Merge tag 'usb-3.3-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/usb
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / net / core / skbuff.c
blobda0c97f2fab446b551e79abd65bb485316ae53fe
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 + sizeof(struct skb_shared_info), gfp_mask);
956 if (!data)
957 goto nodata;
959 /* Copy only real data... and, alas, header. This should be
960 * optimized for the cases when header is void.
962 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
964 memcpy((struct skb_shared_info *)(data + size),
965 skb_shinfo(skb),
966 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
968 if (fastpath) {
969 kfree(skb->head);
970 } else {
971 /* copy this zero copy skb frags */
972 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
973 if (skb_copy_ubufs(skb, gfp_mask))
974 goto nofrags;
976 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
977 skb_frag_ref(skb, i);
979 if (skb_has_frag_list(skb))
980 skb_clone_fraglist(skb);
982 skb_release_data(skb);
984 off = (data + nhead) - skb->head;
986 skb->head = data;
987 adjust_others:
988 skb->data += off;
989 #ifdef NET_SKBUFF_DATA_USES_OFFSET
990 skb->end = size;
991 off = nhead;
992 #else
993 skb->end = skb->head + size;
994 #endif
995 /* {transport,network,mac}_header and tail are relative to skb->head */
996 skb->tail += off;
997 skb->transport_header += off;
998 skb->network_header += off;
999 if (skb_mac_header_was_set(skb))
1000 skb->mac_header += off;
1001 /* Only adjust this if it actually is csum_start rather than csum */
1002 if (skb->ip_summed == CHECKSUM_PARTIAL)
1003 skb->csum_start += nhead;
1004 skb->cloned = 0;
1005 skb->hdr_len = 0;
1006 skb->nohdr = 0;
1007 atomic_set(&skb_shinfo(skb)->dataref, 1);
1008 return 0;
1010 nofrags:
1011 kfree(data);
1012 nodata:
1013 return -ENOMEM;
1015 EXPORT_SYMBOL(pskb_expand_head);
1017 /* Make private copy of skb with writable head and some headroom */
1019 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1021 struct sk_buff *skb2;
1022 int delta = headroom - skb_headroom(skb);
1024 if (delta <= 0)
1025 skb2 = pskb_copy(skb, GFP_ATOMIC);
1026 else {
1027 skb2 = skb_clone(skb, GFP_ATOMIC);
1028 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1029 GFP_ATOMIC)) {
1030 kfree_skb(skb2);
1031 skb2 = NULL;
1034 return skb2;
1036 EXPORT_SYMBOL(skb_realloc_headroom);
1039 * skb_copy_expand - copy and expand sk_buff
1040 * @skb: buffer to copy
1041 * @newheadroom: new free bytes at head
1042 * @newtailroom: new free bytes at tail
1043 * @gfp_mask: allocation priority
1045 * Make a copy of both an &sk_buff and its data and while doing so
1046 * allocate additional space.
1048 * This is used when the caller wishes to modify the data and needs a
1049 * private copy of the data to alter as well as more space for new fields.
1050 * Returns %NULL on failure or the pointer to the buffer
1051 * on success. The returned buffer has a reference count of 1.
1053 * You must pass %GFP_ATOMIC as the allocation priority if this function
1054 * is called from an interrupt.
1056 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1057 int newheadroom, int newtailroom,
1058 gfp_t gfp_mask)
1061 * Allocate the copy buffer
1063 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1064 gfp_mask);
1065 int oldheadroom = skb_headroom(skb);
1066 int head_copy_len, head_copy_off;
1067 int off;
1069 if (!n)
1070 return NULL;
1072 skb_reserve(n, newheadroom);
1074 /* Set the tail pointer and length */
1075 skb_put(n, skb->len);
1077 head_copy_len = oldheadroom;
1078 head_copy_off = 0;
1079 if (newheadroom <= head_copy_len)
1080 head_copy_len = newheadroom;
1081 else
1082 head_copy_off = newheadroom - head_copy_len;
1084 /* Copy the linear header and data. */
1085 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1086 skb->len + head_copy_len))
1087 BUG();
1089 copy_skb_header(n, skb);
1091 off = newheadroom - oldheadroom;
1092 if (n->ip_summed == CHECKSUM_PARTIAL)
1093 n->csum_start += off;
1094 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1095 n->transport_header += off;
1096 n->network_header += off;
1097 if (skb_mac_header_was_set(skb))
1098 n->mac_header += off;
1099 #endif
1101 return n;
1103 EXPORT_SYMBOL(skb_copy_expand);
1106 * skb_pad - zero pad the tail of an skb
1107 * @skb: buffer to pad
1108 * @pad: space to pad
1110 * Ensure that a buffer is followed by a padding area that is zero
1111 * filled. Used by network drivers which may DMA or transfer data
1112 * beyond the buffer end onto the wire.
1114 * May return error in out of memory cases. The skb is freed on error.
1117 int skb_pad(struct sk_buff *skb, int pad)
1119 int err;
1120 int ntail;
1122 /* If the skbuff is non linear tailroom is always zero.. */
1123 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1124 memset(skb->data+skb->len, 0, pad);
1125 return 0;
1128 ntail = skb->data_len + pad - (skb->end - skb->tail);
1129 if (likely(skb_cloned(skb) || ntail > 0)) {
1130 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1131 if (unlikely(err))
1132 goto free_skb;
1135 /* FIXME: The use of this function with non-linear skb's really needs
1136 * to be audited.
1138 err = skb_linearize(skb);
1139 if (unlikely(err))
1140 goto free_skb;
1142 memset(skb->data + skb->len, 0, pad);
1143 return 0;
1145 free_skb:
1146 kfree_skb(skb);
1147 return err;
1149 EXPORT_SYMBOL(skb_pad);
1152 * skb_put - add data to a buffer
1153 * @skb: buffer to use
1154 * @len: amount of data to add
1156 * This function extends the used data area of the buffer. If this would
1157 * exceed the total buffer size the kernel will panic. A pointer to the
1158 * first byte of the extra data is returned.
1160 unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1162 unsigned char *tmp = skb_tail_pointer(skb);
1163 SKB_LINEAR_ASSERT(skb);
1164 skb->tail += len;
1165 skb->len += len;
1166 if (unlikely(skb->tail > skb->end))
1167 skb_over_panic(skb, len, __builtin_return_address(0));
1168 return tmp;
1170 EXPORT_SYMBOL(skb_put);
1173 * skb_push - add data to the start of a buffer
1174 * @skb: buffer to use
1175 * @len: amount of data to add
1177 * This function extends the used data area of the buffer at the buffer
1178 * start. If this would exceed the total buffer headroom the kernel will
1179 * panic. A pointer to the first byte of the extra data is returned.
1181 unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1183 skb->data -= len;
1184 skb->len += len;
1185 if (unlikely(skb->data<skb->head))
1186 skb_under_panic(skb, len, __builtin_return_address(0));
1187 return skb->data;
1189 EXPORT_SYMBOL(skb_push);
1192 * skb_pull - remove data from the start of a buffer
1193 * @skb: buffer to use
1194 * @len: amount of data to remove
1196 * This function removes data from the start of a buffer, returning
1197 * the memory to the headroom. A pointer to the next data in the buffer
1198 * is returned. Once the data has been pulled future pushes will overwrite
1199 * the old data.
1201 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1203 return skb_pull_inline(skb, len);
1205 EXPORT_SYMBOL(skb_pull);
1208 * skb_trim - remove end from a buffer
1209 * @skb: buffer to alter
1210 * @len: new length
1212 * Cut the length of a buffer down by removing data from the tail. If
1213 * the buffer is already under the length specified it is not modified.
1214 * The skb must be linear.
1216 void skb_trim(struct sk_buff *skb, unsigned int len)
1218 if (skb->len > len)
1219 __skb_trim(skb, len);
1221 EXPORT_SYMBOL(skb_trim);
1223 /* Trims skb to length len. It can change skb pointers.
1226 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1228 struct sk_buff **fragp;
1229 struct sk_buff *frag;
1230 int offset = skb_headlen(skb);
1231 int nfrags = skb_shinfo(skb)->nr_frags;
1232 int i;
1233 int err;
1235 if (skb_cloned(skb) &&
1236 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1237 return err;
1239 i = 0;
1240 if (offset >= len)
1241 goto drop_pages;
1243 for (; i < nfrags; i++) {
1244 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1246 if (end < len) {
1247 offset = end;
1248 continue;
1251 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1253 drop_pages:
1254 skb_shinfo(skb)->nr_frags = i;
1256 for (; i < nfrags; i++)
1257 skb_frag_unref(skb, i);
1259 if (skb_has_frag_list(skb))
1260 skb_drop_fraglist(skb);
1261 goto done;
1264 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1265 fragp = &frag->next) {
1266 int end = offset + frag->len;
1268 if (skb_shared(frag)) {
1269 struct sk_buff *nfrag;
1271 nfrag = skb_clone(frag, GFP_ATOMIC);
1272 if (unlikely(!nfrag))
1273 return -ENOMEM;
1275 nfrag->next = frag->next;
1276 kfree_skb(frag);
1277 frag = nfrag;
1278 *fragp = frag;
1281 if (end < len) {
1282 offset = end;
1283 continue;
1286 if (end > len &&
1287 unlikely((err = pskb_trim(frag, len - offset))))
1288 return err;
1290 if (frag->next)
1291 skb_drop_list(&frag->next);
1292 break;
1295 done:
1296 if (len > skb_headlen(skb)) {
1297 skb->data_len -= skb->len - len;
1298 skb->len = len;
1299 } else {
1300 skb->len = len;
1301 skb->data_len = 0;
1302 skb_set_tail_pointer(skb, len);
1305 return 0;
1307 EXPORT_SYMBOL(___pskb_trim);
1310 * __pskb_pull_tail - advance tail of skb header
1311 * @skb: buffer to reallocate
1312 * @delta: number of bytes to advance tail
1314 * The function makes a sense only on a fragmented &sk_buff,
1315 * it expands header moving its tail forward and copying necessary
1316 * data from fragmented part.
1318 * &sk_buff MUST have reference count of 1.
1320 * Returns %NULL (and &sk_buff does not change) if pull failed
1321 * or value of new tail of skb in the case of success.
1323 * All the pointers pointing into skb header may change and must be
1324 * reloaded after call to this function.
1327 /* Moves tail of skb head forward, copying data from fragmented part,
1328 * when it is necessary.
1329 * 1. It may fail due to malloc failure.
1330 * 2. It may change skb pointers.
1332 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1334 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1336 /* If skb has not enough free space at tail, get new one
1337 * plus 128 bytes for future expansions. If we have enough
1338 * room at tail, reallocate without expansion only if skb is cloned.
1340 int i, k, eat = (skb->tail + delta) - skb->end;
1342 if (eat > 0 || skb_cloned(skb)) {
1343 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1344 GFP_ATOMIC))
1345 return NULL;
1348 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1349 BUG();
1351 /* Optimization: no fragments, no reasons to preestimate
1352 * size of pulled pages. Superb.
1354 if (!skb_has_frag_list(skb))
1355 goto pull_pages;
1357 /* Estimate size of pulled pages. */
1358 eat = delta;
1359 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1360 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1362 if (size >= eat)
1363 goto pull_pages;
1364 eat -= size;
1367 /* If we need update frag list, we are in troubles.
1368 * Certainly, it possible to add an offset to skb data,
1369 * but taking into account that pulling is expected to
1370 * be very rare operation, it is worth to fight against
1371 * further bloating skb head and crucify ourselves here instead.
1372 * Pure masohism, indeed. 8)8)
1374 if (eat) {
1375 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1376 struct sk_buff *clone = NULL;
1377 struct sk_buff *insp = NULL;
1379 do {
1380 BUG_ON(!list);
1382 if (list->len <= eat) {
1383 /* Eaten as whole. */
1384 eat -= list->len;
1385 list = list->next;
1386 insp = list;
1387 } else {
1388 /* Eaten partially. */
1390 if (skb_shared(list)) {
1391 /* Sucks! We need to fork list. :-( */
1392 clone = skb_clone(list, GFP_ATOMIC);
1393 if (!clone)
1394 return NULL;
1395 insp = list->next;
1396 list = clone;
1397 } else {
1398 /* This may be pulled without
1399 * problems. */
1400 insp = list;
1402 if (!pskb_pull(list, eat)) {
1403 kfree_skb(clone);
1404 return NULL;
1406 break;
1408 } while (eat);
1410 /* Free pulled out fragments. */
1411 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1412 skb_shinfo(skb)->frag_list = list->next;
1413 kfree_skb(list);
1415 /* And insert new clone at head. */
1416 if (clone) {
1417 clone->next = list;
1418 skb_shinfo(skb)->frag_list = clone;
1421 /* Success! Now we may commit changes to skb data. */
1423 pull_pages:
1424 eat = delta;
1425 k = 0;
1426 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1427 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1429 if (size <= eat) {
1430 skb_frag_unref(skb, i);
1431 eat -= size;
1432 } else {
1433 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1434 if (eat) {
1435 skb_shinfo(skb)->frags[k].page_offset += eat;
1436 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1437 eat = 0;
1439 k++;
1442 skb_shinfo(skb)->nr_frags = k;
1444 skb->tail += delta;
1445 skb->data_len -= delta;
1447 return skb_tail_pointer(skb);
1449 EXPORT_SYMBOL(__pskb_pull_tail);
1452 * skb_copy_bits - copy bits from skb to kernel buffer
1453 * @skb: source skb
1454 * @offset: offset in source
1455 * @to: destination buffer
1456 * @len: number of bytes to copy
1458 * Copy the specified number of bytes from the source skb to the
1459 * destination buffer.
1461 * CAUTION ! :
1462 * If its prototype is ever changed,
1463 * check arch/{*}/net/{*}.S files,
1464 * since it is called from BPF assembly code.
1466 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1468 int start = skb_headlen(skb);
1469 struct sk_buff *frag_iter;
1470 int i, copy;
1472 if (offset > (int)skb->len - len)
1473 goto fault;
1475 /* Copy header. */
1476 if ((copy = start - offset) > 0) {
1477 if (copy > len)
1478 copy = len;
1479 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1480 if ((len -= copy) == 0)
1481 return 0;
1482 offset += copy;
1483 to += copy;
1486 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1487 int end;
1489 WARN_ON(start > offset + len);
1491 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1492 if ((copy = end - offset) > 0) {
1493 u8 *vaddr;
1495 if (copy > len)
1496 copy = len;
1498 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1499 memcpy(to,
1500 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1501 offset - start, copy);
1502 kunmap_skb_frag(vaddr);
1504 if ((len -= copy) == 0)
1505 return 0;
1506 offset += copy;
1507 to += copy;
1509 start = end;
1512 skb_walk_frags(skb, frag_iter) {
1513 int end;
1515 WARN_ON(start > offset + len);
1517 end = start + frag_iter->len;
1518 if ((copy = end - offset) > 0) {
1519 if (copy > len)
1520 copy = len;
1521 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1522 goto fault;
1523 if ((len -= copy) == 0)
1524 return 0;
1525 offset += copy;
1526 to += copy;
1528 start = end;
1531 if (!len)
1532 return 0;
1534 fault:
1535 return -EFAULT;
1537 EXPORT_SYMBOL(skb_copy_bits);
1540 * Callback from splice_to_pipe(), if we need to release some pages
1541 * at the end of the spd in case we error'ed out in filling the pipe.
1543 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1545 put_page(spd->pages[i]);
1548 static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1549 unsigned int *offset,
1550 struct sk_buff *skb, struct sock *sk)
1552 struct page *p = sk->sk_sndmsg_page;
1553 unsigned int off;
1555 if (!p) {
1556 new_page:
1557 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1558 if (!p)
1559 return NULL;
1561 off = sk->sk_sndmsg_off = 0;
1562 /* hold one ref to this page until it's full */
1563 } else {
1564 unsigned int mlen;
1566 off = sk->sk_sndmsg_off;
1567 mlen = PAGE_SIZE - off;
1568 if (mlen < 64 && mlen < *len) {
1569 put_page(p);
1570 goto new_page;
1573 *len = min_t(unsigned int, *len, mlen);
1576 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1577 sk->sk_sndmsg_off += *len;
1578 *offset = off;
1579 get_page(p);
1581 return p;
1585 * Fill page/offset/length into spd, if it can hold more pages.
1587 static inline int spd_fill_page(struct splice_pipe_desc *spd,
1588 struct pipe_inode_info *pipe, struct page *page,
1589 unsigned int *len, unsigned int offset,
1590 struct sk_buff *skb, int linear,
1591 struct sock *sk)
1593 if (unlikely(spd->nr_pages == pipe->buffers))
1594 return 1;
1596 if (linear) {
1597 page = linear_to_page(page, len, &offset, skb, sk);
1598 if (!page)
1599 return 1;
1600 } else
1601 get_page(page);
1603 spd->pages[spd->nr_pages] = page;
1604 spd->partial[spd->nr_pages].len = *len;
1605 spd->partial[spd->nr_pages].offset = offset;
1606 spd->nr_pages++;
1608 return 0;
1611 static inline void __segment_seek(struct page **page, unsigned int *poff,
1612 unsigned int *plen, unsigned int off)
1614 unsigned long n;
1616 *poff += off;
1617 n = *poff / PAGE_SIZE;
1618 if (n)
1619 *page = nth_page(*page, n);
1621 *poff = *poff % PAGE_SIZE;
1622 *plen -= off;
1625 static inline int __splice_segment(struct page *page, unsigned int poff,
1626 unsigned int plen, unsigned int *off,
1627 unsigned int *len, struct sk_buff *skb,
1628 struct splice_pipe_desc *spd, int linear,
1629 struct sock *sk,
1630 struct pipe_inode_info *pipe)
1632 if (!*len)
1633 return 1;
1635 /* skip this segment if already processed */
1636 if (*off >= plen) {
1637 *off -= plen;
1638 return 0;
1641 /* ignore any bits we already processed */
1642 if (*off) {
1643 __segment_seek(&page, &poff, &plen, *off);
1644 *off = 0;
1647 do {
1648 unsigned int flen = min(*len, plen);
1650 /* the linear region may spread across several pages */
1651 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1653 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1654 return 1;
1656 __segment_seek(&page, &poff, &plen, flen);
1657 *len -= flen;
1659 } while (*len && plen);
1661 return 0;
1665 * Map linear and fragment data from the skb to spd. It reports failure if the
1666 * pipe is full or if we already spliced the requested length.
1668 static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1669 unsigned int *offset, unsigned int *len,
1670 struct splice_pipe_desc *spd, struct sock *sk)
1672 int seg;
1675 * map the linear part
1677 if (__splice_segment(virt_to_page(skb->data),
1678 (unsigned long) skb->data & (PAGE_SIZE - 1),
1679 skb_headlen(skb),
1680 offset, len, skb, spd, 1, sk, pipe))
1681 return 1;
1684 * then map the fragments
1686 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1687 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1689 if (__splice_segment(skb_frag_page(f),
1690 f->page_offset, skb_frag_size(f),
1691 offset, len, skb, spd, 0, sk, pipe))
1692 return 1;
1695 return 0;
1699 * Map data from the skb to a pipe. Should handle both the linear part,
1700 * the fragments, and the frag list. It does NOT handle frag lists within
1701 * the frag list, if such a thing exists. We'd probably need to recurse to
1702 * handle that cleanly.
1704 int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1705 struct pipe_inode_info *pipe, unsigned int tlen,
1706 unsigned int flags)
1708 struct partial_page partial[PIPE_DEF_BUFFERS];
1709 struct page *pages[PIPE_DEF_BUFFERS];
1710 struct splice_pipe_desc spd = {
1711 .pages = pages,
1712 .partial = partial,
1713 .flags = flags,
1714 .ops = &sock_pipe_buf_ops,
1715 .spd_release = sock_spd_release,
1717 struct sk_buff *frag_iter;
1718 struct sock *sk = skb->sk;
1719 int ret = 0;
1721 if (splice_grow_spd(pipe, &spd))
1722 return -ENOMEM;
1725 * __skb_splice_bits() only fails if the output has no room left,
1726 * so no point in going over the frag_list for the error case.
1728 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1729 goto done;
1730 else if (!tlen)
1731 goto done;
1734 * now see if we have a frag_list to map
1736 skb_walk_frags(skb, frag_iter) {
1737 if (!tlen)
1738 break;
1739 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1740 break;
1743 done:
1744 if (spd.nr_pages) {
1746 * Drop the socket lock, otherwise we have reverse
1747 * locking dependencies between sk_lock and i_mutex
1748 * here as compared to sendfile(). We enter here
1749 * with the socket lock held, and splice_to_pipe() will
1750 * grab the pipe inode lock. For sendfile() emulation,
1751 * we call into ->sendpage() with the i_mutex lock held
1752 * and networking will grab the socket lock.
1754 release_sock(sk);
1755 ret = splice_to_pipe(pipe, &spd);
1756 lock_sock(sk);
1759 splice_shrink_spd(pipe, &spd);
1760 return ret;
1764 * skb_store_bits - store bits from kernel buffer to skb
1765 * @skb: destination buffer
1766 * @offset: offset in destination
1767 * @from: source buffer
1768 * @len: number of bytes to copy
1770 * Copy the specified number of bytes from the source buffer to the
1771 * destination skb. This function handles all the messy bits of
1772 * traversing fragment lists and such.
1775 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1777 int start = skb_headlen(skb);
1778 struct sk_buff *frag_iter;
1779 int i, copy;
1781 if (offset > (int)skb->len - len)
1782 goto fault;
1784 if ((copy = start - offset) > 0) {
1785 if (copy > len)
1786 copy = len;
1787 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1788 if ((len -= copy) == 0)
1789 return 0;
1790 offset += copy;
1791 from += copy;
1794 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1795 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1796 int end;
1798 WARN_ON(start > offset + len);
1800 end = start + skb_frag_size(frag);
1801 if ((copy = end - offset) > 0) {
1802 u8 *vaddr;
1804 if (copy > len)
1805 copy = len;
1807 vaddr = kmap_skb_frag(frag);
1808 memcpy(vaddr + frag->page_offset + offset - start,
1809 from, copy);
1810 kunmap_skb_frag(vaddr);
1812 if ((len -= copy) == 0)
1813 return 0;
1814 offset += copy;
1815 from += copy;
1817 start = end;
1820 skb_walk_frags(skb, frag_iter) {
1821 int end;
1823 WARN_ON(start > offset + len);
1825 end = start + frag_iter->len;
1826 if ((copy = end - offset) > 0) {
1827 if (copy > len)
1828 copy = len;
1829 if (skb_store_bits(frag_iter, offset - start,
1830 from, copy))
1831 goto fault;
1832 if ((len -= copy) == 0)
1833 return 0;
1834 offset += copy;
1835 from += copy;
1837 start = end;
1839 if (!len)
1840 return 0;
1842 fault:
1843 return -EFAULT;
1845 EXPORT_SYMBOL(skb_store_bits);
1847 /* Checksum skb data. */
1849 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1850 int len, __wsum csum)
1852 int start = skb_headlen(skb);
1853 int i, copy = start - offset;
1854 struct sk_buff *frag_iter;
1855 int pos = 0;
1857 /* Checksum header. */
1858 if (copy > 0) {
1859 if (copy > len)
1860 copy = len;
1861 csum = csum_partial(skb->data + offset, copy, csum);
1862 if ((len -= copy) == 0)
1863 return csum;
1864 offset += copy;
1865 pos = copy;
1868 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1869 int end;
1871 WARN_ON(start > offset + len);
1873 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1874 if ((copy = end - offset) > 0) {
1875 __wsum csum2;
1876 u8 *vaddr;
1877 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1879 if (copy > len)
1880 copy = len;
1881 vaddr = kmap_skb_frag(frag);
1882 csum2 = csum_partial(vaddr + frag->page_offset +
1883 offset - start, copy, 0);
1884 kunmap_skb_frag(vaddr);
1885 csum = csum_block_add(csum, csum2, pos);
1886 if (!(len -= copy))
1887 return csum;
1888 offset += copy;
1889 pos += copy;
1891 start = end;
1894 skb_walk_frags(skb, frag_iter) {
1895 int end;
1897 WARN_ON(start > offset + len);
1899 end = start + frag_iter->len;
1900 if ((copy = end - offset) > 0) {
1901 __wsum csum2;
1902 if (copy > len)
1903 copy = len;
1904 csum2 = skb_checksum(frag_iter, offset - start,
1905 copy, 0);
1906 csum = csum_block_add(csum, csum2, pos);
1907 if ((len -= copy) == 0)
1908 return csum;
1909 offset += copy;
1910 pos += copy;
1912 start = end;
1914 BUG_ON(len);
1916 return csum;
1918 EXPORT_SYMBOL(skb_checksum);
1920 /* Both of above in one bottle. */
1922 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1923 u8 *to, int len, __wsum csum)
1925 int start = skb_headlen(skb);
1926 int i, copy = start - offset;
1927 struct sk_buff *frag_iter;
1928 int pos = 0;
1930 /* Copy header. */
1931 if (copy > 0) {
1932 if (copy > len)
1933 copy = len;
1934 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1935 copy, csum);
1936 if ((len -= copy) == 0)
1937 return csum;
1938 offset += copy;
1939 to += copy;
1940 pos = copy;
1943 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1944 int end;
1946 WARN_ON(start > offset + len);
1948 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1949 if ((copy = end - offset) > 0) {
1950 __wsum csum2;
1951 u8 *vaddr;
1952 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1954 if (copy > len)
1955 copy = len;
1956 vaddr = kmap_skb_frag(frag);
1957 csum2 = csum_partial_copy_nocheck(vaddr +
1958 frag->page_offset +
1959 offset - start, to,
1960 copy, 0);
1961 kunmap_skb_frag(vaddr);
1962 csum = csum_block_add(csum, csum2, pos);
1963 if (!(len -= copy))
1964 return csum;
1965 offset += copy;
1966 to += copy;
1967 pos += copy;
1969 start = end;
1972 skb_walk_frags(skb, frag_iter) {
1973 __wsum csum2;
1974 int end;
1976 WARN_ON(start > offset + len);
1978 end = start + frag_iter->len;
1979 if ((copy = end - offset) > 0) {
1980 if (copy > len)
1981 copy = len;
1982 csum2 = skb_copy_and_csum_bits(frag_iter,
1983 offset - start,
1984 to, copy, 0);
1985 csum = csum_block_add(csum, csum2, pos);
1986 if ((len -= copy) == 0)
1987 return csum;
1988 offset += copy;
1989 to += copy;
1990 pos += copy;
1992 start = end;
1994 BUG_ON(len);
1995 return csum;
1997 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1999 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2001 __wsum csum;
2002 long csstart;
2004 if (skb->ip_summed == CHECKSUM_PARTIAL)
2005 csstart = skb_checksum_start_offset(skb);
2006 else
2007 csstart = skb_headlen(skb);
2009 BUG_ON(csstart > skb_headlen(skb));
2011 skb_copy_from_linear_data(skb, to, csstart);
2013 csum = 0;
2014 if (csstart != skb->len)
2015 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2016 skb->len - csstart, 0);
2018 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2019 long csstuff = csstart + skb->csum_offset;
2021 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2024 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2027 * skb_dequeue - remove from the head of the queue
2028 * @list: list to dequeue from
2030 * Remove the head of the list. The list lock is taken so the function
2031 * may be used safely with other locking list functions. The head item is
2032 * returned or %NULL if the list is empty.
2035 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2037 unsigned long flags;
2038 struct sk_buff *result;
2040 spin_lock_irqsave(&list->lock, flags);
2041 result = __skb_dequeue(list);
2042 spin_unlock_irqrestore(&list->lock, flags);
2043 return result;
2045 EXPORT_SYMBOL(skb_dequeue);
2048 * skb_dequeue_tail - remove from the tail of the queue
2049 * @list: list to dequeue from
2051 * Remove the tail of the list. The list lock is taken so the function
2052 * may be used safely with other locking list functions. The tail item is
2053 * returned or %NULL if the list is empty.
2055 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2057 unsigned long flags;
2058 struct sk_buff *result;
2060 spin_lock_irqsave(&list->lock, flags);
2061 result = __skb_dequeue_tail(list);
2062 spin_unlock_irqrestore(&list->lock, flags);
2063 return result;
2065 EXPORT_SYMBOL(skb_dequeue_tail);
2068 * skb_queue_purge - empty a list
2069 * @list: list to empty
2071 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2072 * the list and one reference dropped. This function takes the list
2073 * lock and is atomic with respect to other list locking functions.
2075 void skb_queue_purge(struct sk_buff_head *list)
2077 struct sk_buff *skb;
2078 while ((skb = skb_dequeue(list)) != NULL)
2079 kfree_skb(skb);
2081 EXPORT_SYMBOL(skb_queue_purge);
2084 * skb_queue_head - queue a buffer at the list head
2085 * @list: list to use
2086 * @newsk: buffer to queue
2088 * Queue a buffer at the start of the list. This function takes the
2089 * list lock and can be used safely with other locking &sk_buff functions
2090 * safely.
2092 * A buffer cannot be placed on two lists at the same time.
2094 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2096 unsigned long flags;
2098 spin_lock_irqsave(&list->lock, flags);
2099 __skb_queue_head(list, newsk);
2100 spin_unlock_irqrestore(&list->lock, flags);
2102 EXPORT_SYMBOL(skb_queue_head);
2105 * skb_queue_tail - queue a buffer at the list tail
2106 * @list: list to use
2107 * @newsk: buffer to queue
2109 * Queue a buffer at the tail of the list. This function takes the
2110 * list lock and can be used safely with other locking &sk_buff functions
2111 * safely.
2113 * A buffer cannot be placed on two lists at the same time.
2115 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2117 unsigned long flags;
2119 spin_lock_irqsave(&list->lock, flags);
2120 __skb_queue_tail(list, newsk);
2121 spin_unlock_irqrestore(&list->lock, flags);
2123 EXPORT_SYMBOL(skb_queue_tail);
2126 * skb_unlink - remove a buffer from a list
2127 * @skb: buffer to remove
2128 * @list: list to use
2130 * Remove a packet from a list. The list locks are taken and this
2131 * function is atomic with respect to other list locked calls
2133 * You must know what list the SKB is on.
2135 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2137 unsigned long flags;
2139 spin_lock_irqsave(&list->lock, flags);
2140 __skb_unlink(skb, list);
2141 spin_unlock_irqrestore(&list->lock, flags);
2143 EXPORT_SYMBOL(skb_unlink);
2146 * skb_append - append a buffer
2147 * @old: buffer to insert after
2148 * @newsk: buffer to insert
2149 * @list: list to use
2151 * Place a packet after a given packet in a list. The list locks are taken
2152 * and this function is atomic with respect to other list locked calls.
2153 * A buffer cannot be placed on two lists at the same time.
2155 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2157 unsigned long flags;
2159 spin_lock_irqsave(&list->lock, flags);
2160 __skb_queue_after(list, old, newsk);
2161 spin_unlock_irqrestore(&list->lock, flags);
2163 EXPORT_SYMBOL(skb_append);
2166 * skb_insert - insert a buffer
2167 * @old: buffer to insert before
2168 * @newsk: buffer to insert
2169 * @list: list to use
2171 * Place a packet before a given packet in a list. The list locks are
2172 * taken and this function is atomic with respect to other list locked
2173 * calls.
2175 * A buffer cannot be placed on two lists at the same time.
2177 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2179 unsigned long flags;
2181 spin_lock_irqsave(&list->lock, flags);
2182 __skb_insert(newsk, old->prev, old, list);
2183 spin_unlock_irqrestore(&list->lock, flags);
2185 EXPORT_SYMBOL(skb_insert);
2187 static inline void skb_split_inside_header(struct sk_buff *skb,
2188 struct sk_buff* skb1,
2189 const u32 len, const int pos)
2191 int i;
2193 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2194 pos - len);
2195 /* And move data appendix as is. */
2196 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2197 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2199 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2200 skb_shinfo(skb)->nr_frags = 0;
2201 skb1->data_len = skb->data_len;
2202 skb1->len += skb1->data_len;
2203 skb->data_len = 0;
2204 skb->len = len;
2205 skb_set_tail_pointer(skb, len);
2208 static inline void skb_split_no_header(struct sk_buff *skb,
2209 struct sk_buff* skb1,
2210 const u32 len, int pos)
2212 int i, k = 0;
2213 const int nfrags = skb_shinfo(skb)->nr_frags;
2215 skb_shinfo(skb)->nr_frags = 0;
2216 skb1->len = skb1->data_len = skb->len - len;
2217 skb->len = len;
2218 skb->data_len = len - pos;
2220 for (i = 0; i < nfrags; i++) {
2221 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2223 if (pos + size > len) {
2224 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2226 if (pos < len) {
2227 /* Split frag.
2228 * We have two variants in this case:
2229 * 1. Move all the frag to the second
2230 * part, if it is possible. F.e.
2231 * this approach is mandatory for TUX,
2232 * where splitting is expensive.
2233 * 2. Split is accurately. We make this.
2235 skb_frag_ref(skb, i);
2236 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2237 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2238 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2239 skb_shinfo(skb)->nr_frags++;
2241 k++;
2242 } else
2243 skb_shinfo(skb)->nr_frags++;
2244 pos += size;
2246 skb_shinfo(skb1)->nr_frags = k;
2250 * skb_split - Split fragmented skb to two parts at length len.
2251 * @skb: the buffer to split
2252 * @skb1: the buffer to receive the second part
2253 * @len: new length for skb
2255 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2257 int pos = skb_headlen(skb);
2259 if (len < pos) /* Split line is inside header. */
2260 skb_split_inside_header(skb, skb1, len, pos);
2261 else /* Second chunk has no header, nothing to copy. */
2262 skb_split_no_header(skb, skb1, len, pos);
2264 EXPORT_SYMBOL(skb_split);
2266 /* Shifting from/to a cloned skb is a no-go.
2268 * Caller cannot keep skb_shinfo related pointers past calling here!
2270 static int skb_prepare_for_shift(struct sk_buff *skb)
2272 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2276 * skb_shift - Shifts paged data partially from skb to another
2277 * @tgt: buffer into which tail data gets added
2278 * @skb: buffer from which the paged data comes from
2279 * @shiftlen: shift up to this many bytes
2281 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2282 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2283 * It's up to caller to free skb if everything was shifted.
2285 * If @tgt runs out of frags, the whole operation is aborted.
2287 * Skb cannot include anything else but paged data while tgt is allowed
2288 * to have non-paged data as well.
2290 * TODO: full sized shift could be optimized but that would need
2291 * specialized skb free'er to handle frags without up-to-date nr_frags.
2293 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2295 int from, to, merge, todo;
2296 struct skb_frag_struct *fragfrom, *fragto;
2298 BUG_ON(shiftlen > skb->len);
2299 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2301 todo = shiftlen;
2302 from = 0;
2303 to = skb_shinfo(tgt)->nr_frags;
2304 fragfrom = &skb_shinfo(skb)->frags[from];
2306 /* Actual merge is delayed until the point when we know we can
2307 * commit all, so that we don't have to undo partial changes
2309 if (!to ||
2310 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2311 fragfrom->page_offset)) {
2312 merge = -1;
2313 } else {
2314 merge = to - 1;
2316 todo -= skb_frag_size(fragfrom);
2317 if (todo < 0) {
2318 if (skb_prepare_for_shift(skb) ||
2319 skb_prepare_for_shift(tgt))
2320 return 0;
2322 /* All previous frag pointers might be stale! */
2323 fragfrom = &skb_shinfo(skb)->frags[from];
2324 fragto = &skb_shinfo(tgt)->frags[merge];
2326 skb_frag_size_add(fragto, shiftlen);
2327 skb_frag_size_sub(fragfrom, shiftlen);
2328 fragfrom->page_offset += shiftlen;
2330 goto onlymerged;
2333 from++;
2336 /* Skip full, not-fitting skb to avoid expensive operations */
2337 if ((shiftlen == skb->len) &&
2338 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2339 return 0;
2341 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2342 return 0;
2344 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2345 if (to == MAX_SKB_FRAGS)
2346 return 0;
2348 fragfrom = &skb_shinfo(skb)->frags[from];
2349 fragto = &skb_shinfo(tgt)->frags[to];
2351 if (todo >= skb_frag_size(fragfrom)) {
2352 *fragto = *fragfrom;
2353 todo -= skb_frag_size(fragfrom);
2354 from++;
2355 to++;
2357 } else {
2358 __skb_frag_ref(fragfrom);
2359 fragto->page = fragfrom->page;
2360 fragto->page_offset = fragfrom->page_offset;
2361 skb_frag_size_set(fragto, todo);
2363 fragfrom->page_offset += todo;
2364 skb_frag_size_sub(fragfrom, todo);
2365 todo = 0;
2367 to++;
2368 break;
2372 /* Ready to "commit" this state change to tgt */
2373 skb_shinfo(tgt)->nr_frags = to;
2375 if (merge >= 0) {
2376 fragfrom = &skb_shinfo(skb)->frags[0];
2377 fragto = &skb_shinfo(tgt)->frags[merge];
2379 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2380 __skb_frag_unref(fragfrom);
2383 /* Reposition in the original skb */
2384 to = 0;
2385 while (from < skb_shinfo(skb)->nr_frags)
2386 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2387 skb_shinfo(skb)->nr_frags = to;
2389 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2391 onlymerged:
2392 /* Most likely the tgt won't ever need its checksum anymore, skb on
2393 * the other hand might need it if it needs to be resent
2395 tgt->ip_summed = CHECKSUM_PARTIAL;
2396 skb->ip_summed = CHECKSUM_PARTIAL;
2398 /* Yak, is it really working this way? Some helper please? */
2399 skb->len -= shiftlen;
2400 skb->data_len -= shiftlen;
2401 skb->truesize -= shiftlen;
2402 tgt->len += shiftlen;
2403 tgt->data_len += shiftlen;
2404 tgt->truesize += shiftlen;
2406 return shiftlen;
2410 * skb_prepare_seq_read - Prepare a sequential read of skb data
2411 * @skb: the buffer to read
2412 * @from: lower offset of data to be read
2413 * @to: upper offset of data to be read
2414 * @st: state variable
2416 * Initializes the specified state variable. Must be called before
2417 * invoking skb_seq_read() for the first time.
2419 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2420 unsigned int to, struct skb_seq_state *st)
2422 st->lower_offset = from;
2423 st->upper_offset = to;
2424 st->root_skb = st->cur_skb = skb;
2425 st->frag_idx = st->stepped_offset = 0;
2426 st->frag_data = NULL;
2428 EXPORT_SYMBOL(skb_prepare_seq_read);
2431 * skb_seq_read - Sequentially read skb data
2432 * @consumed: number of bytes consumed by the caller so far
2433 * @data: destination pointer for data to be returned
2434 * @st: state variable
2436 * Reads a block of skb data at &consumed relative to the
2437 * lower offset specified to skb_prepare_seq_read(). Assigns
2438 * the head of the data block to &data and returns the length
2439 * of the block or 0 if the end of the skb data or the upper
2440 * offset has been reached.
2442 * The caller is not required to consume all of the data
2443 * returned, i.e. &consumed is typically set to the number
2444 * of bytes already consumed and the next call to
2445 * skb_seq_read() will return the remaining part of the block.
2447 * Note 1: The size of each block of data returned can be arbitrary,
2448 * this limitation is the cost for zerocopy seqeuental
2449 * reads of potentially non linear data.
2451 * Note 2: Fragment lists within fragments are not implemented
2452 * at the moment, state->root_skb could be replaced with
2453 * a stack for this purpose.
2455 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2456 struct skb_seq_state *st)
2458 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2459 skb_frag_t *frag;
2461 if (unlikely(abs_offset >= st->upper_offset))
2462 return 0;
2464 next_skb:
2465 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2467 if (abs_offset < block_limit && !st->frag_data) {
2468 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2469 return block_limit - abs_offset;
2472 if (st->frag_idx == 0 && !st->frag_data)
2473 st->stepped_offset += skb_headlen(st->cur_skb);
2475 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2476 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2477 block_limit = skb_frag_size(frag) + st->stepped_offset;
2479 if (abs_offset < block_limit) {
2480 if (!st->frag_data)
2481 st->frag_data = kmap_skb_frag(frag);
2483 *data = (u8 *) st->frag_data + frag->page_offset +
2484 (abs_offset - st->stepped_offset);
2486 return block_limit - abs_offset;
2489 if (st->frag_data) {
2490 kunmap_skb_frag(st->frag_data);
2491 st->frag_data = NULL;
2494 st->frag_idx++;
2495 st->stepped_offset += skb_frag_size(frag);
2498 if (st->frag_data) {
2499 kunmap_skb_frag(st->frag_data);
2500 st->frag_data = NULL;
2503 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2504 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2505 st->frag_idx = 0;
2506 goto next_skb;
2507 } else if (st->cur_skb->next) {
2508 st->cur_skb = st->cur_skb->next;
2509 st->frag_idx = 0;
2510 goto next_skb;
2513 return 0;
2515 EXPORT_SYMBOL(skb_seq_read);
2518 * skb_abort_seq_read - Abort a sequential read of skb data
2519 * @st: state variable
2521 * Must be called if skb_seq_read() was not called until it
2522 * returned 0.
2524 void skb_abort_seq_read(struct skb_seq_state *st)
2526 if (st->frag_data)
2527 kunmap_skb_frag(st->frag_data);
2529 EXPORT_SYMBOL(skb_abort_seq_read);
2531 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2533 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2534 struct ts_config *conf,
2535 struct ts_state *state)
2537 return skb_seq_read(offset, text, TS_SKB_CB(state));
2540 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2542 skb_abort_seq_read(TS_SKB_CB(state));
2546 * skb_find_text - Find a text pattern in skb data
2547 * @skb: the buffer to look in
2548 * @from: search offset
2549 * @to: search limit
2550 * @config: textsearch configuration
2551 * @state: uninitialized textsearch state variable
2553 * Finds a pattern in the skb data according to the specified
2554 * textsearch configuration. Use textsearch_next() to retrieve
2555 * subsequent occurrences of the pattern. Returns the offset
2556 * to the first occurrence or UINT_MAX if no match was found.
2558 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2559 unsigned int to, struct ts_config *config,
2560 struct ts_state *state)
2562 unsigned int ret;
2564 config->get_next_block = skb_ts_get_next_block;
2565 config->finish = skb_ts_finish;
2567 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2569 ret = textsearch_find(config, state);
2570 return (ret <= to - from ? ret : UINT_MAX);
2572 EXPORT_SYMBOL(skb_find_text);
2575 * skb_append_datato_frags: - append the user data to a skb
2576 * @sk: sock structure
2577 * @skb: skb structure to be appened with user data.
2578 * @getfrag: call back function to be used for getting the user data
2579 * @from: pointer to user message iov
2580 * @length: length of the iov message
2582 * Description: This procedure append the user data in the fragment part
2583 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2585 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2586 int (*getfrag)(void *from, char *to, int offset,
2587 int len, int odd, struct sk_buff *skb),
2588 void *from, int length)
2590 int frg_cnt = 0;
2591 skb_frag_t *frag = NULL;
2592 struct page *page = NULL;
2593 int copy, left;
2594 int offset = 0;
2595 int ret;
2597 do {
2598 /* Return error if we don't have space for new frag */
2599 frg_cnt = skb_shinfo(skb)->nr_frags;
2600 if (frg_cnt >= MAX_SKB_FRAGS)
2601 return -EFAULT;
2603 /* allocate a new page for next frag */
2604 page = alloc_pages(sk->sk_allocation, 0);
2606 /* If alloc_page fails just return failure and caller will
2607 * free previous allocated pages by doing kfree_skb()
2609 if (page == NULL)
2610 return -ENOMEM;
2612 /* initialize the next frag */
2613 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2614 skb->truesize += PAGE_SIZE;
2615 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2617 /* get the new initialized frag */
2618 frg_cnt = skb_shinfo(skb)->nr_frags;
2619 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2621 /* copy the user data to page */
2622 left = PAGE_SIZE - frag->page_offset;
2623 copy = (length > left)? left : length;
2625 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2626 offset, copy, 0, skb);
2627 if (ret < 0)
2628 return -EFAULT;
2630 /* copy was successful so update the size parameters */
2631 skb_frag_size_add(frag, copy);
2632 skb->len += copy;
2633 skb->data_len += copy;
2634 offset += copy;
2635 length -= copy;
2637 } while (length > 0);
2639 return 0;
2641 EXPORT_SYMBOL(skb_append_datato_frags);
2644 * skb_pull_rcsum - pull skb and update receive checksum
2645 * @skb: buffer to update
2646 * @len: length of data pulled
2648 * This function performs an skb_pull on the packet and updates
2649 * the CHECKSUM_COMPLETE checksum. It should be used on
2650 * receive path processing instead of skb_pull unless you know
2651 * that the checksum difference is zero (e.g., a valid IP header)
2652 * or you are setting ip_summed to CHECKSUM_NONE.
2654 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2656 BUG_ON(len > skb->len);
2657 skb->len -= len;
2658 BUG_ON(skb->len < skb->data_len);
2659 skb_postpull_rcsum(skb, skb->data, len);
2660 return skb->data += len;
2662 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2665 * skb_segment - Perform protocol segmentation on skb.
2666 * @skb: buffer to segment
2667 * @features: features for the output path (see dev->features)
2669 * This function performs segmentation on the given skb. It returns
2670 * a pointer to the first in a list of new skbs for the segments.
2671 * In case of error it returns ERR_PTR(err).
2673 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2675 struct sk_buff *segs = NULL;
2676 struct sk_buff *tail = NULL;
2677 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2678 unsigned int mss = skb_shinfo(skb)->gso_size;
2679 unsigned int doffset = skb->data - skb_mac_header(skb);
2680 unsigned int offset = doffset;
2681 unsigned int headroom;
2682 unsigned int len;
2683 int sg = !!(features & NETIF_F_SG);
2684 int nfrags = skb_shinfo(skb)->nr_frags;
2685 int err = -ENOMEM;
2686 int i = 0;
2687 int pos;
2689 __skb_push(skb, doffset);
2690 headroom = skb_headroom(skb);
2691 pos = skb_headlen(skb);
2693 do {
2694 struct sk_buff *nskb;
2695 skb_frag_t *frag;
2696 int hsize;
2697 int size;
2699 len = skb->len - offset;
2700 if (len > mss)
2701 len = mss;
2703 hsize = skb_headlen(skb) - offset;
2704 if (hsize < 0)
2705 hsize = 0;
2706 if (hsize > len || !sg)
2707 hsize = len;
2709 if (!hsize && i >= nfrags) {
2710 BUG_ON(fskb->len != len);
2712 pos += len;
2713 nskb = skb_clone(fskb, GFP_ATOMIC);
2714 fskb = fskb->next;
2716 if (unlikely(!nskb))
2717 goto err;
2719 hsize = skb_end_pointer(nskb) - nskb->head;
2720 if (skb_cow_head(nskb, doffset + headroom)) {
2721 kfree_skb(nskb);
2722 goto err;
2725 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2726 hsize;
2727 skb_release_head_state(nskb);
2728 __skb_push(nskb, doffset);
2729 } else {
2730 nskb = alloc_skb(hsize + doffset + headroom,
2731 GFP_ATOMIC);
2733 if (unlikely(!nskb))
2734 goto err;
2736 skb_reserve(nskb, headroom);
2737 __skb_put(nskb, doffset);
2740 if (segs)
2741 tail->next = nskb;
2742 else
2743 segs = nskb;
2744 tail = nskb;
2746 __copy_skb_header(nskb, skb);
2747 nskb->mac_len = skb->mac_len;
2749 /* nskb and skb might have different headroom */
2750 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2751 nskb->csum_start += skb_headroom(nskb) - headroom;
2753 skb_reset_mac_header(nskb);
2754 skb_set_network_header(nskb, skb->mac_len);
2755 nskb->transport_header = (nskb->network_header +
2756 skb_network_header_len(skb));
2757 skb_copy_from_linear_data(skb, nskb->data, doffset);
2759 if (fskb != skb_shinfo(skb)->frag_list)
2760 continue;
2762 if (!sg) {
2763 nskb->ip_summed = CHECKSUM_NONE;
2764 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2765 skb_put(nskb, len),
2766 len, 0);
2767 continue;
2770 frag = skb_shinfo(nskb)->frags;
2772 skb_copy_from_linear_data_offset(skb, offset,
2773 skb_put(nskb, hsize), hsize);
2775 while (pos < offset + len && i < nfrags) {
2776 *frag = skb_shinfo(skb)->frags[i];
2777 __skb_frag_ref(frag);
2778 size = skb_frag_size(frag);
2780 if (pos < offset) {
2781 frag->page_offset += offset - pos;
2782 skb_frag_size_sub(frag, offset - pos);
2785 skb_shinfo(nskb)->nr_frags++;
2787 if (pos + size <= offset + len) {
2788 i++;
2789 pos += size;
2790 } else {
2791 skb_frag_size_sub(frag, pos + size - (offset + len));
2792 goto skip_fraglist;
2795 frag++;
2798 if (pos < offset + len) {
2799 struct sk_buff *fskb2 = fskb;
2801 BUG_ON(pos + fskb->len != offset + len);
2803 pos += fskb->len;
2804 fskb = fskb->next;
2806 if (fskb2->next) {
2807 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2808 if (!fskb2)
2809 goto err;
2810 } else
2811 skb_get(fskb2);
2813 SKB_FRAG_ASSERT(nskb);
2814 skb_shinfo(nskb)->frag_list = fskb2;
2817 skip_fraglist:
2818 nskb->data_len = len - hsize;
2819 nskb->len += nskb->data_len;
2820 nskb->truesize += nskb->data_len;
2821 } while ((offset += len) < skb->len);
2823 return segs;
2825 err:
2826 while ((skb = segs)) {
2827 segs = skb->next;
2828 kfree_skb(skb);
2830 return ERR_PTR(err);
2832 EXPORT_SYMBOL_GPL(skb_segment);
2834 int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2836 struct sk_buff *p = *head;
2837 struct sk_buff *nskb;
2838 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2839 struct skb_shared_info *pinfo = skb_shinfo(p);
2840 unsigned int headroom;
2841 unsigned int len = skb_gro_len(skb);
2842 unsigned int offset = skb_gro_offset(skb);
2843 unsigned int headlen = skb_headlen(skb);
2845 if (p->len + len >= 65536)
2846 return -E2BIG;
2848 if (pinfo->frag_list)
2849 goto merge;
2850 else if (headlen <= offset) {
2851 skb_frag_t *frag;
2852 skb_frag_t *frag2;
2853 int i = skbinfo->nr_frags;
2854 int nr_frags = pinfo->nr_frags + i;
2856 offset -= headlen;
2858 if (nr_frags > MAX_SKB_FRAGS)
2859 return -E2BIG;
2861 pinfo->nr_frags = nr_frags;
2862 skbinfo->nr_frags = 0;
2864 frag = pinfo->frags + nr_frags;
2865 frag2 = skbinfo->frags + i;
2866 do {
2867 *--frag = *--frag2;
2868 } while (--i);
2870 frag->page_offset += offset;
2871 skb_frag_size_sub(frag, offset);
2873 skb->truesize -= skb->data_len;
2874 skb->len -= skb->data_len;
2875 skb->data_len = 0;
2877 NAPI_GRO_CB(skb)->free = 1;
2878 goto done;
2879 } else if (skb_gro_len(p) != pinfo->gso_size)
2880 return -E2BIG;
2882 headroom = skb_headroom(p);
2883 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2884 if (unlikely(!nskb))
2885 return -ENOMEM;
2887 __copy_skb_header(nskb, p);
2888 nskb->mac_len = p->mac_len;
2890 skb_reserve(nskb, headroom);
2891 __skb_put(nskb, skb_gro_offset(p));
2893 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2894 skb_set_network_header(nskb, skb_network_offset(p));
2895 skb_set_transport_header(nskb, skb_transport_offset(p));
2897 __skb_pull(p, skb_gro_offset(p));
2898 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2899 p->data - skb_mac_header(p));
2901 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2902 skb_shinfo(nskb)->frag_list = p;
2903 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2904 pinfo->gso_size = 0;
2905 skb_header_release(p);
2906 nskb->prev = p;
2908 nskb->data_len += p->len;
2909 nskb->truesize += p->len;
2910 nskb->len += p->len;
2912 *head = nskb;
2913 nskb->next = p->next;
2914 p->next = NULL;
2916 p = nskb;
2918 merge:
2919 if (offset > headlen) {
2920 unsigned int eat = offset - headlen;
2922 skbinfo->frags[0].page_offset += eat;
2923 skb_frag_size_sub(&skbinfo->frags[0], eat);
2924 skb->data_len -= eat;
2925 skb->len -= eat;
2926 offset = headlen;
2929 __skb_pull(skb, offset);
2931 p->prev->next = skb;
2932 p->prev = skb;
2933 skb_header_release(skb);
2935 done:
2936 NAPI_GRO_CB(p)->count++;
2937 p->data_len += len;
2938 p->truesize += len;
2939 p->len += len;
2941 NAPI_GRO_CB(skb)->same_flow = 1;
2942 return 0;
2944 EXPORT_SYMBOL_GPL(skb_gro_receive);
2946 void __init skb_init(void)
2948 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2949 sizeof(struct sk_buff),
2951 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2952 NULL);
2953 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2954 (2*sizeof(struct sk_buff)) +
2955 sizeof(atomic_t),
2957 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2958 NULL);
2962 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2963 * @skb: Socket buffer containing the buffers to be mapped
2964 * @sg: The scatter-gather list to map into
2965 * @offset: The offset into the buffer's contents to start mapping
2966 * @len: Length of buffer space to be mapped
2968 * Fill the specified scatter-gather list with mappings/pointers into a
2969 * region of the buffer space attached to a socket buffer.
2971 static int
2972 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2974 int start = skb_headlen(skb);
2975 int i, copy = start - offset;
2976 struct sk_buff *frag_iter;
2977 int elt = 0;
2979 if (copy > 0) {
2980 if (copy > len)
2981 copy = len;
2982 sg_set_buf(sg, skb->data + offset, copy);
2983 elt++;
2984 if ((len -= copy) == 0)
2985 return elt;
2986 offset += copy;
2989 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2990 int end;
2992 WARN_ON(start > offset + len);
2994 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2995 if ((copy = end - offset) > 0) {
2996 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2998 if (copy > len)
2999 copy = len;
3000 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3001 frag->page_offset+offset-start);
3002 elt++;
3003 if (!(len -= copy))
3004 return elt;
3005 offset += copy;
3007 start = end;
3010 skb_walk_frags(skb, frag_iter) {
3011 int end;
3013 WARN_ON(start > offset + len);
3015 end = start + frag_iter->len;
3016 if ((copy = end - offset) > 0) {
3017 if (copy > len)
3018 copy = len;
3019 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3020 copy);
3021 if ((len -= copy) == 0)
3022 return elt;
3023 offset += copy;
3025 start = end;
3027 BUG_ON(len);
3028 return elt;
3031 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3033 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3035 sg_mark_end(&sg[nsg - 1]);
3037 return nsg;
3039 EXPORT_SYMBOL_GPL(skb_to_sgvec);
3042 * skb_cow_data - Check that a socket buffer's data buffers are writable
3043 * @skb: The socket buffer to check.
3044 * @tailbits: Amount of trailing space to be added
3045 * @trailer: Returned pointer to the skb where the @tailbits space begins
3047 * Make sure that the data buffers attached to a socket buffer are
3048 * writable. If they are not, private copies are made of the data buffers
3049 * and the socket buffer is set to use these instead.
3051 * If @tailbits is given, make sure that there is space to write @tailbits
3052 * bytes of data beyond current end of socket buffer. @trailer will be
3053 * set to point to the skb in which this space begins.
3055 * The number of scatterlist elements required to completely map the
3056 * COW'd and extended socket buffer will be returned.
3058 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3060 int copyflag;
3061 int elt;
3062 struct sk_buff *skb1, **skb_p;
3064 /* If skb is cloned or its head is paged, reallocate
3065 * head pulling out all the pages (pages are considered not writable
3066 * at the moment even if they are anonymous).
3068 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3069 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3070 return -ENOMEM;
3072 /* Easy case. Most of packets will go this way. */
3073 if (!skb_has_frag_list(skb)) {
3074 /* A little of trouble, not enough of space for trailer.
3075 * This should not happen, when stack is tuned to generate
3076 * good frames. OK, on miss we reallocate and reserve even more
3077 * space, 128 bytes is fair. */
3079 if (skb_tailroom(skb) < tailbits &&
3080 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3081 return -ENOMEM;
3083 /* Voila! */
3084 *trailer = skb;
3085 return 1;
3088 /* Misery. We are in troubles, going to mincer fragments... */
3090 elt = 1;
3091 skb_p = &skb_shinfo(skb)->frag_list;
3092 copyflag = 0;
3094 while ((skb1 = *skb_p) != NULL) {
3095 int ntail = 0;
3097 /* The fragment is partially pulled by someone,
3098 * this can happen on input. Copy it and everything
3099 * after it. */
3101 if (skb_shared(skb1))
3102 copyflag = 1;
3104 /* If the skb is the last, worry about trailer. */
3106 if (skb1->next == NULL && tailbits) {
3107 if (skb_shinfo(skb1)->nr_frags ||
3108 skb_has_frag_list(skb1) ||
3109 skb_tailroom(skb1) < tailbits)
3110 ntail = tailbits + 128;
3113 if (copyflag ||
3114 skb_cloned(skb1) ||
3115 ntail ||
3116 skb_shinfo(skb1)->nr_frags ||
3117 skb_has_frag_list(skb1)) {
3118 struct sk_buff *skb2;
3120 /* Fuck, we are miserable poor guys... */
3121 if (ntail == 0)
3122 skb2 = skb_copy(skb1, GFP_ATOMIC);
3123 else
3124 skb2 = skb_copy_expand(skb1,
3125 skb_headroom(skb1),
3126 ntail,
3127 GFP_ATOMIC);
3128 if (unlikely(skb2 == NULL))
3129 return -ENOMEM;
3131 if (skb1->sk)
3132 skb_set_owner_w(skb2, skb1->sk);
3134 /* Looking around. Are we still alive?
3135 * OK, link new skb, drop old one */
3137 skb2->next = skb1->next;
3138 *skb_p = skb2;
3139 kfree_skb(skb1);
3140 skb1 = skb2;
3142 elt++;
3143 *trailer = skb1;
3144 skb_p = &skb1->next;
3147 return elt;
3149 EXPORT_SYMBOL_GPL(skb_cow_data);
3151 static void sock_rmem_free(struct sk_buff *skb)
3153 struct sock *sk = skb->sk;
3155 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3159 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3161 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3163 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3164 (unsigned)sk->sk_rcvbuf)
3165 return -ENOMEM;
3167 skb_orphan(skb);
3168 skb->sk = sk;
3169 skb->destructor = sock_rmem_free;
3170 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3172 /* before exiting rcu section, make sure dst is refcounted */
3173 skb_dst_force(skb);
3175 skb_queue_tail(&sk->sk_error_queue, skb);
3176 if (!sock_flag(sk, SOCK_DEAD))
3177 sk->sk_data_ready(sk, skb->len);
3178 return 0;
3180 EXPORT_SYMBOL(sock_queue_err_skb);
3182 void skb_tstamp_tx(struct sk_buff *orig_skb,
3183 struct skb_shared_hwtstamps *hwtstamps)
3185 struct sock *sk = orig_skb->sk;
3186 struct sock_exterr_skb *serr;
3187 struct sk_buff *skb;
3188 int err;
3190 if (!sk)
3191 return;
3193 skb = skb_clone(orig_skb, GFP_ATOMIC);
3194 if (!skb)
3195 return;
3197 if (hwtstamps) {
3198 *skb_hwtstamps(skb) =
3199 *hwtstamps;
3200 } else {
3202 * no hardware time stamps available,
3203 * so keep the shared tx_flags and only
3204 * store software time stamp
3206 skb->tstamp = ktime_get_real();
3209 serr = SKB_EXT_ERR(skb);
3210 memset(serr, 0, sizeof(*serr));
3211 serr->ee.ee_errno = ENOMSG;
3212 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3214 err = sock_queue_err_skb(sk, skb);
3216 if (err)
3217 kfree_skb(skb);
3219 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3221 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3223 struct sock *sk = skb->sk;
3224 struct sock_exterr_skb *serr;
3225 int err;
3227 skb->wifi_acked_valid = 1;
3228 skb->wifi_acked = acked;
3230 serr = SKB_EXT_ERR(skb);
3231 memset(serr, 0, sizeof(*serr));
3232 serr->ee.ee_errno = ENOMSG;
3233 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3235 err = sock_queue_err_skb(sk, skb);
3236 if (err)
3237 kfree_skb(skb);
3239 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3243 * skb_partial_csum_set - set up and verify partial csum values for packet
3244 * @skb: the skb to set
3245 * @start: the number of bytes after skb->data to start checksumming.
3246 * @off: the offset from start to place the checksum.
3248 * For untrusted partially-checksummed packets, we need to make sure the values
3249 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3251 * This function checks and sets those values and skb->ip_summed: if this
3252 * returns false you should drop the packet.
3254 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3256 if (unlikely(start > skb_headlen(skb)) ||
3257 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3258 if (net_ratelimit())
3259 printk(KERN_WARNING
3260 "bad partial csum: csum=%u/%u len=%u\n",
3261 start, off, skb_headlen(skb));
3262 return false;
3264 skb->ip_summed = CHECKSUM_PARTIAL;
3265 skb->csum_start = skb_headroom(skb) + start;
3266 skb->csum_offset = off;
3267 return true;
3269 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3271 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3273 if (net_ratelimit())
3274 pr_warning("%s: received packets cannot be forwarded"
3275 " while LRO is enabled\n", skb->dev->name);
3277 EXPORT_SYMBOL(__skb_warn_lro_forwarding);