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>
63 #include <linux/kcov.h>
65 #include <net/protocol.h>
68 #include <net/checksum.h>
69 #include <net/ip6_checksum.h>
72 #include <net/mptcp.h>
74 #include <linux/uaccess.h>
75 #include <trace/events/skb.h>
76 #include <linux/highmem.h>
77 #include <linux/capability.h>
78 #include <linux/user_namespace.h>
79 #include <linux/indirect_call_wrapper.h>
83 struct kmem_cache *skbuff_head_cache __ro_after_init;
84 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
85 #ifdef CONFIG_SKB_EXTENSIONS
86 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
88 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
89 EXPORT_SYMBOL(sysctl_max_skb_frags);
92 * skb_panic - private function for out-of-line support
96 * @msg: skb_over_panic or skb_under_panic
98 * Out-of-line support for skb_put() and skb_push().
99 * Called via the wrapper skb_over_panic() or skb_under_panic().
100 * Keep out of line to prevent kernel bloat.
101 * __builtin_return_address is not used because it is not always reliable.
103 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
106 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
107 msg, addr, skb->len, sz, skb->head, skb->data,
108 (unsigned long)skb->tail, (unsigned long)skb->end,
109 skb->dev ? skb->dev->name : "<NULL>");
113 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
115 skb_panic(skb, sz, addr, __func__);
118 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
120 skb_panic(skb, sz, addr, __func__);
124 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
125 * the caller if emergency pfmemalloc reserves are being used. If it is and
126 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
127 * may be used. Otherwise, the packet data may be discarded until enough
130 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
131 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
133 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
134 unsigned long ip, bool *pfmemalloc)
137 bool ret_pfmemalloc = false;
140 * Try a regular allocation, when that fails and we're not entitled
141 * to the reserves, fail.
143 obj = kmalloc_node_track_caller(size,
144 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
146 if (obj || !(gfp_pfmemalloc_allowed(flags)))
149 /* Try again but now we are using pfmemalloc reserves */
150 ret_pfmemalloc = true;
151 obj = kmalloc_node_track_caller(size, flags, node);
155 *pfmemalloc = ret_pfmemalloc;
160 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
161 * 'private' fields and also do memory statistics to find all the
167 * __alloc_skb - allocate a network buffer
168 * @size: size to allocate
169 * @gfp_mask: allocation mask
170 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
171 * instead of head cache and allocate a cloned (child) skb.
172 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
173 * allocations in case the data is required for writeback
174 * @node: numa node to allocate memory on
176 * Allocate a new &sk_buff. The returned buffer has no headroom and a
177 * tail room of at least size bytes. The object has a reference count
178 * of one. The return is the buffer. On a failure the return is %NULL.
180 * Buffers may only be allocated from interrupts using a @gfp_mask of
183 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
186 struct kmem_cache *cache;
187 struct skb_shared_info *shinfo;
192 cache = (flags & SKB_ALLOC_FCLONE)
193 ? skbuff_fclone_cache : skbuff_head_cache;
195 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
196 gfp_mask |= __GFP_MEMALLOC;
199 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
204 /* We do our best to align skb_shared_info on a separate cache
205 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
206 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
207 * Both skb->head and skb_shared_info are cache line aligned.
209 size = SKB_DATA_ALIGN(size);
210 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
211 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
214 /* kmalloc(size) might give us more room than requested.
215 * Put skb_shared_info exactly at the end of allocated zone,
216 * to allow max possible filling before reallocation.
218 size = SKB_WITH_OVERHEAD(ksize(data));
219 prefetchw(data + size);
222 * Only clear those fields we need to clear, not those that we will
223 * actually initialise below. Hence, don't put any more fields after
224 * the tail pointer in struct sk_buff!
226 memset(skb, 0, offsetof(struct sk_buff, tail));
227 /* Account for allocated memory : skb + skb->head */
228 skb->truesize = SKB_TRUESIZE(size);
229 skb->pfmemalloc = pfmemalloc;
230 refcount_set(&skb->users, 1);
233 skb_reset_tail_pointer(skb);
234 skb->end = skb->tail + size;
235 skb->mac_header = (typeof(skb->mac_header))~0U;
236 skb->transport_header = (typeof(skb->transport_header))~0U;
238 /* make sure we initialize shinfo sequentially */
239 shinfo = skb_shinfo(skb);
240 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
241 atomic_set(&shinfo->dataref, 1);
243 if (flags & SKB_ALLOC_FCLONE) {
244 struct sk_buff_fclones *fclones;
246 fclones = container_of(skb, struct sk_buff_fclones, skb1);
248 skb->fclone = SKB_FCLONE_ORIG;
249 refcount_set(&fclones->fclone_ref, 1);
251 fclones->skb2.fclone = SKB_FCLONE_CLONE;
254 skb_set_kcov_handle(skb, kcov_common_handle());
259 kmem_cache_free(cache, skb);
263 EXPORT_SYMBOL(__alloc_skb);
265 /* Caller must provide SKB that is memset cleared */
266 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
267 void *data, unsigned int frag_size)
269 struct skb_shared_info *shinfo;
270 unsigned int size = frag_size ? : ksize(data);
272 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
274 /* Assumes caller memset cleared SKB */
275 skb->truesize = SKB_TRUESIZE(size);
276 refcount_set(&skb->users, 1);
279 skb_reset_tail_pointer(skb);
280 skb->end = skb->tail + size;
281 skb->mac_header = (typeof(skb->mac_header))~0U;
282 skb->transport_header = (typeof(skb->transport_header))~0U;
284 /* make sure we initialize shinfo sequentially */
285 shinfo = skb_shinfo(skb);
286 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
287 atomic_set(&shinfo->dataref, 1);
289 skb_set_kcov_handle(skb, kcov_common_handle());
295 * __build_skb - build a network buffer
296 * @data: data buffer provided by caller
297 * @frag_size: size of data, or 0 if head was kmalloced
299 * Allocate a new &sk_buff. Caller provides space holding head and
300 * skb_shared_info. @data must have been allocated by kmalloc() only if
301 * @frag_size is 0, otherwise data should come from the page allocator
303 * The return is the new skb buffer.
304 * On a failure the return is %NULL, and @data is not freed.
306 * Before IO, driver allocates only data buffer where NIC put incoming frame
307 * Driver should add room at head (NET_SKB_PAD) and
308 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
309 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
310 * before giving packet to stack.
311 * RX rings only contains data buffers, not full skbs.
313 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
317 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
321 memset(skb, 0, offsetof(struct sk_buff, tail));
323 return __build_skb_around(skb, data, frag_size);
326 /* build_skb() is wrapper over __build_skb(), that specifically
327 * takes care of skb->head and skb->pfmemalloc
328 * This means that if @frag_size is not zero, then @data must be backed
329 * by a page fragment, not kmalloc() or vmalloc()
331 struct sk_buff *build_skb(void *data, unsigned int frag_size)
333 struct sk_buff *skb = __build_skb(data, frag_size);
335 if (skb && frag_size) {
337 if (page_is_pfmemalloc(virt_to_head_page(data)))
342 EXPORT_SYMBOL(build_skb);
345 * build_skb_around - build a network buffer around provided skb
346 * @skb: sk_buff provide by caller, must be memset cleared
347 * @data: data buffer provided by caller
348 * @frag_size: size of data, or 0 if head was kmalloced
350 struct sk_buff *build_skb_around(struct sk_buff *skb,
351 void *data, unsigned int frag_size)
356 skb = __build_skb_around(skb, data, frag_size);
358 if (skb && frag_size) {
360 if (page_is_pfmemalloc(virt_to_head_page(data)))
365 EXPORT_SYMBOL(build_skb_around);
367 #define NAPI_SKB_CACHE_SIZE 64
369 struct napi_alloc_cache {
370 struct page_frag_cache page;
371 unsigned int skb_count;
372 void *skb_cache[NAPI_SKB_CACHE_SIZE];
375 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
376 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
378 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
380 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
382 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
385 void *napi_alloc_frag(unsigned int fragsz)
387 fragsz = SKB_DATA_ALIGN(fragsz);
389 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
391 EXPORT_SYMBOL(napi_alloc_frag);
394 * netdev_alloc_frag - allocate a page fragment
395 * @fragsz: fragment size
397 * Allocates a frag from a page for receive buffer.
398 * Uses GFP_ATOMIC allocations.
400 void *netdev_alloc_frag(unsigned int fragsz)
402 struct page_frag_cache *nc;
405 fragsz = SKB_DATA_ALIGN(fragsz);
406 if (in_irq() || irqs_disabled()) {
407 nc = this_cpu_ptr(&netdev_alloc_cache);
408 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
411 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
416 EXPORT_SYMBOL(netdev_alloc_frag);
419 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
420 * @dev: network device to receive on
421 * @len: length to allocate
422 * @gfp_mask: get_free_pages mask, passed to alloc_skb
424 * Allocate a new &sk_buff and assign it a usage count of one. The
425 * buffer has NET_SKB_PAD headroom built in. Users should allocate
426 * the headroom they think they need without accounting for the
427 * built in space. The built in space is used for optimisations.
429 * %NULL is returned if there is no free memory.
431 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
434 struct page_frag_cache *nc;
441 /* If requested length is either too small or too big,
442 * we use kmalloc() for skb->head allocation.
444 if (len <= SKB_WITH_OVERHEAD(1024) ||
445 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
446 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
447 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
453 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
454 len = SKB_DATA_ALIGN(len);
456 if (sk_memalloc_socks())
457 gfp_mask |= __GFP_MEMALLOC;
459 if (in_irq() || irqs_disabled()) {
460 nc = this_cpu_ptr(&netdev_alloc_cache);
461 data = page_frag_alloc(nc, len, gfp_mask);
462 pfmemalloc = nc->pfmemalloc;
465 nc = this_cpu_ptr(&napi_alloc_cache.page);
466 data = page_frag_alloc(nc, len, gfp_mask);
467 pfmemalloc = nc->pfmemalloc;
474 skb = __build_skb(data, len);
475 if (unlikely(!skb)) {
485 skb_reserve(skb, NET_SKB_PAD);
491 EXPORT_SYMBOL(__netdev_alloc_skb);
494 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
495 * @napi: napi instance this buffer was allocated for
496 * @len: length to allocate
497 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
499 * Allocate a new sk_buff for use in NAPI receive. This buffer will
500 * attempt to allocate the head from a special reserved region used
501 * only for NAPI Rx allocation. By doing this we can save several
502 * CPU cycles by avoiding having to disable and re-enable IRQs.
504 * %NULL is returned if there is no free memory.
506 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
509 struct napi_alloc_cache *nc;
513 len += NET_SKB_PAD + NET_IP_ALIGN;
515 /* If requested length is either too small or too big,
516 * we use kmalloc() for skb->head allocation.
518 if (len <= SKB_WITH_OVERHEAD(1024) ||
519 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
520 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
521 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
527 nc = this_cpu_ptr(&napi_alloc_cache);
528 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
529 len = SKB_DATA_ALIGN(len);
531 if (sk_memalloc_socks())
532 gfp_mask |= __GFP_MEMALLOC;
534 data = page_frag_alloc(&nc->page, len, gfp_mask);
538 skb = __build_skb(data, len);
539 if (unlikely(!skb)) {
544 if (nc->page.pfmemalloc)
549 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
550 skb->dev = napi->dev;
555 EXPORT_SYMBOL(__napi_alloc_skb);
557 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
558 int size, unsigned int truesize)
560 skb_fill_page_desc(skb, i, page, off, size);
562 skb->data_len += size;
563 skb->truesize += truesize;
565 EXPORT_SYMBOL(skb_add_rx_frag);
567 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
568 unsigned int truesize)
570 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
572 skb_frag_size_add(frag, size);
574 skb->data_len += size;
575 skb->truesize += truesize;
577 EXPORT_SYMBOL(skb_coalesce_rx_frag);
579 static void skb_drop_list(struct sk_buff **listp)
581 kfree_skb_list(*listp);
585 static inline void skb_drop_fraglist(struct sk_buff *skb)
587 skb_drop_list(&skb_shinfo(skb)->frag_list);
590 static void skb_clone_fraglist(struct sk_buff *skb)
592 struct sk_buff *list;
594 skb_walk_frags(skb, list)
598 static void skb_free_head(struct sk_buff *skb)
600 unsigned char *head = skb->head;
608 static void skb_release_data(struct sk_buff *skb)
610 struct skb_shared_info *shinfo = skb_shinfo(skb);
614 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
618 for (i = 0; i < shinfo->nr_frags; i++)
619 __skb_frag_unref(&shinfo->frags[i]);
621 if (shinfo->frag_list)
622 kfree_skb_list(shinfo->frag_list);
624 skb_zcopy_clear(skb, true);
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=%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);
918 /* if reaching here SKB is ready to free */
919 trace_consume_skb(skb);
921 /* if SKB is a clone, don't handle this case */
922 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
927 _kfree_skb_defer(skb);
929 EXPORT_SYMBOL(napi_consume_skb);
931 /* Make sure a field is enclosed inside headers_start/headers_end section */
932 #define CHECK_SKB_FIELD(field) \
933 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
934 offsetof(struct sk_buff, headers_start)); \
935 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
936 offsetof(struct sk_buff, headers_end)); \
938 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
940 new->tstamp = old->tstamp;
941 /* We do not copy old->sk */
943 memcpy(new->cb, old->cb, sizeof(old->cb));
944 skb_dst_copy(new, old);
945 __skb_ext_copy(new, old);
946 __nf_copy(new, old, false);
948 /* Note : this field could be in headers_start/headers_end section
949 * It is not yet because we do not want to have a 16 bit hole
951 new->queue_mapping = old->queue_mapping;
953 memcpy(&new->headers_start, &old->headers_start,
954 offsetof(struct sk_buff, headers_end) -
955 offsetof(struct sk_buff, headers_start));
956 CHECK_SKB_FIELD(protocol);
957 CHECK_SKB_FIELD(csum);
958 CHECK_SKB_FIELD(hash);
959 CHECK_SKB_FIELD(priority);
960 CHECK_SKB_FIELD(skb_iif);
961 CHECK_SKB_FIELD(vlan_proto);
962 CHECK_SKB_FIELD(vlan_tci);
963 CHECK_SKB_FIELD(transport_header);
964 CHECK_SKB_FIELD(network_header);
965 CHECK_SKB_FIELD(mac_header);
966 CHECK_SKB_FIELD(inner_protocol);
967 CHECK_SKB_FIELD(inner_transport_header);
968 CHECK_SKB_FIELD(inner_network_header);
969 CHECK_SKB_FIELD(inner_mac_header);
970 CHECK_SKB_FIELD(mark);
971 #ifdef CONFIG_NETWORK_SECMARK
972 CHECK_SKB_FIELD(secmark);
974 #ifdef CONFIG_NET_RX_BUSY_POLL
975 CHECK_SKB_FIELD(napi_id);
978 CHECK_SKB_FIELD(sender_cpu);
980 #ifdef CONFIG_NET_SCHED
981 CHECK_SKB_FIELD(tc_index);
987 * You should not add any new code to this function. Add it to
988 * __copy_skb_header above instead.
990 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
992 #define C(x) n->x = skb->x
994 n->next = n->prev = NULL;
996 __copy_skb_header(n, skb);
1001 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1006 n->destructor = NULL;
1013 refcount_set(&n->users, 1);
1015 atomic_inc(&(skb_shinfo(skb)->dataref));
1023 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1024 * @first: first sk_buff of the msg
1026 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1030 n = alloc_skb(0, GFP_ATOMIC);
1034 n->len = first->len;
1035 n->data_len = first->len;
1036 n->truesize = first->truesize;
1038 skb_shinfo(n)->frag_list = first;
1040 __copy_skb_header(n, first);
1041 n->destructor = NULL;
1045 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1048 * skb_morph - morph one skb into another
1049 * @dst: the skb to receive the contents
1050 * @src: the skb to supply the contents
1052 * This is identical to skb_clone except that the target skb is
1053 * supplied by the user.
1055 * The target skb is returned upon exit.
1057 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1059 skb_release_all(dst);
1060 return __skb_clone(dst, src);
1062 EXPORT_SYMBOL_GPL(skb_morph);
1064 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1066 unsigned long max_pg, num_pg, new_pg, old_pg;
1067 struct user_struct *user;
1069 if (capable(CAP_IPC_LOCK) || !size)
1072 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1073 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1074 user = mmp->user ? : current_user();
1077 old_pg = atomic_long_read(&user->locked_vm);
1078 new_pg = old_pg + num_pg;
1079 if (new_pg > max_pg)
1081 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1085 mmp->user = get_uid(user);
1086 mmp->num_pg = num_pg;
1088 mmp->num_pg += num_pg;
1093 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1095 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1098 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1099 free_uid(mmp->user);
1102 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1104 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1106 struct ubuf_info *uarg;
1107 struct sk_buff *skb;
1109 WARN_ON_ONCE(!in_task());
1111 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1115 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1116 uarg = (void *)skb->cb;
1117 uarg->mmp.user = NULL;
1119 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1124 uarg->callback = sock_zerocopy_callback;
1125 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1127 uarg->bytelen = size;
1129 refcount_set(&uarg->refcnt, 1);
1134 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1136 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1138 return container_of((void *)uarg, struct sk_buff, cb);
1141 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1142 struct ubuf_info *uarg)
1145 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1148 /* realloc only when socket is locked (TCP, UDP cork),
1149 * so uarg->len and sk_zckey access is serialized
1151 if (!sock_owned_by_user(sk)) {
1156 bytelen = uarg->bytelen + size;
1157 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1158 /* TCP can create new skb to attach new uarg */
1159 if (sk->sk_type == SOCK_STREAM)
1164 next = (u32)atomic_read(&sk->sk_zckey);
1165 if ((u32)(uarg->id + uarg->len) == next) {
1166 if (mm_account_pinned_pages(&uarg->mmp, size))
1169 uarg->bytelen = bytelen;
1170 atomic_set(&sk->sk_zckey, ++next);
1172 /* no extra ref when appending to datagram (MSG_MORE) */
1173 if (sk->sk_type == SOCK_STREAM)
1174 sock_zerocopy_get(uarg);
1181 return sock_zerocopy_alloc(sk, size);
1183 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1185 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1187 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1191 old_lo = serr->ee.ee_info;
1192 old_hi = serr->ee.ee_data;
1193 sum_len = old_hi - old_lo + 1ULL + len;
1195 if (sum_len >= (1ULL << 32))
1198 if (lo != old_hi + 1)
1201 serr->ee.ee_data += len;
1205 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1207 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1208 struct sock_exterr_skb *serr;
1209 struct sock *sk = skb->sk;
1210 struct sk_buff_head *q;
1211 unsigned long flags;
1215 mm_unaccount_pinned_pages(&uarg->mmp);
1217 /* if !len, there was only 1 call, and it was aborted
1218 * so do not queue a completion notification
1220 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1225 hi = uarg->id + len - 1;
1227 serr = SKB_EXT_ERR(skb);
1228 memset(serr, 0, sizeof(*serr));
1229 serr->ee.ee_errno = 0;
1230 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1231 serr->ee.ee_data = hi;
1232 serr->ee.ee_info = lo;
1234 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1236 q = &sk->sk_error_queue;
1237 spin_lock_irqsave(&q->lock, flags);
1238 tail = skb_peek_tail(q);
1239 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1240 !skb_zerocopy_notify_extend(tail, lo, len)) {
1241 __skb_queue_tail(q, skb);
1244 spin_unlock_irqrestore(&q->lock, flags);
1246 sk->sk_error_report(sk);
1252 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1254 void sock_zerocopy_put(struct ubuf_info *uarg)
1256 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1258 uarg->callback(uarg, uarg->zerocopy);
1260 consume_skb(skb_from_uarg(uarg));
1263 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1265 void sock_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 sock_zerocopy_put(uarg);
1277 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1279 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1281 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1283 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1285 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1286 struct msghdr *msg, int len,
1287 struct ubuf_info *uarg)
1289 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1290 struct iov_iter orig_iter = msg->msg_iter;
1291 int err, orig_len = skb->len;
1293 /* An skb can only point to one uarg. This edge case happens when
1294 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1296 if (orig_uarg && uarg != orig_uarg)
1299 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1300 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1301 struct sock *save_sk = skb->sk;
1303 /* Streams do not free skb on error. Reset to prev state. */
1304 msg->msg_iter = orig_iter;
1306 ___pskb_trim(skb, orig_len);
1311 skb_zcopy_set(skb, uarg, NULL);
1312 return skb->len - orig_len;
1314 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1316 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1319 if (skb_zcopy(orig)) {
1320 if (skb_zcopy(nskb)) {
1321 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1326 if (skb_uarg(nskb) == skb_uarg(orig))
1328 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1331 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1337 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1338 * @skb: the skb to modify
1339 * @gfp_mask: allocation priority
1341 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1342 * It will copy all frags into kernel and drop the reference
1343 * to userspace pages.
1345 * If this function is called from an interrupt gfp_mask() must be
1348 * Returns 0 on success or a negative error code on failure
1349 * to allocate kernel memory to copy to.
1351 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1353 int num_frags = skb_shinfo(skb)->nr_frags;
1354 struct page *page, *head = NULL;
1358 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1364 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1365 for (i = 0; i < new_frags; i++) {
1366 page = alloc_page(gfp_mask);
1369 struct page *next = (struct page *)page_private(head);
1375 set_page_private(page, (unsigned long)head);
1381 for (i = 0; i < num_frags; i++) {
1382 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1383 u32 p_off, p_len, copied;
1387 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1388 p, p_off, p_len, copied) {
1390 vaddr = kmap_atomic(p);
1392 while (done < p_len) {
1393 if (d_off == PAGE_SIZE) {
1395 page = (struct page *)page_private(page);
1397 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1398 memcpy(page_address(page) + d_off,
1399 vaddr + p_off + done, copy);
1403 kunmap_atomic(vaddr);
1407 /* skb frags release userspace buffers */
1408 for (i = 0; i < num_frags; i++)
1409 skb_frag_unref(skb, i);
1411 /* skb frags point to kernel buffers */
1412 for (i = 0; i < new_frags - 1; i++) {
1413 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1414 head = (struct page *)page_private(head);
1416 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1417 skb_shinfo(skb)->nr_frags = new_frags;
1420 skb_zcopy_clear(skb, false);
1423 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1426 * skb_clone - duplicate an sk_buff
1427 * @skb: buffer to clone
1428 * @gfp_mask: allocation priority
1430 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1431 * copies share the same packet data but not structure. The new
1432 * buffer has a reference count of 1. If the allocation fails the
1433 * function returns %NULL otherwise the new buffer is returned.
1435 * If this function is called from an interrupt gfp_mask() must be
1439 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1441 struct sk_buff_fclones *fclones = container_of(skb,
1442 struct sk_buff_fclones,
1446 if (skb_orphan_frags(skb, gfp_mask))
1449 if (skb->fclone == SKB_FCLONE_ORIG &&
1450 refcount_read(&fclones->fclone_ref) == 1) {
1452 refcount_set(&fclones->fclone_ref, 2);
1454 if (skb_pfmemalloc(skb))
1455 gfp_mask |= __GFP_MEMALLOC;
1457 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1461 n->fclone = SKB_FCLONE_UNAVAILABLE;
1464 return __skb_clone(n, skb);
1466 EXPORT_SYMBOL(skb_clone);
1468 void skb_headers_offset_update(struct sk_buff *skb, int off)
1470 /* Only adjust this if it actually is csum_start rather than csum */
1471 if (skb->ip_summed == CHECKSUM_PARTIAL)
1472 skb->csum_start += off;
1473 /* {transport,network,mac}_header and tail are relative to skb->head */
1474 skb->transport_header += off;
1475 skb->network_header += off;
1476 if (skb_mac_header_was_set(skb))
1477 skb->mac_header += off;
1478 skb->inner_transport_header += off;
1479 skb->inner_network_header += off;
1480 skb->inner_mac_header += off;
1482 EXPORT_SYMBOL(skb_headers_offset_update);
1484 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1486 __copy_skb_header(new, old);
1488 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1489 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1490 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1492 EXPORT_SYMBOL(skb_copy_header);
1494 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1496 if (skb_pfmemalloc(skb))
1497 return SKB_ALLOC_RX;
1502 * skb_copy - create private copy of an sk_buff
1503 * @skb: buffer to copy
1504 * @gfp_mask: allocation priority
1506 * Make a copy of both an &sk_buff and its data. This is used when the
1507 * caller wishes to modify the data and needs a private copy of the
1508 * data to alter. Returns %NULL on failure or the pointer to the buffer
1509 * on success. The returned buffer has a reference count of 1.
1511 * As by-product this function converts non-linear &sk_buff to linear
1512 * one, so that &sk_buff becomes completely private and caller is allowed
1513 * to modify all the data of returned buffer. This means that this
1514 * function is not recommended for use in circumstances when only
1515 * header is going to be modified. Use pskb_copy() instead.
1518 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1524 if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
1527 headerlen = skb_headroom(skb);
1528 size = skb_end_offset(skb) + skb->data_len;
1529 n = __alloc_skb(size, gfp_mask,
1530 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1534 /* Set the data pointer */
1535 skb_reserve(n, headerlen);
1536 /* Set the tail pointer and length */
1537 skb_put(n, skb->len);
1539 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1541 skb_copy_header(n, skb);
1544 EXPORT_SYMBOL(skb_copy);
1547 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1548 * @skb: buffer to copy
1549 * @headroom: headroom of new skb
1550 * @gfp_mask: allocation priority
1551 * @fclone: if true allocate the copy of the skb from the fclone
1552 * cache instead of the head cache; it is recommended to set this
1553 * to true for the cases where the copy will likely be cloned
1555 * Make a copy of both an &sk_buff and part of its data, located
1556 * in header. Fragmented data remain shared. This is used when
1557 * the caller wishes to modify only header of &sk_buff and needs
1558 * private copy of the header to alter. Returns %NULL on failure
1559 * or the pointer to the buffer on success.
1560 * The returned buffer has a reference count of 1.
1563 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1564 gfp_t gfp_mask, bool fclone)
1566 unsigned int size = skb_headlen(skb) + headroom;
1567 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1568 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1573 /* Set the data pointer */
1574 skb_reserve(n, headroom);
1575 /* Set the tail pointer and length */
1576 skb_put(n, skb_headlen(skb));
1577 /* Copy the bytes */
1578 skb_copy_from_linear_data(skb, n->data, n->len);
1580 n->truesize += skb->data_len;
1581 n->data_len = skb->data_len;
1584 if (skb_shinfo(skb)->nr_frags) {
1587 if (skb_orphan_frags(skb, gfp_mask) ||
1588 skb_zerocopy_clone(n, skb, gfp_mask)) {
1593 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1594 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1595 skb_frag_ref(skb, i);
1597 skb_shinfo(n)->nr_frags = i;
1600 if (skb_has_frag_list(skb)) {
1601 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1602 skb_clone_fraglist(n);
1605 skb_copy_header(n, skb);
1609 EXPORT_SYMBOL(__pskb_copy_fclone);
1612 * pskb_expand_head - reallocate header of &sk_buff
1613 * @skb: buffer to reallocate
1614 * @nhead: room to add at head
1615 * @ntail: room to add at tail
1616 * @gfp_mask: allocation priority
1618 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1619 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1620 * reference count of 1. Returns zero in the case of success or error,
1621 * if expansion failed. In the last case, &sk_buff is not changed.
1623 * All the pointers pointing into skb header may change and must be
1624 * reloaded after call to this function.
1627 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1630 int i, osize = skb_end_offset(skb);
1631 int size = osize + nhead + ntail;
1637 BUG_ON(skb_shared(skb));
1639 size = SKB_DATA_ALIGN(size);
1641 if (skb_pfmemalloc(skb))
1642 gfp_mask |= __GFP_MEMALLOC;
1643 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1644 gfp_mask, NUMA_NO_NODE, NULL);
1647 size = SKB_WITH_OVERHEAD(ksize(data));
1649 /* Copy only real data... and, alas, header. This should be
1650 * optimized for the cases when header is void.
1652 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1654 memcpy((struct skb_shared_info *)(data + size),
1656 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1659 * if shinfo is shared we must drop the old head gracefully, but if it
1660 * is not we can just drop the old head and let the existing refcount
1661 * be since all we did is relocate the values
1663 if (skb_cloned(skb)) {
1664 if (skb_orphan_frags(skb, gfp_mask))
1667 refcount_inc(&skb_uarg(skb)->refcnt);
1668 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1669 skb_frag_ref(skb, i);
1671 if (skb_has_frag_list(skb))
1672 skb_clone_fraglist(skb);
1674 skb_release_data(skb);
1678 off = (data + nhead) - skb->head;
1683 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1687 skb->end = skb->head + size;
1690 skb_headers_offset_update(skb, nhead);
1694 atomic_set(&skb_shinfo(skb)->dataref, 1);
1696 skb_metadata_clear(skb);
1698 /* It is not generally safe to change skb->truesize.
1699 * For the moment, we really care of rx path, or
1700 * when skb is orphaned (not attached to a socket).
1702 if (!skb->sk || skb->destructor == sock_edemux)
1703 skb->truesize += size - osize;
1712 EXPORT_SYMBOL(pskb_expand_head);
1714 /* Make private copy of skb with writable head and some headroom */
1716 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1718 struct sk_buff *skb2;
1719 int delta = headroom - skb_headroom(skb);
1722 skb2 = pskb_copy(skb, GFP_ATOMIC);
1724 skb2 = skb_clone(skb, GFP_ATOMIC);
1725 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1733 EXPORT_SYMBOL(skb_realloc_headroom);
1736 * skb_copy_expand - copy and expand sk_buff
1737 * @skb: buffer to copy
1738 * @newheadroom: new free bytes at head
1739 * @newtailroom: new free bytes at tail
1740 * @gfp_mask: allocation priority
1742 * Make a copy of both an &sk_buff and its data and while doing so
1743 * allocate additional space.
1745 * This is used when the caller wishes to modify the data and needs a
1746 * private copy of the data to alter as well as more space for new fields.
1747 * Returns %NULL on failure or the pointer to the buffer
1748 * on success. The returned buffer has a reference count of 1.
1750 * You must pass %GFP_ATOMIC as the allocation priority if this function
1751 * is called from an interrupt.
1753 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1754 int newheadroom, int newtailroom,
1758 * Allocate the copy buffer
1760 int head_copy_len, head_copy_off;
1764 if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
1767 oldheadroom = skb_headroom(skb);
1768 n = __alloc_skb(newheadroom + skb->len + newtailroom,
1769 gfp_mask, skb_alloc_rx_flag(skb),
1774 skb_reserve(n, newheadroom);
1776 /* Set the tail pointer and length */
1777 skb_put(n, skb->len);
1779 head_copy_len = oldheadroom;
1781 if (newheadroom <= head_copy_len)
1782 head_copy_len = newheadroom;
1784 head_copy_off = newheadroom - head_copy_len;
1786 /* Copy the linear header and data. */
1787 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1788 skb->len + head_copy_len));
1790 skb_copy_header(n, skb);
1792 skb_headers_offset_update(n, newheadroom - oldheadroom);
1796 EXPORT_SYMBOL(skb_copy_expand);
1799 * __skb_pad - zero pad the tail of an skb
1800 * @skb: buffer to pad
1801 * @pad: space to pad
1802 * @free_on_error: free buffer on error
1804 * Ensure that a buffer is followed by a padding area that is zero
1805 * filled. Used by network drivers which may DMA or transfer data
1806 * beyond the buffer end onto the wire.
1808 * May return error in out of memory cases. The skb is freed on error
1809 * if @free_on_error is true.
1812 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1817 /* If the skbuff is non linear tailroom is always zero.. */
1818 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1819 memset(skb->data+skb->len, 0, pad);
1823 ntail = skb->data_len + pad - (skb->end - skb->tail);
1824 if (likely(skb_cloned(skb) || ntail > 0)) {
1825 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1830 /* FIXME: The use of this function with non-linear skb's really needs
1833 err = skb_linearize(skb);
1837 memset(skb->data + skb->len, 0, pad);
1845 EXPORT_SYMBOL(__skb_pad);
1848 * pskb_put - add data to the tail of a potentially fragmented buffer
1849 * @skb: start of the buffer to use
1850 * @tail: tail fragment of the buffer to use
1851 * @len: amount of data to add
1853 * This function extends the used data area of the potentially
1854 * fragmented buffer. @tail must be the last fragment of @skb -- or
1855 * @skb itself. If this would exceed the total buffer size the kernel
1856 * will panic. A pointer to the first byte of the extra data is
1860 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1863 skb->data_len += len;
1866 return skb_put(tail, len);
1868 EXPORT_SYMBOL_GPL(pskb_put);
1871 * skb_put - add data to a buffer
1872 * @skb: buffer to use
1873 * @len: amount of data to add
1875 * This function extends the used data area of the buffer. If this would
1876 * exceed the total buffer size the kernel will panic. A pointer to the
1877 * first byte of the extra data is returned.
1879 void *skb_put(struct sk_buff *skb, unsigned int len)
1881 void *tmp = skb_tail_pointer(skb);
1882 SKB_LINEAR_ASSERT(skb);
1885 if (unlikely(skb->tail > skb->end))
1886 skb_over_panic(skb, len, __builtin_return_address(0));
1889 EXPORT_SYMBOL(skb_put);
1892 * skb_push - add data to the start of a buffer
1893 * @skb: buffer to use
1894 * @len: amount of data to add
1896 * This function extends the used data area of the buffer at the buffer
1897 * start. If this would exceed the total buffer headroom the kernel will
1898 * panic. A pointer to the first byte of the extra data is returned.
1900 void *skb_push(struct sk_buff *skb, unsigned int len)
1904 if (unlikely(skb->data < skb->head))
1905 skb_under_panic(skb, len, __builtin_return_address(0));
1908 EXPORT_SYMBOL(skb_push);
1911 * skb_pull - remove data from the start of a buffer
1912 * @skb: buffer to use
1913 * @len: amount of data to remove
1915 * This function removes data from the start of a buffer, returning
1916 * the memory to the headroom. A pointer to the next data in the buffer
1917 * is returned. Once the data has been pulled future pushes will overwrite
1920 void *skb_pull(struct sk_buff *skb, unsigned int len)
1922 return skb_pull_inline(skb, len);
1924 EXPORT_SYMBOL(skb_pull);
1927 * skb_trim - remove end from a buffer
1928 * @skb: buffer to alter
1931 * Cut the length of a buffer down by removing data from the tail. If
1932 * the buffer is already under the length specified it is not modified.
1933 * The skb must be linear.
1935 void skb_trim(struct sk_buff *skb, unsigned int len)
1938 __skb_trim(skb, len);
1940 EXPORT_SYMBOL(skb_trim);
1942 /* Trims skb to length len. It can change skb pointers.
1945 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1947 struct sk_buff **fragp;
1948 struct sk_buff *frag;
1949 int offset = skb_headlen(skb);
1950 int nfrags = skb_shinfo(skb)->nr_frags;
1954 if (skb_cloned(skb) &&
1955 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1962 for (; i < nfrags; i++) {
1963 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1970 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1973 skb_shinfo(skb)->nr_frags = i;
1975 for (; i < nfrags; i++)
1976 skb_frag_unref(skb, i);
1978 if (skb_has_frag_list(skb))
1979 skb_drop_fraglist(skb);
1983 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1984 fragp = &frag->next) {
1985 int end = offset + frag->len;
1987 if (skb_shared(frag)) {
1988 struct sk_buff *nfrag;
1990 nfrag = skb_clone(frag, GFP_ATOMIC);
1991 if (unlikely(!nfrag))
1994 nfrag->next = frag->next;
2006 unlikely((err = pskb_trim(frag, len - offset))))
2010 skb_drop_list(&frag->next);
2015 if (len > skb_headlen(skb)) {
2016 skb->data_len -= skb->len - len;
2021 skb_set_tail_pointer(skb, len);
2024 if (!skb->sk || skb->destructor == sock_edemux)
2028 EXPORT_SYMBOL(___pskb_trim);
2030 /* Note : use pskb_trim_rcsum() instead of calling this directly
2032 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2034 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2035 int delta = skb->len - len;
2037 skb->csum = csum_block_sub(skb->csum,
2038 skb_checksum(skb, len, delta, 0),
2040 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2041 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2042 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2044 if (offset + sizeof(__sum16) > hdlen)
2047 return __pskb_trim(skb, len);
2049 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2052 * __pskb_pull_tail - advance tail of skb header
2053 * @skb: buffer to reallocate
2054 * @delta: number of bytes to advance tail
2056 * The function makes a sense only on a fragmented &sk_buff,
2057 * it expands header moving its tail forward and copying necessary
2058 * data from fragmented part.
2060 * &sk_buff MUST have reference count of 1.
2062 * Returns %NULL (and &sk_buff does not change) if pull failed
2063 * or value of new tail of skb in the case of success.
2065 * All the pointers pointing into skb header may change and must be
2066 * reloaded after call to this function.
2069 /* Moves tail of skb head forward, copying data from fragmented part,
2070 * when it is necessary.
2071 * 1. It may fail due to malloc failure.
2072 * 2. It may change skb pointers.
2074 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2076 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2078 /* If skb has not enough free space at tail, get new one
2079 * plus 128 bytes for future expansions. If we have enough
2080 * room at tail, reallocate without expansion only if skb is cloned.
2082 int i, k, eat = (skb->tail + delta) - skb->end;
2084 if (eat > 0 || skb_cloned(skb)) {
2085 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2090 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2091 skb_tail_pointer(skb), delta));
2093 /* Optimization: no fragments, no reasons to preestimate
2094 * size of pulled pages. Superb.
2096 if (!skb_has_frag_list(skb))
2099 /* Estimate size of pulled pages. */
2101 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2102 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2109 /* If we need update frag list, we are in troubles.
2110 * Certainly, it is possible to add an offset to skb data,
2111 * but taking into account that pulling is expected to
2112 * be very rare operation, it is worth to fight against
2113 * further bloating skb head and crucify ourselves here instead.
2114 * Pure masohism, indeed. 8)8)
2117 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2118 struct sk_buff *clone = NULL;
2119 struct sk_buff *insp = NULL;
2122 if (list->len <= eat) {
2123 /* Eaten as whole. */
2128 /* Eaten partially. */
2129 if (skb_is_gso(skb) && !list->head_frag &&
2131 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2133 if (skb_shared(list)) {
2134 /* Sucks! We need to fork list. :-( */
2135 clone = skb_clone(list, GFP_ATOMIC);
2141 /* This may be pulled without
2145 if (!pskb_pull(list, eat)) {
2153 /* Free pulled out fragments. */
2154 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2155 skb_shinfo(skb)->frag_list = list->next;
2158 /* And insert new clone at head. */
2161 skb_shinfo(skb)->frag_list = clone;
2164 /* Success! Now we may commit changes to skb data. */
2169 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2170 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2173 skb_frag_unref(skb, i);
2176 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2178 *frag = skb_shinfo(skb)->frags[i];
2180 skb_frag_off_add(frag, eat);
2181 skb_frag_size_sub(frag, eat);
2189 skb_shinfo(skb)->nr_frags = k;
2193 skb->data_len -= delta;
2196 skb_zcopy_clear(skb, false);
2198 return skb_tail_pointer(skb);
2200 EXPORT_SYMBOL(__pskb_pull_tail);
2203 * skb_copy_bits - copy bits from skb to kernel buffer
2205 * @offset: offset in source
2206 * @to: destination buffer
2207 * @len: number of bytes to copy
2209 * Copy the specified number of bytes from the source skb to the
2210 * destination buffer.
2213 * If its prototype is ever changed,
2214 * check arch/{*}/net/{*}.S files,
2215 * since it is called from BPF assembly code.
2217 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2219 int start = skb_headlen(skb);
2220 struct sk_buff *frag_iter;
2223 if (offset > (int)skb->len - len)
2227 if ((copy = start - offset) > 0) {
2230 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2231 if ((len -= copy) == 0)
2237 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2239 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2241 WARN_ON(start > offset + len);
2243 end = start + skb_frag_size(f);
2244 if ((copy = end - offset) > 0) {
2245 u32 p_off, p_len, copied;
2252 skb_frag_foreach_page(f,
2253 skb_frag_off(f) + offset - start,
2254 copy, p, p_off, p_len, copied) {
2255 vaddr = kmap_atomic(p);
2256 memcpy(to + copied, vaddr + p_off, p_len);
2257 kunmap_atomic(vaddr);
2260 if ((len -= copy) == 0)
2268 skb_walk_frags(skb, frag_iter) {
2271 WARN_ON(start > offset + len);
2273 end = start + frag_iter->len;
2274 if ((copy = end - offset) > 0) {
2277 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2279 if ((len -= copy) == 0)
2293 EXPORT_SYMBOL(skb_copy_bits);
2296 * Callback from splice_to_pipe(), if we need to release some pages
2297 * at the end of the spd in case we error'ed out in filling the pipe.
2299 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2301 put_page(spd->pages[i]);
2304 static struct page *linear_to_page(struct page *page, unsigned int *len,
2305 unsigned int *offset,
2308 struct page_frag *pfrag = sk_page_frag(sk);
2310 if (!sk_page_frag_refill(sk, pfrag))
2313 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2315 memcpy(page_address(pfrag->page) + pfrag->offset,
2316 page_address(page) + *offset, *len);
2317 *offset = pfrag->offset;
2318 pfrag->offset += *len;
2323 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2325 unsigned int offset)
2327 return spd->nr_pages &&
2328 spd->pages[spd->nr_pages - 1] == page &&
2329 (spd->partial[spd->nr_pages - 1].offset +
2330 spd->partial[spd->nr_pages - 1].len == offset);
2334 * Fill page/offset/length into spd, if it can hold more pages.
2336 static bool spd_fill_page(struct splice_pipe_desc *spd,
2337 struct pipe_inode_info *pipe, struct page *page,
2338 unsigned int *len, unsigned int offset,
2342 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2346 page = linear_to_page(page, len, &offset, sk);
2350 if (spd_can_coalesce(spd, page, offset)) {
2351 spd->partial[spd->nr_pages - 1].len += *len;
2355 spd->pages[spd->nr_pages] = page;
2356 spd->partial[spd->nr_pages].len = *len;
2357 spd->partial[spd->nr_pages].offset = offset;
2363 static bool __splice_segment(struct page *page, unsigned int poff,
2364 unsigned int plen, unsigned int *off,
2366 struct splice_pipe_desc *spd, bool linear,
2368 struct pipe_inode_info *pipe)
2373 /* skip this segment if already processed */
2379 /* ignore any bits we already processed */
2385 unsigned int flen = min(*len, plen);
2387 if (spd_fill_page(spd, pipe, page, &flen, poff,
2393 } while (*len && plen);
2399 * Map linear and fragment data from the skb to spd. It reports true if the
2400 * pipe is full or if we already spliced the requested length.
2402 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2403 unsigned int *offset, unsigned int *len,
2404 struct splice_pipe_desc *spd, struct sock *sk)
2407 struct sk_buff *iter;
2409 /* map the linear part :
2410 * If skb->head_frag is set, this 'linear' part is backed by a
2411 * fragment, and if the head is not shared with any clones then
2412 * we can avoid a copy since we own the head portion of this page.
2414 if (__splice_segment(virt_to_page(skb->data),
2415 (unsigned long) skb->data & (PAGE_SIZE - 1),
2418 skb_head_is_locked(skb),
2423 * then map the fragments
2425 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2426 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2428 if (__splice_segment(skb_frag_page(f),
2429 skb_frag_off(f), skb_frag_size(f),
2430 offset, len, spd, false, sk, pipe))
2434 skb_walk_frags(skb, iter) {
2435 if (*offset >= iter->len) {
2436 *offset -= iter->len;
2439 /* __skb_splice_bits() only fails if the output has no room
2440 * left, so no point in going over the frag_list for the error
2443 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2451 * Map data from the skb to a pipe. Should handle both the linear part,
2452 * the fragments, and the frag list.
2454 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2455 struct pipe_inode_info *pipe, unsigned int tlen,
2458 struct partial_page partial[MAX_SKB_FRAGS];
2459 struct page *pages[MAX_SKB_FRAGS];
2460 struct splice_pipe_desc spd = {
2463 .nr_pages_max = MAX_SKB_FRAGS,
2464 .ops = &nosteal_pipe_buf_ops,
2465 .spd_release = sock_spd_release,
2469 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2472 ret = splice_to_pipe(pipe, &spd);
2476 EXPORT_SYMBOL_GPL(skb_splice_bits);
2478 /* Send skb data on a socket. Socket must be locked. */
2479 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2482 unsigned int orig_len = len;
2483 struct sk_buff *head = skb;
2484 unsigned short fragidx;
2489 /* Deal with head data */
2490 while (offset < skb_headlen(skb) && len) {
2494 slen = min_t(int, len, skb_headlen(skb) - offset);
2495 kv.iov_base = skb->data + offset;
2497 memset(&msg, 0, sizeof(msg));
2498 msg.msg_flags = MSG_DONTWAIT;
2500 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2508 /* All the data was skb head? */
2512 /* Make offset relative to start of frags */
2513 offset -= skb_headlen(skb);
2515 /* Find where we are in frag list */
2516 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2517 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2519 if (offset < skb_frag_size(frag))
2522 offset -= skb_frag_size(frag);
2525 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2526 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2528 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2531 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2532 skb_frag_off(frag) + offset,
2533 slen, MSG_DONTWAIT);
2546 /* Process any frag lists */
2549 if (skb_has_frag_list(skb)) {
2550 skb = skb_shinfo(skb)->frag_list;
2553 } else if (skb->next) {
2560 return orig_len - len;
2563 return orig_len == len ? ret : orig_len - len;
2565 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2568 * skb_store_bits - store bits from kernel buffer to skb
2569 * @skb: destination buffer
2570 * @offset: offset in destination
2571 * @from: source buffer
2572 * @len: number of bytes to copy
2574 * Copy the specified number of bytes from the source buffer to the
2575 * destination skb. This function handles all the messy bits of
2576 * traversing fragment lists and such.
2579 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2581 int start = skb_headlen(skb);
2582 struct sk_buff *frag_iter;
2585 if (offset > (int)skb->len - len)
2588 if ((copy = start - offset) > 0) {
2591 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2592 if ((len -= copy) == 0)
2598 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2599 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2602 WARN_ON(start > offset + len);
2604 end = start + skb_frag_size(frag);
2605 if ((copy = end - offset) > 0) {
2606 u32 p_off, p_len, copied;
2613 skb_frag_foreach_page(frag,
2614 skb_frag_off(frag) + offset - start,
2615 copy, p, p_off, p_len, copied) {
2616 vaddr = kmap_atomic(p);
2617 memcpy(vaddr + p_off, from + copied, p_len);
2618 kunmap_atomic(vaddr);
2621 if ((len -= copy) == 0)
2629 skb_walk_frags(skb, frag_iter) {
2632 WARN_ON(start > offset + len);
2634 end = start + frag_iter->len;
2635 if ((copy = end - offset) > 0) {
2638 if (skb_store_bits(frag_iter, offset - start,
2641 if ((len -= copy) == 0)
2654 EXPORT_SYMBOL(skb_store_bits);
2656 /* Checksum skb data. */
2657 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2658 __wsum csum, const struct skb_checksum_ops *ops)
2660 int start = skb_headlen(skb);
2661 int i, copy = start - offset;
2662 struct sk_buff *frag_iter;
2665 /* Checksum header. */
2669 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2670 skb->data + offset, copy, csum);
2671 if ((len -= copy) == 0)
2677 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2679 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2681 WARN_ON(start > offset + len);
2683 end = start + skb_frag_size(frag);
2684 if ((copy = end - offset) > 0) {
2685 u32 p_off, p_len, copied;
2693 skb_frag_foreach_page(frag,
2694 skb_frag_off(frag) + offset - start,
2695 copy, p, p_off, p_len, copied) {
2696 vaddr = kmap_atomic(p);
2697 csum2 = INDIRECT_CALL_1(ops->update,
2699 vaddr + p_off, p_len, 0);
2700 kunmap_atomic(vaddr);
2701 csum = INDIRECT_CALL_1(ops->combine,
2702 csum_block_add_ext, csum,
2714 skb_walk_frags(skb, frag_iter) {
2717 WARN_ON(start > offset + len);
2719 end = start + frag_iter->len;
2720 if ((copy = end - offset) > 0) {
2724 csum2 = __skb_checksum(frag_iter, offset - start,
2726 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2727 csum, csum2, pos, copy);
2728 if ((len -= copy) == 0)
2739 EXPORT_SYMBOL(__skb_checksum);
2741 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2742 int len, __wsum csum)
2744 const struct skb_checksum_ops ops = {
2745 .update = csum_partial_ext,
2746 .combine = csum_block_add_ext,
2749 return __skb_checksum(skb, offset, len, csum, &ops);
2751 EXPORT_SYMBOL(skb_checksum);
2753 /* Both of above in one bottle. */
2755 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2758 int start = skb_headlen(skb);
2759 int i, copy = start - offset;
2760 struct sk_buff *frag_iter;
2768 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2770 if ((len -= copy) == 0)
2777 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2780 WARN_ON(start > offset + len);
2782 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2783 if ((copy = end - offset) > 0) {
2784 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2785 u32 p_off, p_len, copied;
2793 skb_frag_foreach_page(frag,
2794 skb_frag_off(frag) + offset - start,
2795 copy, p, p_off, p_len, copied) {
2796 vaddr = kmap_atomic(p);
2797 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2800 kunmap_atomic(vaddr);
2801 csum = csum_block_add(csum, csum2, pos);
2813 skb_walk_frags(skb, frag_iter) {
2817 WARN_ON(start > offset + len);
2819 end = start + frag_iter->len;
2820 if ((copy = end - offset) > 0) {
2823 csum2 = skb_copy_and_csum_bits(frag_iter,
2826 csum = csum_block_add(csum, csum2, pos);
2827 if ((len -= copy) == 0)
2838 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2840 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2844 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2845 /* See comments in __skb_checksum_complete(). */
2847 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2848 !skb->csum_complete_sw)
2849 netdev_rx_csum_fault(skb->dev, skb);
2851 if (!skb_shared(skb))
2852 skb->csum_valid = !sum;
2855 EXPORT_SYMBOL(__skb_checksum_complete_head);
2857 /* This function assumes skb->csum already holds pseudo header's checksum,
2858 * which has been changed from the hardware checksum, for example, by
2859 * __skb_checksum_validate_complete(). And, the original skb->csum must
2860 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2862 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2863 * zero. The new checksum is stored back into skb->csum unless the skb is
2866 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2871 csum = skb_checksum(skb, 0, skb->len, 0);
2873 sum = csum_fold(csum_add(skb->csum, csum));
2874 /* This check is inverted, because we already knew the hardware
2875 * checksum is invalid before calling this function. So, if the
2876 * re-computed checksum is valid instead, then we have a mismatch
2877 * between the original skb->csum and skb_checksum(). This means either
2878 * the original hardware checksum is incorrect or we screw up skb->csum
2879 * when moving skb->data around.
2882 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2883 !skb->csum_complete_sw)
2884 netdev_rx_csum_fault(skb->dev, skb);
2887 if (!skb_shared(skb)) {
2888 /* Save full packet checksum */
2890 skb->ip_summed = CHECKSUM_COMPLETE;
2891 skb->csum_complete_sw = 1;
2892 skb->csum_valid = !sum;
2897 EXPORT_SYMBOL(__skb_checksum_complete);
2899 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2901 net_warn_ratelimited(
2902 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2907 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2908 int offset, int len)
2910 net_warn_ratelimited(
2911 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2916 static const struct skb_checksum_ops default_crc32c_ops = {
2917 .update = warn_crc32c_csum_update,
2918 .combine = warn_crc32c_csum_combine,
2921 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2922 &default_crc32c_ops;
2923 EXPORT_SYMBOL(crc32c_csum_stub);
2926 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2927 * @from: source buffer
2929 * Calculates the amount of linear headroom needed in the 'to' skb passed
2930 * into skb_zerocopy().
2933 skb_zerocopy_headlen(const struct sk_buff *from)
2935 unsigned int hlen = 0;
2937 if (!from->head_frag ||
2938 skb_headlen(from) < L1_CACHE_BYTES ||
2939 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
2940 hlen = skb_headlen(from);
2945 if (skb_has_frag_list(from))
2950 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2953 * skb_zerocopy - Zero copy skb to skb
2954 * @to: destination buffer
2955 * @from: source buffer
2956 * @len: number of bytes to copy from source buffer
2957 * @hlen: size of linear headroom in destination buffer
2959 * Copies up to `len` bytes from `from` to `to` by creating references
2960 * to the frags in the source buffer.
2962 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2963 * headroom in the `to` buffer.
2966 * 0: everything is OK
2967 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2968 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2971 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2974 int plen = 0; /* length of skb->head fragment */
2977 unsigned int offset;
2979 BUG_ON(!from->head_frag && !hlen);
2981 /* dont bother with small payloads */
2982 if (len <= skb_tailroom(to))
2983 return skb_copy_bits(from, 0, skb_put(to, len), len);
2986 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2991 plen = min_t(int, skb_headlen(from), len);
2993 page = virt_to_head_page(from->head);
2994 offset = from->data - (unsigned char *)page_address(page);
2995 __skb_fill_page_desc(to, 0, page, offset, plen);
3002 to->truesize += len + plen;
3003 to->len += len + plen;
3004 to->data_len += len + plen;
3006 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3010 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3012 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3017 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3018 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3020 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3022 skb_frag_ref(to, j);
3025 skb_shinfo(to)->nr_frags = j;
3029 EXPORT_SYMBOL_GPL(skb_zerocopy);
3031 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3036 if (skb->ip_summed == CHECKSUM_PARTIAL)
3037 csstart = skb_checksum_start_offset(skb);
3039 csstart = skb_headlen(skb);
3041 BUG_ON(csstart > skb_headlen(skb));
3043 skb_copy_from_linear_data(skb, to, csstart);
3046 if (csstart != skb->len)
3047 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3048 skb->len - csstart);
3050 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3051 long csstuff = csstart + skb->csum_offset;
3053 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3056 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3059 * skb_dequeue - remove from the head of the queue
3060 * @list: list to dequeue from
3062 * Remove the head of the list. The list lock is taken so the function
3063 * may be used safely with other locking list functions. The head item is
3064 * returned or %NULL if the list is empty.
3067 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3069 unsigned long flags;
3070 struct sk_buff *result;
3072 spin_lock_irqsave(&list->lock, flags);
3073 result = __skb_dequeue(list);
3074 spin_unlock_irqrestore(&list->lock, flags);
3077 EXPORT_SYMBOL(skb_dequeue);
3080 * skb_dequeue_tail - remove from the tail of the queue
3081 * @list: list to dequeue from
3083 * Remove the tail of the list. The list lock is taken so the function
3084 * may be used safely with other locking list functions. The tail item is
3085 * returned or %NULL if the list is empty.
3087 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3089 unsigned long flags;
3090 struct sk_buff *result;
3092 spin_lock_irqsave(&list->lock, flags);
3093 result = __skb_dequeue_tail(list);
3094 spin_unlock_irqrestore(&list->lock, flags);
3097 EXPORT_SYMBOL(skb_dequeue_tail);
3100 * skb_queue_purge - empty a list
3101 * @list: list to empty
3103 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3104 * the list and one reference dropped. This function takes the list
3105 * lock and is atomic with respect to other list locking functions.
3107 void skb_queue_purge(struct sk_buff_head *list)
3109 struct sk_buff *skb;
3110 while ((skb = skb_dequeue(list)) != NULL)
3113 EXPORT_SYMBOL(skb_queue_purge);
3116 * skb_rbtree_purge - empty a skb rbtree
3117 * @root: root of the rbtree to empty
3118 * Return value: the sum of truesizes of all purged skbs.
3120 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3121 * the list and one reference dropped. This function does not take
3122 * any lock. Synchronization should be handled by the caller (e.g., TCP
3123 * out-of-order queue is protected by the socket lock).
3125 unsigned int skb_rbtree_purge(struct rb_root *root)
3127 struct rb_node *p = rb_first(root);
3128 unsigned int sum = 0;
3131 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3134 rb_erase(&skb->rbnode, root);
3135 sum += skb->truesize;
3142 * skb_queue_head - queue a buffer at the list head
3143 * @list: list to use
3144 * @newsk: buffer to queue
3146 * Queue a buffer at the start of the list. This function takes the
3147 * list lock and can be used safely with other locking &sk_buff functions
3150 * A buffer cannot be placed on two lists at the same time.
3152 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3154 unsigned long flags;
3156 spin_lock_irqsave(&list->lock, flags);
3157 __skb_queue_head(list, newsk);
3158 spin_unlock_irqrestore(&list->lock, flags);
3160 EXPORT_SYMBOL(skb_queue_head);
3163 * skb_queue_tail - queue a buffer at the list tail
3164 * @list: list to use
3165 * @newsk: buffer to queue
3167 * Queue a buffer at the tail of the list. This function takes the
3168 * list lock and can be used safely with other locking &sk_buff functions
3171 * A buffer cannot be placed on two lists at the same time.
3173 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3175 unsigned long flags;
3177 spin_lock_irqsave(&list->lock, flags);
3178 __skb_queue_tail(list, newsk);
3179 spin_unlock_irqrestore(&list->lock, flags);
3181 EXPORT_SYMBOL(skb_queue_tail);
3184 * skb_unlink - remove a buffer from a list
3185 * @skb: buffer to remove
3186 * @list: list to use
3188 * Remove a packet from a list. The list locks are taken and this
3189 * function is atomic with respect to other list locked calls
3191 * You must know what list the SKB is on.
3193 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3195 unsigned long flags;
3197 spin_lock_irqsave(&list->lock, flags);
3198 __skb_unlink(skb, list);
3199 spin_unlock_irqrestore(&list->lock, flags);
3201 EXPORT_SYMBOL(skb_unlink);
3204 * skb_append - append a buffer
3205 * @old: buffer to insert after
3206 * @newsk: buffer to insert
3207 * @list: list to use
3209 * Place a packet after a given packet in a list. The list locks are taken
3210 * and this function is atomic with respect to other list locked calls.
3211 * A buffer cannot be placed on two lists at the same time.
3213 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3215 unsigned long flags;
3217 spin_lock_irqsave(&list->lock, flags);
3218 __skb_queue_after(list, old, newsk);
3219 spin_unlock_irqrestore(&list->lock, flags);
3221 EXPORT_SYMBOL(skb_append);
3223 static inline void skb_split_inside_header(struct sk_buff *skb,
3224 struct sk_buff* skb1,
3225 const u32 len, const int pos)
3229 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3231 /* And move data appendix as is. */
3232 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3233 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3235 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3236 skb_shinfo(skb)->nr_frags = 0;
3237 skb1->data_len = skb->data_len;
3238 skb1->len += skb1->data_len;
3241 skb_set_tail_pointer(skb, len);
3244 static inline void skb_split_no_header(struct sk_buff *skb,
3245 struct sk_buff* skb1,
3246 const u32 len, int pos)
3249 const int nfrags = skb_shinfo(skb)->nr_frags;
3251 skb_shinfo(skb)->nr_frags = 0;
3252 skb1->len = skb1->data_len = skb->len - len;
3254 skb->data_len = len - pos;
3256 for (i = 0; i < nfrags; i++) {
3257 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3259 if (pos + size > len) {
3260 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3264 * We have two variants in this case:
3265 * 1. Move all the frag to the second
3266 * part, if it is possible. F.e.
3267 * this approach is mandatory for TUX,
3268 * where splitting is expensive.
3269 * 2. Split is accurately. We make this.
3271 skb_frag_ref(skb, i);
3272 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3273 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3274 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3275 skb_shinfo(skb)->nr_frags++;
3279 skb_shinfo(skb)->nr_frags++;
3282 skb_shinfo(skb1)->nr_frags = k;
3286 * skb_split - Split fragmented skb to two parts at length len.
3287 * @skb: the buffer to split
3288 * @skb1: the buffer to receive the second part
3289 * @len: new length for skb
3291 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3293 int pos = skb_headlen(skb);
3295 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3297 skb_zerocopy_clone(skb1, skb, 0);
3298 if (len < pos) /* Split line is inside header. */
3299 skb_split_inside_header(skb, skb1, len, pos);
3300 else /* Second chunk has no header, nothing to copy. */
3301 skb_split_no_header(skb, skb1, len, pos);
3303 EXPORT_SYMBOL(skb_split);
3305 /* Shifting from/to a cloned skb is a no-go.
3307 * Caller cannot keep skb_shinfo related pointers past calling here!
3309 static int skb_prepare_for_shift(struct sk_buff *skb)
3313 if (skb_cloned(skb)) {
3314 /* Save and restore truesize: pskb_expand_head() may reallocate
3315 * memory where ksize(kmalloc(S)) != ksize(kmalloc(S)), but we
3316 * cannot change truesize at this point.
3318 unsigned int save_truesize = skb->truesize;
3320 ret = pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3321 skb->truesize = save_truesize;
3327 * skb_shift - Shifts paged data partially from skb to another
3328 * @tgt: buffer into which tail data gets added
3329 * @skb: buffer from which the paged data comes from
3330 * @shiftlen: shift up to this many bytes
3332 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3333 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3334 * It's up to caller to free skb if everything was shifted.
3336 * If @tgt runs out of frags, the whole operation is aborted.
3338 * Skb cannot include anything else but paged data while tgt is allowed
3339 * to have non-paged data as well.
3341 * TODO: full sized shift could be optimized but that would need
3342 * specialized skb free'er to handle frags without up-to-date nr_frags.
3344 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3346 int from, to, merge, todo;
3347 skb_frag_t *fragfrom, *fragto;
3349 BUG_ON(shiftlen > skb->len);
3351 if (skb_headlen(skb))
3353 if (skb_zcopy(tgt) || skb_zcopy(skb))
3358 to = skb_shinfo(tgt)->nr_frags;
3359 fragfrom = &skb_shinfo(skb)->frags[from];
3361 /* Actual merge is delayed until the point when we know we can
3362 * commit all, so that we don't have to undo partial changes
3365 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3366 skb_frag_off(fragfrom))) {
3371 todo -= skb_frag_size(fragfrom);
3373 if (skb_prepare_for_shift(skb) ||
3374 skb_prepare_for_shift(tgt))
3377 /* All previous frag pointers might be stale! */
3378 fragfrom = &skb_shinfo(skb)->frags[from];
3379 fragto = &skb_shinfo(tgt)->frags[merge];
3381 skb_frag_size_add(fragto, shiftlen);
3382 skb_frag_size_sub(fragfrom, shiftlen);
3383 skb_frag_off_add(fragfrom, shiftlen);
3391 /* Skip full, not-fitting skb to avoid expensive operations */
3392 if ((shiftlen == skb->len) &&
3393 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3396 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3399 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3400 if (to == MAX_SKB_FRAGS)
3403 fragfrom = &skb_shinfo(skb)->frags[from];
3404 fragto = &skb_shinfo(tgt)->frags[to];
3406 if (todo >= skb_frag_size(fragfrom)) {
3407 *fragto = *fragfrom;
3408 todo -= skb_frag_size(fragfrom);
3413 __skb_frag_ref(fragfrom);
3414 skb_frag_page_copy(fragto, fragfrom);
3415 skb_frag_off_copy(fragto, fragfrom);
3416 skb_frag_size_set(fragto, todo);
3418 skb_frag_off_add(fragfrom, todo);
3419 skb_frag_size_sub(fragfrom, todo);
3427 /* Ready to "commit" this state change to tgt */
3428 skb_shinfo(tgt)->nr_frags = to;
3431 fragfrom = &skb_shinfo(skb)->frags[0];
3432 fragto = &skb_shinfo(tgt)->frags[merge];
3434 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3435 __skb_frag_unref(fragfrom);
3438 /* Reposition in the original skb */
3440 while (from < skb_shinfo(skb)->nr_frags)
3441 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3442 skb_shinfo(skb)->nr_frags = to;
3444 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3447 /* Most likely the tgt won't ever need its checksum anymore, skb on
3448 * the other hand might need it if it needs to be resent
3450 tgt->ip_summed = CHECKSUM_PARTIAL;
3451 skb->ip_summed = CHECKSUM_PARTIAL;
3453 /* Yak, is it really working this way? Some helper please? */
3454 skb->len -= shiftlen;
3455 skb->data_len -= shiftlen;
3456 skb->truesize -= shiftlen;
3457 tgt->len += shiftlen;
3458 tgt->data_len += shiftlen;
3459 tgt->truesize += shiftlen;
3465 * skb_prepare_seq_read - Prepare a sequential read of skb data
3466 * @skb: the buffer to read
3467 * @from: lower offset of data to be read
3468 * @to: upper offset of data to be read
3469 * @st: state variable
3471 * Initializes the specified state variable. Must be called before
3472 * invoking skb_seq_read() for the first time.
3474 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3475 unsigned int to, struct skb_seq_state *st)
3477 st->lower_offset = from;
3478 st->upper_offset = to;
3479 st->root_skb = st->cur_skb = skb;
3480 st->frag_idx = st->stepped_offset = 0;
3481 st->frag_data = NULL;
3483 EXPORT_SYMBOL(skb_prepare_seq_read);
3486 * skb_seq_read - Sequentially read skb data
3487 * @consumed: number of bytes consumed by the caller so far
3488 * @data: destination pointer for data to be returned
3489 * @st: state variable
3491 * Reads a block of skb data at @consumed relative to the
3492 * lower offset specified to skb_prepare_seq_read(). Assigns
3493 * the head of the data block to @data and returns the length
3494 * of the block or 0 if the end of the skb data or the upper
3495 * offset has been reached.
3497 * The caller is not required to consume all of the data
3498 * returned, i.e. @consumed is typically set to the number
3499 * of bytes already consumed and the next call to
3500 * skb_seq_read() will return the remaining part of the block.
3502 * Note 1: The size of each block of data returned can be arbitrary,
3503 * this limitation is the cost for zerocopy sequential
3504 * reads of potentially non linear data.
3506 * Note 2: Fragment lists within fragments are not implemented
3507 * at the moment, state->root_skb could be replaced with
3508 * a stack for this purpose.
3510 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3511 struct skb_seq_state *st)
3513 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3516 if (unlikely(abs_offset >= st->upper_offset)) {
3517 if (st->frag_data) {
3518 kunmap_atomic(st->frag_data);
3519 st->frag_data = NULL;
3525 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3527 if (abs_offset < block_limit && !st->frag_data) {
3528 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3529 return block_limit - abs_offset;
3532 if (st->frag_idx == 0 && !st->frag_data)
3533 st->stepped_offset += skb_headlen(st->cur_skb);
3535 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3536 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3537 block_limit = skb_frag_size(frag) + st->stepped_offset;
3539 if (abs_offset < block_limit) {
3541 st->frag_data = kmap_atomic(skb_frag_page(frag));
3543 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3544 (abs_offset - st->stepped_offset);
3546 return block_limit - abs_offset;
3549 if (st->frag_data) {
3550 kunmap_atomic(st->frag_data);
3551 st->frag_data = NULL;
3555 st->stepped_offset += skb_frag_size(frag);
3558 if (st->frag_data) {
3559 kunmap_atomic(st->frag_data);
3560 st->frag_data = NULL;
3563 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3564 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3567 } else if (st->cur_skb->next) {
3568 st->cur_skb = st->cur_skb->next;
3575 EXPORT_SYMBOL(skb_seq_read);
3578 * skb_abort_seq_read - Abort a sequential read of skb data
3579 * @st: state variable
3581 * Must be called if skb_seq_read() was not called until it
3584 void skb_abort_seq_read(struct skb_seq_state *st)
3587 kunmap_atomic(st->frag_data);
3589 EXPORT_SYMBOL(skb_abort_seq_read);
3591 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3593 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3594 struct ts_config *conf,
3595 struct ts_state *state)
3597 return skb_seq_read(offset, text, TS_SKB_CB(state));
3600 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3602 skb_abort_seq_read(TS_SKB_CB(state));
3606 * skb_find_text - Find a text pattern in skb data
3607 * @skb: the buffer to look in
3608 * @from: search offset
3610 * @config: textsearch configuration
3612 * Finds a pattern in the skb data according to the specified
3613 * textsearch configuration. Use textsearch_next() to retrieve
3614 * subsequent occurrences of the pattern. Returns the offset
3615 * to the first occurrence or UINT_MAX if no match was found.
3617 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3618 unsigned int to, struct ts_config *config)
3620 struct ts_state state;
3623 config->get_next_block = skb_ts_get_next_block;
3624 config->finish = skb_ts_finish;
3626 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3628 ret = textsearch_find(config, &state);
3629 return (ret <= to - from ? ret : UINT_MAX);
3631 EXPORT_SYMBOL(skb_find_text);
3633 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3634 int offset, size_t size)
3636 int i = skb_shinfo(skb)->nr_frags;
3638 if (skb_can_coalesce(skb, i, page, offset)) {
3639 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3640 } else if (i < MAX_SKB_FRAGS) {
3642 skb_fill_page_desc(skb, i, page, offset, size);
3649 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3652 * skb_pull_rcsum - pull skb and update receive checksum
3653 * @skb: buffer to update
3654 * @len: length of data pulled
3656 * This function performs an skb_pull on the packet and updates
3657 * the CHECKSUM_COMPLETE checksum. It should be used on
3658 * receive path processing instead of skb_pull unless you know
3659 * that the checksum difference is zero (e.g., a valid IP header)
3660 * or you are setting ip_summed to CHECKSUM_NONE.
3662 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3664 unsigned char *data = skb->data;
3666 BUG_ON(len > skb->len);
3667 __skb_pull(skb, len);
3668 skb_postpull_rcsum(skb, data, len);
3671 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3673 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3675 skb_frag_t head_frag;
3678 page = virt_to_head_page(frag_skb->head);
3679 __skb_frag_set_page(&head_frag, page);
3680 skb_frag_off_set(&head_frag, frag_skb->data -
3681 (unsigned char *)page_address(page));
3682 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3686 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3687 netdev_features_t features,
3688 unsigned int offset)
3690 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3691 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3692 unsigned int delta_truesize = 0;
3693 unsigned int delta_len = 0;
3694 struct sk_buff *tail = NULL;
3695 struct sk_buff *nskb, *tmp;
3698 skb_push(skb, -skb_network_offset(skb) + offset);
3700 /* Ensure the head is writeable before touching the shared info */
3701 err = skb_unclone(skb, GFP_ATOMIC);
3705 skb_shinfo(skb)->frag_list = NULL;
3709 list_skb = list_skb->next;
3712 delta_truesize += nskb->truesize;
3713 if (skb_shared(nskb)) {
3714 tmp = skb_clone(nskb, GFP_ATOMIC);
3718 err = skb_unclone(nskb, GFP_ATOMIC);
3729 if (unlikely(err)) {
3730 nskb->next = list_skb;
3736 delta_len += nskb->len;
3738 skb_push(nskb, -skb_network_offset(nskb) + offset);
3740 skb_release_head_state(nskb);
3741 __copy_skb_header(nskb, skb);
3743 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3744 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3745 nskb->data - tnl_hlen,
3748 if (skb_needs_linearize(nskb, features) &&
3749 __skb_linearize(nskb))
3753 skb->truesize = skb->truesize - delta_truesize;
3754 skb->data_len = skb->data_len - delta_len;
3755 skb->len = skb->len - delta_len;
3761 if (skb_needs_linearize(skb, features) &&
3762 __skb_linearize(skb))
3770 kfree_skb_list(skb->next);
3772 return ERR_PTR(-ENOMEM);
3774 EXPORT_SYMBOL_GPL(skb_segment_list);
3776 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3778 if (unlikely(p->len + skb->len >= 65536))
3781 if (NAPI_GRO_CB(p)->last == p)
3782 skb_shinfo(p)->frag_list = skb;
3784 NAPI_GRO_CB(p)->last->next = skb;
3786 skb_pull(skb, skb_gro_offset(skb));
3788 NAPI_GRO_CB(p)->last = skb;
3789 NAPI_GRO_CB(p)->count++;
3790 p->data_len += skb->len;
3791 p->truesize += skb->truesize;
3794 NAPI_GRO_CB(skb)->same_flow = 1;
3800 * skb_segment - Perform protocol segmentation on skb.
3801 * @head_skb: buffer to segment
3802 * @features: features for the output path (see dev->features)
3804 * This function performs segmentation on the given skb. It returns
3805 * a pointer to the first in a list of new skbs for the segments.
3806 * In case of error it returns ERR_PTR(err).
3808 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3809 netdev_features_t features)
3811 struct sk_buff *segs = NULL;
3812 struct sk_buff *tail = NULL;
3813 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3814 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3815 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3816 unsigned int offset = doffset;
3817 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3818 unsigned int partial_segs = 0;
3819 unsigned int headroom;
3820 unsigned int len = head_skb->len;
3821 struct sk_buff *frag_skb;
3829 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
3830 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
3831 struct sk_buff *check_skb;
3833 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
3834 if (skb_headlen(check_skb) && !check_skb->head_frag) {
3835 /* gso_size is untrusted, and we have a frag_list with
3836 * a linear non head_frag item.
3838 * If head_skb's headlen does not fit requested gso_size,
3839 * it means that the frag_list members do NOT terminate
3840 * on exact gso_size boundaries. Hence we cannot perform
3841 * skb_frag_t page sharing. Therefore we must fallback to
3842 * copying the frag_list skbs; we do so by disabling SG.
3844 features &= ~NETIF_F_SG;
3850 __skb_push(head_skb, doffset);
3851 proto = skb_network_protocol(head_skb, NULL);
3852 if (unlikely(!proto))
3853 return ERR_PTR(-EINVAL);
3855 sg = !!(features & NETIF_F_SG);
3856 csum = !!can_checksum_protocol(features, proto);
3858 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3859 if (!(features & NETIF_F_GSO_PARTIAL)) {
3860 struct sk_buff *iter;
3861 unsigned int frag_len;
3864 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3867 /* If we get here then all the required
3868 * GSO features except frag_list are supported.
3869 * Try to split the SKB to multiple GSO SKBs
3870 * with no frag_list.
3871 * Currently we can do that only when the buffers don't
3872 * have a linear part and all the buffers except
3873 * the last are of the same length.
3875 frag_len = list_skb->len;
3876 skb_walk_frags(head_skb, iter) {
3877 if (frag_len != iter->len && iter->next)
3879 if (skb_headlen(iter) && !iter->head_frag)
3885 if (len != frag_len)
3889 /* GSO partial only requires that we trim off any excess that
3890 * doesn't fit into an MSS sized block, so take care of that
3892 * Cap len to not accidentally hit GSO_BY_FRAGS.
3894 partial_segs = min(len, GSO_BY_FRAGS - 1U) / mss;
3895 if (partial_segs > 1)
3896 mss *= partial_segs;
3902 headroom = skb_headroom(head_skb);
3903 pos = skb_headlen(head_skb);
3905 if (skb_orphan_frags(head_skb, GFP_ATOMIC))
3906 return ERR_PTR(-ENOMEM);
3908 nfrags = skb_shinfo(head_skb)->nr_frags;
3909 frag = skb_shinfo(head_skb)->frags;
3910 frag_skb = head_skb;
3913 struct sk_buff *nskb;
3914 skb_frag_t *nskb_frag;
3918 if (unlikely(mss == GSO_BY_FRAGS)) {
3919 len = list_skb->len;
3921 len = head_skb->len - offset;
3926 hsize = skb_headlen(head_skb) - offset;
3929 if (hsize > len || !sg)
3932 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3933 (skb_headlen(list_skb) == len || sg)) {
3934 BUG_ON(skb_headlen(list_skb) > len);
3936 nskb = skb_clone(list_skb, GFP_ATOMIC);
3937 if (unlikely(!nskb))
3941 nfrags = skb_shinfo(list_skb)->nr_frags;
3942 frag = skb_shinfo(list_skb)->frags;
3943 frag_skb = list_skb;
3944 pos += skb_headlen(list_skb);
3946 while (pos < offset + len) {
3947 BUG_ON(i >= nfrags);
3949 size = skb_frag_size(frag);
3950 if (pos + size > offset + len)
3958 list_skb = list_skb->next;
3960 if (unlikely(pskb_trim(nskb, len))) {
3965 hsize = skb_end_offset(nskb);
3966 if (skb_cow_head(nskb, doffset + headroom)) {
3971 nskb->truesize += skb_end_offset(nskb) - hsize;
3972 skb_release_head_state(nskb);
3973 __skb_push(nskb, doffset);
3975 nskb = __alloc_skb(hsize + doffset + headroom,
3976 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3979 if (unlikely(!nskb))
3982 skb_reserve(nskb, headroom);
3983 __skb_put(nskb, doffset);
3992 __copy_skb_header(nskb, head_skb);
3994 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3995 skb_reset_mac_len(nskb);
3997 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3998 nskb->data - tnl_hlen,
3999 doffset + tnl_hlen);
4001 if (nskb->len == len + doffset)
4002 goto perform_csum_check;
4006 if (!nskb->remcsum_offload)
4007 nskb->ip_summed = CHECKSUM_NONE;
4008 SKB_GSO_CB(nskb)->csum =
4009 skb_copy_and_csum_bits(head_skb, offset,
4013 SKB_GSO_CB(nskb)->csum_start =
4014 skb_headroom(nskb) + doffset;
4016 if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4022 nskb_frag = skb_shinfo(nskb)->frags;
4024 skb_copy_from_linear_data_offset(head_skb, offset,
4025 skb_put(nskb, hsize), hsize);
4027 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
4030 if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4033 while (pos < offset + len) {
4035 if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
4036 skb_zerocopy_clone(nskb, list_skb,
4041 nfrags = skb_shinfo(list_skb)->nr_frags;
4042 frag = skb_shinfo(list_skb)->frags;
4043 frag_skb = list_skb;
4044 if (!skb_headlen(list_skb)) {
4047 BUG_ON(!list_skb->head_frag);
4049 /* to make room for head_frag. */
4054 list_skb = list_skb->next;
4057 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4059 net_warn_ratelimited(
4060 "skb_segment: too many frags: %u %u\n",
4066 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4067 __skb_frag_ref(nskb_frag);
4068 size = skb_frag_size(nskb_frag);
4071 skb_frag_off_add(nskb_frag, offset - pos);
4072 skb_frag_size_sub(nskb_frag, offset - pos);
4075 skb_shinfo(nskb)->nr_frags++;
4077 if (pos + size <= offset + len) {
4082 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4090 nskb->data_len = len - hsize;
4091 nskb->len += nskb->data_len;
4092 nskb->truesize += nskb->data_len;
4096 if (skb_has_shared_frag(nskb) &&
4097 __skb_linearize(nskb))
4100 if (!nskb->remcsum_offload)
4101 nskb->ip_summed = CHECKSUM_NONE;
4102 SKB_GSO_CB(nskb)->csum =
4103 skb_checksum(nskb, doffset,
4104 nskb->len - doffset, 0);
4105 SKB_GSO_CB(nskb)->csum_start =
4106 skb_headroom(nskb) + doffset;
4108 } while ((offset += len) < head_skb->len);
4110 /* Some callers want to get the end of the list.
4111 * Put it in segs->prev to avoid walking the list.
4112 * (see validate_xmit_skb_list() for example)
4117 struct sk_buff *iter;
4118 int type = skb_shinfo(head_skb)->gso_type;
4119 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4121 /* Update type to add partial and then remove dodgy if set */
4122 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4123 type &= ~SKB_GSO_DODGY;
4125 /* Update GSO info and prepare to start updating headers on
4126 * our way back down the stack of protocols.
4128 for (iter = segs; iter; iter = iter->next) {
4129 skb_shinfo(iter)->gso_size = gso_size;
4130 skb_shinfo(iter)->gso_segs = partial_segs;
4131 skb_shinfo(iter)->gso_type = type;
4132 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4135 if (tail->len - doffset <= gso_size)
4136 skb_shinfo(tail)->gso_size = 0;
4137 else if (tail != segs)
4138 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4141 /* Following permits correct backpressure, for protocols
4142 * using skb_set_owner_w().
4143 * Idea is to tranfert ownership from head_skb to last segment.
4145 if (head_skb->destructor == sock_wfree) {
4146 swap(tail->truesize, head_skb->truesize);
4147 swap(tail->destructor, head_skb->destructor);
4148 swap(tail->sk, head_skb->sk);
4153 kfree_skb_list(segs);
4154 return ERR_PTR(err);
4156 EXPORT_SYMBOL_GPL(skb_segment);
4158 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4160 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4161 unsigned int offset = skb_gro_offset(skb);
4162 unsigned int headlen = skb_headlen(skb);
4163 unsigned int len = skb_gro_len(skb);
4164 unsigned int delta_truesize;
4167 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4170 lp = NAPI_GRO_CB(p)->last;
4171 pinfo = skb_shinfo(lp);
4173 if (headlen <= offset) {
4176 int i = skbinfo->nr_frags;
4177 int nr_frags = pinfo->nr_frags + i;
4179 if (nr_frags > MAX_SKB_FRAGS)
4183 pinfo->nr_frags = nr_frags;
4184 skbinfo->nr_frags = 0;
4186 frag = pinfo->frags + nr_frags;
4187 frag2 = skbinfo->frags + i;
4192 skb_frag_off_add(frag, offset);
4193 skb_frag_size_sub(frag, offset);
4195 /* all fragments truesize : remove (head size + sk_buff) */
4196 delta_truesize = skb->truesize -
4197 SKB_TRUESIZE(skb_end_offset(skb));
4199 skb->truesize -= skb->data_len;
4200 skb->len -= skb->data_len;
4203 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4205 } else if (skb->head_frag) {
4206 int nr_frags = pinfo->nr_frags;
4207 skb_frag_t *frag = pinfo->frags + nr_frags;
4208 struct page *page = virt_to_head_page(skb->head);
4209 unsigned int first_size = headlen - offset;
4210 unsigned int first_offset;
4212 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4215 first_offset = skb->data -
4216 (unsigned char *)page_address(page) +
4219 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4221 __skb_frag_set_page(frag, page);
4222 skb_frag_off_set(frag, first_offset);
4223 skb_frag_size_set(frag, first_size);
4225 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4226 /* We dont need to clear skbinfo->nr_frags here */
4228 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4229 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4234 delta_truesize = skb->truesize;
4235 if (offset > headlen) {
4236 unsigned int eat = offset - headlen;
4238 skb_frag_off_add(&skbinfo->frags[0], eat);
4239 skb_frag_size_sub(&skbinfo->frags[0], eat);
4240 skb->data_len -= eat;
4245 __skb_pull(skb, offset);
4247 if (NAPI_GRO_CB(p)->last == p)
4248 skb_shinfo(p)->frag_list = skb;
4250 NAPI_GRO_CB(p)->last->next = skb;
4251 NAPI_GRO_CB(p)->last = skb;
4252 __skb_header_release(skb);
4256 NAPI_GRO_CB(p)->count++;
4258 p->truesize += delta_truesize;
4261 lp->data_len += len;
4262 lp->truesize += delta_truesize;
4265 NAPI_GRO_CB(skb)->same_flow = 1;
4269 #ifdef CONFIG_SKB_EXTENSIONS
4270 #define SKB_EXT_ALIGN_VALUE 8
4271 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4273 static const u8 skb_ext_type_len[] = {
4274 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4275 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4278 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4280 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4281 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4283 #if IS_ENABLED(CONFIG_MPTCP)
4284 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4288 static __always_inline unsigned int skb_ext_total_length(void)
4290 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4291 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4292 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4295 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4297 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4298 skb_ext_type_len[TC_SKB_EXT] +
4300 #if IS_ENABLED(CONFIG_MPTCP)
4301 skb_ext_type_len[SKB_EXT_MPTCP] +
4306 static void skb_extensions_init(void)
4308 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4309 BUILD_BUG_ON(skb_ext_total_length() > 255);
4311 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4312 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4314 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4318 static void skb_extensions_init(void) {}
4321 void __init skb_init(void)
4323 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4324 sizeof(struct sk_buff),
4326 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4327 offsetof(struct sk_buff, cb),
4328 sizeof_field(struct sk_buff, cb),
4330 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4331 sizeof(struct sk_buff_fclones),
4333 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4335 skb_extensions_init();
4339 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4340 unsigned int recursion_level)
4342 int start = skb_headlen(skb);
4343 int i, copy = start - offset;
4344 struct sk_buff *frag_iter;
4347 if (unlikely(recursion_level >= 24))
4353 sg_set_buf(sg, skb->data + offset, copy);
4355 if ((len -= copy) == 0)
4360 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4363 WARN_ON(start > offset + len);
4365 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4366 if ((copy = end - offset) > 0) {
4367 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4368 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4373 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4374 skb_frag_off(frag) + offset - start);
4383 skb_walk_frags(skb, frag_iter) {
4386 WARN_ON(start > offset + len);
4388 end = start + frag_iter->len;
4389 if ((copy = end - offset) > 0) {
4390 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4395 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4396 copy, recursion_level + 1);
4397 if (unlikely(ret < 0))
4400 if ((len -= copy) == 0)
4411 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4412 * @skb: Socket buffer containing the buffers to be mapped
4413 * @sg: The scatter-gather list to map into
4414 * @offset: The offset into the buffer's contents to start mapping
4415 * @len: Length of buffer space to be mapped
4417 * Fill the specified scatter-gather list with mappings/pointers into a
4418 * region of the buffer space attached to a socket buffer. Returns either
4419 * the number of scatterlist items used, or -EMSGSIZE if the contents
4422 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4424 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4429 sg_mark_end(&sg[nsg - 1]);
4433 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4435 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4436 * sglist without mark the sg which contain last skb data as the end.
4437 * So the caller can mannipulate sg list as will when padding new data after
4438 * the first call without calling sg_unmark_end to expend sg list.
4440 * Scenario to use skb_to_sgvec_nomark:
4442 * 2. skb_to_sgvec_nomark(payload1)
4443 * 3. skb_to_sgvec_nomark(payload2)
4445 * This is equivalent to:
4447 * 2. skb_to_sgvec(payload1)
4449 * 4. skb_to_sgvec(payload2)
4451 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4452 * is more preferable.
4454 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4455 int offset, int len)
4457 return __skb_to_sgvec(skb, sg, offset, len, 0);
4459 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4464 * skb_cow_data - Check that a socket buffer's data buffers are writable
4465 * @skb: The socket buffer to check.
4466 * @tailbits: Amount of trailing space to be added
4467 * @trailer: Returned pointer to the skb where the @tailbits space begins
4469 * Make sure that the data buffers attached to a socket buffer are
4470 * writable. If they are not, private copies are made of the data buffers
4471 * and the socket buffer is set to use these instead.
4473 * If @tailbits is given, make sure that there is space to write @tailbits
4474 * bytes of data beyond current end of socket buffer. @trailer will be
4475 * set to point to the skb in which this space begins.
4477 * The number of scatterlist elements required to completely map the
4478 * COW'd and extended socket buffer will be returned.
4480 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4484 struct sk_buff *skb1, **skb_p;
4486 /* If skb is cloned or its head is paged, reallocate
4487 * head pulling out all the pages (pages are considered not writable
4488 * at the moment even if they are anonymous).
4490 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4491 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4494 /* Easy case. Most of packets will go this way. */
4495 if (!skb_has_frag_list(skb)) {
4496 /* A little of trouble, not enough of space for trailer.
4497 * This should not happen, when stack is tuned to generate
4498 * good frames. OK, on miss we reallocate and reserve even more
4499 * space, 128 bytes is fair. */
4501 if (skb_tailroom(skb) < tailbits &&
4502 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4510 /* Misery. We are in troubles, going to mincer fragments... */
4513 skb_p = &skb_shinfo(skb)->frag_list;
4516 while ((skb1 = *skb_p) != NULL) {
4519 /* The fragment is partially pulled by someone,
4520 * this can happen on input. Copy it and everything
4523 if (skb_shared(skb1))
4526 /* If the skb is the last, worry about trailer. */
4528 if (skb1->next == NULL && tailbits) {
4529 if (skb_shinfo(skb1)->nr_frags ||
4530 skb_has_frag_list(skb1) ||
4531 skb_tailroom(skb1) < tailbits)
4532 ntail = tailbits + 128;
4538 skb_shinfo(skb1)->nr_frags ||
4539 skb_has_frag_list(skb1)) {
4540 struct sk_buff *skb2;
4542 /* Fuck, we are miserable poor guys... */
4544 skb2 = skb_copy(skb1, GFP_ATOMIC);
4546 skb2 = skb_copy_expand(skb1,
4550 if (unlikely(skb2 == NULL))
4554 skb_set_owner_w(skb2, skb1->sk);
4556 /* Looking around. Are we still alive?
4557 * OK, link new skb, drop old one */
4559 skb2->next = skb1->next;
4566 skb_p = &skb1->next;
4571 EXPORT_SYMBOL_GPL(skb_cow_data);
4573 static void sock_rmem_free(struct sk_buff *skb)
4575 struct sock *sk = skb->sk;
4577 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4580 static void skb_set_err_queue(struct sk_buff *skb)
4582 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4583 * So, it is safe to (mis)use it to mark skbs on the error queue.
4585 skb->pkt_type = PACKET_OUTGOING;
4586 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4590 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4592 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4594 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4595 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4600 skb->destructor = sock_rmem_free;
4601 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4602 skb_set_err_queue(skb);
4604 /* before exiting rcu section, make sure dst is refcounted */
4607 skb_queue_tail(&sk->sk_error_queue, skb);
4608 if (!sock_flag(sk, SOCK_DEAD))
4609 sk->sk_error_report(sk);
4612 EXPORT_SYMBOL(sock_queue_err_skb);
4614 static bool is_icmp_err_skb(const struct sk_buff *skb)
4616 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4617 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4620 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4622 struct sk_buff_head *q = &sk->sk_error_queue;
4623 struct sk_buff *skb, *skb_next = NULL;
4624 bool icmp_next = false;
4625 unsigned long flags;
4627 spin_lock_irqsave(&q->lock, flags);
4628 skb = __skb_dequeue(q);
4629 if (skb && (skb_next = skb_peek(q))) {
4630 icmp_next = is_icmp_err_skb(skb_next);
4632 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4634 spin_unlock_irqrestore(&q->lock, flags);
4636 if (is_icmp_err_skb(skb) && !icmp_next)
4640 sk->sk_error_report(sk);
4644 EXPORT_SYMBOL(sock_dequeue_err_skb);
4647 * skb_clone_sk - create clone of skb, and take reference to socket
4648 * @skb: the skb to clone
4650 * This function creates a clone of a buffer that holds a reference on
4651 * sk_refcnt. Buffers created via this function are meant to be
4652 * returned using sock_queue_err_skb, or free via kfree_skb.
4654 * When passing buffers allocated with this function to sock_queue_err_skb
4655 * it is necessary to wrap the call with sock_hold/sock_put in order to
4656 * prevent the socket from being released prior to being enqueued on
4657 * the sk_error_queue.
4659 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4661 struct sock *sk = skb->sk;
4662 struct sk_buff *clone;
4664 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4667 clone = skb_clone(skb, GFP_ATOMIC);
4674 clone->destructor = sock_efree;
4678 EXPORT_SYMBOL(skb_clone_sk);
4680 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4685 struct sock_exterr_skb *serr;
4688 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4690 serr = SKB_EXT_ERR(skb);
4691 memset(serr, 0, sizeof(*serr));
4692 serr->ee.ee_errno = ENOMSG;
4693 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4694 serr->ee.ee_info = tstype;
4695 serr->opt_stats = opt_stats;
4696 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4697 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4698 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4699 if (sk->sk_protocol == IPPROTO_TCP &&
4700 sk->sk_type == SOCK_STREAM)
4701 serr->ee.ee_data -= sk->sk_tskey;
4704 err = sock_queue_err_skb(sk, skb);
4710 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4714 if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
4717 read_lock_bh(&sk->sk_callback_lock);
4718 ret = sk->sk_socket && sk->sk_socket->file &&
4719 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4720 read_unlock_bh(&sk->sk_callback_lock);
4724 void skb_complete_tx_timestamp(struct sk_buff *skb,
4725 struct skb_shared_hwtstamps *hwtstamps)
4727 struct sock *sk = skb->sk;
4729 if (!skb_may_tx_timestamp(sk, false))
4732 /* Take a reference to prevent skb_orphan() from freeing the socket,
4733 * but only if the socket refcount is not zero.
4735 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4736 *skb_hwtstamps(skb) = *hwtstamps;
4737 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4745 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4747 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4748 struct skb_shared_hwtstamps *hwtstamps,
4749 struct sock *sk, int tstype)
4751 struct sk_buff *skb;
4752 bool tsonly, opt_stats = false;
4757 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4758 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4761 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4762 if (!skb_may_tx_timestamp(sk, tsonly))
4767 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4768 sk->sk_protocol == IPPROTO_TCP &&
4769 sk->sk_type == SOCK_STREAM) {
4770 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4774 skb = alloc_skb(0, GFP_ATOMIC);
4776 skb = skb_clone(orig_skb, GFP_ATOMIC);
4778 if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
4787 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4789 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4793 *skb_hwtstamps(skb) = *hwtstamps;
4795 skb->tstamp = ktime_get_real();
4797 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4799 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4801 void skb_tstamp_tx(struct sk_buff *orig_skb,
4802 struct skb_shared_hwtstamps *hwtstamps)
4804 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4807 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4809 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4811 struct sock *sk = skb->sk;
4812 struct sock_exterr_skb *serr;
4815 skb->wifi_acked_valid = 1;
4816 skb->wifi_acked = acked;
4818 serr = SKB_EXT_ERR(skb);
4819 memset(serr, 0, sizeof(*serr));
4820 serr->ee.ee_errno = ENOMSG;
4821 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4823 /* Take a reference to prevent skb_orphan() from freeing the socket,
4824 * but only if the socket refcount is not zero.
4826 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4827 err = sock_queue_err_skb(sk, skb);
4833 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4836 * skb_partial_csum_set - set up and verify partial csum values for packet
4837 * @skb: the skb to set
4838 * @start: the number of bytes after skb->data to start checksumming.
4839 * @off: the offset from start to place the checksum.
4841 * For untrusted partially-checksummed packets, we need to make sure the values
4842 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4844 * This function checks and sets those values and skb->ip_summed: if this
4845 * returns false you should drop the packet.
4847 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4849 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4850 u32 csum_start = skb_headroom(skb) + (u32)start;
4852 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4853 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4854 start, off, skb_headroom(skb), skb_headlen(skb));
4857 skb->ip_summed = CHECKSUM_PARTIAL;
4858 skb->csum_start = csum_start;
4859 skb->csum_offset = off;
4860 skb_set_transport_header(skb, start);
4863 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4865 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4868 if (skb_headlen(skb) >= len)
4871 /* If we need to pullup then pullup to the max, so we
4872 * won't need to do it again.
4877 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4880 if (skb_headlen(skb) < len)
4886 #define MAX_TCP_HDR_LEN (15 * 4)
4888 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4889 typeof(IPPROTO_IP) proto,
4896 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4897 off + MAX_TCP_HDR_LEN);
4898 if (!err && !skb_partial_csum_set(skb, off,
4899 offsetof(struct tcphdr,
4902 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4905 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4906 off + sizeof(struct udphdr));
4907 if (!err && !skb_partial_csum_set(skb, off,
4908 offsetof(struct udphdr,
4911 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4914 return ERR_PTR(-EPROTO);
4917 /* This value should be large enough to cover a tagged ethernet header plus
4918 * maximally sized IP and TCP or UDP headers.
4920 #define MAX_IP_HDR_LEN 128
4922 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4931 err = skb_maybe_pull_tail(skb,
4932 sizeof(struct iphdr),
4937 if (ip_is_fragment(ip_hdr(skb)))
4940 off = ip_hdrlen(skb);
4947 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4949 return PTR_ERR(csum);
4952 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4955 ip_hdr(skb)->protocol, 0);
4962 /* This value should be large enough to cover a tagged ethernet header plus
4963 * an IPv6 header, all options, and a maximal TCP or UDP header.
4965 #define MAX_IPV6_HDR_LEN 256
4967 #define OPT_HDR(type, skb, off) \
4968 (type *)(skb_network_header(skb) + (off))
4970 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4983 off = sizeof(struct ipv6hdr);
4985 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4989 nexthdr = ipv6_hdr(skb)->nexthdr;
4991 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4992 while (off <= len && !done) {
4994 case IPPROTO_DSTOPTS:
4995 case IPPROTO_HOPOPTS:
4996 case IPPROTO_ROUTING: {
4997 struct ipv6_opt_hdr *hp;
4999 err = skb_maybe_pull_tail(skb,
5001 sizeof(struct ipv6_opt_hdr),
5006 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5007 nexthdr = hp->nexthdr;
5008 off += ipv6_optlen(hp);
5012 struct ip_auth_hdr *hp;
5014 err = skb_maybe_pull_tail(skb,
5016 sizeof(struct ip_auth_hdr),
5021 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5022 nexthdr = hp->nexthdr;
5023 off += ipv6_authlen(hp);
5026 case IPPROTO_FRAGMENT: {
5027 struct frag_hdr *hp;
5029 err = skb_maybe_pull_tail(skb,
5031 sizeof(struct frag_hdr),
5036 hp = OPT_HDR(struct frag_hdr, skb, off);
5038 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5041 nexthdr = hp->nexthdr;
5042 off += sizeof(struct frag_hdr);
5053 if (!done || fragment)
5056 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5058 return PTR_ERR(csum);
5061 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5062 &ipv6_hdr(skb)->daddr,
5063 skb->len - off, nexthdr, 0);
5071 * skb_checksum_setup - set up partial checksum offset
5072 * @skb: the skb to set up
5073 * @recalculate: if true the pseudo-header checksum will be recalculated
5075 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5079 switch (skb->protocol) {
5080 case htons(ETH_P_IP):
5081 err = skb_checksum_setup_ipv4(skb, recalculate);
5084 case htons(ETH_P_IPV6):
5085 err = skb_checksum_setup_ipv6(skb, recalculate);
5095 EXPORT_SYMBOL(skb_checksum_setup);
5098 * skb_checksum_maybe_trim - maybe trims the given skb
5099 * @skb: the skb to check
5100 * @transport_len: the data length beyond the network header
5102 * Checks whether the given skb has data beyond the given transport length.
5103 * If so, returns a cloned skb trimmed to this transport length.
5104 * Otherwise returns the provided skb. Returns NULL in error cases
5105 * (e.g. transport_len exceeds skb length or out-of-memory).
5107 * Caller needs to set the skb transport header and free any returned skb if it
5108 * differs from the provided skb.
5110 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5111 unsigned int transport_len)
5113 struct sk_buff *skb_chk;
5114 unsigned int len = skb_transport_offset(skb) + transport_len;
5119 else if (skb->len == len)
5122 skb_chk = skb_clone(skb, GFP_ATOMIC);
5126 ret = pskb_trim_rcsum(skb_chk, len);
5136 * skb_checksum_trimmed - validate checksum of an skb
5137 * @skb: the skb to check
5138 * @transport_len: the data length beyond the network header
5139 * @skb_chkf: checksum function to use
5141 * Applies the given checksum function skb_chkf to the provided skb.
5142 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5144 * If the skb has data beyond the given transport length, then a
5145 * trimmed & cloned skb is checked and returned.
5147 * Caller needs to set the skb transport header and free any returned skb if it
5148 * differs from the provided skb.
5150 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5151 unsigned int transport_len,
5152 __sum16(*skb_chkf)(struct sk_buff *skb))
5154 struct sk_buff *skb_chk;
5155 unsigned int offset = skb_transport_offset(skb);
5158 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5162 if (!pskb_may_pull(skb_chk, offset))
5165 skb_pull_rcsum(skb_chk, offset);
5166 ret = skb_chkf(skb_chk);
5167 skb_push_rcsum(skb_chk, offset);
5175 if (skb_chk && skb_chk != skb)
5181 EXPORT_SYMBOL(skb_checksum_trimmed);
5183 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5185 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5188 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5190 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5193 skb_release_head_state(skb);
5194 kmem_cache_free(skbuff_head_cache, skb);
5199 EXPORT_SYMBOL(kfree_skb_partial);
5202 * skb_try_coalesce - try to merge skb to prior one
5204 * @from: buffer to add
5205 * @fragstolen: pointer to boolean
5206 * @delta_truesize: how much more was allocated than was requested
5208 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5209 bool *fragstolen, int *delta_truesize)
5211 struct skb_shared_info *to_shinfo, *from_shinfo;
5212 int i, delta, len = from->len;
5214 *fragstolen = false;
5219 if (len <= skb_tailroom(to)) {
5221 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5222 *delta_truesize = 0;
5226 to_shinfo = skb_shinfo(to);
5227 from_shinfo = skb_shinfo(from);
5228 if (to_shinfo->frag_list || from_shinfo->frag_list)
5230 if (skb_zcopy(to) || skb_zcopy(from))
5233 if (skb_headlen(from) != 0) {
5235 unsigned int offset;
5237 if (to_shinfo->nr_frags +
5238 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5241 if (skb_head_is_locked(from))
5244 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5246 page = virt_to_head_page(from->head);
5247 offset = from->data - (unsigned char *)page_address(page);
5249 skb_fill_page_desc(to, to_shinfo->nr_frags,
5250 page, offset, skb_headlen(from));
5253 if (to_shinfo->nr_frags +
5254 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5257 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5260 WARN_ON_ONCE(delta < len);
5262 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5264 from_shinfo->nr_frags * sizeof(skb_frag_t));
5265 to_shinfo->nr_frags += from_shinfo->nr_frags;
5267 if (!skb_cloned(from))
5268 from_shinfo->nr_frags = 0;
5270 /* if the skb is not cloned this does nothing
5271 * since we set nr_frags to 0.
5273 for (i = 0; i < from_shinfo->nr_frags; i++)
5274 __skb_frag_ref(&from_shinfo->frags[i]);
5276 to->truesize += delta;
5278 to->data_len += len;
5280 *delta_truesize = delta;
5283 EXPORT_SYMBOL(skb_try_coalesce);
5286 * skb_scrub_packet - scrub an skb
5288 * @skb: buffer to clean
5289 * @xnet: packet is crossing netns
5291 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5292 * into/from a tunnel. Some information have to be cleared during these
5294 * skb_scrub_packet can also be used to clean a skb before injecting it in
5295 * another namespace (@xnet == true). We have to clear all information in the
5296 * skb that could impact namespace isolation.
5298 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5300 skb->pkt_type = PACKET_HOST;
5306 nf_reset_trace(skb);
5308 #ifdef CONFIG_NET_SWITCHDEV
5309 skb->offload_fwd_mark = 0;
5310 skb->offload_l3_fwd_mark = 0;
5320 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5323 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5327 * skb_gso_transport_seglen is used to determine the real size of the
5328 * individual segments, including Layer4 headers (TCP/UDP).
5330 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5332 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5334 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5335 unsigned int thlen = 0;
5337 if (skb->encapsulation) {
5338 thlen = skb_inner_transport_header(skb) -
5339 skb_transport_header(skb);
5341 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5342 thlen += inner_tcp_hdrlen(skb);
5343 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5344 thlen = tcp_hdrlen(skb);
5345 } else if (unlikely(skb_is_gso_sctp(skb))) {
5346 thlen = sizeof(struct sctphdr);
5347 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5348 thlen = sizeof(struct udphdr);
5350 /* UFO sets gso_size to the size of the fragmentation
5351 * payload, i.e. the size of the L4 (UDP) header is already
5354 return thlen + shinfo->gso_size;
5358 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5362 * skb_gso_network_seglen is used to determine the real size of the
5363 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5365 * The MAC/L2 header is not accounted for.
5367 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5369 unsigned int hdr_len = skb_transport_header(skb) -
5370 skb_network_header(skb);
5372 return hdr_len + skb_gso_transport_seglen(skb);
5376 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5380 * skb_gso_mac_seglen is used to determine the real size of the
5381 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5382 * headers (TCP/UDP).
5384 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5386 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5388 return hdr_len + skb_gso_transport_seglen(skb);
5392 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5394 * There are a couple of instances where we have a GSO skb, and we
5395 * want to determine what size it would be after it is segmented.
5397 * We might want to check:
5398 * - L3+L4+payload size (e.g. IP forwarding)
5399 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5401 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5405 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5406 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5408 * @max_len: The maximum permissible length.
5410 * Returns true if the segmented length <= max length.
5412 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5413 unsigned int seg_len,
5414 unsigned int max_len) {
5415 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5416 const struct sk_buff *iter;
5418 if (shinfo->gso_size != GSO_BY_FRAGS)
5419 return seg_len <= max_len;
5421 /* Undo this so we can re-use header sizes */
5422 seg_len -= GSO_BY_FRAGS;
5424 skb_walk_frags(skb, iter) {
5425 if (seg_len + skb_headlen(iter) > max_len)
5433 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5436 * @mtu: MTU to validate against
5438 * skb_gso_validate_network_len validates if a given skb will fit a
5439 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5442 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5444 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5446 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5449 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5452 * @len: length to validate against
5454 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5455 * length once split, including L2, L3 and L4 headers and the payload.
5457 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5459 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5461 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5463 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5465 int mac_len, meta_len;
5468 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5473 mac_len = skb->data - skb_mac_header(skb);
5474 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5475 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5476 mac_len - VLAN_HLEN - ETH_TLEN);
5479 meta_len = skb_metadata_len(skb);
5481 meta = skb_metadata_end(skb) - meta_len;
5482 memmove(meta + VLAN_HLEN, meta, meta_len);
5485 skb->mac_header += VLAN_HLEN;
5489 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5491 struct vlan_hdr *vhdr;
5494 if (unlikely(skb_vlan_tag_present(skb))) {
5495 /* vlan_tci is already set-up so leave this for another time */
5499 skb = skb_share_check(skb, GFP_ATOMIC);
5502 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5503 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5506 vhdr = (struct vlan_hdr *)skb->data;
5507 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5508 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5510 skb_pull_rcsum(skb, VLAN_HLEN);
5511 vlan_set_encap_proto(skb, vhdr);
5513 skb = skb_reorder_vlan_header(skb);
5517 skb_reset_network_header(skb);
5518 skb_reset_transport_header(skb);
5519 skb_reset_mac_len(skb);
5527 EXPORT_SYMBOL(skb_vlan_untag);
5529 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5531 if (!pskb_may_pull(skb, write_len))
5534 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5537 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5539 EXPORT_SYMBOL(skb_ensure_writable);
5541 /* remove VLAN header from packet and update csum accordingly.
5542 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5544 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5546 struct vlan_hdr *vhdr;
5547 int offset = skb->data - skb_mac_header(skb);
5550 if (WARN_ONCE(offset,
5551 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5556 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5560 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5562 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5563 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5565 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5566 __skb_pull(skb, VLAN_HLEN);
5568 vlan_set_encap_proto(skb, vhdr);
5569 skb->mac_header += VLAN_HLEN;
5571 if (skb_network_offset(skb) < ETH_HLEN)
5572 skb_set_network_header(skb, ETH_HLEN);
5574 skb_reset_mac_len(skb);
5578 EXPORT_SYMBOL(__skb_vlan_pop);
5580 /* Pop a vlan tag either from hwaccel or from payload.
5581 * Expects skb->data at mac header.
5583 int skb_vlan_pop(struct sk_buff *skb)
5589 if (likely(skb_vlan_tag_present(skb))) {
5590 __vlan_hwaccel_clear_tag(skb);
5592 if (unlikely(!eth_type_vlan(skb->protocol)))
5595 err = __skb_vlan_pop(skb, &vlan_tci);
5599 /* move next vlan tag to hw accel tag */
5600 if (likely(!eth_type_vlan(skb->protocol)))
5603 vlan_proto = skb->protocol;
5604 err = __skb_vlan_pop(skb, &vlan_tci);
5608 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5611 EXPORT_SYMBOL(skb_vlan_pop);
5613 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5614 * Expects skb->data at mac header.
5616 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5618 if (skb_vlan_tag_present(skb)) {
5619 int offset = skb->data - skb_mac_header(skb);
5622 if (WARN_ONCE(offset,
5623 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5628 err = __vlan_insert_tag(skb, skb->vlan_proto,
5629 skb_vlan_tag_get(skb));
5633 skb->protocol = skb->vlan_proto;
5634 skb->mac_len += VLAN_HLEN;
5636 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5638 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5641 EXPORT_SYMBOL(skb_vlan_push);
5644 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5646 * @skb: Socket buffer to modify
5648 * Drop the Ethernet header of @skb.
5650 * Expects that skb->data points to the mac header and that no VLAN tags are
5653 * Returns 0 on success, -errno otherwise.
5655 int skb_eth_pop(struct sk_buff *skb)
5657 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5658 skb_network_offset(skb) < ETH_HLEN)
5661 skb_pull_rcsum(skb, ETH_HLEN);
5662 skb_reset_mac_header(skb);
5663 skb_reset_mac_len(skb);
5667 EXPORT_SYMBOL(skb_eth_pop);
5670 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5672 * @skb: Socket buffer to modify
5673 * @dst: Destination MAC address of the new header
5674 * @src: Source MAC address of the new header
5676 * Prepend @skb with a new Ethernet header.
5678 * Expects that skb->data points to the mac header, which must be empty.
5680 * Returns 0 on success, -errno otherwise.
5682 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5683 const unsigned char *src)
5688 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5691 err = skb_cow_head(skb, sizeof(*eth));
5695 skb_push(skb, sizeof(*eth));
5696 skb_reset_mac_header(skb);
5697 skb_reset_mac_len(skb);
5700 ether_addr_copy(eth->h_dest, dst);
5701 ether_addr_copy(eth->h_source, src);
5702 eth->h_proto = skb->protocol;
5704 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5708 EXPORT_SYMBOL(skb_eth_push);
5710 /* Update the ethertype of hdr and the skb csum value if required. */
5711 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5714 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5715 __be16 diff[] = { ~hdr->h_proto, ethertype };
5717 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5720 hdr->h_proto = ethertype;
5724 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5728 * @mpls_lse: MPLS label stack entry to push
5729 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5730 * @mac_len: length of the MAC header
5731 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5734 * Expects skb->data at mac header.
5736 * Returns 0 on success, -errno otherwise.
5738 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5739 int mac_len, bool ethernet)
5741 struct mpls_shim_hdr *lse;
5744 if (unlikely(!eth_p_mpls(mpls_proto)))
5747 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5748 if (skb->encapsulation)
5751 err = skb_cow_head(skb, MPLS_HLEN);
5755 if (!skb->inner_protocol) {
5756 skb_set_inner_network_header(skb, skb_network_offset(skb));
5757 skb_set_inner_protocol(skb, skb->protocol);
5760 skb_push(skb, MPLS_HLEN);
5761 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5763 skb_reset_mac_header(skb);
5764 skb_set_network_header(skb, mac_len);
5765 skb_reset_mac_len(skb);
5767 lse = mpls_hdr(skb);
5768 lse->label_stack_entry = mpls_lse;
5769 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5771 if (ethernet && mac_len >= ETH_HLEN)
5772 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5773 skb->protocol = mpls_proto;
5777 EXPORT_SYMBOL_GPL(skb_mpls_push);
5780 * skb_mpls_pop() - pop the outermost MPLS header
5783 * @next_proto: ethertype of header after popped MPLS header
5784 * @mac_len: length of the MAC header
5785 * @ethernet: flag to indicate if the packet is ethernet
5787 * Expects skb->data at mac header.
5789 * Returns 0 on success, -errno otherwise.
5791 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5796 if (unlikely(!eth_p_mpls(skb->protocol)))
5799 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5803 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5804 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5807 __skb_pull(skb, MPLS_HLEN);
5808 skb_reset_mac_header(skb);
5809 skb_set_network_header(skb, mac_len);
5811 if (ethernet && mac_len >= ETH_HLEN) {
5814 /* use mpls_hdr() to get ethertype to account for VLANs. */
5815 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5816 skb_mod_eth_type(skb, hdr, next_proto);
5818 skb->protocol = next_proto;
5822 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5825 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5828 * @mpls_lse: new MPLS label stack entry to update to
5830 * Expects skb->data at mac header.
5832 * Returns 0 on success, -errno otherwise.
5834 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5838 if (unlikely(!eth_p_mpls(skb->protocol)))
5841 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5845 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5846 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5848 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5851 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5855 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5858 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5862 * Expects skb->data at mac header.
5864 * Returns 0 on success, -errno otherwise.
5866 int skb_mpls_dec_ttl(struct sk_buff *skb)
5871 if (unlikely(!eth_p_mpls(skb->protocol)))
5874 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5877 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5878 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5882 lse &= ~MPLS_LS_TTL_MASK;
5883 lse |= ttl << MPLS_LS_TTL_SHIFT;
5885 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5887 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5890 * alloc_skb_with_frags - allocate skb with page frags
5892 * @header_len: size of linear part
5893 * @data_len: needed length in frags
5894 * @max_page_order: max page order desired.
5895 * @errcode: pointer to error code if any
5896 * @gfp_mask: allocation mask
5898 * This can be used to allocate a paged skb, given a maximal order for frags.
5900 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5901 unsigned long data_len,
5906 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5907 unsigned long chunk;
5908 struct sk_buff *skb;
5912 *errcode = -EMSGSIZE;
5913 /* Note this test could be relaxed, if we succeed to allocate
5914 * high order pages...
5916 if (npages > MAX_SKB_FRAGS)
5919 *errcode = -ENOBUFS;
5920 skb = alloc_skb(header_len, gfp_mask);
5924 skb->truesize += npages << PAGE_SHIFT;
5926 for (i = 0; npages > 0; i++) {
5927 int order = max_page_order;
5930 if (npages >= 1 << order) {
5931 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5937 /* Do not retry other high order allocations */
5943 page = alloc_page(gfp_mask);
5947 chunk = min_t(unsigned long, data_len,
5948 PAGE_SIZE << order);
5949 skb_fill_page_desc(skb, i, page, 0, chunk);
5951 npages -= 1 << order;
5959 EXPORT_SYMBOL(alloc_skb_with_frags);
5961 /* carve out the first off bytes from skb when off < headlen */
5962 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5963 const int headlen, gfp_t gfp_mask)
5966 int size = skb_end_offset(skb);
5967 int new_hlen = headlen - off;
5970 size = SKB_DATA_ALIGN(size);
5972 if (skb_pfmemalloc(skb))
5973 gfp_mask |= __GFP_MEMALLOC;
5974 data = kmalloc_reserve(size +
5975 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5976 gfp_mask, NUMA_NO_NODE, NULL);
5980 size = SKB_WITH_OVERHEAD(ksize(data));
5982 /* Copy real data, and all frags */
5983 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5986 memcpy((struct skb_shared_info *)(data + size),
5988 offsetof(struct skb_shared_info,
5989 frags[skb_shinfo(skb)->nr_frags]));
5990 if (skb_cloned(skb)) {
5991 /* drop the old head gracefully */
5992 if (skb_orphan_frags(skb, gfp_mask)) {
5996 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5997 skb_frag_ref(skb, i);
5998 if (skb_has_frag_list(skb))
5999 skb_clone_fraglist(skb);
6000 skb_release_data(skb);
6002 /* we can reuse existing recount- all we did was
6011 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6014 skb->end = skb->head + size;
6016 skb_set_tail_pointer(skb, skb_headlen(skb));
6017 skb_headers_offset_update(skb, 0);
6021 atomic_set(&skb_shinfo(skb)->dataref, 1);
6026 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6028 /* carve out the first eat bytes from skb's frag_list. May recurse into
6031 static int pskb_carve_frag_list(struct sk_buff *skb,
6032 struct skb_shared_info *shinfo, int eat,
6035 struct sk_buff *list = shinfo->frag_list;
6036 struct sk_buff *clone = NULL;
6037 struct sk_buff *insp = NULL;
6041 pr_err("Not enough bytes to eat. Want %d\n", eat);
6044 if (list->len <= eat) {
6045 /* Eaten as whole. */
6050 /* Eaten partially. */
6051 if (skb_shared(list)) {
6052 clone = skb_clone(list, gfp_mask);
6058 /* This may be pulled without problems. */
6061 if (pskb_carve(list, eat, gfp_mask) < 0) {
6069 /* Free pulled out fragments. */
6070 while ((list = shinfo->frag_list) != insp) {
6071 shinfo->frag_list = list->next;
6074 /* And insert new clone at head. */
6077 shinfo->frag_list = clone;
6082 /* carve off first len bytes from skb. Split line (off) is in the
6083 * non-linear part of skb
6085 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6086 int pos, gfp_t gfp_mask)
6089 int size = skb_end_offset(skb);
6091 const int nfrags = skb_shinfo(skb)->nr_frags;
6092 struct skb_shared_info *shinfo;
6094 size = SKB_DATA_ALIGN(size);
6096 if (skb_pfmemalloc(skb))
6097 gfp_mask |= __GFP_MEMALLOC;
6098 data = kmalloc_reserve(size +
6099 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6100 gfp_mask, NUMA_NO_NODE, NULL);
6104 size = SKB_WITH_OVERHEAD(ksize(data));
6106 memcpy((struct skb_shared_info *)(data + size),
6107 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6108 if (skb_orphan_frags(skb, gfp_mask)) {
6112 shinfo = (struct skb_shared_info *)(data + size);
6113 for (i = 0; i < nfrags; i++) {
6114 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6116 if (pos + fsize > off) {
6117 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6121 * We have two variants in this case:
6122 * 1. Move all the frag to the second
6123 * part, if it is possible. F.e.
6124 * this approach is mandatory for TUX,
6125 * where splitting is expensive.
6126 * 2. Split is accurately. We make this.
6128 skb_frag_off_add(&shinfo->frags[0], off - pos);
6129 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6131 skb_frag_ref(skb, i);
6136 shinfo->nr_frags = k;
6137 if (skb_has_frag_list(skb))
6138 skb_clone_fraglist(skb);
6140 /* split line is in frag list */
6141 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6142 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6143 if (skb_has_frag_list(skb))
6144 kfree_skb_list(skb_shinfo(skb)->frag_list);
6148 skb_release_data(skb);
6153 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6156 skb->end = skb->head + size;
6158 skb_reset_tail_pointer(skb);
6159 skb_headers_offset_update(skb, 0);
6164 skb->data_len = skb->len;
6165 atomic_set(&skb_shinfo(skb)->dataref, 1);
6169 /* remove len bytes from the beginning of the skb */
6170 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6172 int headlen = skb_headlen(skb);
6175 return pskb_carve_inside_header(skb, len, headlen, gfp);
6177 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6180 /* Extract to_copy bytes starting at off from skb, and return this in
6183 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6184 int to_copy, gfp_t gfp)
6186 struct sk_buff *clone = skb_clone(skb, gfp);
6191 if (pskb_carve(clone, off, gfp) < 0 ||
6192 pskb_trim(clone, to_copy)) {
6198 EXPORT_SYMBOL(pskb_extract);
6201 * skb_condense - try to get rid of fragments/frag_list if possible
6204 * Can be used to save memory before skb is added to a busy queue.
6205 * If packet has bytes in frags and enough tail room in skb->head,
6206 * pull all of them, so that we can free the frags right now and adjust
6209 * We do not reallocate skb->head thus can not fail.
6210 * Caller must re-evaluate skb->truesize if needed.
6212 void skb_condense(struct sk_buff *skb)
6214 if (skb->data_len) {
6215 if (skb->data_len > skb->end - skb->tail ||
6219 /* Nice, we can free page frag(s) right now */
6220 __pskb_pull_tail(skb, skb->data_len);
6222 /* At this point, skb->truesize might be over estimated,
6223 * because skb had a fragment, and fragments do not tell
6225 * When we pulled its content into skb->head, fragment
6226 * was freed, but __pskb_pull_tail() could not possibly
6227 * adjust skb->truesize, not knowing the frag truesize.
6229 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6232 #ifdef CONFIG_SKB_EXTENSIONS
6233 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6235 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6239 * __skb_ext_alloc - allocate a new skb extensions storage
6241 * @flags: See kmalloc().
6243 * Returns the newly allocated pointer. The pointer can later attached to a
6244 * skb via __skb_ext_set().
6245 * Note: caller must handle the skb_ext as an opaque data.
6247 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6249 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6252 memset(new->offset, 0, sizeof(new->offset));
6253 refcount_set(&new->refcnt, 1);
6259 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6260 unsigned int old_active)
6262 struct skb_ext *new;
6264 if (refcount_read(&old->refcnt) == 1)
6267 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6271 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6272 refcount_set(&new->refcnt, 1);
6275 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6276 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6279 for (i = 0; i < sp->len; i++)
6280 xfrm_state_hold(sp->xvec[i]);
6288 * __skb_ext_set - attach the specified extension storage to this skb
6291 * @ext: extension storage previously allocated via __skb_ext_alloc()
6293 * Existing extensions, if any, are cleared.
6295 * Returns the pointer to the extension.
6297 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6298 struct skb_ext *ext)
6300 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6303 newlen = newoff + skb_ext_type_len[id];
6304 ext->chunks = newlen;
6305 ext->offset[id] = newoff;
6306 skb->extensions = ext;
6307 skb->active_extensions = 1 << id;
6308 return skb_ext_get_ptr(ext, id);
6312 * skb_ext_add - allocate space for given extension, COW if needed
6314 * @id: extension to allocate space for
6316 * Allocates enough space for the given extension.
6317 * If the extension is already present, a pointer to that extension
6320 * If the skb was cloned, COW applies and the returned memory can be
6321 * modified without changing the extension space of clones buffers.
6323 * Returns pointer to the extension or NULL on allocation failure.
6325 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6327 struct skb_ext *new, *old = NULL;
6328 unsigned int newlen, newoff;
6330 if (skb->active_extensions) {
6331 old = skb->extensions;
6333 new = skb_ext_maybe_cow(old, skb->active_extensions);
6337 if (__skb_ext_exist(new, id))
6340 newoff = new->chunks;
6342 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6344 new = __skb_ext_alloc(GFP_ATOMIC);
6349 newlen = newoff + skb_ext_type_len[id];
6350 new->chunks = newlen;
6351 new->offset[id] = newoff;
6353 skb->extensions = new;
6354 skb->active_extensions |= 1 << id;
6355 return skb_ext_get_ptr(new, id);
6357 EXPORT_SYMBOL(skb_ext_add);
6360 static void skb_ext_put_sp(struct sec_path *sp)
6364 for (i = 0; i < sp->len; i++)
6365 xfrm_state_put(sp->xvec[i]);
6369 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6371 struct skb_ext *ext = skb->extensions;
6373 skb->active_extensions &= ~(1 << id);
6374 if (skb->active_extensions == 0) {
6375 skb->extensions = NULL;
6378 } else if (id == SKB_EXT_SEC_PATH &&
6379 refcount_read(&ext->refcnt) == 1) {
6380 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6387 EXPORT_SYMBOL(__skb_ext_del);
6389 void __skb_ext_put(struct skb_ext *ext)
6391 /* If this is last clone, nothing can increment
6392 * it after check passes. Avoids one atomic op.
6394 if (refcount_read(&ext->refcnt) == 1)
6397 if (!refcount_dec_and_test(&ext->refcnt))
6401 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6402 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6405 kmem_cache_free(skbuff_ext_cache, ext);
6407 EXPORT_SYMBOL(__skb_ext_put);
6408 #endif /* CONFIG_SKB_EXTENSIONS */