SG.
3778 */
3779 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3780 features &= ~NETIF_F_SG;
3781 }
3782
3783 __skb_push(head_skb, doffset);
3784 proto = skb_network_protocol(head_skb, NULL);
3785 if (unlikely(!proto))
3786 return ERR_PTR(-EINVAL);
3787
3788 sg = !!(features & NETIF_F_SG);
3789 csum = !!can_checksum_protocol(features, proto);
3790
3791 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3792 if (!(features & NETIF_F_GSO_PARTIAL)) {
3793 struct sk_buff *iter;
3794 unsigned int frag_len;
3795
3796 if (!list_skb ||
3797 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3798 goto normal;
3799
3800 /* If we get here then all the required
3801 * GSO features except frag_list are supported.
3802 * Try to split the SKB to multiple GSO SKBs
3803 * with no frag_list.
3804 * Currently we can do that only when the buffers don't
3805 * have a linear part and all the buffers except
3806 * the last are of the same length.
3807 */
3808 frag_len = list_skb->len;
3809 skb_walk_frags(head_skb, iter) {
3810 if (frag_len != iter->len && iter->next)
3811 goto normal;
3812 if (skb_headlen(iter) && !iter->head_frag)
3813 goto normal;
3814
3815 len -= iter->len;
3816 }
3817
3818 if (len != frag_len)
3819 goto normal;
3820 }
3821
3822 /* GSO partial only requires that we trim off any excess that
3823 * doesn't fit into an MSS sized block, so take care of that
3824 * now.
3825 */
3826 partial_segs = len / mss;
3827 if (partial_segs > 1)
3828 mss *= partial_segs;
3829 else
3830 partial_segs = 0;
3831 }
3832
3833normal:
3834 headroom = skb_headroom(head_skb);
3835 pos = skb_headlen(head_skb);
3836
3837 do {
3838 struct sk_buff *nskb;
3839 skb_frag_t *nskb_frag;
3840 int hsize;
3841 int size;
3842
3843 if (unlikely(mss == GSO_BY_FRAGS)) {
3844 len = list_skb->len;
3845 } else {
3846 len = head_skb->len - offset;
3847 if (len > mss)
3848 len = mss;
3849 }
3850
3851 hsize = skb_headlen(head_skb) - offset;
3852 if (hsize < 0)
3853 hsize = 0;
3854 if (hsize > len || !sg)
3855 hsize = len;
3856
3857 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3858 (skb_headlen(list_skb) == len || sg)) {
3859 BUG_ON(skb_headlen(list_skb) > len);
3860
3861 i = 0;
3862 nfrags = skb_shinfo(list_skb)->nr_frags;
3863 frag = skb_shinfo(list_skb)->frags;
3864 frag_skb = list_skb;
3865 pos += skb_headlen(list_skb);
3866
3867 while (pos < offset + len) {
3868 BUG_ON(i >= nfrags);
3869
3870 size = skb_frag_size(frag);
3871 if (pos + size > offset + len)
3872 break;
3873
3874 i++;
3875 pos += size;
3876 frag++;
3877 }
3878
3879 nskb = skb_clone(list_skb, GFP_ATOMIC);
3880 list_skb = list_skb->next;
3881
3882 if (unlikely(!nskb))
3883 goto err;
3884
3885 if (unlikely(pskb_trim(nskb, len))) {
3886 kfree_skb(nskb);
3887 goto err;
3888 }
3889
3890 hsize = skb_end_offset(nskb);
3891 if (skb_cow_head(nskb, doffset + headroom)) {
3892 kfree_skb(nskb);
3893 goto err;
3894 }
3895
3896 nskb->truesize += skb_end_offset(nskb) - hsize;
3897 skb_release_head_state(nskb);
3898 __skb_push(nskb, doffset);
3899 } else {
3900 nskb = __alloc_skb(hsize + doffset + headroom,
3901 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3902 NUMA_NO_NODE);
3903
3904 if (unlikely(!nskb))
3905 goto err;
3906
3907 skb_reserve(nskb, headroom);
3908 __skb_put(nskb, doffset);
3909 }
3910
3911 if (segs)
3912 tail->next = nskb;
3913 else
3914 segs = nskb;
3915 tail = nskb;
3916
3917 __copy_skb_header(nskb, head_skb);
3918
3919 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3920 skb_reset_mac_len(nskb);
3921
3922 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3923 nskb->data - tnl_hlen,
3924 doffset + tnl_hlen);
3925
3926 if (nskb->len == len + doffset)
3927 goto perform_csum_check;
3928
3929 if (!sg) {
3930 if (!csum) {
3931 if (!nskb->remcsum_offload)
3932 nskb->ip_summed = CHECKSUM_NONE;
3933 SKB_GSO_CB(nskb)->csum =
3934 skb_copy_and_csum_bits(head_skb, offset,
3935 skb_put(nskb,
3936 len),
3937 len, 0);
3938 SKB_GSO_CB(nskb)->csum_start =
3939 skb_headroom(nskb) + doffset;
3940 } else {
3941 skb_copy_bits(head_skb, offset,
3942 skb_put(nskb, len),
3943 len);
3944 }
3945 continue;
3946 }
3947
3948 nskb_frag = skb_shinfo(nskb)->frags;
3949
3950 skb_copy_from_linear_data_offset(head_skb, offset,
3951 skb_put(nskb, hsize), hsize);
3952
3953 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3954 SKBTX_SHARED_FRAG;
3955
3956 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3957 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3958 goto err;
3959
3960 while (pos < offset + len) {
3961 if (i >= nfrags) {
3962 i = 0;
3963 nfrags = skb_shinfo(list_skb)->nr_frags;
3964 frag = skb_shinfo(list_skb)->frags;
3965 frag_skb = list_skb;
3966 if (!skb_headlen(list_skb)) {
3967 BUG_ON(!nfrags);
3968 } else {
3969 BUG_ON(!list_skb->head_frag);
3970
3971 /* to make room for head_frag. */
3972 i--;
3973 frag--;
3974 }
3975 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3976 skb_zerocopy_clone(nskb, frag_skb,
3977 GFP_ATOMIC))
3978 goto err;
3979
3980 list_skb = list_skb->next;
3981 }
3982
3983 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3984 MAX_SKB_FRAGS)) {
3985 net_warn_ratelimited(
3986 "skb_segment: too many frags: %u %u\n",
3987 pos, mss);
3988 err = -EINVAL;
3989 goto err;
3990 }
3991
3992 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3993 __skb_frag_ref(nskb_frag);
3994 size = skb_frag_size(nskb_frag);
3995
3996 if (pos < offset) {
3997 skb_frag_off_add(nskb_frag, offset - pos);
3998 skb_frag_size_sub(nskb_frag, offset - pos);
3999 }
4000
4001 skb_shinfo(nskb)->nr_frags++;
4002
4003 if (pos + size <= offset + len) {
4004 i++;
4005 frag++;
4006 pos += size;
4007 } else {
4008 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4009 goto skip_fraglist;
4010 }
4011
4012 nskb_frag++;
4013 }
4014
4015skip_fraglist:
4016 nskb->data_len = len - hsize;
4017 nskb->len += nskb->data_len;
4018 nskb->truesize += nskb->data_len;
4019
4020perform_csum_check:
4021 if (!csum) {
4022 if (skb_has_shared_frag(nskb) &&
4023 __skb_linearize(nskb))
4024 goto err;
4025
4026 if (!nskb->remcsum_offload)
4027 nskb->ip_summed = CHECKSUM_NONE;
4028 SKB_GSO_CB(nskb)->csum =
4029 skb_checksum(nskb, doffset,
4030 nskb->len - doffset, 0);
4031 SKB_GSO_CB(nskb)->csum_start =
4032 skb_headroom(nskb) + doffset;
4033 }
4034 } while ((offset += len) < head_skb->len);
4035
4036 /* Some callers want to get the end of the list.
4037 * Put it in segs->prev to avoid walking the list.
4038 * (see validate_xmit_skb_list() for example)
4039 */
4040 segs->prev = tail;
4041
4042 if (partial_segs) {
4043 struct sk_buff *iter;
4044 int type = skb_shinfo(head_skb)->gso_type;
4045 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4046
4047 /* Update type to add partial and then remove dodgy if set */
4048 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4049 type &= ~SKB_GSO_DODGY;
4050
4051 /* Update GSO info and prepare to start updating headers on
4052 * our way back down the stack of protocols.
4053 */
4054 for (iter = segs; iter; iter = iter->next) {
4055 skb_shinfo(iter)->gso_size = gso_size;
4056 skb_shinfo(iter)->gso_segs = partial_segs;
4057 skb_shinfo(iter)->gso_type = type;
4058 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4059 }
4060
4061 if (tail->len - doffset <= gso_size)
4062 skb_shinfo(tail)->gso_size = 0;
4063 else if (tail != segs)
4064 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4065 }
4066
4067 /* Following permits correct backpressure, for protocols
4068 * using skb_set_owner_w().
4069 * Idea is to tranfert ownership from head_skb to last segment.
4070 */
4071 if (head_skb->destructor == sock_wfree) {
4072 swap(tail->truesize, head_skb->truesize);
4073 swap(tail->destructor, head_skb->destructor);
4074 swap(tail->sk, head_skb->sk);
4075 }
4076 return segs;
4077
4078err:
4079 kfree_skb_list(segs);
4080 return ERR_PTR(err);
4081}
4082EXPORT_SYMBOL_GPL(skb_segment);
4083
4084int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4085{
4086 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4087 unsigned int offset = skb_gro_offset(skb);
4088 unsigned int headlen = skb_headlen(skb);
4089 unsigned int len = skb_gro_len(skb);
4090 unsigned int delta_truesize;
4091 struct sk_buff *lp;
4092
4093 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4094 return -E2BIG;
4095
4096 lp = NAPI_GRO_CB(p)->last;
4097 pinfo = skb_shinfo(lp);
4098
4099 if (headlen <= offset) {
4100 skb_frag_t *frag;
4101 skb_frag_t *frag2;
4102 int i = skbinfo->nr_frags;
4103 int nr_frags = pinfo->nr_frags + i;
4104
4105 if (nr_frags > MAX_SKB_FRAGS)
4106 goto merge;
4107
4108 offset -= headlen;
4109 pinfo->nr_frags = nr_frags;
4110 skbinfo->nr_frags = 0;
4111
4112 frag = pinfo->frags + nr_frags;
4113 frag2 = skbinfo->frags + i;
4114 do {
4115 *--frag = *--frag2;
4116 } while (--i);
4117
4118 skb_frag_off_add(frag, offset);
4119 skb_frag_size_sub(frag, offset);
4120
4121 /* all fragments truesize : remove (head size + sk_buff) */
4122 delta_truesize = skb->truesize -
4123 SKB_TRUESIZE(skb_end_offset(skb));
4124
4125 skb->truesize -= skb->data_len;
4126 skb->len -= skb->data_len;
4127 skb->data_len = 0;
4128
4129 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4130 goto done;
4131 } else if (skb->head_frag) {
4132 int nr_frags = pinfo->nr_frags;
4133 skb_frag_t *frag = pinfo->frags + nr_frags;
4134 struct page *page = virt_to_head_page(skb->head);
4135 unsigned int first_size = headlen - offset;
4136 unsigned int first_offset;
4137
4138 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4139 goto merge;
4140
4141 first_offset = skb->data -
4142 (unsigned char *)page_address(page) +
4143 offset;
4144
4145 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4146
4147 __skb_frag_set_page(frag, page);
4148 skb_frag_off_set(frag, first_offset);
4149 skb_frag_size_set(frag, first_size);
4150
4151 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4152 /* We dont need to clear skbinfo->nr_frags here */
4153
4154 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4155 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4156 goto done;
4157 }
4158
4159merge:
4160 delta_truesize = skb->truesize;
4161 if (offset > headlen) {
4162 unsigned int eat = offset - headlen;
4163
4164 skb_frag_off_add(&skbinfo->frags[0], eat);
4165 skb_frag_size_sub(&skbinfo->frags[0], eat);
4166 skb->data_len -= eat;
4167 skb->len -= eat;
4168 offset = headlen;
4169 }
4170
4171 __skb_pull(skb, offset);
4172
4173 if (NAPI_GRO_CB(p)->last == p)
4174 skb_shinfo(p)->frag_list = skb;
4175 else
4176 NAPI_GRO_CB(p)->last->next = skb;
4177 NAPI_GRO_CB(p)->last = skb;
4178 __skb_header_release(skb);
4179 lp = p;
4180
4181done:
4182 NAPI_GRO_CB(p)->count++;
4183 p->data_len += len;
4184 p->truesize += delta_truesize;
4185 p->len += len;
4186 if (lp != p) {
4187 lp->data_len += len;
4188 lp->truesize += delta_truesize;
4189 lp->len += len;
4190 }
4191 NAPI_GRO_CB(skb)->same_flow = 1;
4192 return 0;
4193}
4194
4195#ifdef CONFIG_SKB_EXTENSIONS
4196#define SKB_EXT_ALIGN_VALUE 8
4197#define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4198
4199static const u8 skb_ext_type_len[] = {
4200#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4201 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4202#endif
4203#ifdef CONFIG_XFRM
4204 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4205#endif
4206#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4207 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4208#endif
4209#if IS_ENABLED(CONFIG_MPTCP)
4210 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4211#endif
4212};
4213
4214static __always_inline unsigned int skb_ext_total_length(void)
4215{
4216 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4217#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4218 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4219#endif
4220#ifdef CONFIG_XFRM
4221 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4222#endif
4223#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4224 skb_ext_type_len[TC_SKB_EXT] +
4225#endif
4226#if IS_ENABLED(CONFIG_MPTCP)
4227 skb_ext_type_len[SKB_EXT_MPTCP] +
4228#endif
4229 0;
4230}
4231
4232static void skb_extensions_init(void)
4233{
4234 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4235 BUILD_BUG_ON(skb_ext_total_length() > 255);
4236
4237 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4238 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4239 0,
4240 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4241 NULL);
4242}
4243#else
4244static void skb_extensions_init(void) {}
4245#endif
4246
4247void __init skb_init(void)
4248{
4249 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4250 sizeof(struct sk_buff),
4251 0,
4252 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4253 offsetof(struct sk_buff, cb),
4254 sizeof_field(struct sk_buff, cb),
4255 NULL);
4256 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4257 sizeof(struct sk_buff_fclones),
4258 0,
4259 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4260 NULL);
4261 skb_extensions_init();
4262}
4263
4264static int
4265__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4266 unsigned int recursion_level)
4267{
4268 int start = skb_headlen(skb);
4269 int i, copy = start - offset;
4270 struct sk_buff *frag_iter;
4271 int elt = 0;
4272
4273 if (unlikely(recursion_level >= 24))
4274 return -EMSGSIZE;
4275
4276 if (copy > 0) {
4277 if (copy > len)
4278 copy = len;
4279 sg_set_buf(sg, skb->data + offset, copy);
4280 elt++;
4281 if ((len -= copy) == 0)
4282 return elt;
4283 offset += copy;
4284 }
4285
4286 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4287 int end;
4288
4289 WARN_ON(start > offset + len);
4290
4291 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4292 if ((copy = end - offset) > 0) {
4293 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4294 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4295 return -EMSGSIZE;
4296
4297 if (copy > len)
4298 copy = len;
4299 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4300 skb_frag_off(frag) + offset - start);
4301 elt++;
4302 if (!(len -= copy))
4303 return elt;
4304 offset += copy;
4305 }
4306 start = end;
4307 }
4308
4309 skb_walk_frags(skb, frag_iter) {
4310 int end, ret;
4311
4312 WARN_ON(start > offset + len);
4313
4314 end = start + frag_iter->len;
4315 if ((copy = end - offset) > 0) {
4316 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4317 return -EMSGSIZE;
4318
4319 if (copy > len)
4320 copy = len;
4321 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4322 copy, recursion_level + 1);
4323 if (unlikely(ret < 0))
4324 return ret;
4325 elt += ret;
4326 if ((len -= copy) == 0)
4327 return elt;
4328 offset += copy;
4329 }
4330 start = end;
4331 }
4332 BUG_ON(len);
4333 return elt;
4334}
4335
4336/**
4337 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4338 * @skb: Socket buffer containing the buffers to be mapped
4339 * @sg: The scatter-gather list to map into
4340 * @offset: The offset into the buffer's contents to start mapping
4341 * @len: Length of buffer space to be mapped
4342 *
4343 * Fill the specified scatter-gather list with mappings/pointers into a
4344 * region of the buffer space attached to a socket buffer. Returns either
4345 * the number of scatterlist items used, or -EMSGSIZE if the contents
4346 * could not fit.
4347 */
4348int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4349{
4350 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4351
4352 if (nsg <= 0)
4353 return nsg;
4354
4355 sg_mark_end(&sg[nsg - 1]);
4356
4357 return nsg;
4358}
4359EXPORT_SYMBOL_GPL(skb_to_sgvec);
4360
4361/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4362 * sglist without mark the sg which contain last skb data as the end.
4363 * So the caller can mannipulate sg list as will when padding new data after
4364 * the first call without calling sg_unmark_end to expend sg list.
4365 *
4366 * Scenario to use skb_to_sgvec_nomark:
4367 * 1. sg_init_table
4368 * 2. skb_to_sgvec_nomark(payload1)
4369 * 3. skb_to_sgvec_nomark(payload2)
4370 *
4371 * This is equivalent to:
4372 * 1. sg_init_table
4373 * 2. skb_to_sgvec(payload1)
4374 * 3. sg_unmark_end
4375 * 4. skb_to_sgvec(payload2)
4376 *
4377 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4378 * is more preferable.
4379 */
4380int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4381 int offset, int len)
4382{
4383 return __skb_to_sgvec(skb, sg, offset, len, 0);
4384}
4385EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4386
4387
4388
4389/**
4390 * skb_cow_data - Check that a socket buffer's data buffers are writable
4391 * @skb: The socket buffer to check.
4392 * @tailbits: Amount of trailing space to be added
4393 * @trailer: Returned pointer to the skb where the @tailbits space begins
4394 *
4395 * Make sure that the data buffers attached to a socket buffer are
4396 * writable. If they are not, private copies are made of the data buffers
4397 * and the socket buffer is set to use these instead.
4398 *
4399 * If @tailbits is given, make sure that there is space to write @tailbits
4400 * bytes of data beyond current end of socket buffer. @trailer will be
4401 * set to point to the skb in which this space begins.
4402 *
4403 * The number of scatterlist elements required to completely map the
4404 * COW'd and extended socket buffer will be returned.
4405 */
4406int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4407{
4408 int copyflag;
4409 int elt;
4410 struct sk_buff *skb1, **skb_p;
4411
4412 /* If skb is cloned or its head is paged, reallocate
4413 * head pulling out all the pages (pages are considered not writable
4414 * at the moment even if they are anonymous).
4415 */
4416 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4417 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4418 return -ENOMEM;
4419
4420 /* Easy case. Most of packets will go this way. */
4421 if (!skb_has_frag_list(skb)) {
4422 /* A little of trouble, not enough of space for trailer.
4423 * This should not happen, when stack is tuned to generate
4424 * good frames. OK, on miss we reallocate and reserve even more
4425 * space, 128 bytes is fair. */
4426
4427 if (skb_tailroom(skb) < tailbits &&
4428 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4429 return -ENOMEM;
4430
4431 /* Voila! */
4432 *trailer = skb;
4433 return 1;
4434 }
4435
4436 /* Misery. We are in troubles, going to mincer fragments... */
4437
4438 elt = 1;
4439 skb_p = &skb_shinfo(skb)->frag_list;
4440 copyflag = 0;
4441
4442 while ((skb1 = *skb_p) != NULL) {
4443 int ntail = 0;
4444
4445 /* The fragment is partially pulled by someone,
4446 * this can happen on input. Copy it and everything
4447 * after it. */
4448
4449 if (skb_shared(skb1))
4450 copyflag = 1;
4451
4452 /* If the skb is the last, worry about trailer. */
4453
4454 if (skb1->next == NULL && tailbits) {
4455 if (skb_shinfo(skb1)->nr_frags ||
4456 skb_has_frag_list(skb1) ||
4457 skb_tailroom(skb1) < tailbits)
4458 ntail = tailbits + 128;
4459 }
4460
4461 if (copyflag ||
4462 skb_cloned(skb1) ||
4463 ntail ||
4464 skb_shinfo(skb1)->nr_frags ||
4465 skb_has_frag_list(skb1)) {
4466 struct sk_buff *skb2;
4467
4468 /* Fuck, we are miserable poor guys... */
4469 if (ntail == 0)
4470 skb2 = skb_copy(skb1, GFP_ATOMIC);
4471 else
4472 skb2 = skb_copy_expand(skb1,
4473 skb_headroom(skb1),
4474 ntail,
4475 GFP_ATOMIC);
4476 if (unlikely(skb2 == NULL))
4477 return -ENOMEM;
4478
4479 if (skb1->sk)
4480 skb_set_owner_w(skb2, skb1->sk);
4481
4482 /* Looking around. Are we still alive?
4483 * OK, link new skb, drop old one */
4484
4485 skb2->next = skb1->next;
4486 *skb_p = skb2;
4487 kfree_skb(skb1);
4488 skb1 = skb2;
4489 }
4490 elt++;
4491 *trailer = skb1;
4492 skb_p = &skb1->next;
4493 }
4494
4495 return elt;
4496}
4497EXPORT_SYMBOL_GPL(skb_cow_data);
4498
4499static void sock_rmem_free(struct sk_buff *skb)
4500{
4501 struct sock *sk = skb->sk;
4502
4503 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4504}
4505
4506static void skb_set_err_queue(struct sk_buff *skb)
4507{
4508 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4509 * So, it is safe to (mis)use it to mark skbs on the error queue.
4510 */
4511 skb->pkt_type = PACKET_OUTGOING;
4512 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4513}
4514
4515/*
4516 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4517 */
4518int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4519{
4520 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4521 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4522 return -ENOMEM;
4523
4524 skb_orphan(skb);
4525 skb->sk = sk;
4526 skb->destructor = sock_rmem_free;
4527 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4528 skb_set_err_queue(skb);
4529
4530 /* before exiting rcu section, make sure dst is refcounted */
4531 skb_dst_force(skb);
4532
4533 skb_queue_tail(&sk->sk_error_queue, skb);
4534 if (!sock_flag(sk, SOCK_DEAD))
4535 sk->sk_error_report(sk);
4536 return 0;
4537}
4538EXPORT_SYMBOL(sock_queue_err_skb);
4539
4540static bool is_icmp_err_skb(const struct sk_buff *skb)
4541{
4542 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4543 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4544}
4545
4546struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4547{
4548 struct sk_buff_head *q = &sk->sk_error_queue;
4549 struct sk_buff *skb, *skb_next = NULL;
4550 bool icmp_next = false;
4551 unsigned long flags;
4552
4553 spin_lock_irqsave(&q->lock, flags);
4554 skb = __skb_dequeue(q);
4555 if (skb && (skb_next = skb_peek(q))) {
4556 icmp_next = is_icmp_err_skb(skb_next);
4557 if (icmp_next)
4558 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4559 }
4560 spin_unlock_irqrestore(&q->lock, flags);
4561
4562 if (is_icmp_err_skb(skb) && !icmp_next)
4563 sk->sk_err = 0;
4564
4565 if (skb_next)
4566 sk->sk_error_report(sk);
4567
4568 return skb;
4569}
4570EXPORT_SYMBOL(sock_dequeue_err_skb);
4571
4572/**
4573 * skb_clone_sk - create clone of skb, and take reference to socket
4574 * @skb: the skb to clone
4575 *
4576 * This function creates a clone of a buffer that holds a reference on
4577 * sk_refcnt. Buffers created via this function are meant to be
4578 * returned using sock_queue_err_skb, or free via kfree_skb.
4579 *
4580 * When passing buffers allocated with this function to sock_queue_err_skb
4581 * it is necessary to wrap the call with sock_hold/sock_put in order to
4582 * prevent the socket from being released prior to being enqueued on
4583 * the sk_error_queue.
4584 */
4585struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4586{
4587 struct sock *sk = skb->sk;
4588 struct sk_buff *clone;
4589
4590 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4591 return NULL;
4592
4593 clone = skb_clone(skb, GFP_ATOMIC);
4594 if (!clone) {
4595 sock_put(sk);
4596 return NULL;
4597 }
4598
4599 clone->sk = sk;
4600 clone->destructor = sock_efree;
4601
4602 return clone;
4603}
4604EXPORT_SYMBOL(skb_clone_sk);
4605
4606static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4607 struct sock *sk,
4608 int tstype,
4609 bool opt_stats)
4610{
4611 struct sock_exterr_skb *serr;
4612 int err;
4613
4614 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4615
4616 serr = SKB_EXT_ERR(skb);
4617 memset(serr, 0, sizeof(*serr));
4618 serr->ee.ee_errno = ENOMSG;
4619 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4620 serr->ee.ee_info = tstype;
4621 serr->opt_stats = opt_stats;
4622 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4623 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4624 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4625 if (sk->sk_protocol == IPPROTO_TCP &&
4626 sk->sk_type == SOCK_STREAM)
4627 serr->ee.ee_data -= sk->sk_tskey;
4628 }
4629
4630 err = sock_queue_err_skb(sk, skb);
4631
4632 if (err)
4633 kfree_skb(skb);
4634}
4635
4636static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4637{
4638 bool ret;
4639
4640 if (likely(sysctl_tstamp_allow_data || tsonly))
4641 return true;
4642
4643 read_lock_bh(&sk->sk_callback_lock);
4644 ret = sk->sk_socket && sk->sk_socket->file &&
4645 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4646 read_unlock_bh(&sk->sk_callback_lock);
4647 return ret;
4648}
4649
4650void skb_complete_tx_timestamp(struct sk_buff *skb,
4651 struct skb_shared_hwtstamps *hwtstamps)
4652{
4653 struct sock *sk = skb->sk;
4654
4655 if (!skb_may_tx_timestamp(sk, false))
4656 goto err;
4657
4658 /* Take a reference to prevent skb_orphan() from freeing the socket,
4659 * but only if the socket refcount is not zero.
4660 */
4661 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4662 *skb_hwtstamps(skb) = *hwtstamps;
4663 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4664 sock_put(sk);
4665 return;
4666 }
4667
4668err:
4669 kfree_skb(skb);
4670}
4671EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4672
4673void __skb_tstamp_tx(struct sk_buff *orig_skb,
4674 struct skb_shared_hwtstamps *hwtstamps,
4675 struct sock *sk, int tstype)
4676{
4677 struct sk_buff *skb;
4678 bool tsonly, opt_stats = false;
4679
4680 if (!sk)
4681 return;
4682
4683 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4684 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4685 return;
4686
4687 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4688 if (!skb_may_tx_timestamp(sk, tsonly))
4689 return;
4690
4691 if (tsonly) {
4692#ifdef CONFIG_INET
4693 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4694 sk->sk_protocol == IPPROTO_TCP &&
4695 sk->sk_type == SOCK_STREAM) {
4696 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4697 opt_stats = true;
4698 } else
4699#endif
4700 skb = alloc_skb(0, GFP_ATOMIC);
4701 } else {
4702 skb = skb_clone(orig_skb, GFP_ATOMIC);
4703 }
4704 if (!skb)
4705 return;
4706
4707 if (tsonly) {
4708 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4709 SKBTX_ANY_TSTAMP;
4710 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4711 }
4712
4713 if (hwtstamps)
4714 *skb_hwtstamps(skb) = *hwtstamps;
4715 else
4716 skb->tstamp = ktime_get_real();
4717
4718 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4719}
4720EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4721
4722void skb_tstamp_tx(struct sk_buff *orig_skb,
4723 struct skb_shared_hwtstamps *hwtstamps)
4724{
4725 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4726 SCM_TSTAMP_SND);
4727}
4728EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4729
4730void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4731{
4732 struct sock *sk = skb->sk;
4733 struct sock_exterr_skb *serr;
4734 int err = 1;
4735
4736 skb->wifi_acked_valid = 1;
4737 skb->wifi_acked = acked;
4738
4739 serr = SKB_EXT_ERR(skb);
4740 memset(serr, 0, sizeof(*serr));
4741 serr->ee.ee_errno = ENOMSG;
4742 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4743
4744 /* Take a reference to prevent skb_orphan() from freeing the socket,
4745 * but only if the socket refcount is not zero.
4746 */
4747 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4748 err = sock_queue_err_skb(sk, skb);
4749 sock_put(sk);
4750 }
4751 if (err)
4752 kfree_skb(skb);
4753}
4754EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4755
4756/**
4757 * skb_partial_csum_set - set up and verify partial csum values for packet
4758 * @skb: the skb to set
4759 * @start: the number of bytes after skb->data to start checksumming.
4760 * @off: the offset from start to place the checksum.
4761 *
4762 * For untrusted partially-checksummed packets, we need to make sure the values
4763 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4764 *
4765 * This function checks and sets those values and skb->ip_summed: if this
4766 * returns false you should drop the packet.
4767 */
4768bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4769{
4770 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4771 u32 csum_start = skb_headroom(skb) + (u32)start;
4772
4773 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4774 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4775 start, off, skb_headroom(skb), skb_headlen(skb));
4776 return false;
4777 }
4778 skb->ip_summed = CHECKSUM_PARTIAL;
4779 skb->csum_start = csum_start;
4780 skb->csum_offset = off;
4781 skb_set_transport_header(skb, start);
4782 return true;
4783}
4784EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4785
4786static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4787 unsigned int max)
4788{
4789 if (skb_headlen(skb) >= len)
4790 return 0;
4791
4792 /* If we need to pullup then pullup to the max, so we
4793 * won't need to do it again.
4794 */
4795 if (max > skb->len)
4796 max = skb->len;
4797
4798 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4799 return -ENOMEM;
4800
4801 if (skb_headlen(skb) < len)
4802 return -EPROTO;
4803
4804 return 0;
4805}
4806
4807#define MAX_TCP_HDR_LEN (15 * 4)
4808
4809static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4810 typeof(IPPROTO_IP) proto,
4811 unsigned int off)
4812{
4813 int err;
4814
4815 switch (proto) {
4816 case IPPROTO_TCP:
4817 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4818 off + MAX_TCP_HDR_LEN);
4819 if (!err && !skb_partial_csum_set(skb, off,
4820 offsetof(struct tcphdr,
4821 check)))
4822 err = -EPROTO;
4823 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4824
4825 case IPPROTO_UDP:
4826 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4827 off + sizeof(struct udphdr));
4828 if (!err && !skb_partial_csum_set(skb, off,
4829 offsetof(struct udphdr,
4830 check)))
4831 err = -EPROTO;
4832 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4833 }
4834
4835 return ERR_PTR(-EPROTO);
4836}
4837
4838/* This value should be large enough to cover a tagged ethernet header plus
4839 * maximally sized IP and TCP or UDP headers.
4840 */
4841#define MAX_IP_HDR_LEN 128
4842
4843static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4844{
4845 unsigned int off;
4846 bool fragment;
4847 __sum16 *csum;
4848 int err;
4849
4850 fragment = false;
4851
4852 err = skb_maybe_pull_tail(skb,
4853 sizeof(struct iphdr),
4854 MAX_IP_HDR_LEN);
4855 if (err < 0)
4856 goto out;
4857
4858 if (ip_is_fragment(ip_hdr(skb)))
4859 fragment = true;
4860
4861 off = ip_hdrlen(skb);
4862
4863 err = -EPROTO;
4864
4865 if (fragment)
4866 goto out;
4867
4868 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4869 if (IS_ERR(csum))
4870 return PTR_ERR(csum);
4871
4872 if (recalculate)
4873 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4874 ip_hdr(skb)->daddr,
4875 skb->len - off,
4876 ip_hdr(skb)->protocol, 0);
4877 err = 0;
4878
4879out:
4880 return err;
4881}
4882
4883/* This value should be large enough to cover a tagged ethernet header plus
4884 * an IPv6 header, all options, and a maximal TCP or UDP header.
4885 */
4886#define MAX_IPV6_HDR_LEN 256
4887
4888#define OPT_HDR(type, skb, off) \
4889 (type *)(skb_network_header(skb) + (off))
4890
4891static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4892{
4893 int err;
4894 u8 nexthdr;
4895 unsigned int off;
4896 unsigned int len;
4897 bool fragment;
4898 bool done;
4899 __sum16 *csum;
4900
4901 fragment = false;
4902 done = false;
4903
4904 off = sizeof(struct ipv6hdr);
4905
4906 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4907 if (err < 0)
4908 goto out;
4909
4910 nexthdr = ipv6_hdr(skb)->nexthdr;
4911
4912 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4913 while (off <= len && !done) {
4914 switch (nexthdr) {
4915 case IPPROTO_DSTOPTS:
4916 case IPPROTO_HOPOPTS:
4917 case IPPROTO_ROUTING: {
4918 struct ipv6_opt_hdr *hp;
4919
4920 err = skb_maybe_pull_tail(skb,
4921 off +
4922 sizeof(struct ipv6_opt_hdr),
4923 MAX_IPV6_HDR_LEN);
4924 if (err < 0)
4925 goto out;
4926
4927 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4928 nexthdr = hp->nexthdr;
4929 off += ipv6_optlen(hp);
4930 break;
4931 }
4932 case IPPROTO_AH: {
4933 struct ip_auth_hdr *hp;
4934
4935 err = skb_maybe_pull_tail(skb,
4936 off +
4937 sizeof(struct ip_auth_hdr),
4938 MAX_IPV6_HDR_LEN);
4939 if (err < 0)
4940 goto out;
4941
4942 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4943 nexthdr = hp->nexthdr;
4944 off += ipv6_authlen(hp);
4945 break;
4946 }
4947 case IPPROTO_FRAGMENT: {
4948 struct frag_hdr *hp;
4949
4950 err = skb_maybe_pull_tail(skb,
4951 off +
4952 sizeof(struct frag_hdr),
4953 MAX_IPV6_HDR_LEN);
4954 if (err < 0)
4955 goto out;
4956
4957 hp = OPT_HDR(struct frag_hdr, skb, off);
4958
4959 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4960 fragment = true;
4961
4962 nexthdr = hp->nexthdr;
4963 off += sizeof(struct frag_hdr);
4964 break;
4965 }
4966 default:
4967 done = true;
4968 break;
4969 }
4970 }
4971
4972 err = -EPROTO;
4973
4974 if (!done || fragment)
4975 goto out;
4976
4977 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4978 if (IS_ERR(csum))
4979 return PTR_ERR(csum);
4980
4981 if (recalculate)
4982 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4983 &ipv6_hdr(skb)->daddr,
4984 skb->len - off, nexthdr, 0);
4985 err = 0;
4986
4987out:
4988 return err;
4989}
4990
4991/**
4992 * skb_checksum_setup - set up partial checksum offset
4993 * @skb: the skb to set up
4994 * @recalculate: if true the pseudo-header checksum will be recalculated
4995 */
4996int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4997{
4998 int err;
4999
5000 switch (skb->protocol) {
5001 case htons(ETH_P_IP):
5002 err = skb_checksum_setup_ipv4(skb, recalculate);
5003 break;
5004
5005 case htons(ETH_P_IPV6):
5006 err = skb_checksum_setup_ipv6(skb, recalculate);
5007 break;
5008
5009 default:
5010 err = -EPROTO;
5011 break;
5012 }
5013
5014 return err;
5015}
5016EXPORT_SYMBOL(skb_checksum_setup);
5017
5018/**
5019 * skb_checksum_maybe_trim - maybe trims the given skb
5020 * @skb: the skb to check
5021 * @transport_len: the data length beyond the network header
5022 *
5023 * Checks whether the given skb has data beyond the given transport length.
5024 * If so, returns a cloned skb trimmed to this transport length.
5025 * Otherwise returns the provided skb. Returns NULL in error cases
5026 * (e.g. transport_len exceeds skb length or out-of-memory).
5027 *
5028 * Caller needs to set the skb transport header and free any returned skb if it
5029 * differs from the provided skb.
5030 */
5031static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5032 unsigned int transport_len)
5033{
5034 struct sk_buff *skb_chk;
5035 unsigned int len = skb_transport_offset(skb) + transport_len;
5036 int ret;
5037
5038 if (skb->len < len)
5039 return NULL;
5040 else if (skb->len == len)
5041 return skb;
5042
5043 skb_chk = skb_clone(skb, GFP_ATOMIC);
5044 if (!skb_chk)
5045 return NULL;
5046
5047 ret = pskb_trim_rcsum(skb_chk, len);
5048 if (ret) {
5049 kfree_skb(skb_chk);
5050 return NULL;
5051 }
5052
5053 return skb_chk;
5054}
5055
5056/**
5057 * skb_checksum_trimmed - validate checksum of an skb
5058 * @skb: the skb to check
5059 * @transport_len: the data length beyond the network header
5060 * @skb_chkf: checksum function to use
5061 *
5062 * Applies the given checksum function skb_chkf to the provided skb.
5063 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5064 *
5065 * If the skb has data beyond the given transport length, then a
5066 * trimmed & cloned skb is checked and returned.
5067 *
5068 * Caller needs to set the skb transport header and free any returned skb if it
5069 * differs from the provided skb.
5070 */
5071struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5072 unsigned int transport_len,
5073 __sum16(*skb_chkf)(struct sk_buff *skb))
5074{
5075 struct sk_buff *skb_chk;
5076 unsigned int offset = skb_transport_offset(skb);
5077 __sum16 ret;
5078
5079 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5080 if (!skb_chk)
5081 goto err;
5082
5083 if (!pskb_may_pull(skb_chk, offset))
5084 goto err;
5085
5086 skb_pull_rcsum(skb_chk, offset);
5087 ret = skb_chkf(skb_chk);
5088 skb_push_rcsum(skb_chk, offset);
5089
5090 if (ret)
5091 goto err;
5092
5093 return skb_chk;
5094
5095err:
5096 if (skb_chk && skb_chk != skb)
5097 kfree_skb(skb_chk);
5098
5099 return NULL;
5100
5101}
5102EXPORT_SYMBOL(skb_checksum_trimmed);
5103
5104void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5105{
5106 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5107 skb->dev->name);
5108}
5109EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5110
5111void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5112{
5113 if (head_stolen) {
5114 skb_release_head_state(skb);
5115 kmem_cache_free(skbuff_head_cache, skb);
5116 } else {
5117 __kfree_skb(skb);
5118 }
5119}
5120EXPORT_SYMBOL(kfree_skb_partial);
5121
5122/**
5123 * skb_try_coalesce - try to merge skb to prior one
5124 * @to: prior buffer
5125 * @from: buffer to add
5126 * @fragstolen: pointer to boolean
5127 * @delta_truesize: how much more was allocated than was requested
5128 */
5129bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5130 bool *fragstolen, int *delta_truesize)
5131{
5132 struct skb_shared_info *to_shinfo, *from_shinfo;
5133 int i, delta, len = from->len;
5134
5135 *fragstolen = false;
5136
5137 if (skb_cloned(to))
5138 return false;
5139
5140 if (len <= skb_tailroom(to)) {
5141 if (len)
5142 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5143 *delta_truesize = 0;
5144 return true;
5145 }
5146
5147 to_shinfo = skb_shinfo(to);
5148 from_shinfo = skb_shinfo(from);
5149 if (to_shinfo->frag_list || from_shinfo->frag_list)
5150 return false;
5151 if (skb_zcopy(to) || skb_zcopy(from))
5152 return false;
5153
5154 if (skb_headlen(from) != 0) {
5155 struct page *page;
5156 unsigned int offset;
5157
5158 if (to_shinfo->nr_frags +
5159 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5160 return false;
5161
5162 if (skb_head_is_locked(from))
5163 return false;
5164
5165 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5166
5167 page = virt_to_head_page(from->head);
5168 offset = from->data - (unsigned char *)page_address(page);
5169
5170 skb_fill_page_desc(to, to_shinfo->nr_frags,
5171 page, offset, skb_headlen(from));
5172 *fragstolen = true;
5173 } else {
5174 if (to_shinfo->nr_frags +
5175 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5176 return false;
5177
5178 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5179 }
5180
5181 WARN_ON_ONCE(delta < len);
5182
5183 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5184 from_shinfo->frags,
5185 from_shinfo->nr_frags * sizeof(skb_frag_t));
5186 to_shinfo->nr_frags += from_shinfo->nr_frags;
5187
5188 if (!skb_cloned(from))
5189 from_shinfo->nr_frags = 0;
5190
5191 /* if the skb is not cloned this does nothing
5192 * since we set nr_frags to 0.
5193 */
5194 for (i = 0; i < from_shinfo->nr_frags; i++)
5195 __skb_frag_ref(&from_shinfo->frags[i]);
5196
5197 to->truesize += delta;
5198 to->len += len;
5199 to->data_len += len;
5200
5201 *delta_truesize = delta;
5202 return true;
5203}
5204EXPORT_SYMBOL(skb_try_coalesce);
5205
5206/**
5207 * skb_scrub_packet - scrub an skb
5208 *
5209 * @skb: buffer to clean
5210 * @xnet: packet is crossing netns
5211 *
5212 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5213 * into/from a tunnel. Some information have to be cleared during these
5214 * operations.
5215 * skb_scrub_packet can also be used to clean a skb before injecting it in
5216 * another namespace (@xnet == true). We have to clear all information in the
5217 * skb that could impact namespace isolation.
5218 */
5219void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5220{
5221 skb->pkt_type = PACKET_HOST;
5222 skb->skb_iif = 0;
5223 skb->ignore_df = 0;
5224 skb_dst_drop(skb);
5225 skb_ext_reset(skb);
5226 nf_reset_ct(skb);
5227 nf_reset_trace(skb);
5228
5229#ifdef CONFIG_NET_SWITCHDEV
5230 skb->offload_fwd_mark = 0;
5231 skb->offload_l3_fwd_mark = 0;
5232#endif
5233
5234 if (!xnet)
5235 return;
5236
5237 ipvs_reset(skb);
5238 skb->mark = 0;
5239 skb->tstamp = 0;
5240}
5241EXPORT_SYMBOL_GPL(skb_scrub_packet);
5242
5243/**
5244 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5245 *
5246 * @skb: GSO skb
5247 *
5248 * skb_gso_transport_seglen is used to determine the real size of the
5249 * individual segments, including Layer4 headers (TCP/UDP).
5250 *
5251 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5252 */
5253static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5254{
5255 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5256 unsigned int thlen = 0;
5257
5258 if (skb->encapsulation) {
5259 thlen = skb_inner_transport_header(skb) -
5260 skb_transport_header(skb);
5261
5262 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5263 thlen += inner_tcp_hdrlen(skb);
5264 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5265 thlen = tcp_hdrlen(skb);
5266 } else if (unlikely(skb_is_gso_sctp(skb))) {
5267 thlen = sizeof(struct sctphdr);
5268 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5269 thlen = sizeof(struct udphdr);
5270 }
5271 /* UFO sets gso_size to the size of the fragmentation
5272 * payload, i.e. the size of the L4 (UDP) header is already
5273 * accounted for.
5274 */
5275 return thlen + shinfo->gso_size;
5276}
5277
5278/**
5279 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5280 *
5281 * @skb: GSO skb
5282 *
5283 * skb_gso_network_seglen is used to determine the real size of the
5284 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5285 *
5286 * The MAC/L2 header is not accounted for.
5287 */
5288static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5289{
5290 unsigned int hdr_len = skb_transport_header(skb) -
5291 skb_network_header(skb);
5292
5293 return hdr_len + skb_gso_transport_seglen(skb);
5294}
5295
5296/**
5297 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5298 *
5299 * @skb: GSO skb
5300 *
5301 * skb_gso_mac_seglen is used to determine the real size of the
5302 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5303 * headers (TCP/UDP).
5304 */
5305static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5306{
5307 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5308
5309 return hdr_len + skb_gso_transport_seglen(skb);
5310}
5311
5312/**
5313 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5314 *
5315 * There are a couple of instances where we have a GSO skb, and we
5316 * want to determine what size it would be after it is segmented.
5317 *
5318 * We might want to check:
5319 * - L3+L4+payload size (e.g. IP forwarding)
5320 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5321 *
5322 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5323 *
5324 * @skb: GSO skb
5325 *
5326 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5327 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5328 *
5329 * @max_len: The maximum permissible length.
5330 *
5331 * Returns true if the segmented length <= max length.
5332 */
5333static inline bool skb_gso_size_check(const struct sk_buff *skb,
5334 unsigned int seg_len,
5335 unsigned int max_len) {
5336 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5337 const struct sk_buff *iter;
5338
5339 if (shinfo->gso_size != GSO_BY_FRAGS)
5340 return seg_len <= max_len;
5341
5342 /* Undo this so we can re-use header sizes */
5343 seg_len -= GSO_BY_FRAGS;
5344
5345 skb_walk_frags(skb, iter) {
5346 if (seg_len + skb_headlen(iter) > max_len)
5347 return false;
5348 }
5349
5350 return true;
5351}
5352
5353/**
5354 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5355 *
5356 * @skb: GSO skb
5357 * @mtu: MTU to validate against
5358 *
5359 * skb_gso_validate_network_len validates if a given skb will fit a
5360 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5361 * payload.
5362 */
5363bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5364{
5365 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5366}
5367EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5368
5369/**
5370 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5371 *
5372 * @skb: GSO skb
5373 * @len: length to validate against
5374 *
5375 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5376 * length once split, including L2, L3 and L4 headers and the payload.
5377 */
5378bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5379{
5380 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5381}
5382EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5383
5384static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5385{
5386 int mac_len, meta_len;
5387 void *meta;
5388
5389 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5390 kfree_skb(skb);
5391 return NULL;
5392 }
5393
5394 mac_len = skb->data - skb_mac_header(skb);
5395 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5396 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5397 mac_len - VLAN_HLEN - ETH_TLEN);
5398 }
5399
5400 meta_len = skb_metadata_len(skb);
5401 if (meta_len) {
5402 meta = skb_metadata_end(skb) - meta_len;
5403 memmove(meta + VLAN_HLEN, meta, meta_len);
5404 }
5405
5406 skb->mac_header += VLAN_HLEN;
5407 return skb;
5408}
5409
5410struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5411{
5412 struct vlan_hdr *vhdr;
5413 u16 vlan_tci;
5414
5415 if (unlikely(skb_vlan_tag_present(skb))) {
5416 /* vlan_tci is already set-up so leave this for another time */
5417 return skb;
5418 }
5419
5420 skb = skb_share_check(skb, GFP_ATOMIC);
5421 if (unlikely(!skb))
5422 goto err_free;
5423 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5424 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5425 goto err_free;
5426
5427 vhdr = (struct vlan_hdr *)skb->data;
5428 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5429 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5430
5431 skb_pull_rcsum(skb, VLAN_HLEN);
5432 vlan_set_encap_proto(skb, vhdr);
5433
5434 skb = skb_reorder_vlan_header(skb);
5435 if (unlikely(!skb))
5436 goto err_free;
5437
5438 skb_reset_network_header(skb);
5439 skb_reset_transport_header(skb);
5440 skb_reset_mac_len(skb);
5441
5442 return skb;
5443
5444err_free:
5445 kfree_skb(skb);
5446 return NULL;
5447}
5448EXPORT_SYMBOL(skb_vlan_untag);
5449
5450int skb_ensure_writable(struct sk_buff *skb, int write_len)
5451{
5452 if (!pskb_may_pull(skb, write_len))
5453 return -ENOMEM;
5454
5455 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5456 return 0;
5457
5458 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5459}
5460EXPORT_SYMBOL(skb_ensure_writable);
5461
5462/* remove VLAN header from packet and update csum accordingly.
5463 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5464 */
5465int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5466{
5467 struct vlan_hdr *vhdr;
5468 int offset = skb->data - skb_mac_header(skb);
5469 int err;
5470
5471 if (WARN_ONCE(offset,
5472 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5473 offset)) {
5474 return -EINVAL;
5475 }
5476
5477 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5478 if (unlikely(err))
5479 return err;
5480
5481 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5482
5483 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5484 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5485
5486 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5487 __skb_pull(skb, VLAN_HLEN);
5488
5489 vlan_set_encap_proto(skb, vhdr);
5490 skb->mac_header += VLAN_HLEN;
5491
5492 if (skb_network_offset(skb) < ETH_HLEN)
5493 skb_set_network_header(skb, ETH_HLEN);
5494
5495 skb_reset_mac_len(skb);
5496
5497 return err;
5498}
5499EXPORT_SYMBOL(__skb_vlan_pop);
5500
5501/* Pop a vlan tag either from hwaccel or from payload.
5502 * Expects skb->data at mac header.
5503 */
5504int skb_vlan_pop(struct sk_buff *skb)
5505{
5506 u16 vlan_tci;
5507 __be16 vlan_proto;
5508 int err;
5509
5510 if (likely(skb_vlan_tag_present(skb))) {
5511 __vlan_hwaccel_clear_tag(skb);
5512 } else {
5513 if (unlikely(!eth_type_vlan(skb->protocol)))
5514 return 0;
5515
5516 err = __skb_vlan_pop(skb, &vlan_tci);
5517 if (err)
5518 return err;
5519 }
5520 /* move next vlan tag to hw accel tag */
5521 if (likely(!eth_type_vlan(skb->protocol)))
5522 return 0;
5523
5524 vlan_proto = skb->protocol;
5525 err = __skb_vlan_pop(skb, &vlan_tci);
5526 if (unlikely(err))
5527 return err;
5528
5529 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5530 return 0;
5531}
5532EXPORT_SYMBOL(skb_vlan_pop);
5533
5534/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5535 * Expects skb->data at mac header.
5536 */
5537int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5538{
5539 if (skb_vlan_tag_present(skb)) {
5540 int offset = skb->data - skb_mac_header(skb);
5541 int err;
5542
5543 if (WARN_ONCE(offset,
5544 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5545 offset)) {
5546 return -EINVAL;
5547 }
5548
5549 err = __vlan_insert_tag(skb, skb->vlan_proto,
5550 skb_vlan_tag_get(skb));
5551 if (err)
5552 return err;
5553
5554 skb->protocol = skb->vlan_proto;
5555 skb->mac_len += VLAN_HLEN;
5556
5557 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5558 }
5559 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5560 return 0;
5561}
5562EXPORT_SYMBOL(skb_vlan_push);
5563
5564/* Update the ethertype of hdr and the skb csum value if required. */
5565static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5566 __be16 ethertype)
5567{
5568 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5569 __be16 diff[] = { ~hdr->h_proto, ethertype };
5570
5571 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5572 }
5573
5574 hdr->h_proto = ethertype;
5575}
5576
5577/**
5578 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5579 * the packet
5580 *
5581 * @skb: buffer
5582 * @mpls_lse: MPLS label stack entry to push
5583 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5584 * @mac_len: length of the MAC header
5585 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5586 * ethernet
5587 *
5588 * Expects skb->data at mac header.
5589 *
5590 * Returns 0 on success, -errno otherwise.
5591 */
5592int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5593 int mac_len, bool ethernet)
5594{
5595 struct mpls_shim_hdr *lse;
5596 int err;
5597
5598 if (unlikely(!eth_p_mpls(mpls_proto)))
5599 return -EINVAL;
5600
5601 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5602 if (skb->encapsulation)
5603 return -EINVAL;
5604
5605 err = skb_cow_head(skb, MPLS_HLEN);
5606 if (unlikely(err))
5607 return err;
5608
5609 if (!skb->inner_protocol) {
5610 skb_set_inner_network_header(skb, skb_network_offset(skb));
5611 skb_set_inner_protocol(skb, skb->protocol);
5612 }
5613
5614 skb_push(skb, MPLS_HLEN);
5615 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5616 mac_len);
5617 skb_reset_mac_header(skb);
5618 skb_set_network_header(skb, mac_len);
5619 skb_reset_mac_len(skb);
5620
5621 lse = mpls_hdr(skb);
5622 lse->label_stack_entry = mpls_lse;
5623 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5624
5625 if (ethernet && mac_len >= ETH_HLEN)
5626 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5627 skb->protocol = mpls_proto;
5628
5629 return 0;
5630}
5631EXPORT_SYMBOL_GPL(skb_mpls_push);
5632
5633/**
5634 * skb_mpls_pop() - pop the outermost MPLS header
5635 *
5636 * @skb: buffer
5637 * @next_proto: ethertype of header after popped MPLS header
5638 * @mac_len: length of the MAC header
5639 * @ethernet: flag to indicate if the packet is ethernet
5640 *
5641 * Expects skb->data at mac header.
5642 *
5643 * Returns 0 on success, -errno otherwise.
5644 */
5645int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5646 bool ethernet)
5647{
5648 int err;
5649
5650 if (unlikely(!eth_p_mpls(skb->protocol)))
5651 return 0;
5652
5653 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5654 if (unlikely(err))
5655 return err;
5656
5657 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5658 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5659 mac_len);
5660
5661 __skb_pull(skb, MPLS_HLEN);
5662 skb_reset_mac_header(skb);
5663 skb_set_network_header(skb, mac_len);
5664
5665 if (ethernet && mac_len >= ETH_HLEN) {
5666 struct ethhdr *hdr;
5667
5668 /* use mpls_hdr() to get ethertype to account for VLANs. */
5669 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5670 skb_mod_eth_type(skb, hdr, next_proto);
5671 }
5672 skb->protocol = next_proto;
5673
5674 return 0;
5675}
5676EXPORT_SYMBOL_GPL(skb_mpls_pop);
5677
5678/**
5679 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5680 *
5681 * @skb: buffer
5682 * @mpls_lse: new MPLS label stack entry to update to
5683 *
5684 * Expects skb->data at mac header.
5685 *
5686 * Returns 0 on success, -errno otherwise.
5687 */
5688int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5689{
5690 int err;
5691
5692 if (unlikely(!eth_p_mpls(skb->protocol)))
5693 return -EINVAL;
5694
5695 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5696 if (unlikely(err))
5697 return err;
5698
5699 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5700 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5701
5702 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5703 }
5704
5705 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5706
5707 return 0;
5708}
5709EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5710
5711/**
5712 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5713 *
5714 * @skb: buffer
5715 *
5716 * Expects skb->data at mac header.
5717 *
5718 * Returns 0 on success, -errno otherwise.
5719 */
5720int skb_mpls_dec_ttl(struct sk_buff *skb)
5721{
5722 u32 lse;
5723 u8 ttl;
5724
5725 if (unlikely(!eth_p_mpls(skb->protocol)))
5726 return -EINVAL;
5727
5728 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5729 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5730 if (!--ttl)
5731 return -EINVAL;
5732
5733 lse &= ~MPLS_LS_TTL_MASK;
5734 lse |= ttl << MPLS_LS_TTL_SHIFT;
5735
5736 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5737}
5738EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5739
5740/**
5741 * alloc_skb_with_frags - allocate skb with page frags
5742 *
5743 * @header_len: size of linear part
5744 * @data_len: needed length in frags
5745 * @max_page_order: max page order desired.
5746 * @errcode: pointer to error code if any
5747 * @gfp_mask: allocation mask
5748 *
5749 * This can be used to allocate a paged skb, given a maximal order for frags.
5750 */
5751struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5752 unsigned long data_len,
5753 int max_page_order,
5754 int *errcode,
5755 gfp_t gfp_mask)
5756{
5757 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5758 unsigned long chunk;
5759 struct sk_buff *skb;
5760 struct page *page;
5761 int i;
5762
5763 *errcode = -EMSGSIZE;
5764 /* Note this test could be relaxed, if we succeed to allocate
5765 * high order pages...
5766 */
5767 if (npages > MAX_SKB_FRAGS)
5768 return NULL;
5769
5770 *errcode = -ENOBUFS;
5771 skb = alloc_skb(header_len, gfp_mask);
5772 if (!skb)
5773 return NULL;
5774
5775 skb->truesize += npages << PAGE_SHIFT;
5776
5777 for (i = 0; npages > 0; i++) {
5778 int order = max_page_order;
5779
5780 while (order) {
5781 if (npages >= 1 << order) {
5782 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5783 __GFP_COMP |
5784 __GFP_NOWARN,
5785 order);
5786 if (page)
5787 goto fill_page;
5788 /* Do not retry other high order allocations */
5789 order = 1;
5790 max_page_order = 0;
5791 }
5792 order--;
5793 }
5794 page = alloc_page(gfp_mask);
5795 if (!page)
5796 goto failure;
5797fill_page:
5798 chunk = min_t(unsigned long, data_len,
5799 PAGE_SIZE << order);
5800 skb_fill_page_desc(skb, i, page, 0, chunk);
5801 data_len -= chunk;
5802 npages -= 1 << order;
5803 }
5804 return skb;
5805
5806failure:
5807 kfree_skb(skb);
5808 return NULL;
5809}
5810EXPORT_SYMBOL(alloc_skb_with_frags);
5811
5812/* carve out the first off bytes from skb when off < headlen */
5813static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5814 const int headlen, gfp_t gfp_mask)
5815{
5816 int i;
5817 int size = skb_end_offset(skb);
5818 int new_hlen = headlen - off;
5819 u8 *data;
5820
5821 size = SKB_DATA_ALIGN(size);
5822
5823 if (skb_pfmemalloc(skb))
5824 gfp_mask |= __GFP_MEMALLOC;
5825 data = kmalloc_reserve(size +
5826 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5827 gfp_mask, NUMA_NO_NODE, NULL);
5828 if (!data)
5829 return -ENOMEM;
5830
5831 size = SKB_WITH_OVERHEAD(ksize(data));
5832
5833 /* Copy real data, and all frags */
5834 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5835 skb->len -= off;
5836
5837 memcpy((struct skb_shared_info *)(data + size),
5838 skb_shinfo(skb),
5839 offsetof(struct skb_shared_info,
5840 frags[skb_shinfo(skb)->nr_frags]));
5841 if (skb_cloned(skb)) {
5842 /* drop the old head gracefully */
5843 if (skb_orphan_frags(skb, gfp_mask)) {
5844 kfree(data);
5845 return -ENOMEM;
5846 }
5847 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5848 skb_frag_ref(skb, i);
5849 if (skb_has_frag_list(skb))
5850 skb_clone_fraglist(skb);
5851 skb_release_data(skb);
5852 } else {
5853 /* we can reuse existing recount- all we did was
5854 * relocate values
5855 */
5856 skb_free_head(skb);
5857 }
5858
5859 skb->head = data;
5860 skb->data = data;
5861 skb->head_frag = 0;
5862#ifdef NET_SKBUFF_DATA_USES_OFFSET
5863 skb->end = size;
5864#else
5865 skb->end = skb->head + size;
5866#endif
5867 skb_set_tail_pointer(skb, skb_headlen(skb));
5868 skb_headers_offset_update(skb, 0);
5869 skb->cloned = 0;
5870 skb->hdr_len = 0;
5871 skb->nohdr = 0;
5872 atomic_set(&skb_shinfo(skb)->dataref, 1);
5873
5874 return 0;
5875}
5876
5877static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5878
5879/* carve out the first eat bytes from skb's frag_list. May recurse into
5880 * pskb_carve()
5881 */
5882static int pskb_carve_frag_list(struct sk_buff *skb,
5883 struct skb_shared_info *shinfo, int eat,
5884 gfp_t gfp_mask)
5885{
5886 struct sk_buff *list = shinfo->frag_list;
5887 struct sk_buff *clone = NULL;
5888 struct sk_buff *insp = NULL;
5889
5890 do {
5891 if (!list) {
5892 pr_err("Not enough bytes to eat. Want %d\n", eat);
5893 return -EFAULT;
5894 }
5895 if (list->len <= eat) {
5896 /* Eaten as whole. */
5897 eat -= list->len;
5898 list = list->next;
5899 insp = list;
5900 } else {
5901 /* Eaten partially. */
5902 if (skb_shared(list)) {
5903 clone = skb_clone(list, gfp_mask);
5904 if (!clone)
5905 return -ENOMEM;
5906 insp = list->next;
5907 list = clone;
5908 } else {
5909 /* This may be pulled without problems. */
5910 insp = list;
5911 }
5912 if (pskb_carve(list, eat, gfp_mask) < 0) {
5913 kfree_skb(clone);
5914 return -ENOMEM;
5915 }
5916 break;
5917 }
5918 } while (eat);
5919
5920 /* Free pulled out fragments. */
5921 while ((list = shinfo->frag_list) != insp) {
5922 shinfo->frag_list = list->next;
5923 kfree_skb(list);
5924 }
5925 /* And insert new clone at head. */
5926 if (clone) {
5927 clone->next = list;
5928 shinfo->frag_list = clone;
5929 }
5930 return 0;
5931}
5932
5933/* carve off first len bytes from skb. Split line (off) is in the
5934 * non-linear part of skb
5935 */
5936static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5937 int pos, gfp_t gfp_mask)
5938{
5939 int i, k = 0;
5940 int size = skb_end_offset(skb);
5941 u8 *data;
5942 const int nfrags = skb_shinfo(skb)->nr_frags;
5943 struct skb_shared_info *shinfo;
5944
5945 size = SKB_DATA_ALIGN(size);
5946
5947 if (skb_pfmemalloc(skb))
5948 gfp_mask |= __GFP_MEMALLOC;
5949 data = kmalloc_reserve(size +
5950 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5951 gfp_mask, NUMA_NO_NODE, NULL);
5952 if (!data)
5953 return -ENOMEM;
5954
5955 size = SKB_WITH_OVERHEAD(ksize(data));
5956
5957 memcpy((struct skb_shared_info *)(data + size),
5958 skb_shinfo(skb), offsetof(struct skb_shared_info,
5959 frags[skb_shinfo(skb)->nr_frags]));
5960 if (skb_orphan_frags(skb, gfp_mask)) {
5961 kfree(data);
5962 return -ENOMEM;
5963 }
5964 shinfo = (struct skb_shared_info *)(data + size);
5965 for (i = 0; i < nfrags; i++) {
5966 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5967
5968 if (pos + fsize > off) {
5969 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5970
5971 if (pos < off) {
5972 /* Split frag.
5973 * We have two variants in this case:
5974 * 1. Move all the frag to the second
5975 * part, if it is possible. F.e.
5976 * this approach is mandatory for TUX,
5977 * where splitting is expensive.
5978 * 2. Split is accurately. We make this.
5979 */
5980 skb_frag_off_add(&shinfo->frags[0], off - pos);
5981 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5982 }
5983 skb_frag_ref(skb, i);
5984 k++;
5985 }
5986 pos += fsize;
5987 }
5988 shinfo->nr_frags = k;
5989 if (skb_has_frag_list(skb))
5990 skb_clone_fraglist(skb);
5991
5992 /* split line is in frag list */
5993 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
5994 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
5995 if (skb_has_frag_list(skb))
5996 kfree_skb_list(skb_shinfo(skb)->frag_list);
5997 kfree(data);
5998 return -ENOMEM;
5999 }
6000 skb_release_data(skb);
6001
6002 skb->head = data;
6003 skb->head_frag = 0;
6004 skb->data = data;
6005#ifdef NET_SKBUFF_DATA_USES_OFFSET
6006 skb->end = size;
6007#else
6008 skb->end = skb->head + size;
6009#endif
6010 skb_reset_tail_pointer(skb);
6011 skb_headers_offset_update(skb, 0);
6012 skb->cloned = 0;
6013 skb->hdr_len = 0;
6014 skb->nohdr = 0;
6015 skb->len -= off;
6016 skb->data_len = skb->len;
6017 atomic_set(&skb_shinfo(skb)->dataref, 1);
6018 return 0;
6019}
6020
6021/* remove len bytes from the beginning of the skb */
6022static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6023{
6024 int headlen = skb_headlen(skb);
6025
6026 if (len < headlen)
6027 return pskb_carve_inside_header(skb, len, headlen, gfp);
6028 else
6029 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6030}
6031
6032/* Extract to_copy bytes starting at off from skb, and return this in
6033 * a new skb
6034 */
6035struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6036 int to_copy, gfp_t gfp)
6037{
6038 struct sk_buff *clone = skb_clone(skb, gfp);
6039
6040 if (!clone)
6041 return NULL;
6042
6043 if (pskb_carve(clone, off, gfp) < 0 ||
6044 pskb_trim(clone, to_copy)) {
6045 kfree_skb(clone);
6046 return NULL;
6047 }
6048 return clone;
6049}
6050EXPORT_SYMBOL(pskb_extract);
6051
6052/**
6053 * skb_condense - try to get rid of fragments/frag_list if possible
6054 * @skb: buffer
6055 *
6056 * Can be used to save memory before skb is added to a busy queue.
6057 * If packet has bytes in frags and enough tail room in skb->head,
6058 * pull all of them, so that we can free the frags right now and adjust
6059 * truesize.
6060 * Notes:
6061 * We do not reallocate skb->head thus can not fail.
6062 * Caller must re-evaluate skb->truesize if needed.
6063 */
6064void skb_condense(struct sk_buff *skb)
6065{
6066 if (skb->data_len) {
6067 if (skb->data_len > skb->end - skb->tail ||
6068 skb_cloned(skb))
6069 return;
6070
6071 /* Nice, we can free page frag(s) right now */
6072 __pskb_pull_tail(skb, skb->data_len);
6073 }
6074 /* At this point, skb->truesize might be over estimated,
6075 * because skb had a fragment, and fragments do not tell
6076 * their truesize.
6077 * When we pulled its content into skb->head, fragment
6078 * was freed, but __pskb_pull_tail() could not possibly
6079 * adjust skb->truesize, not knowing the frag truesize.
6080 */
6081 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6082}
6083
6084#ifdef CONFIG_SKB_EXTENSIONS
6085static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6086{
6087 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6088}
6089
6090/**
6091 * __skb_ext_alloc - allocate a new skb extensions storage
6092 *
6093 * @flags: See kmalloc().
6094 *
6095 * Returns the newly allocated pointer. The pointer can later attached to a
6096 * skb via __skb_ext_set().
6097 * Note: caller must handle the skb_ext as an opaque data.
6098 */
6099struct skb_ext *__skb_ext_alloc(gfp_t flags)
6100{
6101 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6102
6103 if (new) {
6104 memset(new->offset, 0, sizeof(new->offset));
6105 refcount_set(&new->refcnt, 1);
6106 }
6107
6108 return new;
6109}
6110
6111static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6112 unsigned int old_active)
6113{
6114 struct skb_ext *new;
6115
6116 if (refcount_read(&old->refcnt) == 1)
6117 return old;
6118
6119 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6120 if (!new)
6121 return NULL;
6122
6123 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6124 refcount_set(&new->refcnt, 1);
6125
6126#ifdef CONFIG_XFRM
6127 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6128 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6129 unsigned int i;
6130
6131 for (i = 0; i < sp->len; i++)
6132 xfrm_state_hold(sp->xvec[i]);
6133 }
6134#endif
6135 __skb_ext_put(old);
6136 return new;
6137}
6138
6139/**
6140 * __skb_ext_set - attach the specified extension storage to this skb
6141 * @skb: buffer
6142 * @id: extension id
6143 * @ext: extension storage previously allocated via __skb_ext_alloc()
6144 *
6145 * Existing extensions, if any, are cleared.
6146 *
6147 * Returns the pointer to the extension.
6148 */
6149void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6150 struct skb_ext *ext)
6151{
6152 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6153
6154 skb_ext_put(skb);
6155 newlen = newoff + skb_ext_type_len[id];
6156 ext->chunks = newlen;
6157 ext->offset[id] = newoff;
6158 skb->extensions = ext;
6159 skb->active_extensions = 1 << id;
6160 return skb_ext_get_ptr(ext, id);
6161}
6162
6163/**
6164 * skb_ext_add - allocate space for given extension, COW if needed
6165 * @skb: buffer
6166 * @id: extension to allocate space for
6167 *
6168 * Allocates enough space for the given extension.
6169 * If the extension is already present, a pointer to that extension
6170 * is returned.
6171 *
6172 * If the skb was cloned, COW applies and the returned memory can be
6173 * modified without changing the extension space of clones buffers.
6174 *
6175 * Returns pointer to the extension or NULL on allocation failure.
6176 */
6177void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6178{
6179 struct skb_ext *new, *old = NULL;
6180 unsigned int newlen, newoff;
6181
6182 if (skb->active_extensions) {
6183 old = skb->extensions;
6184
6185 new = skb_ext_maybe_cow(old, skb->active_extensions);
6186 if (!new)
6187 return NULL;
6188
6189 if (__skb_ext_exist(new, id))
6190 goto set_active;
6191
6192 newoff = new->chunks;
6193 } else {
6194 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6195
6196 new = __skb_ext_alloc(GFP_ATOMIC);
6197 if (!new)
6198 return NULL;
6199 }
6200
6201 newlen = newoff + skb_ext_type_len[id];
6202 new->chunks = newlen;
6203 new->offset[id] = newoff;
6204set_active:
6205 skb->extensions = new;
6206 skb->active_extensions |= 1 << id;
6207 return skb_ext_get_ptr(new, id);
6208}
6209EXPORT_SYMBOL(skb_ext_add);
6210
6211#ifdef CONFIG_XFRM
6212static void skb_ext_put_sp(struct sec_path *sp)
6213{
6214 unsigned int i;
6215
6216 for (i = 0; i < sp->len; i++)
6217 xfrm_state_put(sp->xvec[i]);
6218}
6219#endif
6220
6221void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6222{
6223 struct skb_ext *ext = skb->extensions;
6224
6225 skb->active_extensions &= ~(1 << id);
6226 if (skb->active_extensions == 0) {
6227 skb->extensions = NULL;
6228 __skb_ext_put(ext);
6229#ifdef CONFIG_XFRM
6230 } else if (id == SKB_EXT_SEC_PATH &&
6231 refcount_read(&ext->refcnt) == 1) {
6232 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6233
6234 skb_ext_put_sp(sp);
6235 sp->len = 0;
6236#endif
6237 }
6238}
6239EXPORT_SYMBOL(__skb_ext_del);
6240
6241void __skb_ext_put(struct skb_ext *ext)
6242{
6243 /* If this is last clone, nothing can increment
6244 * it after check passes. Avoids one atomic op.
6245 */
6246 if (refcount_read(&ext->refcnt) == 1)
6247 goto free_now;
6248
6249 if (!refcount_dec_and_test(&ext->refcnt))
6250 return;
6251free_now:
6252#ifdef CONFIG_XFRM
6253 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6254 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6255#endif
6256
6257 kmem_cache_free(skbuff_ext_cache, ext);
6258}
6259EXPORT_SYMBOL(__skb_ext_put);
6260#endif /* CONFIG_SKB_EXTENSIONS */