1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * Routines having to do with the 'struct sk_buff' memory handlers.
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
9 * Alan Cox : Fixed the worst of the load
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
32 * The functions in this file will not compile correctly with gcc 2.4.x
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
41 #include <linux/interrupt.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
64 #include <net/protocol.h>
67 #include <net/checksum.h>
68 #include <net/ip6_checksum.h>
71 #include <net/mptcp.h>
73 #include <linux/uaccess.h>
74 #include <trace/events/skb.h>
75 #include <linux/highmem.h>
76 #include <linux/capability.h>
77 #include <linux/user_namespace.h>
78 #include <linux/indirect_call_wrapper.h>
82 struct kmem_cache *skbuff_head_cache __ro_after_init;
83 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84 #ifdef CONFIG_SKB_EXTENSIONS
85 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
87 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88 EXPORT_SYMBOL(sysctl_max_skb_frags);
91 * skb_panic - private function for out-of-line support
95 * @msg: skb_over_panic or skb_under_panic
97 * Out-of-line support for skb_put() and skb_push().
98 * Called via the wrapper skb_over_panic() or skb_under_panic().
99 * Keep out of line to prevent kernel bloat.
100 * __builtin_return_address is not used because it is not always reliable.
102 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
105 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
112 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
117 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
119 skb_panic(skb, sz, addr, __func__);
123 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
124 * the caller if emergency pfmemalloc reserves are being used. If it is and
125 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
126 * may be used. Otherwise, the packet data may be discarded until enough
129 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
132 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
136 bool ret_pfmemalloc = false;
139 * Try a regular allocation, when that fails and we're not entitled
140 * to the reserves, fail.
142 obj = kmalloc_node_track_caller(size,
143 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
145 if (obj || !(gfp_pfmemalloc_allowed(flags)))
148 /* Try again but now we are using pfmemalloc reserves */
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
154 *pfmemalloc = ret_pfmemalloc;
159 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
160 * 'private' fields and also do memory statistics to find all the
166 * __alloc_skb - allocate a network buffer
167 * @size: size to allocate
168 * @gfp_mask: allocation mask
169 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
170 * instead of head cache and allocate a cloned (child) skb.
171 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
172 * allocations in case the data is required for writeback
173 * @node: numa node to allocate memory on
175 * Allocate a new &sk_buff. The returned buffer has no headroom and a
176 * tail room of at least size bytes. The object has a reference count
177 * of one. The return is the buffer. On a failure the return is %NULL.
179 * Buffers may only be allocated from interrupts using a @gfp_mask of
182 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
198 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
203 /* We do our best to align skb_shared_info on a separate cache
204 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
205 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
206 * Both skb->head and skb_shared_info are cache line aligned.
208 size = SKB_DATA_ALIGN(size);
209 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
213 /* kmalloc(size) might give us more room than requested.
214 * Put skb_shared_info exactly at the end of allocated zone,
215 * to allow max possible filling before reallocation.
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
221 * Only clear those fields we need to clear, not those that we will
222 * actually initialise below. Hence, don't put any more fields after
223 * the tail pointer in struct sk_buff!
225 memset(skb, 0, offsetof(struct sk_buff, tail));
226 /* Account for allocated memory : skb + skb->head */
227 skb->truesize = SKB_TRUESIZE(size);
228 skb->pfmemalloc = pfmemalloc;
229 refcount_set(&skb->users, 1);
232 skb_reset_tail_pointer(skb);
233 skb->end = skb->tail + size;
234 skb->mac_header = (typeof(skb->mac_header))~0U;
235 skb->transport_header = (typeof(skb->transport_header))~0U;
237 /* make sure we initialize shinfo sequentially */
238 shinfo = skb_shinfo(skb);
239 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240 atomic_set(&shinfo->dataref, 1);
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
253 skb_set_kcov_handle(skb, kcov_common_handle());
258 kmem_cache_free(cache, skb);
262 EXPORT_SYMBOL(__alloc_skb);
264 /* Caller must provide SKB that is memset cleared */
265 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
266 void *data, unsigned int frag_size)
268 struct skb_shared_info *shinfo;
269 unsigned int size = frag_size ? : ksize(data);
271 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
273 /* Assumes caller memset cleared SKB */
274 skb->truesize = SKB_TRUESIZE(size);
275 refcount_set(&skb->users, 1);
278 skb_reset_tail_pointer(skb);
279 skb->end = skb->tail + size;
280 skb->mac_header = (typeof(skb->mac_header))~0U;
281 skb->transport_header = (typeof(skb->transport_header))~0U;
283 /* make sure we initialize shinfo sequentially */
284 shinfo = skb_shinfo(skb);
285 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
286 atomic_set(&shinfo->dataref, 1);
288 skb_set_kcov_handle(skb, kcov_common_handle());
294 * __build_skb - build a network buffer
295 * @data: data buffer provided by caller
296 * @frag_size: size of data, or 0 if head was kmalloced
298 * Allocate a new &sk_buff. Caller provides space holding head and
299 * skb_shared_info. @data must have been allocated by kmalloc() only if
300 * @frag_size is 0, otherwise data should come from the page allocator
302 * The return is the new skb buffer.
303 * On a failure the return is %NULL, and @data is not freed.
305 * Before IO, driver allocates only data buffer where NIC put incoming frame
306 * Driver should add room at head (NET_SKB_PAD) and
307 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
308 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
309 * before giving packet to stack.
310 * RX rings only contains data buffers, not full skbs.
312 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
316 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
320 memset(skb, 0, offsetof(struct sk_buff, tail));
322 return __build_skb_around(skb, data, frag_size);
325 /* build_skb() is wrapper over __build_skb(), that specifically
326 * takes care of skb->head and skb->pfmemalloc
327 * This means that if @frag_size is not zero, then @data must be backed
328 * by a page fragment, not kmalloc() or vmalloc()
330 struct sk_buff *build_skb(void *data, unsigned int frag_size)
332 struct sk_buff *skb = __build_skb(data, frag_size);
334 if (skb && frag_size) {
336 if (page_is_pfmemalloc(virt_to_head_page(data)))
341 EXPORT_SYMBOL(build_skb);
344 * build_skb_around - build a network buffer around provided skb
345 * @skb: sk_buff provide by caller, must be memset cleared
346 * @data: data buffer provided by caller
347 * @frag_size: size of data, or 0 if head was kmalloced
349 struct sk_buff *build_skb_around(struct sk_buff *skb,
350 void *data, unsigned int frag_size)
355 skb = __build_skb_around(skb, data, frag_size);
357 if (skb && frag_size) {
359 if (page_is_pfmemalloc(virt_to_head_page(data)))
364 EXPORT_SYMBOL(build_skb_around);
366 #define NAPI_SKB_CACHE_SIZE 64
368 struct napi_alloc_cache {
369 struct page_frag_cache page;
370 unsigned int skb_count;
371 void *skb_cache[NAPI_SKB_CACHE_SIZE];
374 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
375 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
377 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
379 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
381 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
384 void *napi_alloc_frag(unsigned int fragsz)
386 fragsz = SKB_DATA_ALIGN(fragsz);
388 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
390 EXPORT_SYMBOL(napi_alloc_frag);
393 * netdev_alloc_frag - allocate a page fragment
394 * @fragsz: fragment size
396 * Allocates a frag from a page for receive buffer.
397 * Uses GFP_ATOMIC allocations.
399 void *netdev_alloc_frag(unsigned int fragsz)
401 struct page_frag_cache *nc;
404 fragsz = SKB_DATA_ALIGN(fragsz);
405 if (in_irq() || irqs_disabled()) {
406 nc = this_cpu_ptr(&netdev_alloc_cache);
407 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
410 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
415 EXPORT_SYMBOL(netdev_alloc_frag);
418 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
419 * @dev: network device to receive on
420 * @len: length to allocate
421 * @gfp_mask: get_free_pages mask, passed to alloc_skb
423 * Allocate a new &sk_buff and assign it a usage count of one. The
424 * buffer has NET_SKB_PAD headroom built in. Users should allocate
425 * the headroom they think they need without accounting for the
426 * built in space. The built in space is used for optimisations.
428 * %NULL is returned if there is no free memory.
430 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
433 struct page_frag_cache *nc;
440 /* If requested length is either too small or too big,
441 * we use kmalloc() for skb->head allocation.
443 if (len <= SKB_WITH_OVERHEAD(1024) ||
444 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
445 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
446 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
452 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
453 len = SKB_DATA_ALIGN(len);
455 if (sk_memalloc_socks())
456 gfp_mask |= __GFP_MEMALLOC;
458 if (in_irq() || irqs_disabled()) {
459 nc = this_cpu_ptr(&netdev_alloc_cache);
460 data = page_frag_alloc(nc, len, gfp_mask);
461 pfmemalloc = nc->pfmemalloc;
464 nc = this_cpu_ptr(&napi_alloc_cache.page);
465 data = page_frag_alloc(nc, len, gfp_mask);
466 pfmemalloc = nc->pfmemalloc;
473 skb = __build_skb(data, len);
474 if (unlikely(!skb)) {
484 skb_reserve(skb, NET_SKB_PAD);
490 EXPORT_SYMBOL(__netdev_alloc_skb);
493 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
494 * @napi: napi instance this buffer was allocated for
495 * @len: length to allocate
496 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
498 * Allocate a new sk_buff for use in NAPI receive. This buffer will
499 * attempt to allocate the head from a special reserved region used
500 * only for NAPI Rx allocation. By doing this we can save several
501 * CPU cycles by avoiding having to disable and re-enable IRQs.
503 * %NULL is returned if there is no free memory.
505 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
508 struct napi_alloc_cache *nc;
512 len += NET_SKB_PAD + NET_IP_ALIGN;
514 /* If requested length is either too small or too big,
515 * we use kmalloc() for skb->head allocation.
517 if (len <= SKB_WITH_OVERHEAD(1024) ||
518 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
519 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
520 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
526 nc = this_cpu_ptr(&napi_alloc_cache);
527 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
528 len = SKB_DATA_ALIGN(len);
530 if (sk_memalloc_socks())
531 gfp_mask |= __GFP_MEMALLOC;
533 data = page_frag_alloc(&nc->page, len, gfp_mask);
537 skb = __build_skb(data, len);
538 if (unlikely(!skb)) {
543 if (nc->page.pfmemalloc)
548 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
549 skb->dev = napi->dev;
554 EXPORT_SYMBOL(__napi_alloc_skb);
556 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
557 int size, unsigned int truesize)
559 skb_fill_page_desc(skb, i, page, off, size);
561 skb->data_len += size;
562 skb->truesize += truesize;
564 EXPORT_SYMBOL(skb_add_rx_frag);
566 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
567 unsigned int truesize)
569 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
571 skb_frag_size_add(frag, size);
573 skb->data_len += size;
574 skb->truesize += truesize;
576 EXPORT_SYMBOL(skb_coalesce_rx_frag);
578 static void skb_drop_list(struct sk_buff **listp)
580 kfree_skb_list(*listp);
584 static inline void skb_drop_fraglist(struct sk_buff *skb)
586 skb_drop_list(&skb_shinfo(skb)->frag_list);
589 static void skb_clone_fraglist(struct sk_buff *skb)
591 struct sk_buff *list;
593 skb_walk_frags(skb, list)
597 static void skb_free_head(struct sk_buff *skb)
599 unsigned char *head = skb->head;
607 static void skb_release_data(struct sk_buff *skb)
609 struct skb_shared_info *shinfo = skb_shinfo(skb);
613 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
617 for (i = 0; i < shinfo->nr_frags; i++)
618 __skb_frag_unref(&shinfo->frags[i]);
620 if (shinfo->frag_list)
621 kfree_skb_list(shinfo->frag_list);
623 skb_zcopy_clear(skb, true);
628 * Free an skbuff by memory without cleaning the state.
630 static void kfree_skbmem(struct sk_buff *skb)
632 struct sk_buff_fclones *fclones;
634 switch (skb->fclone) {
635 case SKB_FCLONE_UNAVAILABLE:
636 kmem_cache_free(skbuff_head_cache, skb);
639 case SKB_FCLONE_ORIG:
640 fclones = container_of(skb, struct sk_buff_fclones, skb1);
642 /* We usually free the clone (TX completion) before original skb
643 * This test would have no chance to be true for the clone,
644 * while here, branch prediction will be good.
646 if (refcount_read(&fclones->fclone_ref) == 1)
650 default: /* SKB_FCLONE_CLONE */
651 fclones = container_of(skb, struct sk_buff_fclones, skb2);
654 if (!refcount_dec_and_test(&fclones->fclone_ref))
657 kmem_cache_free(skbuff_fclone_cache, fclones);
660 void skb_release_head_state(struct sk_buff *skb)
663 if (skb->destructor) {
665 skb->destructor(skb);
667 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
668 nf_conntrack_put(skb_nfct(skb));
673 /* Free everything but the sk_buff shell. */
674 static void skb_release_all(struct sk_buff *skb)
676 skb_release_head_state(skb);
677 if (likely(skb->head))
678 skb_release_data(skb);
682 * __kfree_skb - private function
685 * Free an sk_buff. Release anything attached to the buffer.
686 * Clean the state. This is an internal helper function. Users should
687 * always call kfree_skb
690 void __kfree_skb(struct sk_buff *skb)
692 skb_release_all(skb);
695 EXPORT_SYMBOL(__kfree_skb);
698 * kfree_skb - free an sk_buff
699 * @skb: buffer to free
701 * Drop a reference to the buffer and free it if the usage count has
704 void kfree_skb(struct sk_buff *skb)
709 trace_kfree_skb(skb, __builtin_return_address(0));
712 EXPORT_SYMBOL(kfree_skb);
714 void kfree_skb_list(struct sk_buff *segs)
717 struct sk_buff *next = segs->next;
723 EXPORT_SYMBOL(kfree_skb_list);
725 /* Dump skb information and contents.
727 * Must only be called from net_ratelimit()-ed paths.
729 * Dumps whole packets if full_pkt, only headers otherwise.
731 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
733 struct skb_shared_info *sh = skb_shinfo(skb);
734 struct net_device *dev = skb->dev;
735 struct sock *sk = skb->sk;
736 struct sk_buff *list_skb;
737 bool has_mac, has_trans;
738 int headroom, tailroom;
744 len = min_t(int, skb->len, MAX_HEADER + 128);
746 headroom = skb_headroom(skb);
747 tailroom = skb_tailroom(skb);
749 has_mac = skb_mac_header_was_set(skb);
750 has_trans = skb_transport_header_was_set(skb);
752 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
753 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
754 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
755 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
756 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
757 level, skb->len, headroom, skb_headlen(skb), tailroom,
758 has_mac ? skb->mac_header : -1,
759 has_mac ? skb_mac_header_len(skb) : -1,
761 has_trans ? skb_network_header_len(skb) : -1,
762 has_trans ? skb->transport_header : -1,
763 sh->tx_flags, sh->nr_frags,
764 sh->gso_size, sh->gso_type, sh->gso_segs,
765 skb->csum, skb->ip_summed, skb->csum_complete_sw,
766 skb->csum_valid, skb->csum_level,
767 skb->hash, skb->sw_hash, skb->l4_hash,
768 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
771 printk("%sdev name=%s feat=0x%pNF\n",
772 level, dev->name, &dev->features);
774 printk("%ssk family=%hu type=%u proto=%u\n",
775 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
777 if (full_pkt && headroom)
778 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb->head, headroom, false);
781 seg_len = min_t(int, skb_headlen(skb), len);
783 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
784 16, 1, skb->data, seg_len, false);
787 if (full_pkt && tailroom)
788 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
789 16, 1, skb_tail_pointer(skb), tailroom, false);
791 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
792 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
793 u32 p_off, p_len, copied;
797 skb_frag_foreach_page(frag, skb_frag_off(frag),
798 skb_frag_size(frag), p, p_off, p_len,
800 seg_len = min_t(int, p_len, len);
801 vaddr = kmap_atomic(p);
802 print_hex_dump(level, "skb frag: ",
804 16, 1, vaddr + p_off, seg_len, false);
805 kunmap_atomic(vaddr);
812 if (full_pkt && skb_has_frag_list(skb)) {
813 printk("skb fraglist:\n");
814 skb_walk_frags(skb, list_skb)
815 skb_dump(level, list_skb, true);
818 EXPORT_SYMBOL(skb_dump);
821 * skb_tx_error - report an sk_buff xmit error
822 * @skb: buffer that triggered an error
824 * Report xmit error if a device callback is tracking this skb.
825 * skb must be freed afterwards.
827 void skb_tx_error(struct sk_buff *skb)
829 skb_zcopy_clear(skb, true);
831 EXPORT_SYMBOL(skb_tx_error);
833 #ifdef CONFIG_TRACEPOINTS
835 * consume_skb - free an skbuff
836 * @skb: buffer to free
838 * Drop a ref to the buffer and free it if the usage count has hit zero
839 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
840 * is being dropped after a failure and notes that
842 void consume_skb(struct sk_buff *skb)
847 trace_consume_skb(skb);
850 EXPORT_SYMBOL(consume_skb);
854 * __consume_stateless_skb - free an skbuff, assuming it is stateless
855 * @skb: buffer to free
857 * Alike consume_skb(), but this variant assumes that this is the last
858 * skb reference and all the head states have been already dropped
860 void __consume_stateless_skb(struct sk_buff *skb)
862 trace_consume_skb(skb);
863 skb_release_data(skb);
867 void __kfree_skb_flush(void)
869 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
871 /* flush skb_cache if containing objects */
873 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
879 static inline void _kfree_skb_defer(struct sk_buff *skb)
881 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
883 /* drop skb->head and call any destructors for packet */
884 skb_release_all(skb);
886 /* record skb to CPU local list */
887 nc->skb_cache[nc->skb_count++] = skb;
890 /* SLUB writes into objects when freeing */
894 /* flush skb_cache if it is filled */
895 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
896 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
901 void __kfree_skb_defer(struct sk_buff *skb)
903 _kfree_skb_defer(skb);
906 void napi_consume_skb(struct sk_buff *skb, int budget)
908 /* Zero budget indicate non-NAPI context called us, like netpoll */
909 if (unlikely(!budget)) {
910 dev_consume_skb_any(skb);
914 lockdep_assert_in_softirq();
919 /* if reaching here SKB is ready to free */
920 trace_consume_skb(skb);
922 /* if SKB is a clone, don't handle this case */
923 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
928 _kfree_skb_defer(skb);
930 EXPORT_SYMBOL(napi_consume_skb);
932 /* Make sure a field is enclosed inside headers_start/headers_end section */
933 #define CHECK_SKB_FIELD(field) \
934 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
935 offsetof(struct sk_buff, headers_start)); \
936 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
937 offsetof(struct sk_buff, headers_end)); \
939 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
941 new->tstamp = old->tstamp;
942 /* We do not copy old->sk */
944 memcpy(new->cb, old->cb, sizeof(old->cb));
945 skb_dst_copy(new, old);
946 __skb_ext_copy(new, old);
947 __nf_copy(new, old, false);
949 /* Note : this field could be in headers_start/headers_end section
950 * It is not yet because we do not want to have a 16 bit hole
952 new->queue_mapping = old->queue_mapping;
954 memcpy(&new->headers_start, &old->headers_start,
955 offsetof(struct sk_buff, headers_end) -
956 offsetof(struct sk_buff, headers_start));
957 CHECK_SKB_FIELD(protocol);
958 CHECK_SKB_FIELD(csum);
959 CHECK_SKB_FIELD(hash);
960 CHECK_SKB_FIELD(priority);
961 CHECK_SKB_FIELD(skb_iif);
962 CHECK_SKB_FIELD(vlan_proto);
963 CHECK_SKB_FIELD(vlan_tci);
964 CHECK_SKB_FIELD(transport_header);
965 CHECK_SKB_FIELD(network_header);
966 CHECK_SKB_FIELD(mac_header);
967 CHECK_SKB_FIELD(inner_protocol);
968 CHECK_SKB_FIELD(inner_transport_header);
969 CHECK_SKB_FIELD(inner_network_header);
970 CHECK_SKB_FIELD(inner_mac_header);
971 CHECK_SKB_FIELD(mark);
972 #ifdef CONFIG_NETWORK_SECMARK
973 CHECK_SKB_FIELD(secmark);
975 #ifdef CONFIG_NET_RX_BUSY_POLL
976 CHECK_SKB_FIELD(napi_id);
979 CHECK_SKB_FIELD(sender_cpu);
981 #ifdef CONFIG_NET_SCHED
982 CHECK_SKB_FIELD(tc_index);
988 * You should not add any new code to this function. Add it to
989 * __copy_skb_header above instead.
991 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
993 #define C(x) n->x = skb->x
995 n->next = n->prev = NULL;
997 __copy_skb_header(n, skb);
1002 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1007 n->destructor = NULL;
1014 refcount_set(&n->users, 1);
1016 atomic_inc(&(skb_shinfo(skb)->dataref));
1024 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1025 * @first: first sk_buff of the msg
1027 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1031 n = alloc_skb(0, GFP_ATOMIC);
1035 n->len = first->len;
1036 n->data_len = first->len;
1037 n->truesize = first->truesize;
1039 skb_shinfo(n)->frag_list = first;
1041 __copy_skb_header(n, first);
1042 n->destructor = NULL;
1046 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1049 * skb_morph - morph one skb into another
1050 * @dst: the skb to receive the contents
1051 * @src: the skb to supply the contents
1053 * This is identical to skb_clone except that the target skb is
1054 * supplied by the user.
1056 * The target skb is returned upon exit.
1058 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1060 skb_release_all(dst);
1061 return __skb_clone(dst, src);
1063 EXPORT_SYMBOL_GPL(skb_morph);
1065 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1067 unsigned long max_pg, num_pg, new_pg, old_pg;
1068 struct user_struct *user;
1070 if (capable(CAP_IPC_LOCK) || !size)
1073 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1074 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1075 user = mmp->user ? : current_user();
1078 old_pg = atomic_long_read(&user->locked_vm);
1079 new_pg = old_pg + num_pg;
1080 if (new_pg > max_pg)
1082 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1086 mmp->user = get_uid(user);
1087 mmp->num_pg = num_pg;
1089 mmp->num_pg += num_pg;
1094 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1096 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1099 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1100 free_uid(mmp->user);
1103 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1105 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1107 struct ubuf_info *uarg;
1108 struct sk_buff *skb;
1110 WARN_ON_ONCE(!in_task());
1112 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1116 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1117 uarg = (void *)skb->cb;
1118 uarg->mmp.user = NULL;
1120 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1125 uarg->callback = sock_zerocopy_callback;
1126 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1128 uarg->bytelen = size;
1130 refcount_set(&uarg->refcnt, 1);
1135 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1137 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1139 return container_of((void *)uarg, struct sk_buff, cb);
1142 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1143 struct ubuf_info *uarg)
1146 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1149 /* realloc only when socket is locked (TCP, UDP cork),
1150 * so uarg->len and sk_zckey access is serialized
1152 if (!sock_owned_by_user(sk)) {
1157 bytelen = uarg->bytelen + size;
1158 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1159 /* TCP can create new skb to attach new uarg */
1160 if (sk->sk_type == SOCK_STREAM)
1165 next = (u32)atomic_read(&sk->sk_zckey);
1166 if ((u32)(uarg->id + uarg->len) == next) {
1167 if (mm_account_pinned_pages(&uarg->mmp, size))
1170 uarg->bytelen = bytelen;
1171 atomic_set(&sk->sk_zckey, ++next);
1173 /* no extra ref when appending to datagram (MSG_MORE) */
1174 if (sk->sk_type == SOCK_STREAM)
1175 sock_zerocopy_get(uarg);
1182 return sock_zerocopy_alloc(sk, size);
1184 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1186 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1188 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1192 old_lo = serr->ee.ee_info;
1193 old_hi = serr->ee.ee_data;
1194 sum_len = old_hi - old_lo + 1ULL + len;
1196 if (sum_len >= (1ULL << 32))
1199 if (lo != old_hi + 1)
1202 serr->ee.ee_data += len;
1206 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1208 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1209 struct sock_exterr_skb *serr;
1210 struct sock *sk = skb->sk;
1211 struct sk_buff_head *q;
1212 unsigned long flags;
1216 mm_unaccount_pinned_pages(&uarg->mmp);
1218 /* if !len, there was only 1 call, and it was aborted
1219 * so do not queue a completion notification
1221 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1226 hi = uarg->id + len - 1;
1228 serr = SKB_EXT_ERR(skb);
1229 memset(serr, 0, sizeof(*serr));
1230 serr->ee.ee_errno = 0;
1231 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1232 serr->ee.ee_data = hi;
1233 serr->ee.ee_info = lo;
1235 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1237 q = &sk->sk_error_queue;
1238 spin_lock_irqsave(&q->lock, flags);
1239 tail = skb_peek_tail(q);
1240 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1241 !skb_zerocopy_notify_extend(tail, lo, len)) {
1242 __skb_queue_tail(q, skb);
1245 spin_unlock_irqrestore(&q->lock, flags);
1247 sk->sk_error_report(sk);
1253 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1255 void sock_zerocopy_put(struct ubuf_info *uarg)
1257 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1259 uarg->callback(uarg, uarg->zerocopy);
1261 consume_skb(skb_from_uarg(uarg));
1264 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1266 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1269 struct sock *sk = skb_from_uarg(uarg)->sk;
1271 atomic_dec(&sk->sk_zckey);
1275 sock_zerocopy_put(uarg);
1278 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1280 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1282 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1284 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1286 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1287 struct msghdr *msg, int len,
1288 struct ubuf_info *uarg)
1290 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1291 struct iov_iter orig_iter = msg->msg_iter;
1292 int err, orig_len = skb->len;
1294 /* An skb can only point to one uarg. This edge case happens when
1295 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1297 if (orig_uarg && uarg != orig_uarg)
1300 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1301 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1302 struct sock *save_sk = skb->sk;
1304 /* Streams do not free skb on error. Reset to prev state. */
1305 msg->msg_iter = orig_iter;
1307 ___pskb_trim(skb, orig_len);
1312 skb_zcopy_set(skb, uarg, NULL);
1313 return skb->len - orig_len;
1315 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1317 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1320 if (skb_zcopy(orig)) {
1321 if (skb_zcopy(nskb)) {
1322 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1327 if (skb_uarg(nskb) == skb_uarg(orig))
1329 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1332 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1338 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1339 * @skb: the skb to modify
1340 * @gfp_mask: allocation priority
1342 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1343 * It will copy all frags into kernel and drop the reference
1344 * to userspace pages.
1346 * If this function is called from an interrupt gfp_mask() must be
1349 * Returns 0 on success or a negative error code on failure
1350 * to allocate kernel memory to copy to.
1352 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1354 int num_frags = skb_shinfo(skb)->nr_frags;
1355 struct page *page, *head = NULL;
1359 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1365 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1366 for (i = 0; i < new_frags; i++) {
1367 page = alloc_page(gfp_mask);
1370 struct page *next = (struct page *)page_private(head);
1376 set_page_private(page, (unsigned long)head);
1382 for (i = 0; i < num_frags; i++) {
1383 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1384 u32 p_off, p_len, copied;
1388 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1389 p, p_off, p_len, copied) {
1391 vaddr = kmap_atomic(p);
1393 while (done < p_len) {
1394 if (d_off == PAGE_SIZE) {
1396 page = (struct page *)page_private(page);
1398 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1399 memcpy(page_address(page) + d_off,
1400 vaddr + p_off + done, copy);
1404 kunmap_atomic(vaddr);
1408 /* skb frags release userspace buffers */
1409 for (i = 0; i < num_frags; i++)
1410 skb_frag_unref(skb, i);
1412 /* skb frags point to kernel buffers */
1413 for (i = 0; i < new_frags - 1; i++) {
1414 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1415 head = (struct page *)page_private(head);
1417 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1418 skb_shinfo(skb)->nr_frags = new_frags;
1421 skb_zcopy_clear(skb, false);
1424 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1427 * skb_clone - duplicate an sk_buff
1428 * @skb: buffer to clone
1429 * @gfp_mask: allocation priority
1431 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1432 * copies share the same packet data but not structure. The new
1433 * buffer has a reference count of 1. If the allocation fails the
1434 * function returns %NULL otherwise the new buffer is returned.
1436 * If this function is called from an interrupt gfp_mask() must be
1440 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1442 struct sk_buff_fclones *fclones = container_of(skb,
1443 struct sk_buff_fclones,
1447 if (skb_orphan_frags(skb, gfp_mask))
1450 if (skb->fclone == SKB_FCLONE_ORIG &&
1451 refcount_read(&fclones->fclone_ref) == 1) {
1453 refcount_set(&fclones->fclone_ref, 2);
1455 if (skb_pfmemalloc(skb))
1456 gfp_mask |= __GFP_MEMALLOC;
1458 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1462 n->fclone = SKB_FCLONE_UNAVAILABLE;
1465 return __skb_clone(n, skb);
1467 EXPORT_SYMBOL(skb_clone);
1469 void skb_headers_offset_update(struct sk_buff *skb, int off)
1471 /* Only adjust this if it actually is csum_start rather than csum */
1472 if (skb->ip_summed == CHECKSUM_PARTIAL)
1473 skb->csum_start += off;
1474 /* {transport,network,mac}_header and tail are relative to skb->head */
1475 skb->transport_header += off;
1476 skb->network_header += off;
1477 if (skb_mac_header_was_set(skb))
1478 skb->mac_header += off;
1479 skb->inner_transport_header += off;
1480 skb->inner_network_header += off;
1481 skb->inner_mac_header += off;
1483 EXPORT_SYMBOL(skb_headers_offset_update);
1485 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1487 __copy_skb_header(new, old);
1489 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1490 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1491 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1493 EXPORT_SYMBOL(skb_copy_header);
1495 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1497 if (skb_pfmemalloc(skb))
1498 return SKB_ALLOC_RX;
1503 * skb_copy - create private copy of an sk_buff
1504 * @skb: buffer to copy
1505 * @gfp_mask: allocation priority
1507 * Make a copy of both an &sk_buff and its data. This is used when the
1508 * caller wishes to modify the data and needs a private copy of the
1509 * data to alter. Returns %NULL on failure or the pointer to the buffer
1510 * on success. The returned buffer has a reference count of 1.
1512 * As by-product this function converts non-linear &sk_buff to linear
1513 * one, so that &sk_buff becomes completely private and caller is allowed
1514 * to modify all the data of returned buffer. This means that this
1515 * function is not recommended for use in circumstances when only
1516 * header is going to be modified. Use pskb_copy() instead.
1519 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1521 int headerlen = skb_headroom(skb);
1522 unsigned int size = skb_end_offset(skb) + skb->data_len;
1523 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1524 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1529 /* Set the data pointer */
1530 skb_reserve(n, headerlen);
1531 /* Set the tail pointer and length */
1532 skb_put(n, skb->len);
1534 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1536 skb_copy_header(n, skb);
1539 EXPORT_SYMBOL(skb_copy);
1542 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1543 * @skb: buffer to copy
1544 * @headroom: headroom of new skb
1545 * @gfp_mask: allocation priority
1546 * @fclone: if true allocate the copy of the skb from the fclone
1547 * cache instead of the head cache; it is recommended to set this
1548 * to true for the cases where the copy will likely be cloned
1550 * Make a copy of both an &sk_buff and part of its data, located
1551 * in header. Fragmented data remain shared. This is used when
1552 * the caller wishes to modify only header of &sk_buff and needs
1553 * private copy of the header to alter. Returns %NULL on failure
1554 * or the pointer to the buffer on success.
1555 * The returned buffer has a reference count of 1.
1558 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1559 gfp_t gfp_mask, bool fclone)
1561 unsigned int size = skb_headlen(skb) + headroom;
1562 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1563 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1568 /* Set the data pointer */
1569 skb_reserve(n, headroom);
1570 /* Set the tail pointer and length */
1571 skb_put(n, skb_headlen(skb));
1572 /* Copy the bytes */
1573 skb_copy_from_linear_data(skb, n->data, n->len);
1575 n->truesize += skb->data_len;
1576 n->data_len = skb->data_len;
1579 if (skb_shinfo(skb)->nr_frags) {
1582 if (skb_orphan_frags(skb, gfp_mask) ||
1583 skb_zerocopy_clone(n, skb, gfp_mask)) {
1588 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1589 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1590 skb_frag_ref(skb, i);
1592 skb_shinfo(n)->nr_frags = i;
1595 if (skb_has_frag_list(skb)) {
1596 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1597 skb_clone_fraglist(n);
1600 skb_copy_header(n, skb);
1604 EXPORT_SYMBOL(__pskb_copy_fclone);
1607 * pskb_expand_head - reallocate header of &sk_buff
1608 * @skb: buffer to reallocate
1609 * @nhead: room to add at head
1610 * @ntail: room to add at tail
1611 * @gfp_mask: allocation priority
1613 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1614 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1615 * reference count of 1. Returns zero in the case of success or error,
1616 * if expansion failed. In the last case, &sk_buff is not changed.
1618 * All the pointers pointing into skb header may change and must be
1619 * reloaded after call to this function.
1622 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1625 int i, osize = skb_end_offset(skb);
1626 int size = osize + nhead + ntail;
1632 BUG_ON(skb_shared(skb));
1634 size = SKB_DATA_ALIGN(size);
1636 if (skb_pfmemalloc(skb))
1637 gfp_mask |= __GFP_MEMALLOC;
1638 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1639 gfp_mask, NUMA_NO_NODE, NULL);
1642 size = SKB_WITH_OVERHEAD(ksize(data));
1644 /* Copy only real data... and, alas, header. This should be
1645 * optimized for the cases when header is void.
1647 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1649 memcpy((struct skb_shared_info *)(data + size),
1651 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1654 * if shinfo is shared we must drop the old head gracefully, but if it
1655 * is not we can just drop the old head and let the existing refcount
1656 * be since all we did is relocate the values
1658 if (skb_cloned(skb)) {
1659 if (skb_orphan_frags(skb, gfp_mask))
1662 refcount_inc(&skb_uarg(skb)->refcnt);
1663 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1664 skb_frag_ref(skb, i);
1666 if (skb_has_frag_list(skb))
1667 skb_clone_fraglist(skb);
1669 skb_release_data(skb);
1673 off = (data + nhead) - skb->head;
1678 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1682 skb->end = skb->head + size;
1685 skb_headers_offset_update(skb, nhead);
1689 atomic_set(&skb_shinfo(skb)->dataref, 1);
1691 skb_metadata_clear(skb);
1693 /* It is not generally safe to change skb->truesize.
1694 * For the moment, we really care of rx path, or
1695 * when skb is orphaned (not attached to a socket).
1697 if (!skb->sk || skb->destructor == sock_edemux)
1698 skb->truesize += size - osize;
1707 EXPORT_SYMBOL(pskb_expand_head);
1709 /* Make private copy of skb with writable head and some headroom */
1711 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1713 struct sk_buff *skb2;
1714 int delta = headroom - skb_headroom(skb);
1717 skb2 = pskb_copy(skb, GFP_ATOMIC);
1719 skb2 = skb_clone(skb, GFP_ATOMIC);
1720 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1728 EXPORT_SYMBOL(skb_realloc_headroom);
1731 * skb_copy_expand - copy and expand sk_buff
1732 * @skb: buffer to copy
1733 * @newheadroom: new free bytes at head
1734 * @newtailroom: new free bytes at tail
1735 * @gfp_mask: allocation priority
1737 * Make a copy of both an &sk_buff and its data and while doing so
1738 * allocate additional space.
1740 * This is used when the caller wishes to modify the data and needs a
1741 * private copy of the data to alter as well as more space for new fields.
1742 * Returns %NULL on failure or the pointer to the buffer
1743 * on success. The returned buffer has a reference count of 1.
1745 * You must pass %GFP_ATOMIC as the allocation priority if this function
1746 * is called from an interrupt.
1748 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1749 int newheadroom, int newtailroom,
1753 * Allocate the copy buffer
1755 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1756 gfp_mask, skb_alloc_rx_flag(skb),
1758 int oldheadroom = skb_headroom(skb);
1759 int head_copy_len, head_copy_off;
1764 skb_reserve(n, newheadroom);
1766 /* Set the tail pointer and length */
1767 skb_put(n, skb->len);
1769 head_copy_len = oldheadroom;
1771 if (newheadroom <= head_copy_len)
1772 head_copy_len = newheadroom;
1774 head_copy_off = newheadroom - head_copy_len;
1776 /* Copy the linear header and data. */
1777 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1778 skb->len + head_copy_len));
1780 skb_copy_header(n, skb);
1782 skb_headers_offset_update(n, newheadroom - oldheadroom);
1786 EXPORT_SYMBOL(skb_copy_expand);
1789 * __skb_pad - zero pad the tail of an skb
1790 * @skb: buffer to pad
1791 * @pad: space to pad
1792 * @free_on_error: free buffer on error
1794 * Ensure that a buffer is followed by a padding area that is zero
1795 * filled. Used by network drivers which may DMA or transfer data
1796 * beyond the buffer end onto the wire.
1798 * May return error in out of memory cases. The skb is freed on error
1799 * if @free_on_error is true.
1802 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1807 /* If the skbuff is non linear tailroom is always zero.. */
1808 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1809 memset(skb->data+skb->len, 0, pad);
1813 ntail = skb->data_len + pad - (skb->end - skb->tail);
1814 if (likely(skb_cloned(skb) || ntail > 0)) {
1815 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1820 /* FIXME: The use of this function with non-linear skb's really needs
1823 err = skb_linearize(skb);
1827 memset(skb->data + skb->len, 0, pad);
1835 EXPORT_SYMBOL(__skb_pad);
1838 * pskb_put - add data to the tail of a potentially fragmented buffer
1839 * @skb: start of the buffer to use
1840 * @tail: tail fragment of the buffer to use
1841 * @len: amount of data to add
1843 * This function extends the used data area of the potentially
1844 * fragmented buffer. @tail must be the last fragment of @skb -- or
1845 * @skb itself. If this would exceed the total buffer size the kernel
1846 * will panic. A pointer to the first byte of the extra data is
1850 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1853 skb->data_len += len;
1856 return skb_put(tail, len);
1858 EXPORT_SYMBOL_GPL(pskb_put);
1861 * skb_put - add data to a buffer
1862 * @skb: buffer to use
1863 * @len: amount of data to add
1865 * This function extends the used data area of the buffer. If this would
1866 * exceed the total buffer size the kernel will panic. A pointer to the
1867 * first byte of the extra data is returned.
1869 void *skb_put(struct sk_buff *skb, unsigned int len)
1871 void *tmp = skb_tail_pointer(skb);
1872 SKB_LINEAR_ASSERT(skb);
1875 if (unlikely(skb->tail > skb->end))
1876 skb_over_panic(skb, len, __builtin_return_address(0));
1879 EXPORT_SYMBOL(skb_put);
1882 * skb_push - add data to the start of a buffer
1883 * @skb: buffer to use
1884 * @len: amount of data to add
1886 * This function extends the used data area of the buffer at the buffer
1887 * start. If this would exceed the total buffer headroom the kernel will
1888 * panic. A pointer to the first byte of the extra data is returned.
1890 void *skb_push(struct sk_buff *skb, unsigned int len)
1894 if (unlikely(skb->data < skb->head))
1895 skb_under_panic(skb, len, __builtin_return_address(0));
1898 EXPORT_SYMBOL(skb_push);
1901 * skb_pull - remove data from the start of a buffer
1902 * @skb: buffer to use
1903 * @len: amount of data to remove
1905 * This function removes data from the start of a buffer, returning
1906 * the memory to the headroom. A pointer to the next data in the buffer
1907 * is returned. Once the data has been pulled future pushes will overwrite
1910 void *skb_pull(struct sk_buff *skb, unsigned int len)
1912 return skb_pull_inline(skb, len);
1914 EXPORT_SYMBOL(skb_pull);
1917 * skb_trim - remove end from a buffer
1918 * @skb: buffer to alter
1921 * Cut the length of a buffer down by removing data from the tail. If
1922 * the buffer is already under the length specified it is not modified.
1923 * The skb must be linear.
1925 void skb_trim(struct sk_buff *skb, unsigned int len)
1928 __skb_trim(skb, len);
1930 EXPORT_SYMBOL(skb_trim);
1932 /* Trims skb to length len. It can change skb pointers.
1935 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1937 struct sk_buff **fragp;
1938 struct sk_buff *frag;
1939 int offset = skb_headlen(skb);
1940 int nfrags = skb_shinfo(skb)->nr_frags;
1944 if (skb_cloned(skb) &&
1945 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1952 for (; i < nfrags; i++) {
1953 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1960 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1963 skb_shinfo(skb)->nr_frags = i;
1965 for (; i < nfrags; i++)
1966 skb_frag_unref(skb, i);
1968 if (skb_has_frag_list(skb))
1969 skb_drop_fraglist(skb);
1973 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1974 fragp = &frag->next) {
1975 int end = offset + frag->len;
1977 if (skb_shared(frag)) {
1978 struct sk_buff *nfrag;
1980 nfrag = skb_clone(frag, GFP_ATOMIC);
1981 if (unlikely(!nfrag))
1984 nfrag->next = frag->next;
1996 unlikely((err = pskb_trim(frag, len - offset))))
2000 skb_drop_list(&frag->next);
2005 if (len > skb_headlen(skb)) {
2006 skb->data_len -= skb->len - len;
2011 skb_set_tail_pointer(skb, len);
2014 if (!skb->sk || skb->destructor == sock_edemux)
2018 EXPORT_SYMBOL(___pskb_trim);
2020 /* Note : use pskb_trim_rcsum() instead of calling this directly
2022 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2024 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2025 int delta = skb->len - len;
2027 skb->csum = csum_block_sub(skb->csum,
2028 skb_checksum(skb, len, delta, 0),
2030 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2031 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2032 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2034 if (offset + sizeof(__sum16) > hdlen)
2037 return __pskb_trim(skb, len);
2039 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2042 * __pskb_pull_tail - advance tail of skb header
2043 * @skb: buffer to reallocate
2044 * @delta: number of bytes to advance tail
2046 * The function makes a sense only on a fragmented &sk_buff,
2047 * it expands header moving its tail forward and copying necessary
2048 * data from fragmented part.
2050 * &sk_buff MUST have reference count of 1.
2052 * Returns %NULL (and &sk_buff does not change) if pull failed
2053 * or value of new tail of skb in the case of success.
2055 * All the pointers pointing into skb header may change and must be
2056 * reloaded after call to this function.
2059 /* Moves tail of skb head forward, copying data from fragmented part,
2060 * when it is necessary.
2061 * 1. It may fail due to malloc failure.
2062 * 2. It may change skb pointers.
2064 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2066 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2068 /* If skb has not enough free space at tail, get new one
2069 * plus 128 bytes for future expansions. If we have enough
2070 * room at tail, reallocate without expansion only if skb is cloned.
2072 int i, k, eat = (skb->tail + delta) - skb->end;
2074 if (eat > 0 || skb_cloned(skb)) {
2075 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2080 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2081 skb_tail_pointer(skb), delta));
2083 /* Optimization: no fragments, no reasons to preestimate
2084 * size of pulled pages. Superb.
2086 if (!skb_has_frag_list(skb))
2089 /* Estimate size of pulled pages. */
2091 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2092 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2099 /* If we need update frag list, we are in troubles.
2100 * Certainly, it is possible to add an offset to skb data,
2101 * but taking into account that pulling is expected to
2102 * be very rare operation, it is worth to fight against
2103 * further bloating skb head and crucify ourselves here instead.
2104 * Pure masohism, indeed. 8)8)
2107 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2108 struct sk_buff *clone = NULL;
2109 struct sk_buff *insp = NULL;
2112 if (list->len <= eat) {
2113 /* Eaten as whole. */
2118 /* Eaten partially. */
2120 if (skb_shared(list)) {
2121 /* Sucks! We need to fork list. :-( */
2122 clone = skb_clone(list, GFP_ATOMIC);
2128 /* This may be pulled without
2132 if (!pskb_pull(list, eat)) {
2140 /* Free pulled out fragments. */
2141 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2142 skb_shinfo(skb)->frag_list = list->next;
2145 /* And insert new clone at head. */
2148 skb_shinfo(skb)->frag_list = clone;
2151 /* Success! Now we may commit changes to skb data. */
2156 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2157 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2160 skb_frag_unref(skb, i);
2163 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2165 *frag = skb_shinfo(skb)->frags[i];
2167 skb_frag_off_add(frag, eat);
2168 skb_frag_size_sub(frag, eat);
2176 skb_shinfo(skb)->nr_frags = k;
2180 skb->data_len -= delta;
2183 skb_zcopy_clear(skb, false);
2185 return skb_tail_pointer(skb);
2187 EXPORT_SYMBOL(__pskb_pull_tail);
2190 * skb_copy_bits - copy bits from skb to kernel buffer
2192 * @offset: offset in source
2193 * @to: destination buffer
2194 * @len: number of bytes to copy
2196 * Copy the specified number of bytes from the source skb to the
2197 * destination buffer.
2200 * If its prototype is ever changed,
2201 * check arch/{*}/net/{*}.S files,
2202 * since it is called from BPF assembly code.
2204 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2206 int start = skb_headlen(skb);
2207 struct sk_buff *frag_iter;
2210 if (offset > (int)skb->len - len)
2214 if ((copy = start - offset) > 0) {
2217 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2218 if ((len -= copy) == 0)
2224 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2226 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2228 WARN_ON(start > offset + len);
2230 end = start + skb_frag_size(f);
2231 if ((copy = end - offset) > 0) {
2232 u32 p_off, p_len, copied;
2239 skb_frag_foreach_page(f,
2240 skb_frag_off(f) + offset - start,
2241 copy, p, p_off, p_len, copied) {
2242 vaddr = kmap_atomic(p);
2243 memcpy(to + copied, vaddr + p_off, p_len);
2244 kunmap_atomic(vaddr);
2247 if ((len -= copy) == 0)
2255 skb_walk_frags(skb, frag_iter) {
2258 WARN_ON(start > offset + len);
2260 end = start + frag_iter->len;
2261 if ((copy = end - offset) > 0) {
2264 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2266 if ((len -= copy) == 0)
2280 EXPORT_SYMBOL(skb_copy_bits);
2283 * Callback from splice_to_pipe(), if we need to release some pages
2284 * at the end of the spd in case we error'ed out in filling the pipe.
2286 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2288 put_page(spd->pages[i]);
2291 static struct page *linear_to_page(struct page *page, unsigned int *len,
2292 unsigned int *offset,
2295 struct page_frag *pfrag = sk_page_frag(sk);
2297 if (!sk_page_frag_refill(sk, pfrag))
2300 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2302 memcpy(page_address(pfrag->page) + pfrag->offset,
2303 page_address(page) + *offset, *len);
2304 *offset = pfrag->offset;
2305 pfrag->offset += *len;
2310 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2312 unsigned int offset)
2314 return spd->nr_pages &&
2315 spd->pages[spd->nr_pages - 1] == page &&
2316 (spd->partial[spd->nr_pages - 1].offset +
2317 spd->partial[spd->nr_pages - 1].len == offset);
2321 * Fill page/offset/length into spd, if it can hold more pages.
2323 static bool spd_fill_page(struct splice_pipe_desc *spd,
2324 struct pipe_inode_info *pipe, struct page *page,
2325 unsigned int *len, unsigned int offset,
2329 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2333 page = linear_to_page(page, len, &offset, sk);
2337 if (spd_can_coalesce(spd, page, offset)) {
2338 spd->partial[spd->nr_pages - 1].len += *len;
2342 spd->pages[spd->nr_pages] = page;
2343 spd->partial[spd->nr_pages].len = *len;
2344 spd->partial[spd->nr_pages].offset = offset;
2350 static bool __splice_segment(struct page *page, unsigned int poff,
2351 unsigned int plen, unsigned int *off,
2353 struct splice_pipe_desc *spd, bool linear,
2355 struct pipe_inode_info *pipe)
2360 /* skip this segment if already processed */
2366 /* ignore any bits we already processed */
2372 unsigned int flen = min(*len, plen);
2374 if (spd_fill_page(spd, pipe, page, &flen, poff,
2380 } while (*len && plen);
2386 * Map linear and fragment data from the skb to spd. It reports true if the
2387 * pipe is full or if we already spliced the requested length.
2389 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2390 unsigned int *offset, unsigned int *len,
2391 struct splice_pipe_desc *spd, struct sock *sk)
2394 struct sk_buff *iter;
2396 /* map the linear part :
2397 * If skb->head_frag is set, this 'linear' part is backed by a
2398 * fragment, and if the head is not shared with any clones then
2399 * we can avoid a copy since we own the head portion of this page.
2401 if (__splice_segment(virt_to_page(skb->data),
2402 (unsigned long) skb->data & (PAGE_SIZE - 1),
2405 skb_head_is_locked(skb),
2410 * then map the fragments
2412 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2413 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2415 if (__splice_segment(skb_frag_page(f),
2416 skb_frag_off(f), skb_frag_size(f),
2417 offset, len, spd, false, sk, pipe))
2421 skb_walk_frags(skb, iter) {
2422 if (*offset >= iter->len) {
2423 *offset -= iter->len;
2426 /* __skb_splice_bits() only fails if the output has no room
2427 * left, so no point in going over the frag_list for the error
2430 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2438 * Map data from the skb to a pipe. Should handle both the linear part,
2439 * the fragments, and the frag list.
2441 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2442 struct pipe_inode_info *pipe, unsigned int tlen,
2445 struct partial_page partial[MAX_SKB_FRAGS];
2446 struct page *pages[MAX_SKB_FRAGS];
2447 struct splice_pipe_desc spd = {
2450 .nr_pages_max = MAX_SKB_FRAGS,
2451 .ops = &nosteal_pipe_buf_ops,
2452 .spd_release = sock_spd_release,
2456 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2459 ret = splice_to_pipe(pipe, &spd);
2463 EXPORT_SYMBOL_GPL(skb_splice_bits);
2465 /* Send skb data on a socket. Socket must be locked. */
2466 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2469 unsigned int orig_len = len;
2470 struct sk_buff *head = skb;
2471 unsigned short fragidx;
2476 /* Deal with head data */
2477 while (offset < skb_headlen(skb) && len) {
2481 slen = min_t(int, len, skb_headlen(skb) - offset);
2482 kv.iov_base = skb->data + offset;
2484 memset(&msg, 0, sizeof(msg));
2485 msg.msg_flags = MSG_DONTWAIT;
2487 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2495 /* All the data was skb head? */
2499 /* Make offset relative to start of frags */
2500 offset -= skb_headlen(skb);
2502 /* Find where we are in frag list */
2503 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2504 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2506 if (offset < skb_frag_size(frag))
2509 offset -= skb_frag_size(frag);
2512 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2513 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2515 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2518 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2519 skb_frag_off(frag) + offset,
2520 slen, MSG_DONTWAIT);
2533 /* Process any frag lists */
2536 if (skb_has_frag_list(skb)) {
2537 skb = skb_shinfo(skb)->frag_list;
2540 } else if (skb->next) {
2547 return orig_len - len;
2550 return orig_len == len ? ret : orig_len - len;
2552 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2555 * skb_store_bits - store bits from kernel buffer to skb
2556 * @skb: destination buffer
2557 * @offset: offset in destination
2558 * @from: source buffer
2559 * @len: number of bytes to copy
2561 * Copy the specified number of bytes from the source buffer to the
2562 * destination skb. This function handles all the messy bits of
2563 * traversing fragment lists and such.
2566 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2568 int start = skb_headlen(skb);
2569 struct sk_buff *frag_iter;
2572 if (offset > (int)skb->len - len)
2575 if ((copy = start - offset) > 0) {
2578 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2579 if ((len -= copy) == 0)
2585 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2586 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2589 WARN_ON(start > offset + len);
2591 end = start + skb_frag_size(frag);
2592 if ((copy = end - offset) > 0) {
2593 u32 p_off, p_len, copied;
2600 skb_frag_foreach_page(frag,
2601 skb_frag_off(frag) + offset - start,
2602 copy, p, p_off, p_len, copied) {
2603 vaddr = kmap_atomic(p);
2604 memcpy(vaddr + p_off, from + copied, p_len);
2605 kunmap_atomic(vaddr);
2608 if ((len -= copy) == 0)
2616 skb_walk_frags(skb, frag_iter) {
2619 WARN_ON(start > offset + len);
2621 end = start + frag_iter->len;
2622 if ((copy = end - offset) > 0) {
2625 if (skb_store_bits(frag_iter, offset - start,
2628 if ((len -= copy) == 0)
2641 EXPORT_SYMBOL(skb_store_bits);
2643 /* Checksum skb data. */
2644 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2645 __wsum csum, const struct skb_checksum_ops *ops)
2647 int start = skb_headlen(skb);
2648 int i, copy = start - offset;
2649 struct sk_buff *frag_iter;
2652 /* Checksum header. */
2656 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2657 skb->data + offset, copy, csum);
2658 if ((len -= copy) == 0)
2664 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2666 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2668 WARN_ON(start > offset + len);
2670 end = start + skb_frag_size(frag);
2671 if ((copy = end - offset) > 0) {
2672 u32 p_off, p_len, copied;
2680 skb_frag_foreach_page(frag,
2681 skb_frag_off(frag) + offset - start,
2682 copy, p, p_off, p_len, copied) {
2683 vaddr = kmap_atomic(p);
2684 csum2 = INDIRECT_CALL_1(ops->update,
2686 vaddr + p_off, p_len, 0);
2687 kunmap_atomic(vaddr);
2688 csum = INDIRECT_CALL_1(ops->combine,
2689 csum_block_add_ext, csum,
2701 skb_walk_frags(skb, frag_iter) {
2704 WARN_ON(start > offset + len);
2706 end = start + frag_iter->len;
2707 if ((copy = end - offset) > 0) {
2711 csum2 = __skb_checksum(frag_iter, offset - start,
2713 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2714 csum, csum2, pos, copy);
2715 if ((len -= copy) == 0)
2726 EXPORT_SYMBOL(__skb_checksum);
2728 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2729 int len, __wsum csum)
2731 const struct skb_checksum_ops ops = {
2732 .update = csum_partial_ext,
2733 .combine = csum_block_add_ext,
2736 return __skb_checksum(skb, offset, len, csum, &ops);
2738 EXPORT_SYMBOL(skb_checksum);
2740 /* Both of above in one bottle. */
2742 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2745 int start = skb_headlen(skb);
2746 int i, copy = start - offset;
2747 struct sk_buff *frag_iter;
2755 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2757 if ((len -= copy) == 0)
2764 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2767 WARN_ON(start > offset + len);
2769 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2770 if ((copy = end - offset) > 0) {
2771 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2772 u32 p_off, p_len, copied;
2780 skb_frag_foreach_page(frag,
2781 skb_frag_off(frag) + offset - start,
2782 copy, p, p_off, p_len, copied) {
2783 vaddr = kmap_atomic(p);
2784 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2787 kunmap_atomic(vaddr);
2788 csum = csum_block_add(csum, csum2, pos);
2800 skb_walk_frags(skb, frag_iter) {
2804 WARN_ON(start > offset + len);
2806 end = start + frag_iter->len;
2807 if ((copy = end - offset) > 0) {
2810 csum2 = skb_copy_and_csum_bits(frag_iter,
2813 csum = csum_block_add(csum, csum2, pos);
2814 if ((len -= copy) == 0)
2825 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2827 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2831 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2832 /* See comments in __skb_checksum_complete(). */
2834 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2835 !skb->csum_complete_sw)
2836 netdev_rx_csum_fault(skb->dev, skb);
2838 if (!skb_shared(skb))
2839 skb->csum_valid = !sum;
2842 EXPORT_SYMBOL(__skb_checksum_complete_head);
2844 /* This function assumes skb->csum already holds pseudo header's checksum,
2845 * which has been changed from the hardware checksum, for example, by
2846 * __skb_checksum_validate_complete(). And, the original skb->csum must
2847 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2849 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2850 * zero. The new checksum is stored back into skb->csum unless the skb is
2853 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2858 csum = skb_checksum(skb, 0, skb->len, 0);
2860 sum = csum_fold(csum_add(skb->csum, csum));
2861 /* This check is inverted, because we already knew the hardware
2862 * checksum is invalid before calling this function. So, if the
2863 * re-computed checksum is valid instead, then we have a mismatch
2864 * between the original skb->csum and skb_checksum(). This means either
2865 * the original hardware checksum is incorrect or we screw up skb->csum
2866 * when moving skb->data around.
2869 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2870 !skb->csum_complete_sw)
2871 netdev_rx_csum_fault(skb->dev, skb);
2874 if (!skb_shared(skb)) {
2875 /* Save full packet checksum */
2877 skb->ip_summed = CHECKSUM_COMPLETE;
2878 skb->csum_complete_sw = 1;
2879 skb->csum_valid = !sum;
2884 EXPORT_SYMBOL(__skb_checksum_complete);
2886 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2888 net_warn_ratelimited(
2889 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2894 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2895 int offset, int len)
2897 net_warn_ratelimited(
2898 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2903 static const struct skb_checksum_ops default_crc32c_ops = {
2904 .update = warn_crc32c_csum_update,
2905 .combine = warn_crc32c_csum_combine,
2908 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2909 &default_crc32c_ops;
2910 EXPORT_SYMBOL(crc32c_csum_stub);
2913 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2914 * @from: source buffer
2916 * Calculates the amount of linear headroom needed in the 'to' skb passed
2917 * into skb_zerocopy().
2920 skb_zerocopy_headlen(const struct sk_buff *from)
2922 unsigned int hlen = 0;
2924 if (!from->head_frag ||
2925 skb_headlen(from) < L1_CACHE_BYTES ||
2926 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2927 hlen = skb_headlen(from);
2929 if (skb_has_frag_list(from))
2934 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2937 * skb_zerocopy - Zero copy skb to skb
2938 * @to: destination buffer
2939 * @from: source buffer
2940 * @len: number of bytes to copy from source buffer
2941 * @hlen: size of linear headroom in destination buffer
2943 * Copies up to `len` bytes from `from` to `to` by creating references
2944 * to the frags in the source buffer.
2946 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2947 * headroom in the `to` buffer.
2950 * 0: everything is OK
2951 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2952 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2955 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2958 int plen = 0; /* length of skb->head fragment */
2961 unsigned int offset;
2963 BUG_ON(!from->head_frag && !hlen);
2965 /* dont bother with small payloads */
2966 if (len <= skb_tailroom(to))
2967 return skb_copy_bits(from, 0, skb_put(to, len), len);
2970 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2975 plen = min_t(int, skb_headlen(from), len);
2977 page = virt_to_head_page(from->head);
2978 offset = from->data - (unsigned char *)page_address(page);
2979 __skb_fill_page_desc(to, 0, page, offset, plen);
2986 to->truesize += len + plen;
2987 to->len += len + plen;
2988 to->data_len += len + plen;
2990 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2994 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2996 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3001 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3002 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3004 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3006 skb_frag_ref(to, j);
3009 skb_shinfo(to)->nr_frags = j;
3013 EXPORT_SYMBOL_GPL(skb_zerocopy);
3015 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3020 if (skb->ip_summed == CHECKSUM_PARTIAL)
3021 csstart = skb_checksum_start_offset(skb);
3023 csstart = skb_headlen(skb);
3025 BUG_ON(csstart > skb_headlen(skb));
3027 skb_copy_from_linear_data(skb, to, csstart);
3030 if (csstart != skb->len)
3031 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3032 skb->len - csstart);
3034 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3035 long csstuff = csstart + skb->csum_offset;
3037 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3040 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3043 * skb_dequeue - remove from the head of the queue
3044 * @list: list to dequeue from
3046 * Remove the head of the list. The list lock is taken so the function
3047 * may be used safely with other locking list functions. The head item is
3048 * returned or %NULL if the list is empty.
3051 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3053 unsigned long flags;
3054 struct sk_buff *result;
3056 spin_lock_irqsave(&list->lock, flags);
3057 result = __skb_dequeue(list);
3058 spin_unlock_irqrestore(&list->lock, flags);
3061 EXPORT_SYMBOL(skb_dequeue);
3064 * skb_dequeue_tail - remove from the tail of the queue
3065 * @list: list to dequeue from
3067 * Remove the tail of the list. The list lock is taken so the function
3068 * may be used safely with other locking list functions. The tail item is
3069 * returned or %NULL if the list is empty.
3071 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3073 unsigned long flags;
3074 struct sk_buff *result;
3076 spin_lock_irqsave(&list->lock, flags);
3077 result = __skb_dequeue_tail(list);
3078 spin_unlock_irqrestore(&list->lock, flags);
3081 EXPORT_SYMBOL(skb_dequeue_tail);
3084 * skb_queue_purge - empty a list
3085 * @list: list to empty
3087 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3088 * the list and one reference dropped. This function takes the list
3089 * lock and is atomic with respect to other list locking functions.
3091 void skb_queue_purge(struct sk_buff_head *list)
3093 struct sk_buff *skb;
3094 while ((skb = skb_dequeue(list)) != NULL)
3097 EXPORT_SYMBOL(skb_queue_purge);
3100 * skb_rbtree_purge - empty a skb rbtree
3101 * @root: root of the rbtree to empty
3102 * Return value: the sum of truesizes of all purged skbs.
3104 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3105 * the list and one reference dropped. This function does not take
3106 * any lock. Synchronization should be handled by the caller (e.g., TCP
3107 * out-of-order queue is protected by the socket lock).
3109 unsigned int skb_rbtree_purge(struct rb_root *root)
3111 struct rb_node *p = rb_first(root);
3112 unsigned int sum = 0;
3115 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3118 rb_erase(&skb->rbnode, root);
3119 sum += skb->truesize;
3126 * skb_queue_head - queue a buffer at the list head
3127 * @list: list to use
3128 * @newsk: buffer to queue
3130 * Queue a buffer at the start of the list. This function takes the
3131 * list lock and can be used safely with other locking &sk_buff functions
3134 * A buffer cannot be placed on two lists at the same time.
3136 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3138 unsigned long flags;
3140 spin_lock_irqsave(&list->lock, flags);
3141 __skb_queue_head(list, newsk);
3142 spin_unlock_irqrestore(&list->lock, flags);
3144 EXPORT_SYMBOL(skb_queue_head);
3147 * skb_queue_tail - queue a buffer at the list tail
3148 * @list: list to use
3149 * @newsk: buffer to queue
3151 * Queue a buffer at the tail of the list. This function takes the
3152 * list lock and can be used safely with other locking &sk_buff functions
3155 * A buffer cannot be placed on two lists at the same time.
3157 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3159 unsigned long flags;
3161 spin_lock_irqsave(&list->lock, flags);
3162 __skb_queue_tail(list, newsk);
3163 spin_unlock_irqrestore(&list->lock, flags);
3165 EXPORT_SYMBOL(skb_queue_tail);
3168 * skb_unlink - remove a buffer from a list
3169 * @skb: buffer to remove
3170 * @list: list to use
3172 * Remove a packet from a list. The list locks are taken and this
3173 * function is atomic with respect to other list locked calls
3175 * You must know what list the SKB is on.
3177 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3179 unsigned long flags;
3181 spin_lock_irqsave(&list->lock, flags);
3182 __skb_unlink(skb, list);
3183 spin_unlock_irqrestore(&list->lock, flags);
3185 EXPORT_SYMBOL(skb_unlink);
3188 * skb_append - append a buffer
3189 * @old: buffer to insert after
3190 * @newsk: buffer to insert
3191 * @list: list to use
3193 * Place a packet after a given packet in a list. The list locks are taken
3194 * and this function is atomic with respect to other list locked calls.
3195 * A buffer cannot be placed on two lists at the same time.
3197 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3199 unsigned long flags;
3201 spin_lock_irqsave(&list->lock, flags);
3202 __skb_queue_after(list, old, newsk);
3203 spin_unlock_irqrestore(&list->lock, flags);
3205 EXPORT_SYMBOL(skb_append);
3207 static inline void skb_split_inside_header(struct sk_buff *skb,
3208 struct sk_buff* skb1,
3209 const u32 len, const int pos)
3213 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3215 /* And move data appendix as is. */
3216 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3217 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3219 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3220 skb_shinfo(skb)->nr_frags = 0;
3221 skb1->data_len = skb->data_len;
3222 skb1->len += skb1->data_len;
3225 skb_set_tail_pointer(skb, len);
3228 static inline void skb_split_no_header(struct sk_buff *skb,
3229 struct sk_buff* skb1,
3230 const u32 len, int pos)
3233 const int nfrags = skb_shinfo(skb)->nr_frags;
3235 skb_shinfo(skb)->nr_frags = 0;
3236 skb1->len = skb1->data_len = skb->len - len;
3238 skb->data_len = len - pos;
3240 for (i = 0; i < nfrags; i++) {
3241 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3243 if (pos + size > len) {
3244 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3248 * We have two variants in this case:
3249 * 1. Move all the frag to the second
3250 * part, if it is possible. F.e.
3251 * this approach is mandatory for TUX,
3252 * where splitting is expensive.
3253 * 2. Split is accurately. We make this.
3255 skb_frag_ref(skb, i);
3256 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3257 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3258 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3259 skb_shinfo(skb)->nr_frags++;
3263 skb_shinfo(skb)->nr_frags++;
3266 skb_shinfo(skb1)->nr_frags = k;
3270 * skb_split - Split fragmented skb to two parts at length len.
3271 * @skb: the buffer to split
3272 * @skb1: the buffer to receive the second part
3273 * @len: new length for skb
3275 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3277 int pos = skb_headlen(skb);
3279 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3281 skb_zerocopy_clone(skb1, skb, 0);
3282 if (len < pos) /* Split line is inside header. */
3283 skb_split_inside_header(skb, skb1, len, pos);
3284 else /* Second chunk has no header, nothing to copy. */
3285 skb_split_no_header(skb, skb1, len, pos);
3287 EXPORT_SYMBOL(skb_split);
3289 /* Shifting from/to a cloned skb is a no-go.
3291 * Caller cannot keep skb_shinfo related pointers past calling here!
3293 static int skb_prepare_for_shift(struct sk_buff *skb)
3295 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3299 * skb_shift - Shifts paged data partially from skb to another
3300 * @tgt: buffer into which tail data gets added
3301 * @skb: buffer from which the paged data comes from
3302 * @shiftlen: shift up to this many bytes
3304 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3305 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3306 * It's up to caller to free skb if everything was shifted.
3308 * If @tgt runs out of frags, the whole operation is aborted.
3310 * Skb cannot include anything else but paged data while tgt is allowed
3311 * to have non-paged data as well.
3313 * TODO: full sized shift could be optimized but that would need
3314 * specialized skb free'er to handle frags without up-to-date nr_frags.
3316 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3318 int from, to, merge, todo;
3319 skb_frag_t *fragfrom, *fragto;
3321 BUG_ON(shiftlen > skb->len);
3323 if (skb_headlen(skb))
3325 if (skb_zcopy(tgt) || skb_zcopy(skb))
3330 to = skb_shinfo(tgt)->nr_frags;
3331 fragfrom = &skb_shinfo(skb)->frags[from];
3333 /* Actual merge is delayed until the point when we know we can
3334 * commit all, so that we don't have to undo partial changes
3337 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3338 skb_frag_off(fragfrom))) {
3343 todo -= skb_frag_size(fragfrom);
3345 if (skb_prepare_for_shift(skb) ||
3346 skb_prepare_for_shift(tgt))
3349 /* All previous frag pointers might be stale! */
3350 fragfrom = &skb_shinfo(skb)->frags[from];
3351 fragto = &skb_shinfo(tgt)->frags[merge];
3353 skb_frag_size_add(fragto, shiftlen);
3354 skb_frag_size_sub(fragfrom, shiftlen);
3355 skb_frag_off_add(fragfrom, shiftlen);
3363 /* Skip full, not-fitting skb to avoid expensive operations */
3364 if ((shiftlen == skb->len) &&
3365 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3368 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3371 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3372 if (to == MAX_SKB_FRAGS)
3375 fragfrom = &skb_shinfo(skb)->frags[from];
3376 fragto = &skb_shinfo(tgt)->frags[to];
3378 if (todo >= skb_frag_size(fragfrom)) {
3379 *fragto = *fragfrom;
3380 todo -= skb_frag_size(fragfrom);
3385 __skb_frag_ref(fragfrom);
3386 skb_frag_page_copy(fragto, fragfrom);
3387 skb_frag_off_copy(fragto, fragfrom);
3388 skb_frag_size_set(fragto, todo);
3390 skb_frag_off_add(fragfrom, todo);
3391 skb_frag_size_sub(fragfrom, todo);
3399 /* Ready to "commit" this state change to tgt */
3400 skb_shinfo(tgt)->nr_frags = to;
3403 fragfrom = &skb_shinfo(skb)->frags[0];
3404 fragto = &skb_shinfo(tgt)->frags[merge];
3406 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3407 __skb_frag_unref(fragfrom);
3410 /* Reposition in the original skb */
3412 while (from < skb_shinfo(skb)->nr_frags)
3413 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3414 skb_shinfo(skb)->nr_frags = to;
3416 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3419 /* Most likely the tgt won't ever need its checksum anymore, skb on
3420 * the other hand might need it if it needs to be resent
3422 tgt->ip_summed = CHECKSUM_PARTIAL;
3423 skb->ip_summed = CHECKSUM_PARTIAL;
3425 /* Yak, is it really working this way? Some helper please? */
3426 skb->len -= shiftlen;
3427 skb->data_len -= shiftlen;
3428 skb->truesize -= shiftlen;
3429 tgt->len += shiftlen;
3430 tgt->data_len += shiftlen;
3431 tgt->truesize += shiftlen;
3437 * skb_prepare_seq_read - Prepare a sequential read of skb data
3438 * @skb: the buffer to read
3439 * @from: lower offset of data to be read
3440 * @to: upper offset of data to be read
3441 * @st: state variable
3443 * Initializes the specified state variable. Must be called before
3444 * invoking skb_seq_read() for the first time.
3446 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3447 unsigned int to, struct skb_seq_state *st)
3449 st->lower_offset = from;
3450 st->upper_offset = to;
3451 st->root_skb = st->cur_skb = skb;
3452 st->frag_idx = st->stepped_offset = 0;
3453 st->frag_data = NULL;
3456 EXPORT_SYMBOL(skb_prepare_seq_read);
3459 * skb_seq_read - Sequentially read skb data
3460 * @consumed: number of bytes consumed by the caller so far
3461 * @data: destination pointer for data to be returned
3462 * @st: state variable
3464 * Reads a block of skb data at @consumed relative to the
3465 * lower offset specified to skb_prepare_seq_read(). Assigns
3466 * the head of the data block to @data and returns the length
3467 * of the block or 0 if the end of the skb data or the upper
3468 * offset has been reached.
3470 * The caller is not required to consume all of the data
3471 * returned, i.e. @consumed is typically set to the number
3472 * of bytes already consumed and the next call to
3473 * skb_seq_read() will return the remaining part of the block.
3475 * Note 1: The size of each block of data returned can be arbitrary,
3476 * this limitation is the cost for zerocopy sequential
3477 * reads of potentially non linear data.
3479 * Note 2: Fragment lists within fragments are not implemented
3480 * at the moment, state->root_skb could be replaced with
3481 * a stack for this purpose.
3483 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3484 struct skb_seq_state *st)
3486 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3489 if (unlikely(abs_offset >= st->upper_offset)) {
3490 if (st->frag_data) {
3491 kunmap_atomic(st->frag_data);
3492 st->frag_data = NULL;
3498 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3500 if (abs_offset < block_limit && !st->frag_data) {
3501 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3502 return block_limit - abs_offset;
3505 if (st->frag_idx == 0 && !st->frag_data)
3506 st->stepped_offset += skb_headlen(st->cur_skb);
3508 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3509 unsigned int pg_idx, pg_off, pg_sz;
3511 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3514 pg_off = skb_frag_off(frag);
3515 pg_sz = skb_frag_size(frag);
3517 if (skb_frag_must_loop(skb_frag_page(frag))) {
3518 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3519 pg_off = offset_in_page(pg_off + st->frag_off);
3520 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3521 PAGE_SIZE - pg_off);
3524 block_limit = pg_sz + st->stepped_offset;
3525 if (abs_offset < block_limit) {
3527 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3529 *data = (u8 *)st->frag_data + pg_off +
3530 (abs_offset - st->stepped_offset);
3532 return block_limit - abs_offset;
3535 if (st->frag_data) {
3536 kunmap_atomic(st->frag_data);
3537 st->frag_data = NULL;
3540 st->stepped_offset += pg_sz;
3541 st->frag_off += pg_sz;
3542 if (st->frag_off == skb_frag_size(frag)) {
3548 if (st->frag_data) {
3549 kunmap_atomic(st->frag_data);
3550 st->frag_data = NULL;
3553 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3554 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3557 } else if (st->cur_skb->next) {
3558 st->cur_skb = st->cur_skb->next;
3565 EXPORT_SYMBOL(skb_seq_read);
3568 * skb_abort_seq_read - Abort a sequential read of skb data
3569 * @st: state variable
3571 * Must be called if skb_seq_read() was not called until it
3574 void skb_abort_seq_read(struct skb_seq_state *st)
3577 kunmap_atomic(st->frag_data);
3579 EXPORT_SYMBOL(skb_abort_seq_read);
3581 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3583 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3584 struct ts_config *conf,
3585 struct ts_state *state)
3587 return skb_seq_read(offset, text, TS_SKB_CB(state));
3590 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3592 skb_abort_seq_read(TS_SKB_CB(state));
3596 * skb_find_text - Find a text pattern in skb data
3597 * @skb: the buffer to look in
3598 * @from: search offset
3600 * @config: textsearch configuration
3602 * Finds a pattern in the skb data according to the specified
3603 * textsearch configuration. Use textsearch_next() to retrieve
3604 * subsequent occurrences of the pattern. Returns the offset
3605 * to the first occurrence or UINT_MAX if no match was found.
3607 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3608 unsigned int to, struct ts_config *config)
3610 struct ts_state state;
3613 config->get_next_block = skb_ts_get_next_block;
3614 config->finish = skb_ts_finish;
3616 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3618 ret = textsearch_find(config, &state);
3619 return (ret <= to - from ? ret : UINT_MAX);
3621 EXPORT_SYMBOL(skb_find_text);
3623 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3624 int offset, size_t size)
3626 int i = skb_shinfo(skb)->nr_frags;
3628 if (skb_can_coalesce(skb, i, page, offset)) {
3629 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3630 } else if (i < MAX_SKB_FRAGS) {
3632 skb_fill_page_desc(skb, i, page, offset, size);
3639 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3642 * skb_pull_rcsum - pull skb and update receive checksum
3643 * @skb: buffer to update
3644 * @len: length of data pulled
3646 * This function performs an skb_pull on the packet and updates
3647 * the CHECKSUM_COMPLETE checksum. It should be used on
3648 * receive path processing instead of skb_pull unless you know
3649 * that the checksum difference is zero (e.g., a valid IP header)
3650 * or you are setting ip_summed to CHECKSUM_NONE.
3652 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3654 unsigned char *data = skb->data;
3656 BUG_ON(len > skb->len);
3657 __skb_pull(skb, len);
3658 skb_postpull_rcsum(skb, data, len);
3661 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3663 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3665 skb_frag_t head_frag;
3668 page = virt_to_head_page(frag_skb->head);
3669 __skb_frag_set_page(&head_frag, page);
3670 skb_frag_off_set(&head_frag, frag_skb->data -
3671 (unsigned char *)page_address(page));
3672 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3676 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3677 netdev_features_t features,
3678 unsigned int offset)
3680 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3681 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3682 unsigned int delta_truesize = 0;
3683 unsigned int delta_len = 0;
3684 struct sk_buff *tail = NULL;
3685 struct sk_buff *nskb, *tmp;
3688 skb_push(skb, -skb_network_offset(skb) + offset);
3690 skb_shinfo(skb)->frag_list = NULL;
3694 list_skb = list_skb->next;
3697 if (skb_shared(nskb)) {
3698 tmp = skb_clone(nskb, GFP_ATOMIC);
3702 err = skb_unclone(nskb, GFP_ATOMIC);
3713 if (unlikely(err)) {
3714 nskb->next = list_skb;
3720 delta_len += nskb->len;
3721 delta_truesize += nskb->truesize;
3723 skb_push(nskb, -skb_network_offset(nskb) + offset);
3725 skb_release_head_state(nskb);
3726 __copy_skb_header(nskb, skb);
3728 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3729 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3730 nskb->data - tnl_hlen,
3733 if (skb_needs_linearize(nskb, features) &&
3734 __skb_linearize(nskb))
3739 skb->truesize = skb->truesize - delta_truesize;
3740 skb->data_len = skb->data_len - delta_len;
3741 skb->len = skb->len - delta_len;
3747 if (skb_needs_linearize(skb, features) &&
3748 __skb_linearize(skb))
3756 kfree_skb_list(skb->next);
3758 return ERR_PTR(-ENOMEM);
3760 EXPORT_SYMBOL_GPL(skb_segment_list);
3762 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3764 if (unlikely(p->len + skb->len >= 65536))
3767 if (NAPI_GRO_CB(p)->last == p)
3768 skb_shinfo(p)->frag_list = skb;
3770 NAPI_GRO_CB(p)->last->next = skb;
3772 skb_pull(skb, skb_gro_offset(skb));
3774 NAPI_GRO_CB(p)->last = skb;
3775 NAPI_GRO_CB(p)->count++;
3776 p->data_len += skb->len;
3777 p->truesize += skb->truesize;
3780 NAPI_GRO_CB(skb)->same_flow = 1;
3786 * skb_segment - Perform protocol segmentation on skb.
3787 * @head_skb: buffer to segment
3788 * @features: features for the output path (see dev->features)
3790 * This function performs segmentation on the given skb. It returns
3791 * a pointer to the first in a list of new skbs for the segments.
3792 * In case of error it returns ERR_PTR(err).
3794 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3795 netdev_features_t features)
3797 struct sk_buff *segs = NULL;
3798 struct sk_buff *tail = NULL;
3799 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3800 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3801 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3802 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3803 struct sk_buff *frag_skb = head_skb;
3804 unsigned int offset = doffset;
3805 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3806 unsigned int partial_segs = 0;
3807 unsigned int headroom;
3808 unsigned int len = head_skb->len;
3811 int nfrags = skb_shinfo(head_skb)->nr_frags;
3816 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3817 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3818 /* gso_size is untrusted, and we have a frag_list with a linear
3819 * non head_frag head.
3821 * (we assume checking the first list_skb member suffices;
3822 * i.e if either of the list_skb members have non head_frag
3823 * head, then the first one has too).
3825 * If head_skb's headlen does not fit requested gso_size, it
3826 * means that the frag_list members do NOT terminate on exact
3827 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3828 * sharing. Therefore we must fallback to copying the frag_list
3829 * skbs; we do so by disabling SG.
3831 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3832 features &= ~NETIF_F_SG;
3835 __skb_push(head_skb, doffset);
3836 proto = skb_network_protocol(head_skb, NULL);
3837 if (unlikely(!proto))
3838 return ERR_PTR(-EINVAL);
3840 sg = !!(features & NETIF_F_SG);
3841 csum = !!can_checksum_protocol(features, proto);
3843 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3844 if (!(features & NETIF_F_GSO_PARTIAL)) {
3845 struct sk_buff *iter;
3846 unsigned int frag_len;
3849 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3852 /* If we get here then all the required
3853 * GSO features except frag_list are supported.
3854 * Try to split the SKB to multiple GSO SKBs
3855 * with no frag_list.
3856 * Currently we can do that only when the buffers don't
3857 * have a linear part and all the buffers except
3858 * the last are of the same length.
3860 frag_len = list_skb->len;
3861 skb_walk_frags(head_skb, iter) {
3862 if (frag_len != iter->len && iter->next)
3864 if (skb_headlen(iter) && !iter->head_frag)
3870 if (len != frag_len)
3874 /* GSO partial only requires that we trim off any excess that
3875 * doesn't fit into an MSS sized block, so take care of that
3878 partial_segs = len / mss;
3879 if (partial_segs > 1)
3880 mss *= partial_segs;
3886 headroom = skb_headroom(head_skb);
3887 pos = skb_headlen(head_skb);
3890 struct sk_buff *nskb;
3891 skb_frag_t *nskb_frag;
3895 if (unlikely(mss == GSO_BY_FRAGS)) {
3896 len = list_skb->len;
3898 len = head_skb->len - offset;
3903 hsize = skb_headlen(head_skb) - offset;
3906 if (hsize > len || !sg)
3909 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3910 (skb_headlen(list_skb) == len || sg)) {
3911 BUG_ON(skb_headlen(list_skb) > len);
3914 nfrags = skb_shinfo(list_skb)->nr_frags;
3915 frag = skb_shinfo(list_skb)->frags;
3916 frag_skb = list_skb;
3917 pos += skb_headlen(list_skb);
3919 while (pos < offset + len) {
3920 BUG_ON(i >= nfrags);
3922 size = skb_frag_size(frag);
3923 if (pos + size > offset + len)
3931 nskb = skb_clone(list_skb, GFP_ATOMIC);
3932 list_skb = list_skb->next;
3934 if (unlikely(!nskb))
3937 if (unlikely(pskb_trim(nskb, len))) {
3942 hsize = skb_end_offset(nskb);
3943 if (skb_cow_head(nskb, doffset + headroom)) {
3948 nskb->truesize += skb_end_offset(nskb) - hsize;
3949 skb_release_head_state(nskb);
3950 __skb_push(nskb, doffset);
3952 nskb = __alloc_skb(hsize + doffset + headroom,
3953 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3956 if (unlikely(!nskb))
3959 skb_reserve(nskb, headroom);
3960 __skb_put(nskb, doffset);
3969 __copy_skb_header(nskb, head_skb);
3971 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3972 skb_reset_mac_len(nskb);
3974 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3975 nskb->data - tnl_hlen,
3976 doffset + tnl_hlen);
3978 if (nskb->len == len + doffset)
3979 goto perform_csum_check;
3983 if (!nskb->remcsum_offload)
3984 nskb->ip_summed = CHECKSUM_NONE;
3985 SKB_GSO_CB(nskb)->csum =
3986 skb_copy_and_csum_bits(head_skb, offset,
3990 SKB_GSO_CB(nskb)->csum_start =
3991 skb_headroom(nskb) + doffset;
3993 skb_copy_bits(head_skb, offset,
4000 nskb_frag = skb_shinfo(nskb)->frags;
4002 skb_copy_from_linear_data_offset(head_skb, offset,
4003 skb_put(nskb, hsize), hsize);
4005 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
4008 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4009 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4012 while (pos < offset + len) {
4015 nfrags = skb_shinfo(list_skb)->nr_frags;
4016 frag = skb_shinfo(list_skb)->frags;
4017 frag_skb = list_skb;
4018 if (!skb_headlen(list_skb)) {
4021 BUG_ON(!list_skb->head_frag);
4023 /* to make room for head_frag. */
4027 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4028 skb_zerocopy_clone(nskb, frag_skb,
4032 list_skb = list_skb->next;
4035 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4037 net_warn_ratelimited(
4038 "skb_segment: too many frags: %u %u\n",
4044 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4045 __skb_frag_ref(nskb_frag);
4046 size = skb_frag_size(nskb_frag);
4049 skb_frag_off_add(nskb_frag, offset - pos);
4050 skb_frag_size_sub(nskb_frag, offset - pos);
4053 skb_shinfo(nskb)->nr_frags++;
4055 if (pos + size <= offset + len) {
4060 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4068 nskb->data_len = len - hsize;
4069 nskb->len += nskb->data_len;
4070 nskb->truesize += nskb->data_len;
4074 if (skb_has_shared_frag(nskb) &&
4075 __skb_linearize(nskb))
4078 if (!nskb->remcsum_offload)
4079 nskb->ip_summed = CHECKSUM_NONE;
4080 SKB_GSO_CB(nskb)->csum =
4081 skb_checksum(nskb, doffset,
4082 nskb->len - doffset, 0);
4083 SKB_GSO_CB(nskb)->csum_start =
4084 skb_headroom(nskb) + doffset;
4086 } while ((offset += len) < head_skb->len);
4088 /* Some callers want to get the end of the list.
4089 * Put it in segs->prev to avoid walking the list.
4090 * (see validate_xmit_skb_list() for example)
4095 struct sk_buff *iter;
4096 int type = skb_shinfo(head_skb)->gso_type;
4097 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4099 /* Update type to add partial and then remove dodgy if set */
4100 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4101 type &= ~SKB_GSO_DODGY;
4103 /* Update GSO info and prepare to start updating headers on
4104 * our way back down the stack of protocols.
4106 for (iter = segs; iter; iter = iter->next) {
4107 skb_shinfo(iter)->gso_size = gso_size;
4108 skb_shinfo(iter)->gso_segs = partial_segs;
4109 skb_shinfo(iter)->gso_type = type;
4110 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4113 if (tail->len - doffset <= gso_size)
4114 skb_shinfo(tail)->gso_size = 0;
4115 else if (tail != segs)
4116 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4119 /* Following permits correct backpressure, for protocols
4120 * using skb_set_owner_w().
4121 * Idea is to tranfert ownership from head_skb to last segment.
4123 if (head_skb->destructor == sock_wfree) {
4124 swap(tail->truesize, head_skb->truesize);
4125 swap(tail->destructor, head_skb->destructor);
4126 swap(tail->sk, head_skb->sk);
4131 kfree_skb_list(segs);
4132 return ERR_PTR(err);
4134 EXPORT_SYMBOL_GPL(skb_segment);
4136 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4138 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4139 unsigned int offset = skb_gro_offset(skb);
4140 unsigned int headlen = skb_headlen(skb);
4141 unsigned int len = skb_gro_len(skb);
4142 unsigned int delta_truesize;
4145 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4148 lp = NAPI_GRO_CB(p)->last;
4149 pinfo = skb_shinfo(lp);
4151 if (headlen <= offset) {
4154 int i = skbinfo->nr_frags;
4155 int nr_frags = pinfo->nr_frags + i;
4157 if (nr_frags > MAX_SKB_FRAGS)
4161 pinfo->nr_frags = nr_frags;
4162 skbinfo->nr_frags = 0;
4164 frag = pinfo->frags + nr_frags;
4165 frag2 = skbinfo->frags + i;
4170 skb_frag_off_add(frag, offset);
4171 skb_frag_size_sub(frag, offset);
4173 /* all fragments truesize : remove (head size + sk_buff) */
4174 delta_truesize = skb->truesize -
4175 SKB_TRUESIZE(skb_end_offset(skb));
4177 skb->truesize -= skb->data_len;
4178 skb->len -= skb->data_len;
4181 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4183 } else if (skb->head_frag) {
4184 int nr_frags = pinfo->nr_frags;
4185 skb_frag_t *frag = pinfo->frags + nr_frags;
4186 struct page *page = virt_to_head_page(skb->head);
4187 unsigned int first_size = headlen - offset;
4188 unsigned int first_offset;
4190 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4193 first_offset = skb->data -
4194 (unsigned char *)page_address(page) +
4197 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4199 __skb_frag_set_page(frag, page);
4200 skb_frag_off_set(frag, first_offset);
4201 skb_frag_size_set(frag, first_size);
4203 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4204 /* We dont need to clear skbinfo->nr_frags here */
4206 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4207 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4212 delta_truesize = skb->truesize;
4213 if (offset > headlen) {
4214 unsigned int eat = offset - headlen;
4216 skb_frag_off_add(&skbinfo->frags[0], eat);
4217 skb_frag_size_sub(&skbinfo->frags[0], eat);
4218 skb->data_len -= eat;
4223 __skb_pull(skb, offset);
4225 if (NAPI_GRO_CB(p)->last == p)
4226 skb_shinfo(p)->frag_list = skb;
4228 NAPI_GRO_CB(p)->last->next = skb;
4229 NAPI_GRO_CB(p)->last = skb;
4230 __skb_header_release(skb);
4234 NAPI_GRO_CB(p)->count++;
4236 p->truesize += delta_truesize;
4239 lp->data_len += len;
4240 lp->truesize += delta_truesize;
4243 NAPI_GRO_CB(skb)->same_flow = 1;
4247 #ifdef CONFIG_SKB_EXTENSIONS
4248 #define SKB_EXT_ALIGN_VALUE 8
4249 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4251 static const u8 skb_ext_type_len[] = {
4252 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4253 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4256 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4258 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4259 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4261 #if IS_ENABLED(CONFIG_MPTCP)
4262 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4266 static __always_inline unsigned int skb_ext_total_length(void)
4268 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4269 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4270 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4273 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4275 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4276 skb_ext_type_len[TC_SKB_EXT] +
4278 #if IS_ENABLED(CONFIG_MPTCP)
4279 skb_ext_type_len[SKB_EXT_MPTCP] +
4284 static void skb_extensions_init(void)
4286 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4287 BUILD_BUG_ON(skb_ext_total_length() > 255);
4289 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4290 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4292 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4296 static void skb_extensions_init(void) {}
4299 void __init skb_init(void)
4301 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4302 sizeof(struct sk_buff),
4304 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4305 offsetof(struct sk_buff, cb),
4306 sizeof_field(struct sk_buff, cb),
4308 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4309 sizeof(struct sk_buff_fclones),
4311 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4313 skb_extensions_init();
4317 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4318 unsigned int recursion_level)
4320 int start = skb_headlen(skb);
4321 int i, copy = start - offset;
4322 struct sk_buff *frag_iter;
4325 if (unlikely(recursion_level >= 24))
4331 sg_set_buf(sg, skb->data + offset, copy);
4333 if ((len -= copy) == 0)
4338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4341 WARN_ON(start > offset + len);
4343 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4344 if ((copy = end - offset) > 0) {
4345 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4346 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4351 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4352 skb_frag_off(frag) + offset - start);
4361 skb_walk_frags(skb, frag_iter) {
4364 WARN_ON(start > offset + len);
4366 end = start + frag_iter->len;
4367 if ((copy = end - offset) > 0) {
4368 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4373 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4374 copy, recursion_level + 1);
4375 if (unlikely(ret < 0))
4378 if ((len -= copy) == 0)
4389 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4390 * @skb: Socket buffer containing the buffers to be mapped
4391 * @sg: The scatter-gather list to map into
4392 * @offset: The offset into the buffer's contents to start mapping
4393 * @len: Length of buffer space to be mapped
4395 * Fill the specified scatter-gather list with mappings/pointers into a
4396 * region of the buffer space attached to a socket buffer. Returns either
4397 * the number of scatterlist items used, or -EMSGSIZE if the contents
4400 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4402 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4407 sg_mark_end(&sg[nsg - 1]);
4411 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4413 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4414 * sglist without mark the sg which contain last skb data as the end.
4415 * So the caller can mannipulate sg list as will when padding new data after
4416 * the first call without calling sg_unmark_end to expend sg list.
4418 * Scenario to use skb_to_sgvec_nomark:
4420 * 2. skb_to_sgvec_nomark(payload1)
4421 * 3. skb_to_sgvec_nomark(payload2)
4423 * This is equivalent to:
4425 * 2. skb_to_sgvec(payload1)
4427 * 4. skb_to_sgvec(payload2)
4429 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4430 * is more preferable.
4432 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4433 int offset, int len)
4435 return __skb_to_sgvec(skb, sg, offset, len, 0);
4437 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4442 * skb_cow_data - Check that a socket buffer's data buffers are writable
4443 * @skb: The socket buffer to check.
4444 * @tailbits: Amount of trailing space to be added
4445 * @trailer: Returned pointer to the skb where the @tailbits space begins
4447 * Make sure that the data buffers attached to a socket buffer are
4448 * writable. If they are not, private copies are made of the data buffers
4449 * and the socket buffer is set to use these instead.
4451 * If @tailbits is given, make sure that there is space to write @tailbits
4452 * bytes of data beyond current end of socket buffer. @trailer will be
4453 * set to point to the skb in which this space begins.
4455 * The number of scatterlist elements required to completely map the
4456 * COW'd and extended socket buffer will be returned.
4458 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4462 struct sk_buff *skb1, **skb_p;
4464 /* If skb is cloned or its head is paged, reallocate
4465 * head pulling out all the pages (pages are considered not writable
4466 * at the moment even if they are anonymous).
4468 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4469 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4472 /* Easy case. Most of packets will go this way. */
4473 if (!skb_has_frag_list(skb)) {
4474 /* A little of trouble, not enough of space for trailer.
4475 * This should not happen, when stack is tuned to generate
4476 * good frames. OK, on miss we reallocate and reserve even more
4477 * space, 128 bytes is fair. */
4479 if (skb_tailroom(skb) < tailbits &&
4480 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4488 /* Misery. We are in troubles, going to mincer fragments... */
4491 skb_p = &skb_shinfo(skb)->frag_list;
4494 while ((skb1 = *skb_p) != NULL) {
4497 /* The fragment is partially pulled by someone,
4498 * this can happen on input. Copy it and everything
4501 if (skb_shared(skb1))
4504 /* If the skb is the last, worry about trailer. */
4506 if (skb1->next == NULL && tailbits) {
4507 if (skb_shinfo(skb1)->nr_frags ||
4508 skb_has_frag_list(skb1) ||
4509 skb_tailroom(skb1) < tailbits)
4510 ntail = tailbits + 128;
4516 skb_shinfo(skb1)->nr_frags ||
4517 skb_has_frag_list(skb1)) {
4518 struct sk_buff *skb2;
4520 /* Fuck, we are miserable poor guys... */
4522 skb2 = skb_copy(skb1, GFP_ATOMIC);
4524 skb2 = skb_copy_expand(skb1,
4528 if (unlikely(skb2 == NULL))
4532 skb_set_owner_w(skb2, skb1->sk);
4534 /* Looking around. Are we still alive?
4535 * OK, link new skb, drop old one */
4537 skb2->next = skb1->next;
4544 skb_p = &skb1->next;
4549 EXPORT_SYMBOL_GPL(skb_cow_data);
4551 static void sock_rmem_free(struct sk_buff *skb)
4553 struct sock *sk = skb->sk;
4555 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4558 static void skb_set_err_queue(struct sk_buff *skb)
4560 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4561 * So, it is safe to (mis)use it to mark skbs on the error queue.
4563 skb->pkt_type = PACKET_OUTGOING;
4564 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4568 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4570 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4572 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4573 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4578 skb->destructor = sock_rmem_free;
4579 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4580 skb_set_err_queue(skb);
4582 /* before exiting rcu section, make sure dst is refcounted */
4585 skb_queue_tail(&sk->sk_error_queue, skb);
4586 if (!sock_flag(sk, SOCK_DEAD))
4587 sk->sk_error_report(sk);
4590 EXPORT_SYMBOL(sock_queue_err_skb);
4592 static bool is_icmp_err_skb(const struct sk_buff *skb)
4594 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4595 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4598 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4600 struct sk_buff_head *q = &sk->sk_error_queue;
4601 struct sk_buff *skb, *skb_next = NULL;
4602 bool icmp_next = false;
4603 unsigned long flags;
4605 spin_lock_irqsave(&q->lock, flags);
4606 skb = __skb_dequeue(q);
4607 if (skb && (skb_next = skb_peek(q))) {
4608 icmp_next = is_icmp_err_skb(skb_next);
4610 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4612 spin_unlock_irqrestore(&q->lock, flags);
4614 if (is_icmp_err_skb(skb) && !icmp_next)
4618 sk->sk_error_report(sk);
4622 EXPORT_SYMBOL(sock_dequeue_err_skb);
4625 * skb_clone_sk - create clone of skb, and take reference to socket
4626 * @skb: the skb to clone
4628 * This function creates a clone of a buffer that holds a reference on
4629 * sk_refcnt. Buffers created via this function are meant to be
4630 * returned using sock_queue_err_skb, or free via kfree_skb.
4632 * When passing buffers allocated with this function to sock_queue_err_skb
4633 * it is necessary to wrap the call with sock_hold/sock_put in order to
4634 * prevent the socket from being released prior to being enqueued on
4635 * the sk_error_queue.
4637 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4639 struct sock *sk = skb->sk;
4640 struct sk_buff *clone;
4642 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4645 clone = skb_clone(skb, GFP_ATOMIC);
4652 clone->destructor = sock_efree;
4656 EXPORT_SYMBOL(skb_clone_sk);
4658 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4663 struct sock_exterr_skb *serr;
4666 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4668 serr = SKB_EXT_ERR(skb);
4669 memset(serr, 0, sizeof(*serr));
4670 serr->ee.ee_errno = ENOMSG;
4671 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4672 serr->ee.ee_info = tstype;
4673 serr->opt_stats = opt_stats;
4674 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4675 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4676 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4677 if (sk->sk_protocol == IPPROTO_TCP &&
4678 sk->sk_type == SOCK_STREAM)
4679 serr->ee.ee_data -= sk->sk_tskey;
4682 err = sock_queue_err_skb(sk, skb);
4688 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4692 if (likely(sysctl_tstamp_allow_data || tsonly))
4695 read_lock_bh(&sk->sk_callback_lock);
4696 ret = sk->sk_socket && sk->sk_socket->file &&
4697 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4698 read_unlock_bh(&sk->sk_callback_lock);
4702 void skb_complete_tx_timestamp(struct sk_buff *skb,
4703 struct skb_shared_hwtstamps *hwtstamps)
4705 struct sock *sk = skb->sk;
4707 if (!skb_may_tx_timestamp(sk, false))
4710 /* Take a reference to prevent skb_orphan() from freeing the socket,
4711 * but only if the socket refcount is not zero.
4713 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4714 *skb_hwtstamps(skb) = *hwtstamps;
4715 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4723 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4725 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4726 struct skb_shared_hwtstamps *hwtstamps,
4727 struct sock *sk, int tstype)
4729 struct sk_buff *skb;
4730 bool tsonly, opt_stats = false;
4735 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4736 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4739 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4740 if (!skb_may_tx_timestamp(sk, tsonly))
4745 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4746 sk->sk_protocol == IPPROTO_TCP &&
4747 sk->sk_type == SOCK_STREAM) {
4748 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4752 skb = alloc_skb(0, GFP_ATOMIC);
4754 skb = skb_clone(orig_skb, GFP_ATOMIC);
4760 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4762 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4766 *skb_hwtstamps(skb) = *hwtstamps;
4768 skb->tstamp = ktime_get_real();
4770 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4772 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4774 void skb_tstamp_tx(struct sk_buff *orig_skb,
4775 struct skb_shared_hwtstamps *hwtstamps)
4777 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4780 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4782 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4784 struct sock *sk = skb->sk;
4785 struct sock_exterr_skb *serr;
4788 skb->wifi_acked_valid = 1;
4789 skb->wifi_acked = acked;
4791 serr = SKB_EXT_ERR(skb);
4792 memset(serr, 0, sizeof(*serr));
4793 serr->ee.ee_errno = ENOMSG;
4794 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4796 /* Take a reference to prevent skb_orphan() from freeing the socket,
4797 * but only if the socket refcount is not zero.
4799 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4800 err = sock_queue_err_skb(sk, skb);
4806 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4809 * skb_partial_csum_set - set up and verify partial csum values for packet
4810 * @skb: the skb to set
4811 * @start: the number of bytes after skb->data to start checksumming.
4812 * @off: the offset from start to place the checksum.
4814 * For untrusted partially-checksummed packets, we need to make sure the values
4815 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4817 * This function checks and sets those values and skb->ip_summed: if this
4818 * returns false you should drop the packet.
4820 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4822 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4823 u32 csum_start = skb_headroom(skb) + (u32)start;
4825 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4826 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4827 start, off, skb_headroom(skb), skb_headlen(skb));
4830 skb->ip_summed = CHECKSUM_PARTIAL;
4831 skb->csum_start = csum_start;
4832 skb->csum_offset = off;
4833 skb_set_transport_header(skb, start);
4836 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4838 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4841 if (skb_headlen(skb) >= len)
4844 /* If we need to pullup then pullup to the max, so we
4845 * won't need to do it again.
4850 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4853 if (skb_headlen(skb) < len)
4859 #define MAX_TCP_HDR_LEN (15 * 4)
4861 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4862 typeof(IPPROTO_IP) proto,
4869 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4870 off + MAX_TCP_HDR_LEN);
4871 if (!err && !skb_partial_csum_set(skb, off,
4872 offsetof(struct tcphdr,
4875 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4878 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4879 off + sizeof(struct udphdr));
4880 if (!err && !skb_partial_csum_set(skb, off,
4881 offsetof(struct udphdr,
4884 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4887 return ERR_PTR(-EPROTO);
4890 /* This value should be large enough to cover a tagged ethernet header plus
4891 * maximally sized IP and TCP or UDP headers.
4893 #define MAX_IP_HDR_LEN 128
4895 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4904 err = skb_maybe_pull_tail(skb,
4905 sizeof(struct iphdr),
4910 if (ip_is_fragment(ip_hdr(skb)))
4913 off = ip_hdrlen(skb);
4920 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4922 return PTR_ERR(csum);
4925 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4928 ip_hdr(skb)->protocol, 0);
4935 /* This value should be large enough to cover a tagged ethernet header plus
4936 * an IPv6 header, all options, and a maximal TCP or UDP header.
4938 #define MAX_IPV6_HDR_LEN 256
4940 #define OPT_HDR(type, skb, off) \
4941 (type *)(skb_network_header(skb) + (off))
4943 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4956 off = sizeof(struct ipv6hdr);
4958 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4962 nexthdr = ipv6_hdr(skb)->nexthdr;
4964 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4965 while (off <= len && !done) {
4967 case IPPROTO_DSTOPTS:
4968 case IPPROTO_HOPOPTS:
4969 case IPPROTO_ROUTING: {
4970 struct ipv6_opt_hdr *hp;
4972 err = skb_maybe_pull_tail(skb,
4974 sizeof(struct ipv6_opt_hdr),
4979 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4980 nexthdr = hp->nexthdr;
4981 off += ipv6_optlen(hp);
4985 struct ip_auth_hdr *hp;
4987 err = skb_maybe_pull_tail(skb,
4989 sizeof(struct ip_auth_hdr),
4994 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4995 nexthdr = hp->nexthdr;
4996 off += ipv6_authlen(hp);
4999 case IPPROTO_FRAGMENT: {
5000 struct frag_hdr *hp;
5002 err = skb_maybe_pull_tail(skb,
5004 sizeof(struct frag_hdr),
5009 hp = OPT_HDR(struct frag_hdr, skb, off);
5011 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5014 nexthdr = hp->nexthdr;
5015 off += sizeof(struct frag_hdr);
5026 if (!done || fragment)
5029 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5031 return PTR_ERR(csum);
5034 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5035 &ipv6_hdr(skb)->daddr,
5036 skb->len - off, nexthdr, 0);
5044 * skb_checksum_setup - set up partial checksum offset
5045 * @skb: the skb to set up
5046 * @recalculate: if true the pseudo-header checksum will be recalculated
5048 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5052 switch (skb->protocol) {
5053 case htons(ETH_P_IP):
5054 err = skb_checksum_setup_ipv4(skb, recalculate);
5057 case htons(ETH_P_IPV6):
5058 err = skb_checksum_setup_ipv6(skb, recalculate);
5068 EXPORT_SYMBOL(skb_checksum_setup);
5071 * skb_checksum_maybe_trim - maybe trims the given skb
5072 * @skb: the skb to check
5073 * @transport_len: the data length beyond the network header
5075 * Checks whether the given skb has data beyond the given transport length.
5076 * If so, returns a cloned skb trimmed to this transport length.
5077 * Otherwise returns the provided skb. Returns NULL in error cases
5078 * (e.g. transport_len exceeds skb length or out-of-memory).
5080 * Caller needs to set the skb transport header and free any returned skb if it
5081 * differs from the provided skb.
5083 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5084 unsigned int transport_len)
5086 struct sk_buff *skb_chk;
5087 unsigned int len = skb_transport_offset(skb) + transport_len;
5092 else if (skb->len == len)
5095 skb_chk = skb_clone(skb, GFP_ATOMIC);
5099 ret = pskb_trim_rcsum(skb_chk, len);
5109 * skb_checksum_trimmed - validate checksum of an skb
5110 * @skb: the skb to check
5111 * @transport_len: the data length beyond the network header
5112 * @skb_chkf: checksum function to use
5114 * Applies the given checksum function skb_chkf to the provided skb.
5115 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5117 * If the skb has data beyond the given transport length, then a
5118 * trimmed & cloned skb is checked and returned.
5120 * Caller needs to set the skb transport header and free any returned skb if it
5121 * differs from the provided skb.
5123 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5124 unsigned int transport_len,
5125 __sum16(*skb_chkf)(struct sk_buff *skb))
5127 struct sk_buff *skb_chk;
5128 unsigned int offset = skb_transport_offset(skb);
5131 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5135 if (!pskb_may_pull(skb_chk, offset))
5138 skb_pull_rcsum(skb_chk, offset);
5139 ret = skb_chkf(skb_chk);
5140 skb_push_rcsum(skb_chk, offset);
5148 if (skb_chk && skb_chk != skb)
5154 EXPORT_SYMBOL(skb_checksum_trimmed);
5156 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5158 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5161 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5163 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5166 skb_release_head_state(skb);
5167 kmem_cache_free(skbuff_head_cache, skb);
5172 EXPORT_SYMBOL(kfree_skb_partial);
5175 * skb_try_coalesce - try to merge skb to prior one
5177 * @from: buffer to add
5178 * @fragstolen: pointer to boolean
5179 * @delta_truesize: how much more was allocated than was requested
5181 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5182 bool *fragstolen, int *delta_truesize)
5184 struct skb_shared_info *to_shinfo, *from_shinfo;
5185 int i, delta, len = from->len;
5187 *fragstolen = false;
5192 if (len <= skb_tailroom(to)) {
5194 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5195 *delta_truesize = 0;
5199 to_shinfo = skb_shinfo(to);
5200 from_shinfo = skb_shinfo(from);
5201 if (to_shinfo->frag_list || from_shinfo->frag_list)
5203 if (skb_zcopy(to) || skb_zcopy(from))
5206 if (skb_headlen(from) != 0) {
5208 unsigned int offset;
5210 if (to_shinfo->nr_frags +
5211 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5214 if (skb_head_is_locked(from))
5217 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5219 page = virt_to_head_page(from->head);
5220 offset = from->data - (unsigned char *)page_address(page);
5222 skb_fill_page_desc(to, to_shinfo->nr_frags,
5223 page, offset, skb_headlen(from));
5226 if (to_shinfo->nr_frags +
5227 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5230 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5233 WARN_ON_ONCE(delta < len);
5235 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5237 from_shinfo->nr_frags * sizeof(skb_frag_t));
5238 to_shinfo->nr_frags += from_shinfo->nr_frags;
5240 if (!skb_cloned(from))
5241 from_shinfo->nr_frags = 0;
5243 /* if the skb is not cloned this does nothing
5244 * since we set nr_frags to 0.
5246 for (i = 0; i < from_shinfo->nr_frags; i++)
5247 __skb_frag_ref(&from_shinfo->frags[i]);
5249 to->truesize += delta;
5251 to->data_len += len;
5253 *delta_truesize = delta;
5256 EXPORT_SYMBOL(skb_try_coalesce);
5259 * skb_scrub_packet - scrub an skb
5261 * @skb: buffer to clean
5262 * @xnet: packet is crossing netns
5264 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5265 * into/from a tunnel. Some information have to be cleared during these
5267 * skb_scrub_packet can also be used to clean a skb before injecting it in
5268 * another namespace (@xnet == true). We have to clear all information in the
5269 * skb that could impact namespace isolation.
5271 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5273 skb->pkt_type = PACKET_HOST;
5279 nf_reset_trace(skb);
5281 #ifdef CONFIG_NET_SWITCHDEV
5282 skb->offload_fwd_mark = 0;
5283 skb->offload_l3_fwd_mark = 0;
5293 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5296 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5300 * skb_gso_transport_seglen is used to determine the real size of the
5301 * individual segments, including Layer4 headers (TCP/UDP).
5303 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5305 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5307 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5308 unsigned int thlen = 0;
5310 if (skb->encapsulation) {
5311 thlen = skb_inner_transport_header(skb) -
5312 skb_transport_header(skb);
5314 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5315 thlen += inner_tcp_hdrlen(skb);
5316 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5317 thlen = tcp_hdrlen(skb);
5318 } else if (unlikely(skb_is_gso_sctp(skb))) {
5319 thlen = sizeof(struct sctphdr);
5320 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5321 thlen = sizeof(struct udphdr);
5323 /* UFO sets gso_size to the size of the fragmentation
5324 * payload, i.e. the size of the L4 (UDP) header is already
5327 return thlen + shinfo->gso_size;
5331 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5335 * skb_gso_network_seglen is used to determine the real size of the
5336 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5338 * The MAC/L2 header is not accounted for.
5340 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5342 unsigned int hdr_len = skb_transport_header(skb) -
5343 skb_network_header(skb);
5345 return hdr_len + skb_gso_transport_seglen(skb);
5349 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5353 * skb_gso_mac_seglen is used to determine the real size of the
5354 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5355 * headers (TCP/UDP).
5357 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5359 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5361 return hdr_len + skb_gso_transport_seglen(skb);
5365 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5367 * There are a couple of instances where we have a GSO skb, and we
5368 * want to determine what size it would be after it is segmented.
5370 * We might want to check:
5371 * - L3+L4+payload size (e.g. IP forwarding)
5372 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5374 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5378 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5379 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5381 * @max_len: The maximum permissible length.
5383 * Returns true if the segmented length <= max length.
5385 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5386 unsigned int seg_len,
5387 unsigned int max_len) {
5388 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5389 const struct sk_buff *iter;
5391 if (shinfo->gso_size != GSO_BY_FRAGS)
5392 return seg_len <= max_len;
5394 /* Undo this so we can re-use header sizes */
5395 seg_len -= GSO_BY_FRAGS;
5397 skb_walk_frags(skb, iter) {
5398 if (seg_len + skb_headlen(iter) > max_len)
5406 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5409 * @mtu: MTU to validate against
5411 * skb_gso_validate_network_len validates if a given skb will fit a
5412 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5415 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5417 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5419 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5422 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5425 * @len: length to validate against
5427 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5428 * length once split, including L2, L3 and L4 headers and the payload.
5430 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5432 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5434 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5436 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5438 int mac_len, meta_len;
5441 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5446 mac_len = skb->data - skb_mac_header(skb);
5447 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5448 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5449 mac_len - VLAN_HLEN - ETH_TLEN);
5452 meta_len = skb_metadata_len(skb);
5454 meta = skb_metadata_end(skb) - meta_len;
5455 memmove(meta + VLAN_HLEN, meta, meta_len);
5458 skb->mac_header += VLAN_HLEN;
5462 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5464 struct vlan_hdr *vhdr;
5467 if (unlikely(skb_vlan_tag_present(skb))) {
5468 /* vlan_tci is already set-up so leave this for another time */
5472 skb = skb_share_check(skb, GFP_ATOMIC);
5475 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5476 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5479 vhdr = (struct vlan_hdr *)skb->data;
5480 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5481 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5483 skb_pull_rcsum(skb, VLAN_HLEN);
5484 vlan_set_encap_proto(skb, vhdr);
5486 skb = skb_reorder_vlan_header(skb);
5490 skb_reset_network_header(skb);
5491 if (!skb_transport_header_was_set(skb))
5492 skb_reset_transport_header(skb);
5493 skb_reset_mac_len(skb);
5501 EXPORT_SYMBOL(skb_vlan_untag);
5503 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5505 if (!pskb_may_pull(skb, write_len))
5508 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5511 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5513 EXPORT_SYMBOL(skb_ensure_writable);
5515 /* remove VLAN header from packet and update csum accordingly.
5516 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5518 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5520 struct vlan_hdr *vhdr;
5521 int offset = skb->data - skb_mac_header(skb);
5524 if (WARN_ONCE(offset,
5525 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5530 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5534 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5536 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5537 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5539 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5540 __skb_pull(skb, VLAN_HLEN);
5542 vlan_set_encap_proto(skb, vhdr);
5543 skb->mac_header += VLAN_HLEN;
5545 if (skb_network_offset(skb) < ETH_HLEN)
5546 skb_set_network_header(skb, ETH_HLEN);
5548 skb_reset_mac_len(skb);
5552 EXPORT_SYMBOL(__skb_vlan_pop);
5554 /* Pop a vlan tag either from hwaccel or from payload.
5555 * Expects skb->data at mac header.
5557 int skb_vlan_pop(struct sk_buff *skb)
5563 if (likely(skb_vlan_tag_present(skb))) {
5564 __vlan_hwaccel_clear_tag(skb);
5566 if (unlikely(!eth_type_vlan(skb->protocol)))
5569 err = __skb_vlan_pop(skb, &vlan_tci);
5573 /* move next vlan tag to hw accel tag */
5574 if (likely(!eth_type_vlan(skb->protocol)))
5577 vlan_proto = skb->protocol;
5578 err = __skb_vlan_pop(skb, &vlan_tci);
5582 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5585 EXPORT_SYMBOL(skb_vlan_pop);
5587 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5588 * Expects skb->data at mac header.
5590 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5592 if (skb_vlan_tag_present(skb)) {
5593 int offset = skb->data - skb_mac_header(skb);
5596 if (WARN_ONCE(offset,
5597 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5602 err = __vlan_insert_tag(skb, skb->vlan_proto,
5603 skb_vlan_tag_get(skb));
5607 skb->protocol = skb->vlan_proto;
5608 skb->mac_len += VLAN_HLEN;
5610 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5612 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5615 EXPORT_SYMBOL(skb_vlan_push);
5618 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5620 * @skb: Socket buffer to modify
5622 * Drop the Ethernet header of @skb.
5624 * Expects that skb->data points to the mac header and that no VLAN tags are
5627 * Returns 0 on success, -errno otherwise.
5629 int skb_eth_pop(struct sk_buff *skb)
5631 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5632 skb_network_offset(skb) < ETH_HLEN)
5635 skb_pull_rcsum(skb, ETH_HLEN);
5636 skb_reset_mac_header(skb);
5637 skb_reset_mac_len(skb);
5641 EXPORT_SYMBOL(skb_eth_pop);
5644 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5646 * @skb: Socket buffer to modify
5647 * @dst: Destination MAC address of the new header
5648 * @src: Source MAC address of the new header
5650 * Prepend @skb with a new Ethernet header.
5652 * Expects that skb->data points to the mac header, which must be empty.
5654 * Returns 0 on success, -errno otherwise.
5656 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5657 const unsigned char *src)
5662 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5665 err = skb_cow_head(skb, sizeof(*eth));
5669 skb_push(skb, sizeof(*eth));
5670 skb_reset_mac_header(skb);
5671 skb_reset_mac_len(skb);
5674 ether_addr_copy(eth->h_dest, dst);
5675 ether_addr_copy(eth->h_source, src);
5676 eth->h_proto = skb->protocol;
5678 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5682 EXPORT_SYMBOL(skb_eth_push);
5684 /* Update the ethertype of hdr and the skb csum value if required. */
5685 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5688 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5689 __be16 diff[] = { ~hdr->h_proto, ethertype };
5691 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5694 hdr->h_proto = ethertype;
5698 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5702 * @mpls_lse: MPLS label stack entry to push
5703 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5704 * @mac_len: length of the MAC header
5705 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5708 * Expects skb->data at mac header.
5710 * Returns 0 on success, -errno otherwise.
5712 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5713 int mac_len, bool ethernet)
5715 struct mpls_shim_hdr *lse;
5718 if (unlikely(!eth_p_mpls(mpls_proto)))
5721 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5722 if (skb->encapsulation)
5725 err = skb_cow_head(skb, MPLS_HLEN);
5729 if (!skb->inner_protocol) {
5730 skb_set_inner_network_header(skb, skb_network_offset(skb));
5731 skb_set_inner_protocol(skb, skb->protocol);
5734 skb_push(skb, MPLS_HLEN);
5735 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5737 skb_reset_mac_header(skb);
5738 skb_set_network_header(skb, mac_len);
5739 skb_reset_mac_len(skb);
5741 lse = mpls_hdr(skb);
5742 lse->label_stack_entry = mpls_lse;
5743 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5745 if (ethernet && mac_len >= ETH_HLEN)
5746 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5747 skb->protocol = mpls_proto;
5751 EXPORT_SYMBOL_GPL(skb_mpls_push);
5754 * skb_mpls_pop() - pop the outermost MPLS header
5757 * @next_proto: ethertype of header after popped MPLS header
5758 * @mac_len: length of the MAC header
5759 * @ethernet: flag to indicate if the packet is ethernet
5761 * Expects skb->data at mac header.
5763 * Returns 0 on success, -errno otherwise.
5765 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5770 if (unlikely(!eth_p_mpls(skb->protocol)))
5773 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5777 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5778 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5781 __skb_pull(skb, MPLS_HLEN);
5782 skb_reset_mac_header(skb);
5783 skb_set_network_header(skb, mac_len);
5785 if (ethernet && mac_len >= ETH_HLEN) {
5788 /* use mpls_hdr() to get ethertype to account for VLANs. */
5789 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5790 skb_mod_eth_type(skb, hdr, next_proto);
5792 skb->protocol = next_proto;
5796 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5799 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5802 * @mpls_lse: new MPLS label stack entry to update to
5804 * Expects skb->data at mac header.
5806 * Returns 0 on success, -errno otherwise.
5808 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5812 if (unlikely(!eth_p_mpls(skb->protocol)))
5815 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5819 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5820 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5822 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5825 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5829 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5832 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5836 * Expects skb->data at mac header.
5838 * Returns 0 on success, -errno otherwise.
5840 int skb_mpls_dec_ttl(struct sk_buff *skb)
5845 if (unlikely(!eth_p_mpls(skb->protocol)))
5848 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5851 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5852 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5856 lse &= ~MPLS_LS_TTL_MASK;
5857 lse |= ttl << MPLS_LS_TTL_SHIFT;
5859 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5861 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5864 * alloc_skb_with_frags - allocate skb with page frags
5866 * @header_len: size of linear part
5867 * @data_len: needed length in frags
5868 * @max_page_order: max page order desired.
5869 * @errcode: pointer to error code if any
5870 * @gfp_mask: allocation mask
5872 * This can be used to allocate a paged skb, given a maximal order for frags.
5874 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5875 unsigned long data_len,
5880 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5881 unsigned long chunk;
5882 struct sk_buff *skb;
5886 *errcode = -EMSGSIZE;
5887 /* Note this test could be relaxed, if we succeed to allocate
5888 * high order pages...
5890 if (npages > MAX_SKB_FRAGS)
5893 *errcode = -ENOBUFS;
5894 skb = alloc_skb(header_len, gfp_mask);
5898 skb->truesize += npages << PAGE_SHIFT;
5900 for (i = 0; npages > 0; i++) {
5901 int order = max_page_order;
5904 if (npages >= 1 << order) {
5905 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5911 /* Do not retry other high order allocations */
5917 page = alloc_page(gfp_mask);
5921 chunk = min_t(unsigned long, data_len,
5922 PAGE_SIZE << order);
5923 skb_fill_page_desc(skb, i, page, 0, chunk);
5925 npages -= 1 << order;
5933 EXPORT_SYMBOL(alloc_skb_with_frags);
5935 /* carve out the first off bytes from skb when off < headlen */
5936 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5937 const int headlen, gfp_t gfp_mask)
5940 int size = skb_end_offset(skb);
5941 int new_hlen = headlen - off;
5944 size = SKB_DATA_ALIGN(size);
5946 if (skb_pfmemalloc(skb))
5947 gfp_mask |= __GFP_MEMALLOC;
5948 data = kmalloc_reserve(size +
5949 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5950 gfp_mask, NUMA_NO_NODE, NULL);
5954 size = SKB_WITH_OVERHEAD(ksize(data));
5956 /* Copy real data, and all frags */
5957 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5960 memcpy((struct skb_shared_info *)(data + size),
5962 offsetof(struct skb_shared_info,
5963 frags[skb_shinfo(skb)->nr_frags]));
5964 if (skb_cloned(skb)) {
5965 /* drop the old head gracefully */
5966 if (skb_orphan_frags(skb, gfp_mask)) {
5970 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5971 skb_frag_ref(skb, i);
5972 if (skb_has_frag_list(skb))
5973 skb_clone_fraglist(skb);
5974 skb_release_data(skb);
5976 /* we can reuse existing recount- all we did was
5985 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5988 skb->end = skb->head + size;
5990 skb_set_tail_pointer(skb, skb_headlen(skb));
5991 skb_headers_offset_update(skb, 0);
5995 atomic_set(&skb_shinfo(skb)->dataref, 1);
6000 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6002 /* carve out the first eat bytes from skb's frag_list. May recurse into
6005 static int pskb_carve_frag_list(struct sk_buff *skb,
6006 struct skb_shared_info *shinfo, int eat,
6009 struct sk_buff *list = shinfo->frag_list;
6010 struct sk_buff *clone = NULL;
6011 struct sk_buff *insp = NULL;
6015 pr_err("Not enough bytes to eat. Want %d\n", eat);
6018 if (list->len <= eat) {
6019 /* Eaten as whole. */
6024 /* Eaten partially. */
6025 if (skb_shared(list)) {
6026 clone = skb_clone(list, gfp_mask);
6032 /* This may be pulled without problems. */
6035 if (pskb_carve(list, eat, gfp_mask) < 0) {
6043 /* Free pulled out fragments. */
6044 while ((list = shinfo->frag_list) != insp) {
6045 shinfo->frag_list = list->next;
6048 /* And insert new clone at head. */
6051 shinfo->frag_list = clone;
6056 /* carve off first len bytes from skb. Split line (off) is in the
6057 * non-linear part of skb
6059 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6060 int pos, gfp_t gfp_mask)
6063 int size = skb_end_offset(skb);
6065 const int nfrags = skb_shinfo(skb)->nr_frags;
6066 struct skb_shared_info *shinfo;
6068 size = SKB_DATA_ALIGN(size);
6070 if (skb_pfmemalloc(skb))
6071 gfp_mask |= __GFP_MEMALLOC;
6072 data = kmalloc_reserve(size +
6073 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6074 gfp_mask, NUMA_NO_NODE, NULL);
6078 size = SKB_WITH_OVERHEAD(ksize(data));
6080 memcpy((struct skb_shared_info *)(data + size),
6081 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6082 if (skb_orphan_frags(skb, gfp_mask)) {
6086 shinfo = (struct skb_shared_info *)(data + size);
6087 for (i = 0; i < nfrags; i++) {
6088 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6090 if (pos + fsize > off) {
6091 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6095 * We have two variants in this case:
6096 * 1. Move all the frag to the second
6097 * part, if it is possible. F.e.
6098 * this approach is mandatory for TUX,
6099 * where splitting is expensive.
6100 * 2. Split is accurately. We make this.
6102 skb_frag_off_add(&shinfo->frags[0], off - pos);
6103 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6105 skb_frag_ref(skb, i);
6110 shinfo->nr_frags = k;
6111 if (skb_has_frag_list(skb))
6112 skb_clone_fraglist(skb);
6114 /* split line is in frag list */
6115 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6116 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6117 if (skb_has_frag_list(skb))
6118 kfree_skb_list(skb_shinfo(skb)->frag_list);
6122 skb_release_data(skb);
6127 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6130 skb->end = skb->head + size;
6132 skb_reset_tail_pointer(skb);
6133 skb_headers_offset_update(skb, 0);
6138 skb->data_len = skb->len;
6139 atomic_set(&skb_shinfo(skb)->dataref, 1);
6143 /* remove len bytes from the beginning of the skb */
6144 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6146 int headlen = skb_headlen(skb);
6149 return pskb_carve_inside_header(skb, len, headlen, gfp);
6151 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6154 /* Extract to_copy bytes starting at off from skb, and return this in
6157 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6158 int to_copy, gfp_t gfp)
6160 struct sk_buff *clone = skb_clone(skb, gfp);
6165 if (pskb_carve(clone, off, gfp) < 0 ||
6166 pskb_trim(clone, to_copy)) {
6172 EXPORT_SYMBOL(pskb_extract);
6175 * skb_condense - try to get rid of fragments/frag_list if possible
6178 * Can be used to save memory before skb is added to a busy queue.
6179 * If packet has bytes in frags and enough tail room in skb->head,
6180 * pull all of them, so that we can free the frags right now and adjust
6183 * We do not reallocate skb->head thus can not fail.
6184 * Caller must re-evaluate skb->truesize if needed.
6186 void skb_condense(struct sk_buff *skb)
6188 if (skb->data_len) {
6189 if (skb->data_len > skb->end - skb->tail ||
6193 /* Nice, we can free page frag(s) right now */
6194 __pskb_pull_tail(skb, skb->data_len);
6196 /* At this point, skb->truesize might be over estimated,
6197 * because skb had a fragment, and fragments do not tell
6199 * When we pulled its content into skb->head, fragment
6200 * was freed, but __pskb_pull_tail() could not possibly
6201 * adjust skb->truesize, not knowing the frag truesize.
6203 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6206 #ifdef CONFIG_SKB_EXTENSIONS
6207 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6209 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6213 * __skb_ext_alloc - allocate a new skb extensions storage
6215 * @flags: See kmalloc().
6217 * Returns the newly allocated pointer. The pointer can later attached to a
6218 * skb via __skb_ext_set().
6219 * Note: caller must handle the skb_ext as an opaque data.
6221 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6223 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6226 memset(new->offset, 0, sizeof(new->offset));
6227 refcount_set(&new->refcnt, 1);
6233 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6234 unsigned int old_active)
6236 struct skb_ext *new;
6238 if (refcount_read(&old->refcnt) == 1)
6241 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6245 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6246 refcount_set(&new->refcnt, 1);
6249 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6250 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6253 for (i = 0; i < sp->len; i++)
6254 xfrm_state_hold(sp->xvec[i]);
6262 * __skb_ext_set - attach the specified extension storage to this skb
6265 * @ext: extension storage previously allocated via __skb_ext_alloc()
6267 * Existing extensions, if any, are cleared.
6269 * Returns the pointer to the extension.
6271 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6272 struct skb_ext *ext)
6274 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6277 newlen = newoff + skb_ext_type_len[id];
6278 ext->chunks = newlen;
6279 ext->offset[id] = newoff;
6280 skb->extensions = ext;
6281 skb->active_extensions = 1 << id;
6282 return skb_ext_get_ptr(ext, id);
6286 * skb_ext_add - allocate space for given extension, COW if needed
6288 * @id: extension to allocate space for
6290 * Allocates enough space for the given extension.
6291 * If the extension is already present, a pointer to that extension
6294 * If the skb was cloned, COW applies and the returned memory can be
6295 * modified without changing the extension space of clones buffers.
6297 * Returns pointer to the extension or NULL on allocation failure.
6299 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6301 struct skb_ext *new, *old = NULL;
6302 unsigned int newlen, newoff;
6304 if (skb->active_extensions) {
6305 old = skb->extensions;
6307 new = skb_ext_maybe_cow(old, skb->active_extensions);
6311 if (__skb_ext_exist(new, id))
6314 newoff = new->chunks;
6316 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6318 new = __skb_ext_alloc(GFP_ATOMIC);
6323 newlen = newoff + skb_ext_type_len[id];
6324 new->chunks = newlen;
6325 new->offset[id] = newoff;
6327 skb->extensions = new;
6328 skb->active_extensions |= 1 << id;
6329 return skb_ext_get_ptr(new, id);
6331 EXPORT_SYMBOL(skb_ext_add);
6334 static void skb_ext_put_sp(struct sec_path *sp)
6338 for (i = 0; i < sp->len; i++)
6339 xfrm_state_put(sp->xvec[i]);
6343 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6345 struct skb_ext *ext = skb->extensions;
6347 skb->active_extensions &= ~(1 << id);
6348 if (skb->active_extensions == 0) {
6349 skb->extensions = NULL;
6352 } else if (id == SKB_EXT_SEC_PATH &&
6353 refcount_read(&ext->refcnt) == 1) {
6354 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6361 EXPORT_SYMBOL(__skb_ext_del);
6363 void __skb_ext_put(struct skb_ext *ext)
6365 /* If this is last clone, nothing can increment
6366 * it after check passes. Avoids one atomic op.
6368 if (refcount_read(&ext->refcnt) == 1)
6371 if (!refcount_dec_and_test(&ext->refcnt))
6375 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6376 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6379 kmem_cache_free(skbuff_ext_cache, ext);
6381 EXPORT_SYMBOL(__skb_ext_put);
6382 #endif /* CONFIG_SKB_EXTENSIONS */