spi-topcliff-pch: add recovery processing in case wait-event timeout
[zen-stable.git] / net / core / skbuff.c
blob38f49e015c8d8d33bb123abb2af226c423c742ae
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(size);
193 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
194 data = kmalloc_node_track_caller(size, gfp_mask, node);
195 if (!data)
196 goto nodata;
197 /* kmalloc(size) might give us more room than requested.
198 * Put skb_shared_info exactly at the end of allocated zone,
199 * to allow max possible filling before reallocation.
201 size = SKB_WITH_OVERHEAD(ksize(data));
202 prefetchw(data + size);
205 * Only clear those fields we need to clear, not those that we will
206 * actually initialise below. Hence, don't put any more fields after
207 * the tail pointer in struct sk_buff!
209 memset(skb, 0, offsetof(struct sk_buff, tail));
210 /* Account for allocated memory : skb + skb->head */
211 skb->truesize = SKB_TRUESIZE(size);
212 atomic_set(&skb->users, 1);
213 skb->head = data;
214 skb->data = data;
215 skb_reset_tail_pointer(skb);
216 skb->end = skb->tail + size;
217 #ifdef NET_SKBUFF_DATA_USES_OFFSET
218 skb->mac_header = ~0U;
219 #endif
221 /* make sure we initialize shinfo sequentially */
222 shinfo = skb_shinfo(skb);
223 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
224 atomic_set(&shinfo->dataref, 1);
225 kmemcheck_annotate_variable(shinfo->destructor_arg);
227 if (fclone) {
228 struct sk_buff *child = skb + 1;
229 atomic_t *fclone_ref = (atomic_t *) (child + 1);
231 kmemcheck_annotate_bitfield(child, flags1);
232 kmemcheck_annotate_bitfield(child, flags2);
233 skb->fclone = SKB_FCLONE_ORIG;
234 atomic_set(fclone_ref, 1);
236 child->fclone = SKB_FCLONE_UNAVAILABLE;
238 out:
239 return skb;
240 nodata:
241 kmem_cache_free(cache, skb);
242 skb = NULL;
243 goto out;
245 EXPORT_SYMBOL(__alloc_skb);
248 * build_skb - build a network buffer
249 * @data: data buffer provided by caller
251 * Allocate a new &sk_buff. Caller provides space holding head and
252 * skb_shared_info. @data must have been allocated by kmalloc()
253 * The return is the new skb buffer.
254 * On a failure the return is %NULL, and @data is not freed.
255 * Notes :
256 * Before IO, driver allocates only data buffer where NIC put incoming frame
257 * Driver should add room at head (NET_SKB_PAD) and
258 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
259 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
260 * before giving packet to stack.
261 * RX rings only contains data buffers, not full skbs.
263 struct sk_buff *build_skb(void *data)
265 struct skb_shared_info *shinfo;
266 struct sk_buff *skb;
267 unsigned int size;
269 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
270 if (!skb)
271 return NULL;
273 size = ksize(data) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
275 memset(skb, 0, offsetof(struct sk_buff, tail));
276 skb->truesize = SKB_TRUESIZE(size);
277 atomic_set(&skb->users, 1);
278 skb->head = data;
279 skb->data = data;
280 skb_reset_tail_pointer(skb);
281 skb->end = skb->tail + size;
282 #ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb->mac_header = ~0U;
284 #endif
286 /* make sure we initialize shinfo sequentially */
287 shinfo = skb_shinfo(skb);
288 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
289 atomic_set(&shinfo->dataref, 1);
290 kmemcheck_annotate_variable(shinfo->destructor_arg);
292 return skb;
294 EXPORT_SYMBOL(build_skb);
297 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
298 * @dev: network device to receive on
299 * @length: length to allocate
300 * @gfp_mask: get_free_pages mask, passed to alloc_skb
302 * Allocate a new &sk_buff and assign it a usage count of one. The
303 * buffer has unspecified headroom built in. Users should allocate
304 * the headroom they think they need without accounting for the
305 * built in space. The built in space is used for optimisations.
307 * %NULL is returned if there is no free memory.
309 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
310 unsigned int length, gfp_t gfp_mask)
312 struct sk_buff *skb;
314 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
315 if (likely(skb)) {
316 skb_reserve(skb, NET_SKB_PAD);
317 skb->dev = dev;
319 return skb;
321 EXPORT_SYMBOL(__netdev_alloc_skb);
323 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
324 int size)
326 skb_fill_page_desc(skb, i, page, off, size);
327 skb->len += size;
328 skb->data_len += size;
329 skb->truesize += size;
331 EXPORT_SYMBOL(skb_add_rx_frag);
334 * dev_alloc_skb - allocate an skbuff for receiving
335 * @length: length to allocate
337 * Allocate a new &sk_buff and assign it a usage count of one. The
338 * buffer has unspecified headroom built in. Users should allocate
339 * the headroom they think they need without accounting for the
340 * built in space. The built in space is used for optimisations.
342 * %NULL is returned if there is no free memory. Although this function
343 * allocates memory it can be called from an interrupt.
345 struct sk_buff *dev_alloc_skb(unsigned int length)
348 * There is more code here than it seems:
349 * __dev_alloc_skb is an inline
351 return __dev_alloc_skb(length, GFP_ATOMIC);
353 EXPORT_SYMBOL(dev_alloc_skb);
355 static void skb_drop_list(struct sk_buff **listp)
357 struct sk_buff *list = *listp;
359 *listp = NULL;
361 do {
362 struct sk_buff *this = list;
363 list = list->next;
364 kfree_skb(this);
365 } while (list);
368 static inline void skb_drop_fraglist(struct sk_buff *skb)
370 skb_drop_list(&skb_shinfo(skb)->frag_list);
373 static void skb_clone_fraglist(struct sk_buff *skb)
375 struct sk_buff *list;
377 skb_walk_frags(skb, list)
378 skb_get(list);
381 static void skb_release_data(struct sk_buff *skb)
383 if (!skb->cloned ||
384 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
385 &skb_shinfo(skb)->dataref)) {
386 if (skb_shinfo(skb)->nr_frags) {
387 int i;
388 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
389 skb_frag_unref(skb, i);
393 * If skb buf is from userspace, we need to notify the caller
394 * the lower device DMA has done;
396 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
397 struct ubuf_info *uarg;
399 uarg = skb_shinfo(skb)->destructor_arg;
400 if (uarg->callback)
401 uarg->callback(uarg);
404 if (skb_has_frag_list(skb))
405 skb_drop_fraglist(skb);
407 kfree(skb->head);
412 * Free an skbuff by memory without cleaning the state.
414 static void kfree_skbmem(struct sk_buff *skb)
416 struct sk_buff *other;
417 atomic_t *fclone_ref;
419 switch (skb->fclone) {
420 case SKB_FCLONE_UNAVAILABLE:
421 kmem_cache_free(skbuff_head_cache, skb);
422 break;
424 case SKB_FCLONE_ORIG:
425 fclone_ref = (atomic_t *) (skb + 2);
426 if (atomic_dec_and_test(fclone_ref))
427 kmem_cache_free(skbuff_fclone_cache, skb);
428 break;
430 case SKB_FCLONE_CLONE:
431 fclone_ref = (atomic_t *) (skb + 1);
432 other = skb - 1;
434 /* The clone portion is available for
435 * fast-cloning again.
437 skb->fclone = SKB_FCLONE_UNAVAILABLE;
439 if (atomic_dec_and_test(fclone_ref))
440 kmem_cache_free(skbuff_fclone_cache, other);
441 break;
445 static void skb_release_head_state(struct sk_buff *skb)
447 skb_dst_drop(skb);
448 #ifdef CONFIG_XFRM
449 secpath_put(skb->sp);
450 #endif
451 if (skb->destructor) {
452 WARN_ON(in_irq());
453 skb->destructor(skb);
455 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
456 nf_conntrack_put(skb->nfct);
457 #endif
458 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
459 nf_conntrack_put_reasm(skb->nfct_reasm);
460 #endif
461 #ifdef CONFIG_BRIDGE_NETFILTER
462 nf_bridge_put(skb->nf_bridge);
463 #endif
464 /* XXX: IS this still necessary? - JHS */
465 #ifdef CONFIG_NET_SCHED
466 skb->tc_index = 0;
467 #ifdef CONFIG_NET_CLS_ACT
468 skb->tc_verd = 0;
469 #endif
470 #endif
473 /* Free everything but the sk_buff shell. */
474 static void skb_release_all(struct sk_buff *skb)
476 skb_release_head_state(skb);
477 skb_release_data(skb);
481 * __kfree_skb - private function
482 * @skb: buffer
484 * Free an sk_buff. Release anything attached to the buffer.
485 * Clean the state. This is an internal helper function. Users should
486 * always call kfree_skb
489 void __kfree_skb(struct sk_buff *skb)
491 skb_release_all(skb);
492 kfree_skbmem(skb);
494 EXPORT_SYMBOL(__kfree_skb);
497 * kfree_skb - free an sk_buff
498 * @skb: buffer to free
500 * Drop a reference to the buffer and free it if the usage count has
501 * hit zero.
503 void kfree_skb(struct sk_buff *skb)
505 if (unlikely(!skb))
506 return;
507 if (likely(atomic_read(&skb->users) == 1))
508 smp_rmb();
509 else if (likely(!atomic_dec_and_test(&skb->users)))
510 return;
511 trace_kfree_skb(skb, __builtin_return_address(0));
512 __kfree_skb(skb);
514 EXPORT_SYMBOL(kfree_skb);
517 * consume_skb - free an skbuff
518 * @skb: buffer to free
520 * Drop a ref to the buffer and free it if the usage count has hit zero
521 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
522 * is being dropped after a failure and notes that
524 void consume_skb(struct sk_buff *skb)
526 if (unlikely(!skb))
527 return;
528 if (likely(atomic_read(&skb->users) == 1))
529 smp_rmb();
530 else if (likely(!atomic_dec_and_test(&skb->users)))
531 return;
532 trace_consume_skb(skb);
533 __kfree_skb(skb);
535 EXPORT_SYMBOL(consume_skb);
538 * skb_recycle - clean up an skb for reuse
539 * @skb: buffer
541 * Recycles the skb to be reused as a receive buffer. This
542 * function does any necessary reference count dropping, and
543 * cleans up the skbuff as if it just came from __alloc_skb().
545 void skb_recycle(struct sk_buff *skb)
547 struct skb_shared_info *shinfo;
549 skb_release_head_state(skb);
551 shinfo = skb_shinfo(skb);
552 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
553 atomic_set(&shinfo->dataref, 1);
555 memset(skb, 0, offsetof(struct sk_buff, tail));
556 skb->data = skb->head + NET_SKB_PAD;
557 skb_reset_tail_pointer(skb);
559 EXPORT_SYMBOL(skb_recycle);
562 * skb_recycle_check - check if skb can be reused for receive
563 * @skb: buffer
564 * @skb_size: minimum receive buffer size
566 * Checks that the skb passed in is not shared or cloned, and
567 * that it is linear and its head portion at least as large as
568 * skb_size so that it can be recycled as a receive buffer.
569 * If these conditions are met, this function does any necessary
570 * reference count dropping and cleans up the skbuff as if it
571 * just came from __alloc_skb().
573 bool skb_recycle_check(struct sk_buff *skb, int skb_size)
575 if (!skb_is_recycleable(skb, skb_size))
576 return false;
578 skb_recycle(skb);
580 return true;
582 EXPORT_SYMBOL(skb_recycle_check);
584 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
586 new->tstamp = old->tstamp;
587 new->dev = old->dev;
588 new->transport_header = old->transport_header;
589 new->network_header = old->network_header;
590 new->mac_header = old->mac_header;
591 skb_dst_copy(new, old);
592 new->rxhash = old->rxhash;
593 new->ooo_okay = old->ooo_okay;
594 new->l4_rxhash = old->l4_rxhash;
595 #ifdef CONFIG_XFRM
596 new->sp = secpath_get(old->sp);
597 #endif
598 memcpy(new->cb, old->cb, sizeof(old->cb));
599 new->csum = old->csum;
600 new->local_df = old->local_df;
601 new->pkt_type = old->pkt_type;
602 new->ip_summed = old->ip_summed;
603 skb_copy_queue_mapping(new, old);
604 new->priority = old->priority;
605 #if IS_ENABLED(CONFIG_IP_VS)
606 new->ipvs_property = old->ipvs_property;
607 #endif
608 new->protocol = old->protocol;
609 new->mark = old->mark;
610 new->skb_iif = old->skb_iif;
611 __nf_copy(new, old);
612 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
613 new->nf_trace = old->nf_trace;
614 #endif
615 #ifdef CONFIG_NET_SCHED
616 new->tc_index = old->tc_index;
617 #ifdef CONFIG_NET_CLS_ACT
618 new->tc_verd = old->tc_verd;
619 #endif
620 #endif
621 new->vlan_tci = old->vlan_tci;
623 skb_copy_secmark(new, old);
627 * You should not add any new code to this function. Add it to
628 * __copy_skb_header above instead.
630 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
632 #define C(x) n->x = skb->x
634 n->next = n->prev = NULL;
635 n->sk = NULL;
636 __copy_skb_header(n, skb);
638 C(len);
639 C(data_len);
640 C(mac_len);
641 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
642 n->cloned = 1;
643 n->nohdr = 0;
644 n->destructor = NULL;
645 C(tail);
646 C(end);
647 C(head);
648 C(data);
649 C(truesize);
650 atomic_set(&n->users, 1);
652 atomic_inc(&(skb_shinfo(skb)->dataref));
653 skb->cloned = 1;
655 return n;
656 #undef C
660 * skb_morph - morph one skb into another
661 * @dst: the skb to receive the contents
662 * @src: the skb to supply the contents
664 * This is identical to skb_clone except that the target skb is
665 * supplied by the user.
667 * The target skb is returned upon exit.
669 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
671 skb_release_all(dst);
672 return __skb_clone(dst, src);
674 EXPORT_SYMBOL_GPL(skb_morph);
676 /* skb_copy_ubufs - copy userspace skb frags buffers to kernel
677 * @skb: the skb to modify
678 * @gfp_mask: allocation priority
680 * This must be called on SKBTX_DEV_ZEROCOPY skb.
681 * It will copy all frags into kernel and drop the reference
682 * to userspace pages.
684 * If this function is called from an interrupt gfp_mask() must be
685 * %GFP_ATOMIC.
687 * Returns 0 on success or a negative error code on failure
688 * to allocate kernel memory to copy to.
690 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
692 int i;
693 int num_frags = skb_shinfo(skb)->nr_frags;
694 struct page *page, *head = NULL;
695 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
697 for (i = 0; i < num_frags; i++) {
698 u8 *vaddr;
699 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
701 page = alloc_page(GFP_ATOMIC);
702 if (!page) {
703 while (head) {
704 struct page *next = (struct page *)head->private;
705 put_page(head);
706 head = next;
708 return -ENOMEM;
710 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
711 memcpy(page_address(page),
712 vaddr + f->page_offset, skb_frag_size(f));
713 kunmap_skb_frag(vaddr);
714 page->private = (unsigned long)head;
715 head = page;
718 /* skb frags release userspace buffers */
719 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
720 skb_frag_unref(skb, i);
722 uarg->callback(uarg);
724 /* skb frags point to kernel buffers */
725 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
726 __skb_fill_page_desc(skb, i-1, head, 0,
727 skb_shinfo(skb)->frags[i - 1].size);
728 head = (struct page *)head->private;
731 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
732 return 0;
737 * skb_clone - duplicate an sk_buff
738 * @skb: buffer to clone
739 * @gfp_mask: allocation priority
741 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
742 * copies share the same packet data but not structure. The new
743 * buffer has a reference count of 1. If the allocation fails the
744 * function returns %NULL otherwise the new buffer is returned.
746 * If this function is called from an interrupt gfp_mask() must be
747 * %GFP_ATOMIC.
750 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
752 struct sk_buff *n;
754 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
755 if (skb_copy_ubufs(skb, gfp_mask))
756 return NULL;
759 n = skb + 1;
760 if (skb->fclone == SKB_FCLONE_ORIG &&
761 n->fclone == SKB_FCLONE_UNAVAILABLE) {
762 atomic_t *fclone_ref = (atomic_t *) (n + 1);
763 n->fclone = SKB_FCLONE_CLONE;
764 atomic_inc(fclone_ref);
765 } else {
766 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
767 if (!n)
768 return NULL;
770 kmemcheck_annotate_bitfield(n, flags1);
771 kmemcheck_annotate_bitfield(n, flags2);
772 n->fclone = SKB_FCLONE_UNAVAILABLE;
775 return __skb_clone(n, skb);
777 EXPORT_SYMBOL(skb_clone);
779 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
781 #ifndef NET_SKBUFF_DATA_USES_OFFSET
783 * Shift between the two data areas in bytes
785 unsigned long offset = new->data - old->data;
786 #endif
788 __copy_skb_header(new, old);
790 #ifndef NET_SKBUFF_DATA_USES_OFFSET
791 /* {transport,network,mac}_header are relative to skb->head */
792 new->transport_header += offset;
793 new->network_header += offset;
794 if (skb_mac_header_was_set(new))
795 new->mac_header += offset;
796 #endif
797 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
798 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
799 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
803 * skb_copy - create private copy of an sk_buff
804 * @skb: buffer to copy
805 * @gfp_mask: allocation priority
807 * Make a copy of both an &sk_buff and its data. This is used when the
808 * caller wishes to modify the data and needs a private copy of the
809 * data to alter. Returns %NULL on failure or the pointer to the buffer
810 * on success. The returned buffer has a reference count of 1.
812 * As by-product this function converts non-linear &sk_buff to linear
813 * one, so that &sk_buff becomes completely private and caller is allowed
814 * to modify all the data of returned buffer. This means that this
815 * function is not recommended for use in circumstances when only
816 * header is going to be modified. Use pskb_copy() instead.
819 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
821 int headerlen = skb_headroom(skb);
822 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
823 struct sk_buff *n = alloc_skb(size, gfp_mask);
825 if (!n)
826 return NULL;
828 /* Set the data pointer */
829 skb_reserve(n, headerlen);
830 /* Set the tail pointer and length */
831 skb_put(n, skb->len);
833 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
834 BUG();
836 copy_skb_header(n, skb);
837 return n;
839 EXPORT_SYMBOL(skb_copy);
842 * __pskb_copy - create copy of an sk_buff with private head.
843 * @skb: buffer to copy
844 * @headroom: headroom of new skb
845 * @gfp_mask: allocation priority
847 * Make a copy of both an &sk_buff and part of its data, located
848 * in header. Fragmented data remain shared. This is used when
849 * the caller wishes to modify only header of &sk_buff and needs
850 * private copy of the header to alter. Returns %NULL on failure
851 * or the pointer to the buffer on success.
852 * The returned buffer has a reference count of 1.
855 struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
857 unsigned int size = skb_headlen(skb) + headroom;
858 struct sk_buff *n = alloc_skb(size, gfp_mask);
860 if (!n)
861 goto out;
863 /* Set the data pointer */
864 skb_reserve(n, headroom);
865 /* Set the tail pointer and length */
866 skb_put(n, skb_headlen(skb));
867 /* Copy the bytes */
868 skb_copy_from_linear_data(skb, n->data, n->len);
870 n->truesize += skb->data_len;
871 n->data_len = skb->data_len;
872 n->len = skb->len;
874 if (skb_shinfo(skb)->nr_frags) {
875 int i;
877 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
878 if (skb_copy_ubufs(skb, gfp_mask)) {
879 kfree_skb(n);
880 n = NULL;
881 goto out;
884 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
885 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
886 skb_frag_ref(skb, i);
888 skb_shinfo(n)->nr_frags = i;
891 if (skb_has_frag_list(skb)) {
892 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
893 skb_clone_fraglist(n);
896 copy_skb_header(n, skb);
897 out:
898 return n;
900 EXPORT_SYMBOL(__pskb_copy);
903 * pskb_expand_head - reallocate header of &sk_buff
904 * @skb: buffer to reallocate
905 * @nhead: room to add at head
906 * @ntail: room to add at tail
907 * @gfp_mask: allocation priority
909 * Expands (or creates identical copy, if &nhead and &ntail are zero)
910 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
911 * reference count of 1. Returns zero in the case of success or error,
912 * if expansion failed. In the last case, &sk_buff is not changed.
914 * All the pointers pointing into skb header may change and must be
915 * reloaded after call to this function.
918 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
919 gfp_t gfp_mask)
921 int i;
922 u8 *data;
923 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
924 long off;
925 bool fastpath;
927 BUG_ON(nhead < 0);
929 if (skb_shared(skb))
930 BUG();
932 size = SKB_DATA_ALIGN(size);
934 /* Check if we can avoid taking references on fragments if we own
935 * the last reference on skb->head. (see skb_release_data())
937 if (!skb->cloned)
938 fastpath = true;
939 else {
940 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
941 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
944 if (fastpath &&
945 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
946 memmove(skb->head + size, skb_shinfo(skb),
947 offsetof(struct skb_shared_info,
948 frags[skb_shinfo(skb)->nr_frags]));
949 memmove(skb->head + nhead, skb->head,
950 skb_tail_pointer(skb) - skb->head);
951 off = nhead;
952 goto adjust_others;
955 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
956 gfp_mask);
957 if (!data)
958 goto nodata;
959 size = SKB_WITH_OVERHEAD(ksize(data));
961 /* Copy only real data... and, alas, header. This should be
962 * optimized for the cases when header is void.
964 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
966 memcpy((struct skb_shared_info *)(data + size),
967 skb_shinfo(skb),
968 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
970 if (fastpath) {
971 kfree(skb->head);
972 } else {
973 /* copy this zero copy skb frags */
974 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
975 if (skb_copy_ubufs(skb, gfp_mask))
976 goto nofrags;
978 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
979 skb_frag_ref(skb, i);
981 if (skb_has_frag_list(skb))
982 skb_clone_fraglist(skb);
984 skb_release_data(skb);
986 off = (data + nhead) - skb->head;
988 skb->head = data;
989 adjust_others:
990 skb->data += off;
991 #ifdef NET_SKBUFF_DATA_USES_OFFSET
992 skb->end = size;
993 off = nhead;
994 #else
995 skb->end = skb->head + size;
996 #endif
997 /* {transport,network,mac}_header and tail are relative to skb->head */
998 skb->tail += off;
999 skb->transport_header += off;
1000 skb->network_header += off;
1001 if (skb_mac_header_was_set(skb))
1002 skb->mac_header += off;
1003 /* Only adjust this if it actually is csum_start rather than csum */
1004 if (skb->ip_summed == CHECKSUM_PARTIAL)
1005 skb->csum_start += nhead;
1006 skb->cloned = 0;
1007 skb->hdr_len = 0;
1008 skb->nohdr = 0;
1009 atomic_set(&skb_shinfo(skb)->dataref, 1);
1010 return 0;
1012 nofrags:
1013 kfree(data);
1014 nodata:
1015 return -ENOMEM;
1017 EXPORT_SYMBOL(pskb_expand_head);
1019 /* Make private copy of skb with writable head and some headroom */
1021 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1023 struct sk_buff *skb2;
1024 int delta = headroom - skb_headroom(skb);
1026 if (delta <= 0)
1027 skb2 = pskb_copy(skb, GFP_ATOMIC);
1028 else {
1029 skb2 = skb_clone(skb, GFP_ATOMIC);
1030 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1031 GFP_ATOMIC)) {
1032 kfree_skb(skb2);
1033 skb2 = NULL;
1036 return skb2;
1038 EXPORT_SYMBOL(skb_realloc_headroom);
1041 * skb_copy_expand - copy and expand sk_buff
1042 * @skb: buffer to copy
1043 * @newheadroom: new free bytes at head
1044 * @newtailroom: new free bytes at tail
1045 * @gfp_mask: allocation priority
1047 * Make a copy of both an &sk_buff and its data and while doing so
1048 * allocate additional space.
1050 * This is used when the caller wishes to modify the data and needs a
1051 * private copy of the data to alter as well as more space for new fields.
1052 * Returns %NULL on failure or the pointer to the buffer
1053 * on success. The returned buffer has a reference count of 1.
1055 * You must pass %GFP_ATOMIC as the allocation priority if this function
1056 * is called from an interrupt.
1058 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1059 int newheadroom, int newtailroom,
1060 gfp_t gfp_mask)
1063 * Allocate the copy buffer
1065 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1066 gfp_mask);
1067 int oldheadroom = skb_headroom(skb);
1068 int head_copy_len, head_copy_off;
1069 int off;
1071 if (!n)
1072 return NULL;
1074 skb_reserve(n, newheadroom);
1076 /* Set the tail pointer and length */
1077 skb_put(n, skb->len);
1079 head_copy_len = oldheadroom;
1080 head_copy_off = 0;
1081 if (newheadroom <= head_copy_len)
1082 head_copy_len = newheadroom;
1083 else
1084 head_copy_off = newheadroom - head_copy_len;
1086 /* Copy the linear header and data. */
1087 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1088 skb->len + head_copy_len))
1089 BUG();
1091 copy_skb_header(n, skb);
1093 off = newheadroom - oldheadroom;
1094 if (n->ip_summed == CHECKSUM_PARTIAL)
1095 n->csum_start += off;
1096 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1097 n->transport_header += off;
1098 n->network_header += off;
1099 if (skb_mac_header_was_set(skb))
1100 n->mac_header += off;
1101 #endif
1103 return n;
1105 EXPORT_SYMBOL(skb_copy_expand);
1108 * skb_pad - zero pad the tail of an skb
1109 * @skb: buffer to pad
1110 * @pad: space to pad
1112 * Ensure that a buffer is followed by a padding area that is zero
1113 * filled. Used by network drivers which may DMA or transfer data
1114 * beyond the buffer end onto the wire.
1116 * May return error in out of memory cases. The skb is freed on error.
1119 int skb_pad(struct sk_buff *skb, int pad)
1121 int err;
1122 int ntail;
1124 /* If the skbuff is non linear tailroom is always zero.. */
1125 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1126 memset(skb->data+skb->len, 0, pad);
1127 return 0;
1130 ntail = skb->data_len + pad - (skb->end - skb->tail);
1131 if (likely(skb_cloned(skb) || ntail > 0)) {
1132 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1133 if (unlikely(err))
1134 goto free_skb;
1137 /* FIXME: The use of this function with non-linear skb's really needs
1138 * to be audited.
1140 err = skb_linearize(skb);
1141 if (unlikely(err))
1142 goto free_skb;
1144 memset(skb->data + skb->len, 0, pad);
1145 return 0;
1147 free_skb:
1148 kfree_skb(skb);
1149 return err;
1151 EXPORT_SYMBOL(skb_pad);
1154 * skb_put - add data to a buffer
1155 * @skb: buffer to use
1156 * @len: amount of data to add
1158 * This function extends the used data area of the buffer. If this would
1159 * exceed the total buffer size the kernel will panic. A pointer to the
1160 * first byte of the extra data is returned.
1162 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1164 unsigned char *tmp = skb_tail_pointer(skb);
1165 SKB_LINEAR_ASSERT(skb);
1166 skb->tail += len;
1167 skb->len += len;
1168 if (unlikely(skb->tail > skb->end))
1169 skb_over_panic(skb, len, __builtin_return_address(0));
1170 return tmp;
1172 EXPORT_SYMBOL(skb_put);
1175 * skb_push - add data to the start of a buffer
1176 * @skb: buffer to use
1177 * @len: amount of data to add
1179 * This function extends the used data area of the buffer at the buffer
1180 * start. If this would exceed the total buffer headroom the kernel will
1181 * panic. A pointer to the first byte of the extra data is returned.
1183 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1185 skb->data -= len;
1186 skb->len += len;
1187 if (unlikely(skb->data<skb->head))
1188 skb_under_panic(skb, len, __builtin_return_address(0));
1189 return skb->data;
1191 EXPORT_SYMBOL(skb_push);
1194 * skb_pull - remove data from the start of a buffer
1195 * @skb: buffer to use
1196 * @len: amount of data to remove
1198 * This function removes data from the start of a buffer, returning
1199 * the memory to the headroom. A pointer to the next data in the buffer
1200 * is returned. Once the data has been pulled future pushes will overwrite
1201 * the old data.
1203 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1205 return skb_pull_inline(skb, len);
1207 EXPORT_SYMBOL(skb_pull);
1210 * skb_trim - remove end from a buffer
1211 * @skb: buffer to alter
1212 * @len: new length
1214 * Cut the length of a buffer down by removing data from the tail. If
1215 * the buffer is already under the length specified it is not modified.
1216 * The skb must be linear.
1218 void skb_trim(struct sk_buff *skb, unsigned int len)
1220 if (skb->len > len)
1221 __skb_trim(skb, len);
1223 EXPORT_SYMBOL(skb_trim);
1225 /* Trims skb to length len. It can change skb pointers.
1228 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1230 struct sk_buff **fragp;
1231 struct sk_buff *frag;
1232 int offset = skb_headlen(skb);
1233 int nfrags = skb_shinfo(skb)->nr_frags;
1234 int i;
1235 int err;
1237 if (skb_cloned(skb) &&
1238 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1239 return err;
1241 i = 0;
1242 if (offset >= len)
1243 goto drop_pages;
1245 for (; i < nfrags; i++) {
1246 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1248 if (end < len) {
1249 offset = end;
1250 continue;
1253 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1255 drop_pages:
1256 skb_shinfo(skb)->nr_frags = i;
1258 for (; i < nfrags; i++)
1259 skb_frag_unref(skb, i);
1261 if (skb_has_frag_list(skb))
1262 skb_drop_fraglist(skb);
1263 goto done;
1266 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1267 fragp = &frag->next) {
1268 int end = offset + frag->len;
1270 if (skb_shared(frag)) {
1271 struct sk_buff *nfrag;
1273 nfrag = skb_clone(frag, GFP_ATOMIC);
1274 if (unlikely(!nfrag))
1275 return -ENOMEM;
1277 nfrag->next = frag->next;
1278 kfree_skb(frag);
1279 frag = nfrag;
1280 *fragp = frag;
1283 if (end < len) {
1284 offset = end;
1285 continue;
1288 if (end > len &&
1289 unlikely((err = pskb_trim(frag, len - offset))))
1290 return err;
1292 if (frag->next)
1293 skb_drop_list(&frag->next);
1294 break;
1297 done:
1298 if (len > skb_headlen(skb)) {
1299 skb->data_len -= skb->len - len;
1300 skb->len = len;
1301 } else {
1302 skb->len = len;
1303 skb->data_len = 0;
1304 skb_set_tail_pointer(skb, len);
1307 return 0;
1309 EXPORT_SYMBOL(___pskb_trim);
1312 * __pskb_pull_tail - advance tail of skb header
1313 * @skb: buffer to reallocate
1314 * @delta: number of bytes to advance tail
1316 * The function makes a sense only on a fragmented &sk_buff,
1317 * it expands header moving its tail forward and copying necessary
1318 * data from fragmented part.
1320 * &sk_buff MUST have reference count of 1.
1322 * Returns %NULL (and &sk_buff does not change) if pull failed
1323 * or value of new tail of skb in the case of success.
1325 * All the pointers pointing into skb header may change and must be
1326 * reloaded after call to this function.
1329 /* Moves tail of skb head forward, copying data from fragmented part,
1330 * when it is necessary.
1331 * 1. It may fail due to malloc failure.
1332 * 2. It may change skb pointers.
1334 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1336 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1338 /* If skb has not enough free space at tail, get new one
1339 * plus 128 bytes for future expansions. If we have enough
1340 * room at tail, reallocate without expansion only if skb is cloned.
1342 int i, k, eat = (skb->tail + delta) - skb->end;
1344 if (eat > 0 || skb_cloned(skb)) {
1345 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1346 GFP_ATOMIC))
1347 return NULL;
1350 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1351 BUG();
1353 /* Optimization: no fragments, no reasons to preestimate
1354 * size of pulled pages. Superb.
1356 if (!skb_has_frag_list(skb))
1357 goto pull_pages;
1359 /* Estimate size of pulled pages. */
1360 eat = delta;
1361 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1364 if (size >= eat)
1365 goto pull_pages;
1366 eat -= size;
1369 /* If we need update frag list, we are in troubles.
1370 * Certainly, it possible to add an offset to skb data,
1371 * but taking into account that pulling is expected to
1372 * be very rare operation, it is worth to fight against
1373 * further bloating skb head and crucify ourselves here instead.
1374 * Pure masohism, indeed. 8)8)
1376 if (eat) {
1377 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1378 struct sk_buff *clone = NULL;
1379 struct sk_buff *insp = NULL;
1381 do {
1382 BUG_ON(!list);
1384 if (list->len <= eat) {
1385 /* Eaten as whole. */
1386 eat -= list->len;
1387 list = list->next;
1388 insp = list;
1389 } else {
1390 /* Eaten partially. */
1392 if (skb_shared(list)) {
1393 /* Sucks! We need to fork list. :-( */
1394 clone = skb_clone(list, GFP_ATOMIC);
1395 if (!clone)
1396 return NULL;
1397 insp = list->next;
1398 list = clone;
1399 } else {
1400 /* This may be pulled without
1401 * problems. */
1402 insp = list;
1404 if (!pskb_pull(list, eat)) {
1405 kfree_skb(clone);
1406 return NULL;
1408 break;
1410 } while (eat);
1412 /* Free pulled out fragments. */
1413 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1414 skb_shinfo(skb)->frag_list = list->next;
1415 kfree_skb(list);
1417 /* And insert new clone at head. */
1418 if (clone) {
1419 clone->next = list;
1420 skb_shinfo(skb)->frag_list = clone;
1423 /* Success! Now we may commit changes to skb data. */
1425 pull_pages:
1426 eat = delta;
1427 k = 0;
1428 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1429 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1431 if (size <= eat) {
1432 skb_frag_unref(skb, i);
1433 eat -= size;
1434 } else {
1435 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1436 if (eat) {
1437 skb_shinfo(skb)->frags[k].page_offset += eat;
1438 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1439 eat = 0;
1441 k++;
1444 skb_shinfo(skb)->nr_frags = k;
1446 skb->tail += delta;
1447 skb->data_len -= delta;
1449 return skb_tail_pointer(skb);
1451 EXPORT_SYMBOL(__pskb_pull_tail);
1454 * skb_copy_bits - copy bits from skb to kernel buffer
1455 * @skb: source skb
1456 * @offset: offset in source
1457 * @to: destination buffer
1458 * @len: number of bytes to copy
1460 * Copy the specified number of bytes from the source skb to the
1461 * destination buffer.
1463 * CAUTION ! :
1464 * If its prototype is ever changed,
1465 * check arch/{*}/net/{*}.S files,
1466 * since it is called from BPF assembly code.
1468 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1470 int start = skb_headlen(skb);
1471 struct sk_buff *frag_iter;
1472 int i, copy;
1474 if (offset > (int)skb->len - len)
1475 goto fault;
1477 /* Copy header. */
1478 if ((copy = start - offset) > 0) {
1479 if (copy > len)
1480 copy = len;
1481 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1482 if ((len -= copy) == 0)
1483 return 0;
1484 offset += copy;
1485 to += copy;
1488 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1489 int end;
1491 WARN_ON(start > offset + len);
1493 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1494 if ((copy = end - offset) > 0) {
1495 u8 *vaddr;
1497 if (copy > len)
1498 copy = len;
1500 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1501 memcpy(to,
1502 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1503 offset - start, copy);
1504 kunmap_skb_frag(vaddr);
1506 if ((len -= copy) == 0)
1507 return 0;
1508 offset += copy;
1509 to += copy;
1511 start = end;
1514 skb_walk_frags(skb, frag_iter) {
1515 int end;
1517 WARN_ON(start > offset + len);
1519 end = start + frag_iter->len;
1520 if ((copy = end - offset) > 0) {
1521 if (copy > len)
1522 copy = len;
1523 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1524 goto fault;
1525 if ((len -= copy) == 0)
1526 return 0;
1527 offset += copy;
1528 to += copy;
1530 start = end;
1533 if (!len)
1534 return 0;
1536 fault:
1537 return -EFAULT;
1539 EXPORT_SYMBOL(skb_copy_bits);
1542 * Callback from splice_to_pipe(), if we need to release some pages
1543 * at the end of the spd in case we error'ed out in filling the pipe.
1545 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1547 put_page(spd->pages[i]);
1550 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1551 unsigned int *offset,
1552 struct sk_buff *skb, struct sock *sk)
1554 struct page *p = sk->sk_sndmsg_page;
1555 unsigned int off;
1557 if (!p) {
1558 new_page:
1559 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1560 if (!p)
1561 return NULL;
1563 off = sk->sk_sndmsg_off = 0;
1564 /* hold one ref to this page until it's full */
1565 } else {
1566 unsigned int mlen;
1568 off = sk->sk_sndmsg_off;
1569 mlen = PAGE_SIZE - off;
1570 if (mlen < 64 && mlen < *len) {
1571 put_page(p);
1572 goto new_page;
1575 *len = min_t(unsigned int, *len, mlen);
1578 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1579 sk->sk_sndmsg_off += *len;
1580 *offset = off;
1581 get_page(p);
1583 return p;
1587 * Fill page/offset/length into spd, if it can hold more pages.
1589 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1590 struct pipe_inode_info *pipe, struct page *page,
1591 unsigned int *len, unsigned int offset,
1592 struct sk_buff *skb, int linear,
1593 struct sock *sk)
1595 if (unlikely(spd->nr_pages == pipe->buffers))
1596 return 1;
1598 if (linear) {
1599 page = linear_to_page(page, len, &offset, skb, sk);
1600 if (!page)
1601 return 1;
1602 } else
1603 get_page(page);
1605 spd->pages[spd->nr_pages] = page;
1606 spd->partial[spd->nr_pages].len = *len;
1607 spd->partial[spd->nr_pages].offset = offset;
1608 spd->nr_pages++;
1610 return 0;
1613 static inline void __segment_seek(struct page **page, unsigned int *poff,
1614 unsigned int *plen, unsigned int off)
1616 unsigned long n;
1618 *poff += off;
1619 n = *poff / PAGE_SIZE;
1620 if (n)
1621 *page = nth_page(*page, n);
1623 *poff = *poff % PAGE_SIZE;
1624 *plen -= off;
1627 static inline int __splice_segment(struct page *page, unsigned int poff,
1628 unsigned int plen, unsigned int *off,
1629 unsigned int *len, struct sk_buff *skb,
1630 struct splice_pipe_desc *spd, int linear,
1631 struct sock *sk,
1632 struct pipe_inode_info *pipe)
1634 if (!*len)
1635 return 1;
1637 /* skip this segment if already processed */
1638 if (*off >= plen) {
1639 *off -= plen;
1640 return 0;
1643 /* ignore any bits we already processed */
1644 if (*off) {
1645 __segment_seek(&page, &poff, &plen, *off);
1646 *off = 0;
1649 do {
1650 unsigned int flen = min(*len, plen);
1652 /* the linear region may spread across several pages */
1653 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1655 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1656 return 1;
1658 __segment_seek(&page, &poff, &plen, flen);
1659 *len -= flen;
1661 } while (*len && plen);
1663 return 0;
1667 * Map linear and fragment data from the skb to spd. It reports failure if the
1668 * pipe is full or if we already spliced the requested length.
1670 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1671 unsigned int *offset, unsigned int *len,
1672 struct splice_pipe_desc *spd, struct sock *sk)
1674 int seg;
1677 * map the linear part
1679 if (__splice_segment(virt_to_page(skb->data),
1680 (unsigned long) skb->data & (PAGE_SIZE - 1),
1681 skb_headlen(skb),
1682 offset, len, skb, spd, 1, sk, pipe))
1683 return 1;
1686 * then map the fragments
1688 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1689 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1691 if (__splice_segment(skb_frag_page(f),
1692 f->page_offset, skb_frag_size(f),
1693 offset, len, skb, spd, 0, sk, pipe))
1694 return 1;
1697 return 0;
1701 * Map data from the skb to a pipe. Should handle both the linear part,
1702 * the fragments, and the frag list. It does NOT handle frag lists within
1703 * the frag list, if such a thing exists. We'd probably need to recurse to
1704 * handle that cleanly.
1706 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1707 struct pipe_inode_info *pipe, unsigned int tlen,
1708 unsigned int flags)
1710 struct partial_page partial[PIPE_DEF_BUFFERS];
1711 struct page *pages[PIPE_DEF_BUFFERS];
1712 struct splice_pipe_desc spd = {
1713 .pages = pages,
1714 .partial = partial,
1715 .flags = flags,
1716 .ops = &sock_pipe_buf_ops,
1717 .spd_release = sock_spd_release,
1719 struct sk_buff *frag_iter;
1720 struct sock *sk = skb->sk;
1721 int ret = 0;
1723 if (splice_grow_spd(pipe, &spd))
1724 return -ENOMEM;
1727 * __skb_splice_bits() only fails if the output has no room left,
1728 * so no point in going over the frag_list for the error case.
1730 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1731 goto done;
1732 else if (!tlen)
1733 goto done;
1736 * now see if we have a frag_list to map
1738 skb_walk_frags(skb, frag_iter) {
1739 if (!tlen)
1740 break;
1741 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1742 break;
1745 done:
1746 if (spd.nr_pages) {
1748 * Drop the socket lock, otherwise we have reverse
1749 * locking dependencies between sk_lock and i_mutex
1750 * here as compared to sendfile(). We enter here
1751 * with the socket lock held, and splice_to_pipe() will
1752 * grab the pipe inode lock. For sendfile() emulation,
1753 * we call into ->sendpage() with the i_mutex lock held
1754 * and networking will grab the socket lock.
1756 release_sock(sk);
1757 ret = splice_to_pipe(pipe, &spd);
1758 lock_sock(sk);
1761 splice_shrink_spd(pipe, &spd);
1762 return ret;
1766 * skb_store_bits - store bits from kernel buffer to skb
1767 * @skb: destination buffer
1768 * @offset: offset in destination
1769 * @from: source buffer
1770 * @len: number of bytes to copy
1772 * Copy the specified number of bytes from the source buffer to the
1773 * destination skb. This function handles all the messy bits of
1774 * traversing fragment lists and such.
1777 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1779 int start = skb_headlen(skb);
1780 struct sk_buff *frag_iter;
1781 int i, copy;
1783 if (offset > (int)skb->len - len)
1784 goto fault;
1786 if ((copy = start - offset) > 0) {
1787 if (copy > len)
1788 copy = len;
1789 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1790 if ((len -= copy) == 0)
1791 return 0;
1792 offset += copy;
1793 from += copy;
1796 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1797 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1798 int end;
1800 WARN_ON(start > offset + len);
1802 end = start + skb_frag_size(frag);
1803 if ((copy = end - offset) > 0) {
1804 u8 *vaddr;
1806 if (copy > len)
1807 copy = len;
1809 vaddr = kmap_skb_frag(frag);
1810 memcpy(vaddr + frag->page_offset + offset - start,
1811 from, copy);
1812 kunmap_skb_frag(vaddr);
1814 if ((len -= copy) == 0)
1815 return 0;
1816 offset += copy;
1817 from += copy;
1819 start = end;
1822 skb_walk_frags(skb, frag_iter) {
1823 int end;
1825 WARN_ON(start > offset + len);
1827 end = start + frag_iter->len;
1828 if ((copy = end - offset) > 0) {
1829 if (copy > len)
1830 copy = len;
1831 if (skb_store_bits(frag_iter, offset - start,
1832 from, copy))
1833 goto fault;
1834 if ((len -= copy) == 0)
1835 return 0;
1836 offset += copy;
1837 from += copy;
1839 start = end;
1841 if (!len)
1842 return 0;
1844 fault:
1845 return -EFAULT;
1847 EXPORT_SYMBOL(skb_store_bits);
1849 /* Checksum skb data. */
1851 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1852 int len, __wsum csum)
1854 int start = skb_headlen(skb);
1855 int i, copy = start - offset;
1856 struct sk_buff *frag_iter;
1857 int pos = 0;
1859 /* Checksum header. */
1860 if (copy > 0) {
1861 if (copy > len)
1862 copy = len;
1863 csum = csum_partial(skb->data + offset, copy, csum);
1864 if ((len -= copy) == 0)
1865 return csum;
1866 offset += copy;
1867 pos = copy;
1870 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1871 int end;
1873 WARN_ON(start > offset + len);
1875 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1876 if ((copy = end - offset) > 0) {
1877 __wsum csum2;
1878 u8 *vaddr;
1879 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1881 if (copy > len)
1882 copy = len;
1883 vaddr = kmap_skb_frag(frag);
1884 csum2 = csum_partial(vaddr + frag->page_offset +
1885 offset - start, copy, 0);
1886 kunmap_skb_frag(vaddr);
1887 csum = csum_block_add(csum, csum2, pos);
1888 if (!(len -= copy))
1889 return csum;
1890 offset += copy;
1891 pos += copy;
1893 start = end;
1896 skb_walk_frags(skb, frag_iter) {
1897 int end;
1899 WARN_ON(start > offset + len);
1901 end = start + frag_iter->len;
1902 if ((copy = end - offset) > 0) {
1903 __wsum csum2;
1904 if (copy > len)
1905 copy = len;
1906 csum2 = skb_checksum(frag_iter, offset - start,
1907 copy, 0);
1908 csum = csum_block_add(csum, csum2, pos);
1909 if ((len -= copy) == 0)
1910 return csum;
1911 offset += copy;
1912 pos += copy;
1914 start = end;
1916 BUG_ON(len);
1918 return csum;
1920 EXPORT_SYMBOL(skb_checksum);
1922 /* Both of above in one bottle. */
1924 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1925 u8 *to, int len, __wsum csum)
1927 int start = skb_headlen(skb);
1928 int i, copy = start - offset;
1929 struct sk_buff *frag_iter;
1930 int pos = 0;
1932 /* Copy header. */
1933 if (copy > 0) {
1934 if (copy > len)
1935 copy = len;
1936 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1937 copy, csum);
1938 if ((len -= copy) == 0)
1939 return csum;
1940 offset += copy;
1941 to += copy;
1942 pos = copy;
1945 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1946 int end;
1948 WARN_ON(start > offset + len);
1950 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1951 if ((copy = end - offset) > 0) {
1952 __wsum csum2;
1953 u8 *vaddr;
1954 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1956 if (copy > len)
1957 copy = len;
1958 vaddr = kmap_skb_frag(frag);
1959 csum2 = csum_partial_copy_nocheck(vaddr +
1960 frag->page_offset +
1961 offset - start, to,
1962 copy, 0);
1963 kunmap_skb_frag(vaddr);
1964 csum = csum_block_add(csum, csum2, pos);
1965 if (!(len -= copy))
1966 return csum;
1967 offset += copy;
1968 to += copy;
1969 pos += copy;
1971 start = end;
1974 skb_walk_frags(skb, frag_iter) {
1975 __wsum csum2;
1976 int end;
1978 WARN_ON(start > offset + len);
1980 end = start + frag_iter->len;
1981 if ((copy = end - offset) > 0) {
1982 if (copy > len)
1983 copy = len;
1984 csum2 = skb_copy_and_csum_bits(frag_iter,
1985 offset - start,
1986 to, copy, 0);
1987 csum = csum_block_add(csum, csum2, pos);
1988 if ((len -= copy) == 0)
1989 return csum;
1990 offset += copy;
1991 to += copy;
1992 pos += copy;
1994 start = end;
1996 BUG_ON(len);
1997 return csum;
1999 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2001 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2003 __wsum csum;
2004 long csstart;
2006 if (skb->ip_summed == CHECKSUM_PARTIAL)
2007 csstart = skb_checksum_start_offset(skb);
2008 else
2009 csstart = skb_headlen(skb);
2011 BUG_ON(csstart > skb_headlen(skb));
2013 skb_copy_from_linear_data(skb, to, csstart);
2015 csum = 0;
2016 if (csstart != skb->len)
2017 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2018 skb->len - csstart, 0);
2020 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2021 long csstuff = csstart + skb->csum_offset;
2023 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2026 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2029 * skb_dequeue - remove from the head of the queue
2030 * @list: list to dequeue from
2032 * Remove the head of the list. The list lock is taken so the function
2033 * may be used safely with other locking list functions. The head item is
2034 * returned or %NULL if the list is empty.
2037 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2039 unsigned long flags;
2040 struct sk_buff *result;
2042 spin_lock_irqsave(&list->lock, flags);
2043 result = __skb_dequeue(list);
2044 spin_unlock_irqrestore(&list->lock, flags);
2045 return result;
2047 EXPORT_SYMBOL(skb_dequeue);
2050 * skb_dequeue_tail - remove from the tail of the queue
2051 * @list: list to dequeue from
2053 * Remove the tail of the list. The list lock is taken so the function
2054 * may be used safely with other locking list functions. The tail item is
2055 * returned or %NULL if the list is empty.
2057 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2059 unsigned long flags;
2060 struct sk_buff *result;
2062 spin_lock_irqsave(&list->lock, flags);
2063 result = __skb_dequeue_tail(list);
2064 spin_unlock_irqrestore(&list->lock, flags);
2065 return result;
2067 EXPORT_SYMBOL(skb_dequeue_tail);
2070 * skb_queue_purge - empty a list
2071 * @list: list to empty
2073 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2074 * the list and one reference dropped. This function takes the list
2075 * lock and is atomic with respect to other list locking functions.
2077 void skb_queue_purge(struct sk_buff_head *list)
2079 struct sk_buff *skb;
2080 while ((skb = skb_dequeue(list)) != NULL)
2081 kfree_skb(skb);
2083 EXPORT_SYMBOL(skb_queue_purge);
2086 * skb_queue_head - queue a buffer at the list head
2087 * @list: list to use
2088 * @newsk: buffer to queue
2090 * Queue a buffer at the start of the list. This function takes the
2091 * list lock and can be used safely with other locking &sk_buff functions
2092 * safely.
2094 * A buffer cannot be placed on two lists at the same time.
2096 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2098 unsigned long flags;
2100 spin_lock_irqsave(&list->lock, flags);
2101 __skb_queue_head(list, newsk);
2102 spin_unlock_irqrestore(&list->lock, flags);
2104 EXPORT_SYMBOL(skb_queue_head);
2107 * skb_queue_tail - queue a buffer at the list tail
2108 * @list: list to use
2109 * @newsk: buffer to queue
2111 * Queue a buffer at the tail of the list. This function takes the
2112 * list lock and can be used safely with other locking &sk_buff functions
2113 * safely.
2115 * A buffer cannot be placed on two lists at the same time.
2117 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2119 unsigned long flags;
2121 spin_lock_irqsave(&list->lock, flags);
2122 __skb_queue_tail(list, newsk);
2123 spin_unlock_irqrestore(&list->lock, flags);
2125 EXPORT_SYMBOL(skb_queue_tail);
2128 * skb_unlink - remove a buffer from a list
2129 * @skb: buffer to remove
2130 * @list: list to use
2132 * Remove a packet from a list. The list locks are taken and this
2133 * function is atomic with respect to other list locked calls
2135 * You must know what list the SKB is on.
2137 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2139 unsigned long flags;
2141 spin_lock_irqsave(&list->lock, flags);
2142 __skb_unlink(skb, list);
2143 spin_unlock_irqrestore(&list->lock, flags);
2145 EXPORT_SYMBOL(skb_unlink);
2148 * skb_append - append a buffer
2149 * @old: buffer to insert after
2150 * @newsk: buffer to insert
2151 * @list: list to use
2153 * Place a packet after a given packet in a list. The list locks are taken
2154 * and this function is atomic with respect to other list locked calls.
2155 * A buffer cannot be placed on two lists at the same time.
2157 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2159 unsigned long flags;
2161 spin_lock_irqsave(&list->lock, flags);
2162 __skb_queue_after(list, old, newsk);
2163 spin_unlock_irqrestore(&list->lock, flags);
2165 EXPORT_SYMBOL(skb_append);
2168 * skb_insert - insert a buffer
2169 * @old: buffer to insert before
2170 * @newsk: buffer to insert
2171 * @list: list to use
2173 * Place a packet before a given packet in a list. The list locks are
2174 * taken and this function is atomic with respect to other list locked
2175 * calls.
2177 * A buffer cannot be placed on two lists at the same time.
2179 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2181 unsigned long flags;
2183 spin_lock_irqsave(&list->lock, flags);
2184 __skb_insert(newsk, old->prev, old, list);
2185 spin_unlock_irqrestore(&list->lock, flags);
2187 EXPORT_SYMBOL(skb_insert);
2189 static inline void skb_split_inside_header(struct sk_buff *skb,
2190 struct sk_buff* skb1,
2191 const u32 len, const int pos)
2193 int i;
2195 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2196 pos - len);
2197 /* And move data appendix as is. */
2198 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2199 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2201 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2202 skb_shinfo(skb)->nr_frags = 0;
2203 skb1->data_len = skb->data_len;
2204 skb1->len += skb1->data_len;
2205 skb->data_len = 0;
2206 skb->len = len;
2207 skb_set_tail_pointer(skb, len);
2210 static inline void skb_split_no_header(struct sk_buff *skb,
2211 struct sk_buff* skb1,
2212 const u32 len, int pos)
2214 int i, k = 0;
2215 const int nfrags = skb_shinfo(skb)->nr_frags;
2217 skb_shinfo(skb)->nr_frags = 0;
2218 skb1->len = skb1->data_len = skb->len - len;
2219 skb->len = len;
2220 skb->data_len = len - pos;
2222 for (i = 0; i < nfrags; i++) {
2223 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2225 if (pos + size > len) {
2226 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2228 if (pos < len) {
2229 /* Split frag.
2230 * We have two variants in this case:
2231 * 1. Move all the frag to the second
2232 * part, if it is possible. F.e.
2233 * this approach is mandatory for TUX,
2234 * where splitting is expensive.
2235 * 2. Split is accurately. We make this.
2237 skb_frag_ref(skb, i);
2238 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2239 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2240 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2241 skb_shinfo(skb)->nr_frags++;
2243 k++;
2244 } else
2245 skb_shinfo(skb)->nr_frags++;
2246 pos += size;
2248 skb_shinfo(skb1)->nr_frags = k;
2252 * skb_split - Split fragmented skb to two parts at length len.
2253 * @skb: the buffer to split
2254 * @skb1: the buffer to receive the second part
2255 * @len: new length for skb
2257 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2259 int pos = skb_headlen(skb);
2261 if (len < pos) /* Split line is inside header. */
2262 skb_split_inside_header(skb, skb1, len, pos);
2263 else /* Second chunk has no header, nothing to copy. */
2264 skb_split_no_header(skb, skb1, len, pos);
2266 EXPORT_SYMBOL(skb_split);
2268 /* Shifting from/to a cloned skb is a no-go.
2270 * Caller cannot keep skb_shinfo related pointers past calling here!
2272 static int skb_prepare_for_shift(struct sk_buff *skb)
2274 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2278 * skb_shift - Shifts paged data partially from skb to another
2279 * @tgt: buffer into which tail data gets added
2280 * @skb: buffer from which the paged data comes from
2281 * @shiftlen: shift up to this many bytes
2283 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2284 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2285 * It's up to caller to free skb if everything was shifted.
2287 * If @tgt runs out of frags, the whole operation is aborted.
2289 * Skb cannot include anything else but paged data while tgt is allowed
2290 * to have non-paged data as well.
2292 * TODO: full sized shift could be optimized but that would need
2293 * specialized skb free'er to handle frags without up-to-date nr_frags.
2295 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2297 int from, to, merge, todo;
2298 struct skb_frag_struct *fragfrom, *fragto;
2300 BUG_ON(shiftlen > skb->len);
2301 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2303 todo = shiftlen;
2304 from = 0;
2305 to = skb_shinfo(tgt)->nr_frags;
2306 fragfrom = &skb_shinfo(skb)->frags[from];
2308 /* Actual merge is delayed until the point when we know we can
2309 * commit all, so that we don't have to undo partial changes
2311 if (!to ||
2312 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2313 fragfrom->page_offset)) {
2314 merge = -1;
2315 } else {
2316 merge = to - 1;
2318 todo -= skb_frag_size(fragfrom);
2319 if (todo < 0) {
2320 if (skb_prepare_for_shift(skb) ||
2321 skb_prepare_for_shift(tgt))
2322 return 0;
2324 /* All previous frag pointers might be stale! */
2325 fragfrom = &skb_shinfo(skb)->frags[from];
2326 fragto = &skb_shinfo(tgt)->frags[merge];
2328 skb_frag_size_add(fragto, shiftlen);
2329 skb_frag_size_sub(fragfrom, shiftlen);
2330 fragfrom->page_offset += shiftlen;
2332 goto onlymerged;
2335 from++;
2338 /* Skip full, not-fitting skb to avoid expensive operations */
2339 if ((shiftlen == skb->len) &&
2340 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2341 return 0;
2343 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2344 return 0;
2346 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2347 if (to == MAX_SKB_FRAGS)
2348 return 0;
2350 fragfrom = &skb_shinfo(skb)->frags[from];
2351 fragto = &skb_shinfo(tgt)->frags[to];
2353 if (todo >= skb_frag_size(fragfrom)) {
2354 *fragto = *fragfrom;
2355 todo -= skb_frag_size(fragfrom);
2356 from++;
2357 to++;
2359 } else {
2360 __skb_frag_ref(fragfrom);
2361 fragto->page = fragfrom->page;
2362 fragto->page_offset = fragfrom->page_offset;
2363 skb_frag_size_set(fragto, todo);
2365 fragfrom->page_offset += todo;
2366 skb_frag_size_sub(fragfrom, todo);
2367 todo = 0;
2369 to++;
2370 break;
2374 /* Ready to "commit" this state change to tgt */
2375 skb_shinfo(tgt)->nr_frags = to;
2377 if (merge >= 0) {
2378 fragfrom = &skb_shinfo(skb)->frags[0];
2379 fragto = &skb_shinfo(tgt)->frags[merge];
2381 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2382 __skb_frag_unref(fragfrom);
2385 /* Reposition in the original skb */
2386 to = 0;
2387 while (from < skb_shinfo(skb)->nr_frags)
2388 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2389 skb_shinfo(skb)->nr_frags = to;
2391 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2393 onlymerged:
2394 /* Most likely the tgt won't ever need its checksum anymore, skb on
2395 * the other hand might need it if it needs to be resent
2397 tgt->ip_summed = CHECKSUM_PARTIAL;
2398 skb->ip_summed = CHECKSUM_PARTIAL;
2400 /* Yak, is it really working this way? Some helper please? */
2401 skb->len -= shiftlen;
2402 skb->data_len -= shiftlen;
2403 skb->truesize -= shiftlen;
2404 tgt->len += shiftlen;
2405 tgt->data_len += shiftlen;
2406 tgt->truesize += shiftlen;
2408 return shiftlen;
2412 * skb_prepare_seq_read - Prepare a sequential read of skb data
2413 * @skb: the buffer to read
2414 * @from: lower offset of data to be read
2415 * @to: upper offset of data to be read
2416 * @st: state variable
2418 * Initializes the specified state variable. Must be called before
2419 * invoking skb_seq_read() for the first time.
2421 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2422 unsigned int to, struct skb_seq_state *st)
2424 st->lower_offset = from;
2425 st->upper_offset = to;
2426 st->root_skb = st->cur_skb = skb;
2427 st->frag_idx = st->stepped_offset = 0;
2428 st->frag_data = NULL;
2430 EXPORT_SYMBOL(skb_prepare_seq_read);
2433 * skb_seq_read - Sequentially read skb data
2434 * @consumed: number of bytes consumed by the caller so far
2435 * @data: destination pointer for data to be returned
2436 * @st: state variable
2438 * Reads a block of skb data at &consumed relative to the
2439 * lower offset specified to skb_prepare_seq_read(). Assigns
2440 * the head of the data block to &data and returns the length
2441 * of the block or 0 if the end of the skb data or the upper
2442 * offset has been reached.
2444 * The caller is not required to consume all of the data
2445 * returned, i.e. &consumed is typically set to the number
2446 * of bytes already consumed and the next call to
2447 * skb_seq_read() will return the remaining part of the block.
2449 * Note 1: The size of each block of data returned can be arbitrary,
2450 * this limitation is the cost for zerocopy seqeuental
2451 * reads of potentially non linear data.
2453 * Note 2: Fragment lists within fragments are not implemented
2454 * at the moment, state->root_skb could be replaced with
2455 * a stack for this purpose.
2457 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2458 struct skb_seq_state *st)
2460 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2461 skb_frag_t *frag;
2463 if (unlikely(abs_offset >= st->upper_offset))
2464 return 0;
2466 next_skb:
2467 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2469 if (abs_offset < block_limit && !st->frag_data) {
2470 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2471 return block_limit - abs_offset;
2474 if (st->frag_idx == 0 && !st->frag_data)
2475 st->stepped_offset += skb_headlen(st->cur_skb);
2477 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2478 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2479 block_limit = skb_frag_size(frag) + st->stepped_offset;
2481 if (abs_offset < block_limit) {
2482 if (!st->frag_data)
2483 st->frag_data = kmap_skb_frag(frag);
2485 *data = (u8 *) st->frag_data + frag->page_offset +
2486 (abs_offset - st->stepped_offset);
2488 return block_limit - abs_offset;
2491 if (st->frag_data) {
2492 kunmap_skb_frag(st->frag_data);
2493 st->frag_data = NULL;
2496 st->frag_idx++;
2497 st->stepped_offset += skb_frag_size(frag);
2500 if (st->frag_data) {
2501 kunmap_skb_frag(st->frag_data);
2502 st->frag_data = NULL;
2505 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2506 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2507 st->frag_idx = 0;
2508 goto next_skb;
2509 } else if (st->cur_skb->next) {
2510 st->cur_skb = st->cur_skb->next;
2511 st->frag_idx = 0;
2512 goto next_skb;
2515 return 0;
2517 EXPORT_SYMBOL(skb_seq_read);
2520 * skb_abort_seq_read - Abort a sequential read of skb data
2521 * @st: state variable
2523 * Must be called if skb_seq_read() was not called until it
2524 * returned 0.
2526 void skb_abort_seq_read(struct skb_seq_state *st)
2528 if (st->frag_data)
2529 kunmap_skb_frag(st->frag_data);
2531 EXPORT_SYMBOL(skb_abort_seq_read);
2533 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2535 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2536 struct ts_config *conf,
2537 struct ts_state *state)
2539 return skb_seq_read(offset, text, TS_SKB_CB(state));
2542 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2544 skb_abort_seq_read(TS_SKB_CB(state));
2548 * skb_find_text - Find a text pattern in skb data
2549 * @skb: the buffer to look in
2550 * @from: search offset
2551 * @to: search limit
2552 * @config: textsearch configuration
2553 * @state: uninitialized textsearch state variable
2555 * Finds a pattern in the skb data according to the specified
2556 * textsearch configuration. Use textsearch_next() to retrieve
2557 * subsequent occurrences of the pattern. Returns the offset
2558 * to the first occurrence or UINT_MAX if no match was found.
2560 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2561 unsigned int to, struct ts_config *config,
2562 struct ts_state *state)
2564 unsigned int ret;
2566 config->get_next_block = skb_ts_get_next_block;
2567 config->finish = skb_ts_finish;
2569 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2571 ret = textsearch_find(config, state);
2572 return (ret <= to - from ? ret : UINT_MAX);
2574 EXPORT_SYMBOL(skb_find_text);
2577 * skb_append_datato_frags: - append the user data to a skb
2578 * @sk: sock structure
2579 * @skb: skb structure to be appened with user data.
2580 * @getfrag: call back function to be used for getting the user data
2581 * @from: pointer to user message iov
2582 * @length: length of the iov message
2584 * Description: This procedure append the user data in the fragment part
2585 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2587 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2588 int (*getfrag)(void *from, char *to, int offset,
2589 int len, int odd, struct sk_buff *skb),
2590 void *from, int length)
2592 int frg_cnt = 0;
2593 skb_frag_t *frag = NULL;
2594 struct page *page = NULL;
2595 int copy, left;
2596 int offset = 0;
2597 int ret;
2599 do {
2600 /* Return error if we don't have space for new frag */
2601 frg_cnt = skb_shinfo(skb)->nr_frags;
2602 if (frg_cnt >= MAX_SKB_FRAGS)
2603 return -EFAULT;
2605 /* allocate a new page for next frag */
2606 page = alloc_pages(sk->sk_allocation, 0);
2608 /* If alloc_page fails just return failure and caller will
2609 * free previous allocated pages by doing kfree_skb()
2611 if (page == NULL)
2612 return -ENOMEM;
2614 /* initialize the next frag */
2615 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2616 skb->truesize += PAGE_SIZE;
2617 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2619 /* get the new initialized frag */
2620 frg_cnt = skb_shinfo(skb)->nr_frags;
2621 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2623 /* copy the user data to page */
2624 left = PAGE_SIZE - frag->page_offset;
2625 copy = (length > left)? left : length;
2627 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2628 offset, copy, 0, skb);
2629 if (ret < 0)
2630 return -EFAULT;
2632 /* copy was successful so update the size parameters */
2633 skb_frag_size_add(frag, copy);
2634 skb->len += copy;
2635 skb->data_len += copy;
2636 offset += copy;
2637 length -= copy;
2639 } while (length > 0);
2641 return 0;
2643 EXPORT_SYMBOL(skb_append_datato_frags);
2646 * skb_pull_rcsum - pull skb and update receive checksum
2647 * @skb: buffer to update
2648 * @len: length of data pulled
2650 * This function performs an skb_pull on the packet and updates
2651 * the CHECKSUM_COMPLETE checksum. It should be used on
2652 * receive path processing instead of skb_pull unless you know
2653 * that the checksum difference is zero (e.g., a valid IP header)
2654 * or you are setting ip_summed to CHECKSUM_NONE.
2656 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2658 BUG_ON(len > skb->len);
2659 skb->len -= len;
2660 BUG_ON(skb->len < skb->data_len);
2661 skb_postpull_rcsum(skb, skb->data, len);
2662 return skb->data += len;
2664 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2667 * skb_segment - Perform protocol segmentation on skb.
2668 * @skb: buffer to segment
2669 * @features: features for the output path (see dev->features)
2671 * This function performs segmentation on the given skb. It returns
2672 * a pointer to the first in a list of new skbs for the segments.
2673 * In case of error it returns ERR_PTR(err).
2675 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2677 struct sk_buff *segs = NULL;
2678 struct sk_buff *tail = NULL;
2679 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2680 unsigned int mss = skb_shinfo(skb)->gso_size;
2681 unsigned int doffset = skb->data - skb_mac_header(skb);
2682 unsigned int offset = doffset;
2683 unsigned int headroom;
2684 unsigned int len;
2685 int sg = !!(features & NETIF_F_SG);
2686 int nfrags = skb_shinfo(skb)->nr_frags;
2687 int err = -ENOMEM;
2688 int i = 0;
2689 int pos;
2691 __skb_push(skb, doffset);
2692 headroom = skb_headroom(skb);
2693 pos = skb_headlen(skb);
2695 do {
2696 struct sk_buff *nskb;
2697 skb_frag_t *frag;
2698 int hsize;
2699 int size;
2701 len = skb->len - offset;
2702 if (len > mss)
2703 len = mss;
2705 hsize = skb_headlen(skb) - offset;
2706 if (hsize < 0)
2707 hsize = 0;
2708 if (hsize > len || !sg)
2709 hsize = len;
2711 if (!hsize && i >= nfrags) {
2712 BUG_ON(fskb->len != len);
2714 pos += len;
2715 nskb = skb_clone(fskb, GFP_ATOMIC);
2716 fskb = fskb->next;
2718 if (unlikely(!nskb))
2719 goto err;
2721 hsize = skb_end_pointer(nskb) - nskb->head;
2722 if (skb_cow_head(nskb, doffset + headroom)) {
2723 kfree_skb(nskb);
2724 goto err;
2727 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2728 hsize;
2729 skb_release_head_state(nskb);
2730 __skb_push(nskb, doffset);
2731 } else {
2732 nskb = alloc_skb(hsize + doffset + headroom,
2733 GFP_ATOMIC);
2735 if (unlikely(!nskb))
2736 goto err;
2738 skb_reserve(nskb, headroom);
2739 __skb_put(nskb, doffset);
2742 if (segs)
2743 tail->next = nskb;
2744 else
2745 segs = nskb;
2746 tail = nskb;
2748 __copy_skb_header(nskb, skb);
2749 nskb->mac_len = skb->mac_len;
2751 /* nskb and skb might have different headroom */
2752 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2753 nskb->csum_start += skb_headroom(nskb) - headroom;
2755 skb_reset_mac_header(nskb);
2756 skb_set_network_header(nskb, skb->mac_len);
2757 nskb->transport_header = (nskb->network_header +
2758 skb_network_header_len(skb));
2759 skb_copy_from_linear_data(skb, nskb->data, doffset);
2761 if (fskb != skb_shinfo(skb)->frag_list)
2762 continue;
2764 if (!sg) {
2765 nskb->ip_summed = CHECKSUM_NONE;
2766 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2767 skb_put(nskb, len),
2768 len, 0);
2769 continue;
2772 frag = skb_shinfo(nskb)->frags;
2774 skb_copy_from_linear_data_offset(skb, offset,
2775 skb_put(nskb, hsize), hsize);
2777 while (pos < offset + len && i < nfrags) {
2778 *frag = skb_shinfo(skb)->frags[i];
2779 __skb_frag_ref(frag);
2780 size = skb_frag_size(frag);
2782 if (pos < offset) {
2783 frag->page_offset += offset - pos;
2784 skb_frag_size_sub(frag, offset - pos);
2787 skb_shinfo(nskb)->nr_frags++;
2789 if (pos + size <= offset + len) {
2790 i++;
2791 pos += size;
2792 } else {
2793 skb_frag_size_sub(frag, pos + size - (offset + len));
2794 goto skip_fraglist;
2797 frag++;
2800 if (pos < offset + len) {
2801 struct sk_buff *fskb2 = fskb;
2803 BUG_ON(pos + fskb->len != offset + len);
2805 pos += fskb->len;
2806 fskb = fskb->next;
2808 if (fskb2->next) {
2809 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2810 if (!fskb2)
2811 goto err;
2812 } else
2813 skb_get(fskb2);
2815 SKB_FRAG_ASSERT(nskb);
2816 skb_shinfo(nskb)->frag_list = fskb2;
2819 skip_fraglist:
2820 nskb->data_len = len - hsize;
2821 nskb->len += nskb->data_len;
2822 nskb->truesize += nskb->data_len;
2823 } while ((offset += len) < skb->len);
2825 return segs;
2827 err:
2828 while ((skb = segs)) {
2829 segs = skb->next;
2830 kfree_skb(skb);
2832 return ERR_PTR(err);
2834 EXPORT_SYMBOL_GPL(skb_segment);
2836 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2838 struct sk_buff *p = *head;
2839 struct sk_buff *nskb;
2840 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2841 struct skb_shared_info *pinfo = skb_shinfo(p);
2842 unsigned int headroom;
2843 unsigned int len = skb_gro_len(skb);
2844 unsigned int offset = skb_gro_offset(skb);
2845 unsigned int headlen = skb_headlen(skb);
2847 if (p->len + len >= 65536)
2848 return -E2BIG;
2850 if (pinfo->frag_list)
2851 goto merge;
2852 else if (headlen <= offset) {
2853 skb_frag_t *frag;
2854 skb_frag_t *frag2;
2855 int i = skbinfo->nr_frags;
2856 int nr_frags = pinfo->nr_frags + i;
2858 offset -= headlen;
2860 if (nr_frags > MAX_SKB_FRAGS)
2861 return -E2BIG;
2863 pinfo->nr_frags = nr_frags;
2864 skbinfo->nr_frags = 0;
2866 frag = pinfo->frags + nr_frags;
2867 frag2 = skbinfo->frags + i;
2868 do {
2869 *--frag = *--frag2;
2870 } while (--i);
2872 frag->page_offset += offset;
2873 skb_frag_size_sub(frag, offset);
2875 skb->truesize -= skb->data_len;
2876 skb->len -= skb->data_len;
2877 skb->data_len = 0;
2879 NAPI_GRO_CB(skb)->free = 1;
2880 goto done;
2881 } else if (skb_gro_len(p) != pinfo->gso_size)
2882 return -E2BIG;
2884 headroom = skb_headroom(p);
2885 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2886 if (unlikely(!nskb))
2887 return -ENOMEM;
2889 __copy_skb_header(nskb, p);
2890 nskb->mac_len = p->mac_len;
2892 skb_reserve(nskb, headroom);
2893 __skb_put(nskb, skb_gro_offset(p));
2895 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2896 skb_set_network_header(nskb, skb_network_offset(p));
2897 skb_set_transport_header(nskb, skb_transport_offset(p));
2899 __skb_pull(p, skb_gro_offset(p));
2900 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2901 p->data - skb_mac_header(p));
2903 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2904 skb_shinfo(nskb)->frag_list = p;
2905 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2906 pinfo->gso_size = 0;
2907 skb_header_release(p);
2908 nskb->prev = p;
2910 nskb->data_len += p->len;
2911 nskb->truesize += p->len;
2912 nskb->len += p->len;
2914 *head = nskb;
2915 nskb->next = p->next;
2916 p->next = NULL;
2918 p = nskb;
2920 merge:
2921 if (offset > headlen) {
2922 unsigned int eat = offset - headlen;
2924 skbinfo->frags[0].page_offset += eat;
2925 skb_frag_size_sub(&skbinfo->frags[0], eat);
2926 skb->data_len -= eat;
2927 skb->len -= eat;
2928 offset = headlen;
2931 __skb_pull(skb, offset);
2933 p->prev->next = skb;
2934 p->prev = skb;
2935 skb_header_release(skb);
2937 done:
2938 NAPI_GRO_CB(p)->count++;
2939 p->data_len += len;
2940 p->truesize += len;
2941 p->len += len;
2943 NAPI_GRO_CB(skb)->same_flow = 1;
2944 return 0;
2946 EXPORT_SYMBOL_GPL(skb_gro_receive);
2948 void __init skb_init(void)
2950 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2951 sizeof(struct sk_buff),
2953 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2954 NULL);
2955 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2956 (2*sizeof(struct sk_buff)) +
2957 sizeof(atomic_t),
2959 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2960 NULL);
2964 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2965 * @skb: Socket buffer containing the buffers to be mapped
2966 * @sg: The scatter-gather list to map into
2967 * @offset: The offset into the buffer's contents to start mapping
2968 * @len: Length of buffer space to be mapped
2970 * Fill the specified scatter-gather list with mappings/pointers into a
2971 * region of the buffer space attached to a socket buffer.
2973 static int
2974 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2976 int start = skb_headlen(skb);
2977 int i, copy = start - offset;
2978 struct sk_buff *frag_iter;
2979 int elt = 0;
2981 if (copy > 0) {
2982 if (copy > len)
2983 copy = len;
2984 sg_set_buf(sg, skb->data + offset, copy);
2985 elt++;
2986 if ((len -= copy) == 0)
2987 return elt;
2988 offset += copy;
2991 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2992 int end;
2994 WARN_ON(start > offset + len);
2996 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2997 if ((copy = end - offset) > 0) {
2998 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3000 if (copy > len)
3001 copy = len;
3002 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3003 frag->page_offset+offset-start);
3004 elt++;
3005 if (!(len -= copy))
3006 return elt;
3007 offset += copy;
3009 start = end;
3012 skb_walk_frags(skb, frag_iter) {
3013 int end;
3015 WARN_ON(start > offset + len);
3017 end = start + frag_iter->len;
3018 if ((copy = end - offset) > 0) {
3019 if (copy > len)
3020 copy = len;
3021 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3022 copy);
3023 if ((len -= copy) == 0)
3024 return elt;
3025 offset += copy;
3027 start = end;
3029 BUG_ON(len);
3030 return elt;
3033 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3035 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3037 sg_mark_end(&sg[nsg - 1]);
3039 return nsg;
3041 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3044 * skb_cow_data - Check that a socket buffer's data buffers are writable
3045 * @skb: The socket buffer to check.
3046 * @tailbits: Amount of trailing space to be added
3047 * @trailer: Returned pointer to the skb where the @tailbits space begins
3049 * Make sure that the data buffers attached to a socket buffer are
3050 * writable. If they are not, private copies are made of the data buffers
3051 * and the socket buffer is set to use these instead.
3053 * If @tailbits is given, make sure that there is space to write @tailbits
3054 * bytes of data beyond current end of socket buffer. @trailer will be
3055 * set to point to the skb in which this space begins.
3057 * The number of scatterlist elements required to completely map the
3058 * COW'd and extended socket buffer will be returned.
3060 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3062 int copyflag;
3063 int elt;
3064 struct sk_buff *skb1, **skb_p;
3066 /* If skb is cloned or its head is paged, reallocate
3067 * head pulling out all the pages (pages are considered not writable
3068 * at the moment even if they are anonymous).
3070 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3071 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3072 return -ENOMEM;
3074 /* Easy case. Most of packets will go this way. */
3075 if (!skb_has_frag_list(skb)) {
3076 /* A little of trouble, not enough of space for trailer.
3077 * This should not happen, when stack is tuned to generate
3078 * good frames. OK, on miss we reallocate and reserve even more
3079 * space, 128 bytes is fair. */
3081 if (skb_tailroom(skb) < tailbits &&
3082 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3083 return -ENOMEM;
3085 /* Voila! */
3086 *trailer = skb;
3087 return 1;
3090 /* Misery. We are in troubles, going to mincer fragments... */
3092 elt = 1;
3093 skb_p = &skb_shinfo(skb)->frag_list;
3094 copyflag = 0;
3096 while ((skb1 = *skb_p) != NULL) {
3097 int ntail = 0;
3099 /* The fragment is partially pulled by someone,
3100 * this can happen on input. Copy it and everything
3101 * after it. */
3103 if (skb_shared(skb1))
3104 copyflag = 1;
3106 /* If the skb is the last, worry about trailer. */
3108 if (skb1->next == NULL && tailbits) {
3109 if (skb_shinfo(skb1)->nr_frags ||
3110 skb_has_frag_list(skb1) ||
3111 skb_tailroom(skb1) < tailbits)
3112 ntail = tailbits + 128;
3115 if (copyflag ||
3116 skb_cloned(skb1) ||
3117 ntail ||
3118 skb_shinfo(skb1)->nr_frags ||
3119 skb_has_frag_list(skb1)) {
3120 struct sk_buff *skb2;
3122 /* Fuck, we are miserable poor guys... */
3123 if (ntail == 0)
3124 skb2 = skb_copy(skb1, GFP_ATOMIC);
3125 else
3126 skb2 = skb_copy_expand(skb1,
3127 skb_headroom(skb1),
3128 ntail,
3129 GFP_ATOMIC);
3130 if (unlikely(skb2 == NULL))
3131 return -ENOMEM;
3133 if (skb1->sk)
3134 skb_set_owner_w(skb2, skb1->sk);
3136 /* Looking around. Are we still alive?
3137 * OK, link new skb, drop old one */
3139 skb2->next = skb1->next;
3140 *skb_p = skb2;
3141 kfree_skb(skb1);
3142 skb1 = skb2;
3144 elt++;
3145 *trailer = skb1;
3146 skb_p = &skb1->next;
3149 return elt;
3151 EXPORT_SYMBOL_GPL(skb_cow_data);
3153 static void sock_rmem_free(struct sk_buff *skb)
3155 struct sock *sk = skb->sk;
3157 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3161 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3163 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3165 int len = skb->len;
3167 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3168 (unsigned)sk->sk_rcvbuf)
3169 return -ENOMEM;
3171 skb_orphan(skb);
3172 skb->sk = sk;
3173 skb->destructor = sock_rmem_free;
3174 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3176 /* before exiting rcu section, make sure dst is refcounted */
3177 skb_dst_force(skb);
3179 skb_queue_tail(&sk->sk_error_queue, skb);
3180 if (!sock_flag(sk, SOCK_DEAD))
3181 sk->sk_data_ready(sk, len);
3182 return 0;
3184 EXPORT_SYMBOL(sock_queue_err_skb);
3186 void skb_tstamp_tx(struct sk_buff *orig_skb,
3187 struct skb_shared_hwtstamps *hwtstamps)
3189 struct sock *sk = orig_skb->sk;
3190 struct sock_exterr_skb *serr;
3191 struct sk_buff *skb;
3192 int err;
3194 if (!sk)
3195 return;
3197 skb = skb_clone(orig_skb, GFP_ATOMIC);
3198 if (!skb)
3199 return;
3201 if (hwtstamps) {
3202 *skb_hwtstamps(skb) =
3203 *hwtstamps;
3204 } else {
3206 * no hardware time stamps available,
3207 * so keep the shared tx_flags and only
3208 * store software time stamp
3210 skb->tstamp = ktime_get_real();
3213 serr = SKB_EXT_ERR(skb);
3214 memset(serr, 0, sizeof(*serr));
3215 serr->ee.ee_errno = ENOMSG;
3216 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3218 err = sock_queue_err_skb(sk, skb);
3220 if (err)
3221 kfree_skb(skb);
3223 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3225 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3227 struct sock *sk = skb->sk;
3228 struct sock_exterr_skb *serr;
3229 int err;
3231 skb->wifi_acked_valid = 1;
3232 skb->wifi_acked = acked;
3234 serr = SKB_EXT_ERR(skb);
3235 memset(serr, 0, sizeof(*serr));
3236 serr->ee.ee_errno = ENOMSG;
3237 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3239 err = sock_queue_err_skb(sk, skb);
3240 if (err)
3241 kfree_skb(skb);
3243 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3247 * skb_partial_csum_set - set up and verify partial csum values for packet
3248 * @skb: the skb to set
3249 * @start: the number of bytes after skb->data to start checksumming.
3250 * @off: the offset from start to place the checksum.
3252 * For untrusted partially-checksummed packets, we need to make sure the values
3253 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3255 * This function checks and sets those values and skb->ip_summed: if this
3256 * returns false you should drop the packet.
3258 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3260 if (unlikely(start > skb_headlen(skb)) ||
3261 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3262 if (net_ratelimit())
3263 printk(KERN_WARNING
3264 "bad partial csum: csum=%u/%u len=%u\n",
3265 start, off, skb_headlen(skb));
3266 return false;
3268 skb->ip_summed = CHECKSUM_PARTIAL;
3269 skb->csum_start = skb_headroom(skb) + start;
3270 skb->csum_offset = off;
3271 return true;
3273 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3275 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3277 if (net_ratelimit())
3278 pr_warning("%s: received packets cannot be forwarded"
3279 " while LRO is enabled\n", skb->dev->name);
3281 EXPORT_SYMBOL(__skb_warn_lro_forwarding);