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>
8 * Alan Cox : Fixed the worst of the load
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
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 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
45 #include <linux/interrupt.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/tcp.h>
50 #include <linux/udp.h>
51 #include <linux/sctp.h>
52 #include <linux/netdevice.h>
53 #ifdef CONFIG_NET_CLS_ACT
54 #include <net/pkt_sched.h>
56 #include <linux/string.h>
57 #include <linux/skbuff.h>
58 #include <linux/splice.h>
59 #include <linux/cache.h>
60 #include <linux/rtnetlink.h>
61 #include <linux/init.h>
62 #include <linux/scatterlist.h>
63 #include <linux/errqueue.h>
64 #include <linux/prefetch.h>
65 #include <linux/if_vlan.h>
67 #include <net/protocol.h>
70 #include <net/checksum.h>
71 #include <net/ip6_checksum.h>
74 #include <linux/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
80 struct kmem_cache *skbuff_head_cache __ro_after_init;
81 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
82 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83 EXPORT_SYMBOL(sysctl_max_skb_frags);
86 * skb_panic - private function for out-of-line support
90 * @msg: skb_over_panic or skb_under_panic
92 * Out-of-line support for skb_put() and skb_push().
93 * Called via the wrapper skb_over_panic() or skb_under_panic().
94 * Keep out of line to prevent kernel bloat.
95 * __builtin_return_address is not used because it is not always reliable.
97 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
107 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109 skb_panic(skb, sz, addr, __func__);
112 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
119 * the caller if emergency pfmemalloc reserves are being used. If it is and
120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
121 * may be used. Otherwise, the packet data may be discarded until enough
124 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
127 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
131 bool ret_pfmemalloc = false;
134 * Try a regular allocation, when that fails and we're not entitled
135 * to the reserves, fail.
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
143 /* Try again but now we are using pfmemalloc reserves */
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
149 *pfmemalloc = ret_pfmemalloc;
154 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
155 * 'private' fields and also do memory statistics to find all the
161 * __alloc_skb - allocate a network buffer
162 * @size: size to allocate
163 * @gfp_mask: allocation mask
164 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
165 * instead of head cache and allocate a cloned (child) skb.
166 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
167 * allocations in case the data is required for writeback
168 * @node: numa node to allocate memory on
170 * Allocate a new &sk_buff. The returned buffer has no headroom and a
171 * tail room of at least size bytes. The object has a reference count
172 * of one. The return is the buffer. On a failure the return is %NULL.
174 * Buffers may only be allocated from interrupts using a @gfp_mask of
177 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
180 struct kmem_cache *cache;
181 struct skb_shared_info *shinfo;
186 cache = (flags & SKB_ALLOC_FCLONE)
187 ? skbuff_fclone_cache : skbuff_head_cache;
189 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
190 gfp_mask |= __GFP_MEMALLOC;
193 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
198 /* We do our best to align skb_shared_info on a separate cache
199 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
200 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
201 * Both skb->head and skb_shared_info are cache line aligned.
203 size = SKB_DATA_ALIGN(size);
204 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
205 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
208 /* kmalloc(size) might give us more room than requested.
209 * Put skb_shared_info exactly at the end of allocated zone,
210 * to allow max possible filling before reallocation.
212 size = SKB_WITH_OVERHEAD(ksize(data));
213 prefetchw(data + size);
216 * Only clear those fields we need to clear, not those that we will
217 * actually initialise below. Hence, don't put any more fields after
218 * the tail pointer in struct sk_buff!
220 memset(skb, 0, offsetof(struct sk_buff, tail));
221 /* Account for allocated memory : skb + skb->head */
222 skb->truesize = SKB_TRUESIZE(size);
223 skb->pfmemalloc = pfmemalloc;
224 refcount_set(&skb->users, 1);
227 skb_reset_tail_pointer(skb);
228 skb->end = skb->tail + size;
229 skb->mac_header = (typeof(skb->mac_header))~0U;
230 skb->transport_header = (typeof(skb->transport_header))~0U;
232 /* make sure we initialize shinfo sequentially */
233 shinfo = skb_shinfo(skb);
234 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
235 atomic_set(&shinfo->dataref, 1);
237 if (flags & SKB_ALLOC_FCLONE) {
238 struct sk_buff_fclones *fclones;
240 fclones = container_of(skb, struct sk_buff_fclones, skb1);
242 skb->fclone = SKB_FCLONE_ORIG;
243 refcount_set(&fclones->fclone_ref, 1);
245 fclones->skb2.fclone = SKB_FCLONE_CLONE;
250 kmem_cache_free(cache, skb);
254 EXPORT_SYMBOL(__alloc_skb);
257 * __build_skb - build a network buffer
258 * @data: data buffer provided by caller
259 * @frag_size: size of data, or 0 if head was kmalloced
261 * Allocate a new &sk_buff. Caller provides space holding head and
262 * skb_shared_info. @data must have been allocated by kmalloc() only if
263 * @frag_size is 0, otherwise data should come from the page allocator
265 * The return is the new skb buffer.
266 * On a failure the return is %NULL, and @data is not freed.
268 * Before IO, driver allocates only data buffer where NIC put incoming frame
269 * Driver should add room at head (NET_SKB_PAD) and
270 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
271 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
272 * before giving packet to stack.
273 * RX rings only contains data buffers, not full skbs.
275 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
277 struct skb_shared_info *shinfo;
279 unsigned int size = frag_size ? : ksize(data);
281 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
285 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
287 memset(skb, 0, offsetof(struct sk_buff, tail));
288 skb->truesize = SKB_TRUESIZE(size);
289 refcount_set(&skb->users, 1);
292 skb_reset_tail_pointer(skb);
293 skb->end = skb->tail + size;
294 skb->mac_header = (typeof(skb->mac_header))~0U;
295 skb->transport_header = (typeof(skb->transport_header))~0U;
297 /* make sure we initialize shinfo sequentially */
298 shinfo = skb_shinfo(skb);
299 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
300 atomic_set(&shinfo->dataref, 1);
305 /* build_skb() is wrapper over __build_skb(), that specifically
306 * takes care of skb->head and skb->pfmemalloc
307 * This means that if @frag_size is not zero, then @data must be backed
308 * by a page fragment, not kmalloc() or vmalloc()
310 struct sk_buff *build_skb(void *data, unsigned int frag_size)
312 struct sk_buff *skb = __build_skb(data, frag_size);
314 if (skb && frag_size) {
316 if (page_is_pfmemalloc(virt_to_head_page(data)))
321 EXPORT_SYMBOL(build_skb);
323 #define NAPI_SKB_CACHE_SIZE 64
325 struct napi_alloc_cache {
326 struct page_frag_cache page;
327 unsigned int skb_count;
328 void *skb_cache[NAPI_SKB_CACHE_SIZE];
331 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
332 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
334 static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
336 struct page_frag_cache *nc;
340 local_irq_save(flags);
341 nc = this_cpu_ptr(&netdev_alloc_cache);
342 data = page_frag_alloc(nc, fragsz, gfp_mask);
343 local_irq_restore(flags);
348 * netdev_alloc_frag - allocate a page fragment
349 * @fragsz: fragment size
351 * Allocates a frag from a page for receive buffer.
352 * Uses GFP_ATOMIC allocations.
354 void *netdev_alloc_frag(unsigned int fragsz)
356 return __netdev_alloc_frag(fragsz, GFP_ATOMIC);
358 EXPORT_SYMBOL(netdev_alloc_frag);
360 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
362 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
364 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
367 void *napi_alloc_frag(unsigned int fragsz)
369 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
371 EXPORT_SYMBOL(napi_alloc_frag);
374 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
375 * @dev: network device to receive on
376 * @len: length to allocate
377 * @gfp_mask: get_free_pages mask, passed to alloc_skb
379 * Allocate a new &sk_buff and assign it a usage count of one. The
380 * buffer has NET_SKB_PAD headroom built in. Users should allocate
381 * the headroom they think they need without accounting for the
382 * built in space. The built in space is used for optimisations.
384 * %NULL is returned if there is no free memory.
386 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
389 struct page_frag_cache *nc;
397 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
398 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
399 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
405 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
406 len = SKB_DATA_ALIGN(len);
408 if (sk_memalloc_socks())
409 gfp_mask |= __GFP_MEMALLOC;
411 local_irq_save(flags);
413 nc = this_cpu_ptr(&netdev_alloc_cache);
414 data = page_frag_alloc(nc, len, gfp_mask);
415 pfmemalloc = nc->pfmemalloc;
417 local_irq_restore(flags);
422 skb = __build_skb(data, len);
423 if (unlikely(!skb)) {
428 /* use OR instead of assignment to avoid clearing of bits in mask */
434 skb_reserve(skb, NET_SKB_PAD);
440 EXPORT_SYMBOL(__netdev_alloc_skb);
443 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
444 * @napi: napi instance this buffer was allocated for
445 * @len: length to allocate
446 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
448 * Allocate a new sk_buff for use in NAPI receive. This buffer will
449 * attempt to allocate the head from a special reserved region used
450 * only for NAPI Rx allocation. By doing this we can save several
451 * CPU cycles by avoiding having to disable and re-enable IRQs.
453 * %NULL is returned if there is no free memory.
455 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
458 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
462 len += NET_SKB_PAD + NET_IP_ALIGN;
464 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
465 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
466 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
472 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
473 len = SKB_DATA_ALIGN(len);
475 if (sk_memalloc_socks())
476 gfp_mask |= __GFP_MEMALLOC;
478 data = page_frag_alloc(&nc->page, len, gfp_mask);
482 skb = __build_skb(data, len);
483 if (unlikely(!skb)) {
488 /* use OR instead of assignment to avoid clearing of bits in mask */
489 if (nc->page.pfmemalloc)
494 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
495 skb->dev = napi->dev;
500 EXPORT_SYMBOL(__napi_alloc_skb);
502 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
503 int size, unsigned int truesize)
505 skb_fill_page_desc(skb, i, page, off, size);
507 skb->data_len += size;
508 skb->truesize += truesize;
510 EXPORT_SYMBOL(skb_add_rx_frag);
512 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
513 unsigned int truesize)
515 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
517 skb_frag_size_add(frag, size);
519 skb->data_len += size;
520 skb->truesize += truesize;
522 EXPORT_SYMBOL(skb_coalesce_rx_frag);
524 static void skb_drop_list(struct sk_buff **listp)
526 kfree_skb_list(*listp);
530 static inline void skb_drop_fraglist(struct sk_buff *skb)
532 skb_drop_list(&skb_shinfo(skb)->frag_list);
535 static void skb_clone_fraglist(struct sk_buff *skb)
537 struct sk_buff *list;
539 skb_walk_frags(skb, list)
543 static void skb_free_head(struct sk_buff *skb)
545 unsigned char *head = skb->head;
553 static void skb_release_data(struct sk_buff *skb)
555 struct skb_shared_info *shinfo = skb_shinfo(skb);
559 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
563 for (i = 0; i < shinfo->nr_frags; i++)
564 __skb_frag_unref(&shinfo->frags[i]);
566 if (shinfo->frag_list)
567 kfree_skb_list(shinfo->frag_list);
569 skb_zcopy_clear(skb, true);
574 * Free an skbuff by memory without cleaning the state.
576 static void kfree_skbmem(struct sk_buff *skb)
578 struct sk_buff_fclones *fclones;
580 switch (skb->fclone) {
581 case SKB_FCLONE_UNAVAILABLE:
582 kmem_cache_free(skbuff_head_cache, skb);
585 case SKB_FCLONE_ORIG:
586 fclones = container_of(skb, struct sk_buff_fclones, skb1);
588 /* We usually free the clone (TX completion) before original skb
589 * This test would have no chance to be true for the clone,
590 * while here, branch prediction will be good.
592 if (refcount_read(&fclones->fclone_ref) == 1)
596 default: /* SKB_FCLONE_CLONE */
597 fclones = container_of(skb, struct sk_buff_fclones, skb2);
600 if (!refcount_dec_and_test(&fclones->fclone_ref))
603 kmem_cache_free(skbuff_fclone_cache, fclones);
606 void skb_release_head_state(struct sk_buff *skb)
610 if (skb->destructor) {
612 skb->destructor(skb);
614 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
615 nf_conntrack_put(skb_nfct(skb));
617 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
618 nf_bridge_put(skb->nf_bridge);
622 /* Free everything but the sk_buff shell. */
623 static void skb_release_all(struct sk_buff *skb)
625 skb_release_head_state(skb);
626 if (likely(skb->head))
627 skb_release_data(skb);
631 * __kfree_skb - private function
634 * Free an sk_buff. Release anything attached to the buffer.
635 * Clean the state. This is an internal helper function. Users should
636 * always call kfree_skb
639 void __kfree_skb(struct sk_buff *skb)
641 skb_release_all(skb);
644 EXPORT_SYMBOL(__kfree_skb);
647 * kfree_skb - free an sk_buff
648 * @skb: buffer to free
650 * Drop a reference to the buffer and free it if the usage count has
653 void kfree_skb(struct sk_buff *skb)
658 trace_kfree_skb(skb, __builtin_return_address(0));
661 EXPORT_SYMBOL(kfree_skb);
663 void kfree_skb_list(struct sk_buff *segs)
666 struct sk_buff *next = segs->next;
672 EXPORT_SYMBOL(kfree_skb_list);
675 * skb_tx_error - report an sk_buff xmit error
676 * @skb: buffer that triggered an error
678 * Report xmit error if a device callback is tracking this skb.
679 * skb must be freed afterwards.
681 void skb_tx_error(struct sk_buff *skb)
683 skb_zcopy_clear(skb, true);
685 EXPORT_SYMBOL(skb_tx_error);
688 * consume_skb - free an skbuff
689 * @skb: buffer to free
691 * Drop a ref to the buffer and free it if the usage count has hit zero
692 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
693 * is being dropped after a failure and notes that
695 void consume_skb(struct sk_buff *skb)
700 trace_consume_skb(skb);
703 EXPORT_SYMBOL(consume_skb);
706 * consume_stateless_skb - free an skbuff, assuming it is stateless
707 * @skb: buffer to free
709 * Alike consume_skb(), but this variant assumes that this is the last
710 * skb reference and all the head states have been already dropped
712 void __consume_stateless_skb(struct sk_buff *skb)
714 trace_consume_skb(skb);
715 skb_release_data(skb);
719 void __kfree_skb_flush(void)
721 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
723 /* flush skb_cache if containing objects */
725 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
731 static inline void _kfree_skb_defer(struct sk_buff *skb)
733 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
735 /* drop skb->head and call any destructors for packet */
736 skb_release_all(skb);
738 /* record skb to CPU local list */
739 nc->skb_cache[nc->skb_count++] = skb;
742 /* SLUB writes into objects when freeing */
746 /* flush skb_cache if it is filled */
747 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
748 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
753 void __kfree_skb_defer(struct sk_buff *skb)
755 _kfree_skb_defer(skb);
758 void napi_consume_skb(struct sk_buff *skb, int budget)
763 /* Zero budget indicate non-NAPI context called us, like netpoll */
764 if (unlikely(!budget)) {
765 dev_consume_skb_any(skb);
772 /* if reaching here SKB is ready to free */
773 trace_consume_skb(skb);
775 /* if SKB is a clone, don't handle this case */
776 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
781 _kfree_skb_defer(skb);
783 EXPORT_SYMBOL(napi_consume_skb);
785 /* Make sure a field is enclosed inside headers_start/headers_end section */
786 #define CHECK_SKB_FIELD(field) \
787 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
788 offsetof(struct sk_buff, headers_start)); \
789 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
790 offsetof(struct sk_buff, headers_end)); \
792 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
794 new->tstamp = old->tstamp;
795 /* We do not copy old->sk */
797 memcpy(new->cb, old->cb, sizeof(old->cb));
798 skb_dst_copy(new, old);
800 new->sp = secpath_get(old->sp);
802 __nf_copy(new, old, false);
804 /* Note : this field could be in headers_start/headers_end section
805 * It is not yet because we do not want to have a 16 bit hole
807 new->queue_mapping = old->queue_mapping;
809 memcpy(&new->headers_start, &old->headers_start,
810 offsetof(struct sk_buff, headers_end) -
811 offsetof(struct sk_buff, headers_start));
812 CHECK_SKB_FIELD(protocol);
813 CHECK_SKB_FIELD(csum);
814 CHECK_SKB_FIELD(hash);
815 CHECK_SKB_FIELD(priority);
816 CHECK_SKB_FIELD(skb_iif);
817 CHECK_SKB_FIELD(vlan_proto);
818 CHECK_SKB_FIELD(vlan_tci);
819 CHECK_SKB_FIELD(transport_header);
820 CHECK_SKB_FIELD(network_header);
821 CHECK_SKB_FIELD(mac_header);
822 CHECK_SKB_FIELD(inner_protocol);
823 CHECK_SKB_FIELD(inner_transport_header);
824 CHECK_SKB_FIELD(inner_network_header);
825 CHECK_SKB_FIELD(inner_mac_header);
826 CHECK_SKB_FIELD(mark);
827 #ifdef CONFIG_NETWORK_SECMARK
828 CHECK_SKB_FIELD(secmark);
830 #ifdef CONFIG_NET_RX_BUSY_POLL
831 CHECK_SKB_FIELD(napi_id);
834 CHECK_SKB_FIELD(sender_cpu);
836 #ifdef CONFIG_NET_SCHED
837 CHECK_SKB_FIELD(tc_index);
843 * You should not add any new code to this function. Add it to
844 * __copy_skb_header above instead.
846 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
848 #define C(x) n->x = skb->x
850 n->next = n->prev = NULL;
852 __copy_skb_header(n, skb);
857 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
862 n->destructor = NULL;
869 refcount_set(&n->users, 1);
871 atomic_inc(&(skb_shinfo(skb)->dataref));
879 * skb_morph - morph one skb into another
880 * @dst: the skb to receive the contents
881 * @src: the skb to supply the contents
883 * This is identical to skb_clone except that the target skb is
884 * supplied by the user.
886 * The target skb is returned upon exit.
888 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
890 skb_release_all(dst);
891 return __skb_clone(dst, src);
893 EXPORT_SYMBOL_GPL(skb_morph);
895 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
897 unsigned long max_pg, num_pg, new_pg, old_pg;
898 struct user_struct *user;
900 if (capable(CAP_IPC_LOCK) || !size)
903 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
904 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
905 user = mmp->user ? : current_user();
908 old_pg = atomic_long_read(&user->locked_vm);
909 new_pg = old_pg + num_pg;
912 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
916 mmp->user = get_uid(user);
917 mmp->num_pg = num_pg;
919 mmp->num_pg += num_pg;
924 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
926 void mm_unaccount_pinned_pages(struct mmpin *mmp)
929 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
933 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
935 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
937 struct ubuf_info *uarg;
940 WARN_ON_ONCE(!in_task());
942 if (!sock_flag(sk, SOCK_ZEROCOPY))
945 skb = sock_omalloc(sk, 0, GFP_KERNEL);
949 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
950 uarg = (void *)skb->cb;
951 uarg->mmp.user = NULL;
953 if (mm_account_pinned_pages(&uarg->mmp, size)) {
958 uarg->callback = sock_zerocopy_callback;
959 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
961 uarg->bytelen = size;
963 refcount_set(&uarg->refcnt, 1);
968 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
970 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
972 return container_of((void *)uarg, struct sk_buff, cb);
975 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
976 struct ubuf_info *uarg)
979 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
982 /* realloc only when socket is locked (TCP, UDP cork),
983 * so uarg->len and sk_zckey access is serialized
985 if (!sock_owned_by_user(sk)) {
990 bytelen = uarg->bytelen + size;
991 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
992 /* TCP can create new skb to attach new uarg */
993 if (sk->sk_type == SOCK_STREAM)
998 next = (u32)atomic_read(&sk->sk_zckey);
999 if ((u32)(uarg->id + uarg->len) == next) {
1000 if (mm_account_pinned_pages(&uarg->mmp, size))
1003 uarg->bytelen = bytelen;
1004 atomic_set(&sk->sk_zckey, ++next);
1005 sock_zerocopy_get(uarg);
1011 return sock_zerocopy_alloc(sk, size);
1013 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1015 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1017 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1021 old_lo = serr->ee.ee_info;
1022 old_hi = serr->ee.ee_data;
1023 sum_len = old_hi - old_lo + 1ULL + len;
1025 if (sum_len >= (1ULL << 32))
1028 if (lo != old_hi + 1)
1031 serr->ee.ee_data += len;
1035 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1037 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1038 struct sock_exterr_skb *serr;
1039 struct sock *sk = skb->sk;
1040 struct sk_buff_head *q;
1041 unsigned long flags;
1045 mm_unaccount_pinned_pages(&uarg->mmp);
1047 /* if !len, there was only 1 call, and it was aborted
1048 * so do not queue a completion notification
1050 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1055 hi = uarg->id + len - 1;
1057 serr = SKB_EXT_ERR(skb);
1058 memset(serr, 0, sizeof(*serr));
1059 serr->ee.ee_errno = 0;
1060 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1061 serr->ee.ee_data = hi;
1062 serr->ee.ee_info = lo;
1064 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1066 q = &sk->sk_error_queue;
1067 spin_lock_irqsave(&q->lock, flags);
1068 tail = skb_peek_tail(q);
1069 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1070 !skb_zerocopy_notify_extend(tail, lo, len)) {
1071 __skb_queue_tail(q, skb);
1074 spin_unlock_irqrestore(&q->lock, flags);
1076 sk->sk_error_report(sk);
1082 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1084 void sock_zerocopy_put(struct ubuf_info *uarg)
1086 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1088 uarg->callback(uarg, uarg->zerocopy);
1090 consume_skb(skb_from_uarg(uarg));
1093 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1095 void sock_zerocopy_put_abort(struct ubuf_info *uarg)
1098 struct sock *sk = skb_from_uarg(uarg)->sk;
1100 atomic_dec(&sk->sk_zckey);
1103 sock_zerocopy_put(uarg);
1106 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1108 extern int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb,
1109 struct iov_iter *from, size_t length);
1111 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1112 struct msghdr *msg, int len,
1113 struct ubuf_info *uarg)
1115 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1116 struct iov_iter orig_iter = msg->msg_iter;
1117 int err, orig_len = skb->len;
1119 /* An skb can only point to one uarg. This edge case happens when
1120 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1122 if (orig_uarg && uarg != orig_uarg)
1125 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1126 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1127 struct sock *save_sk = skb->sk;
1129 /* Streams do not free skb on error. Reset to prev state. */
1130 msg->msg_iter = orig_iter;
1132 ___pskb_trim(skb, orig_len);
1137 skb_zcopy_set(skb, uarg);
1138 return skb->len - orig_len;
1140 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1142 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1145 if (skb_zcopy(orig)) {
1146 if (skb_zcopy(nskb)) {
1147 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1152 if (skb_uarg(nskb) == skb_uarg(orig))
1154 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1157 skb_zcopy_set(nskb, skb_uarg(orig));
1163 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1164 * @skb: the skb to modify
1165 * @gfp_mask: allocation priority
1167 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1168 * It will copy all frags into kernel and drop the reference
1169 * to userspace pages.
1171 * If this function is called from an interrupt gfp_mask() must be
1174 * Returns 0 on success or a negative error code on failure
1175 * to allocate kernel memory to copy to.
1177 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1179 int num_frags = skb_shinfo(skb)->nr_frags;
1180 struct page *page, *head = NULL;
1184 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1190 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1191 for (i = 0; i < new_frags; i++) {
1192 page = alloc_page(gfp_mask);
1195 struct page *next = (struct page *)page_private(head);
1201 set_page_private(page, (unsigned long)head);
1207 for (i = 0; i < num_frags; i++) {
1208 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1209 u32 p_off, p_len, copied;
1213 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f),
1214 p, p_off, p_len, copied) {
1216 vaddr = kmap_atomic(p);
1218 while (done < p_len) {
1219 if (d_off == PAGE_SIZE) {
1221 page = (struct page *)page_private(page);
1223 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1224 memcpy(page_address(page) + d_off,
1225 vaddr + p_off + done, copy);
1229 kunmap_atomic(vaddr);
1233 /* skb frags release userspace buffers */
1234 for (i = 0; i < num_frags; i++)
1235 skb_frag_unref(skb, i);
1237 /* skb frags point to kernel buffers */
1238 for (i = 0; i < new_frags - 1; i++) {
1239 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1240 head = (struct page *)page_private(head);
1242 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1243 skb_shinfo(skb)->nr_frags = new_frags;
1246 skb_zcopy_clear(skb, false);
1249 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1252 * skb_clone - duplicate an sk_buff
1253 * @skb: buffer to clone
1254 * @gfp_mask: allocation priority
1256 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1257 * copies share the same packet data but not structure. The new
1258 * buffer has a reference count of 1. If the allocation fails the
1259 * function returns %NULL otherwise the new buffer is returned.
1261 * If this function is called from an interrupt gfp_mask() must be
1265 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1267 struct sk_buff_fclones *fclones = container_of(skb,
1268 struct sk_buff_fclones,
1272 if (skb_orphan_frags(skb, gfp_mask))
1275 if (skb->fclone == SKB_FCLONE_ORIG &&
1276 refcount_read(&fclones->fclone_ref) == 1) {
1278 refcount_set(&fclones->fclone_ref, 2);
1280 if (skb_pfmemalloc(skb))
1281 gfp_mask |= __GFP_MEMALLOC;
1283 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1287 n->fclone = SKB_FCLONE_UNAVAILABLE;
1290 return __skb_clone(n, skb);
1292 EXPORT_SYMBOL(skb_clone);
1294 void skb_headers_offset_update(struct sk_buff *skb, int off)
1296 /* Only adjust this if it actually is csum_start rather than csum */
1297 if (skb->ip_summed == CHECKSUM_PARTIAL)
1298 skb->csum_start += off;
1299 /* {transport,network,mac}_header and tail are relative to skb->head */
1300 skb->transport_header += off;
1301 skb->network_header += off;
1302 if (skb_mac_header_was_set(skb))
1303 skb->mac_header += off;
1304 skb->inner_transport_header += off;
1305 skb->inner_network_header += off;
1306 skb->inner_mac_header += off;
1308 EXPORT_SYMBOL(skb_headers_offset_update);
1310 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1312 __copy_skb_header(new, old);
1314 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1315 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1316 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1318 EXPORT_SYMBOL(skb_copy_header);
1320 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1322 if (skb_pfmemalloc(skb))
1323 return SKB_ALLOC_RX;
1328 * skb_copy - create private copy of an sk_buff
1329 * @skb: buffer to copy
1330 * @gfp_mask: allocation priority
1332 * Make a copy of both an &sk_buff and its data. This is used when the
1333 * caller wishes to modify the data and needs a private copy of the
1334 * data to alter. Returns %NULL on failure or the pointer to the buffer
1335 * on success. The returned buffer has a reference count of 1.
1337 * As by-product this function converts non-linear &sk_buff to linear
1338 * one, so that &sk_buff becomes completely private and caller is allowed
1339 * to modify all the data of returned buffer. This means that this
1340 * function is not recommended for use in circumstances when only
1341 * header is going to be modified. Use pskb_copy() instead.
1344 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1346 int headerlen = skb_headroom(skb);
1347 unsigned int size = skb_end_offset(skb) + skb->data_len;
1348 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1349 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1354 /* Set the data pointer */
1355 skb_reserve(n, headerlen);
1356 /* Set the tail pointer and length */
1357 skb_put(n, skb->len);
1359 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1361 skb_copy_header(n, skb);
1364 EXPORT_SYMBOL(skb_copy);
1367 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1368 * @skb: buffer to copy
1369 * @headroom: headroom of new skb
1370 * @gfp_mask: allocation priority
1371 * @fclone: if true allocate the copy of the skb from the fclone
1372 * cache instead of the head cache; it is recommended to set this
1373 * to true for the cases where the copy will likely be cloned
1375 * Make a copy of both an &sk_buff and part of its data, located
1376 * in header. Fragmented data remain shared. This is used when
1377 * the caller wishes to modify only header of &sk_buff and needs
1378 * private copy of the header to alter. Returns %NULL on failure
1379 * or the pointer to the buffer on success.
1380 * The returned buffer has a reference count of 1.
1383 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1384 gfp_t gfp_mask, bool fclone)
1386 unsigned int size = skb_headlen(skb) + headroom;
1387 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1388 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1393 /* Set the data pointer */
1394 skb_reserve(n, headroom);
1395 /* Set the tail pointer and length */
1396 skb_put(n, skb_headlen(skb));
1397 /* Copy the bytes */
1398 skb_copy_from_linear_data(skb, n->data, n->len);
1400 n->truesize += skb->data_len;
1401 n->data_len = skb->data_len;
1404 if (skb_shinfo(skb)->nr_frags) {
1407 if (skb_orphan_frags(skb, gfp_mask) ||
1408 skb_zerocopy_clone(n, skb, gfp_mask)) {
1413 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1414 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1415 skb_frag_ref(skb, i);
1417 skb_shinfo(n)->nr_frags = i;
1420 if (skb_has_frag_list(skb)) {
1421 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1422 skb_clone_fraglist(n);
1425 skb_copy_header(n, skb);
1429 EXPORT_SYMBOL(__pskb_copy_fclone);
1432 * pskb_expand_head - reallocate header of &sk_buff
1433 * @skb: buffer to reallocate
1434 * @nhead: room to add at head
1435 * @ntail: room to add at tail
1436 * @gfp_mask: allocation priority
1438 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1439 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1440 * reference count of 1. Returns zero in the case of success or error,
1441 * if expansion failed. In the last case, &sk_buff is not changed.
1443 * All the pointers pointing into skb header may change and must be
1444 * reloaded after call to this function.
1447 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1450 int i, osize = skb_end_offset(skb);
1451 int size = osize + nhead + ntail;
1457 BUG_ON(skb_shared(skb));
1459 size = SKB_DATA_ALIGN(size);
1461 if (skb_pfmemalloc(skb))
1462 gfp_mask |= __GFP_MEMALLOC;
1463 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1464 gfp_mask, NUMA_NO_NODE, NULL);
1467 size = SKB_WITH_OVERHEAD(ksize(data));
1469 /* Copy only real data... and, alas, header. This should be
1470 * optimized for the cases when header is void.
1472 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1474 memcpy((struct skb_shared_info *)(data + size),
1476 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1479 * if shinfo is shared we must drop the old head gracefully, but if it
1480 * is not we can just drop the old head and let the existing refcount
1481 * be since all we did is relocate the values
1483 if (skb_cloned(skb)) {
1484 if (skb_orphan_frags(skb, gfp_mask))
1487 refcount_inc(&skb_uarg(skb)->refcnt);
1488 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1489 skb_frag_ref(skb, i);
1491 if (skb_has_frag_list(skb))
1492 skb_clone_fraglist(skb);
1494 skb_release_data(skb);
1498 off = (data + nhead) - skb->head;
1503 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1507 skb->end = skb->head + size;
1510 skb_headers_offset_update(skb, nhead);
1514 atomic_set(&skb_shinfo(skb)->dataref, 1);
1516 skb_metadata_clear(skb);
1518 /* It is not generally safe to change skb->truesize.
1519 * For the moment, we really care of rx path, or
1520 * when skb is orphaned (not attached to a socket).
1522 if (!skb->sk || skb->destructor == sock_edemux)
1523 skb->truesize += size - osize;
1532 EXPORT_SYMBOL(pskb_expand_head);
1534 /* Make private copy of skb with writable head and some headroom */
1536 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1538 struct sk_buff *skb2;
1539 int delta = headroom - skb_headroom(skb);
1542 skb2 = pskb_copy(skb, GFP_ATOMIC);
1544 skb2 = skb_clone(skb, GFP_ATOMIC);
1545 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1553 EXPORT_SYMBOL(skb_realloc_headroom);
1556 * skb_copy_expand - copy and expand sk_buff
1557 * @skb: buffer to copy
1558 * @newheadroom: new free bytes at head
1559 * @newtailroom: new free bytes at tail
1560 * @gfp_mask: allocation priority
1562 * Make a copy of both an &sk_buff and its data and while doing so
1563 * allocate additional space.
1565 * This is used when the caller wishes to modify the data and needs a
1566 * private copy of the data to alter as well as more space for new fields.
1567 * Returns %NULL on failure or the pointer to the buffer
1568 * on success. The returned buffer has a reference count of 1.
1570 * You must pass %GFP_ATOMIC as the allocation priority if this function
1571 * is called from an interrupt.
1573 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1574 int newheadroom, int newtailroom,
1578 * Allocate the copy buffer
1580 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1581 gfp_mask, skb_alloc_rx_flag(skb),
1583 int oldheadroom = skb_headroom(skb);
1584 int head_copy_len, head_copy_off;
1589 skb_reserve(n, newheadroom);
1591 /* Set the tail pointer and length */
1592 skb_put(n, skb->len);
1594 head_copy_len = oldheadroom;
1596 if (newheadroom <= head_copy_len)
1597 head_copy_len = newheadroom;
1599 head_copy_off = newheadroom - head_copy_len;
1601 /* Copy the linear header and data. */
1602 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1603 skb->len + head_copy_len));
1605 skb_copy_header(n, skb);
1607 skb_headers_offset_update(n, newheadroom - oldheadroom);
1611 EXPORT_SYMBOL(skb_copy_expand);
1614 * __skb_pad - zero pad the tail of an skb
1615 * @skb: buffer to pad
1616 * @pad: space to pad
1617 * @free_on_error: free buffer on error
1619 * Ensure that a buffer is followed by a padding area that is zero
1620 * filled. Used by network drivers which may DMA or transfer data
1621 * beyond the buffer end onto the wire.
1623 * May return error in out of memory cases. The skb is freed on error
1624 * if @free_on_error is true.
1627 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1632 /* If the skbuff is non linear tailroom is always zero.. */
1633 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1634 memset(skb->data+skb->len, 0, pad);
1638 ntail = skb->data_len + pad - (skb->end - skb->tail);
1639 if (likely(skb_cloned(skb) || ntail > 0)) {
1640 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1645 /* FIXME: The use of this function with non-linear skb's really needs
1648 err = skb_linearize(skb);
1652 memset(skb->data + skb->len, 0, pad);
1660 EXPORT_SYMBOL(__skb_pad);
1663 * pskb_put - add data to the tail of a potentially fragmented buffer
1664 * @skb: start of the buffer to use
1665 * @tail: tail fragment of the buffer to use
1666 * @len: amount of data to add
1668 * This function extends the used data area of the potentially
1669 * fragmented buffer. @tail must be the last fragment of @skb -- or
1670 * @skb itself. If this would exceed the total buffer size the kernel
1671 * will panic. A pointer to the first byte of the extra data is
1675 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1678 skb->data_len += len;
1681 return skb_put(tail, len);
1683 EXPORT_SYMBOL_GPL(pskb_put);
1686 * skb_put - add data to a buffer
1687 * @skb: buffer to use
1688 * @len: amount of data to add
1690 * This function extends the used data area of the buffer. If this would
1691 * exceed the total buffer size the kernel will panic. A pointer to the
1692 * first byte of the extra data is returned.
1694 void *skb_put(struct sk_buff *skb, unsigned int len)
1696 void *tmp = skb_tail_pointer(skb);
1697 SKB_LINEAR_ASSERT(skb);
1700 if (unlikely(skb->tail > skb->end))
1701 skb_over_panic(skb, len, __builtin_return_address(0));
1704 EXPORT_SYMBOL(skb_put);
1707 * skb_push - add data to the start of a buffer
1708 * @skb: buffer to use
1709 * @len: amount of data to add
1711 * This function extends the used data area of the buffer at the buffer
1712 * start. If this would exceed the total buffer headroom the kernel will
1713 * panic. A pointer to the first byte of the extra data is returned.
1715 void *skb_push(struct sk_buff *skb, unsigned int len)
1719 if (unlikely(skb->data < skb->head))
1720 skb_under_panic(skb, len, __builtin_return_address(0));
1723 EXPORT_SYMBOL(skb_push);
1726 * skb_pull - remove data from the start of a buffer
1727 * @skb: buffer to use
1728 * @len: amount of data to remove
1730 * This function removes data from the start of a buffer, returning
1731 * the memory to the headroom. A pointer to the next data in the buffer
1732 * is returned. Once the data has been pulled future pushes will overwrite
1735 void *skb_pull(struct sk_buff *skb, unsigned int len)
1737 return skb_pull_inline(skb, len);
1739 EXPORT_SYMBOL(skb_pull);
1742 * skb_trim - remove end from a buffer
1743 * @skb: buffer to alter
1746 * Cut the length of a buffer down by removing data from the tail. If
1747 * the buffer is already under the length specified it is not modified.
1748 * The skb must be linear.
1750 void skb_trim(struct sk_buff *skb, unsigned int len)
1753 __skb_trim(skb, len);
1755 EXPORT_SYMBOL(skb_trim);
1757 /* Trims skb to length len. It can change skb pointers.
1760 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1762 struct sk_buff **fragp;
1763 struct sk_buff *frag;
1764 int offset = skb_headlen(skb);
1765 int nfrags = skb_shinfo(skb)->nr_frags;
1769 if (skb_cloned(skb) &&
1770 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1777 for (; i < nfrags; i++) {
1778 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1785 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1788 skb_shinfo(skb)->nr_frags = i;
1790 for (; i < nfrags; i++)
1791 skb_frag_unref(skb, i);
1793 if (skb_has_frag_list(skb))
1794 skb_drop_fraglist(skb);
1798 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1799 fragp = &frag->next) {
1800 int end = offset + frag->len;
1802 if (skb_shared(frag)) {
1803 struct sk_buff *nfrag;
1805 nfrag = skb_clone(frag, GFP_ATOMIC);
1806 if (unlikely(!nfrag))
1809 nfrag->next = frag->next;
1821 unlikely((err = pskb_trim(frag, len - offset))))
1825 skb_drop_list(&frag->next);
1830 if (len > skb_headlen(skb)) {
1831 skb->data_len -= skb->len - len;
1836 skb_set_tail_pointer(skb, len);
1839 if (!skb->sk || skb->destructor == sock_edemux)
1843 EXPORT_SYMBOL(___pskb_trim);
1845 /* Note : use pskb_trim_rcsum() instead of calling this directly
1847 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
1849 if (skb->ip_summed == CHECKSUM_COMPLETE) {
1850 int delta = skb->len - len;
1852 skb->csum = csum_sub(skb->csum,
1853 skb_checksum(skb, len, delta, 0));
1855 return __pskb_trim(skb, len);
1857 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
1860 * __pskb_pull_tail - advance tail of skb header
1861 * @skb: buffer to reallocate
1862 * @delta: number of bytes to advance tail
1864 * The function makes a sense only on a fragmented &sk_buff,
1865 * it expands header moving its tail forward and copying necessary
1866 * data from fragmented part.
1868 * &sk_buff MUST have reference count of 1.
1870 * Returns %NULL (and &sk_buff does not change) if pull failed
1871 * or value of new tail of skb in the case of success.
1873 * All the pointers pointing into skb header may change and must be
1874 * reloaded after call to this function.
1877 /* Moves tail of skb head forward, copying data from fragmented part,
1878 * when it is necessary.
1879 * 1. It may fail due to malloc failure.
1880 * 2. It may change skb pointers.
1882 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1884 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
1886 /* If skb has not enough free space at tail, get new one
1887 * plus 128 bytes for future expansions. If we have enough
1888 * room at tail, reallocate without expansion only if skb is cloned.
1890 int i, k, eat = (skb->tail + delta) - skb->end;
1892 if (eat > 0 || skb_cloned(skb)) {
1893 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1898 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
1899 skb_tail_pointer(skb), delta));
1901 /* Optimization: no fragments, no reasons to preestimate
1902 * size of pulled pages. Superb.
1904 if (!skb_has_frag_list(skb))
1907 /* Estimate size of pulled pages. */
1909 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1910 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1917 /* If we need update frag list, we are in troubles.
1918 * Certainly, it is possible to add an offset to skb data,
1919 * but taking into account that pulling is expected to
1920 * be very rare operation, it is worth to fight against
1921 * further bloating skb head and crucify ourselves here instead.
1922 * Pure masohism, indeed. 8)8)
1925 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1926 struct sk_buff *clone = NULL;
1927 struct sk_buff *insp = NULL;
1932 if (list->len <= eat) {
1933 /* Eaten as whole. */
1938 /* Eaten partially. */
1940 if (skb_shared(list)) {
1941 /* Sucks! We need to fork list. :-( */
1942 clone = skb_clone(list, GFP_ATOMIC);
1948 /* This may be pulled without
1952 if (!pskb_pull(list, eat)) {
1960 /* Free pulled out fragments. */
1961 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1962 skb_shinfo(skb)->frag_list = list->next;
1965 /* And insert new clone at head. */
1968 skb_shinfo(skb)->frag_list = clone;
1971 /* Success! Now we may commit changes to skb data. */
1976 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1977 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1980 skb_frag_unref(skb, i);
1983 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1985 skb_shinfo(skb)->frags[k].page_offset += eat;
1986 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1994 skb_shinfo(skb)->nr_frags = k;
1998 skb->data_len -= delta;
2001 skb_zcopy_clear(skb, false);
2003 return skb_tail_pointer(skb);
2005 EXPORT_SYMBOL(__pskb_pull_tail);
2008 * skb_copy_bits - copy bits from skb to kernel buffer
2010 * @offset: offset in source
2011 * @to: destination buffer
2012 * @len: number of bytes to copy
2014 * Copy the specified number of bytes from the source skb to the
2015 * destination buffer.
2018 * If its prototype is ever changed,
2019 * check arch/{*}/net/{*}.S files,
2020 * since it is called from BPF assembly code.
2022 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2024 int start = skb_headlen(skb);
2025 struct sk_buff *frag_iter;
2028 if (offset > (int)skb->len - len)
2032 if ((copy = start - offset) > 0) {
2035 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2036 if ((len -= copy) == 0)
2042 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2044 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2046 WARN_ON(start > offset + len);
2048 end = start + skb_frag_size(f);
2049 if ((copy = end - offset) > 0) {
2050 u32 p_off, p_len, copied;
2057 skb_frag_foreach_page(f,
2058 f->page_offset + offset - start,
2059 copy, p, p_off, p_len, copied) {
2060 vaddr = kmap_atomic(p);
2061 memcpy(to + copied, vaddr + p_off, p_len);
2062 kunmap_atomic(vaddr);
2065 if ((len -= copy) == 0)
2073 skb_walk_frags(skb, frag_iter) {
2076 WARN_ON(start > offset + len);
2078 end = start + frag_iter->len;
2079 if ((copy = end - offset) > 0) {
2082 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2084 if ((len -= copy) == 0)
2098 EXPORT_SYMBOL(skb_copy_bits);
2101 * Callback from splice_to_pipe(), if we need to release some pages
2102 * at the end of the spd in case we error'ed out in filling the pipe.
2104 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2106 put_page(spd->pages[i]);
2109 static struct page *linear_to_page(struct page *page, unsigned int *len,
2110 unsigned int *offset,
2113 struct page_frag *pfrag = sk_page_frag(sk);
2115 if (!sk_page_frag_refill(sk, pfrag))
2118 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2120 memcpy(page_address(pfrag->page) + pfrag->offset,
2121 page_address(page) + *offset, *len);
2122 *offset = pfrag->offset;
2123 pfrag->offset += *len;
2128 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2130 unsigned int offset)
2132 return spd->nr_pages &&
2133 spd->pages[spd->nr_pages - 1] == page &&
2134 (spd->partial[spd->nr_pages - 1].offset +
2135 spd->partial[spd->nr_pages - 1].len == offset);
2139 * Fill page/offset/length into spd, if it can hold more pages.
2141 static bool spd_fill_page(struct splice_pipe_desc *spd,
2142 struct pipe_inode_info *pipe, struct page *page,
2143 unsigned int *len, unsigned int offset,
2147 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2151 page = linear_to_page(page, len, &offset, sk);
2155 if (spd_can_coalesce(spd, page, offset)) {
2156 spd->partial[spd->nr_pages - 1].len += *len;
2160 spd->pages[spd->nr_pages] = page;
2161 spd->partial[spd->nr_pages].len = *len;
2162 spd->partial[spd->nr_pages].offset = offset;
2168 static bool __splice_segment(struct page *page, unsigned int poff,
2169 unsigned int plen, unsigned int *off,
2171 struct splice_pipe_desc *spd, bool linear,
2173 struct pipe_inode_info *pipe)
2178 /* skip this segment if already processed */
2184 /* ignore any bits we already processed */
2190 unsigned int flen = min(*len, plen);
2192 if (spd_fill_page(spd, pipe, page, &flen, poff,
2198 } while (*len && plen);
2204 * Map linear and fragment data from the skb to spd. It reports true if the
2205 * pipe is full or if we already spliced the requested length.
2207 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2208 unsigned int *offset, unsigned int *len,
2209 struct splice_pipe_desc *spd, struct sock *sk)
2212 struct sk_buff *iter;
2214 /* map the linear part :
2215 * If skb->head_frag is set, this 'linear' part is backed by a
2216 * fragment, and if the head is not shared with any clones then
2217 * we can avoid a copy since we own the head portion of this page.
2219 if (__splice_segment(virt_to_page(skb->data),
2220 (unsigned long) skb->data & (PAGE_SIZE - 1),
2223 skb_head_is_locked(skb),
2228 * then map the fragments
2230 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2231 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2233 if (__splice_segment(skb_frag_page(f),
2234 f->page_offset, skb_frag_size(f),
2235 offset, len, spd, false, sk, pipe))
2239 skb_walk_frags(skb, iter) {
2240 if (*offset >= iter->len) {
2241 *offset -= iter->len;
2244 /* __skb_splice_bits() only fails if the output has no room
2245 * left, so no point in going over the frag_list for the error
2248 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2256 * Map data from the skb to a pipe. Should handle both the linear part,
2257 * the fragments, and the frag list.
2259 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2260 struct pipe_inode_info *pipe, unsigned int tlen,
2263 struct partial_page partial[MAX_SKB_FRAGS];
2264 struct page *pages[MAX_SKB_FRAGS];
2265 struct splice_pipe_desc spd = {
2268 .nr_pages_max = MAX_SKB_FRAGS,
2269 .ops = &nosteal_pipe_buf_ops,
2270 .spd_release = sock_spd_release,
2274 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2277 ret = splice_to_pipe(pipe, &spd);
2281 EXPORT_SYMBOL_GPL(skb_splice_bits);
2283 /* Send skb data on a socket. Socket must be locked. */
2284 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2287 unsigned int orig_len = len;
2288 struct sk_buff *head = skb;
2289 unsigned short fragidx;
2294 /* Deal with head data */
2295 while (offset < skb_headlen(skb) && len) {
2299 slen = min_t(int, len, skb_headlen(skb) - offset);
2300 kv.iov_base = skb->data + offset;
2302 memset(&msg, 0, sizeof(msg));
2304 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2312 /* All the data was skb head? */
2316 /* Make offset relative to start of frags */
2317 offset -= skb_headlen(skb);
2319 /* Find where we are in frag list */
2320 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2321 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2323 if (offset < frag->size)
2326 offset -= frag->size;
2329 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2330 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2332 slen = min_t(size_t, len, frag->size - offset);
2335 ret = kernel_sendpage_locked(sk, frag->page.p,
2336 frag->page_offset + offset,
2337 slen, MSG_DONTWAIT);
2350 /* Process any frag lists */
2353 if (skb_has_frag_list(skb)) {
2354 skb = skb_shinfo(skb)->frag_list;
2357 } else if (skb->next) {
2364 return orig_len - len;
2367 return orig_len == len ? ret : orig_len - len;
2369 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2371 /* Send skb data on a socket. */
2372 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2377 ret = skb_send_sock_locked(sk, skb, offset, len);
2382 EXPORT_SYMBOL_GPL(skb_send_sock);
2385 * skb_store_bits - store bits from kernel buffer to skb
2386 * @skb: destination buffer
2387 * @offset: offset in destination
2388 * @from: source buffer
2389 * @len: number of bytes to copy
2391 * Copy the specified number of bytes from the source buffer to the
2392 * destination skb. This function handles all the messy bits of
2393 * traversing fragment lists and such.
2396 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2398 int start = skb_headlen(skb);
2399 struct sk_buff *frag_iter;
2402 if (offset > (int)skb->len - len)
2405 if ((copy = start - offset) > 0) {
2408 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2409 if ((len -= copy) == 0)
2415 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2416 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2419 WARN_ON(start > offset + len);
2421 end = start + skb_frag_size(frag);
2422 if ((copy = end - offset) > 0) {
2423 u32 p_off, p_len, copied;
2430 skb_frag_foreach_page(frag,
2431 frag->page_offset + offset - start,
2432 copy, p, p_off, p_len, copied) {
2433 vaddr = kmap_atomic(p);
2434 memcpy(vaddr + p_off, from + copied, p_len);
2435 kunmap_atomic(vaddr);
2438 if ((len -= copy) == 0)
2446 skb_walk_frags(skb, frag_iter) {
2449 WARN_ON(start > offset + len);
2451 end = start + frag_iter->len;
2452 if ((copy = end - offset) > 0) {
2455 if (skb_store_bits(frag_iter, offset - start,
2458 if ((len -= copy) == 0)
2471 EXPORT_SYMBOL(skb_store_bits);
2473 /* Checksum skb data. */
2474 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2475 __wsum csum, const struct skb_checksum_ops *ops)
2477 int start = skb_headlen(skb);
2478 int i, copy = start - offset;
2479 struct sk_buff *frag_iter;
2482 /* Checksum header. */
2486 csum = ops->update(skb->data + offset, copy, csum);
2487 if ((len -= copy) == 0)
2493 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2495 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2497 WARN_ON(start > offset + len);
2499 end = start + skb_frag_size(frag);
2500 if ((copy = end - offset) > 0) {
2501 u32 p_off, p_len, copied;
2509 skb_frag_foreach_page(frag,
2510 frag->page_offset + offset - start,
2511 copy, p, p_off, p_len, copied) {
2512 vaddr = kmap_atomic(p);
2513 csum2 = ops->update(vaddr + p_off, p_len, 0);
2514 kunmap_atomic(vaddr);
2515 csum = ops->combine(csum, csum2, pos, p_len);
2526 skb_walk_frags(skb, frag_iter) {
2529 WARN_ON(start > offset + len);
2531 end = start + frag_iter->len;
2532 if ((copy = end - offset) > 0) {
2536 csum2 = __skb_checksum(frag_iter, offset - start,
2538 csum = ops->combine(csum, csum2, pos, copy);
2539 if ((len -= copy) == 0)
2550 EXPORT_SYMBOL(__skb_checksum);
2552 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2553 int len, __wsum csum)
2555 const struct skb_checksum_ops ops = {
2556 .update = csum_partial_ext,
2557 .combine = csum_block_add_ext,
2560 return __skb_checksum(skb, offset, len, csum, &ops);
2562 EXPORT_SYMBOL(skb_checksum);
2564 /* Both of above in one bottle. */
2566 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2567 u8 *to, int len, __wsum csum)
2569 int start = skb_headlen(skb);
2570 int i, copy = start - offset;
2571 struct sk_buff *frag_iter;
2578 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2580 if ((len -= copy) == 0)
2587 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2590 WARN_ON(start > offset + len);
2592 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2593 if ((copy = end - offset) > 0) {
2594 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2595 u32 p_off, p_len, copied;
2603 skb_frag_foreach_page(frag,
2604 frag->page_offset + offset - start,
2605 copy, p, p_off, p_len, copied) {
2606 vaddr = kmap_atomic(p);
2607 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2610 kunmap_atomic(vaddr);
2611 csum = csum_block_add(csum, csum2, pos);
2623 skb_walk_frags(skb, frag_iter) {
2627 WARN_ON(start > offset + len);
2629 end = start + frag_iter->len;
2630 if ((copy = end - offset) > 0) {
2633 csum2 = skb_copy_and_csum_bits(frag_iter,
2636 csum = csum_block_add(csum, csum2, pos);
2637 if ((len -= copy) == 0)
2648 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2650 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2652 net_warn_ratelimited(
2653 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2658 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2659 int offset, int len)
2661 net_warn_ratelimited(
2662 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2667 static const struct skb_checksum_ops default_crc32c_ops = {
2668 .update = warn_crc32c_csum_update,
2669 .combine = warn_crc32c_csum_combine,
2672 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2673 &default_crc32c_ops;
2674 EXPORT_SYMBOL(crc32c_csum_stub);
2677 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2678 * @from: source buffer
2680 * Calculates the amount of linear headroom needed in the 'to' skb passed
2681 * into skb_zerocopy().
2684 skb_zerocopy_headlen(const struct sk_buff *from)
2686 unsigned int hlen = 0;
2688 if (!from->head_frag ||
2689 skb_headlen(from) < L1_CACHE_BYTES ||
2690 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2691 hlen = skb_headlen(from);
2693 if (skb_has_frag_list(from))
2698 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2701 * skb_zerocopy - Zero copy skb to skb
2702 * @to: destination buffer
2703 * @from: source buffer
2704 * @len: number of bytes to copy from source buffer
2705 * @hlen: size of linear headroom in destination buffer
2707 * Copies up to `len` bytes from `from` to `to` by creating references
2708 * to the frags in the source buffer.
2710 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2711 * headroom in the `to` buffer.
2714 * 0: everything is OK
2715 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2716 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2719 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2722 int plen = 0; /* length of skb->head fragment */
2725 unsigned int offset;
2727 BUG_ON(!from->head_frag && !hlen);
2729 /* dont bother with small payloads */
2730 if (len <= skb_tailroom(to))
2731 return skb_copy_bits(from, 0, skb_put(to, len), len);
2734 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2739 plen = min_t(int, skb_headlen(from), len);
2741 page = virt_to_head_page(from->head);
2742 offset = from->data - (unsigned char *)page_address(page);
2743 __skb_fill_page_desc(to, 0, page, offset, plen);
2750 to->truesize += len + plen;
2751 to->len += len + plen;
2752 to->data_len += len + plen;
2754 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2758 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2760 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2763 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2764 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2765 len -= skb_shinfo(to)->frags[j].size;
2766 skb_frag_ref(to, j);
2769 skb_shinfo(to)->nr_frags = j;
2773 EXPORT_SYMBOL_GPL(skb_zerocopy);
2775 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2780 if (skb->ip_summed == CHECKSUM_PARTIAL)
2781 csstart = skb_checksum_start_offset(skb);
2783 csstart = skb_headlen(skb);
2785 BUG_ON(csstart > skb_headlen(skb));
2787 skb_copy_from_linear_data(skb, to, csstart);
2790 if (csstart != skb->len)
2791 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2792 skb->len - csstart, 0);
2794 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2795 long csstuff = csstart + skb->csum_offset;
2797 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2800 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2803 * skb_dequeue - remove from the head of the queue
2804 * @list: list to dequeue from
2806 * Remove the head of the list. The list lock is taken so the function
2807 * may be used safely with other locking list functions. The head item is
2808 * returned or %NULL if the list is empty.
2811 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2813 unsigned long flags;
2814 struct sk_buff *result;
2816 spin_lock_irqsave(&list->lock, flags);
2817 result = __skb_dequeue(list);
2818 spin_unlock_irqrestore(&list->lock, flags);
2821 EXPORT_SYMBOL(skb_dequeue);
2824 * skb_dequeue_tail - remove from the tail of the queue
2825 * @list: list to dequeue from
2827 * Remove the tail of the list. The list lock is taken so the function
2828 * may be used safely with other locking list functions. The tail item is
2829 * returned or %NULL if the list is empty.
2831 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2833 unsigned long flags;
2834 struct sk_buff *result;
2836 spin_lock_irqsave(&list->lock, flags);
2837 result = __skb_dequeue_tail(list);
2838 spin_unlock_irqrestore(&list->lock, flags);
2841 EXPORT_SYMBOL(skb_dequeue_tail);
2844 * skb_queue_purge - empty a list
2845 * @list: list to empty
2847 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2848 * the list and one reference dropped. This function takes the list
2849 * lock and is atomic with respect to other list locking functions.
2851 void skb_queue_purge(struct sk_buff_head *list)
2853 struct sk_buff *skb;
2854 while ((skb = skb_dequeue(list)) != NULL)
2857 EXPORT_SYMBOL(skb_queue_purge);
2860 * skb_rbtree_purge - empty a skb rbtree
2861 * @root: root of the rbtree to empty
2862 * Return value: the sum of truesizes of all purged skbs.
2864 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2865 * the list and one reference dropped. This function does not take
2866 * any lock. Synchronization should be handled by the caller (e.g., TCP
2867 * out-of-order queue is protected by the socket lock).
2869 unsigned int skb_rbtree_purge(struct rb_root *root)
2871 struct rb_node *p = rb_first(root);
2872 unsigned int sum = 0;
2875 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
2878 rb_erase(&skb->rbnode, root);
2879 sum += skb->truesize;
2886 * skb_queue_head - queue a buffer at the list head
2887 * @list: list to use
2888 * @newsk: buffer to queue
2890 * Queue a buffer at the start of the list. This function takes the
2891 * list lock and can be used safely with other locking &sk_buff functions
2894 * A buffer cannot be placed on two lists at the same time.
2896 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2898 unsigned long flags;
2900 spin_lock_irqsave(&list->lock, flags);
2901 __skb_queue_head(list, newsk);
2902 spin_unlock_irqrestore(&list->lock, flags);
2904 EXPORT_SYMBOL(skb_queue_head);
2907 * skb_queue_tail - queue a buffer at the list tail
2908 * @list: list to use
2909 * @newsk: buffer to queue
2911 * Queue a buffer at the tail of the list. This function takes the
2912 * list lock and can be used safely with other locking &sk_buff functions
2915 * A buffer cannot be placed on two lists at the same time.
2917 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2919 unsigned long flags;
2921 spin_lock_irqsave(&list->lock, flags);
2922 __skb_queue_tail(list, newsk);
2923 spin_unlock_irqrestore(&list->lock, flags);
2925 EXPORT_SYMBOL(skb_queue_tail);
2928 * skb_unlink - remove a buffer from a list
2929 * @skb: buffer to remove
2930 * @list: list to use
2932 * Remove a packet from a list. The list locks are taken and this
2933 * function is atomic with respect to other list locked calls
2935 * You must know what list the SKB is on.
2937 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2939 unsigned long flags;
2941 spin_lock_irqsave(&list->lock, flags);
2942 __skb_unlink(skb, list);
2943 spin_unlock_irqrestore(&list->lock, flags);
2945 EXPORT_SYMBOL(skb_unlink);
2948 * skb_append - append a buffer
2949 * @old: buffer to insert after
2950 * @newsk: buffer to insert
2951 * @list: list to use
2953 * Place a packet after a given packet in a list. The list locks are taken
2954 * and this function is atomic with respect to other list locked calls.
2955 * A buffer cannot be placed on two lists at the same time.
2957 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2959 unsigned long flags;
2961 spin_lock_irqsave(&list->lock, flags);
2962 __skb_queue_after(list, old, newsk);
2963 spin_unlock_irqrestore(&list->lock, flags);
2965 EXPORT_SYMBOL(skb_append);
2968 * skb_insert - insert a buffer
2969 * @old: buffer to insert before
2970 * @newsk: buffer to insert
2971 * @list: list to use
2973 * Place a packet before a given packet in a list. The list locks are
2974 * taken and this function is atomic with respect to other list locked
2977 * A buffer cannot be placed on two lists at the same time.
2979 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2981 unsigned long flags;
2983 spin_lock_irqsave(&list->lock, flags);
2984 __skb_insert(newsk, old->prev, old, list);
2985 spin_unlock_irqrestore(&list->lock, flags);
2987 EXPORT_SYMBOL(skb_insert);
2989 static inline void skb_split_inside_header(struct sk_buff *skb,
2990 struct sk_buff* skb1,
2991 const u32 len, const int pos)
2995 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2997 /* And move data appendix as is. */
2998 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2999 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3001 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3002 skb_shinfo(skb)->nr_frags = 0;
3003 skb1->data_len = skb->data_len;
3004 skb1->len += skb1->data_len;
3007 skb_set_tail_pointer(skb, len);
3010 static inline void skb_split_no_header(struct sk_buff *skb,
3011 struct sk_buff* skb1,
3012 const u32 len, int pos)
3015 const int nfrags = skb_shinfo(skb)->nr_frags;
3017 skb_shinfo(skb)->nr_frags = 0;
3018 skb1->len = skb1->data_len = skb->len - len;
3020 skb->data_len = len - pos;
3022 for (i = 0; i < nfrags; i++) {
3023 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3025 if (pos + size > len) {
3026 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3030 * We have two variants in this case:
3031 * 1. Move all the frag to the second
3032 * part, if it is possible. F.e.
3033 * this approach is mandatory for TUX,
3034 * where splitting is expensive.
3035 * 2. Split is accurately. We make this.
3037 skb_frag_ref(skb, i);
3038 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
3039 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3040 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3041 skb_shinfo(skb)->nr_frags++;
3045 skb_shinfo(skb)->nr_frags++;
3048 skb_shinfo(skb1)->nr_frags = k;
3052 * skb_split - Split fragmented skb to two parts at length len.
3053 * @skb: the buffer to split
3054 * @skb1: the buffer to receive the second part
3055 * @len: new length for skb
3057 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3059 int pos = skb_headlen(skb);
3061 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3063 skb_zerocopy_clone(skb1, skb, 0);
3064 if (len < pos) /* Split line is inside header. */
3065 skb_split_inside_header(skb, skb1, len, pos);
3066 else /* Second chunk has no header, nothing to copy. */
3067 skb_split_no_header(skb, skb1, len, pos);
3069 EXPORT_SYMBOL(skb_split);
3071 /* Shifting from/to a cloned skb is a no-go.
3073 * Caller cannot keep skb_shinfo related pointers past calling here!
3075 static int skb_prepare_for_shift(struct sk_buff *skb)
3077 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3081 * skb_shift - Shifts paged data partially from skb to another
3082 * @tgt: buffer into which tail data gets added
3083 * @skb: buffer from which the paged data comes from
3084 * @shiftlen: shift up to this many bytes
3086 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3087 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3088 * It's up to caller to free skb if everything was shifted.
3090 * If @tgt runs out of frags, the whole operation is aborted.
3092 * Skb cannot include anything else but paged data while tgt is allowed
3093 * to have non-paged data as well.
3095 * TODO: full sized shift could be optimized but that would need
3096 * specialized skb free'er to handle frags without up-to-date nr_frags.
3098 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3100 int from, to, merge, todo;
3101 struct skb_frag_struct *fragfrom, *fragto;
3103 BUG_ON(shiftlen > skb->len);
3105 if (skb_headlen(skb))
3107 if (skb_zcopy(tgt) || skb_zcopy(skb))
3112 to = skb_shinfo(tgt)->nr_frags;
3113 fragfrom = &skb_shinfo(skb)->frags[from];
3115 /* Actual merge is delayed until the point when we know we can
3116 * commit all, so that we don't have to undo partial changes
3119 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3120 fragfrom->page_offset)) {
3125 todo -= skb_frag_size(fragfrom);
3127 if (skb_prepare_for_shift(skb) ||
3128 skb_prepare_for_shift(tgt))
3131 /* All previous frag pointers might be stale! */
3132 fragfrom = &skb_shinfo(skb)->frags[from];
3133 fragto = &skb_shinfo(tgt)->frags[merge];
3135 skb_frag_size_add(fragto, shiftlen);
3136 skb_frag_size_sub(fragfrom, shiftlen);
3137 fragfrom->page_offset += shiftlen;
3145 /* Skip full, not-fitting skb to avoid expensive operations */
3146 if ((shiftlen == skb->len) &&
3147 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3150 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3153 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3154 if (to == MAX_SKB_FRAGS)
3157 fragfrom = &skb_shinfo(skb)->frags[from];
3158 fragto = &skb_shinfo(tgt)->frags[to];
3160 if (todo >= skb_frag_size(fragfrom)) {
3161 *fragto = *fragfrom;
3162 todo -= skb_frag_size(fragfrom);
3167 __skb_frag_ref(fragfrom);
3168 fragto->page = fragfrom->page;
3169 fragto->page_offset = fragfrom->page_offset;
3170 skb_frag_size_set(fragto, todo);
3172 fragfrom->page_offset += todo;
3173 skb_frag_size_sub(fragfrom, todo);
3181 /* Ready to "commit" this state change to tgt */
3182 skb_shinfo(tgt)->nr_frags = to;
3185 fragfrom = &skb_shinfo(skb)->frags[0];
3186 fragto = &skb_shinfo(tgt)->frags[merge];
3188 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3189 __skb_frag_unref(fragfrom);
3192 /* Reposition in the original skb */
3194 while (from < skb_shinfo(skb)->nr_frags)
3195 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3196 skb_shinfo(skb)->nr_frags = to;
3198 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3201 /* Most likely the tgt won't ever need its checksum anymore, skb on
3202 * the other hand might need it if it needs to be resent
3204 tgt->ip_summed = CHECKSUM_PARTIAL;
3205 skb->ip_summed = CHECKSUM_PARTIAL;
3207 /* Yak, is it really working this way? Some helper please? */
3208 skb->len -= shiftlen;
3209 skb->data_len -= shiftlen;
3210 skb->truesize -= shiftlen;
3211 tgt->len += shiftlen;
3212 tgt->data_len += shiftlen;
3213 tgt->truesize += shiftlen;
3219 * skb_prepare_seq_read - Prepare a sequential read of skb data
3220 * @skb: the buffer to read
3221 * @from: lower offset of data to be read
3222 * @to: upper offset of data to be read
3223 * @st: state variable
3225 * Initializes the specified state variable. Must be called before
3226 * invoking skb_seq_read() for the first time.
3228 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3229 unsigned int to, struct skb_seq_state *st)
3231 st->lower_offset = from;
3232 st->upper_offset = to;
3233 st->root_skb = st->cur_skb = skb;
3234 st->frag_idx = st->stepped_offset = 0;
3235 st->frag_data = NULL;
3237 EXPORT_SYMBOL(skb_prepare_seq_read);
3240 * skb_seq_read - Sequentially read skb data
3241 * @consumed: number of bytes consumed by the caller so far
3242 * @data: destination pointer for data to be returned
3243 * @st: state variable
3245 * Reads a block of skb data at @consumed relative to the
3246 * lower offset specified to skb_prepare_seq_read(). Assigns
3247 * the head of the data block to @data and returns the length
3248 * of the block or 0 if the end of the skb data or the upper
3249 * offset has been reached.
3251 * The caller is not required to consume all of the data
3252 * returned, i.e. @consumed is typically set to the number
3253 * of bytes already consumed and the next call to
3254 * skb_seq_read() will return the remaining part of the block.
3256 * Note 1: The size of each block of data returned can be arbitrary,
3257 * this limitation is the cost for zerocopy sequential
3258 * reads of potentially non linear data.
3260 * Note 2: Fragment lists within fragments are not implemented
3261 * at the moment, state->root_skb could be replaced with
3262 * a stack for this purpose.
3264 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3265 struct skb_seq_state *st)
3267 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3270 if (unlikely(abs_offset >= st->upper_offset)) {
3271 if (st->frag_data) {
3272 kunmap_atomic(st->frag_data);
3273 st->frag_data = NULL;
3279 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3281 if (abs_offset < block_limit && !st->frag_data) {
3282 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3283 return block_limit - abs_offset;
3286 if (st->frag_idx == 0 && !st->frag_data)
3287 st->stepped_offset += skb_headlen(st->cur_skb);
3289 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3290 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3291 block_limit = skb_frag_size(frag) + st->stepped_offset;
3293 if (abs_offset < block_limit) {
3295 st->frag_data = kmap_atomic(skb_frag_page(frag));
3297 *data = (u8 *) st->frag_data + frag->page_offset +
3298 (abs_offset - st->stepped_offset);
3300 return block_limit - abs_offset;
3303 if (st->frag_data) {
3304 kunmap_atomic(st->frag_data);
3305 st->frag_data = NULL;
3309 st->stepped_offset += skb_frag_size(frag);
3312 if (st->frag_data) {
3313 kunmap_atomic(st->frag_data);
3314 st->frag_data = NULL;
3317 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3318 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3321 } else if (st->cur_skb->next) {
3322 st->cur_skb = st->cur_skb->next;
3329 EXPORT_SYMBOL(skb_seq_read);
3332 * skb_abort_seq_read - Abort a sequential read of skb data
3333 * @st: state variable
3335 * Must be called if skb_seq_read() was not called until it
3338 void skb_abort_seq_read(struct skb_seq_state *st)
3341 kunmap_atomic(st->frag_data);
3343 EXPORT_SYMBOL(skb_abort_seq_read);
3345 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3347 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3348 struct ts_config *conf,
3349 struct ts_state *state)
3351 return skb_seq_read(offset, text, TS_SKB_CB(state));
3354 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3356 skb_abort_seq_read(TS_SKB_CB(state));
3360 * skb_find_text - Find a text pattern in skb data
3361 * @skb: the buffer to look in
3362 * @from: search offset
3364 * @config: textsearch configuration
3366 * Finds a pattern in the skb data according to the specified
3367 * textsearch configuration. Use textsearch_next() to retrieve
3368 * subsequent occurrences of the pattern. Returns the offset
3369 * to the first occurrence or UINT_MAX if no match was found.
3371 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3372 unsigned int to, struct ts_config *config)
3374 struct ts_state state;
3377 config->get_next_block = skb_ts_get_next_block;
3378 config->finish = skb_ts_finish;
3380 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3382 ret = textsearch_find(config, &state);
3383 return (ret <= to - from ? ret : UINT_MAX);
3385 EXPORT_SYMBOL(skb_find_text);
3388 * skb_append_datato_frags - append the user data to a skb
3389 * @sk: sock structure
3390 * @skb: skb structure to be appended with user data.
3391 * @getfrag: call back function to be used for getting the user data
3392 * @from: pointer to user message iov
3393 * @length: length of the iov message
3395 * Description: This procedure append the user data in the fragment part
3396 * of the skb if any page alloc fails user this procedure returns -ENOMEM
3398 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
3399 int (*getfrag)(void *from, char *to, int offset,
3400 int len, int odd, struct sk_buff *skb),
3401 void *from, int length)
3403 int frg_cnt = skb_shinfo(skb)->nr_frags;
3407 struct page_frag *pfrag = ¤t->task_frag;
3410 /* Return error if we don't have space for new frag */
3411 if (frg_cnt >= MAX_SKB_FRAGS)
3414 if (!sk_page_frag_refill(sk, pfrag))
3417 /* copy the user data to page */
3418 copy = min_t(int, length, pfrag->size - pfrag->offset);
3420 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
3421 offset, copy, 0, skb);
3425 /* copy was successful so update the size parameters */
3426 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
3429 pfrag->offset += copy;
3430 get_page(pfrag->page);
3432 skb->truesize += copy;
3433 refcount_add(copy, &sk->sk_wmem_alloc);
3435 skb->data_len += copy;
3439 } while (length > 0);
3443 EXPORT_SYMBOL(skb_append_datato_frags);
3445 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3446 int offset, size_t size)
3448 int i = skb_shinfo(skb)->nr_frags;
3450 if (skb_can_coalesce(skb, i, page, offset)) {
3451 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3452 } else if (i < MAX_SKB_FRAGS) {
3454 skb_fill_page_desc(skb, i, page, offset, size);
3461 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3464 * skb_pull_rcsum - pull skb and update receive checksum
3465 * @skb: buffer to update
3466 * @len: length of data pulled
3468 * This function performs an skb_pull on the packet and updates
3469 * the CHECKSUM_COMPLETE checksum. It should be used on
3470 * receive path processing instead of skb_pull unless you know
3471 * that the checksum difference is zero (e.g., a valid IP header)
3472 * or you are setting ip_summed to CHECKSUM_NONE.
3474 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3476 unsigned char *data = skb->data;
3478 BUG_ON(len > skb->len);
3479 __skb_pull(skb, len);
3480 skb_postpull_rcsum(skb, data, len);
3483 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3485 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3487 skb_frag_t head_frag;
3490 page = virt_to_head_page(frag_skb->head);
3491 head_frag.page.p = page;
3492 head_frag.page_offset = frag_skb->data -
3493 (unsigned char *)page_address(page);
3494 head_frag.size = skb_headlen(frag_skb);
3499 * skb_segment - Perform protocol segmentation on skb.
3500 * @head_skb: buffer to segment
3501 * @features: features for the output path (see dev->features)
3503 * This function performs segmentation on the given skb. It returns
3504 * a pointer to the first in a list of new skbs for the segments.
3505 * In case of error it returns ERR_PTR(err).
3507 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3508 netdev_features_t features)
3510 struct sk_buff *segs = NULL;
3511 struct sk_buff *tail = NULL;
3512 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3513 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3514 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3515 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3516 struct sk_buff *frag_skb = head_skb;
3517 unsigned int offset = doffset;
3518 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3519 unsigned int partial_segs = 0;
3520 unsigned int headroom;
3521 unsigned int len = head_skb->len;
3524 int nfrags = skb_shinfo(head_skb)->nr_frags;
3530 __skb_push(head_skb, doffset);
3531 proto = skb_network_protocol(head_skb, &dummy);
3532 if (unlikely(!proto))
3533 return ERR_PTR(-EINVAL);
3535 sg = !!(features & NETIF_F_SG);
3536 csum = !!can_checksum_protocol(features, proto);
3538 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3539 if (!(features & NETIF_F_GSO_PARTIAL)) {
3540 struct sk_buff *iter;
3541 unsigned int frag_len;
3544 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3547 /* If we get here then all the required
3548 * GSO features except frag_list are supported.
3549 * Try to split the SKB to multiple GSO SKBs
3550 * with no frag_list.
3551 * Currently we can do that only when the buffers don't
3552 * have a linear part and all the buffers except
3553 * the last are of the same length.
3555 frag_len = list_skb->len;
3556 skb_walk_frags(head_skb, iter) {
3557 if (frag_len != iter->len && iter->next)
3559 if (skb_headlen(iter) && !iter->head_frag)
3565 if (len != frag_len)
3569 /* GSO partial only requires that we trim off any excess that
3570 * doesn't fit into an MSS sized block, so take care of that
3573 partial_segs = len / mss;
3574 if (partial_segs > 1)
3575 mss *= partial_segs;
3581 headroom = skb_headroom(head_skb);
3582 pos = skb_headlen(head_skb);
3585 struct sk_buff *nskb;
3586 skb_frag_t *nskb_frag;
3590 if (unlikely(mss == GSO_BY_FRAGS)) {
3591 len = list_skb->len;
3593 len = head_skb->len - offset;
3598 hsize = skb_headlen(head_skb) - offset;
3601 if (hsize > len || !sg)
3604 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3605 (skb_headlen(list_skb) == len || sg)) {
3606 BUG_ON(skb_headlen(list_skb) > len);
3609 nfrags = skb_shinfo(list_skb)->nr_frags;
3610 frag = skb_shinfo(list_skb)->frags;
3611 frag_skb = list_skb;
3612 pos += skb_headlen(list_skb);
3614 while (pos < offset + len) {
3615 BUG_ON(i >= nfrags);
3617 size = skb_frag_size(frag);
3618 if (pos + size > offset + len)
3626 nskb = skb_clone(list_skb, GFP_ATOMIC);
3627 list_skb = list_skb->next;
3629 if (unlikely(!nskb))
3632 if (unlikely(pskb_trim(nskb, len))) {
3637 hsize = skb_end_offset(nskb);
3638 if (skb_cow_head(nskb, doffset + headroom)) {
3643 nskb->truesize += skb_end_offset(nskb) - hsize;
3644 skb_release_head_state(nskb);
3645 __skb_push(nskb, doffset);
3647 nskb = __alloc_skb(hsize + doffset + headroom,
3648 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3651 if (unlikely(!nskb))
3654 skb_reserve(nskb, headroom);
3655 __skb_put(nskb, doffset);
3664 __copy_skb_header(nskb, head_skb);
3666 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3667 skb_reset_mac_len(nskb);
3669 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3670 nskb->data - tnl_hlen,
3671 doffset + tnl_hlen);
3673 if (nskb->len == len + doffset)
3674 goto perform_csum_check;
3677 if (!nskb->remcsum_offload)
3678 nskb->ip_summed = CHECKSUM_NONE;
3679 SKB_GSO_CB(nskb)->csum =
3680 skb_copy_and_csum_bits(head_skb, offset,
3683 SKB_GSO_CB(nskb)->csum_start =
3684 skb_headroom(nskb) + doffset;
3688 nskb_frag = skb_shinfo(nskb)->frags;
3690 skb_copy_from_linear_data_offset(head_skb, offset,
3691 skb_put(nskb, hsize), hsize);
3693 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3696 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3697 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3700 while (pos < offset + len) {
3703 nfrags = skb_shinfo(list_skb)->nr_frags;
3704 frag = skb_shinfo(list_skb)->frags;
3705 frag_skb = list_skb;
3706 if (!skb_headlen(list_skb)) {
3709 BUG_ON(!list_skb->head_frag);
3711 /* to make room for head_frag. */
3715 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3716 skb_zerocopy_clone(nskb, frag_skb,
3720 list_skb = list_skb->next;
3723 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3725 net_warn_ratelimited(
3726 "skb_segment: too many frags: %u %u\n",
3732 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3733 __skb_frag_ref(nskb_frag);
3734 size = skb_frag_size(nskb_frag);
3737 nskb_frag->page_offset += offset - pos;
3738 skb_frag_size_sub(nskb_frag, offset - pos);
3741 skb_shinfo(nskb)->nr_frags++;
3743 if (pos + size <= offset + len) {
3748 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3756 nskb->data_len = len - hsize;
3757 nskb->len += nskb->data_len;
3758 nskb->truesize += nskb->data_len;
3762 if (skb_has_shared_frag(nskb) &&
3763 __skb_linearize(nskb))
3766 if (!nskb->remcsum_offload)
3767 nskb->ip_summed = CHECKSUM_NONE;
3768 SKB_GSO_CB(nskb)->csum =
3769 skb_checksum(nskb, doffset,
3770 nskb->len - doffset, 0);
3771 SKB_GSO_CB(nskb)->csum_start =
3772 skb_headroom(nskb) + doffset;
3774 } while ((offset += len) < head_skb->len);
3776 /* Some callers want to get the end of the list.
3777 * Put it in segs->prev to avoid walking the list.
3778 * (see validate_xmit_skb_list() for example)
3783 struct sk_buff *iter;
3784 int type = skb_shinfo(head_skb)->gso_type;
3785 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3787 /* Update type to add partial and then remove dodgy if set */
3788 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3789 type &= ~SKB_GSO_DODGY;
3791 /* Update GSO info and prepare to start updating headers on
3792 * our way back down the stack of protocols.
3794 for (iter = segs; iter; iter = iter->next) {
3795 skb_shinfo(iter)->gso_size = gso_size;
3796 skb_shinfo(iter)->gso_segs = partial_segs;
3797 skb_shinfo(iter)->gso_type = type;
3798 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3801 if (tail->len - doffset <= gso_size)
3802 skb_shinfo(tail)->gso_size = 0;
3803 else if (tail != segs)
3804 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3807 /* Following permits correct backpressure, for protocols
3808 * using skb_set_owner_w().
3809 * Idea is to tranfert ownership from head_skb to last segment.
3811 if (head_skb->destructor == sock_wfree) {
3812 swap(tail->truesize, head_skb->truesize);
3813 swap(tail->destructor, head_skb->destructor);
3814 swap(tail->sk, head_skb->sk);
3819 kfree_skb_list(segs);
3820 return ERR_PTR(err);
3822 EXPORT_SYMBOL_GPL(skb_segment);
3824 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
3826 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3827 unsigned int offset = skb_gro_offset(skb);
3828 unsigned int headlen = skb_headlen(skb);
3829 unsigned int len = skb_gro_len(skb);
3830 unsigned int delta_truesize;
3833 if (unlikely(p->len + len >= 65536))
3836 lp = NAPI_GRO_CB(p)->last;
3837 pinfo = skb_shinfo(lp);
3839 if (headlen <= offset) {
3842 int i = skbinfo->nr_frags;
3843 int nr_frags = pinfo->nr_frags + i;
3845 if (nr_frags > MAX_SKB_FRAGS)
3849 pinfo->nr_frags = nr_frags;
3850 skbinfo->nr_frags = 0;
3852 frag = pinfo->frags + nr_frags;
3853 frag2 = skbinfo->frags + i;
3858 frag->page_offset += offset;
3859 skb_frag_size_sub(frag, offset);
3861 /* all fragments truesize : remove (head size + sk_buff) */
3862 delta_truesize = skb->truesize -
3863 SKB_TRUESIZE(skb_end_offset(skb));
3865 skb->truesize -= skb->data_len;
3866 skb->len -= skb->data_len;
3869 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3871 } else if (skb->head_frag) {
3872 int nr_frags = pinfo->nr_frags;
3873 skb_frag_t *frag = pinfo->frags + nr_frags;
3874 struct page *page = virt_to_head_page(skb->head);
3875 unsigned int first_size = headlen - offset;
3876 unsigned int first_offset;
3878 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3881 first_offset = skb->data -
3882 (unsigned char *)page_address(page) +
3885 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3887 frag->page.p = page;
3888 frag->page_offset = first_offset;
3889 skb_frag_size_set(frag, first_size);
3891 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3892 /* We dont need to clear skbinfo->nr_frags here */
3894 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3895 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3900 delta_truesize = skb->truesize;
3901 if (offset > headlen) {
3902 unsigned int eat = offset - headlen;
3904 skbinfo->frags[0].page_offset += eat;
3905 skb_frag_size_sub(&skbinfo->frags[0], eat);
3906 skb->data_len -= eat;
3911 __skb_pull(skb, offset);
3913 if (NAPI_GRO_CB(p)->last == p)
3914 skb_shinfo(p)->frag_list = skb;
3916 NAPI_GRO_CB(p)->last->next = skb;
3917 NAPI_GRO_CB(p)->last = skb;
3918 __skb_header_release(skb);
3922 NAPI_GRO_CB(p)->count++;
3924 p->truesize += delta_truesize;
3927 lp->data_len += len;
3928 lp->truesize += delta_truesize;
3931 NAPI_GRO_CB(skb)->same_flow = 1;
3934 EXPORT_SYMBOL_GPL(skb_gro_receive);
3936 void __init skb_init(void)
3938 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
3939 sizeof(struct sk_buff),
3941 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3942 offsetof(struct sk_buff, cb),
3943 sizeof_field(struct sk_buff, cb),
3945 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3946 sizeof(struct sk_buff_fclones),
3948 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3953 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
3954 unsigned int recursion_level)
3956 int start = skb_headlen(skb);
3957 int i, copy = start - offset;
3958 struct sk_buff *frag_iter;
3961 if (unlikely(recursion_level >= 24))
3967 sg_set_buf(sg, skb->data + offset, copy);
3969 if ((len -= copy) == 0)
3974 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3977 WARN_ON(start > offset + len);
3979 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3980 if ((copy = end - offset) > 0) {
3981 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3982 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
3987 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3988 frag->page_offset+offset-start);
3997 skb_walk_frags(skb, frag_iter) {
4000 WARN_ON(start > offset + len);
4002 end = start + frag_iter->len;
4003 if ((copy = end - offset) > 0) {
4004 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4009 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4010 copy, recursion_level + 1);
4011 if (unlikely(ret < 0))
4014 if ((len -= copy) == 0)
4025 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4026 * @skb: Socket buffer containing the buffers to be mapped
4027 * @sg: The scatter-gather list to map into
4028 * @offset: The offset into the buffer's contents to start mapping
4029 * @len: Length of buffer space to be mapped
4031 * Fill the specified scatter-gather list with mappings/pointers into a
4032 * region of the buffer space attached to a socket buffer. Returns either
4033 * the number of scatterlist items used, or -EMSGSIZE if the contents
4036 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4038 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4043 sg_mark_end(&sg[nsg - 1]);
4047 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4049 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4050 * sglist without mark the sg which contain last skb data as the end.
4051 * So the caller can mannipulate sg list as will when padding new data after
4052 * the first call without calling sg_unmark_end to expend sg list.
4054 * Scenario to use skb_to_sgvec_nomark:
4056 * 2. skb_to_sgvec_nomark(payload1)
4057 * 3. skb_to_sgvec_nomark(payload2)
4059 * This is equivalent to:
4061 * 2. skb_to_sgvec(payload1)
4063 * 4. skb_to_sgvec(payload2)
4065 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4066 * is more preferable.
4068 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4069 int offset, int len)
4071 return __skb_to_sgvec(skb, sg, offset, len, 0);
4073 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4078 * skb_cow_data - Check that a socket buffer's data buffers are writable
4079 * @skb: The socket buffer to check.
4080 * @tailbits: Amount of trailing space to be added
4081 * @trailer: Returned pointer to the skb where the @tailbits space begins
4083 * Make sure that the data buffers attached to a socket buffer are
4084 * writable. If they are not, private copies are made of the data buffers
4085 * and the socket buffer is set to use these instead.
4087 * If @tailbits is given, make sure that there is space to write @tailbits
4088 * bytes of data beyond current end of socket buffer. @trailer will be
4089 * set to point to the skb in which this space begins.
4091 * The number of scatterlist elements required to completely map the
4092 * COW'd and extended socket buffer will be returned.
4094 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4098 struct sk_buff *skb1, **skb_p;
4100 /* If skb is cloned or its head is paged, reallocate
4101 * head pulling out all the pages (pages are considered not writable
4102 * at the moment even if they are anonymous).
4104 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4105 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4108 /* Easy case. Most of packets will go this way. */
4109 if (!skb_has_frag_list(skb)) {
4110 /* A little of trouble, not enough of space for trailer.
4111 * This should not happen, when stack is tuned to generate
4112 * good frames. OK, on miss we reallocate and reserve even more
4113 * space, 128 bytes is fair. */
4115 if (skb_tailroom(skb) < tailbits &&
4116 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4124 /* Misery. We are in troubles, going to mincer fragments... */
4127 skb_p = &skb_shinfo(skb)->frag_list;
4130 while ((skb1 = *skb_p) != NULL) {
4133 /* The fragment is partially pulled by someone,
4134 * this can happen on input. Copy it and everything
4137 if (skb_shared(skb1))
4140 /* If the skb is the last, worry about trailer. */
4142 if (skb1->next == NULL && tailbits) {
4143 if (skb_shinfo(skb1)->nr_frags ||
4144 skb_has_frag_list(skb1) ||
4145 skb_tailroom(skb1) < tailbits)
4146 ntail = tailbits + 128;
4152 skb_shinfo(skb1)->nr_frags ||
4153 skb_has_frag_list(skb1)) {
4154 struct sk_buff *skb2;
4156 /* Fuck, we are miserable poor guys... */
4158 skb2 = skb_copy(skb1, GFP_ATOMIC);
4160 skb2 = skb_copy_expand(skb1,
4164 if (unlikely(skb2 == NULL))
4168 skb_set_owner_w(skb2, skb1->sk);
4170 /* Looking around. Are we still alive?
4171 * OK, link new skb, drop old one */
4173 skb2->next = skb1->next;
4180 skb_p = &skb1->next;
4185 EXPORT_SYMBOL_GPL(skb_cow_data);
4187 static void sock_rmem_free(struct sk_buff *skb)
4189 struct sock *sk = skb->sk;
4191 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4194 static void skb_set_err_queue(struct sk_buff *skb)
4196 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4197 * So, it is safe to (mis)use it to mark skbs on the error queue.
4199 skb->pkt_type = PACKET_OUTGOING;
4200 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4204 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4206 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4208 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4209 (unsigned int)sk->sk_rcvbuf)
4214 skb->destructor = sock_rmem_free;
4215 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4216 skb_set_err_queue(skb);
4218 /* before exiting rcu section, make sure dst is refcounted */
4221 skb_queue_tail(&sk->sk_error_queue, skb);
4222 if (!sock_flag(sk, SOCK_DEAD))
4223 sk->sk_error_report(sk);
4226 EXPORT_SYMBOL(sock_queue_err_skb);
4228 static bool is_icmp_err_skb(const struct sk_buff *skb)
4230 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4231 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4234 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4236 struct sk_buff_head *q = &sk->sk_error_queue;
4237 struct sk_buff *skb, *skb_next = NULL;
4238 bool icmp_next = false;
4239 unsigned long flags;
4241 spin_lock_irqsave(&q->lock, flags);
4242 skb = __skb_dequeue(q);
4243 if (skb && (skb_next = skb_peek(q))) {
4244 icmp_next = is_icmp_err_skb(skb_next);
4246 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4248 spin_unlock_irqrestore(&q->lock, flags);
4250 if (is_icmp_err_skb(skb) && !icmp_next)
4254 sk->sk_error_report(sk);
4258 EXPORT_SYMBOL(sock_dequeue_err_skb);
4261 * skb_clone_sk - create clone of skb, and take reference to socket
4262 * @skb: the skb to clone
4264 * This function creates a clone of a buffer that holds a reference on
4265 * sk_refcnt. Buffers created via this function are meant to be
4266 * returned using sock_queue_err_skb, or free via kfree_skb.
4268 * When passing buffers allocated with this function to sock_queue_err_skb
4269 * it is necessary to wrap the call with sock_hold/sock_put in order to
4270 * prevent the socket from being released prior to being enqueued on
4271 * the sk_error_queue.
4273 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4275 struct sock *sk = skb->sk;
4276 struct sk_buff *clone;
4278 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4281 clone = skb_clone(skb, GFP_ATOMIC);
4288 clone->destructor = sock_efree;
4292 EXPORT_SYMBOL(skb_clone_sk);
4294 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4299 struct sock_exterr_skb *serr;
4302 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4304 serr = SKB_EXT_ERR(skb);
4305 memset(serr, 0, sizeof(*serr));
4306 serr->ee.ee_errno = ENOMSG;
4307 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4308 serr->ee.ee_info = tstype;
4309 serr->opt_stats = opt_stats;
4310 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4311 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4312 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4313 if (sk->sk_protocol == IPPROTO_TCP &&
4314 sk->sk_type == SOCK_STREAM)
4315 serr->ee.ee_data -= sk->sk_tskey;
4318 err = sock_queue_err_skb(sk, skb);
4324 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4328 if (likely(sysctl_tstamp_allow_data || tsonly))
4331 read_lock_bh(&sk->sk_callback_lock);
4332 ret = sk->sk_socket && sk->sk_socket->file &&
4333 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4334 read_unlock_bh(&sk->sk_callback_lock);
4338 void skb_complete_tx_timestamp(struct sk_buff *skb,
4339 struct skb_shared_hwtstamps *hwtstamps)
4341 struct sock *sk = skb->sk;
4343 if (!skb_may_tx_timestamp(sk, false))
4346 /* Take a reference to prevent skb_orphan() from freeing the socket,
4347 * but only if the socket refcount is not zero.
4349 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4350 *skb_hwtstamps(skb) = *hwtstamps;
4351 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4359 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4361 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4362 struct skb_shared_hwtstamps *hwtstamps,
4363 struct sock *sk, int tstype)
4365 struct sk_buff *skb;
4366 bool tsonly, opt_stats = false;
4371 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4372 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4375 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4376 if (!skb_may_tx_timestamp(sk, tsonly))
4381 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4382 sk->sk_protocol == IPPROTO_TCP &&
4383 sk->sk_type == SOCK_STREAM) {
4384 skb = tcp_get_timestamping_opt_stats(sk);
4388 skb = alloc_skb(0, GFP_ATOMIC);
4390 skb = skb_clone(orig_skb, GFP_ATOMIC);
4396 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4398 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4402 *skb_hwtstamps(skb) = *hwtstamps;
4404 skb->tstamp = ktime_get_real();
4406 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4408 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4410 void skb_tstamp_tx(struct sk_buff *orig_skb,
4411 struct skb_shared_hwtstamps *hwtstamps)
4413 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4416 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4418 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4420 struct sock *sk = skb->sk;
4421 struct sock_exterr_skb *serr;
4424 skb->wifi_acked_valid = 1;
4425 skb->wifi_acked = acked;
4427 serr = SKB_EXT_ERR(skb);
4428 memset(serr, 0, sizeof(*serr));
4429 serr->ee.ee_errno = ENOMSG;
4430 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4432 /* Take a reference to prevent skb_orphan() from freeing the socket,
4433 * but only if the socket refcount is not zero.
4435 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4436 err = sock_queue_err_skb(sk, skb);
4442 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4445 * skb_partial_csum_set - set up and verify partial csum values for packet
4446 * @skb: the skb to set
4447 * @start: the number of bytes after skb->data to start checksumming.
4448 * @off: the offset from start to place the checksum.
4450 * For untrusted partially-checksummed packets, we need to make sure the values
4451 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4453 * This function checks and sets those values and skb->ip_summed: if this
4454 * returns false you should drop the packet.
4456 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4458 if (unlikely(start > skb_headlen(skb)) ||
4459 unlikely((int)start + off > skb_headlen(skb) - 2)) {
4460 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
4461 start, off, skb_headlen(skb));
4464 skb->ip_summed = CHECKSUM_PARTIAL;
4465 skb->csum_start = skb_headroom(skb) + start;
4466 skb->csum_offset = off;
4467 skb_set_transport_header(skb, start);
4470 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4472 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4475 if (skb_headlen(skb) >= len)
4478 /* If we need to pullup then pullup to the max, so we
4479 * won't need to do it again.
4484 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4487 if (skb_headlen(skb) < len)
4493 #define MAX_TCP_HDR_LEN (15 * 4)
4495 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4496 typeof(IPPROTO_IP) proto,
4503 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4504 off + MAX_TCP_HDR_LEN);
4505 if (!err && !skb_partial_csum_set(skb, off,
4506 offsetof(struct tcphdr,
4509 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4512 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4513 off + sizeof(struct udphdr));
4514 if (!err && !skb_partial_csum_set(skb, off,
4515 offsetof(struct udphdr,
4518 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4521 return ERR_PTR(-EPROTO);
4524 /* This value should be large enough to cover a tagged ethernet header plus
4525 * maximally sized IP and TCP or UDP headers.
4527 #define MAX_IP_HDR_LEN 128
4529 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4538 err = skb_maybe_pull_tail(skb,
4539 sizeof(struct iphdr),
4544 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4547 off = ip_hdrlen(skb);
4554 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4556 return PTR_ERR(csum);
4559 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4562 ip_hdr(skb)->protocol, 0);
4569 /* This value should be large enough to cover a tagged ethernet header plus
4570 * an IPv6 header, all options, and a maximal TCP or UDP header.
4572 #define MAX_IPV6_HDR_LEN 256
4574 #define OPT_HDR(type, skb, off) \
4575 (type *)(skb_network_header(skb) + (off))
4577 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4590 off = sizeof(struct ipv6hdr);
4592 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4596 nexthdr = ipv6_hdr(skb)->nexthdr;
4598 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4599 while (off <= len && !done) {
4601 case IPPROTO_DSTOPTS:
4602 case IPPROTO_HOPOPTS:
4603 case IPPROTO_ROUTING: {
4604 struct ipv6_opt_hdr *hp;
4606 err = skb_maybe_pull_tail(skb,
4608 sizeof(struct ipv6_opt_hdr),
4613 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4614 nexthdr = hp->nexthdr;
4615 off += ipv6_optlen(hp);
4619 struct ip_auth_hdr *hp;
4621 err = skb_maybe_pull_tail(skb,
4623 sizeof(struct ip_auth_hdr),
4628 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4629 nexthdr = hp->nexthdr;
4630 off += ipv6_authlen(hp);
4633 case IPPROTO_FRAGMENT: {
4634 struct frag_hdr *hp;
4636 err = skb_maybe_pull_tail(skb,
4638 sizeof(struct frag_hdr),
4643 hp = OPT_HDR(struct frag_hdr, skb, off);
4645 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4648 nexthdr = hp->nexthdr;
4649 off += sizeof(struct frag_hdr);
4660 if (!done || fragment)
4663 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4665 return PTR_ERR(csum);
4668 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4669 &ipv6_hdr(skb)->daddr,
4670 skb->len - off, nexthdr, 0);
4678 * skb_checksum_setup - set up partial checksum offset
4679 * @skb: the skb to set up
4680 * @recalculate: if true the pseudo-header checksum will be recalculated
4682 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4686 switch (skb->protocol) {
4687 case htons(ETH_P_IP):
4688 err = skb_checksum_setup_ipv4(skb, recalculate);
4691 case htons(ETH_P_IPV6):
4692 err = skb_checksum_setup_ipv6(skb, recalculate);
4702 EXPORT_SYMBOL(skb_checksum_setup);
4705 * skb_checksum_maybe_trim - maybe trims the given skb
4706 * @skb: the skb to check
4707 * @transport_len: the data length beyond the network header
4709 * Checks whether the given skb has data beyond the given transport length.
4710 * If so, returns a cloned skb trimmed to this transport length.
4711 * Otherwise returns the provided skb. Returns NULL in error cases
4712 * (e.g. transport_len exceeds skb length or out-of-memory).
4714 * Caller needs to set the skb transport header and free any returned skb if it
4715 * differs from the provided skb.
4717 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4718 unsigned int transport_len)
4720 struct sk_buff *skb_chk;
4721 unsigned int len = skb_transport_offset(skb) + transport_len;
4726 else if (skb->len == len)
4729 skb_chk = skb_clone(skb, GFP_ATOMIC);
4733 ret = pskb_trim_rcsum(skb_chk, len);
4743 * skb_checksum_trimmed - validate checksum of an skb
4744 * @skb: the skb to check
4745 * @transport_len: the data length beyond the network header
4746 * @skb_chkf: checksum function to use
4748 * Applies the given checksum function skb_chkf to the provided skb.
4749 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4751 * If the skb has data beyond the given transport length, then a
4752 * trimmed & cloned skb is checked and returned.
4754 * Caller needs to set the skb transport header and free any returned skb if it
4755 * differs from the provided skb.
4757 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4758 unsigned int transport_len,
4759 __sum16(*skb_chkf)(struct sk_buff *skb))
4761 struct sk_buff *skb_chk;
4762 unsigned int offset = skb_transport_offset(skb);
4765 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4769 if (!pskb_may_pull(skb_chk, offset))
4772 skb_pull_rcsum(skb_chk, offset);
4773 ret = skb_chkf(skb_chk);
4774 skb_push_rcsum(skb_chk, offset);
4782 if (skb_chk && skb_chk != skb)
4788 EXPORT_SYMBOL(skb_checksum_trimmed);
4790 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4792 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4795 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4797 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4800 skb_release_head_state(skb);
4801 kmem_cache_free(skbuff_head_cache, skb);
4806 EXPORT_SYMBOL(kfree_skb_partial);
4809 * skb_try_coalesce - try to merge skb to prior one
4811 * @from: buffer to add
4812 * @fragstolen: pointer to boolean
4813 * @delta_truesize: how much more was allocated than was requested
4815 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4816 bool *fragstolen, int *delta_truesize)
4818 struct skb_shared_info *to_shinfo, *from_shinfo;
4819 int i, delta, len = from->len;
4821 *fragstolen = false;
4826 if (len <= skb_tailroom(to)) {
4828 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4829 *delta_truesize = 0;
4833 to_shinfo = skb_shinfo(to);
4834 from_shinfo = skb_shinfo(from);
4835 if (to_shinfo->frag_list || from_shinfo->frag_list)
4837 if (skb_zcopy(to) || skb_zcopy(from))
4840 if (skb_headlen(from) != 0) {
4842 unsigned int offset;
4844 if (to_shinfo->nr_frags +
4845 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
4848 if (skb_head_is_locked(from))
4851 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4853 page = virt_to_head_page(from->head);
4854 offset = from->data - (unsigned char *)page_address(page);
4856 skb_fill_page_desc(to, to_shinfo->nr_frags,
4857 page, offset, skb_headlen(from));
4860 if (to_shinfo->nr_frags +
4861 from_shinfo->nr_frags > MAX_SKB_FRAGS)
4864 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4867 WARN_ON_ONCE(delta < len);
4869 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
4871 from_shinfo->nr_frags * sizeof(skb_frag_t));
4872 to_shinfo->nr_frags += from_shinfo->nr_frags;
4874 if (!skb_cloned(from))
4875 from_shinfo->nr_frags = 0;
4877 /* if the skb is not cloned this does nothing
4878 * since we set nr_frags to 0.
4880 for (i = 0; i < from_shinfo->nr_frags; i++)
4881 __skb_frag_ref(&from_shinfo->frags[i]);
4883 to->truesize += delta;
4885 to->data_len += len;
4887 *delta_truesize = delta;
4890 EXPORT_SYMBOL(skb_try_coalesce);
4893 * skb_scrub_packet - scrub an skb
4895 * @skb: buffer to clean
4896 * @xnet: packet is crossing netns
4898 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4899 * into/from a tunnel. Some information have to be cleared during these
4901 * skb_scrub_packet can also be used to clean a skb before injecting it in
4902 * another namespace (@xnet == true). We have to clear all information in the
4903 * skb that could impact namespace isolation.
4905 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4907 skb->pkt_type = PACKET_HOST;
4913 nf_reset_trace(skb);
4922 EXPORT_SYMBOL_GPL(skb_scrub_packet);
4925 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4929 * skb_gso_transport_seglen is used to determine the real size of the
4930 * individual segments, including Layer4 headers (TCP/UDP).
4932 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4934 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4936 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4937 unsigned int thlen = 0;
4939 if (skb->encapsulation) {
4940 thlen = skb_inner_transport_header(skb) -
4941 skb_transport_header(skb);
4943 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4944 thlen += inner_tcp_hdrlen(skb);
4945 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4946 thlen = tcp_hdrlen(skb);
4947 } else if (unlikely(skb_is_gso_sctp(skb))) {
4948 thlen = sizeof(struct sctphdr);
4949 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
4950 thlen = sizeof(struct udphdr);
4952 /* UFO sets gso_size to the size of the fragmentation
4953 * payload, i.e. the size of the L4 (UDP) header is already
4956 return thlen + shinfo->gso_size;
4960 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4964 * skb_gso_network_seglen is used to determine the real size of the
4965 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4967 * The MAC/L2 header is not accounted for.
4969 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4971 unsigned int hdr_len = skb_transport_header(skb) -
4972 skb_network_header(skb);
4974 return hdr_len + skb_gso_transport_seglen(skb);
4978 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
4982 * skb_gso_mac_seglen is used to determine the real size of the
4983 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
4984 * headers (TCP/UDP).
4986 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
4988 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
4990 return hdr_len + skb_gso_transport_seglen(skb);
4994 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
4996 * There are a couple of instances where we have a GSO skb, and we
4997 * want to determine what size it would be after it is segmented.
4999 * We might want to check:
5000 * - L3+L4+payload size (e.g. IP forwarding)
5001 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5003 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5005 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5006 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5008 * @max_len: The maximum permissible length.
5010 * Returns true if the segmented length <= max length.
5012 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5013 unsigned int seg_len,
5014 unsigned int max_len) {
5015 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5016 const struct sk_buff *iter;
5018 if (shinfo->gso_size != GSO_BY_FRAGS)
5019 return seg_len <= max_len;
5021 /* Undo this so we can re-use header sizes */
5022 seg_len -= GSO_BY_FRAGS;
5024 skb_walk_frags(skb, iter) {
5025 if (seg_len + skb_headlen(iter) > max_len)
5033 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5036 * @mtu: MTU to validate against
5038 * skb_gso_validate_network_len validates if a given skb will fit a
5039 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5042 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5044 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5046 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5049 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5052 * @len: length to validate against
5054 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5055 * length once split, including L2, L3 and L4 headers and the payload.
5057 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5059 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5061 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5063 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5067 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5072 mac_len = skb->data - skb_mac_header(skb);
5073 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5074 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5075 mac_len - VLAN_HLEN - ETH_TLEN);
5077 skb->mac_header += VLAN_HLEN;
5081 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5083 struct vlan_hdr *vhdr;
5086 if (unlikely(skb_vlan_tag_present(skb))) {
5087 /* vlan_tci is already set-up so leave this for another time */
5091 skb = skb_share_check(skb, GFP_ATOMIC);
5095 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5098 vhdr = (struct vlan_hdr *)skb->data;
5099 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5100 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5102 skb_pull_rcsum(skb, VLAN_HLEN);
5103 vlan_set_encap_proto(skb, vhdr);
5105 skb = skb_reorder_vlan_header(skb);
5109 skb_reset_network_header(skb);
5110 skb_reset_transport_header(skb);
5111 skb_reset_mac_len(skb);
5119 EXPORT_SYMBOL(skb_vlan_untag);
5121 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5123 if (!pskb_may_pull(skb, write_len))
5126 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5129 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5131 EXPORT_SYMBOL(skb_ensure_writable);
5133 /* remove VLAN header from packet and update csum accordingly.
5134 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5136 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5138 struct vlan_hdr *vhdr;
5139 int offset = skb->data - skb_mac_header(skb);
5142 if (WARN_ONCE(offset,
5143 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5148 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5152 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5154 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5155 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5157 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5158 __skb_pull(skb, VLAN_HLEN);
5160 vlan_set_encap_proto(skb, vhdr);
5161 skb->mac_header += VLAN_HLEN;
5163 if (skb_network_offset(skb) < ETH_HLEN)
5164 skb_set_network_header(skb, ETH_HLEN);
5166 skb_reset_mac_len(skb);
5170 EXPORT_SYMBOL(__skb_vlan_pop);
5172 /* Pop a vlan tag either from hwaccel or from payload.
5173 * Expects skb->data at mac header.
5175 int skb_vlan_pop(struct sk_buff *skb)
5181 if (likely(skb_vlan_tag_present(skb))) {
5184 if (unlikely(!eth_type_vlan(skb->protocol)))
5187 err = __skb_vlan_pop(skb, &vlan_tci);
5191 /* move next vlan tag to hw accel tag */
5192 if (likely(!eth_type_vlan(skb->protocol)))
5195 vlan_proto = skb->protocol;
5196 err = __skb_vlan_pop(skb, &vlan_tci);
5200 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5203 EXPORT_SYMBOL(skb_vlan_pop);
5205 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5206 * Expects skb->data at mac header.
5208 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5210 if (skb_vlan_tag_present(skb)) {
5211 int offset = skb->data - skb_mac_header(skb);
5214 if (WARN_ONCE(offset,
5215 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5220 err = __vlan_insert_tag(skb, skb->vlan_proto,
5221 skb_vlan_tag_get(skb));
5225 skb->protocol = skb->vlan_proto;
5226 skb->mac_len += VLAN_HLEN;
5228 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5230 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5233 EXPORT_SYMBOL(skb_vlan_push);
5236 * alloc_skb_with_frags - allocate skb with page frags
5238 * @header_len: size of linear part
5239 * @data_len: needed length in frags
5240 * @max_page_order: max page order desired.
5241 * @errcode: pointer to error code if any
5242 * @gfp_mask: allocation mask
5244 * This can be used to allocate a paged skb, given a maximal order for frags.
5246 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5247 unsigned long data_len,
5252 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5253 unsigned long chunk;
5254 struct sk_buff *skb;
5259 *errcode = -EMSGSIZE;
5260 /* Note this test could be relaxed, if we succeed to allocate
5261 * high order pages...
5263 if (npages > MAX_SKB_FRAGS)
5266 gfp_head = gfp_mask;
5267 if (gfp_head & __GFP_DIRECT_RECLAIM)
5268 gfp_head |= __GFP_RETRY_MAYFAIL;
5270 *errcode = -ENOBUFS;
5271 skb = alloc_skb(header_len, gfp_head);
5275 skb->truesize += npages << PAGE_SHIFT;
5277 for (i = 0; npages > 0; i++) {
5278 int order = max_page_order;
5281 if (npages >= 1 << order) {
5282 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5288 /* Do not retry other high order allocations */
5294 page = alloc_page(gfp_mask);
5298 chunk = min_t(unsigned long, data_len,
5299 PAGE_SIZE << order);
5300 skb_fill_page_desc(skb, i, page, 0, chunk);
5302 npages -= 1 << order;
5310 EXPORT_SYMBOL(alloc_skb_with_frags);
5312 /* carve out the first off bytes from skb when off < headlen */
5313 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5314 const int headlen, gfp_t gfp_mask)
5317 int size = skb_end_offset(skb);
5318 int new_hlen = headlen - off;
5321 size = SKB_DATA_ALIGN(size);
5323 if (skb_pfmemalloc(skb))
5324 gfp_mask |= __GFP_MEMALLOC;
5325 data = kmalloc_reserve(size +
5326 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5327 gfp_mask, NUMA_NO_NODE, NULL);
5331 size = SKB_WITH_OVERHEAD(ksize(data));
5333 /* Copy real data, and all frags */
5334 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5337 memcpy((struct skb_shared_info *)(data + size),
5339 offsetof(struct skb_shared_info,
5340 frags[skb_shinfo(skb)->nr_frags]));
5341 if (skb_cloned(skb)) {
5342 /* drop the old head gracefully */
5343 if (skb_orphan_frags(skb, gfp_mask)) {
5347 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5348 skb_frag_ref(skb, i);
5349 if (skb_has_frag_list(skb))
5350 skb_clone_fraglist(skb);
5351 skb_release_data(skb);
5353 /* we can reuse existing recount- all we did was
5362 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5365 skb->end = skb->head + size;
5367 skb_set_tail_pointer(skb, skb_headlen(skb));
5368 skb_headers_offset_update(skb, 0);
5372 atomic_set(&skb_shinfo(skb)->dataref, 1);
5377 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5379 /* carve out the first eat bytes from skb's frag_list. May recurse into
5382 static int pskb_carve_frag_list(struct sk_buff *skb,
5383 struct skb_shared_info *shinfo, int eat,
5386 struct sk_buff *list = shinfo->frag_list;
5387 struct sk_buff *clone = NULL;
5388 struct sk_buff *insp = NULL;
5392 pr_err("Not enough bytes to eat. Want %d\n", eat);
5395 if (list->len <= eat) {
5396 /* Eaten as whole. */
5401 /* Eaten partially. */
5402 if (skb_shared(list)) {
5403 clone = skb_clone(list, gfp_mask);
5409 /* This may be pulled without problems. */
5412 if (pskb_carve(list, eat, gfp_mask) < 0) {
5420 /* Free pulled out fragments. */
5421 while ((list = shinfo->frag_list) != insp) {
5422 shinfo->frag_list = list->next;
5425 /* And insert new clone at head. */
5428 shinfo->frag_list = clone;
5433 /* carve off first len bytes from skb. Split line (off) is in the
5434 * non-linear part of skb
5436 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5437 int pos, gfp_t gfp_mask)
5440 int size = skb_end_offset(skb);
5442 const int nfrags = skb_shinfo(skb)->nr_frags;
5443 struct skb_shared_info *shinfo;
5445 size = SKB_DATA_ALIGN(size);
5447 if (skb_pfmemalloc(skb))
5448 gfp_mask |= __GFP_MEMALLOC;
5449 data = kmalloc_reserve(size +
5450 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5451 gfp_mask, NUMA_NO_NODE, NULL);
5455 size = SKB_WITH_OVERHEAD(ksize(data));
5457 memcpy((struct skb_shared_info *)(data + size),
5458 skb_shinfo(skb), offsetof(struct skb_shared_info,
5459 frags[skb_shinfo(skb)->nr_frags]));
5460 if (skb_orphan_frags(skb, gfp_mask)) {
5464 shinfo = (struct skb_shared_info *)(data + size);
5465 for (i = 0; i < nfrags; i++) {
5466 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5468 if (pos + fsize > off) {
5469 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5473 * We have two variants in this case:
5474 * 1. Move all the frag to the second
5475 * part, if it is possible. F.e.
5476 * this approach is mandatory for TUX,
5477 * where splitting is expensive.
5478 * 2. Split is accurately. We make this.
5480 shinfo->frags[0].page_offset += off - pos;
5481 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5483 skb_frag_ref(skb, i);
5488 shinfo->nr_frags = k;
5489 if (skb_has_frag_list(skb))
5490 skb_clone_fraglist(skb);
5493 /* split line is in frag list */
5494 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5496 skb_release_data(skb);
5501 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5504 skb->end = skb->head + size;
5506 skb_reset_tail_pointer(skb);
5507 skb_headers_offset_update(skb, 0);
5512 skb->data_len = skb->len;
5513 atomic_set(&skb_shinfo(skb)->dataref, 1);
5517 /* remove len bytes from the beginning of the skb */
5518 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5520 int headlen = skb_headlen(skb);
5523 return pskb_carve_inside_header(skb, len, headlen, gfp);
5525 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5528 /* Extract to_copy bytes starting at off from skb, and return this in
5531 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5532 int to_copy, gfp_t gfp)
5534 struct sk_buff *clone = skb_clone(skb, gfp);
5539 if (pskb_carve(clone, off, gfp) < 0 ||
5540 pskb_trim(clone, to_copy)) {
5546 EXPORT_SYMBOL(pskb_extract);
5549 * skb_condense - try to get rid of fragments/frag_list if possible
5552 * Can be used to save memory before skb is added to a busy queue.
5553 * If packet has bytes in frags and enough tail room in skb->head,
5554 * pull all of them, so that we can free the frags right now and adjust
5557 * We do not reallocate skb->head thus can not fail.
5558 * Caller must re-evaluate skb->truesize if needed.
5560 void skb_condense(struct sk_buff *skb)
5562 if (skb->data_len) {
5563 if (skb->data_len > skb->end - skb->tail ||
5567 /* Nice, we can free page frag(s) right now */
5568 __pskb_pull_tail(skb, skb->data_len);
5570 /* At this point, skb->truesize might be over estimated,
5571 * because skb had a fragment, and fragments do not tell
5573 * When we pulled its content into skb->head, fragment
5574 * was freed, but __pskb_pull_tail() could not possibly
5575 * adjust skb->truesize, not knowing the frag truesize.
5577 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));