MIPS: Yosemite, Emma: Fix off-by-two in arcs_cmdline buffer size check
[linux-2.6/linux-mips.git] / net / core / skbuff.c
blobca4db40e75b84becaab9c9acef48e5677d202a6c
1 /*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
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
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).
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.
36 * The functions in this file will not compile correctly with gcc 2.4.x
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>
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>
68 #include <asm/uaccess.h>
69 #include <asm/system.h>
70 #include <trace/events/skb.h>
72 #include "kmap_skb.h"
74 static struct kmem_cache *skbuff_head_cache __read_mostly;
75 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
77 static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
80 put_page(buf->page);
83 static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
86 get_page(buf->page);
89 static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
92 return 1;
96 /* Pipe buffer operations for a socket. */
97 static 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,
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
110 * reliable.
114 * skb_over_panic - private function
115 * @skb: buffer
116 * @sz: size
117 * @here: address
119 * Out of line support code for skb_put(). Not user callable.
121 static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
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();
132 * skb_under_panic - private function
133 * @skb: buffer
134 * @sz: size
135 * @here: address
137 * Out of line support code for skb_push(). Not user callable.
140 static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
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();
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.
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
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.
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 * %GFP_ATOMIC.
171 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
176 struct sk_buff *skb;
177 u8 *data;
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
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);
187 /* We do our best to align skb_shared_info on a separate cache
188 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
189 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
190 * Both skb->head and skb_shared_info are cache line aligned.
192 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
193 data = kmalloc_node_track_caller(size, gfp_mask, node);
194 if (!data)
195 goto nodata;
196 /* kmalloc(size) might give us more room than requested.
197 * Put skb_shared_info exactly at the end of allocated zone,
198 * to allow max possible filling before reallocation.
200 size = SKB_WITH_OVERHEAD(ksize(data));
201 prefetchw(data + size);
204 * Only clear those fields we need to clear, not those that we will
205 * actually initialise below. Hence, don't put any more fields after
206 * the tail pointer in struct sk_buff!
208 memset(skb, 0, offsetof(struct sk_buff, tail));
209 /* Account for allocated memory : skb + skb->head */
210 skb->truesize = SKB_TRUESIZE(size);
211 atomic_set(&skb->users, 1);
212 skb->head = data;
213 skb->data = data;
214 skb_reset_tail_pointer(skb);
215 skb->end = skb->tail + size;
216 #ifdef NET_SKBUFF_DATA_USES_OFFSET
217 skb->mac_header = ~0U;
218 #endif
220 /* make sure we initialize shinfo sequentially */
221 shinfo = skb_shinfo(skb);
222 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
223 atomic_set(&shinfo->dataref, 1);
224 kmemcheck_annotate_variable(shinfo->destructor_arg);
226 if (fclone) {
227 struct sk_buff *child = skb + 1;
228 atomic_t *fclone_ref = (atomic_t *) (child + 1);
230 kmemcheck_annotate_bitfield(child, flags1);
231 kmemcheck_annotate_bitfield(child, flags2);
232 skb->fclone = SKB_FCLONE_ORIG;
233 atomic_set(fclone_ref, 1);
235 child->fclone = SKB_FCLONE_UNAVAILABLE;
237 out:
238 return skb;
239 nodata:
240 kmem_cache_free(cache, skb);
241 skb = NULL;
242 goto out;
244 EXPORT_SYMBOL(__alloc_skb);
247 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
248 * @dev: network device to receive on
249 * @length: length to allocate
250 * @gfp_mask: get_free_pages mask, passed to alloc_skb
252 * Allocate a new &sk_buff and assign it a usage count of one. The
253 * buffer has unspecified headroom built in. Users should allocate
254 * the headroom they think they need without accounting for the
255 * built in space. The built in space is used for optimisations.
257 * %NULL is returned if there is no free memory.
259 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
260 unsigned int length, gfp_t gfp_mask)
262 struct sk_buff *skb;
264 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
265 if (likely(skb)) {
266 skb_reserve(skb, NET_SKB_PAD);
267 skb->dev = dev;
269 return skb;
271 EXPORT_SYMBOL(__netdev_alloc_skb);
273 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
274 int size)
276 skb_fill_page_desc(skb, i, page, off, size);
277 skb->len += size;
278 skb->data_len += size;
279 skb->truesize += size;
281 EXPORT_SYMBOL(skb_add_rx_frag);
284 * dev_alloc_skb - allocate an skbuff for receiving
285 * @length: length to allocate
287 * Allocate a new &sk_buff and assign it a usage count of one. The
288 * buffer has unspecified headroom built in. Users should allocate
289 * the headroom they think they need without accounting for the
290 * built in space. The built in space is used for optimisations.
292 * %NULL is returned if there is no free memory. Although this function
293 * allocates memory it can be called from an interrupt.
295 struct sk_buff *dev_alloc_skb(unsigned int length)
298 * There is more code here than it seems:
299 * __dev_alloc_skb is an inline
301 return __dev_alloc_skb(length, GFP_ATOMIC);
303 EXPORT_SYMBOL(dev_alloc_skb);
305 static void skb_drop_list(struct sk_buff **listp)
307 struct sk_buff *list = *listp;
309 *listp = NULL;
311 do {
312 struct sk_buff *this = list;
313 list = list->next;
314 kfree_skb(this);
315 } while (list);
318 static inline void skb_drop_fraglist(struct sk_buff *skb)
320 skb_drop_list(&skb_shinfo(skb)->frag_list);
323 static void skb_clone_fraglist(struct sk_buff *skb)
325 struct sk_buff *list;
327 skb_walk_frags(skb, list)
328 skb_get(list);
331 static void skb_release_data(struct sk_buff *skb)
333 if (!skb->cloned ||
334 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
335 &skb_shinfo(skb)->dataref)) {
336 if (skb_shinfo(skb)->nr_frags) {
337 int i;
338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
339 skb_frag_unref(skb, i);
343 * If skb buf is from userspace, we need to notify the caller
344 * the lower device DMA has done;
346 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
347 struct ubuf_info *uarg;
349 uarg = skb_shinfo(skb)->destructor_arg;
350 if (uarg->callback)
351 uarg->callback(uarg);
354 if (skb_has_frag_list(skb))
355 skb_drop_fraglist(skb);
357 kfree(skb->head);
362 * Free an skbuff by memory without cleaning the state.
364 static void kfree_skbmem(struct sk_buff *skb)
366 struct sk_buff *other;
367 atomic_t *fclone_ref;
369 switch (skb->fclone) {
370 case SKB_FCLONE_UNAVAILABLE:
371 kmem_cache_free(skbuff_head_cache, skb);
372 break;
374 case SKB_FCLONE_ORIG:
375 fclone_ref = (atomic_t *) (skb + 2);
376 if (atomic_dec_and_test(fclone_ref))
377 kmem_cache_free(skbuff_fclone_cache, skb);
378 break;
380 case SKB_FCLONE_CLONE:
381 fclone_ref = (atomic_t *) (skb + 1);
382 other = skb - 1;
384 /* The clone portion is available for
385 * fast-cloning again.
387 skb->fclone = SKB_FCLONE_UNAVAILABLE;
389 if (atomic_dec_and_test(fclone_ref))
390 kmem_cache_free(skbuff_fclone_cache, other);
391 break;
395 static void skb_release_head_state(struct sk_buff *skb)
397 skb_dst_drop(skb);
398 #ifdef CONFIG_XFRM
399 secpath_put(skb->sp);
400 #endif
401 if (skb->destructor) {
402 WARN_ON(in_irq());
403 skb->destructor(skb);
405 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
406 nf_conntrack_put(skb->nfct);
407 #endif
408 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
409 nf_conntrack_put_reasm(skb->nfct_reasm);
410 #endif
411 #ifdef CONFIG_BRIDGE_NETFILTER
412 nf_bridge_put(skb->nf_bridge);
413 #endif
414 /* XXX: IS this still necessary? - JHS */
415 #ifdef CONFIG_NET_SCHED
416 skb->tc_index = 0;
417 #ifdef CONFIG_NET_CLS_ACT
418 skb->tc_verd = 0;
419 #endif
420 #endif
423 /* Free everything but the sk_buff shell. */
424 static void skb_release_all(struct sk_buff *skb)
426 skb_release_head_state(skb);
427 skb_release_data(skb);
431 * __kfree_skb - private function
432 * @skb: buffer
434 * Free an sk_buff. Release anything attached to the buffer.
435 * Clean the state. This is an internal helper function. Users should
436 * always call kfree_skb
439 void __kfree_skb(struct sk_buff *skb)
441 skb_release_all(skb);
442 kfree_skbmem(skb);
444 EXPORT_SYMBOL(__kfree_skb);
447 * kfree_skb - free an sk_buff
448 * @skb: buffer to free
450 * Drop a reference to the buffer and free it if the usage count has
451 * hit zero.
453 void kfree_skb(struct sk_buff *skb)
455 if (unlikely(!skb))
456 return;
457 if (likely(atomic_read(&skb->users) == 1))
458 smp_rmb();
459 else if (likely(!atomic_dec_and_test(&skb->users)))
460 return;
461 trace_kfree_skb(skb, __builtin_return_address(0));
462 __kfree_skb(skb);
464 EXPORT_SYMBOL(kfree_skb);
467 * consume_skb - free an skbuff
468 * @skb: buffer to free
470 * Drop a ref to the buffer and free it if the usage count has hit zero
471 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
472 * is being dropped after a failure and notes that
474 void consume_skb(struct sk_buff *skb)
476 if (unlikely(!skb))
477 return;
478 if (likely(atomic_read(&skb->users) == 1))
479 smp_rmb();
480 else if (likely(!atomic_dec_and_test(&skb->users)))
481 return;
482 trace_consume_skb(skb);
483 __kfree_skb(skb);
485 EXPORT_SYMBOL(consume_skb);
488 * skb_recycle - clean up an skb for reuse
489 * @skb: buffer
491 * Recycles the skb to be reused as a receive buffer. This
492 * function does any necessary reference count dropping, and
493 * cleans up the skbuff as if it just came from __alloc_skb().
495 void skb_recycle(struct sk_buff *skb)
497 struct skb_shared_info *shinfo;
499 skb_release_head_state(skb);
501 shinfo = skb_shinfo(skb);
502 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
503 atomic_set(&shinfo->dataref, 1);
505 memset(skb, 0, offsetof(struct sk_buff, tail));
506 skb->data = skb->head + NET_SKB_PAD;
507 skb_reset_tail_pointer(skb);
509 EXPORT_SYMBOL(skb_recycle);
512 * skb_recycle_check - check if skb can be reused for receive
513 * @skb: buffer
514 * @skb_size: minimum receive buffer size
516 * Checks that the skb passed in is not shared or cloned, and
517 * that it is linear and its head portion at least as large as
518 * skb_size so that it can be recycled as a receive buffer.
519 * If these conditions are met, this function does any necessary
520 * reference count dropping and cleans up the skbuff as if it
521 * just came from __alloc_skb().
523 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
525 if (!skb_is_recycleable(skb, skb_size))
526 return false;
528 skb_recycle(skb);
530 return true;
532 EXPORT_SYMBOL(skb_recycle_check);
534 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
536 new->tstamp = old->tstamp;
537 new->dev = old->dev;
538 new->transport_header = old->transport_header;
539 new->network_header = old->network_header;
540 new->mac_header = old->mac_header;
541 skb_dst_copy(new, old);
542 new->rxhash = old->rxhash;
543 new->ooo_okay = old->ooo_okay;
544 new->l4_rxhash = old->l4_rxhash;
545 #ifdef CONFIG_XFRM
546 new->sp = secpath_get(old->sp);
547 #endif
548 memcpy(new->cb, old->cb, sizeof(old->cb));
549 new->csum = old->csum;
550 new->local_df = old->local_df;
551 new->pkt_type = old->pkt_type;
552 new->ip_summed = old->ip_summed;
553 skb_copy_queue_mapping(new, old);
554 new->priority = old->priority;
555 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
556 new->ipvs_property = old->ipvs_property;
557 #endif
558 new->protocol = old->protocol;
559 new->mark = old->mark;
560 new->skb_iif = old->skb_iif;
561 __nf_copy(new, old);
562 #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
563 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
564 new->nf_trace = old->nf_trace;
565 #endif
566 #ifdef CONFIG_NET_SCHED
567 new->tc_index = old->tc_index;
568 #ifdef CONFIG_NET_CLS_ACT
569 new->tc_verd = old->tc_verd;
570 #endif
571 #endif
572 new->vlan_tci = old->vlan_tci;
574 skb_copy_secmark(new, old);
578 * You should not add any new code to this function. Add it to
579 * __copy_skb_header above instead.
581 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
583 #define C(x) n->x = skb->x
585 n->next = n->prev = NULL;
586 n->sk = NULL;
587 __copy_skb_header(n, skb);
589 C(len);
590 C(data_len);
591 C(mac_len);
592 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
593 n->cloned = 1;
594 n->nohdr = 0;
595 n->destructor = NULL;
596 C(tail);
597 C(end);
598 C(head);
599 C(data);
600 C(truesize);
601 atomic_set(&n->users, 1);
603 atomic_inc(&(skb_shinfo(skb)->dataref));
604 skb->cloned = 1;
606 return n;
607 #undef C
611 * skb_morph - morph one skb into another
612 * @dst: the skb to receive the contents
613 * @src: the skb to supply the contents
615 * This is identical to skb_clone except that the target skb is
616 * supplied by the user.
618 * The target skb is returned upon exit.
620 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
622 skb_release_all(dst);
623 return __skb_clone(dst, src);
625 EXPORT_SYMBOL_GPL(skb_morph);
627 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
628 * @skb: the skb to modify
629 * @gfp_mask: allocation priority
631 * This must be called on SKBTX_DEV_ZEROCOPY skb.
632 * It will copy all frags into kernel and drop the reference
633 * to userspace pages.
635 * If this function is called from an interrupt gfp_mask() must be
636 * %GFP_ATOMIC.
638 * Returns 0 on success or a negative error code on failure
639 * to allocate kernel memory to copy to.
641 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
643 int i;
644 int num_frags = skb_shinfo(skb)->nr_frags;
645 struct page *page, *head = NULL;
646 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
648 for (i = 0; i < num_frags; i++) {
649 u8 *vaddr;
650 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
652 page = alloc_page(GFP_ATOMIC);
653 if (!page) {
654 while (head) {
655 struct page *next = (struct page *)head->private;
656 put_page(head);
657 head = next;
659 return -ENOMEM;
661 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
662 memcpy(page_address(page),
663 vaddr + f->page_offset, skb_frag_size(f));
664 kunmap_skb_frag(vaddr);
665 page->private = (unsigned long)head;
666 head = page;
669 /* skb frags release userspace buffers */
670 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
671 skb_frag_unref(skb, i);
673 uarg->callback(uarg);
675 /* skb frags point to kernel buffers */
676 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
677 __skb_fill_page_desc(skb, i-1, head, 0,
678 skb_shinfo(skb)->frags[i - 1].size);
679 head = (struct page *)head->private;
682 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
683 return 0;
688 * skb_clone - duplicate an sk_buff
689 * @skb: buffer to clone
690 * @gfp_mask: allocation priority
692 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
693 * copies share the same packet data but not structure. The new
694 * buffer has a reference count of 1. If the allocation fails the
695 * function returns %NULL otherwise the new buffer is returned.
697 * If this function is called from an interrupt gfp_mask() must be
698 * %GFP_ATOMIC.
701 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
703 struct sk_buff *n;
705 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
706 if (skb_copy_ubufs(skb, gfp_mask))
707 return NULL;
710 n = skb + 1;
711 if (skb->fclone == SKB_FCLONE_ORIG &&
712 n->fclone == SKB_FCLONE_UNAVAILABLE) {
713 atomic_t *fclone_ref = (atomic_t *) (n + 1);
714 n->fclone = SKB_FCLONE_CLONE;
715 atomic_inc(fclone_ref);
716 } else {
717 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
718 if (!n)
719 return NULL;
721 kmemcheck_annotate_bitfield(n, flags1);
722 kmemcheck_annotate_bitfield(n, flags2);
723 n->fclone = SKB_FCLONE_UNAVAILABLE;
726 return __skb_clone(n, skb);
728 EXPORT_SYMBOL(skb_clone);
730 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
732 #ifndef NET_SKBUFF_DATA_USES_OFFSET
734 * Shift between the two data areas in bytes
736 unsigned long offset = new->data - old->data;
737 #endif
739 __copy_skb_header(new, old);
741 #ifndef NET_SKBUFF_DATA_USES_OFFSET
742 /* {transport,network,mac}_header are relative to skb->head */
743 new->transport_header += offset;
744 new->network_header += offset;
745 if (skb_mac_header_was_set(new))
746 new->mac_header += offset;
747 #endif
748 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
749 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
750 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
754 * skb_copy - create private copy of an sk_buff
755 * @skb: buffer to copy
756 * @gfp_mask: allocation priority
758 * Make a copy of both an &sk_buff and its data. This is used when the
759 * caller wishes to modify the data and needs a private copy of the
760 * data to alter. Returns %NULL on failure or the pointer to the buffer
761 * on success. The returned buffer has a reference count of 1.
763 * As by-product this function converts non-linear &sk_buff to linear
764 * one, so that &sk_buff becomes completely private and caller is allowed
765 * to modify all the data of returned buffer. This means that this
766 * function is not recommended for use in circumstances when only
767 * header is going to be modified. Use pskb_copy() instead.
770 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
772 int headerlen = skb_headroom(skb);
773 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
774 struct sk_buff *n = alloc_skb(size, gfp_mask);
776 if (!n)
777 return NULL;
779 /* Set the data pointer */
780 skb_reserve(n, headerlen);
781 /* Set the tail pointer and length */
782 skb_put(n, skb->len);
784 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
785 BUG();
787 copy_skb_header(n, skb);
788 return n;
790 EXPORT_SYMBOL(skb_copy);
793 * pskb_copy - create copy of an sk_buff with private head.
794 * @skb: buffer to copy
795 * @gfp_mask: allocation priority
797 * Make a copy of both an &sk_buff and part of its data, located
798 * in header. Fragmented data remain shared. This is used when
799 * the caller wishes to modify only header of &sk_buff and needs
800 * private copy of the header to alter. Returns %NULL on failure
801 * or the pointer to the buffer on success.
802 * The returned buffer has a reference count of 1.
805 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
807 unsigned int size = skb_end_pointer(skb) - skb->head;
808 struct sk_buff *n = alloc_skb(size, gfp_mask);
810 if (!n)
811 goto out;
813 /* Set the data pointer */
814 skb_reserve(n, skb_headroom(skb));
815 /* Set the tail pointer and length */
816 skb_put(n, skb_headlen(skb));
817 /* Copy the bytes */
818 skb_copy_from_linear_data(skb, n->data, n->len);
820 n->truesize += skb->data_len;
821 n->data_len = skb->data_len;
822 n->len = skb->len;
824 if (skb_shinfo(skb)->nr_frags) {
825 int i;
827 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
828 if (skb_copy_ubufs(skb, gfp_mask)) {
829 kfree_skb(n);
830 n = NULL;
831 goto out;
834 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
835 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
836 skb_frag_ref(skb, i);
838 skb_shinfo(n)->nr_frags = i;
841 if (skb_has_frag_list(skb)) {
842 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
843 skb_clone_fraglist(n);
846 copy_skb_header(n, skb);
847 out:
848 return n;
850 EXPORT_SYMBOL(pskb_copy);
853 * pskb_expand_head - reallocate header of &sk_buff
854 * @skb: buffer to reallocate
855 * @nhead: room to add at head
856 * @ntail: room to add at tail
857 * @gfp_mask: allocation priority
859 * Expands (or creates identical copy, if &nhead and &ntail are zero)
860 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
861 * reference count of 1. Returns zero in the case of success or error,
862 * if expansion failed. In the last case, &sk_buff is not changed.
864 * All the pointers pointing into skb header may change and must be
865 * reloaded after call to this function.
868 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
869 gfp_t gfp_mask)
871 int i;
872 u8 *data;
873 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
874 long off;
875 bool fastpath;
877 BUG_ON(nhead < 0);
879 if (skb_shared(skb))
880 BUG();
882 size = SKB_DATA_ALIGN(size);
884 /* Check if we can avoid taking references on fragments if we own
885 * the last reference on skb->head. (see skb_release_data())
887 if (!skb->cloned)
888 fastpath = true;
889 else {
890 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
891 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
894 if (fastpath &&
895 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
896 memmove(skb->head + size, skb_shinfo(skb),
897 offsetof(struct skb_shared_info,
898 frags[skb_shinfo(skb)->nr_frags]));
899 memmove(skb->head + nhead, skb->head,
900 skb_tail_pointer(skb) - skb->head);
901 off = nhead;
902 goto adjust_others;
905 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
906 if (!data)
907 goto nodata;
909 /* Copy only real data... and, alas, header. This should be
910 * optimized for the cases when header is void.
912 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
914 memcpy((struct skb_shared_info *)(data + size),
915 skb_shinfo(skb),
916 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
918 if (fastpath) {
919 kfree(skb->head);
920 } else {
921 /* copy this zero copy skb frags */
922 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
923 if (skb_copy_ubufs(skb, gfp_mask))
924 goto nofrags;
926 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
927 skb_frag_ref(skb, i);
929 if (skb_has_frag_list(skb))
930 skb_clone_fraglist(skb);
932 skb_release_data(skb);
934 off = (data + nhead) - skb->head;
936 skb->head = data;
937 adjust_others:
938 skb->data += off;
939 #ifdef NET_SKBUFF_DATA_USES_OFFSET
940 skb->end = size;
941 off = nhead;
942 #else
943 skb->end = skb->head + size;
944 #endif
945 /* {transport,network,mac}_header and tail are relative to skb->head */
946 skb->tail += off;
947 skb->transport_header += off;
948 skb->network_header += off;
949 if (skb_mac_header_was_set(skb))
950 skb->mac_header += off;
951 /* Only adjust this if it actually is csum_start rather than csum */
952 if (skb->ip_summed == CHECKSUM_PARTIAL)
953 skb->csum_start += nhead;
954 skb->cloned = 0;
955 skb->hdr_len = 0;
956 skb->nohdr = 0;
957 atomic_set(&skb_shinfo(skb)->dataref, 1);
958 return 0;
960 nofrags:
961 kfree(data);
962 nodata:
963 return -ENOMEM;
965 EXPORT_SYMBOL(pskb_expand_head);
967 /* Make private copy of skb with writable head and some headroom */
969 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
971 struct sk_buff *skb2;
972 int delta = headroom - skb_headroom(skb);
974 if (delta <= 0)
975 skb2 = pskb_copy(skb, GFP_ATOMIC);
976 else {
977 skb2 = skb_clone(skb, GFP_ATOMIC);
978 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
979 GFP_ATOMIC)) {
980 kfree_skb(skb2);
981 skb2 = NULL;
984 return skb2;
986 EXPORT_SYMBOL(skb_realloc_headroom);
989 * skb_copy_expand - copy and expand sk_buff
990 * @skb: buffer to copy
991 * @newheadroom: new free bytes at head
992 * @newtailroom: new free bytes at tail
993 * @gfp_mask: allocation priority
995 * Make a copy of both an &sk_buff and its data and while doing so
996 * allocate additional space.
998 * This is used when the caller wishes to modify the data and needs a
999 * private copy of the data to alter as well as more space for new fields.
1000 * Returns %NULL on failure or the pointer to the buffer
1001 * on success. The returned buffer has a reference count of 1.
1003 * You must pass %GFP_ATOMIC as the allocation priority if this function
1004 * is called from an interrupt.
1006 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1007 int newheadroom, int newtailroom,
1008 gfp_t gfp_mask)
1011 * Allocate the copy buffer
1013 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1014 gfp_mask);
1015 int oldheadroom = skb_headroom(skb);
1016 int head_copy_len, head_copy_off;
1017 int off;
1019 if (!n)
1020 return NULL;
1022 skb_reserve(n, newheadroom);
1024 /* Set the tail pointer and length */
1025 skb_put(n, skb->len);
1027 head_copy_len = oldheadroom;
1028 head_copy_off = 0;
1029 if (newheadroom <= head_copy_len)
1030 head_copy_len = newheadroom;
1031 else
1032 head_copy_off = newheadroom - head_copy_len;
1034 /* Copy the linear header and data. */
1035 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1036 skb->len + head_copy_len))
1037 BUG();
1039 copy_skb_header(n, skb);
1041 off = newheadroom - oldheadroom;
1042 if (n->ip_summed == CHECKSUM_PARTIAL)
1043 n->csum_start += off;
1044 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1045 n->transport_header += off;
1046 n->network_header += off;
1047 if (skb_mac_header_was_set(skb))
1048 n->mac_header += off;
1049 #endif
1051 return n;
1053 EXPORT_SYMBOL(skb_copy_expand);
1056 * skb_pad - zero pad the tail of an skb
1057 * @skb: buffer to pad
1058 * @pad: space to pad
1060 * Ensure that a buffer is followed by a padding area that is zero
1061 * filled. Used by network drivers which may DMA or transfer data
1062 * beyond the buffer end onto the wire.
1064 * May return error in out of memory cases. The skb is freed on error.
1067 int skb_pad(struct sk_buff *skb, int pad)
1069 int err;
1070 int ntail;
1072 /* If the skbuff is non linear tailroom is always zero.. */
1073 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1074 memset(skb->data+skb->len, 0, pad);
1075 return 0;
1078 ntail = skb->data_len + pad - (skb->end - skb->tail);
1079 if (likely(skb_cloned(skb) || ntail > 0)) {
1080 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1081 if (unlikely(err))
1082 goto free_skb;
1085 /* FIXME: The use of this function with non-linear skb's really needs
1086 * to be audited.
1088 err = skb_linearize(skb);
1089 if (unlikely(err))
1090 goto free_skb;
1092 memset(skb->data + skb->len, 0, pad);
1093 return 0;
1095 free_skb:
1096 kfree_skb(skb);
1097 return err;
1099 EXPORT_SYMBOL(skb_pad);
1102 * skb_put - add data to a buffer
1103 * @skb: buffer to use
1104 * @len: amount of data to add
1106 * This function extends the used data area of the buffer. If this would
1107 * exceed the total buffer size the kernel will panic. A pointer to the
1108 * first byte of the extra data is returned.
1110 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1112 unsigned char *tmp = skb_tail_pointer(skb);
1113 SKB_LINEAR_ASSERT(skb);
1114 skb->tail += len;
1115 skb->len += len;
1116 if (unlikely(skb->tail > skb->end))
1117 skb_over_panic(skb, len, __builtin_return_address(0));
1118 return tmp;
1120 EXPORT_SYMBOL(skb_put);
1123 * skb_push - add data to the start of a buffer
1124 * @skb: buffer to use
1125 * @len: amount of data to add
1127 * This function extends the used data area of the buffer at the buffer
1128 * start. If this would exceed the total buffer headroom the kernel will
1129 * panic. A pointer to the first byte of the extra data is returned.
1131 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1133 skb->data -= len;
1134 skb->len += len;
1135 if (unlikely(skb->data<skb->head))
1136 skb_under_panic(skb, len, __builtin_return_address(0));
1137 return skb->data;
1139 EXPORT_SYMBOL(skb_push);
1142 * skb_pull - remove data from the start of a buffer
1143 * @skb: buffer to use
1144 * @len: amount of data to remove
1146 * This function removes data from the start of a buffer, returning
1147 * the memory to the headroom. A pointer to the next data in the buffer
1148 * is returned. Once the data has been pulled future pushes will overwrite
1149 * the old data.
1151 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1153 return skb_pull_inline(skb, len);
1155 EXPORT_SYMBOL(skb_pull);
1158 * skb_trim - remove end from a buffer
1159 * @skb: buffer to alter
1160 * @len: new length
1162 * Cut the length of a buffer down by removing data from the tail. If
1163 * the buffer is already under the length specified it is not modified.
1164 * The skb must be linear.
1166 void skb_trim(struct sk_buff *skb, unsigned int len)
1168 if (skb->len > len)
1169 __skb_trim(skb, len);
1171 EXPORT_SYMBOL(skb_trim);
1173 /* Trims skb to length len. It can change skb pointers.
1176 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1178 struct sk_buff **fragp;
1179 struct sk_buff *frag;
1180 int offset = skb_headlen(skb);
1181 int nfrags = skb_shinfo(skb)->nr_frags;
1182 int i;
1183 int err;
1185 if (skb_cloned(skb) &&
1186 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1187 return err;
1189 i = 0;
1190 if (offset >= len)
1191 goto drop_pages;
1193 for (; i < nfrags; i++) {
1194 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1196 if (end < len) {
1197 offset = end;
1198 continue;
1201 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1203 drop_pages:
1204 skb_shinfo(skb)->nr_frags = i;
1206 for (; i < nfrags; i++)
1207 skb_frag_unref(skb, i);
1209 if (skb_has_frag_list(skb))
1210 skb_drop_fraglist(skb);
1211 goto done;
1214 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1215 fragp = &frag->next) {
1216 int end = offset + frag->len;
1218 if (skb_shared(frag)) {
1219 struct sk_buff *nfrag;
1221 nfrag = skb_clone(frag, GFP_ATOMIC);
1222 if (unlikely(!nfrag))
1223 return -ENOMEM;
1225 nfrag->next = frag->next;
1226 kfree_skb(frag);
1227 frag = nfrag;
1228 *fragp = frag;
1231 if (end < len) {
1232 offset = end;
1233 continue;
1236 if (end > len &&
1237 unlikely((err = pskb_trim(frag, len - offset))))
1238 return err;
1240 if (frag->next)
1241 skb_drop_list(&frag->next);
1242 break;
1245 done:
1246 if (len > skb_headlen(skb)) {
1247 skb->data_len -= skb->len - len;
1248 skb->len = len;
1249 } else {
1250 skb->len = len;
1251 skb->data_len = 0;
1252 skb_set_tail_pointer(skb, len);
1255 return 0;
1257 EXPORT_SYMBOL(___pskb_trim);
1260 * __pskb_pull_tail - advance tail of skb header
1261 * @skb: buffer to reallocate
1262 * @delta: number of bytes to advance tail
1264 * The function makes a sense only on a fragmented &sk_buff,
1265 * it expands header moving its tail forward and copying necessary
1266 * data from fragmented part.
1268 * &sk_buff MUST have reference count of 1.
1270 * Returns %NULL (and &sk_buff does not change) if pull failed
1271 * or value of new tail of skb in the case of success.
1273 * All the pointers pointing into skb header may change and must be
1274 * reloaded after call to this function.
1277 /* Moves tail of skb head forward, copying data from fragmented part,
1278 * when it is necessary.
1279 * 1. It may fail due to malloc failure.
1280 * 2. It may change skb pointers.
1282 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1284 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1286 /* If skb has not enough free space at tail, get new one
1287 * plus 128 bytes for future expansions. If we have enough
1288 * room at tail, reallocate without expansion only if skb is cloned.
1290 int i, k, eat = (skb->tail + delta) - skb->end;
1292 if (eat > 0 || skb_cloned(skb)) {
1293 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1294 GFP_ATOMIC))
1295 return NULL;
1298 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1299 BUG();
1301 /* Optimization: no fragments, no reasons to preestimate
1302 * size of pulled pages. Superb.
1304 if (!skb_has_frag_list(skb))
1305 goto pull_pages;
1307 /* Estimate size of pulled pages. */
1308 eat = delta;
1309 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1310 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1312 if (size >= eat)
1313 goto pull_pages;
1314 eat -= size;
1317 /* If we need update frag list, we are in troubles.
1318 * Certainly, it possible to add an offset to skb data,
1319 * but taking into account that pulling is expected to
1320 * be very rare operation, it is worth to fight against
1321 * further bloating skb head and crucify ourselves here instead.
1322 * Pure masohism, indeed. 8)8)
1324 if (eat) {
1325 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1326 struct sk_buff *clone = NULL;
1327 struct sk_buff *insp = NULL;
1329 do {
1330 BUG_ON(!list);
1332 if (list->len <= eat) {
1333 /* Eaten as whole. */
1334 eat -= list->len;
1335 list = list->next;
1336 insp = list;
1337 } else {
1338 /* Eaten partially. */
1340 if (skb_shared(list)) {
1341 /* Sucks! We need to fork list. :-( */
1342 clone = skb_clone(list, GFP_ATOMIC);
1343 if (!clone)
1344 return NULL;
1345 insp = list->next;
1346 list = clone;
1347 } else {
1348 /* This may be pulled without
1349 * problems. */
1350 insp = list;
1352 if (!pskb_pull(list, eat)) {
1353 kfree_skb(clone);
1354 return NULL;
1356 break;
1358 } while (eat);
1360 /* Free pulled out fragments. */
1361 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1362 skb_shinfo(skb)->frag_list = list->next;
1363 kfree_skb(list);
1365 /* And insert new clone at head. */
1366 if (clone) {
1367 clone->next = list;
1368 skb_shinfo(skb)->frag_list = clone;
1371 /* Success! Now we may commit changes to skb data. */
1373 pull_pages:
1374 eat = delta;
1375 k = 0;
1376 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1377 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1379 if (size <= eat) {
1380 skb_frag_unref(skb, i);
1381 eat -= size;
1382 } else {
1383 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1384 if (eat) {
1385 skb_shinfo(skb)->frags[k].page_offset += eat;
1386 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1387 eat = 0;
1389 k++;
1392 skb_shinfo(skb)->nr_frags = k;
1394 skb->tail += delta;
1395 skb->data_len -= delta;
1397 return skb_tail_pointer(skb);
1399 EXPORT_SYMBOL(__pskb_pull_tail);
1402 * skb_copy_bits - copy bits from skb to kernel buffer
1403 * @skb: source skb
1404 * @offset: offset in source
1405 * @to: destination buffer
1406 * @len: number of bytes to copy
1408 * Copy the specified number of bytes from the source skb to the
1409 * destination buffer.
1411 * CAUTION ! :
1412 * If its prototype is ever changed,
1413 * check arch/{*}/net/{*}.S files,
1414 * since it is called from BPF assembly code.
1416 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1418 int start = skb_headlen(skb);
1419 struct sk_buff *frag_iter;
1420 int i, copy;
1422 if (offset > (int)skb->len - len)
1423 goto fault;
1425 /* Copy header. */
1426 if ((copy = start - offset) > 0) {
1427 if (copy > len)
1428 copy = len;
1429 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1430 if ((len -= copy) == 0)
1431 return 0;
1432 offset += copy;
1433 to += copy;
1436 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1437 int end;
1439 WARN_ON(start > offset + len);
1441 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1442 if ((copy = end - offset) > 0) {
1443 u8 *vaddr;
1445 if (copy > len)
1446 copy = len;
1448 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1449 memcpy(to,
1450 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1451 offset - start, copy);
1452 kunmap_skb_frag(vaddr);
1454 if ((len -= copy) == 0)
1455 return 0;
1456 offset += copy;
1457 to += copy;
1459 start = end;
1462 skb_walk_frags(skb, frag_iter) {
1463 int end;
1465 WARN_ON(start > offset + len);
1467 end = start + frag_iter->len;
1468 if ((copy = end - offset) > 0) {
1469 if (copy > len)
1470 copy = len;
1471 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1472 goto fault;
1473 if ((len -= copy) == 0)
1474 return 0;
1475 offset += copy;
1476 to += copy;
1478 start = end;
1481 if (!len)
1482 return 0;
1484 fault:
1485 return -EFAULT;
1487 EXPORT_SYMBOL(skb_copy_bits);
1490 * Callback from splice_to_pipe(), if we need to release some pages
1491 * at the end of the spd in case we error'ed out in filling the pipe.
1493 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1495 put_page(spd->pages[i]);
1498 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1499 unsigned int *offset,
1500 struct sk_buff *skb, struct sock *sk)
1502 struct page *p = sk->sk_sndmsg_page;
1503 unsigned int off;
1505 if (!p) {
1506 new_page:
1507 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1508 if (!p)
1509 return NULL;
1511 off = sk->sk_sndmsg_off = 0;
1512 /* hold one ref to this page until it's full */
1513 } else {
1514 unsigned int mlen;
1516 off = sk->sk_sndmsg_off;
1517 mlen = PAGE_SIZE - off;
1518 if (mlen < 64 && mlen < *len) {
1519 put_page(p);
1520 goto new_page;
1523 *len = min_t(unsigned int, *len, mlen);
1526 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1527 sk->sk_sndmsg_off += *len;
1528 *offset = off;
1529 get_page(p);
1531 return p;
1535 * Fill page/offset/length into spd, if it can hold more pages.
1537 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1538 struct pipe_inode_info *pipe, struct page *page,
1539 unsigned int *len, unsigned int offset,
1540 struct sk_buff *skb, int linear,
1541 struct sock *sk)
1543 if (unlikely(spd->nr_pages == pipe->buffers))
1544 return 1;
1546 if (linear) {
1547 page = linear_to_page(page, len, &offset, skb, sk);
1548 if (!page)
1549 return 1;
1550 } else
1551 get_page(page);
1553 spd->pages[spd->nr_pages] = page;
1554 spd->partial[spd->nr_pages].len = *len;
1555 spd->partial[spd->nr_pages].offset = offset;
1556 spd->nr_pages++;
1558 return 0;
1561 static inline void __segment_seek(struct page **page, unsigned int *poff,
1562 unsigned int *plen, unsigned int off)
1564 unsigned long n;
1566 *poff += off;
1567 n = *poff / PAGE_SIZE;
1568 if (n)
1569 *page = nth_page(*page, n);
1571 *poff = *poff % PAGE_SIZE;
1572 *plen -= off;
1575 static inline int __splice_segment(struct page *page, unsigned int poff,
1576 unsigned int plen, unsigned int *off,
1577 unsigned int *len, struct sk_buff *skb,
1578 struct splice_pipe_desc *spd, int linear,
1579 struct sock *sk,
1580 struct pipe_inode_info *pipe)
1582 if (!*len)
1583 return 1;
1585 /* skip this segment if already processed */
1586 if (*off >= plen) {
1587 *off -= plen;
1588 return 0;
1591 /* ignore any bits we already processed */
1592 if (*off) {
1593 __segment_seek(&page, &poff, &plen, *off);
1594 *off = 0;
1597 do {
1598 unsigned int flen = min(*len, plen);
1600 /* the linear region may spread across several pages */
1601 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1603 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1604 return 1;
1606 __segment_seek(&page, &poff, &plen, flen);
1607 *len -= flen;
1609 } while (*len && plen);
1611 return 0;
1615 * Map linear and fragment data from the skb to spd. It reports failure if the
1616 * pipe is full or if we already spliced the requested length.
1618 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1619 unsigned int *offset, unsigned int *len,
1620 struct splice_pipe_desc *spd, struct sock *sk)
1622 int seg;
1625 * map the linear part
1627 if (__splice_segment(virt_to_page(skb->data),
1628 (unsigned long) skb->data & (PAGE_SIZE - 1),
1629 skb_headlen(skb),
1630 offset, len, skb, spd, 1, sk, pipe))
1631 return 1;
1634 * then map the fragments
1636 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1637 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1639 if (__splice_segment(skb_frag_page(f),
1640 f->page_offset, skb_frag_size(f),
1641 offset, len, skb, spd, 0, sk, pipe))
1642 return 1;
1645 return 0;
1649 * Map data from the skb to a pipe. Should handle both the linear part,
1650 * the fragments, and the frag list. It does NOT handle frag lists within
1651 * the frag list, if such a thing exists. We'd probably need to recurse to
1652 * handle that cleanly.
1654 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1655 struct pipe_inode_info *pipe, unsigned int tlen,
1656 unsigned int flags)
1658 struct partial_page partial[PIPE_DEF_BUFFERS];
1659 struct page *pages[PIPE_DEF_BUFFERS];
1660 struct splice_pipe_desc spd = {
1661 .pages = pages,
1662 .partial = partial,
1663 .flags = flags,
1664 .ops = &sock_pipe_buf_ops,
1665 .spd_release = sock_spd_release,
1667 struct sk_buff *frag_iter;
1668 struct sock *sk = skb->sk;
1669 int ret = 0;
1671 if (splice_grow_spd(pipe, &spd))
1672 return -ENOMEM;
1675 * __skb_splice_bits() only fails if the output has no room left,
1676 * so no point in going over the frag_list for the error case.
1678 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1679 goto done;
1680 else if (!tlen)
1681 goto done;
1684 * now see if we have a frag_list to map
1686 skb_walk_frags(skb, frag_iter) {
1687 if (!tlen)
1688 break;
1689 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1690 break;
1693 done:
1694 if (spd.nr_pages) {
1696 * Drop the socket lock, otherwise we have reverse
1697 * locking dependencies between sk_lock and i_mutex
1698 * here as compared to sendfile(). We enter here
1699 * with the socket lock held, and splice_to_pipe() will
1700 * grab the pipe inode lock. For sendfile() emulation,
1701 * we call into ->sendpage() with the i_mutex lock held
1702 * and networking will grab the socket lock.
1704 release_sock(sk);
1705 ret = splice_to_pipe(pipe, &spd);
1706 lock_sock(sk);
1709 splice_shrink_spd(pipe, &spd);
1710 return ret;
1714 * skb_store_bits - store bits from kernel buffer to skb
1715 * @skb: destination buffer
1716 * @offset: offset in destination
1717 * @from: source buffer
1718 * @len: number of bytes to copy
1720 * Copy the specified number of bytes from the source buffer to the
1721 * destination skb. This function handles all the messy bits of
1722 * traversing fragment lists and such.
1725 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1727 int start = skb_headlen(skb);
1728 struct sk_buff *frag_iter;
1729 int i, copy;
1731 if (offset > (int)skb->len - len)
1732 goto fault;
1734 if ((copy = start - offset) > 0) {
1735 if (copy > len)
1736 copy = len;
1737 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1738 if ((len -= copy) == 0)
1739 return 0;
1740 offset += copy;
1741 from += copy;
1744 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1745 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1746 int end;
1748 WARN_ON(start > offset + len);
1750 end = start + skb_frag_size(frag);
1751 if ((copy = end - offset) > 0) {
1752 u8 *vaddr;
1754 if (copy > len)
1755 copy = len;
1757 vaddr = kmap_skb_frag(frag);
1758 memcpy(vaddr + frag->page_offset + offset - start,
1759 from, copy);
1760 kunmap_skb_frag(vaddr);
1762 if ((len -= copy) == 0)
1763 return 0;
1764 offset += copy;
1765 from += copy;
1767 start = end;
1770 skb_walk_frags(skb, frag_iter) {
1771 int end;
1773 WARN_ON(start > offset + len);
1775 end = start + frag_iter->len;
1776 if ((copy = end - offset) > 0) {
1777 if (copy > len)
1778 copy = len;
1779 if (skb_store_bits(frag_iter, offset - start,
1780 from, copy))
1781 goto fault;
1782 if ((len -= copy) == 0)
1783 return 0;
1784 offset += copy;
1785 from += copy;
1787 start = end;
1789 if (!len)
1790 return 0;
1792 fault:
1793 return -EFAULT;
1795 EXPORT_SYMBOL(skb_store_bits);
1797 /* Checksum skb data. */
1799 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1800 int len, __wsum csum)
1802 int start = skb_headlen(skb);
1803 int i, copy = start - offset;
1804 struct sk_buff *frag_iter;
1805 int pos = 0;
1807 /* Checksum header. */
1808 if (copy > 0) {
1809 if (copy > len)
1810 copy = len;
1811 csum = csum_partial(skb->data + offset, copy, csum);
1812 if ((len -= copy) == 0)
1813 return csum;
1814 offset += copy;
1815 pos = copy;
1818 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1819 int end;
1821 WARN_ON(start > offset + len);
1823 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1824 if ((copy = end - offset) > 0) {
1825 __wsum csum2;
1826 u8 *vaddr;
1827 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1829 if (copy > len)
1830 copy = len;
1831 vaddr = kmap_skb_frag(frag);
1832 csum2 = csum_partial(vaddr + frag->page_offset +
1833 offset - start, copy, 0);
1834 kunmap_skb_frag(vaddr);
1835 csum = csum_block_add(csum, csum2, pos);
1836 if (!(len -= copy))
1837 return csum;
1838 offset += copy;
1839 pos += copy;
1841 start = end;
1844 skb_walk_frags(skb, frag_iter) {
1845 int end;
1847 WARN_ON(start > offset + len);
1849 end = start + frag_iter->len;
1850 if ((copy = end - offset) > 0) {
1851 __wsum csum2;
1852 if (copy > len)
1853 copy = len;
1854 csum2 = skb_checksum(frag_iter, offset - start,
1855 copy, 0);
1856 csum = csum_block_add(csum, csum2, pos);
1857 if ((len -= copy) == 0)
1858 return csum;
1859 offset += copy;
1860 pos += copy;
1862 start = end;
1864 BUG_ON(len);
1866 return csum;
1868 EXPORT_SYMBOL(skb_checksum);
1870 /* Both of above in one bottle. */
1872 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1873 u8 *to, int len, __wsum csum)
1875 int start = skb_headlen(skb);
1876 int i, copy = start - offset;
1877 struct sk_buff *frag_iter;
1878 int pos = 0;
1880 /* Copy header. */
1881 if (copy > 0) {
1882 if (copy > len)
1883 copy = len;
1884 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1885 copy, csum);
1886 if ((len -= copy) == 0)
1887 return csum;
1888 offset += copy;
1889 to += copy;
1890 pos = copy;
1893 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1894 int end;
1896 WARN_ON(start > offset + len);
1898 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1899 if ((copy = end - offset) > 0) {
1900 __wsum csum2;
1901 u8 *vaddr;
1902 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1904 if (copy > len)
1905 copy = len;
1906 vaddr = kmap_skb_frag(frag);
1907 csum2 = csum_partial_copy_nocheck(vaddr +
1908 frag->page_offset +
1909 offset - start, to,
1910 copy, 0);
1911 kunmap_skb_frag(vaddr);
1912 csum = csum_block_add(csum, csum2, pos);
1913 if (!(len -= copy))
1914 return csum;
1915 offset += copy;
1916 to += copy;
1917 pos += copy;
1919 start = end;
1922 skb_walk_frags(skb, frag_iter) {
1923 __wsum csum2;
1924 int end;
1926 WARN_ON(start > offset + len);
1928 end = start + frag_iter->len;
1929 if ((copy = end - offset) > 0) {
1930 if (copy > len)
1931 copy = len;
1932 csum2 = skb_copy_and_csum_bits(frag_iter,
1933 offset - start,
1934 to, copy, 0);
1935 csum = csum_block_add(csum, csum2, pos);
1936 if ((len -= copy) == 0)
1937 return csum;
1938 offset += copy;
1939 to += copy;
1940 pos += copy;
1942 start = end;
1944 BUG_ON(len);
1945 return csum;
1947 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1949 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1951 __wsum csum;
1952 long csstart;
1954 if (skb->ip_summed == CHECKSUM_PARTIAL)
1955 csstart = skb_checksum_start_offset(skb);
1956 else
1957 csstart = skb_headlen(skb);
1959 BUG_ON(csstart > skb_headlen(skb));
1961 skb_copy_from_linear_data(skb, to, csstart);
1963 csum = 0;
1964 if (csstart != skb->len)
1965 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1966 skb->len - csstart, 0);
1968 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1969 long csstuff = csstart + skb->csum_offset;
1971 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1974 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1977 * skb_dequeue - remove from the head of the queue
1978 * @list: list to dequeue from
1980 * Remove the head of the list. The list lock is taken so the function
1981 * may be used safely with other locking list functions. The head item is
1982 * returned or %NULL if the list is empty.
1985 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1987 unsigned long flags;
1988 struct sk_buff *result;
1990 spin_lock_irqsave(&list->lock, flags);
1991 result = __skb_dequeue(list);
1992 spin_unlock_irqrestore(&list->lock, flags);
1993 return result;
1995 EXPORT_SYMBOL(skb_dequeue);
1998 * skb_dequeue_tail - remove from the tail of the queue
1999 * @list: list to dequeue from
2001 * Remove the tail of the list. The list lock is taken so the function
2002 * may be used safely with other locking list functions. The tail item is
2003 * returned or %NULL if the list is empty.
2005 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2007 unsigned long flags;
2008 struct sk_buff *result;
2010 spin_lock_irqsave(&list->lock, flags);
2011 result = __skb_dequeue_tail(list);
2012 spin_unlock_irqrestore(&list->lock, flags);
2013 return result;
2015 EXPORT_SYMBOL(skb_dequeue_tail);
2018 * skb_queue_purge - empty a list
2019 * @list: list to empty
2021 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2022 * the list and one reference dropped. This function takes the list
2023 * lock and is atomic with respect to other list locking functions.
2025 void skb_queue_purge(struct sk_buff_head *list)
2027 struct sk_buff *skb;
2028 while ((skb = skb_dequeue(list)) != NULL)
2029 kfree_skb(skb);
2031 EXPORT_SYMBOL(skb_queue_purge);
2034 * skb_queue_head - queue a buffer at the list head
2035 * @list: list to use
2036 * @newsk: buffer to queue
2038 * Queue a buffer at the start of the list. This function takes the
2039 * list lock and can be used safely with other locking &sk_buff functions
2040 * safely.
2042 * A buffer cannot be placed on two lists at the same time.
2044 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2046 unsigned long flags;
2048 spin_lock_irqsave(&list->lock, flags);
2049 __skb_queue_head(list, newsk);
2050 spin_unlock_irqrestore(&list->lock, flags);
2052 EXPORT_SYMBOL(skb_queue_head);
2055 * skb_queue_tail - queue a buffer at the list tail
2056 * @list: list to use
2057 * @newsk: buffer to queue
2059 * Queue a buffer at the tail of the list. This function takes the
2060 * list lock and can be used safely with other locking &sk_buff functions
2061 * safely.
2063 * A buffer cannot be placed on two lists at the same time.
2065 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2067 unsigned long flags;
2069 spin_lock_irqsave(&list->lock, flags);
2070 __skb_queue_tail(list, newsk);
2071 spin_unlock_irqrestore(&list->lock, flags);
2073 EXPORT_SYMBOL(skb_queue_tail);
2076 * skb_unlink - remove a buffer from a list
2077 * @skb: buffer to remove
2078 * @list: list to use
2080 * Remove a packet from a list. The list locks are taken and this
2081 * function is atomic with respect to other list locked calls
2083 * You must know what list the SKB is on.
2085 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2087 unsigned long flags;
2089 spin_lock_irqsave(&list->lock, flags);
2090 __skb_unlink(skb, list);
2091 spin_unlock_irqrestore(&list->lock, flags);
2093 EXPORT_SYMBOL(skb_unlink);
2096 * skb_append - append a buffer
2097 * @old: buffer to insert after
2098 * @newsk: buffer to insert
2099 * @list: list to use
2101 * Place a packet after a given packet in a list. The list locks are taken
2102 * and this function is atomic with respect to other list locked calls.
2103 * A buffer cannot be placed on two lists at the same time.
2105 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2107 unsigned long flags;
2109 spin_lock_irqsave(&list->lock, flags);
2110 __skb_queue_after(list, old, newsk);
2111 spin_unlock_irqrestore(&list->lock, flags);
2113 EXPORT_SYMBOL(skb_append);
2116 * skb_insert - insert a buffer
2117 * @old: buffer to insert before
2118 * @newsk: buffer to insert
2119 * @list: list to use
2121 * Place a packet before a given packet in a list. The list locks are
2122 * taken and this function is atomic with respect to other list locked
2123 * calls.
2125 * A buffer cannot be placed on two lists at the same time.
2127 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2129 unsigned long flags;
2131 spin_lock_irqsave(&list->lock, flags);
2132 __skb_insert(newsk, old->prev, old, list);
2133 spin_unlock_irqrestore(&list->lock, flags);
2135 EXPORT_SYMBOL(skb_insert);
2137 static inline void skb_split_inside_header(struct sk_buff *skb,
2138 struct sk_buff* skb1,
2139 const u32 len, const int pos)
2141 int i;
2143 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2144 pos - len);
2145 /* And move data appendix as is. */
2146 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2147 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2149 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2150 skb_shinfo(skb)->nr_frags = 0;
2151 skb1->data_len = skb->data_len;
2152 skb1->len += skb1->data_len;
2153 skb->data_len = 0;
2154 skb->len = len;
2155 skb_set_tail_pointer(skb, len);
2158 static inline void skb_split_no_header(struct sk_buff *skb,
2159 struct sk_buff* skb1,
2160 const u32 len, int pos)
2162 int i, k = 0;
2163 const int nfrags = skb_shinfo(skb)->nr_frags;
2165 skb_shinfo(skb)->nr_frags = 0;
2166 skb1->len = skb1->data_len = skb->len - len;
2167 skb->len = len;
2168 skb->data_len = len - pos;
2170 for (i = 0; i < nfrags; i++) {
2171 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2173 if (pos + size > len) {
2174 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2176 if (pos < len) {
2177 /* Split frag.
2178 * We have two variants in this case:
2179 * 1. Move all the frag to the second
2180 * part, if it is possible. F.e.
2181 * this approach is mandatory for TUX,
2182 * where splitting is expensive.
2183 * 2. Split is accurately. We make this.
2185 skb_frag_ref(skb, i);
2186 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2187 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2188 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2189 skb_shinfo(skb)->nr_frags++;
2191 k++;
2192 } else
2193 skb_shinfo(skb)->nr_frags++;
2194 pos += size;
2196 skb_shinfo(skb1)->nr_frags = k;
2200 * skb_split - Split fragmented skb to two parts at length len.
2201 * @skb: the buffer to split
2202 * @skb1: the buffer to receive the second part
2203 * @len: new length for skb
2205 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2207 int pos = skb_headlen(skb);
2209 if (len < pos) /* Split line is inside header. */
2210 skb_split_inside_header(skb, skb1, len, pos);
2211 else /* Second chunk has no header, nothing to copy. */
2212 skb_split_no_header(skb, skb1, len, pos);
2214 EXPORT_SYMBOL(skb_split);
2216 /* Shifting from/to a cloned skb is a no-go.
2218 * Caller cannot keep skb_shinfo related pointers past calling here!
2220 static int skb_prepare_for_shift(struct sk_buff *skb)
2222 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2226 * skb_shift - Shifts paged data partially from skb to another
2227 * @tgt: buffer into which tail data gets added
2228 * @skb: buffer from which the paged data comes from
2229 * @shiftlen: shift up to this many bytes
2231 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2232 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2233 * It's up to caller to free skb if everything was shifted.
2235 * If @tgt runs out of frags, the whole operation is aborted.
2237 * Skb cannot include anything else but paged data while tgt is allowed
2238 * to have non-paged data as well.
2240 * TODO: full sized shift could be optimized but that would need
2241 * specialized skb free'er to handle frags without up-to-date nr_frags.
2243 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2245 int from, to, merge, todo;
2246 struct skb_frag_struct *fragfrom, *fragto;
2248 BUG_ON(shiftlen > skb->len);
2249 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2251 todo = shiftlen;
2252 from = 0;
2253 to = skb_shinfo(tgt)->nr_frags;
2254 fragfrom = &skb_shinfo(skb)->frags[from];
2256 /* Actual merge is delayed until the point when we know we can
2257 * commit all, so that we don't have to undo partial changes
2259 if (!to ||
2260 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2261 fragfrom->page_offset)) {
2262 merge = -1;
2263 } else {
2264 merge = to - 1;
2266 todo -= skb_frag_size(fragfrom);
2267 if (todo < 0) {
2268 if (skb_prepare_for_shift(skb) ||
2269 skb_prepare_for_shift(tgt))
2270 return 0;
2272 /* All previous frag pointers might be stale! */
2273 fragfrom = &skb_shinfo(skb)->frags[from];
2274 fragto = &skb_shinfo(tgt)->frags[merge];
2276 skb_frag_size_add(fragto, shiftlen);
2277 skb_frag_size_sub(fragfrom, shiftlen);
2278 fragfrom->page_offset += shiftlen;
2280 goto onlymerged;
2283 from++;
2286 /* Skip full, not-fitting skb to avoid expensive operations */
2287 if ((shiftlen == skb->len) &&
2288 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2289 return 0;
2291 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2292 return 0;
2294 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2295 if (to == MAX_SKB_FRAGS)
2296 return 0;
2298 fragfrom = &skb_shinfo(skb)->frags[from];
2299 fragto = &skb_shinfo(tgt)->frags[to];
2301 if (todo >= skb_frag_size(fragfrom)) {
2302 *fragto = *fragfrom;
2303 todo -= skb_frag_size(fragfrom);
2304 from++;
2305 to++;
2307 } else {
2308 __skb_frag_ref(fragfrom);
2309 fragto->page = fragfrom->page;
2310 fragto->page_offset = fragfrom->page_offset;
2311 skb_frag_size_set(fragto, todo);
2313 fragfrom->page_offset += todo;
2314 skb_frag_size_sub(fragfrom, todo);
2315 todo = 0;
2317 to++;
2318 break;
2322 /* Ready to "commit" this state change to tgt */
2323 skb_shinfo(tgt)->nr_frags = to;
2325 if (merge >= 0) {
2326 fragfrom = &skb_shinfo(skb)->frags[0];
2327 fragto = &skb_shinfo(tgt)->frags[merge];
2329 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2330 __skb_frag_unref(fragfrom);
2333 /* Reposition in the original skb */
2334 to = 0;
2335 while (from < skb_shinfo(skb)->nr_frags)
2336 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2337 skb_shinfo(skb)->nr_frags = to;
2339 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2341 onlymerged:
2342 /* Most likely the tgt won't ever need its checksum anymore, skb on
2343 * the other hand might need it if it needs to be resent
2345 tgt->ip_summed = CHECKSUM_PARTIAL;
2346 skb->ip_summed = CHECKSUM_PARTIAL;
2348 /* Yak, is it really working this way? Some helper please? */
2349 skb->len -= shiftlen;
2350 skb->data_len -= shiftlen;
2351 skb->truesize -= shiftlen;
2352 tgt->len += shiftlen;
2353 tgt->data_len += shiftlen;
2354 tgt->truesize += shiftlen;
2356 return shiftlen;
2360 * skb_prepare_seq_read - Prepare a sequential read of skb data
2361 * @skb: the buffer to read
2362 * @from: lower offset of data to be read
2363 * @to: upper offset of data to be read
2364 * @st: state variable
2366 * Initializes the specified state variable. Must be called before
2367 * invoking skb_seq_read() for the first time.
2369 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2370 unsigned int to, struct skb_seq_state *st)
2372 st->lower_offset = from;
2373 st->upper_offset = to;
2374 st->root_skb = st->cur_skb = skb;
2375 st->frag_idx = st->stepped_offset = 0;
2376 st->frag_data = NULL;
2378 EXPORT_SYMBOL(skb_prepare_seq_read);
2381 * skb_seq_read - Sequentially read skb data
2382 * @consumed: number of bytes consumed by the caller so far
2383 * @data: destination pointer for data to be returned
2384 * @st: state variable
2386 * Reads a block of skb data at &consumed relative to the
2387 * lower offset specified to skb_prepare_seq_read(). Assigns
2388 * the head of the data block to &data and returns the length
2389 * of the block or 0 if the end of the skb data or the upper
2390 * offset has been reached.
2392 * The caller is not required to consume all of the data
2393 * returned, i.e. &consumed is typically set to the number
2394 * of bytes already consumed and the next call to
2395 * skb_seq_read() will return the remaining part of the block.
2397 * Note 1: The size of each block of data returned can be arbitrary,
2398 * this limitation is the cost for zerocopy seqeuental
2399 * reads of potentially non linear data.
2401 * Note 2: Fragment lists within fragments are not implemented
2402 * at the moment, state->root_skb could be replaced with
2403 * a stack for this purpose.
2405 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2406 struct skb_seq_state *st)
2408 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2409 skb_frag_t *frag;
2411 if (unlikely(abs_offset >= st->upper_offset))
2412 return 0;
2414 next_skb:
2415 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2417 if (abs_offset < block_limit && !st->frag_data) {
2418 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2419 return block_limit - abs_offset;
2422 if (st->frag_idx == 0 && !st->frag_data)
2423 st->stepped_offset += skb_headlen(st->cur_skb);
2425 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2426 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2427 block_limit = skb_frag_size(frag) + st->stepped_offset;
2429 if (abs_offset < block_limit) {
2430 if (!st->frag_data)
2431 st->frag_data = kmap_skb_frag(frag);
2433 *data = (u8 *) st->frag_data + frag->page_offset +
2434 (abs_offset - st->stepped_offset);
2436 return block_limit - abs_offset;
2439 if (st->frag_data) {
2440 kunmap_skb_frag(st->frag_data);
2441 st->frag_data = NULL;
2444 st->frag_idx++;
2445 st->stepped_offset += skb_frag_size(frag);
2448 if (st->frag_data) {
2449 kunmap_skb_frag(st->frag_data);
2450 st->frag_data = NULL;
2453 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2454 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2455 st->frag_idx = 0;
2456 goto next_skb;
2457 } else if (st->cur_skb->next) {
2458 st->cur_skb = st->cur_skb->next;
2459 st->frag_idx = 0;
2460 goto next_skb;
2463 return 0;
2465 EXPORT_SYMBOL(skb_seq_read);
2468 * skb_abort_seq_read - Abort a sequential read of skb data
2469 * @st: state variable
2471 * Must be called if skb_seq_read() was not called until it
2472 * returned 0.
2474 void skb_abort_seq_read(struct skb_seq_state *st)
2476 if (st->frag_data)
2477 kunmap_skb_frag(st->frag_data);
2479 EXPORT_SYMBOL(skb_abort_seq_read);
2481 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2483 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2484 struct ts_config *conf,
2485 struct ts_state *state)
2487 return skb_seq_read(offset, text, TS_SKB_CB(state));
2490 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2492 skb_abort_seq_read(TS_SKB_CB(state));
2496 * skb_find_text - Find a text pattern in skb data
2497 * @skb: the buffer to look in
2498 * @from: search offset
2499 * @to: search limit
2500 * @config: textsearch configuration
2501 * @state: uninitialized textsearch state variable
2503 * Finds a pattern in the skb data according to the specified
2504 * textsearch configuration. Use textsearch_next() to retrieve
2505 * subsequent occurrences of the pattern. Returns the offset
2506 * to the first occurrence or UINT_MAX if no match was found.
2508 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2509 unsigned int to, struct ts_config *config,
2510 struct ts_state *state)
2512 unsigned int ret;
2514 config->get_next_block = skb_ts_get_next_block;
2515 config->finish = skb_ts_finish;
2517 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2519 ret = textsearch_find(config, state);
2520 return (ret <= to - from ? ret : UINT_MAX);
2522 EXPORT_SYMBOL(skb_find_text);
2525 * skb_append_datato_frags: - append the user data to a skb
2526 * @sk: sock structure
2527 * @skb: skb structure to be appened with user data.
2528 * @getfrag: call back function to be used for getting the user data
2529 * @from: pointer to user message iov
2530 * @length: length of the iov message
2532 * Description: This procedure append the user data in the fragment part
2533 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2535 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2536 int (*getfrag)(void *from, char *to, int offset,
2537 int len, int odd, struct sk_buff *skb),
2538 void *from, int length)
2540 int frg_cnt = 0;
2541 skb_frag_t *frag = NULL;
2542 struct page *page = NULL;
2543 int copy, left;
2544 int offset = 0;
2545 int ret;
2547 do {
2548 /* Return error if we don't have space for new frag */
2549 frg_cnt = skb_shinfo(skb)->nr_frags;
2550 if (frg_cnt >= MAX_SKB_FRAGS)
2551 return -EFAULT;
2553 /* allocate a new page for next frag */
2554 page = alloc_pages(sk->sk_allocation, 0);
2556 /* If alloc_page fails just return failure and caller will
2557 * free previous allocated pages by doing kfree_skb()
2559 if (page == NULL)
2560 return -ENOMEM;
2562 /* initialize the next frag */
2563 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2564 skb->truesize += PAGE_SIZE;
2565 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2567 /* get the new initialized frag */
2568 frg_cnt = skb_shinfo(skb)->nr_frags;
2569 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2571 /* copy the user data to page */
2572 left = PAGE_SIZE - frag->page_offset;
2573 copy = (length > left)? left : length;
2575 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2576 offset, copy, 0, skb);
2577 if (ret < 0)
2578 return -EFAULT;
2580 /* copy was successful so update the size parameters */
2581 skb_frag_size_add(frag, copy);
2582 skb->len += copy;
2583 skb->data_len += copy;
2584 offset += copy;
2585 length -= copy;
2587 } while (length > 0);
2589 return 0;
2591 EXPORT_SYMBOL(skb_append_datato_frags);
2594 * skb_pull_rcsum - pull skb and update receive checksum
2595 * @skb: buffer to update
2596 * @len: length of data pulled
2598 * This function performs an skb_pull on the packet and updates
2599 * the CHECKSUM_COMPLETE checksum. It should be used on
2600 * receive path processing instead of skb_pull unless you know
2601 * that the checksum difference is zero (e.g., a valid IP header)
2602 * or you are setting ip_summed to CHECKSUM_NONE.
2604 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2606 BUG_ON(len > skb->len);
2607 skb->len -= len;
2608 BUG_ON(skb->len < skb->data_len);
2609 skb_postpull_rcsum(skb, skb->data, len);
2610 return skb->data += len;
2612 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2615 * skb_segment - Perform protocol segmentation on skb.
2616 * @skb: buffer to segment
2617 * @features: features for the output path (see dev->features)
2619 * This function performs segmentation on the given skb. It returns
2620 * a pointer to the first in a list of new skbs for the segments.
2621 * In case of error it returns ERR_PTR(err).
2623 struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2625 struct sk_buff *segs = NULL;
2626 struct sk_buff *tail = NULL;
2627 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2628 unsigned int mss = skb_shinfo(skb)->gso_size;
2629 unsigned int doffset = skb->data - skb_mac_header(skb);
2630 unsigned int offset = doffset;
2631 unsigned int headroom;
2632 unsigned int len;
2633 int sg = !!(features & NETIF_F_SG);
2634 int nfrags = skb_shinfo(skb)->nr_frags;
2635 int err = -ENOMEM;
2636 int i = 0;
2637 int pos;
2639 __skb_push(skb, doffset);
2640 headroom = skb_headroom(skb);
2641 pos = skb_headlen(skb);
2643 do {
2644 struct sk_buff *nskb;
2645 skb_frag_t *frag;
2646 int hsize;
2647 int size;
2649 len = skb->len - offset;
2650 if (len > mss)
2651 len = mss;
2653 hsize = skb_headlen(skb) - offset;
2654 if (hsize < 0)
2655 hsize = 0;
2656 if (hsize > len || !sg)
2657 hsize = len;
2659 if (!hsize && i >= nfrags) {
2660 BUG_ON(fskb->len != len);
2662 pos += len;
2663 nskb = skb_clone(fskb, GFP_ATOMIC);
2664 fskb = fskb->next;
2666 if (unlikely(!nskb))
2667 goto err;
2669 hsize = skb_end_pointer(nskb) - nskb->head;
2670 if (skb_cow_head(nskb, doffset + headroom)) {
2671 kfree_skb(nskb);
2672 goto err;
2675 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2676 hsize;
2677 skb_release_head_state(nskb);
2678 __skb_push(nskb, doffset);
2679 } else {
2680 nskb = alloc_skb(hsize + doffset + headroom,
2681 GFP_ATOMIC);
2683 if (unlikely(!nskb))
2684 goto err;
2686 skb_reserve(nskb, headroom);
2687 __skb_put(nskb, doffset);
2690 if (segs)
2691 tail->next = nskb;
2692 else
2693 segs = nskb;
2694 tail = nskb;
2696 __copy_skb_header(nskb, skb);
2697 nskb->mac_len = skb->mac_len;
2699 /* nskb and skb might have different headroom */
2700 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2701 nskb->csum_start += skb_headroom(nskb) - headroom;
2703 skb_reset_mac_header(nskb);
2704 skb_set_network_header(nskb, skb->mac_len);
2705 nskb->transport_header = (nskb->network_header +
2706 skb_network_header_len(skb));
2707 skb_copy_from_linear_data(skb, nskb->data, doffset);
2709 if (fskb != skb_shinfo(skb)->frag_list)
2710 continue;
2712 if (!sg) {
2713 nskb->ip_summed = CHECKSUM_NONE;
2714 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2715 skb_put(nskb, len),
2716 len, 0);
2717 continue;
2720 frag = skb_shinfo(nskb)->frags;
2722 skb_copy_from_linear_data_offset(skb, offset,
2723 skb_put(nskb, hsize), hsize);
2725 while (pos < offset + len && i < nfrags) {
2726 *frag = skb_shinfo(skb)->frags[i];
2727 __skb_frag_ref(frag);
2728 size = skb_frag_size(frag);
2730 if (pos < offset) {
2731 frag->page_offset += offset - pos;
2732 skb_frag_size_sub(frag, offset - pos);
2735 skb_shinfo(nskb)->nr_frags++;
2737 if (pos + size <= offset + len) {
2738 i++;
2739 pos += size;
2740 } else {
2741 skb_frag_size_sub(frag, pos + size - (offset + len));
2742 goto skip_fraglist;
2745 frag++;
2748 if (pos < offset + len) {
2749 struct sk_buff *fskb2 = fskb;
2751 BUG_ON(pos + fskb->len != offset + len);
2753 pos += fskb->len;
2754 fskb = fskb->next;
2756 if (fskb2->next) {
2757 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2758 if (!fskb2)
2759 goto err;
2760 } else
2761 skb_get(fskb2);
2763 SKB_FRAG_ASSERT(nskb);
2764 skb_shinfo(nskb)->frag_list = fskb2;
2767 skip_fraglist:
2768 nskb->data_len = len - hsize;
2769 nskb->len += nskb->data_len;
2770 nskb->truesize += nskb->data_len;
2771 } while ((offset += len) < skb->len);
2773 return segs;
2775 err:
2776 while ((skb = segs)) {
2777 segs = skb->next;
2778 kfree_skb(skb);
2780 return ERR_PTR(err);
2782 EXPORT_SYMBOL_GPL(skb_segment);
2784 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2786 struct sk_buff *p = *head;
2787 struct sk_buff *nskb;
2788 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2789 struct skb_shared_info *pinfo = skb_shinfo(p);
2790 unsigned int headroom;
2791 unsigned int len = skb_gro_len(skb);
2792 unsigned int offset = skb_gro_offset(skb);
2793 unsigned int headlen = skb_headlen(skb);
2795 if (p->len + len >= 65536)
2796 return -E2BIG;
2798 if (pinfo->frag_list)
2799 goto merge;
2800 else if (headlen <= offset) {
2801 skb_frag_t *frag;
2802 skb_frag_t *frag2;
2803 int i = skbinfo->nr_frags;
2804 int nr_frags = pinfo->nr_frags + i;
2806 offset -= headlen;
2808 if (nr_frags > MAX_SKB_FRAGS)
2809 return -E2BIG;
2811 pinfo->nr_frags = nr_frags;
2812 skbinfo->nr_frags = 0;
2814 frag = pinfo->frags + nr_frags;
2815 frag2 = skbinfo->frags + i;
2816 do {
2817 *--frag = *--frag2;
2818 } while (--i);
2820 frag->page_offset += offset;
2821 skb_frag_size_sub(frag, offset);
2823 skb->truesize -= skb->data_len;
2824 skb->len -= skb->data_len;
2825 skb->data_len = 0;
2827 NAPI_GRO_CB(skb)->free = 1;
2828 goto done;
2829 } else if (skb_gro_len(p) != pinfo->gso_size)
2830 return -E2BIG;
2832 headroom = skb_headroom(p);
2833 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2834 if (unlikely(!nskb))
2835 return -ENOMEM;
2837 __copy_skb_header(nskb, p);
2838 nskb->mac_len = p->mac_len;
2840 skb_reserve(nskb, headroom);
2841 __skb_put(nskb, skb_gro_offset(p));
2843 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2844 skb_set_network_header(nskb, skb_network_offset(p));
2845 skb_set_transport_header(nskb, skb_transport_offset(p));
2847 __skb_pull(p, skb_gro_offset(p));
2848 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2849 p->data - skb_mac_header(p));
2851 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2852 skb_shinfo(nskb)->frag_list = p;
2853 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2854 pinfo->gso_size = 0;
2855 skb_header_release(p);
2856 nskb->prev = p;
2858 nskb->data_len += p->len;
2859 nskb->truesize += p->len;
2860 nskb->len += p->len;
2862 *head = nskb;
2863 nskb->next = p->next;
2864 p->next = NULL;
2866 p = nskb;
2868 merge:
2869 if (offset > headlen) {
2870 unsigned int eat = offset - headlen;
2872 skbinfo->frags[0].page_offset += eat;
2873 skb_frag_size_sub(&skbinfo->frags[0], eat);
2874 skb->data_len -= eat;
2875 skb->len -= eat;
2876 offset = headlen;
2879 __skb_pull(skb, offset);
2881 p->prev->next = skb;
2882 p->prev = skb;
2883 skb_header_release(skb);
2885 done:
2886 NAPI_GRO_CB(p)->count++;
2887 p->data_len += len;
2888 p->truesize += len;
2889 p->len += len;
2891 NAPI_GRO_CB(skb)->same_flow = 1;
2892 return 0;
2894 EXPORT_SYMBOL_GPL(skb_gro_receive);
2896 void __init skb_init(void)
2898 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2899 sizeof(struct sk_buff),
2901 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2902 NULL);
2903 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2904 (2*sizeof(struct sk_buff)) +
2905 sizeof(atomic_t),
2907 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2908 NULL);
2912 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2913 * @skb: Socket buffer containing the buffers to be mapped
2914 * @sg: The scatter-gather list to map into
2915 * @offset: The offset into the buffer's contents to start mapping
2916 * @len: Length of buffer space to be mapped
2918 * Fill the specified scatter-gather list with mappings/pointers into a
2919 * region of the buffer space attached to a socket buffer.
2921 static int
2922 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2924 int start = skb_headlen(skb);
2925 int i, copy = start - offset;
2926 struct sk_buff *frag_iter;
2927 int elt = 0;
2929 if (copy > 0) {
2930 if (copy > len)
2931 copy = len;
2932 sg_set_buf(sg, skb->data + offset, copy);
2933 elt++;
2934 if ((len -= copy) == 0)
2935 return elt;
2936 offset += copy;
2939 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2940 int end;
2942 WARN_ON(start > offset + len);
2944 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2945 if ((copy = end - offset) > 0) {
2946 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2948 if (copy > len)
2949 copy = len;
2950 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
2951 frag->page_offset+offset-start);
2952 elt++;
2953 if (!(len -= copy))
2954 return elt;
2955 offset += copy;
2957 start = end;
2960 skb_walk_frags(skb, frag_iter) {
2961 int end;
2963 WARN_ON(start > offset + len);
2965 end = start + frag_iter->len;
2966 if ((copy = end - offset) > 0) {
2967 if (copy > len)
2968 copy = len;
2969 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2970 copy);
2971 if ((len -= copy) == 0)
2972 return elt;
2973 offset += copy;
2975 start = end;
2977 BUG_ON(len);
2978 return elt;
2981 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2983 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2985 sg_mark_end(&sg[nsg - 1]);
2987 return nsg;
2989 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2992 * skb_cow_data - Check that a socket buffer's data buffers are writable
2993 * @skb: The socket buffer to check.
2994 * @tailbits: Amount of trailing space to be added
2995 * @trailer: Returned pointer to the skb where the @tailbits space begins
2997 * Make sure that the data buffers attached to a socket buffer are
2998 * writable. If they are not, private copies are made of the data buffers
2999 * and the socket buffer is set to use these instead.
3001 * If @tailbits is given, make sure that there is space to write @tailbits
3002 * bytes of data beyond current end of socket buffer. @trailer will be
3003 * set to point to the skb in which this space begins.
3005 * The number of scatterlist elements required to completely map the
3006 * COW'd and extended socket buffer will be returned.
3008 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3010 int copyflag;
3011 int elt;
3012 struct sk_buff *skb1, **skb_p;
3014 /* If skb is cloned or its head is paged, reallocate
3015 * head pulling out all the pages (pages are considered not writable
3016 * at the moment even if they are anonymous).
3018 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3019 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3020 return -ENOMEM;
3022 /* Easy case. Most of packets will go this way. */
3023 if (!skb_has_frag_list(skb)) {
3024 /* A little of trouble, not enough of space for trailer.
3025 * This should not happen, when stack is tuned to generate
3026 * good frames. OK, on miss we reallocate and reserve even more
3027 * space, 128 bytes is fair. */
3029 if (skb_tailroom(skb) < tailbits &&
3030 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3031 return -ENOMEM;
3033 /* Voila! */
3034 *trailer = skb;
3035 return 1;
3038 /* Misery. We are in troubles, going to mincer fragments... */
3040 elt = 1;
3041 skb_p = &skb_shinfo(skb)->frag_list;
3042 copyflag = 0;
3044 while ((skb1 = *skb_p) != NULL) {
3045 int ntail = 0;
3047 /* The fragment is partially pulled by someone,
3048 * this can happen on input. Copy it and everything
3049 * after it. */
3051 if (skb_shared(skb1))
3052 copyflag = 1;
3054 /* If the skb is the last, worry about trailer. */
3056 if (skb1->next == NULL && tailbits) {
3057 if (skb_shinfo(skb1)->nr_frags ||
3058 skb_has_frag_list(skb1) ||
3059 skb_tailroom(skb1) < tailbits)
3060 ntail = tailbits + 128;
3063 if (copyflag ||
3064 skb_cloned(skb1) ||
3065 ntail ||
3066 skb_shinfo(skb1)->nr_frags ||
3067 skb_has_frag_list(skb1)) {
3068 struct sk_buff *skb2;
3070 /* Fuck, we are miserable poor guys... */
3071 if (ntail == 0)
3072 skb2 = skb_copy(skb1, GFP_ATOMIC);
3073 else
3074 skb2 = skb_copy_expand(skb1,
3075 skb_headroom(skb1),
3076 ntail,
3077 GFP_ATOMIC);
3078 if (unlikely(skb2 == NULL))
3079 return -ENOMEM;
3081 if (skb1->sk)
3082 skb_set_owner_w(skb2, skb1->sk);
3084 /* Looking around. Are we still alive?
3085 * OK, link new skb, drop old one */
3087 skb2->next = skb1->next;
3088 *skb_p = skb2;
3089 kfree_skb(skb1);
3090 skb1 = skb2;
3092 elt++;
3093 *trailer = skb1;
3094 skb_p = &skb1->next;
3097 return elt;
3099 EXPORT_SYMBOL_GPL(skb_cow_data);
3101 static void sock_rmem_free(struct sk_buff *skb)
3103 struct sock *sk = skb->sk;
3105 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3109 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3111 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3113 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3114 (unsigned)sk->sk_rcvbuf)
3115 return -ENOMEM;
3117 skb_orphan(skb);
3118 skb->sk = sk;
3119 skb->destructor = sock_rmem_free;
3120 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3122 /* before exiting rcu section, make sure dst is refcounted */
3123 skb_dst_force(skb);
3125 skb_queue_tail(&sk->sk_error_queue, skb);
3126 if (!sock_flag(sk, SOCK_DEAD))
3127 sk->sk_data_ready(sk, skb->len);
3128 return 0;
3130 EXPORT_SYMBOL(sock_queue_err_skb);
3132 void skb_tstamp_tx(struct sk_buff *orig_skb,
3133 struct skb_shared_hwtstamps *hwtstamps)
3135 struct sock *sk = orig_skb->sk;
3136 struct sock_exterr_skb *serr;
3137 struct sk_buff *skb;
3138 int err;
3140 if (!sk)
3141 return;
3143 skb = skb_clone(orig_skb, GFP_ATOMIC);
3144 if (!skb)
3145 return;
3147 if (hwtstamps) {
3148 *skb_hwtstamps(skb) =
3149 *hwtstamps;
3150 } else {
3152 * no hardware time stamps available,
3153 * so keep the shared tx_flags and only
3154 * store software time stamp
3156 skb->tstamp = ktime_get_real();
3159 serr = SKB_EXT_ERR(skb);
3160 memset(serr, 0, sizeof(*serr));
3161 serr->ee.ee_errno = ENOMSG;
3162 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3164 err = sock_queue_err_skb(sk, skb);
3166 if (err)
3167 kfree_skb(skb);
3169 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3173 * skb_partial_csum_set - set up and verify partial csum values for packet
3174 * @skb: the skb to set
3175 * @start: the number of bytes after skb->data to start checksumming.
3176 * @off: the offset from start to place the checksum.
3178 * For untrusted partially-checksummed packets, we need to make sure the values
3179 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3181 * This function checks and sets those values and skb->ip_summed: if this
3182 * returns false you should drop the packet.
3184 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3186 if (unlikely(start > skb_headlen(skb)) ||
3187 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3188 if (net_ratelimit())
3189 printk(KERN_WARNING
3190 "bad partial csum: csum=%u/%u len=%u\n",
3191 start, off, skb_headlen(skb));
3192 return false;
3194 skb->ip_summed = CHECKSUM_PARTIAL;
3195 skb->csum_start = skb_headroom(skb) + start;
3196 skb->csum_offset = off;
3197 return true;
3199 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3201 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3203 if (net_ratelimit())
3204 pr_warning("%s: received packets cannot be forwarded"
3205 " while LRO is enabled\n", skb->dev->name);
3207 EXPORT_SYMBOL(__skb_warn_lro_forwarding);