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 skb_zcopy_clear(skb, true);
619 for (i = 0; i < shinfo->nr_frags; i++)
620 __skb_frag_unref(&shinfo->frags[i]);
622 if (shinfo->frag_list)
623 kfree_skb_list(shinfo->frag_list);
629 * Free an skbuff by memory without cleaning the state.
631 static void kfree_skbmem(struct sk_buff *skb)
633 struct sk_buff_fclones *fclones;
635 switch (skb->fclone) {
636 case SKB_FCLONE_UNAVAILABLE:
637 kmem_cache_free(skbuff_head_cache, skb);
640 case SKB_FCLONE_ORIG:
641 fclones = container_of(skb, struct sk_buff_fclones, skb1);
643 /* We usually free the clone (TX completion) before original skb
644 * This test would have no chance to be true for the clone,
645 * while here, branch prediction will be good.
647 if (refcount_read(&fclones->fclone_ref) == 1)
651 default: /* SKB_FCLONE_CLONE */
652 fclones = container_of(skb, struct sk_buff_fclones, skb2);
655 if (!refcount_dec_and_test(&fclones->fclone_ref))
658 kmem_cache_free(skbuff_fclone_cache, fclones);
661 void skb_release_head_state(struct sk_buff *skb)
664 if (skb->destructor) {
666 skb->destructor(skb);
668 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
669 nf_conntrack_put(skb_nfct(skb));
674 /* Free everything but the sk_buff shell. */
675 static void skb_release_all(struct sk_buff *skb)
677 skb_release_head_state(skb);
678 if (likely(skb->head))
679 skb_release_data(skb);
683 * __kfree_skb - private function
686 * Free an sk_buff. Release anything attached to the buffer.
687 * Clean the state. This is an internal helper function. Users should
688 * always call kfree_skb
691 void __kfree_skb(struct sk_buff *skb)
693 skb_release_all(skb);
696 EXPORT_SYMBOL(__kfree_skb);
699 * kfree_skb - free an sk_buff
700 * @skb: buffer to free
702 * Drop a reference to the buffer and free it if the usage count has
705 void kfree_skb(struct sk_buff *skb)
710 trace_kfree_skb(skb, __builtin_return_address(0));
713 EXPORT_SYMBOL(kfree_skb);
715 void kfree_skb_list(struct sk_buff *segs)
718 struct sk_buff *next = segs->next;
724 EXPORT_SYMBOL(kfree_skb_list);
726 /* Dump skb information and contents.
728 * Must only be called from net_ratelimit()-ed paths.
730 * Dumps whole packets if full_pkt, only headers otherwise.
732 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
734 struct skb_shared_info *sh = skb_shinfo(skb);
735 struct net_device *dev = skb->dev;
736 struct sock *sk = skb->sk;
737 struct sk_buff *list_skb;
738 bool has_mac, has_trans;
739 int headroom, tailroom;
745 len = min_t(int, skb->len, MAX_HEADER + 128);
747 headroom = skb_headroom(skb);
748 tailroom = skb_tailroom(skb);
750 has_mac = skb_mac_header_was_set(skb);
751 has_trans = skb_transport_header_was_set(skb);
753 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
754 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
755 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
756 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
757 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
758 level, skb->len, headroom, skb_headlen(skb), tailroom,
759 has_mac ? skb->mac_header : -1,
760 has_mac ? skb_mac_header_len(skb) : -1,
762 has_trans ? skb_network_header_len(skb) : -1,
763 has_trans ? skb->transport_header : -1,
764 sh->tx_flags, sh->nr_frags,
765 sh->gso_size, sh->gso_type, sh->gso_segs,
766 skb->csum, skb->ip_summed, skb->csum_complete_sw,
767 skb->csum_valid, skb->csum_level,
768 skb->hash, skb->sw_hash, skb->l4_hash,
769 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
772 printk("%sdev name=%s feat=0x%pNF\n",
773 level, dev->name, &dev->features);
775 printk("%ssk family=%hu type=%u proto=%u\n",
776 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
778 if (full_pkt && headroom)
779 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
780 16, 1, skb->head, headroom, false);
782 seg_len = min_t(int, skb_headlen(skb), len);
784 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
785 16, 1, skb->data, seg_len, false);
788 if (full_pkt && tailroom)
789 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
790 16, 1, skb_tail_pointer(skb), tailroom, false);
792 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
793 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
794 u32 p_off, p_len, copied;
798 skb_frag_foreach_page(frag, skb_frag_off(frag),
799 skb_frag_size(frag), p, p_off, p_len,
801 seg_len = min_t(int, p_len, len);
802 vaddr = kmap_atomic(p);
803 print_hex_dump(level, "skb frag: ",
805 16, 1, vaddr + p_off, seg_len, false);
806 kunmap_atomic(vaddr);
813 if (full_pkt && skb_has_frag_list(skb)) {
814 printk("skb fraglist:\n");
815 skb_walk_frags(skb, list_skb)
816 skb_dump(level, list_skb, true);
819 EXPORT_SYMBOL(skb_dump);
822 * skb_tx_error - report an sk_buff xmit error
823 * @skb: buffer that triggered an error
825 * Report xmit error if a device callback is tracking this skb.
826 * skb must be freed afterwards.
828 void skb_tx_error(struct sk_buff *skb)
830 skb_zcopy_clear(skb, true);
832 EXPORT_SYMBOL(skb_tx_error);
834 #ifdef CONFIG_TRACEPOINTS
836 * consume_skb - free an skbuff
837 * @skb: buffer to free
839 * Drop a ref to the buffer and free it if the usage count has hit zero
840 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
841 * is being dropped after a failure and notes that
843 void consume_skb(struct sk_buff *skb)
848 trace_consume_skb(skb);
851 EXPORT_SYMBOL(consume_skb);
855 * __consume_stateless_skb - free an skbuff, assuming it is stateless
856 * @skb: buffer to free
858 * Alike consume_skb(), but this variant assumes that this is the last
859 * skb reference and all the head states have been already dropped
861 void __consume_stateless_skb(struct sk_buff *skb)
863 trace_consume_skb(skb);
864 skb_release_data(skb);
868 void __kfree_skb_flush(void)
870 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
872 /* flush skb_cache if containing objects */
874 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
880 static inline void _kfree_skb_defer(struct sk_buff *skb)
882 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
884 /* drop skb->head and call any destructors for packet */
885 skb_release_all(skb);
887 /* record skb to CPU local list */
888 nc->skb_cache[nc->skb_count++] = skb;
891 /* SLUB writes into objects when freeing */
895 /* flush skb_cache if it is filled */
896 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
897 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
902 void __kfree_skb_defer(struct sk_buff *skb)
904 _kfree_skb_defer(skb);
907 void napi_consume_skb(struct sk_buff *skb, int budget)
909 /* Zero budget indicate non-NAPI context called us, like netpoll */
910 if (unlikely(!budget)) {
911 dev_consume_skb_any(skb);
915 lockdep_assert_in_softirq();
920 /* if reaching here SKB is ready to free */
921 trace_consume_skb(skb);
923 /* if SKB is a clone, don't handle this case */
924 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
929 _kfree_skb_defer(skb);
931 EXPORT_SYMBOL(napi_consume_skb);
933 /* Make sure a field is enclosed inside headers_start/headers_end section */
934 #define CHECK_SKB_FIELD(field) \
935 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
936 offsetof(struct sk_buff, headers_start)); \
937 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
938 offsetof(struct sk_buff, headers_end)); \
940 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
942 new->tstamp = old->tstamp;
943 /* We do not copy old->sk */
945 memcpy(new->cb, old->cb, sizeof(old->cb));
946 skb_dst_copy(new, old);
947 __skb_ext_copy(new, old);
948 __nf_copy(new, old, false);
950 /* Note : this field could be in headers_start/headers_end section
951 * It is not yet because we do not want to have a 16 bit hole
953 new->queue_mapping = old->queue_mapping;
955 memcpy(&new->headers_start, &old->headers_start,
956 offsetof(struct sk_buff, headers_end) -
957 offsetof(struct sk_buff, headers_start));
958 CHECK_SKB_FIELD(protocol);
959 CHECK_SKB_FIELD(csum);
960 CHECK_SKB_FIELD(hash);
961 CHECK_SKB_FIELD(priority);
962 CHECK_SKB_FIELD(skb_iif);
963 CHECK_SKB_FIELD(vlan_proto);
964 CHECK_SKB_FIELD(vlan_tci);
965 CHECK_SKB_FIELD(transport_header);
966 CHECK_SKB_FIELD(network_header);
967 CHECK_SKB_FIELD(mac_header);
968 CHECK_SKB_FIELD(inner_protocol);
969 CHECK_SKB_FIELD(inner_transport_header);
970 CHECK_SKB_FIELD(inner_network_header);
971 CHECK_SKB_FIELD(inner_mac_header);
972 CHECK_SKB_FIELD(mark);
973 #ifdef CONFIG_NETWORK_SECMARK
974 CHECK_SKB_FIELD(secmark);
976 #ifdef CONFIG_NET_RX_BUSY_POLL
977 CHECK_SKB_FIELD(napi_id);
980 CHECK_SKB_FIELD(sender_cpu);
982 #ifdef CONFIG_NET_SCHED
983 CHECK_SKB_FIELD(tc_index);
989 * You should not add any new code to this function. Add it to
990 * __copy_skb_header above instead.
992 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
994 #define C(x) n->x = skb->x
996 n->next = n->prev = NULL;
998 __copy_skb_header(n, skb);
1003 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1008 n->destructor = NULL;
1015 refcount_set(&n->users, 1);
1017 atomic_inc(&(skb_shinfo(skb)->dataref));
1025 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1026 * @first: first sk_buff of the msg
1028 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1032 n = alloc_skb(0, GFP_ATOMIC);
1036 n->len = first->len;
1037 n->data_len = first->len;
1038 n->truesize = first->truesize;
1040 skb_shinfo(n)->frag_list = first;
1042 __copy_skb_header(n, first);
1043 n->destructor = NULL;
1047 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1050 * skb_morph - morph one skb into another
1051 * @dst: the skb to receive the contents
1052 * @src: the skb to supply the contents
1054 * This is identical to skb_clone except that the target skb is
1055 * supplied by the user.
1057 * The target skb is returned upon exit.
1059 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1061 skb_release_all(dst);
1062 return __skb_clone(dst, src);
1064 EXPORT_SYMBOL_GPL(skb_morph);
1066 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1068 unsigned long max_pg, num_pg, new_pg, old_pg;
1069 struct user_struct *user;
1071 if (capable(CAP_IPC_LOCK) || !size)
1074 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1075 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1076 user = mmp->user ? : current_user();
1079 old_pg = atomic_long_read(&user->locked_vm);
1080 new_pg = old_pg + num_pg;
1081 if (new_pg > max_pg)
1083 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1087 mmp->user = get_uid(user);
1088 mmp->num_pg = num_pg;
1090 mmp->num_pg += num_pg;
1095 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1097 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1100 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1101 free_uid(mmp->user);
1104 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1106 struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1108 struct ubuf_info *uarg;
1109 struct sk_buff *skb;
1111 WARN_ON_ONCE(!in_task());
1113 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1117 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1118 uarg = (void *)skb->cb;
1119 uarg->mmp.user = NULL;
1121 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1126 uarg->callback = msg_zerocopy_callback;
1127 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1129 uarg->bytelen = size;
1131 uarg->flags = SKBFL_ZEROCOPY_FRAG;
1132 refcount_set(&uarg->refcnt, 1);
1137 EXPORT_SYMBOL_GPL(msg_zerocopy_alloc);
1139 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1141 return container_of((void *)uarg, struct sk_buff, cb);
1144 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1145 struct ubuf_info *uarg)
1148 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1151 /* realloc only when socket is locked (TCP, UDP cork),
1152 * so uarg->len and sk_zckey access is serialized
1154 if (!sock_owned_by_user(sk)) {
1159 bytelen = uarg->bytelen + size;
1160 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1161 /* TCP can create new skb to attach new uarg */
1162 if (sk->sk_type == SOCK_STREAM)
1167 next = (u32)atomic_read(&sk->sk_zckey);
1168 if ((u32)(uarg->id + uarg->len) == next) {
1169 if (mm_account_pinned_pages(&uarg->mmp, size))
1172 uarg->bytelen = bytelen;
1173 atomic_set(&sk->sk_zckey, ++next);
1175 /* no extra ref when appending to datagram (MSG_MORE) */
1176 if (sk->sk_type == SOCK_STREAM)
1177 net_zcopy_get(uarg);
1184 return msg_zerocopy_alloc(sk, size);
1186 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1188 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1190 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1194 old_lo = serr->ee.ee_info;
1195 old_hi = serr->ee.ee_data;
1196 sum_len = old_hi - old_lo + 1ULL + len;
1198 if (sum_len >= (1ULL << 32))
1201 if (lo != old_hi + 1)
1204 serr->ee.ee_data += len;
1208 static void __msg_zerocopy_callback(struct ubuf_info *uarg)
1210 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1211 struct sock_exterr_skb *serr;
1212 struct sock *sk = skb->sk;
1213 struct sk_buff_head *q;
1214 unsigned long flags;
1218 mm_unaccount_pinned_pages(&uarg->mmp);
1220 /* if !len, there was only 1 call, and it was aborted
1221 * so do not queue a completion notification
1223 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1228 hi = uarg->id + len - 1;
1230 serr = SKB_EXT_ERR(skb);
1231 memset(serr, 0, sizeof(*serr));
1232 serr->ee.ee_errno = 0;
1233 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1234 serr->ee.ee_data = hi;
1235 serr->ee.ee_info = lo;
1236 if (!uarg->zerocopy)
1237 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1239 q = &sk->sk_error_queue;
1240 spin_lock_irqsave(&q->lock, flags);
1241 tail = skb_peek_tail(q);
1242 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1243 !skb_zerocopy_notify_extend(tail, lo, len)) {
1244 __skb_queue_tail(q, skb);
1247 spin_unlock_irqrestore(&q->lock, flags);
1249 sk->sk_error_report(sk);
1256 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1259 uarg->zerocopy = uarg->zerocopy & success;
1261 if (refcount_dec_and_test(&uarg->refcnt))
1262 __msg_zerocopy_callback(uarg);
1264 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1266 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1268 struct sock *sk = skb_from_uarg(uarg)->sk;
1270 atomic_dec(&sk->sk_zckey);
1274 msg_zerocopy_callback(NULL, uarg, true);
1276 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1278 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1280 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1282 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1284 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1285 struct msghdr *msg, int len,
1286 struct ubuf_info *uarg)
1288 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1289 struct iov_iter orig_iter = msg->msg_iter;
1290 int err, orig_len = skb->len;
1292 /* An skb can only point to one uarg. This edge case happens when
1293 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1295 if (orig_uarg && uarg != orig_uarg)
1298 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1299 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1300 struct sock *save_sk = skb->sk;
1302 /* Streams do not free skb on error. Reset to prev state. */
1303 msg->msg_iter = orig_iter;
1305 ___pskb_trim(skb, orig_len);
1310 skb_zcopy_set(skb, uarg, NULL);
1311 return skb->len - orig_len;
1313 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1315 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1318 if (skb_zcopy(orig)) {
1319 if (skb_zcopy(nskb)) {
1320 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1325 if (skb_uarg(nskb) == skb_uarg(orig))
1327 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1330 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1336 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1337 * @skb: the skb to modify
1338 * @gfp_mask: allocation priority
1340 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1341 * It will copy all frags into kernel and drop the reference
1342 * to userspace pages.
1344 * If this function is called from an interrupt gfp_mask() must be
1347 * Returns 0 on success or a negative error code on failure
1348 * to allocate kernel memory to copy to.
1350 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1352 int num_frags = skb_shinfo(skb)->nr_frags;
1353 struct page *page, *head = NULL;
1357 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1363 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1364 for (i = 0; i < new_frags; i++) {
1365 page = alloc_page(gfp_mask);
1368 struct page *next = (struct page *)page_private(head);
1374 set_page_private(page, (unsigned long)head);
1380 for (i = 0; i < num_frags; i++) {
1381 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1382 u32 p_off, p_len, copied;
1386 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1387 p, p_off, p_len, copied) {
1389 vaddr = kmap_atomic(p);
1391 while (done < p_len) {
1392 if (d_off == PAGE_SIZE) {
1394 page = (struct page *)page_private(page);
1396 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1397 memcpy(page_address(page) + d_off,
1398 vaddr + p_off + done, copy);
1402 kunmap_atomic(vaddr);
1406 /* skb frags release userspace buffers */
1407 for (i = 0; i < num_frags; i++)
1408 skb_frag_unref(skb, i);
1410 /* skb frags point to kernel buffers */
1411 for (i = 0; i < new_frags - 1; i++) {
1412 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1413 head = (struct page *)page_private(head);
1415 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1416 skb_shinfo(skb)->nr_frags = new_frags;
1419 skb_zcopy_clear(skb, false);
1422 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1425 * skb_clone - duplicate an sk_buff
1426 * @skb: buffer to clone
1427 * @gfp_mask: allocation priority
1429 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1430 * copies share the same packet data but not structure. The new
1431 * buffer has a reference count of 1. If the allocation fails the
1432 * function returns %NULL otherwise the new buffer is returned.
1434 * If this function is called from an interrupt gfp_mask() must be
1438 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1440 struct sk_buff_fclones *fclones = container_of(skb,
1441 struct sk_buff_fclones,
1445 if (skb_orphan_frags(skb, gfp_mask))
1448 if (skb->fclone == SKB_FCLONE_ORIG &&
1449 refcount_read(&fclones->fclone_ref) == 1) {
1451 refcount_set(&fclones->fclone_ref, 2);
1453 if (skb_pfmemalloc(skb))
1454 gfp_mask |= __GFP_MEMALLOC;
1456 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1460 n->fclone = SKB_FCLONE_UNAVAILABLE;
1463 return __skb_clone(n, skb);
1465 EXPORT_SYMBOL(skb_clone);
1467 void skb_headers_offset_update(struct sk_buff *skb, int off)
1469 /* Only adjust this if it actually is csum_start rather than csum */
1470 if (skb->ip_summed == CHECKSUM_PARTIAL)
1471 skb->csum_start += off;
1472 /* {transport,network,mac}_header and tail are relative to skb->head */
1473 skb->transport_header += off;
1474 skb->network_header += off;
1475 if (skb_mac_header_was_set(skb))
1476 skb->mac_header += off;
1477 skb->inner_transport_header += off;
1478 skb->inner_network_header += off;
1479 skb->inner_mac_header += off;
1481 EXPORT_SYMBOL(skb_headers_offset_update);
1483 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1485 __copy_skb_header(new, old);
1487 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1488 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1489 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1491 EXPORT_SYMBOL(skb_copy_header);
1493 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1495 if (skb_pfmemalloc(skb))
1496 return SKB_ALLOC_RX;
1501 * skb_copy - create private copy of an sk_buff
1502 * @skb: buffer to copy
1503 * @gfp_mask: allocation priority
1505 * Make a copy of both an &sk_buff and its data. This is used when the
1506 * caller wishes to modify the data and needs a private copy of the
1507 * data to alter. Returns %NULL on failure or the pointer to the buffer
1508 * on success. The returned buffer has a reference count of 1.
1510 * As by-product this function converts non-linear &sk_buff to linear
1511 * one, so that &sk_buff becomes completely private and caller is allowed
1512 * to modify all the data of returned buffer. This means that this
1513 * function is not recommended for use in circumstances when only
1514 * header is going to be modified. Use pskb_copy() instead.
1517 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1519 int headerlen = skb_headroom(skb);
1520 unsigned int size = skb_end_offset(skb) + skb->data_len;
1521 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1522 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1527 /* Set the data pointer */
1528 skb_reserve(n, headerlen);
1529 /* Set the tail pointer and length */
1530 skb_put(n, skb->len);
1532 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1534 skb_copy_header(n, skb);
1537 EXPORT_SYMBOL(skb_copy);
1540 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1541 * @skb: buffer to copy
1542 * @headroom: headroom of new skb
1543 * @gfp_mask: allocation priority
1544 * @fclone: if true allocate the copy of the skb from the fclone
1545 * cache instead of the head cache; it is recommended to set this
1546 * to true for the cases where the copy will likely be cloned
1548 * Make a copy of both an &sk_buff and part of its data, located
1549 * in header. Fragmented data remain shared. This is used when
1550 * the caller wishes to modify only header of &sk_buff and needs
1551 * private copy of the header to alter. Returns %NULL on failure
1552 * or the pointer to the buffer on success.
1553 * The returned buffer has a reference count of 1.
1556 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1557 gfp_t gfp_mask, bool fclone)
1559 unsigned int size = skb_headlen(skb) + headroom;
1560 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1561 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1566 /* Set the data pointer */
1567 skb_reserve(n, headroom);
1568 /* Set the tail pointer and length */
1569 skb_put(n, skb_headlen(skb));
1570 /* Copy the bytes */
1571 skb_copy_from_linear_data(skb, n->data, n->len);
1573 n->truesize += skb->data_len;
1574 n->data_len = skb->data_len;
1577 if (skb_shinfo(skb)->nr_frags) {
1580 if (skb_orphan_frags(skb, gfp_mask) ||
1581 skb_zerocopy_clone(n, skb, gfp_mask)) {
1586 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1587 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1588 skb_frag_ref(skb, i);
1590 skb_shinfo(n)->nr_frags = i;
1593 if (skb_has_frag_list(skb)) {
1594 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1595 skb_clone_fraglist(n);
1598 skb_copy_header(n, skb);
1602 EXPORT_SYMBOL(__pskb_copy_fclone);
1605 * pskb_expand_head - reallocate header of &sk_buff
1606 * @skb: buffer to reallocate
1607 * @nhead: room to add at head
1608 * @ntail: room to add at tail
1609 * @gfp_mask: allocation priority
1611 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1612 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1613 * reference count of 1. Returns zero in the case of success or error,
1614 * if expansion failed. In the last case, &sk_buff is not changed.
1616 * All the pointers pointing into skb header may change and must be
1617 * reloaded after call to this function.
1620 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1623 int i, osize = skb_end_offset(skb);
1624 int size = osize + nhead + ntail;
1630 BUG_ON(skb_shared(skb));
1632 size = SKB_DATA_ALIGN(size);
1634 if (skb_pfmemalloc(skb))
1635 gfp_mask |= __GFP_MEMALLOC;
1636 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1637 gfp_mask, NUMA_NO_NODE, NULL);
1640 size = SKB_WITH_OVERHEAD(ksize(data));
1642 /* Copy only real data... and, alas, header. This should be
1643 * optimized for the cases when header is void.
1645 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1647 memcpy((struct skb_shared_info *)(data + size),
1649 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1652 * if shinfo is shared we must drop the old head gracefully, but if it
1653 * is not we can just drop the old head and let the existing refcount
1654 * be since all we did is relocate the values
1656 if (skb_cloned(skb)) {
1657 if (skb_orphan_frags(skb, gfp_mask))
1660 refcount_inc(&skb_uarg(skb)->refcnt);
1661 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1662 skb_frag_ref(skb, i);
1664 if (skb_has_frag_list(skb))
1665 skb_clone_fraglist(skb);
1667 skb_release_data(skb);
1671 off = (data + nhead) - skb->head;
1676 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1680 skb->end = skb->head + size;
1683 skb_headers_offset_update(skb, nhead);
1687 atomic_set(&skb_shinfo(skb)->dataref, 1);
1689 skb_metadata_clear(skb);
1691 /* It is not generally safe to change skb->truesize.
1692 * For the moment, we really care of rx path, or
1693 * when skb is orphaned (not attached to a socket).
1695 if (!skb->sk || skb->destructor == sock_edemux)
1696 skb->truesize += size - osize;
1705 EXPORT_SYMBOL(pskb_expand_head);
1707 /* Make private copy of skb with writable head and some headroom */
1709 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1711 struct sk_buff *skb2;
1712 int delta = headroom - skb_headroom(skb);
1715 skb2 = pskb_copy(skb, GFP_ATOMIC);
1717 skb2 = skb_clone(skb, GFP_ATOMIC);
1718 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1726 EXPORT_SYMBOL(skb_realloc_headroom);
1729 * skb_copy_expand - copy and expand sk_buff
1730 * @skb: buffer to copy
1731 * @newheadroom: new free bytes at head
1732 * @newtailroom: new free bytes at tail
1733 * @gfp_mask: allocation priority
1735 * Make a copy of both an &sk_buff and its data and while doing so
1736 * allocate additional space.
1738 * This is used when the caller wishes to modify the data and needs a
1739 * private copy of the data to alter as well as more space for new fields.
1740 * Returns %NULL on failure or the pointer to the buffer
1741 * on success. The returned buffer has a reference count of 1.
1743 * You must pass %GFP_ATOMIC as the allocation priority if this function
1744 * is called from an interrupt.
1746 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1747 int newheadroom, int newtailroom,
1751 * Allocate the copy buffer
1753 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1754 gfp_mask, skb_alloc_rx_flag(skb),
1756 int oldheadroom = skb_headroom(skb);
1757 int head_copy_len, head_copy_off;
1762 skb_reserve(n, newheadroom);
1764 /* Set the tail pointer and length */
1765 skb_put(n, skb->len);
1767 head_copy_len = oldheadroom;
1769 if (newheadroom <= head_copy_len)
1770 head_copy_len = newheadroom;
1772 head_copy_off = newheadroom - head_copy_len;
1774 /* Copy the linear header and data. */
1775 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1776 skb->len + head_copy_len));
1778 skb_copy_header(n, skb);
1780 skb_headers_offset_update(n, newheadroom - oldheadroom);
1784 EXPORT_SYMBOL(skb_copy_expand);
1787 * __skb_pad - zero pad the tail of an skb
1788 * @skb: buffer to pad
1789 * @pad: space to pad
1790 * @free_on_error: free buffer on error
1792 * Ensure that a buffer is followed by a padding area that is zero
1793 * filled. Used by network drivers which may DMA or transfer data
1794 * beyond the buffer end onto the wire.
1796 * May return error in out of memory cases. The skb is freed on error
1797 * if @free_on_error is true.
1800 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1805 /* If the skbuff is non linear tailroom is always zero.. */
1806 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1807 memset(skb->data+skb->len, 0, pad);
1811 ntail = skb->data_len + pad - (skb->end - skb->tail);
1812 if (likely(skb_cloned(skb) || ntail > 0)) {
1813 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1818 /* FIXME: The use of this function with non-linear skb's really needs
1821 err = skb_linearize(skb);
1825 memset(skb->data + skb->len, 0, pad);
1833 EXPORT_SYMBOL(__skb_pad);
1836 * pskb_put - add data to the tail of a potentially fragmented buffer
1837 * @skb: start of the buffer to use
1838 * @tail: tail fragment of the buffer to use
1839 * @len: amount of data to add
1841 * This function extends the used data area of the potentially
1842 * fragmented buffer. @tail must be the last fragment of @skb -- or
1843 * @skb itself. If this would exceed the total buffer size the kernel
1844 * will panic. A pointer to the first byte of the extra data is
1848 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1851 skb->data_len += len;
1854 return skb_put(tail, len);
1856 EXPORT_SYMBOL_GPL(pskb_put);
1859 * skb_put - add data to a buffer
1860 * @skb: buffer to use
1861 * @len: amount of data to add
1863 * This function extends the used data area of the buffer. If this would
1864 * exceed the total buffer size the kernel will panic. A pointer to the
1865 * first byte of the extra data is returned.
1867 void *skb_put(struct sk_buff *skb, unsigned int len)
1869 void *tmp = skb_tail_pointer(skb);
1870 SKB_LINEAR_ASSERT(skb);
1873 if (unlikely(skb->tail > skb->end))
1874 skb_over_panic(skb, len, __builtin_return_address(0));
1877 EXPORT_SYMBOL(skb_put);
1880 * skb_push - add data to the start of a buffer
1881 * @skb: buffer to use
1882 * @len: amount of data to add
1884 * This function extends the used data area of the buffer at the buffer
1885 * start. If this would exceed the total buffer headroom the kernel will
1886 * panic. A pointer to the first byte of the extra data is returned.
1888 void *skb_push(struct sk_buff *skb, unsigned int len)
1892 if (unlikely(skb->data < skb->head))
1893 skb_under_panic(skb, len, __builtin_return_address(0));
1896 EXPORT_SYMBOL(skb_push);
1899 * skb_pull - remove data from the start of a buffer
1900 * @skb: buffer to use
1901 * @len: amount of data to remove
1903 * This function removes data from the start of a buffer, returning
1904 * the memory to the headroom. A pointer to the next data in the buffer
1905 * is returned. Once the data has been pulled future pushes will overwrite
1908 void *skb_pull(struct sk_buff *skb, unsigned int len)
1910 return skb_pull_inline(skb, len);
1912 EXPORT_SYMBOL(skb_pull);
1915 * skb_trim - remove end from a buffer
1916 * @skb: buffer to alter
1919 * Cut the length of a buffer down by removing data from the tail. If
1920 * the buffer is already under the length specified it is not modified.
1921 * The skb must be linear.
1923 void skb_trim(struct sk_buff *skb, unsigned int len)
1926 __skb_trim(skb, len);
1928 EXPORT_SYMBOL(skb_trim);
1930 /* Trims skb to length len. It can change skb pointers.
1933 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1935 struct sk_buff **fragp;
1936 struct sk_buff *frag;
1937 int offset = skb_headlen(skb);
1938 int nfrags = skb_shinfo(skb)->nr_frags;
1942 if (skb_cloned(skb) &&
1943 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1950 for (; i < nfrags; i++) {
1951 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1958 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1961 skb_shinfo(skb)->nr_frags = i;
1963 for (; i < nfrags; i++)
1964 skb_frag_unref(skb, i);
1966 if (skb_has_frag_list(skb))
1967 skb_drop_fraglist(skb);
1971 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1972 fragp = &frag->next) {
1973 int end = offset + frag->len;
1975 if (skb_shared(frag)) {
1976 struct sk_buff *nfrag;
1978 nfrag = skb_clone(frag, GFP_ATOMIC);
1979 if (unlikely(!nfrag))
1982 nfrag->next = frag->next;
1994 unlikely((err = pskb_trim(frag, len - offset))))
1998 skb_drop_list(&frag->next);
2003 if (len > skb_headlen(skb)) {
2004 skb->data_len -= skb->len - len;
2009 skb_set_tail_pointer(skb, len);
2012 if (!skb->sk || skb->destructor == sock_edemux)
2016 EXPORT_SYMBOL(___pskb_trim);
2018 /* Note : use pskb_trim_rcsum() instead of calling this directly
2020 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2022 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2023 int delta = skb->len - len;
2025 skb->csum = csum_block_sub(skb->csum,
2026 skb_checksum(skb, len, delta, 0),
2028 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2029 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2030 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2032 if (offset + sizeof(__sum16) > hdlen)
2035 return __pskb_trim(skb, len);
2037 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2040 * __pskb_pull_tail - advance tail of skb header
2041 * @skb: buffer to reallocate
2042 * @delta: number of bytes to advance tail
2044 * The function makes a sense only on a fragmented &sk_buff,
2045 * it expands header moving its tail forward and copying necessary
2046 * data from fragmented part.
2048 * &sk_buff MUST have reference count of 1.
2050 * Returns %NULL (and &sk_buff does not change) if pull failed
2051 * or value of new tail of skb in the case of success.
2053 * All the pointers pointing into skb header may change and must be
2054 * reloaded after call to this function.
2057 /* Moves tail of skb head forward, copying data from fragmented part,
2058 * when it is necessary.
2059 * 1. It may fail due to malloc failure.
2060 * 2. It may change skb pointers.
2062 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2064 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2066 /* If skb has not enough free space at tail, get new one
2067 * plus 128 bytes for future expansions. If we have enough
2068 * room at tail, reallocate without expansion only if skb is cloned.
2070 int i, k, eat = (skb->tail + delta) - skb->end;
2072 if (eat > 0 || skb_cloned(skb)) {
2073 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2078 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2079 skb_tail_pointer(skb), delta));
2081 /* Optimization: no fragments, no reasons to preestimate
2082 * size of pulled pages. Superb.
2084 if (!skb_has_frag_list(skb))
2087 /* Estimate size of pulled pages. */
2089 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2090 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2097 /* If we need update frag list, we are in troubles.
2098 * Certainly, it is possible to add an offset to skb data,
2099 * but taking into account that pulling is expected to
2100 * be very rare operation, it is worth to fight against
2101 * further bloating skb head and crucify ourselves here instead.
2102 * Pure masohism, indeed. 8)8)
2105 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2106 struct sk_buff *clone = NULL;
2107 struct sk_buff *insp = NULL;
2110 if (list->len <= eat) {
2111 /* Eaten as whole. */
2116 /* Eaten partially. */
2118 if (skb_shared(list)) {
2119 /* Sucks! We need to fork list. :-( */
2120 clone = skb_clone(list, GFP_ATOMIC);
2126 /* This may be pulled without
2130 if (!pskb_pull(list, eat)) {
2138 /* Free pulled out fragments. */
2139 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2140 skb_shinfo(skb)->frag_list = list->next;
2143 /* And insert new clone at head. */
2146 skb_shinfo(skb)->frag_list = clone;
2149 /* Success! Now we may commit changes to skb data. */
2154 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2155 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2158 skb_frag_unref(skb, i);
2161 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2163 *frag = skb_shinfo(skb)->frags[i];
2165 skb_frag_off_add(frag, eat);
2166 skb_frag_size_sub(frag, eat);
2174 skb_shinfo(skb)->nr_frags = k;
2178 skb->data_len -= delta;
2181 skb_zcopy_clear(skb, false);
2183 return skb_tail_pointer(skb);
2185 EXPORT_SYMBOL(__pskb_pull_tail);
2188 * skb_copy_bits - copy bits from skb to kernel buffer
2190 * @offset: offset in source
2191 * @to: destination buffer
2192 * @len: number of bytes to copy
2194 * Copy the specified number of bytes from the source skb to the
2195 * destination buffer.
2198 * If its prototype is ever changed,
2199 * check arch/{*}/net/{*}.S files,
2200 * since it is called from BPF assembly code.
2202 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2204 int start = skb_headlen(skb);
2205 struct sk_buff *frag_iter;
2208 if (offset > (int)skb->len - len)
2212 if ((copy = start - offset) > 0) {
2215 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2216 if ((len -= copy) == 0)
2222 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2224 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2226 WARN_ON(start > offset + len);
2228 end = start + skb_frag_size(f);
2229 if ((copy = end - offset) > 0) {
2230 u32 p_off, p_len, copied;
2237 skb_frag_foreach_page(f,
2238 skb_frag_off(f) + offset - start,
2239 copy, p, p_off, p_len, copied) {
2240 vaddr = kmap_atomic(p);
2241 memcpy(to + copied, vaddr + p_off, p_len);
2242 kunmap_atomic(vaddr);
2245 if ((len -= copy) == 0)
2253 skb_walk_frags(skb, frag_iter) {
2256 WARN_ON(start > offset + len);
2258 end = start + frag_iter->len;
2259 if ((copy = end - offset) > 0) {
2262 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2264 if ((len -= copy) == 0)
2278 EXPORT_SYMBOL(skb_copy_bits);
2281 * Callback from splice_to_pipe(), if we need to release some pages
2282 * at the end of the spd in case we error'ed out in filling the pipe.
2284 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2286 put_page(spd->pages[i]);
2289 static struct page *linear_to_page(struct page *page, unsigned int *len,
2290 unsigned int *offset,
2293 struct page_frag *pfrag = sk_page_frag(sk);
2295 if (!sk_page_frag_refill(sk, pfrag))
2298 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2300 memcpy(page_address(pfrag->page) + pfrag->offset,
2301 page_address(page) + *offset, *len);
2302 *offset = pfrag->offset;
2303 pfrag->offset += *len;
2308 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2310 unsigned int offset)
2312 return spd->nr_pages &&
2313 spd->pages[spd->nr_pages - 1] == page &&
2314 (spd->partial[spd->nr_pages - 1].offset +
2315 spd->partial[spd->nr_pages - 1].len == offset);
2319 * Fill page/offset/length into spd, if it can hold more pages.
2321 static bool spd_fill_page(struct splice_pipe_desc *spd,
2322 struct pipe_inode_info *pipe, struct page *page,
2323 unsigned int *len, unsigned int offset,
2327 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2331 page = linear_to_page(page, len, &offset, sk);
2335 if (spd_can_coalesce(spd, page, offset)) {
2336 spd->partial[spd->nr_pages - 1].len += *len;
2340 spd->pages[spd->nr_pages] = page;
2341 spd->partial[spd->nr_pages].len = *len;
2342 spd->partial[spd->nr_pages].offset = offset;
2348 static bool __splice_segment(struct page *page, unsigned int poff,
2349 unsigned int plen, unsigned int *off,
2351 struct splice_pipe_desc *spd, bool linear,
2353 struct pipe_inode_info *pipe)
2358 /* skip this segment if already processed */
2364 /* ignore any bits we already processed */
2370 unsigned int flen = min(*len, plen);
2372 if (spd_fill_page(spd, pipe, page, &flen, poff,
2378 } while (*len && plen);
2384 * Map linear and fragment data from the skb to spd. It reports true if the
2385 * pipe is full or if we already spliced the requested length.
2387 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2388 unsigned int *offset, unsigned int *len,
2389 struct splice_pipe_desc *spd, struct sock *sk)
2392 struct sk_buff *iter;
2394 /* map the linear part :
2395 * If skb->head_frag is set, this 'linear' part is backed by a
2396 * fragment, and if the head is not shared with any clones then
2397 * we can avoid a copy since we own the head portion of this page.
2399 if (__splice_segment(virt_to_page(skb->data),
2400 (unsigned long) skb->data & (PAGE_SIZE - 1),
2403 skb_head_is_locked(skb),
2408 * then map the fragments
2410 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2411 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2413 if (__splice_segment(skb_frag_page(f),
2414 skb_frag_off(f), skb_frag_size(f),
2415 offset, len, spd, false, sk, pipe))
2419 skb_walk_frags(skb, iter) {
2420 if (*offset >= iter->len) {
2421 *offset -= iter->len;
2424 /* __skb_splice_bits() only fails if the output has no room
2425 * left, so no point in going over the frag_list for the error
2428 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2436 * Map data from the skb to a pipe. Should handle both the linear part,
2437 * the fragments, and the frag list.
2439 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2440 struct pipe_inode_info *pipe, unsigned int tlen,
2443 struct partial_page partial[MAX_SKB_FRAGS];
2444 struct page *pages[MAX_SKB_FRAGS];
2445 struct splice_pipe_desc spd = {
2448 .nr_pages_max = MAX_SKB_FRAGS,
2449 .ops = &nosteal_pipe_buf_ops,
2450 .spd_release = sock_spd_release,
2454 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2457 ret = splice_to_pipe(pipe, &spd);
2461 EXPORT_SYMBOL_GPL(skb_splice_bits);
2463 /* Send skb data on a socket. Socket must be locked. */
2464 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2467 unsigned int orig_len = len;
2468 struct sk_buff *head = skb;
2469 unsigned short fragidx;
2474 /* Deal with head data */
2475 while (offset < skb_headlen(skb) && len) {
2479 slen = min_t(int, len, skb_headlen(skb) - offset);
2480 kv.iov_base = skb->data + offset;
2482 memset(&msg, 0, sizeof(msg));
2483 msg.msg_flags = MSG_DONTWAIT;
2485 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2493 /* All the data was skb head? */
2497 /* Make offset relative to start of frags */
2498 offset -= skb_headlen(skb);
2500 /* Find where we are in frag list */
2501 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2502 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2504 if (offset < skb_frag_size(frag))
2507 offset -= skb_frag_size(frag);
2510 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2511 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2513 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2516 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2517 skb_frag_off(frag) + offset,
2518 slen, MSG_DONTWAIT);
2531 /* Process any frag lists */
2534 if (skb_has_frag_list(skb)) {
2535 skb = skb_shinfo(skb)->frag_list;
2538 } else if (skb->next) {
2545 return orig_len - len;
2548 return orig_len == len ? ret : orig_len - len;
2550 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2553 * skb_store_bits - store bits from kernel buffer to skb
2554 * @skb: destination buffer
2555 * @offset: offset in destination
2556 * @from: source buffer
2557 * @len: number of bytes to copy
2559 * Copy the specified number of bytes from the source buffer to the
2560 * destination skb. This function handles all the messy bits of
2561 * traversing fragment lists and such.
2564 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2566 int start = skb_headlen(skb);
2567 struct sk_buff *frag_iter;
2570 if (offset > (int)skb->len - len)
2573 if ((copy = start - offset) > 0) {
2576 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2577 if ((len -= copy) == 0)
2583 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2584 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2587 WARN_ON(start > offset + len);
2589 end = start + skb_frag_size(frag);
2590 if ((copy = end - offset) > 0) {
2591 u32 p_off, p_len, copied;
2598 skb_frag_foreach_page(frag,
2599 skb_frag_off(frag) + offset - start,
2600 copy, p, p_off, p_len, copied) {
2601 vaddr = kmap_atomic(p);
2602 memcpy(vaddr + p_off, from + copied, p_len);
2603 kunmap_atomic(vaddr);
2606 if ((len -= copy) == 0)
2614 skb_walk_frags(skb, frag_iter) {
2617 WARN_ON(start > offset + len);
2619 end = start + frag_iter->len;
2620 if ((copy = end - offset) > 0) {
2623 if (skb_store_bits(frag_iter, offset - start,
2626 if ((len -= copy) == 0)
2639 EXPORT_SYMBOL(skb_store_bits);
2641 /* Checksum skb data. */
2642 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2643 __wsum csum, const struct skb_checksum_ops *ops)
2645 int start = skb_headlen(skb);
2646 int i, copy = start - offset;
2647 struct sk_buff *frag_iter;
2650 /* Checksum header. */
2654 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2655 skb->data + offset, copy, csum);
2656 if ((len -= copy) == 0)
2662 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2664 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2666 WARN_ON(start > offset + len);
2668 end = start + skb_frag_size(frag);
2669 if ((copy = end - offset) > 0) {
2670 u32 p_off, p_len, copied;
2678 skb_frag_foreach_page(frag,
2679 skb_frag_off(frag) + offset - start,
2680 copy, p, p_off, p_len, copied) {
2681 vaddr = kmap_atomic(p);
2682 csum2 = INDIRECT_CALL_1(ops->update,
2684 vaddr + p_off, p_len, 0);
2685 kunmap_atomic(vaddr);
2686 csum = INDIRECT_CALL_1(ops->combine,
2687 csum_block_add_ext, csum,
2699 skb_walk_frags(skb, frag_iter) {
2702 WARN_ON(start > offset + len);
2704 end = start + frag_iter->len;
2705 if ((copy = end - offset) > 0) {
2709 csum2 = __skb_checksum(frag_iter, offset - start,
2711 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2712 csum, csum2, pos, copy);
2713 if ((len -= copy) == 0)
2724 EXPORT_SYMBOL(__skb_checksum);
2726 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2727 int len, __wsum csum)
2729 const struct skb_checksum_ops ops = {
2730 .update = csum_partial_ext,
2731 .combine = csum_block_add_ext,
2734 return __skb_checksum(skb, offset, len, csum, &ops);
2736 EXPORT_SYMBOL(skb_checksum);
2738 /* Both of above in one bottle. */
2740 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2743 int start = skb_headlen(skb);
2744 int i, copy = start - offset;
2745 struct sk_buff *frag_iter;
2753 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2755 if ((len -= copy) == 0)
2762 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2765 WARN_ON(start > offset + len);
2767 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2768 if ((copy = end - offset) > 0) {
2769 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2770 u32 p_off, p_len, copied;
2778 skb_frag_foreach_page(frag,
2779 skb_frag_off(frag) + offset - start,
2780 copy, p, p_off, p_len, copied) {
2781 vaddr = kmap_atomic(p);
2782 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2785 kunmap_atomic(vaddr);
2786 csum = csum_block_add(csum, csum2, pos);
2798 skb_walk_frags(skb, frag_iter) {
2802 WARN_ON(start > offset + len);
2804 end = start + frag_iter->len;
2805 if ((copy = end - offset) > 0) {
2808 csum2 = skb_copy_and_csum_bits(frag_iter,
2811 csum = csum_block_add(csum, csum2, pos);
2812 if ((len -= copy) == 0)
2823 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2825 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2829 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2830 /* See comments in __skb_checksum_complete(). */
2832 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2833 !skb->csum_complete_sw)
2834 netdev_rx_csum_fault(skb->dev, skb);
2836 if (!skb_shared(skb))
2837 skb->csum_valid = !sum;
2840 EXPORT_SYMBOL(__skb_checksum_complete_head);
2842 /* This function assumes skb->csum already holds pseudo header's checksum,
2843 * which has been changed from the hardware checksum, for example, by
2844 * __skb_checksum_validate_complete(). And, the original skb->csum must
2845 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2847 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2848 * zero. The new checksum is stored back into skb->csum unless the skb is
2851 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2856 csum = skb_checksum(skb, 0, skb->len, 0);
2858 sum = csum_fold(csum_add(skb->csum, csum));
2859 /* This check is inverted, because we already knew the hardware
2860 * checksum is invalid before calling this function. So, if the
2861 * re-computed checksum is valid instead, then we have a mismatch
2862 * between the original skb->csum and skb_checksum(). This means either
2863 * the original hardware checksum is incorrect or we screw up skb->csum
2864 * when moving skb->data around.
2867 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2868 !skb->csum_complete_sw)
2869 netdev_rx_csum_fault(skb->dev, skb);
2872 if (!skb_shared(skb)) {
2873 /* Save full packet checksum */
2875 skb->ip_summed = CHECKSUM_COMPLETE;
2876 skb->csum_complete_sw = 1;
2877 skb->csum_valid = !sum;
2882 EXPORT_SYMBOL(__skb_checksum_complete);
2884 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2886 net_warn_ratelimited(
2887 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2892 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2893 int offset, int len)
2895 net_warn_ratelimited(
2896 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2901 static const struct skb_checksum_ops default_crc32c_ops = {
2902 .update = warn_crc32c_csum_update,
2903 .combine = warn_crc32c_csum_combine,
2906 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2907 &default_crc32c_ops;
2908 EXPORT_SYMBOL(crc32c_csum_stub);
2911 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2912 * @from: source buffer
2914 * Calculates the amount of linear headroom needed in the 'to' skb passed
2915 * into skb_zerocopy().
2918 skb_zerocopy_headlen(const struct sk_buff *from)
2920 unsigned int hlen = 0;
2922 if (!from->head_frag ||
2923 skb_headlen(from) < L1_CACHE_BYTES ||
2924 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2925 hlen = skb_headlen(from);
2927 if (skb_has_frag_list(from))
2932 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2935 * skb_zerocopy - Zero copy skb to skb
2936 * @to: destination buffer
2937 * @from: source buffer
2938 * @len: number of bytes to copy from source buffer
2939 * @hlen: size of linear headroom in destination buffer
2941 * Copies up to `len` bytes from `from` to `to` by creating references
2942 * to the frags in the source buffer.
2944 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2945 * headroom in the `to` buffer.
2948 * 0: everything is OK
2949 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2950 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2953 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2956 int plen = 0; /* length of skb->head fragment */
2959 unsigned int offset;
2961 BUG_ON(!from->head_frag && !hlen);
2963 /* dont bother with small payloads */
2964 if (len <= skb_tailroom(to))
2965 return skb_copy_bits(from, 0, skb_put(to, len), len);
2968 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2973 plen = min_t(int, skb_headlen(from), len);
2975 page = virt_to_head_page(from->head);
2976 offset = from->data - (unsigned char *)page_address(page);
2977 __skb_fill_page_desc(to, 0, page, offset, plen);
2984 to->truesize += len + plen;
2985 to->len += len + plen;
2986 to->data_len += len + plen;
2988 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2992 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2994 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2999 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3000 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3002 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3004 skb_frag_ref(to, j);
3007 skb_shinfo(to)->nr_frags = j;
3011 EXPORT_SYMBOL_GPL(skb_zerocopy);
3013 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3018 if (skb->ip_summed == CHECKSUM_PARTIAL)
3019 csstart = skb_checksum_start_offset(skb);
3021 csstart = skb_headlen(skb);
3023 BUG_ON(csstart > skb_headlen(skb));
3025 skb_copy_from_linear_data(skb, to, csstart);
3028 if (csstart != skb->len)
3029 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3030 skb->len - csstart);
3032 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3033 long csstuff = csstart + skb->csum_offset;
3035 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3038 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3041 * skb_dequeue - remove from the head of the queue
3042 * @list: list to dequeue from
3044 * Remove the head of the list. The list lock is taken so the function
3045 * may be used safely with other locking list functions. The head item is
3046 * returned or %NULL if the list is empty.
3049 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3051 unsigned long flags;
3052 struct sk_buff *result;
3054 spin_lock_irqsave(&list->lock, flags);
3055 result = __skb_dequeue(list);
3056 spin_unlock_irqrestore(&list->lock, flags);
3059 EXPORT_SYMBOL(skb_dequeue);
3062 * skb_dequeue_tail - remove from the tail of the queue
3063 * @list: list to dequeue from
3065 * Remove the tail of the list. The list lock is taken so the function
3066 * may be used safely with other locking list functions. The tail item is
3067 * returned or %NULL if the list is empty.
3069 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3071 unsigned long flags;
3072 struct sk_buff *result;
3074 spin_lock_irqsave(&list->lock, flags);
3075 result = __skb_dequeue_tail(list);
3076 spin_unlock_irqrestore(&list->lock, flags);
3079 EXPORT_SYMBOL(skb_dequeue_tail);
3082 * skb_queue_purge - empty a list
3083 * @list: list to empty
3085 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3086 * the list and one reference dropped. This function takes the list
3087 * lock and is atomic with respect to other list locking functions.
3089 void skb_queue_purge(struct sk_buff_head *list)
3091 struct sk_buff *skb;
3092 while ((skb = skb_dequeue(list)) != NULL)
3095 EXPORT_SYMBOL(skb_queue_purge);
3098 * skb_rbtree_purge - empty a skb rbtree
3099 * @root: root of the rbtree to empty
3100 * Return value: the sum of truesizes of all purged skbs.
3102 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3103 * the list and one reference dropped. This function does not take
3104 * any lock. Synchronization should be handled by the caller (e.g., TCP
3105 * out-of-order queue is protected by the socket lock).
3107 unsigned int skb_rbtree_purge(struct rb_root *root)
3109 struct rb_node *p = rb_first(root);
3110 unsigned int sum = 0;
3113 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3116 rb_erase(&skb->rbnode, root);
3117 sum += skb->truesize;
3124 * skb_queue_head - queue a buffer at the list head
3125 * @list: list to use
3126 * @newsk: buffer to queue
3128 * Queue a buffer at the start of the list. This function takes the
3129 * list lock and can be used safely with other locking &sk_buff functions
3132 * A buffer cannot be placed on two lists at the same time.
3134 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3136 unsigned long flags;
3138 spin_lock_irqsave(&list->lock, flags);
3139 __skb_queue_head(list, newsk);
3140 spin_unlock_irqrestore(&list->lock, flags);
3142 EXPORT_SYMBOL(skb_queue_head);
3145 * skb_queue_tail - queue a buffer at the list tail
3146 * @list: list to use
3147 * @newsk: buffer to queue
3149 * Queue a buffer at the tail of the list. This function takes the
3150 * list lock and can be used safely with other locking &sk_buff functions
3153 * A buffer cannot be placed on two lists at the same time.
3155 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3157 unsigned long flags;
3159 spin_lock_irqsave(&list->lock, flags);
3160 __skb_queue_tail(list, newsk);
3161 spin_unlock_irqrestore(&list->lock, flags);
3163 EXPORT_SYMBOL(skb_queue_tail);
3166 * skb_unlink - remove a buffer from a list
3167 * @skb: buffer to remove
3168 * @list: list to use
3170 * Remove a packet from a list. The list locks are taken and this
3171 * function is atomic with respect to other list locked calls
3173 * You must know what list the SKB is on.
3175 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3177 unsigned long flags;
3179 spin_lock_irqsave(&list->lock, flags);
3180 __skb_unlink(skb, list);
3181 spin_unlock_irqrestore(&list->lock, flags);
3183 EXPORT_SYMBOL(skb_unlink);
3186 * skb_append - append a buffer
3187 * @old: buffer to insert after
3188 * @newsk: buffer to insert
3189 * @list: list to use
3191 * Place a packet after a given packet in a list. The list locks are taken
3192 * and this function is atomic with respect to other list locked calls.
3193 * A buffer cannot be placed on two lists at the same time.
3195 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3197 unsigned long flags;
3199 spin_lock_irqsave(&list->lock, flags);
3200 __skb_queue_after(list, old, newsk);
3201 spin_unlock_irqrestore(&list->lock, flags);
3203 EXPORT_SYMBOL(skb_append);
3205 static inline void skb_split_inside_header(struct sk_buff *skb,
3206 struct sk_buff* skb1,
3207 const u32 len, const int pos)
3211 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3213 /* And move data appendix as is. */
3214 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3215 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3217 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3218 skb_shinfo(skb)->nr_frags = 0;
3219 skb1->data_len = skb->data_len;
3220 skb1->len += skb1->data_len;
3223 skb_set_tail_pointer(skb, len);
3226 static inline void skb_split_no_header(struct sk_buff *skb,
3227 struct sk_buff* skb1,
3228 const u32 len, int pos)
3231 const int nfrags = skb_shinfo(skb)->nr_frags;
3233 skb_shinfo(skb)->nr_frags = 0;
3234 skb1->len = skb1->data_len = skb->len - len;
3236 skb->data_len = len - pos;
3238 for (i = 0; i < nfrags; i++) {
3239 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3241 if (pos + size > len) {
3242 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3246 * We have two variants in this case:
3247 * 1. Move all the frag to the second
3248 * part, if it is possible. F.e.
3249 * this approach is mandatory for TUX,
3250 * where splitting is expensive.
3251 * 2. Split is accurately. We make this.
3253 skb_frag_ref(skb, i);
3254 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3255 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3256 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3257 skb_shinfo(skb)->nr_frags++;
3261 skb_shinfo(skb)->nr_frags++;
3264 skb_shinfo(skb1)->nr_frags = k;
3268 * skb_split - Split fragmented skb to two parts at length len.
3269 * @skb: the buffer to split
3270 * @skb1: the buffer to receive the second part
3271 * @len: new length for skb
3273 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3275 int pos = skb_headlen(skb);
3277 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG;
3278 skb_zerocopy_clone(skb1, skb, 0);
3279 if (len < pos) /* Split line is inside header. */
3280 skb_split_inside_header(skb, skb1, len, pos);
3281 else /* Second chunk has no header, nothing to copy. */
3282 skb_split_no_header(skb, skb1, len, pos);
3284 EXPORT_SYMBOL(skb_split);
3286 /* Shifting from/to a cloned skb is a no-go.
3288 * Caller cannot keep skb_shinfo related pointers past calling here!
3290 static int skb_prepare_for_shift(struct sk_buff *skb)
3294 if (skb_cloned(skb)) {
3295 /* Save and restore truesize: pskb_expand_head() may reallocate
3296 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3297 * cannot change truesize at this point.
3299 unsigned int save_truesize = skb->truesize;
3301 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3302 skb->truesize = save_truesize;
3308 * skb_shift - Shifts paged data partially from skb to another
3309 * @tgt: buffer into which tail data gets added
3310 * @skb: buffer from which the paged data comes from
3311 * @shiftlen: shift up to this many bytes
3313 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3314 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3315 * It's up to caller to free skb if everything was shifted.
3317 * If @tgt runs out of frags, the whole operation is aborted.
3319 * Skb cannot include anything else but paged data while tgt is allowed
3320 * to have non-paged data as well.
3322 * TODO: full sized shift could be optimized but that would need
3323 * specialized skb free'er to handle frags without up-to-date nr_frags.
3325 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3327 int from, to, merge, todo;
3328 skb_frag_t *fragfrom, *fragto;
3330 BUG_ON(shiftlen > skb->len);
3332 if (skb_headlen(skb))
3334 if (skb_zcopy(tgt) || skb_zcopy(skb))
3339 to = skb_shinfo(tgt)->nr_frags;
3340 fragfrom = &skb_shinfo(skb)->frags[from];
3342 /* Actual merge is delayed until the point when we know we can
3343 * commit all, so that we don't have to undo partial changes
3346 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3347 skb_frag_off(fragfrom))) {
3352 todo -= skb_frag_size(fragfrom);
3354 if (skb_prepare_for_shift(skb) ||
3355 skb_prepare_for_shift(tgt))
3358 /* All previous frag pointers might be stale! */
3359 fragfrom = &skb_shinfo(skb)->frags[from];
3360 fragto = &skb_shinfo(tgt)->frags[merge];
3362 skb_frag_size_add(fragto, shiftlen);
3363 skb_frag_size_sub(fragfrom, shiftlen);
3364 skb_frag_off_add(fragfrom, shiftlen);
3372 /* Skip full, not-fitting skb to avoid expensive operations */
3373 if ((shiftlen == skb->len) &&
3374 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3377 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3380 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3381 if (to == MAX_SKB_FRAGS)
3384 fragfrom = &skb_shinfo(skb)->frags[from];
3385 fragto = &skb_shinfo(tgt)->frags[to];
3387 if (todo >= skb_frag_size(fragfrom)) {
3388 *fragto = *fragfrom;
3389 todo -= skb_frag_size(fragfrom);
3394 __skb_frag_ref(fragfrom);
3395 skb_frag_page_copy(fragto, fragfrom);
3396 skb_frag_off_copy(fragto, fragfrom);
3397 skb_frag_size_set(fragto, todo);
3399 skb_frag_off_add(fragfrom, todo);
3400 skb_frag_size_sub(fragfrom, todo);
3408 /* Ready to "commit" this state change to tgt */
3409 skb_shinfo(tgt)->nr_frags = to;
3412 fragfrom = &skb_shinfo(skb)->frags[0];
3413 fragto = &skb_shinfo(tgt)->frags[merge];
3415 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3416 __skb_frag_unref(fragfrom);
3419 /* Reposition in the original skb */
3421 while (from < skb_shinfo(skb)->nr_frags)
3422 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3423 skb_shinfo(skb)->nr_frags = to;
3425 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3428 /* Most likely the tgt won't ever need its checksum anymore, skb on
3429 * the other hand might need it if it needs to be resent
3431 tgt->ip_summed = CHECKSUM_PARTIAL;
3432 skb->ip_summed = CHECKSUM_PARTIAL;
3434 /* Yak, is it really working this way? Some helper please? */
3435 skb->len -= shiftlen;
3436 skb->data_len -= shiftlen;
3437 skb->truesize -= shiftlen;
3438 tgt->len += shiftlen;
3439 tgt->data_len += shiftlen;
3440 tgt->truesize += shiftlen;
3446 * skb_prepare_seq_read - Prepare a sequential read of skb data
3447 * @skb: the buffer to read
3448 * @from: lower offset of data to be read
3449 * @to: upper offset of data to be read
3450 * @st: state variable
3452 * Initializes the specified state variable. Must be called before
3453 * invoking skb_seq_read() for the first time.
3455 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3456 unsigned int to, struct skb_seq_state *st)
3458 st->lower_offset = from;
3459 st->upper_offset = to;
3460 st->root_skb = st->cur_skb = skb;
3461 st->frag_idx = st->stepped_offset = 0;
3462 st->frag_data = NULL;
3465 EXPORT_SYMBOL(skb_prepare_seq_read);
3468 * skb_seq_read - Sequentially read skb data
3469 * @consumed: number of bytes consumed by the caller so far
3470 * @data: destination pointer for data to be returned
3471 * @st: state variable
3473 * Reads a block of skb data at @consumed relative to the
3474 * lower offset specified to skb_prepare_seq_read(). Assigns
3475 * the head of the data block to @data and returns the length
3476 * of the block or 0 if the end of the skb data or the upper
3477 * offset has been reached.
3479 * The caller is not required to consume all of the data
3480 * returned, i.e. @consumed is typically set to the number
3481 * of bytes already consumed and the next call to
3482 * skb_seq_read() will return the remaining part of the block.
3484 * Note 1: The size of each block of data returned can be arbitrary,
3485 * this limitation is the cost for zerocopy sequential
3486 * reads of potentially non linear data.
3488 * Note 2: Fragment lists within fragments are not implemented
3489 * at the moment, state->root_skb could be replaced with
3490 * a stack for this purpose.
3492 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3493 struct skb_seq_state *st)
3495 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3498 if (unlikely(abs_offset >= st->upper_offset)) {
3499 if (st->frag_data) {
3500 kunmap_atomic(st->frag_data);
3501 st->frag_data = NULL;
3507 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3509 if (abs_offset < block_limit && !st->frag_data) {
3510 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3511 return block_limit - abs_offset;
3514 if (st->frag_idx == 0 && !st->frag_data)
3515 st->stepped_offset += skb_headlen(st->cur_skb);
3517 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3518 unsigned int pg_idx, pg_off, pg_sz;
3520 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3523 pg_off = skb_frag_off(frag);
3524 pg_sz = skb_frag_size(frag);
3526 if (skb_frag_must_loop(skb_frag_page(frag))) {
3527 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3528 pg_off = offset_in_page(pg_off + st->frag_off);
3529 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3530 PAGE_SIZE - pg_off);
3533 block_limit = pg_sz + st->stepped_offset;
3534 if (abs_offset < block_limit) {
3536 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3538 *data = (u8 *)st->frag_data + pg_off +
3539 (abs_offset - st->stepped_offset);
3541 return block_limit - abs_offset;
3544 if (st->frag_data) {
3545 kunmap_atomic(st->frag_data);
3546 st->frag_data = NULL;
3549 st->stepped_offset += pg_sz;
3550 st->frag_off += pg_sz;
3551 if (st->frag_off == skb_frag_size(frag)) {
3557 if (st->frag_data) {
3558 kunmap_atomic(st->frag_data);
3559 st->frag_data = NULL;
3562 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3563 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3566 } else if (st->cur_skb->next) {
3567 st->cur_skb = st->cur_skb->next;
3574 EXPORT_SYMBOL(skb_seq_read);
3577 * skb_abort_seq_read - Abort a sequential read of skb data
3578 * @st: state variable
3580 * Must be called if skb_seq_read() was not called until it
3583 void skb_abort_seq_read(struct skb_seq_state *st)
3586 kunmap_atomic(st->frag_data);
3588 EXPORT_SYMBOL(skb_abort_seq_read);
3590 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3592 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3593 struct ts_config *conf,
3594 struct ts_state *state)
3596 return skb_seq_read(offset, text, TS_SKB_CB(state));
3599 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3601 skb_abort_seq_read(TS_SKB_CB(state));
3605 * skb_find_text - Find a text pattern in skb data
3606 * @skb: the buffer to look in
3607 * @from: search offset
3609 * @config: textsearch configuration
3611 * Finds a pattern in the skb data according to the specified
3612 * textsearch configuration. Use textsearch_next() to retrieve
3613 * subsequent occurrences of the pattern. Returns the offset
3614 * to the first occurrence or UINT_MAX if no match was found.
3616 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3617 unsigned int to, struct ts_config *config)
3619 struct ts_state state;
3622 config->get_next_block = skb_ts_get_next_block;
3623 config->finish = skb_ts_finish;
3625 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3627 ret = textsearch_find(config, &state);
3628 return (ret <= to - from ? ret : UINT_MAX);
3630 EXPORT_SYMBOL(skb_find_text);
3632 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3633 int offset, size_t size)
3635 int i = skb_shinfo(skb)->nr_frags;
3637 if (skb_can_coalesce(skb, i, page, offset)) {
3638 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3639 } else if (i < MAX_SKB_FRAGS) {
3641 skb_fill_page_desc(skb, i, page, offset, size);
3648 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3651 * skb_pull_rcsum - pull skb and update receive checksum
3652 * @skb: buffer to update
3653 * @len: length of data pulled
3655 * This function performs an skb_pull on the packet and updates
3656 * the CHECKSUM_COMPLETE checksum. It should be used on
3657 * receive path processing instead of skb_pull unless you know
3658 * that the checksum difference is zero (e.g., a valid IP header)
3659 * or you are setting ip_summed to CHECKSUM_NONE.
3661 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3663 unsigned char *data = skb->data;
3665 BUG_ON(len > skb->len);
3666 __skb_pull(skb, len);
3667 skb_postpull_rcsum(skb, data, len);
3670 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3672 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3674 skb_frag_t head_frag;
3677 page = virt_to_head_page(frag_skb->head);
3678 __skb_frag_set_page(&head_frag, page);
3679 skb_frag_off_set(&head_frag, frag_skb->data -
3680 (unsigned char *)page_address(page));
3681 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3685 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3686 netdev_features_t features,
3687 unsigned int offset)
3689 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3690 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3691 unsigned int delta_truesize = 0;
3692 unsigned int delta_len = 0;
3693 struct sk_buff *tail = NULL;
3694 struct sk_buff *nskb, *tmp;
3697 skb_push(skb, -skb_network_offset(skb) + offset);
3699 skb_shinfo(skb)->frag_list = NULL;
3703 list_skb = list_skb->next;
3706 if (skb_shared(nskb)) {
3707 tmp = skb_clone(nskb, GFP_ATOMIC);
3711 err = skb_unclone(nskb, GFP_ATOMIC);
3722 if (unlikely(err)) {
3723 nskb->next = list_skb;
3729 delta_len += nskb->len;
3730 delta_truesize += nskb->truesize;
3732 skb_push(nskb, -skb_network_offset(nskb) + offset);
3734 skb_release_head_state(nskb);
3735 __copy_skb_header(nskb, skb);
3737 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3738 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3739 nskb->data - tnl_hlen,
3742 if (skb_needs_linearize(nskb, features) &&
3743 __skb_linearize(nskb))
3748 skb->truesize = skb->truesize - delta_truesize;
3749 skb->data_len = skb->data_len - delta_len;
3750 skb->len = skb->len - delta_len;
3756 if (skb_needs_linearize(skb, features) &&
3757 __skb_linearize(skb))
3765 kfree_skb_list(skb->next);
3767 return ERR_PTR(-ENOMEM);
3769 EXPORT_SYMBOL_GPL(skb_segment_list);
3771 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3773 if (unlikely(p->len + skb->len >= 65536))
3776 if (NAPI_GRO_CB(p)->last == p)
3777 skb_shinfo(p)->frag_list = skb;
3779 NAPI_GRO_CB(p)->last->next = skb;
3781 skb_pull(skb, skb_gro_offset(skb));
3783 NAPI_GRO_CB(p)->last = skb;
3784 NAPI_GRO_CB(p)->count++;
3785 p->data_len += skb->len;
3786 p->truesize += skb->truesize;
3789 NAPI_GRO_CB(skb)->same_flow = 1;
3795 * skb_segment - Perform protocol segmentation on skb.
3796 * @head_skb: buffer to segment
3797 * @features: features for the output path (see dev->features)
3799 * This function performs segmentation on the given skb. It returns
3800 * a pointer to the first in a list of new skbs for the segments.
3801 * In case of error it returns ERR_PTR(err).
3803 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3804 netdev_features_t features)
3806 struct sk_buff *segs = NULL;
3807 struct sk_buff *tail = NULL;
3808 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3809 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3810 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3811 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3812 struct sk_buff *frag_skb = head_skb;
3813 unsigned int offset = doffset;
3814 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3815 unsigned int partial_segs = 0;
3816 unsigned int headroom;
3817 unsigned int len = head_skb->len;
3820 int nfrags = skb_shinfo(head_skb)->nr_frags;
3825 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3826 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3827 /* gso_size is untrusted, and we have a frag_list with a linear
3828 * non head_frag head.
3830 * (we assume checking the first list_skb member suffices;
3831 * i.e if either of the list_skb members have non head_frag
3832 * head, then the first one has too).
3834 * If head_skb's headlen does not fit requested gso_size, it
3835 * means that the frag_list members do NOT terminate on exact
3836 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3837 * sharing. Therefore we must fallback to copying the frag_list
3838 * skbs; we do so by disabling SG.
3840 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3841 features &= ~NETIF_F_SG;
3844 __skb_push(head_skb, doffset);
3845 proto = skb_network_protocol(head_skb, NULL);
3846 if (unlikely(!proto))
3847 return ERR_PTR(-EINVAL);
3849 sg = !!(features & NETIF_F_SG);
3850 csum = !!can_checksum_protocol(features, proto);
3852 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3853 if (!(features & NETIF_F_GSO_PARTIAL)) {
3854 struct sk_buff *iter;
3855 unsigned int frag_len;
3858 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3861 /* If we get here then all the required
3862 * GSO features except frag_list are supported.
3863 * Try to split the SKB to multiple GSO SKBs
3864 * with no frag_list.
3865 * Currently we can do that only when the buffers don't
3866 * have a linear part and all the buffers except
3867 * the last are of the same length.
3869 frag_len = list_skb->len;
3870 skb_walk_frags(head_skb, iter) {
3871 if (frag_len != iter->len && iter->next)
3873 if (skb_headlen(iter) && !iter->head_frag)
3879 if (len != frag_len)
3883 /* GSO partial only requires that we trim off any excess that
3884 * doesn't fit into an MSS sized block, so take care of that
3887 partial_segs = len / mss;
3888 if (partial_segs > 1)
3889 mss *= partial_segs;
3895 headroom = skb_headroom(head_skb);
3896 pos = skb_headlen(head_skb);
3899 struct sk_buff *nskb;
3900 skb_frag_t *nskb_frag;
3904 if (unlikely(mss == GSO_BY_FRAGS)) {
3905 len = list_skb->len;
3907 len = head_skb->len - offset;
3912 hsize = skb_headlen(head_skb) - offset;
3914 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
3915 (skb_headlen(list_skb) == len || sg)) {
3916 BUG_ON(skb_headlen(list_skb) > len);
3919 nfrags = skb_shinfo(list_skb)->nr_frags;
3920 frag = skb_shinfo(list_skb)->frags;
3921 frag_skb = list_skb;
3922 pos += skb_headlen(list_skb);
3924 while (pos < offset + len) {
3925 BUG_ON(i >= nfrags);
3927 size = skb_frag_size(frag);
3928 if (pos + size > offset + len)
3936 nskb = skb_clone(list_skb, GFP_ATOMIC);
3937 list_skb = list_skb->next;
3939 if (unlikely(!nskb))
3942 if (unlikely(pskb_trim(nskb, len))) {
3947 hsize = skb_end_offset(nskb);
3948 if (skb_cow_head(nskb, doffset + headroom)) {
3953 nskb->truesize += skb_end_offset(nskb) - hsize;
3954 skb_release_head_state(nskb);
3955 __skb_push(nskb, doffset);
3959 if (hsize > len || !sg)
3962 nskb = __alloc_skb(hsize + doffset + headroom,
3963 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3966 if (unlikely(!nskb))
3969 skb_reserve(nskb, headroom);
3970 __skb_put(nskb, doffset);
3979 __copy_skb_header(nskb, head_skb);
3981 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3982 skb_reset_mac_len(nskb);
3984 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3985 nskb->data - tnl_hlen,
3986 doffset + tnl_hlen);
3988 if (nskb->len == len + doffset)
3989 goto perform_csum_check;
3993 if (!nskb->remcsum_offload)
3994 nskb->ip_summed = CHECKSUM_NONE;
3995 SKB_GSO_CB(nskb)->csum =
3996 skb_copy_and_csum_bits(head_skb, offset,
4000 SKB_GSO_CB(nskb)->csum_start =
4001 skb_headroom(nskb) + doffset;
4003 skb_copy_bits(head_skb, offset,
4010 nskb_frag = skb_shinfo(nskb)->frags;
4012 skb_copy_from_linear_data_offset(head_skb, offset,
4013 skb_put(nskb, hsize), hsize);
4015 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4018 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4019 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4022 while (pos < offset + len) {
4025 nfrags = skb_shinfo(list_skb)->nr_frags;
4026 frag = skb_shinfo(list_skb)->frags;
4027 frag_skb = list_skb;
4028 if (!skb_headlen(list_skb)) {
4031 BUG_ON(!list_skb->head_frag);
4033 /* to make room for head_frag. */
4037 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4038 skb_zerocopy_clone(nskb, frag_skb,
4042 list_skb = list_skb->next;
4045 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4047 net_warn_ratelimited(
4048 "skb_segment: too many frags: %u %u\n",
4054 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4055 __skb_frag_ref(nskb_frag);
4056 size = skb_frag_size(nskb_frag);
4059 skb_frag_off_add(nskb_frag, offset - pos);
4060 skb_frag_size_sub(nskb_frag, offset - pos);
4063 skb_shinfo(nskb)->nr_frags++;
4065 if (pos + size <= offset + len) {
4070 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4078 nskb->data_len = len - hsize;
4079 nskb->len += nskb->data_len;
4080 nskb->truesize += nskb->data_len;
4084 if (skb_has_shared_frag(nskb) &&
4085 __skb_linearize(nskb))
4088 if (!nskb->remcsum_offload)
4089 nskb->ip_summed = CHECKSUM_NONE;
4090 SKB_GSO_CB(nskb)->csum =
4091 skb_checksum(nskb, doffset,
4092 nskb->len - doffset, 0);
4093 SKB_GSO_CB(nskb)->csum_start =
4094 skb_headroom(nskb) + doffset;
4096 } while ((offset += len) < head_skb->len);
4098 /* Some callers want to get the end of the list.
4099 * Put it in segs->prev to avoid walking the list.
4100 * (see validate_xmit_skb_list() for example)
4105 struct sk_buff *iter;
4106 int type = skb_shinfo(head_skb)->gso_type;
4107 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4109 /* Update type to add partial and then remove dodgy if set */
4110 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4111 type &= ~SKB_GSO_DODGY;
4113 /* Update GSO info and prepare to start updating headers on
4114 * our way back down the stack of protocols.
4116 for (iter = segs; iter; iter = iter->next) {
4117 skb_shinfo(iter)->gso_size = gso_size;
4118 skb_shinfo(iter)->gso_segs = partial_segs;
4119 skb_shinfo(iter)->gso_type = type;
4120 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4123 if (tail->len - doffset <= gso_size)
4124 skb_shinfo(tail)->gso_size = 0;
4125 else if (tail != segs)
4126 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4129 /* Following permits correct backpressure, for protocols
4130 * using skb_set_owner_w().
4131 * Idea is to tranfert ownership from head_skb to last segment.
4133 if (head_skb->destructor == sock_wfree) {
4134 swap(tail->truesize, head_skb->truesize);
4135 swap(tail->destructor, head_skb->destructor);
4136 swap(tail->sk, head_skb->sk);
4141 kfree_skb_list(segs);
4142 return ERR_PTR(err);
4144 EXPORT_SYMBOL_GPL(skb_segment);
4146 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4148 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4149 unsigned int offset = skb_gro_offset(skb);
4150 unsigned int headlen = skb_headlen(skb);
4151 unsigned int len = skb_gro_len(skb);
4152 unsigned int delta_truesize;
4155 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4158 lp = NAPI_GRO_CB(p)->last;
4159 pinfo = skb_shinfo(lp);
4161 if (headlen <= offset) {
4164 int i = skbinfo->nr_frags;
4165 int nr_frags = pinfo->nr_frags + i;
4167 if (nr_frags > MAX_SKB_FRAGS)
4171 pinfo->nr_frags = nr_frags;
4172 skbinfo->nr_frags = 0;
4174 frag = pinfo->frags + nr_frags;
4175 frag2 = skbinfo->frags + i;
4180 skb_frag_off_add(frag, offset);
4181 skb_frag_size_sub(frag, offset);
4183 /* all fragments truesize : remove (head size + sk_buff) */
4184 delta_truesize = skb->truesize -
4185 SKB_TRUESIZE(skb_end_offset(skb));
4187 skb->truesize -= skb->data_len;
4188 skb->len -= skb->data_len;
4191 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4193 } else if (skb->head_frag) {
4194 int nr_frags = pinfo->nr_frags;
4195 skb_frag_t *frag = pinfo->frags + nr_frags;
4196 struct page *page = virt_to_head_page(skb->head);
4197 unsigned int first_size = headlen - offset;
4198 unsigned int first_offset;
4200 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4203 first_offset = skb->data -
4204 (unsigned char *)page_address(page) +
4207 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4209 __skb_frag_set_page(frag, page);
4210 skb_frag_off_set(frag, first_offset);
4211 skb_frag_size_set(frag, first_size);
4213 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4214 /* We dont need to clear skbinfo->nr_frags here */
4216 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4217 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4222 delta_truesize = skb->truesize;
4223 if (offset > headlen) {
4224 unsigned int eat = offset - headlen;
4226 skb_frag_off_add(&skbinfo->frags[0], eat);
4227 skb_frag_size_sub(&skbinfo->frags[0], eat);
4228 skb->data_len -= eat;
4233 __skb_pull(skb, offset);
4235 if (NAPI_GRO_CB(p)->last == p)
4236 skb_shinfo(p)->frag_list = skb;
4238 NAPI_GRO_CB(p)->last->next = skb;
4239 NAPI_GRO_CB(p)->last = skb;
4240 __skb_header_release(skb);
4244 NAPI_GRO_CB(p)->count++;
4246 p->truesize += delta_truesize;
4249 lp->data_len += len;
4250 lp->truesize += delta_truesize;
4253 NAPI_GRO_CB(skb)->same_flow = 1;
4257 #ifdef CONFIG_SKB_EXTENSIONS
4258 #define SKB_EXT_ALIGN_VALUE 8
4259 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4261 static const u8 skb_ext_type_len[] = {
4262 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4263 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4266 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4268 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4269 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4271 #if IS_ENABLED(CONFIG_MPTCP)
4272 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4276 static __always_inline unsigned int skb_ext_total_length(void)
4278 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4279 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4280 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4283 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4285 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4286 skb_ext_type_len[TC_SKB_EXT] +
4288 #if IS_ENABLED(CONFIG_MPTCP)
4289 skb_ext_type_len[SKB_EXT_MPTCP] +
4294 static void skb_extensions_init(void)
4296 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4297 BUILD_BUG_ON(skb_ext_total_length() > 255);
4299 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4300 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4302 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4306 static void skb_extensions_init(void) {}
4309 void __init skb_init(void)
4311 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4312 sizeof(struct sk_buff),
4314 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4315 offsetof(struct sk_buff, cb),
4316 sizeof_field(struct sk_buff, cb),
4318 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4319 sizeof(struct sk_buff_fclones),
4321 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4323 skb_extensions_init();
4327 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4328 unsigned int recursion_level)
4330 int start = skb_headlen(skb);
4331 int i, copy = start - offset;
4332 struct sk_buff *frag_iter;
4335 if (unlikely(recursion_level >= 24))
4341 sg_set_buf(sg, skb->data + offset, copy);
4343 if ((len -= copy) == 0)
4348 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4351 WARN_ON(start > offset + len);
4353 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4354 if ((copy = end - offset) > 0) {
4355 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4356 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4361 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4362 skb_frag_off(frag) + offset - start);
4371 skb_walk_frags(skb, frag_iter) {
4374 WARN_ON(start > offset + len);
4376 end = start + frag_iter->len;
4377 if ((copy = end - offset) > 0) {
4378 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4383 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4384 copy, recursion_level + 1);
4385 if (unlikely(ret < 0))
4388 if ((len -= copy) == 0)
4399 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4400 * @skb: Socket buffer containing the buffers to be mapped
4401 * @sg: The scatter-gather list to map into
4402 * @offset: The offset into the buffer's contents to start mapping
4403 * @len: Length of buffer space to be mapped
4405 * Fill the specified scatter-gather list with mappings/pointers into a
4406 * region of the buffer space attached to a socket buffer. Returns either
4407 * the number of scatterlist items used, or -EMSGSIZE if the contents
4410 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4412 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4417 sg_mark_end(&sg[nsg - 1]);
4421 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4423 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4424 * sglist without mark the sg which contain last skb data as the end.
4425 * So the caller can mannipulate sg list as will when padding new data after
4426 * the first call without calling sg_unmark_end to expend sg list.
4428 * Scenario to use skb_to_sgvec_nomark:
4430 * 2. skb_to_sgvec_nomark(payload1)
4431 * 3. skb_to_sgvec_nomark(payload2)
4433 * This is equivalent to:
4435 * 2. skb_to_sgvec(payload1)
4437 * 4. skb_to_sgvec(payload2)
4439 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4440 * is more preferable.
4442 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4443 int offset, int len)
4445 return __skb_to_sgvec(skb, sg, offset, len, 0);
4447 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4452 * skb_cow_data - Check that a socket buffer's data buffers are writable
4453 * @skb: The socket buffer to check.
4454 * @tailbits: Amount of trailing space to be added
4455 * @trailer: Returned pointer to the skb where the @tailbits space begins
4457 * Make sure that the data buffers attached to a socket buffer are
4458 * writable. If they are not, private copies are made of the data buffers
4459 * and the socket buffer is set to use these instead.
4461 * If @tailbits is given, make sure that there is space to write @tailbits
4462 * bytes of data beyond current end of socket buffer. @trailer will be
4463 * set to point to the skb in which this space begins.
4465 * The number of scatterlist elements required to completely map the
4466 * COW'd and extended socket buffer will be returned.
4468 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4472 struct sk_buff *skb1, **skb_p;
4474 /* If skb is cloned or its head is paged, reallocate
4475 * head pulling out all the pages (pages are considered not writable
4476 * at the moment even if they are anonymous).
4478 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4479 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4482 /* Easy case. Most of packets will go this way. */
4483 if (!skb_has_frag_list(skb)) {
4484 /* A little of trouble, not enough of space for trailer.
4485 * This should not happen, when stack is tuned to generate
4486 * good frames. OK, on miss we reallocate and reserve even more
4487 * space, 128 bytes is fair. */
4489 if (skb_tailroom(skb) < tailbits &&
4490 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4498 /* Misery. We are in troubles, going to mincer fragments... */
4501 skb_p = &skb_shinfo(skb)->frag_list;
4504 while ((skb1 = *skb_p) != NULL) {
4507 /* The fragment is partially pulled by someone,
4508 * this can happen on input. Copy it and everything
4511 if (skb_shared(skb1))
4514 /* If the skb is the last, worry about trailer. */
4516 if (skb1->next == NULL && tailbits) {
4517 if (skb_shinfo(skb1)->nr_frags ||
4518 skb_has_frag_list(skb1) ||
4519 skb_tailroom(skb1) < tailbits)
4520 ntail = tailbits + 128;
4526 skb_shinfo(skb1)->nr_frags ||
4527 skb_has_frag_list(skb1)) {
4528 struct sk_buff *skb2;
4530 /* Fuck, we are miserable poor guys... */
4532 skb2 = skb_copy(skb1, GFP_ATOMIC);
4534 skb2 = skb_copy_expand(skb1,
4538 if (unlikely(skb2 == NULL))
4542 skb_set_owner_w(skb2, skb1->sk);
4544 /* Looking around. Are we still alive?
4545 * OK, link new skb, drop old one */
4547 skb2->next = skb1->next;
4554 skb_p = &skb1->next;
4559 EXPORT_SYMBOL_GPL(skb_cow_data);
4561 static void sock_rmem_free(struct sk_buff *skb)
4563 struct sock *sk = skb->sk;
4565 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4568 static void skb_set_err_queue(struct sk_buff *skb)
4570 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4571 * So, it is safe to (mis)use it to mark skbs on the error queue.
4573 skb->pkt_type = PACKET_OUTGOING;
4574 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4578 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4580 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4582 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4583 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4588 skb->destructor = sock_rmem_free;
4589 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4590 skb_set_err_queue(skb);
4592 /* before exiting rcu section, make sure dst is refcounted */
4595 skb_queue_tail(&sk->sk_error_queue, skb);
4596 if (!sock_flag(sk, SOCK_DEAD))
4597 sk->sk_error_report(sk);
4600 EXPORT_SYMBOL(sock_queue_err_skb);
4602 static bool is_icmp_err_skb(const struct sk_buff *skb)
4604 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4605 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4608 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4610 struct sk_buff_head *q = &sk->sk_error_queue;
4611 struct sk_buff *skb, *skb_next = NULL;
4612 bool icmp_next = false;
4613 unsigned long flags;
4615 spin_lock_irqsave(&q->lock, flags);
4616 skb = __skb_dequeue(q);
4617 if (skb && (skb_next = skb_peek(q))) {
4618 icmp_next = is_icmp_err_skb(skb_next);
4620 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4622 spin_unlock_irqrestore(&q->lock, flags);
4624 if (is_icmp_err_skb(skb) && !icmp_next)
4628 sk->sk_error_report(sk);
4632 EXPORT_SYMBOL(sock_dequeue_err_skb);
4635 * skb_clone_sk - create clone of skb, and take reference to socket
4636 * @skb: the skb to clone
4638 * This function creates a clone of a buffer that holds a reference on
4639 * sk_refcnt. Buffers created via this function are meant to be
4640 * returned using sock_queue_err_skb, or free via kfree_skb.
4642 * When passing buffers allocated with this function to sock_queue_err_skb
4643 * it is necessary to wrap the call with sock_hold/sock_put in order to
4644 * prevent the socket from being released prior to being enqueued on
4645 * the sk_error_queue.
4647 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4649 struct sock *sk = skb->sk;
4650 struct sk_buff *clone;
4652 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4655 clone = skb_clone(skb, GFP_ATOMIC);
4662 clone->destructor = sock_efree;
4666 EXPORT_SYMBOL(skb_clone_sk);
4668 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4673 struct sock_exterr_skb *serr;
4676 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4678 serr = SKB_EXT_ERR(skb);
4679 memset(serr, 0, sizeof(*serr));
4680 serr->ee.ee_errno = ENOMSG;
4681 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4682 serr->ee.ee_info = tstype;
4683 serr->opt_stats = opt_stats;
4684 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4685 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4686 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4687 if (sk->sk_protocol == IPPROTO_TCP &&
4688 sk->sk_type == SOCK_STREAM)
4689 serr->ee.ee_data -= sk->sk_tskey;
4692 err = sock_queue_err_skb(sk, skb);
4698 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4702 if (likely(sysctl_tstamp_allow_data || tsonly))
4705 read_lock_bh(&sk->sk_callback_lock);
4706 ret = sk->sk_socket && sk->sk_socket->file &&
4707 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4708 read_unlock_bh(&sk->sk_callback_lock);
4712 void skb_complete_tx_timestamp(struct sk_buff *skb,
4713 struct skb_shared_hwtstamps *hwtstamps)
4715 struct sock *sk = skb->sk;
4717 if (!skb_may_tx_timestamp(sk, false))
4720 /* Take a reference to prevent skb_orphan() from freeing the socket,
4721 * but only if the socket refcount is not zero.
4723 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4724 *skb_hwtstamps(skb) = *hwtstamps;
4725 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4733 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4735 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4736 const struct sk_buff *ack_skb,
4737 struct skb_shared_hwtstamps *hwtstamps,
4738 struct sock *sk, int tstype)
4740 struct sk_buff *skb;
4741 bool tsonly, opt_stats = false;
4746 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4747 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4750 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4751 if (!skb_may_tx_timestamp(sk, tsonly))
4756 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4757 sk->sk_protocol == IPPROTO_TCP &&
4758 sk->sk_type == SOCK_STREAM) {
4759 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
4764 skb = alloc_skb(0, GFP_ATOMIC);
4766 skb = skb_clone(orig_skb, GFP_ATOMIC);
4772 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4774 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4778 *skb_hwtstamps(skb) = *hwtstamps;
4780 skb->tstamp = ktime_get_real();
4782 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4784 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4786 void skb_tstamp_tx(struct sk_buff *orig_skb,
4787 struct skb_shared_hwtstamps *hwtstamps)
4789 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
4792 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4794 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4796 struct sock *sk = skb->sk;
4797 struct sock_exterr_skb *serr;
4800 skb->wifi_acked_valid = 1;
4801 skb->wifi_acked = acked;
4803 serr = SKB_EXT_ERR(skb);
4804 memset(serr, 0, sizeof(*serr));
4805 serr->ee.ee_errno = ENOMSG;
4806 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4808 /* Take a reference to prevent skb_orphan() from freeing the socket,
4809 * but only if the socket refcount is not zero.
4811 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4812 err = sock_queue_err_skb(sk, skb);
4818 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4821 * skb_partial_csum_set - set up and verify partial csum values for packet
4822 * @skb: the skb to set
4823 * @start: the number of bytes after skb->data to start checksumming.
4824 * @off: the offset from start to place the checksum.
4826 * For untrusted partially-checksummed packets, we need to make sure the values
4827 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4829 * This function checks and sets those values and skb->ip_summed: if this
4830 * returns false you should drop the packet.
4832 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4834 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4835 u32 csum_start = skb_headroom(skb) + (u32)start;
4837 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4838 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4839 start, off, skb_headroom(skb), skb_headlen(skb));
4842 skb->ip_summed = CHECKSUM_PARTIAL;
4843 skb->csum_start = csum_start;
4844 skb->csum_offset = off;
4845 skb_set_transport_header(skb, start);
4848 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4850 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4853 if (skb_headlen(skb) >= len)
4856 /* If we need to pullup then pullup to the max, so we
4857 * won't need to do it again.
4862 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4865 if (skb_headlen(skb) < len)
4871 #define MAX_TCP_HDR_LEN (15 * 4)
4873 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4874 typeof(IPPROTO_IP) proto,
4881 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4882 off + MAX_TCP_HDR_LEN);
4883 if (!err && !skb_partial_csum_set(skb, off,
4884 offsetof(struct tcphdr,
4887 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4890 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4891 off + sizeof(struct udphdr));
4892 if (!err && !skb_partial_csum_set(skb, off,
4893 offsetof(struct udphdr,
4896 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4899 return ERR_PTR(-EPROTO);
4902 /* This value should be large enough to cover a tagged ethernet header plus
4903 * maximally sized IP and TCP or UDP headers.
4905 #define MAX_IP_HDR_LEN 128
4907 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4916 err = skb_maybe_pull_tail(skb,
4917 sizeof(struct iphdr),
4922 if (ip_is_fragment(ip_hdr(skb)))
4925 off = ip_hdrlen(skb);
4932 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4934 return PTR_ERR(csum);
4937 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4940 ip_hdr(skb)->protocol, 0);
4947 /* This value should be large enough to cover a tagged ethernet header plus
4948 * an IPv6 header, all options, and a maximal TCP or UDP header.
4950 #define MAX_IPV6_HDR_LEN 256
4952 #define OPT_HDR(type, skb, off) \
4953 (type *)(skb_network_header(skb) + (off))
4955 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4968 off = sizeof(struct ipv6hdr);
4970 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4974 nexthdr = ipv6_hdr(skb)->nexthdr;
4976 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4977 while (off <= len && !done) {
4979 case IPPROTO_DSTOPTS:
4980 case IPPROTO_HOPOPTS:
4981 case IPPROTO_ROUTING: {
4982 struct ipv6_opt_hdr *hp;
4984 err = skb_maybe_pull_tail(skb,
4986 sizeof(struct ipv6_opt_hdr),
4991 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4992 nexthdr = hp->nexthdr;
4993 off += ipv6_optlen(hp);
4997 struct ip_auth_hdr *hp;
4999 err = skb_maybe_pull_tail(skb,
5001 sizeof(struct ip_auth_hdr),
5006 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5007 nexthdr = hp->nexthdr;
5008 off += ipv6_authlen(hp);
5011 case IPPROTO_FRAGMENT: {
5012 struct frag_hdr *hp;
5014 err = skb_maybe_pull_tail(skb,
5016 sizeof(struct frag_hdr),
5021 hp = OPT_HDR(struct frag_hdr, skb, off);
5023 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5026 nexthdr = hp->nexthdr;
5027 off += sizeof(struct frag_hdr);
5038 if (!done || fragment)
5041 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5043 return PTR_ERR(csum);
5046 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5047 &ipv6_hdr(skb)->daddr,
5048 skb->len - off, nexthdr, 0);
5056 * skb_checksum_setup - set up partial checksum offset
5057 * @skb: the skb to set up
5058 * @recalculate: if true the pseudo-header checksum will be recalculated
5060 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5064 switch (skb->protocol) {
5065 case htons(ETH_P_IP):
5066 err = skb_checksum_setup_ipv4(skb, recalculate);
5069 case htons(ETH_P_IPV6):
5070 err = skb_checksum_setup_ipv6(skb, recalculate);
5080 EXPORT_SYMBOL(skb_checksum_setup);
5083 * skb_checksum_maybe_trim - maybe trims the given skb
5084 * @skb: the skb to check
5085 * @transport_len: the data length beyond the network header
5087 * Checks whether the given skb has data beyond the given transport length.
5088 * If so, returns a cloned skb trimmed to this transport length.
5089 * Otherwise returns the provided skb. Returns NULL in error cases
5090 * (e.g. transport_len exceeds skb length or out-of-memory).
5092 * Caller needs to set the skb transport header and free any returned skb if it
5093 * differs from the provided skb.
5095 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5096 unsigned int transport_len)
5098 struct sk_buff *skb_chk;
5099 unsigned int len = skb_transport_offset(skb) + transport_len;
5104 else if (skb->len == len)
5107 skb_chk = skb_clone(skb, GFP_ATOMIC);
5111 ret = pskb_trim_rcsum(skb_chk, len);
5121 * skb_checksum_trimmed - validate checksum of an skb
5122 * @skb: the skb to check
5123 * @transport_len: the data length beyond the network header
5124 * @skb_chkf: checksum function to use
5126 * Applies the given checksum function skb_chkf to the provided skb.
5127 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5129 * If the skb has data beyond the given transport length, then a
5130 * trimmed & cloned skb is checked and returned.
5132 * Caller needs to set the skb transport header and free any returned skb if it
5133 * differs from the provided skb.
5135 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5136 unsigned int transport_len,
5137 __sum16(*skb_chkf)(struct sk_buff *skb))
5139 struct sk_buff *skb_chk;
5140 unsigned int offset = skb_transport_offset(skb);
5143 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5147 if (!pskb_may_pull(skb_chk, offset))
5150 skb_pull_rcsum(skb_chk, offset);
5151 ret = skb_chkf(skb_chk);
5152 skb_push_rcsum(skb_chk, offset);
5160 if (skb_chk && skb_chk != skb)
5166 EXPORT_SYMBOL(skb_checksum_trimmed);
5168 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5170 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5173 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5175 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5178 skb_release_head_state(skb);
5179 kmem_cache_free(skbuff_head_cache, skb);
5184 EXPORT_SYMBOL(kfree_skb_partial);
5187 * skb_try_coalesce - try to merge skb to prior one
5189 * @from: buffer to add
5190 * @fragstolen: pointer to boolean
5191 * @delta_truesize: how much more was allocated than was requested
5193 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5194 bool *fragstolen, int *delta_truesize)
5196 struct skb_shared_info *to_shinfo, *from_shinfo;
5197 int i, delta, len = from->len;
5199 *fragstolen = false;
5204 if (len <= skb_tailroom(to)) {
5206 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5207 *delta_truesize = 0;
5211 to_shinfo = skb_shinfo(to);
5212 from_shinfo = skb_shinfo(from);
5213 if (to_shinfo->frag_list || from_shinfo->frag_list)
5215 if (skb_zcopy(to) || skb_zcopy(from))
5218 if (skb_headlen(from) != 0) {
5220 unsigned int offset;
5222 if (to_shinfo->nr_frags +
5223 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5226 if (skb_head_is_locked(from))
5229 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5231 page = virt_to_head_page(from->head);
5232 offset = from->data - (unsigned char *)page_address(page);
5234 skb_fill_page_desc(to, to_shinfo->nr_frags,
5235 page, offset, skb_headlen(from));
5238 if (to_shinfo->nr_frags +
5239 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5242 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5245 WARN_ON_ONCE(delta < len);
5247 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5249 from_shinfo->nr_frags * sizeof(skb_frag_t));
5250 to_shinfo->nr_frags += from_shinfo->nr_frags;
5252 if (!skb_cloned(from))
5253 from_shinfo->nr_frags = 0;
5255 /* if the skb is not cloned this does nothing
5256 * since we set nr_frags to 0.
5258 for (i = 0; i < from_shinfo->nr_frags; i++)
5259 __skb_frag_ref(&from_shinfo->frags[i]);
5261 to->truesize += delta;
5263 to->data_len += len;
5265 *delta_truesize = delta;
5268 EXPORT_SYMBOL(skb_try_coalesce);
5271 * skb_scrub_packet - scrub an skb
5273 * @skb: buffer to clean
5274 * @xnet: packet is crossing netns
5276 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5277 * into/from a tunnel. Some information have to be cleared during these
5279 * skb_scrub_packet can also be used to clean a skb before injecting it in
5280 * another namespace (@xnet == true). We have to clear all information in the
5281 * skb that could impact namespace isolation.
5283 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5285 skb->pkt_type = PACKET_HOST;
5291 nf_reset_trace(skb);
5293 #ifdef CONFIG_NET_SWITCHDEV
5294 skb->offload_fwd_mark = 0;
5295 skb->offload_l3_fwd_mark = 0;
5305 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5308 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5312 * skb_gso_transport_seglen is used to determine the real size of the
5313 * individual segments, including Layer4 headers (TCP/UDP).
5315 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5317 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5319 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5320 unsigned int thlen = 0;
5322 if (skb->encapsulation) {
5323 thlen = skb_inner_transport_header(skb) -
5324 skb_transport_header(skb);
5326 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5327 thlen += inner_tcp_hdrlen(skb);
5328 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5329 thlen = tcp_hdrlen(skb);
5330 } else if (unlikely(skb_is_gso_sctp(skb))) {
5331 thlen = sizeof(struct sctphdr);
5332 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5333 thlen = sizeof(struct udphdr);
5335 /* UFO sets gso_size to the size of the fragmentation
5336 * payload, i.e. the size of the L4 (UDP) header is already
5339 return thlen + shinfo->gso_size;
5343 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5347 * skb_gso_network_seglen is used to determine the real size of the
5348 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5350 * The MAC/L2 header is not accounted for.
5352 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5354 unsigned int hdr_len = skb_transport_header(skb) -
5355 skb_network_header(skb);
5357 return hdr_len + skb_gso_transport_seglen(skb);
5361 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5365 * skb_gso_mac_seglen is used to determine the real size of the
5366 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5367 * headers (TCP/UDP).
5369 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5371 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5373 return hdr_len + skb_gso_transport_seglen(skb);
5377 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5379 * There are a couple of instances where we have a GSO skb, and we
5380 * want to determine what size it would be after it is segmented.
5382 * We might want to check:
5383 * - L3+L4+payload size (e.g. IP forwarding)
5384 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5386 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5390 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5391 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5393 * @max_len: The maximum permissible length.
5395 * Returns true if the segmented length <= max length.
5397 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5398 unsigned int seg_len,
5399 unsigned int max_len) {
5400 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5401 const struct sk_buff *iter;
5403 if (shinfo->gso_size != GSO_BY_FRAGS)
5404 return seg_len <= max_len;
5406 /* Undo this so we can re-use header sizes */
5407 seg_len -= GSO_BY_FRAGS;
5409 skb_walk_frags(skb, iter) {
5410 if (seg_len + skb_headlen(iter) > max_len)
5418 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5421 * @mtu: MTU to validate against
5423 * skb_gso_validate_network_len validates if a given skb will fit a
5424 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5427 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5429 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5431 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5434 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5437 * @len: length to validate against
5439 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5440 * length once split, including L2, L3 and L4 headers and the payload.
5442 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5444 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5446 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5448 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5450 int mac_len, meta_len;
5453 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5458 mac_len = skb->data - skb_mac_header(skb);
5459 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5460 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5461 mac_len - VLAN_HLEN - ETH_TLEN);
5464 meta_len = skb_metadata_len(skb);
5466 meta = skb_metadata_end(skb) - meta_len;
5467 memmove(meta + VLAN_HLEN, meta, meta_len);
5470 skb->mac_header += VLAN_HLEN;
5474 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5476 struct vlan_hdr *vhdr;
5479 if (unlikely(skb_vlan_tag_present(skb))) {
5480 /* vlan_tci is already set-up so leave this for another time */
5484 skb = skb_share_check(skb, GFP_ATOMIC);
5487 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5488 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5491 vhdr = (struct vlan_hdr *)skb->data;
5492 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5493 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5495 skb_pull_rcsum(skb, VLAN_HLEN);
5496 vlan_set_encap_proto(skb, vhdr);
5498 skb = skb_reorder_vlan_header(skb);
5502 skb_reset_network_header(skb);
5503 if (!skb_transport_header_was_set(skb))
5504 skb_reset_transport_header(skb);
5505 skb_reset_mac_len(skb);
5513 EXPORT_SYMBOL(skb_vlan_untag);
5515 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5517 if (!pskb_may_pull(skb, write_len))
5520 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5523 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5525 EXPORT_SYMBOL(skb_ensure_writable);
5527 /* remove VLAN header from packet and update csum accordingly.
5528 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5530 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5532 struct vlan_hdr *vhdr;
5533 int offset = skb->data - skb_mac_header(skb);
5536 if (WARN_ONCE(offset,
5537 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5542 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5546 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5548 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5549 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5551 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5552 __skb_pull(skb, VLAN_HLEN);
5554 vlan_set_encap_proto(skb, vhdr);
5555 skb->mac_header += VLAN_HLEN;
5557 if (skb_network_offset(skb) < ETH_HLEN)
5558 skb_set_network_header(skb, ETH_HLEN);
5560 skb_reset_mac_len(skb);
5564 EXPORT_SYMBOL(__skb_vlan_pop);
5566 /* Pop a vlan tag either from hwaccel or from payload.
5567 * Expects skb->data at mac header.
5569 int skb_vlan_pop(struct sk_buff *skb)
5575 if (likely(skb_vlan_tag_present(skb))) {
5576 __vlan_hwaccel_clear_tag(skb);
5578 if (unlikely(!eth_type_vlan(skb->protocol)))
5581 err = __skb_vlan_pop(skb, &vlan_tci);
5585 /* move next vlan tag to hw accel tag */
5586 if (likely(!eth_type_vlan(skb->protocol)))
5589 vlan_proto = skb->protocol;
5590 err = __skb_vlan_pop(skb, &vlan_tci);
5594 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5597 EXPORT_SYMBOL(skb_vlan_pop);
5599 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5600 * Expects skb->data at mac header.
5602 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5604 if (skb_vlan_tag_present(skb)) {
5605 int offset = skb->data - skb_mac_header(skb);
5608 if (WARN_ONCE(offset,
5609 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5614 err = __vlan_insert_tag(skb, skb->vlan_proto,
5615 skb_vlan_tag_get(skb));
5619 skb->protocol = skb->vlan_proto;
5620 skb->mac_len += VLAN_HLEN;
5622 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5624 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5627 EXPORT_SYMBOL(skb_vlan_push);
5630 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5632 * @skb: Socket buffer to modify
5634 * Drop the Ethernet header of @skb.
5636 * Expects that skb->data points to the mac header and that no VLAN tags are
5639 * Returns 0 on success, -errno otherwise.
5641 int skb_eth_pop(struct sk_buff *skb)
5643 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5644 skb_network_offset(skb) < ETH_HLEN)
5647 skb_pull_rcsum(skb, ETH_HLEN);
5648 skb_reset_mac_header(skb);
5649 skb_reset_mac_len(skb);
5653 EXPORT_SYMBOL(skb_eth_pop);
5656 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5658 * @skb: Socket buffer to modify
5659 * @dst: Destination MAC address of the new header
5660 * @src: Source MAC address of the new header
5662 * Prepend @skb with a new Ethernet header.
5664 * Expects that skb->data points to the mac header, which must be empty.
5666 * Returns 0 on success, -errno otherwise.
5668 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5669 const unsigned char *src)
5674 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5677 err = skb_cow_head(skb, sizeof(*eth));
5681 skb_push(skb, sizeof(*eth));
5682 skb_reset_mac_header(skb);
5683 skb_reset_mac_len(skb);
5686 ether_addr_copy(eth->h_dest, dst);
5687 ether_addr_copy(eth->h_source, src);
5688 eth->h_proto = skb->protocol;
5690 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5694 EXPORT_SYMBOL(skb_eth_push);
5696 /* Update the ethertype of hdr and the skb csum value if required. */
5697 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5700 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5701 __be16 diff[] = { ~hdr->h_proto, ethertype };
5703 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5706 hdr->h_proto = ethertype;
5710 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5714 * @mpls_lse: MPLS label stack entry to push
5715 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5716 * @mac_len: length of the MAC header
5717 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5720 * Expects skb->data at mac header.
5722 * Returns 0 on success, -errno otherwise.
5724 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5725 int mac_len, bool ethernet)
5727 struct mpls_shim_hdr *lse;
5730 if (unlikely(!eth_p_mpls(mpls_proto)))
5733 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5734 if (skb->encapsulation)
5737 err = skb_cow_head(skb, MPLS_HLEN);
5741 if (!skb->inner_protocol) {
5742 skb_set_inner_network_header(skb, skb_network_offset(skb));
5743 skb_set_inner_protocol(skb, skb->protocol);
5746 skb_push(skb, MPLS_HLEN);
5747 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5749 skb_reset_mac_header(skb);
5750 skb_set_network_header(skb, mac_len);
5751 skb_reset_mac_len(skb);
5753 lse = mpls_hdr(skb);
5754 lse->label_stack_entry = mpls_lse;
5755 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5757 if (ethernet && mac_len >= ETH_HLEN)
5758 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5759 skb->protocol = mpls_proto;
5763 EXPORT_SYMBOL_GPL(skb_mpls_push);
5766 * skb_mpls_pop() - pop the outermost MPLS header
5769 * @next_proto: ethertype of header after popped MPLS header
5770 * @mac_len: length of the MAC header
5771 * @ethernet: flag to indicate if the packet is ethernet
5773 * Expects skb->data at mac header.
5775 * Returns 0 on success, -errno otherwise.
5777 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5782 if (unlikely(!eth_p_mpls(skb->protocol)))
5785 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5789 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5790 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5793 __skb_pull(skb, MPLS_HLEN);
5794 skb_reset_mac_header(skb);
5795 skb_set_network_header(skb, mac_len);
5797 if (ethernet && mac_len >= ETH_HLEN) {
5800 /* use mpls_hdr() to get ethertype to account for VLANs. */
5801 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5802 skb_mod_eth_type(skb, hdr, next_proto);
5804 skb->protocol = next_proto;
5808 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5811 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5814 * @mpls_lse: new MPLS label stack entry to update to
5816 * Expects skb->data at mac header.
5818 * Returns 0 on success, -errno otherwise.
5820 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5824 if (unlikely(!eth_p_mpls(skb->protocol)))
5827 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5831 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5832 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5834 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5837 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5841 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5844 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5848 * Expects skb->data at mac header.
5850 * Returns 0 on success, -errno otherwise.
5852 int skb_mpls_dec_ttl(struct sk_buff *skb)
5857 if (unlikely(!eth_p_mpls(skb->protocol)))
5860 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5863 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5864 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5868 lse &= ~MPLS_LS_TTL_MASK;
5869 lse |= ttl << MPLS_LS_TTL_SHIFT;
5871 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5873 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5876 * alloc_skb_with_frags - allocate skb with page frags
5878 * @header_len: size of linear part
5879 * @data_len: needed length in frags
5880 * @max_page_order: max page order desired.
5881 * @errcode: pointer to error code if any
5882 * @gfp_mask: allocation mask
5884 * This can be used to allocate a paged skb, given a maximal order for frags.
5886 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5887 unsigned long data_len,
5892 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5893 unsigned long chunk;
5894 struct sk_buff *skb;
5898 *errcode = -EMSGSIZE;
5899 /* Note this test could be relaxed, if we succeed to allocate
5900 * high order pages...
5902 if (npages > MAX_SKB_FRAGS)
5905 *errcode = -ENOBUFS;
5906 skb = alloc_skb(header_len, gfp_mask);
5910 skb->truesize += npages << PAGE_SHIFT;
5912 for (i = 0; npages > 0; i++) {
5913 int order = max_page_order;
5916 if (npages >= 1 << order) {
5917 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5923 /* Do not retry other high order allocations */
5929 page = alloc_page(gfp_mask);
5933 chunk = min_t(unsigned long, data_len,
5934 PAGE_SIZE << order);
5935 skb_fill_page_desc(skb, i, page, 0, chunk);
5937 npages -= 1 << order;
5945 EXPORT_SYMBOL(alloc_skb_with_frags);
5947 /* carve out the first off bytes from skb when off < headlen */
5948 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5949 const int headlen, gfp_t gfp_mask)
5952 int size = skb_end_offset(skb);
5953 int new_hlen = headlen - off;
5956 size = SKB_DATA_ALIGN(size);
5958 if (skb_pfmemalloc(skb))
5959 gfp_mask |= __GFP_MEMALLOC;
5960 data = kmalloc_reserve(size +
5961 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5962 gfp_mask, NUMA_NO_NODE, NULL);
5966 size = SKB_WITH_OVERHEAD(ksize(data));
5968 /* Copy real data, and all frags */
5969 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5972 memcpy((struct skb_shared_info *)(data + size),
5974 offsetof(struct skb_shared_info,
5975 frags[skb_shinfo(skb)->nr_frags]));
5976 if (skb_cloned(skb)) {
5977 /* drop the old head gracefully */
5978 if (skb_orphan_frags(skb, gfp_mask)) {
5982 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5983 skb_frag_ref(skb, i);
5984 if (skb_has_frag_list(skb))
5985 skb_clone_fraglist(skb);
5986 skb_release_data(skb);
5988 /* we can reuse existing recount- all we did was
5997 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6000 skb->end = skb->head + size;
6002 skb_set_tail_pointer(skb, skb_headlen(skb));
6003 skb_headers_offset_update(skb, 0);
6007 atomic_set(&skb_shinfo(skb)->dataref, 1);
6012 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6014 /* carve out the first eat bytes from skb's frag_list. May recurse into
6017 static int pskb_carve_frag_list(struct sk_buff *skb,
6018 struct skb_shared_info *shinfo, int eat,
6021 struct sk_buff *list = shinfo->frag_list;
6022 struct sk_buff *clone = NULL;
6023 struct sk_buff *insp = NULL;
6027 pr_err("Not enough bytes to eat. Want %d\n", eat);
6030 if (list->len <= eat) {
6031 /* Eaten as whole. */
6036 /* Eaten partially. */
6037 if (skb_shared(list)) {
6038 clone = skb_clone(list, gfp_mask);
6044 /* This may be pulled without problems. */
6047 if (pskb_carve(list, eat, gfp_mask) < 0) {
6055 /* Free pulled out fragments. */
6056 while ((list = shinfo->frag_list) != insp) {
6057 shinfo->frag_list = list->next;
6060 /* And insert new clone at head. */
6063 shinfo->frag_list = clone;
6068 /* carve off first len bytes from skb. Split line (off) is in the
6069 * non-linear part of skb
6071 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6072 int pos, gfp_t gfp_mask)
6075 int size = skb_end_offset(skb);
6077 const int nfrags = skb_shinfo(skb)->nr_frags;
6078 struct skb_shared_info *shinfo;
6080 size = SKB_DATA_ALIGN(size);
6082 if (skb_pfmemalloc(skb))
6083 gfp_mask |= __GFP_MEMALLOC;
6084 data = kmalloc_reserve(size +
6085 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6086 gfp_mask, NUMA_NO_NODE, NULL);
6090 size = SKB_WITH_OVERHEAD(ksize(data));
6092 memcpy((struct skb_shared_info *)(data + size),
6093 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6094 if (skb_orphan_frags(skb, gfp_mask)) {
6098 shinfo = (struct skb_shared_info *)(data + size);
6099 for (i = 0; i < nfrags; i++) {
6100 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6102 if (pos + fsize > off) {
6103 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6107 * We have two variants in this case:
6108 * 1. Move all the frag to the second
6109 * part, if it is possible. F.e.
6110 * this approach is mandatory for TUX,
6111 * where splitting is expensive.
6112 * 2. Split is accurately. We make this.
6114 skb_frag_off_add(&shinfo->frags[0], off - pos);
6115 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6117 skb_frag_ref(skb, i);
6122 shinfo->nr_frags = k;
6123 if (skb_has_frag_list(skb))
6124 skb_clone_fraglist(skb);
6126 /* split line is in frag list */
6127 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6128 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6129 if (skb_has_frag_list(skb))
6130 kfree_skb_list(skb_shinfo(skb)->frag_list);
6134 skb_release_data(skb);
6139 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6142 skb->end = skb->head + size;
6144 skb_reset_tail_pointer(skb);
6145 skb_headers_offset_update(skb, 0);
6150 skb->data_len = skb->len;
6151 atomic_set(&skb_shinfo(skb)->dataref, 1);
6155 /* remove len bytes from the beginning of the skb */
6156 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6158 int headlen = skb_headlen(skb);
6161 return pskb_carve_inside_header(skb, len, headlen, gfp);
6163 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6166 /* Extract to_copy bytes starting at off from skb, and return this in
6169 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6170 int to_copy, gfp_t gfp)
6172 struct sk_buff *clone = skb_clone(skb, gfp);
6177 if (pskb_carve(clone, off, gfp) < 0 ||
6178 pskb_trim(clone, to_copy)) {
6184 EXPORT_SYMBOL(pskb_extract);
6187 * skb_condense - try to get rid of fragments/frag_list if possible
6190 * Can be used to save memory before skb is added to a busy queue.
6191 * If packet has bytes in frags and enough tail room in skb->head,
6192 * pull all of them, so that we can free the frags right now and adjust
6195 * We do not reallocate skb->head thus can not fail.
6196 * Caller must re-evaluate skb->truesize if needed.
6198 void skb_condense(struct sk_buff *skb)
6200 if (skb->data_len) {
6201 if (skb->data_len > skb->end - skb->tail ||
6205 /* Nice, we can free page frag(s) right now */
6206 __pskb_pull_tail(skb, skb->data_len);
6208 /* At this point, skb->truesize might be over estimated,
6209 * because skb had a fragment, and fragments do not tell
6211 * When we pulled its content into skb->head, fragment
6212 * was freed, but __pskb_pull_tail() could not possibly
6213 * adjust skb->truesize, not knowing the frag truesize.
6215 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6218 #ifdef CONFIG_SKB_EXTENSIONS
6219 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6221 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6225 * __skb_ext_alloc - allocate a new skb extensions storage
6227 * @flags: See kmalloc().
6229 * Returns the newly allocated pointer. The pointer can later attached to a
6230 * skb via __skb_ext_set().
6231 * Note: caller must handle the skb_ext as an opaque data.
6233 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6235 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6238 memset(new->offset, 0, sizeof(new->offset));
6239 refcount_set(&new->refcnt, 1);
6245 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6246 unsigned int old_active)
6248 struct skb_ext *new;
6250 if (refcount_read(&old->refcnt) == 1)
6253 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6257 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6258 refcount_set(&new->refcnt, 1);
6261 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6262 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6265 for (i = 0; i < sp->len; i++)
6266 xfrm_state_hold(sp->xvec[i]);
6274 * __skb_ext_set - attach the specified extension storage to this skb
6277 * @ext: extension storage previously allocated via __skb_ext_alloc()
6279 * Existing extensions, if any, are cleared.
6281 * Returns the pointer to the extension.
6283 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6284 struct skb_ext *ext)
6286 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6289 newlen = newoff + skb_ext_type_len[id];
6290 ext->chunks = newlen;
6291 ext->offset[id] = newoff;
6292 skb->extensions = ext;
6293 skb->active_extensions = 1 << id;
6294 return skb_ext_get_ptr(ext, id);
6298 * skb_ext_add - allocate space for given extension, COW if needed
6300 * @id: extension to allocate space for
6302 * Allocates enough space for the given extension.
6303 * If the extension is already present, a pointer to that extension
6306 * If the skb was cloned, COW applies and the returned memory can be
6307 * modified without changing the extension space of clones buffers.
6309 * Returns pointer to the extension or NULL on allocation failure.
6311 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6313 struct skb_ext *new, *old = NULL;
6314 unsigned int newlen, newoff;
6316 if (skb->active_extensions) {
6317 old = skb->extensions;
6319 new = skb_ext_maybe_cow(old, skb->active_extensions);
6323 if (__skb_ext_exist(new, id))
6326 newoff = new->chunks;
6328 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6330 new = __skb_ext_alloc(GFP_ATOMIC);
6335 newlen = newoff + skb_ext_type_len[id];
6336 new->chunks = newlen;
6337 new->offset[id] = newoff;
6339 skb->extensions = new;
6340 skb->active_extensions |= 1 << id;
6341 return skb_ext_get_ptr(new, id);
6343 EXPORT_SYMBOL(skb_ext_add);
6346 static void skb_ext_put_sp(struct sec_path *sp)
6350 for (i = 0; i < sp->len; i++)
6351 xfrm_state_put(sp->xvec[i]);
6355 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6357 struct skb_ext *ext = skb->extensions;
6359 skb->active_extensions &= ~(1 << id);
6360 if (skb->active_extensions == 0) {
6361 skb->extensions = NULL;
6364 } else if (id == SKB_EXT_SEC_PATH &&
6365 refcount_read(&ext->refcnt) == 1) {
6366 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6373 EXPORT_SYMBOL(__skb_ext_del);
6375 void __skb_ext_put(struct skb_ext *ext)
6377 /* If this is last clone, nothing can increment
6378 * it after check passes. Avoids one atomic op.
6380 if (refcount_read(&ext->refcnt) == 1)
6383 if (!refcount_dec_and_test(&ext->refcnt))
6387 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6388 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6391 kmem_cache_free(skbuff_ext_cache, ext);
6393 EXPORT_SYMBOL(__skb_ext_put);
6394 #endif /* CONFIG_SKB_EXTENSIONS */