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 <net/page_pool.h>
76 #include <linux/uaccess.h>
77 #include <trace/events/skb.h>
78 #include <linux/highmem.h>
79 #include <linux/capability.h>
80 #include <linux/user_namespace.h>
81 #include <linux/indirect_call_wrapper.h>
84 #include "sock_destructor.h"
86 struct kmem_cache *skbuff_head_cache __ro_after_init;
87 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
88 #ifdef CONFIG_SKB_EXTENSIONS
89 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
91 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
92 EXPORT_SYMBOL(sysctl_max_skb_frags);
95 #define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
96 const char * const drop_reasons[] = {
97 [SKB_CONSUMED] = "CONSUMED",
98 DEFINE_DROP_REASON(FN, FN)
100 EXPORT_SYMBOL(drop_reasons);
103 * skb_panic - private function for out-of-line support
107 * @msg: skb_over_panic or skb_under_panic
109 * Out-of-line support for skb_put() and skb_push().
110 * Called via the wrapper skb_over_panic() or skb_under_panic().
111 * Keep out of line to prevent kernel bloat.
112 * __builtin_return_address is not used because it is not always reliable.
114 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
117 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
118 msg, addr, skb->len, sz, skb->head, skb->data,
119 (unsigned long)skb->tail, (unsigned long)skb->end,
120 skb->dev ? skb->dev->name : "<NULL>");
124 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
126 skb_panic(skb, sz, addr, __func__);
129 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
131 skb_panic(skb, sz, addr, __func__);
134 #define NAPI_SKB_CACHE_SIZE 64
135 #define NAPI_SKB_CACHE_BULK 16
136 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
138 #if PAGE_SIZE == SZ_4K
140 #define NAPI_HAS_SMALL_PAGE_FRAG 1
141 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc)
143 /* specialized page frag allocator using a single order 0 page
144 * and slicing it into 1K sized fragment. Constrained to systems
145 * with a very limited amount of 1K fragments fitting a single
146 * page - to avoid excessive truesize underestimation
149 struct page_frag_1k {
155 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
160 offset = nc->offset - SZ_1K;
161 if (likely(offset >= 0))
164 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
168 nc->va = page_address(page);
169 nc->pfmemalloc = page_is_pfmemalloc(page);
170 offset = PAGE_SIZE - SZ_1K;
171 page_ref_add(page, offset / SZ_1K);
175 return nc->va + offset;
179 /* the small page is actually unused in this build; add dummy helpers
180 * to please the compiler and avoid later preprocessor's conditionals
182 #define NAPI_HAS_SMALL_PAGE_FRAG 0
183 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false
185 struct page_frag_1k {
188 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
195 struct napi_alloc_cache {
196 struct page_frag_cache page;
197 struct page_frag_1k page_small;
198 unsigned int skb_count;
199 void *skb_cache[NAPI_SKB_CACHE_SIZE];
202 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
203 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
205 /* Double check that napi_get_frags() allocates skbs with
206 * skb->head being backed by slab, not a page fragment.
207 * This is to make sure bug fixed in 3226b158e67c
208 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
209 * does not accidentally come back.
211 void napi_get_frags_check(struct napi_struct *napi)
216 skb = napi_get_frags(napi);
217 WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
218 napi_free_frags(napi);
222 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
224 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
226 fragsz = SKB_DATA_ALIGN(fragsz);
228 return page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
230 EXPORT_SYMBOL(__napi_alloc_frag_align);
232 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
236 fragsz = SKB_DATA_ALIGN(fragsz);
237 if (in_hardirq() || irqs_disabled()) {
238 struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
240 data = page_frag_alloc_align(nc, fragsz, GFP_ATOMIC, align_mask);
242 struct napi_alloc_cache *nc;
245 nc = this_cpu_ptr(&napi_alloc_cache);
246 data = page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC, align_mask);
251 EXPORT_SYMBOL(__netdev_alloc_frag_align);
253 static struct sk_buff *napi_skb_cache_get(void)
255 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
258 if (unlikely(!nc->skb_count)) {
259 nc->skb_count = kmem_cache_alloc_bulk(skbuff_head_cache,
263 if (unlikely(!nc->skb_count))
267 skb = nc->skb_cache[--nc->skb_count];
268 kasan_unpoison_object_data(skbuff_head_cache, skb);
273 static inline void __finalize_skb_around(struct sk_buff *skb, void *data,
276 struct skb_shared_info *shinfo;
278 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
280 /* Assumes caller memset cleared SKB */
281 skb->truesize = SKB_TRUESIZE(size);
282 refcount_set(&skb->users, 1);
285 skb_reset_tail_pointer(skb);
286 skb_set_end_offset(skb, size);
287 skb->mac_header = (typeof(skb->mac_header))~0U;
288 skb->transport_header = (typeof(skb->transport_header))~0U;
289 skb->alloc_cpu = raw_smp_processor_id();
290 /* make sure we initialize shinfo sequentially */
291 shinfo = skb_shinfo(skb);
292 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
293 atomic_set(&shinfo->dataref, 1);
295 skb_set_kcov_handle(skb, kcov_common_handle());
298 static inline void *__slab_build_skb(struct sk_buff *skb, void *data,
303 /* Must find the allocation size (and grow it to match). */
305 /* krealloc() will immediately return "data" when
306 * "ksize(data)" is requested: it is the existing upper
307 * bounds. As a result, GFP_ATOMIC will be ignored. Note
308 * that this "new" pointer needs to be passed back to the
309 * caller for use so the __alloc_size hinting will be
312 resized = krealloc(data, *size, GFP_ATOMIC);
313 WARN_ON_ONCE(resized != data);
317 /* build_skb() variant which can operate on slab buffers.
318 * Note that this should be used sparingly as slab buffers
319 * cannot be combined efficiently by GRO!
321 struct sk_buff *slab_build_skb(void *data)
326 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
330 memset(skb, 0, offsetof(struct sk_buff, tail));
331 data = __slab_build_skb(skb, data, &size);
332 __finalize_skb_around(skb, data, size);
336 EXPORT_SYMBOL(slab_build_skb);
338 /* Caller must provide SKB that is memset cleared */
339 static void __build_skb_around(struct sk_buff *skb, void *data,
340 unsigned int frag_size)
342 unsigned int size = frag_size;
344 /* frag_size == 0 is considered deprecated now. Callers
345 * using slab buffer should use slab_build_skb() instead.
347 if (WARN_ONCE(size == 0, "Use slab_build_skb() instead"))
348 data = __slab_build_skb(skb, data, &size);
350 __finalize_skb_around(skb, data, size);
354 * __build_skb - build a network buffer
355 * @data: data buffer provided by caller
356 * @frag_size: size of data (must not be 0)
358 * Allocate a new &sk_buff. Caller provides space holding head and
359 * skb_shared_info. @data must have been allocated from the page
360 * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc()
361 * allocation is deprecated, and callers should use slab_build_skb()
363 * The return is the new skb buffer.
364 * On a failure the return is %NULL, and @data is not freed.
366 * Before IO, driver allocates only data buffer where NIC put incoming frame
367 * Driver should add room at head (NET_SKB_PAD) and
368 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
369 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
370 * before giving packet to stack.
371 * RX rings only contains data buffers, not full skbs.
373 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
377 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
381 memset(skb, 0, offsetof(struct sk_buff, tail));
382 __build_skb_around(skb, data, frag_size);
387 /* build_skb() is wrapper over __build_skb(), that specifically
388 * takes care of skb->head and skb->pfmemalloc
389 * This means that if @frag_size is not zero, then @data must be backed
390 * by a page fragment, not kmalloc() or vmalloc()
392 struct sk_buff *build_skb(void *data, unsigned int frag_size)
394 struct sk_buff *skb = __build_skb(data, frag_size);
396 if (skb && frag_size) {
398 if (page_is_pfmemalloc(virt_to_head_page(data)))
403 EXPORT_SYMBOL(build_skb);
406 * build_skb_around - build a network buffer around provided skb
407 * @skb: sk_buff provide by caller, must be memset cleared
408 * @data: data buffer provided by caller
409 * @frag_size: size of data, or 0 if head was kmalloced
411 struct sk_buff *build_skb_around(struct sk_buff *skb,
412 void *data, unsigned int frag_size)
417 __build_skb_around(skb, data, frag_size);
421 if (page_is_pfmemalloc(virt_to_head_page(data)))
426 EXPORT_SYMBOL(build_skb_around);
429 * __napi_build_skb - build a network buffer
430 * @data: data buffer provided by caller
431 * @frag_size: size of data, or 0 if head was kmalloced
433 * Version of __build_skb() that uses NAPI percpu caches to obtain
434 * skbuff_head instead of inplace allocation.
436 * Returns a new &sk_buff on success, %NULL on allocation failure.
438 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
442 skb = napi_skb_cache_get();
446 memset(skb, 0, offsetof(struct sk_buff, tail));
447 __build_skb_around(skb, data, frag_size);
453 * napi_build_skb - build a network buffer
454 * @data: data buffer provided by caller
455 * @frag_size: size of data, or 0 if head was kmalloced
457 * Version of __napi_build_skb() that takes care of skb->head_frag
458 * and skb->pfmemalloc when the data is a page or page fragment.
460 * Returns a new &sk_buff on success, %NULL on allocation failure.
462 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
464 struct sk_buff *skb = __napi_build_skb(data, frag_size);
466 if (likely(skb) && frag_size) {
468 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
473 EXPORT_SYMBOL(napi_build_skb);
476 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
477 * the caller if emergency pfmemalloc reserves are being used. If it is and
478 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
479 * may be used. Otherwise, the packet data may be discarded until enough
482 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
486 bool ret_pfmemalloc = false;
489 * Try a regular allocation, when that fails and we're not entitled
490 * to the reserves, fail.
492 obj = kmalloc_node_track_caller(size,
493 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
495 if (obj || !(gfp_pfmemalloc_allowed(flags)))
498 /* Try again but now we are using pfmemalloc reserves */
499 ret_pfmemalloc = true;
500 obj = kmalloc_node_track_caller(size, flags, node);
504 *pfmemalloc = ret_pfmemalloc;
509 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
510 * 'private' fields and also do memory statistics to find all the
516 * __alloc_skb - allocate a network buffer
517 * @size: size to allocate
518 * @gfp_mask: allocation mask
519 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
520 * instead of head cache and allocate a cloned (child) skb.
521 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
522 * allocations in case the data is required for writeback
523 * @node: numa node to allocate memory on
525 * Allocate a new &sk_buff. The returned buffer has no headroom and a
526 * tail room of at least size bytes. The object has a reference count
527 * of one. The return is the buffer. On a failure the return is %NULL.
529 * Buffers may only be allocated from interrupts using a @gfp_mask of
532 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
535 struct kmem_cache *cache;
541 cache = (flags & SKB_ALLOC_FCLONE)
542 ? skbuff_fclone_cache : skbuff_head_cache;
544 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
545 gfp_mask |= __GFP_MEMALLOC;
548 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
549 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
550 skb = napi_skb_cache_get();
552 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
557 /* We do our best to align skb_shared_info on a separate cache
558 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
559 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
560 * Both skb->head and skb_shared_info are cache line aligned.
562 size = SKB_DATA_ALIGN(size);
563 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
564 osize = kmalloc_size_roundup(size);
565 data = kmalloc_reserve(osize, gfp_mask, node, &pfmemalloc);
568 /* kmalloc_size_roundup() might give us more room than requested.
569 * Put skb_shared_info exactly at the end of allocated zone,
570 * to allow max possible filling before reallocation.
572 size = SKB_WITH_OVERHEAD(osize);
573 prefetchw(data + size);
576 * Only clear those fields we need to clear, not those that we will
577 * actually initialise below. Hence, don't put any more fields after
578 * the tail pointer in struct sk_buff!
580 memset(skb, 0, offsetof(struct sk_buff, tail));
581 __build_skb_around(skb, data, osize);
582 skb->pfmemalloc = pfmemalloc;
584 if (flags & SKB_ALLOC_FCLONE) {
585 struct sk_buff_fclones *fclones;
587 fclones = container_of(skb, struct sk_buff_fclones, skb1);
589 skb->fclone = SKB_FCLONE_ORIG;
590 refcount_set(&fclones->fclone_ref, 1);
596 kmem_cache_free(cache, skb);
599 EXPORT_SYMBOL(__alloc_skb);
602 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
603 * @dev: network device to receive on
604 * @len: length to allocate
605 * @gfp_mask: get_free_pages mask, passed to alloc_skb
607 * Allocate a new &sk_buff and assign it a usage count of one. The
608 * buffer has NET_SKB_PAD headroom built in. Users should allocate
609 * the headroom they think they need without accounting for the
610 * built in space. The built in space is used for optimisations.
612 * %NULL is returned if there is no free memory.
614 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
617 struct page_frag_cache *nc;
624 /* If requested length is either too small or too big,
625 * we use kmalloc() for skb->head allocation.
627 if (len <= SKB_WITH_OVERHEAD(1024) ||
628 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
629 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
630 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
636 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
637 len = SKB_DATA_ALIGN(len);
639 if (sk_memalloc_socks())
640 gfp_mask |= __GFP_MEMALLOC;
642 if (in_hardirq() || irqs_disabled()) {
643 nc = this_cpu_ptr(&netdev_alloc_cache);
644 data = page_frag_alloc(nc, len, gfp_mask);
645 pfmemalloc = nc->pfmemalloc;
648 nc = this_cpu_ptr(&napi_alloc_cache.page);
649 data = page_frag_alloc(nc, len, gfp_mask);
650 pfmemalloc = nc->pfmemalloc;
657 skb = __build_skb(data, len);
658 if (unlikely(!skb)) {
668 skb_reserve(skb, NET_SKB_PAD);
674 EXPORT_SYMBOL(__netdev_alloc_skb);
677 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
678 * @napi: napi instance this buffer was allocated for
679 * @len: length to allocate
680 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
682 * Allocate a new sk_buff for use in NAPI receive. This buffer will
683 * attempt to allocate the head from a special reserved region used
684 * only for NAPI Rx allocation. By doing this we can save several
685 * CPU cycles by avoiding having to disable and re-enable IRQs.
687 * %NULL is returned if there is no free memory.
689 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
692 struct napi_alloc_cache *nc;
697 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
698 len += NET_SKB_PAD + NET_IP_ALIGN;
700 /* If requested length is either too small or too big,
701 * we use kmalloc() for skb->head allocation.
702 * When the small frag allocator is available, prefer it over kmalloc
703 * for small fragments
705 if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
706 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
707 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
708 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
715 nc = this_cpu_ptr(&napi_alloc_cache);
717 if (sk_memalloc_socks())
718 gfp_mask |= __GFP_MEMALLOC;
720 if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
721 /* we are artificially inflating the allocation size, but
722 * that is not as bad as it may look like, as:
723 * - 'len' less than GRO_MAX_HEAD makes little sense
724 * - On most systems, larger 'len' values lead to fragment
725 * size above 512 bytes
726 * - kmalloc would use the kmalloc-1k slab for such values
727 * - Builds with smaller GRO_MAX_HEAD will very likely do
728 * little networking, as that implies no WiFi and no
729 * tunnels support, and 32 bits arches.
733 data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
734 pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
736 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
737 len = SKB_DATA_ALIGN(len);
739 data = page_frag_alloc(&nc->page, len, gfp_mask);
740 pfmemalloc = nc->page.pfmemalloc;
746 skb = __napi_build_skb(data, len);
747 if (unlikely(!skb)) {
757 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
758 skb->dev = napi->dev;
763 EXPORT_SYMBOL(__napi_alloc_skb);
765 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
766 int size, unsigned int truesize)
768 skb_fill_page_desc(skb, i, page, off, size);
770 skb->data_len += size;
771 skb->truesize += truesize;
773 EXPORT_SYMBOL(skb_add_rx_frag);
775 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
776 unsigned int truesize)
778 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
780 skb_frag_size_add(frag, size);
782 skb->data_len += size;
783 skb->truesize += truesize;
785 EXPORT_SYMBOL(skb_coalesce_rx_frag);
787 static void skb_drop_list(struct sk_buff **listp)
789 kfree_skb_list(*listp);
793 static inline void skb_drop_fraglist(struct sk_buff *skb)
795 skb_drop_list(&skb_shinfo(skb)->frag_list);
798 static void skb_clone_fraglist(struct sk_buff *skb)
800 struct sk_buff *list;
802 skb_walk_frags(skb, list)
806 static bool skb_pp_recycle(struct sk_buff *skb, void *data)
808 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
810 return page_pool_return_skb_page(virt_to_page(data));
813 static void skb_free_head(struct sk_buff *skb)
815 unsigned char *head = skb->head;
817 if (skb->head_frag) {
818 if (skb_pp_recycle(skb, head))
826 static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
828 struct skb_shared_info *shinfo = skb_shinfo(skb);
832 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
836 if (skb_zcopy(skb)) {
837 bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
839 skb_zcopy_clear(skb, true);
844 for (i = 0; i < shinfo->nr_frags; i++)
845 __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
848 if (shinfo->frag_list)
849 kfree_skb_list_reason(shinfo->frag_list, reason);
853 /* When we clone an SKB we copy the reycling bit. The pp_recycle
854 * bit is only set on the head though, so in order to avoid races
855 * while trying to recycle fragments on __skb_frag_unref() we need
856 * to make one SKB responsible for triggering the recycle path.
857 * So disable the recycling bit if an SKB is cloned and we have
858 * additional references to the fragmented part of the SKB.
859 * Eventually the last SKB will have the recycling bit set and it's
860 * dataref set to 0, which will trigger the recycling
866 * Free an skbuff by memory without cleaning the state.
868 static void kfree_skbmem(struct sk_buff *skb)
870 struct sk_buff_fclones *fclones;
872 switch (skb->fclone) {
873 case SKB_FCLONE_UNAVAILABLE:
874 kmem_cache_free(skbuff_head_cache, skb);
877 case SKB_FCLONE_ORIG:
878 fclones = container_of(skb, struct sk_buff_fclones, skb1);
880 /* We usually free the clone (TX completion) before original skb
881 * This test would have no chance to be true for the clone,
882 * while here, branch prediction will be good.
884 if (refcount_read(&fclones->fclone_ref) == 1)
888 default: /* SKB_FCLONE_CLONE */
889 fclones = container_of(skb, struct sk_buff_fclones, skb2);
892 if (!refcount_dec_and_test(&fclones->fclone_ref))
895 kmem_cache_free(skbuff_fclone_cache, fclones);
898 void skb_release_head_state(struct sk_buff *skb)
901 if (skb->destructor) {
902 DEBUG_NET_WARN_ON_ONCE(in_hardirq());
903 skb->destructor(skb);
905 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
906 nf_conntrack_put(skb_nfct(skb));
911 /* Free everything but the sk_buff shell. */
912 static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
914 skb_release_head_state(skb);
915 if (likely(skb->head))
916 skb_release_data(skb, reason);
920 * __kfree_skb - private function
923 * Free an sk_buff. Release anything attached to the buffer.
924 * Clean the state. This is an internal helper function. Users should
925 * always call kfree_skb
928 void __kfree_skb(struct sk_buff *skb)
930 skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
933 EXPORT_SYMBOL(__kfree_skb);
936 * kfree_skb_reason - free an sk_buff with special reason
937 * @skb: buffer to free
938 * @reason: reason why this skb is dropped
940 * Drop a reference to the buffer and free it if the usage count has
941 * hit zero. Meanwhile, pass the drop reason to 'kfree_skb'
945 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
947 if (unlikely(!skb_unref(skb)))
950 DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX);
952 if (reason == SKB_CONSUMED)
953 trace_consume_skb(skb);
955 trace_kfree_skb(skb, __builtin_return_address(0), reason);
958 EXPORT_SYMBOL(kfree_skb_reason);
960 void kfree_skb_list_reason(struct sk_buff *segs,
961 enum skb_drop_reason reason)
964 struct sk_buff *next = segs->next;
966 kfree_skb_reason(segs, reason);
970 EXPORT_SYMBOL(kfree_skb_list_reason);
972 /* Dump skb information and contents.
974 * Must only be called from net_ratelimit()-ed paths.
976 * Dumps whole packets if full_pkt, only headers otherwise.
978 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
980 struct skb_shared_info *sh = skb_shinfo(skb);
981 struct net_device *dev = skb->dev;
982 struct sock *sk = skb->sk;
983 struct sk_buff *list_skb;
984 bool has_mac, has_trans;
985 int headroom, tailroom;
991 len = min_t(int, skb->len, MAX_HEADER + 128);
993 headroom = skb_headroom(skb);
994 tailroom = skb_tailroom(skb);
996 has_mac = skb_mac_header_was_set(skb);
997 has_trans = skb_transport_header_was_set(skb);
999 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1000 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
1001 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1002 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1003 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
1004 level, skb->len, headroom, skb_headlen(skb), tailroom,
1005 has_mac ? skb->mac_header : -1,
1006 has_mac ? skb_mac_header_len(skb) : -1,
1007 skb->network_header,
1008 has_trans ? skb_network_header_len(skb) : -1,
1009 has_trans ? skb->transport_header : -1,
1010 sh->tx_flags, sh->nr_frags,
1011 sh->gso_size, sh->gso_type, sh->gso_segs,
1012 skb->csum, skb->ip_summed, skb->csum_complete_sw,
1013 skb->csum_valid, skb->csum_level,
1014 skb->hash, skb->sw_hash, skb->l4_hash,
1015 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
1018 printk("%sdev name=%s feat=%pNF\n",
1019 level, dev->name, &dev->features);
1021 printk("%ssk family=%hu type=%u proto=%u\n",
1022 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1024 if (full_pkt && headroom)
1025 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1026 16, 1, skb->head, headroom, false);
1028 seg_len = min_t(int, skb_headlen(skb), len);
1030 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
1031 16, 1, skb->data, seg_len, false);
1034 if (full_pkt && tailroom)
1035 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1036 16, 1, skb_tail_pointer(skb), tailroom, false);
1038 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1039 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1040 u32 p_off, p_len, copied;
1044 skb_frag_foreach_page(frag, skb_frag_off(frag),
1045 skb_frag_size(frag), p, p_off, p_len,
1047 seg_len = min_t(int, p_len, len);
1048 vaddr = kmap_atomic(p);
1049 print_hex_dump(level, "skb frag: ",
1051 16, 1, vaddr + p_off, seg_len, false);
1052 kunmap_atomic(vaddr);
1059 if (full_pkt && skb_has_frag_list(skb)) {
1060 printk("skb fraglist:\n");
1061 skb_walk_frags(skb, list_skb)
1062 skb_dump(level, list_skb, true);
1065 EXPORT_SYMBOL(skb_dump);
1068 * skb_tx_error - report an sk_buff xmit error
1069 * @skb: buffer that triggered an error
1071 * Report xmit error if a device callback is tracking this skb.
1072 * skb must be freed afterwards.
1074 void skb_tx_error(struct sk_buff *skb)
1077 skb_zcopy_downgrade_managed(skb);
1078 skb_zcopy_clear(skb, true);
1081 EXPORT_SYMBOL(skb_tx_error);
1083 #ifdef CONFIG_TRACEPOINTS
1085 * consume_skb - free an skbuff
1086 * @skb: buffer to free
1088 * Drop a ref to the buffer and free it if the usage count has hit zero
1089 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
1090 * is being dropped after a failure and notes that
1092 void consume_skb(struct sk_buff *skb)
1094 if (!skb_unref(skb))
1097 trace_consume_skb(skb);
1100 EXPORT_SYMBOL(consume_skb);
1104 * __consume_stateless_skb - free an skbuff, assuming it is stateless
1105 * @skb: buffer to free
1107 * Alike consume_skb(), but this variant assumes that this is the last
1108 * skb reference and all the head states have been already dropped
1110 void __consume_stateless_skb(struct sk_buff *skb)
1112 trace_consume_skb(skb);
1113 skb_release_data(skb, SKB_CONSUMED);
1117 static void napi_skb_cache_put(struct sk_buff *skb)
1119 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1122 kasan_poison_object_data(skbuff_head_cache, skb);
1123 nc->skb_cache[nc->skb_count++] = skb;
1125 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1126 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1127 kasan_unpoison_object_data(skbuff_head_cache,
1130 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
1131 nc->skb_cache + NAPI_SKB_CACHE_HALF);
1132 nc->skb_count = NAPI_SKB_CACHE_HALF;
1136 void __kfree_skb_defer(struct sk_buff *skb)
1138 skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1139 napi_skb_cache_put(skb);
1142 void napi_skb_free_stolen_head(struct sk_buff *skb)
1144 if (unlikely(skb->slow_gro)) {
1151 napi_skb_cache_put(skb);
1154 void napi_consume_skb(struct sk_buff *skb, int budget)
1156 /* Zero budget indicate non-NAPI context called us, like netpoll */
1157 if (unlikely(!budget)) {
1158 dev_consume_skb_any(skb);
1162 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1164 if (!skb_unref(skb))
1167 /* if reaching here SKB is ready to free */
1168 trace_consume_skb(skb);
1170 /* if SKB is a clone, don't handle this case */
1171 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1176 skb_release_all(skb, SKB_CONSUMED);
1177 napi_skb_cache_put(skb);
1179 EXPORT_SYMBOL(napi_consume_skb);
1181 /* Make sure a field is contained by headers group */
1182 #define CHECK_SKB_FIELD(field) \
1183 BUILD_BUG_ON(offsetof(struct sk_buff, field) != \
1184 offsetof(struct sk_buff, headers.field)); \
1186 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1188 new->tstamp = old->tstamp;
1189 /* We do not copy old->sk */
1190 new->dev = old->dev;
1191 memcpy(new->cb, old->cb, sizeof(old->cb));
1192 skb_dst_copy(new, old);
1193 __skb_ext_copy(new, old);
1194 __nf_copy(new, old, false);
1196 /* Note : this field could be in the headers group.
1197 * It is not yet because we do not want to have a 16 bit hole
1199 new->queue_mapping = old->queue_mapping;
1201 memcpy(&new->headers, &old->headers, sizeof(new->headers));
1202 CHECK_SKB_FIELD(protocol);
1203 CHECK_SKB_FIELD(csum);
1204 CHECK_SKB_FIELD(hash);
1205 CHECK_SKB_FIELD(priority);
1206 CHECK_SKB_FIELD(skb_iif);
1207 CHECK_SKB_FIELD(vlan_proto);
1208 CHECK_SKB_FIELD(vlan_tci);
1209 CHECK_SKB_FIELD(transport_header);
1210 CHECK_SKB_FIELD(network_header);
1211 CHECK_SKB_FIELD(mac_header);
1212 CHECK_SKB_FIELD(inner_protocol);
1213 CHECK_SKB_FIELD(inner_transport_header);
1214 CHECK_SKB_FIELD(inner_network_header);
1215 CHECK_SKB_FIELD(inner_mac_header);
1216 CHECK_SKB_FIELD(mark);
1217 #ifdef CONFIG_NETWORK_SECMARK
1218 CHECK_SKB_FIELD(secmark);
1220 #ifdef CONFIG_NET_RX_BUSY_POLL
1221 CHECK_SKB_FIELD(napi_id);
1223 CHECK_SKB_FIELD(alloc_cpu);
1225 CHECK_SKB_FIELD(sender_cpu);
1227 #ifdef CONFIG_NET_SCHED
1228 CHECK_SKB_FIELD(tc_index);
1234 * You should not add any new code to this function. Add it to
1235 * __copy_skb_header above instead.
1237 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1239 #define C(x) n->x = skb->x
1241 n->next = n->prev = NULL;
1243 __copy_skb_header(n, skb);
1248 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1254 n->destructor = NULL;
1261 refcount_set(&n->users, 1);
1263 atomic_inc(&(skb_shinfo(skb)->dataref));
1271 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1272 * @first: first sk_buff of the msg
1274 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1278 n = alloc_skb(0, GFP_ATOMIC);
1282 n->len = first->len;
1283 n->data_len = first->len;
1284 n->truesize = first->truesize;
1286 skb_shinfo(n)->frag_list = first;
1288 __copy_skb_header(n, first);
1289 n->destructor = NULL;
1293 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1296 * skb_morph - morph one skb into another
1297 * @dst: the skb to receive the contents
1298 * @src: the skb to supply the contents
1300 * This is identical to skb_clone except that the target skb is
1301 * supplied by the user.
1303 * The target skb is returned upon exit.
1305 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1307 skb_release_all(dst, SKB_CONSUMED);
1308 return __skb_clone(dst, src);
1310 EXPORT_SYMBOL_GPL(skb_morph);
1312 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1314 unsigned long max_pg, num_pg, new_pg, old_pg;
1315 struct user_struct *user;
1317 if (capable(CAP_IPC_LOCK) || !size)
1320 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1321 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1322 user = mmp->user ? : current_user();
1324 old_pg = atomic_long_read(&user->locked_vm);
1326 new_pg = old_pg + num_pg;
1327 if (new_pg > max_pg)
1329 } while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1332 mmp->user = get_uid(user);
1333 mmp->num_pg = num_pg;
1335 mmp->num_pg += num_pg;
1340 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1342 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1345 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1346 free_uid(mmp->user);
1349 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1351 static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1353 struct ubuf_info_msgzc *uarg;
1354 struct sk_buff *skb;
1356 WARN_ON_ONCE(!in_task());
1358 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1362 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1363 uarg = (void *)skb->cb;
1364 uarg->mmp.user = NULL;
1366 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1371 uarg->ubuf.callback = msg_zerocopy_callback;
1372 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1374 uarg->bytelen = size;
1376 uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1377 refcount_set(&uarg->ubuf.refcnt, 1);
1383 static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1385 return container_of((void *)uarg, struct sk_buff, cb);
1388 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1389 struct ubuf_info *uarg)
1392 struct ubuf_info_msgzc *uarg_zc;
1393 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1396 /* there might be non MSG_ZEROCOPY users */
1397 if (uarg->callback != msg_zerocopy_callback)
1400 /* realloc only when socket is locked (TCP, UDP cork),
1401 * so uarg->len and sk_zckey access is serialized
1403 if (!sock_owned_by_user(sk)) {
1408 uarg_zc = uarg_to_msgzc(uarg);
1409 bytelen = uarg_zc->bytelen + size;
1410 if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1411 /* TCP can create new skb to attach new uarg */
1412 if (sk->sk_type == SOCK_STREAM)
1417 next = (u32)atomic_read(&sk->sk_zckey);
1418 if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1419 if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1422 uarg_zc->bytelen = bytelen;
1423 atomic_set(&sk->sk_zckey, ++next);
1425 /* no extra ref when appending to datagram (MSG_MORE) */
1426 if (sk->sk_type == SOCK_STREAM)
1427 net_zcopy_get(uarg);
1434 return msg_zerocopy_alloc(sk, size);
1436 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1438 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1440 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1444 old_lo = serr->ee.ee_info;
1445 old_hi = serr->ee.ee_data;
1446 sum_len = old_hi - old_lo + 1ULL + len;
1448 if (sum_len >= (1ULL << 32))
1451 if (lo != old_hi + 1)
1454 serr->ee.ee_data += len;
1458 static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1460 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1461 struct sock_exterr_skb *serr;
1462 struct sock *sk = skb->sk;
1463 struct sk_buff_head *q;
1464 unsigned long flags;
1469 mm_unaccount_pinned_pages(&uarg->mmp);
1471 /* if !len, there was only 1 call, and it was aborted
1472 * so do not queue a completion notification
1474 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1479 hi = uarg->id + len - 1;
1480 is_zerocopy = uarg->zerocopy;
1482 serr = SKB_EXT_ERR(skb);
1483 memset(serr, 0, sizeof(*serr));
1484 serr->ee.ee_errno = 0;
1485 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1486 serr->ee.ee_data = hi;
1487 serr->ee.ee_info = lo;
1489 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1491 q = &sk->sk_error_queue;
1492 spin_lock_irqsave(&q->lock, flags);
1493 tail = skb_peek_tail(q);
1494 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1495 !skb_zerocopy_notify_extend(tail, lo, len)) {
1496 __skb_queue_tail(q, skb);
1499 spin_unlock_irqrestore(&q->lock, flags);
1501 sk_error_report(sk);
1508 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1511 struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1513 uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1515 if (refcount_dec_and_test(&uarg->refcnt))
1516 __msg_zerocopy_callback(uarg_zc);
1518 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1520 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1522 struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1524 atomic_dec(&sk->sk_zckey);
1525 uarg_to_msgzc(uarg)->len--;
1528 msg_zerocopy_callback(NULL, uarg, true);
1530 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1532 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1533 struct msghdr *msg, int len,
1534 struct ubuf_info *uarg)
1536 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1537 int err, orig_len = skb->len;
1539 /* An skb can only point to one uarg. This edge case happens when
1540 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1542 if (orig_uarg && uarg != orig_uarg)
1545 err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1546 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1547 struct sock *save_sk = skb->sk;
1549 /* Streams do not free skb on error. Reset to prev state. */
1550 iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1552 ___pskb_trim(skb, orig_len);
1557 skb_zcopy_set(skb, uarg, NULL);
1558 return skb->len - orig_len;
1560 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1562 void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1566 skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1567 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1568 skb_frag_ref(skb, i);
1570 EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1572 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1575 if (skb_zcopy(orig)) {
1576 if (skb_zcopy(nskb)) {
1577 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1582 if (skb_uarg(nskb) == skb_uarg(orig))
1584 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1587 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1593 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1594 * @skb: the skb to modify
1595 * @gfp_mask: allocation priority
1597 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1598 * It will copy all frags into kernel and drop the reference
1599 * to userspace pages.
1601 * If this function is called from an interrupt gfp_mask() must be
1604 * Returns 0 on success or a negative error code on failure
1605 * to allocate kernel memory to copy to.
1607 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1609 int num_frags = skb_shinfo(skb)->nr_frags;
1610 struct page *page, *head = NULL;
1614 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1620 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1621 for (i = 0; i < new_frags; i++) {
1622 page = alloc_page(gfp_mask);
1625 struct page *next = (struct page *)page_private(head);
1631 set_page_private(page, (unsigned long)head);
1637 for (i = 0; i < num_frags; i++) {
1638 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1639 u32 p_off, p_len, copied;
1643 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1644 p, p_off, p_len, copied) {
1646 vaddr = kmap_atomic(p);
1648 while (done < p_len) {
1649 if (d_off == PAGE_SIZE) {
1651 page = (struct page *)page_private(page);
1653 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1654 memcpy(page_address(page) + d_off,
1655 vaddr + p_off + done, copy);
1659 kunmap_atomic(vaddr);
1663 /* skb frags release userspace buffers */
1664 for (i = 0; i < num_frags; i++)
1665 skb_frag_unref(skb, i);
1667 /* skb frags point to kernel buffers */
1668 for (i = 0; i < new_frags - 1; i++) {
1669 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1670 head = (struct page *)page_private(head);
1672 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1673 skb_shinfo(skb)->nr_frags = new_frags;
1676 skb_zcopy_clear(skb, false);
1679 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1682 * skb_clone - duplicate an sk_buff
1683 * @skb: buffer to clone
1684 * @gfp_mask: allocation priority
1686 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1687 * copies share the same packet data but not structure. The new
1688 * buffer has a reference count of 1. If the allocation fails the
1689 * function returns %NULL otherwise the new buffer is returned.
1691 * If this function is called from an interrupt gfp_mask() must be
1695 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1697 struct sk_buff_fclones *fclones = container_of(skb,
1698 struct sk_buff_fclones,
1702 if (skb_orphan_frags(skb, gfp_mask))
1705 if (skb->fclone == SKB_FCLONE_ORIG &&
1706 refcount_read(&fclones->fclone_ref) == 1) {
1708 refcount_set(&fclones->fclone_ref, 2);
1709 n->fclone = SKB_FCLONE_CLONE;
1711 if (skb_pfmemalloc(skb))
1712 gfp_mask |= __GFP_MEMALLOC;
1714 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1718 n->fclone = SKB_FCLONE_UNAVAILABLE;
1721 return __skb_clone(n, skb);
1723 EXPORT_SYMBOL(skb_clone);
1725 void skb_headers_offset_update(struct sk_buff *skb, int off)
1727 /* Only adjust this if it actually is csum_start rather than csum */
1728 if (skb->ip_summed == CHECKSUM_PARTIAL)
1729 skb->csum_start += off;
1730 /* {transport,network,mac}_header and tail are relative to skb->head */
1731 skb->transport_header += off;
1732 skb->network_header += off;
1733 if (skb_mac_header_was_set(skb))
1734 skb->mac_header += off;
1735 skb->inner_transport_header += off;
1736 skb->inner_network_header += off;
1737 skb->inner_mac_header += off;
1739 EXPORT_SYMBOL(skb_headers_offset_update);
1741 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1743 __copy_skb_header(new, old);
1745 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1746 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1747 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1749 EXPORT_SYMBOL(skb_copy_header);
1751 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1753 if (skb_pfmemalloc(skb))
1754 return SKB_ALLOC_RX;
1759 * skb_copy - create private copy of an sk_buff
1760 * @skb: buffer to copy
1761 * @gfp_mask: allocation priority
1763 * Make a copy of both an &sk_buff and its data. This is used when the
1764 * caller wishes to modify the data and needs a private copy of the
1765 * data to alter. Returns %NULL on failure or the pointer to the buffer
1766 * on success. The returned buffer has a reference count of 1.
1768 * As by-product this function converts non-linear &sk_buff to linear
1769 * one, so that &sk_buff becomes completely private and caller is allowed
1770 * to modify all the data of returned buffer. This means that this
1771 * function is not recommended for use in circumstances when only
1772 * header is going to be modified. Use pskb_copy() instead.
1775 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1777 int headerlen = skb_headroom(skb);
1778 unsigned int size = skb_end_offset(skb) + skb->data_len;
1779 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1780 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1785 /* Set the data pointer */
1786 skb_reserve(n, headerlen);
1787 /* Set the tail pointer and length */
1788 skb_put(n, skb->len);
1790 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1792 skb_copy_header(n, skb);
1795 EXPORT_SYMBOL(skb_copy);
1798 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1799 * @skb: buffer to copy
1800 * @headroom: headroom of new skb
1801 * @gfp_mask: allocation priority
1802 * @fclone: if true allocate the copy of the skb from the fclone
1803 * cache instead of the head cache; it is recommended to set this
1804 * to true for the cases where the copy will likely be cloned
1806 * Make a copy of both an &sk_buff and part of its data, located
1807 * in header. Fragmented data remain shared. This is used when
1808 * the caller wishes to modify only header of &sk_buff and needs
1809 * private copy of the header to alter. Returns %NULL on failure
1810 * or the pointer to the buffer on success.
1811 * The returned buffer has a reference count of 1.
1814 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1815 gfp_t gfp_mask, bool fclone)
1817 unsigned int size = skb_headlen(skb) + headroom;
1818 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1819 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1824 /* Set the data pointer */
1825 skb_reserve(n, headroom);
1826 /* Set the tail pointer and length */
1827 skb_put(n, skb_headlen(skb));
1828 /* Copy the bytes */
1829 skb_copy_from_linear_data(skb, n->data, n->len);
1831 n->truesize += skb->data_len;
1832 n->data_len = skb->data_len;
1835 if (skb_shinfo(skb)->nr_frags) {
1838 if (skb_orphan_frags(skb, gfp_mask) ||
1839 skb_zerocopy_clone(n, skb, gfp_mask)) {
1844 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1845 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1846 skb_frag_ref(skb, i);
1848 skb_shinfo(n)->nr_frags = i;
1851 if (skb_has_frag_list(skb)) {
1852 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1853 skb_clone_fraglist(n);
1856 skb_copy_header(n, skb);
1860 EXPORT_SYMBOL(__pskb_copy_fclone);
1863 * pskb_expand_head - reallocate header of &sk_buff
1864 * @skb: buffer to reallocate
1865 * @nhead: room to add at head
1866 * @ntail: room to add at tail
1867 * @gfp_mask: allocation priority
1869 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1870 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1871 * reference count of 1. Returns zero in the case of success or error,
1872 * if expansion failed. In the last case, &sk_buff is not changed.
1874 * All the pointers pointing into skb header may change and must be
1875 * reloaded after call to this function.
1878 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1881 unsigned int osize = skb_end_offset(skb);
1882 unsigned int size = osize + nhead + ntail;
1889 BUG_ON(skb_shared(skb));
1891 skb_zcopy_downgrade_managed(skb);
1893 if (skb_pfmemalloc(skb))
1894 gfp_mask |= __GFP_MEMALLOC;
1896 size = SKB_DATA_ALIGN(size);
1897 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1898 size = kmalloc_size_roundup(size);
1899 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
1902 size = SKB_WITH_OVERHEAD(size);
1904 /* Copy only real data... and, alas, header. This should be
1905 * optimized for the cases when header is void.
1907 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1909 memcpy((struct skb_shared_info *)(data + size),
1911 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1914 * if shinfo is shared we must drop the old head gracefully, but if it
1915 * is not we can just drop the old head and let the existing refcount
1916 * be since all we did is relocate the values
1918 if (skb_cloned(skb)) {
1919 if (skb_orphan_frags(skb, gfp_mask))
1922 refcount_inc(&skb_uarg(skb)->refcnt);
1923 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1924 skb_frag_ref(skb, i);
1926 if (skb_has_frag_list(skb))
1927 skb_clone_fraglist(skb);
1929 skb_release_data(skb, SKB_CONSUMED);
1933 off = (data + nhead) - skb->head;
1939 skb_set_end_offset(skb, size);
1940 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1944 skb_headers_offset_update(skb, nhead);
1948 atomic_set(&skb_shinfo(skb)->dataref, 1);
1950 skb_metadata_clear(skb);
1952 /* It is not generally safe to change skb->truesize.
1953 * For the moment, we really care of rx path, or
1954 * when skb is orphaned (not attached to a socket).
1956 if (!skb->sk || skb->destructor == sock_edemux)
1957 skb->truesize += size - osize;
1966 EXPORT_SYMBOL(pskb_expand_head);
1968 /* Make private copy of skb with writable head and some headroom */
1970 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1972 struct sk_buff *skb2;
1973 int delta = headroom - skb_headroom(skb);
1976 skb2 = pskb_copy(skb, GFP_ATOMIC);
1978 skb2 = skb_clone(skb, GFP_ATOMIC);
1979 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1987 EXPORT_SYMBOL(skb_realloc_headroom);
1989 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1991 unsigned int saved_end_offset, saved_truesize;
1992 struct skb_shared_info *shinfo;
1995 saved_end_offset = skb_end_offset(skb);
1996 saved_truesize = skb->truesize;
1998 res = pskb_expand_head(skb, 0, 0, pri);
2002 skb->truesize = saved_truesize;
2004 if (likely(skb_end_offset(skb) == saved_end_offset))
2007 shinfo = skb_shinfo(skb);
2009 /* We are about to change back skb->end,
2010 * we need to move skb_shinfo() to its new location.
2012 memmove(skb->head + saved_end_offset,
2014 offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2016 skb_set_end_offset(skb, saved_end_offset);
2022 * skb_expand_head - reallocate header of &sk_buff
2023 * @skb: buffer to reallocate
2024 * @headroom: needed headroom
2026 * Unlike skb_realloc_headroom, this one does not allocate a new skb
2027 * if possible; copies skb->sk to new skb as needed
2028 * and frees original skb in case of failures.
2030 * It expect increased headroom and generates warning otherwise.
2033 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2035 int delta = headroom - skb_headroom(skb);
2036 int osize = skb_end_offset(skb);
2037 struct sock *sk = skb->sk;
2039 if (WARN_ONCE(delta <= 0,
2040 "%s is expecting an increase in the headroom", __func__))
2043 delta = SKB_DATA_ALIGN(delta);
2044 /* pskb_expand_head() might crash, if skb is shared. */
2045 if (skb_shared(skb) || !is_skb_wmem(skb)) {
2046 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2048 if (unlikely(!nskb))
2052 skb_set_owner_w(nskb, sk);
2056 if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2059 if (sk && is_skb_wmem(skb)) {
2060 delta = skb_end_offset(skb) - osize;
2061 refcount_add(delta, &sk->sk_wmem_alloc);
2062 skb->truesize += delta;
2070 EXPORT_SYMBOL(skb_expand_head);
2073 * skb_copy_expand - copy and expand sk_buff
2074 * @skb: buffer to copy
2075 * @newheadroom: new free bytes at head
2076 * @newtailroom: new free bytes at tail
2077 * @gfp_mask: allocation priority
2079 * Make a copy of both an &sk_buff and its data and while doing so
2080 * allocate additional space.
2082 * This is used when the caller wishes to modify the data and needs a
2083 * private copy of the data to alter as well as more space for new fields.
2084 * Returns %NULL on failure or the pointer to the buffer
2085 * on success. The returned buffer has a reference count of 1.
2087 * You must pass %GFP_ATOMIC as the allocation priority if this function
2088 * is called from an interrupt.
2090 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2091 int newheadroom, int newtailroom,
2095 * Allocate the copy buffer
2097 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
2098 gfp_mask, skb_alloc_rx_flag(skb),
2100 int oldheadroom = skb_headroom(skb);
2101 int head_copy_len, head_copy_off;
2106 skb_reserve(n, newheadroom);
2108 /* Set the tail pointer and length */
2109 skb_put(n, skb->len);
2111 head_copy_len = oldheadroom;
2113 if (newheadroom <= head_copy_len)
2114 head_copy_len = newheadroom;
2116 head_copy_off = newheadroom - head_copy_len;
2118 /* Copy the linear header and data. */
2119 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2120 skb->len + head_copy_len));
2122 skb_copy_header(n, skb);
2124 skb_headers_offset_update(n, newheadroom - oldheadroom);
2128 EXPORT_SYMBOL(skb_copy_expand);
2131 * __skb_pad - zero pad the tail of an skb
2132 * @skb: buffer to pad
2133 * @pad: space to pad
2134 * @free_on_error: free buffer on error
2136 * Ensure that a buffer is followed by a padding area that is zero
2137 * filled. Used by network drivers which may DMA or transfer data
2138 * beyond the buffer end onto the wire.
2140 * May return error in out of memory cases. The skb is freed on error
2141 * if @free_on_error is true.
2144 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2149 /* If the skbuff is non linear tailroom is always zero.. */
2150 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2151 memset(skb->data+skb->len, 0, pad);
2155 ntail = skb->data_len + pad - (skb->end - skb->tail);
2156 if (likely(skb_cloned(skb) || ntail > 0)) {
2157 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2162 /* FIXME: The use of this function with non-linear skb's really needs
2165 err = skb_linearize(skb);
2169 memset(skb->data + skb->len, 0, pad);
2177 EXPORT_SYMBOL(__skb_pad);
2180 * pskb_put - add data to the tail of a potentially fragmented buffer
2181 * @skb: start of the buffer to use
2182 * @tail: tail fragment of the buffer to use
2183 * @len: amount of data to add
2185 * This function extends the used data area of the potentially
2186 * fragmented buffer. @tail must be the last fragment of @skb -- or
2187 * @skb itself. If this would exceed the total buffer size the kernel
2188 * will panic. A pointer to the first byte of the extra data is
2192 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2195 skb->data_len += len;
2198 return skb_put(tail, len);
2200 EXPORT_SYMBOL_GPL(pskb_put);
2203 * skb_put - add data to a buffer
2204 * @skb: buffer to use
2205 * @len: amount of data to add
2207 * This function extends the used data area of the buffer. If this would
2208 * exceed the total buffer size the kernel will panic. A pointer to the
2209 * first byte of the extra data is returned.
2211 void *skb_put(struct sk_buff *skb, unsigned int len)
2213 void *tmp = skb_tail_pointer(skb);
2214 SKB_LINEAR_ASSERT(skb);
2217 if (unlikely(skb->tail > skb->end))
2218 skb_over_panic(skb, len, __builtin_return_address(0));
2221 EXPORT_SYMBOL(skb_put);
2224 * skb_push - add data to the start of a buffer
2225 * @skb: buffer to use
2226 * @len: amount of data to add
2228 * This function extends the used data area of the buffer at the buffer
2229 * start. If this would exceed the total buffer headroom the kernel will
2230 * panic. A pointer to the first byte of the extra data is returned.
2232 void *skb_push(struct sk_buff *skb, unsigned int len)
2236 if (unlikely(skb->data < skb->head))
2237 skb_under_panic(skb, len, __builtin_return_address(0));
2240 EXPORT_SYMBOL(skb_push);
2243 * skb_pull - remove data from the start of a buffer
2244 * @skb: buffer to use
2245 * @len: amount of data to remove
2247 * This function removes data from the start of a buffer, returning
2248 * the memory to the headroom. A pointer to the next data in the buffer
2249 * is returned. Once the data has been pulled future pushes will overwrite
2252 void *skb_pull(struct sk_buff *skb, unsigned int len)
2254 return skb_pull_inline(skb, len);
2256 EXPORT_SYMBOL(skb_pull);
2259 * skb_pull_data - remove data from the start of a buffer returning its
2260 * original position.
2261 * @skb: buffer to use
2262 * @len: amount of data to remove
2264 * This function removes data from the start of a buffer, returning
2265 * the memory to the headroom. A pointer to the original data in the buffer
2266 * is returned after checking if there is enough data to pull. Once the
2267 * data has been pulled future pushes will overwrite the old data.
2269 void *skb_pull_data(struct sk_buff *skb, size_t len)
2271 void *data = skb->data;
2280 EXPORT_SYMBOL(skb_pull_data);
2283 * skb_trim - remove end from a buffer
2284 * @skb: buffer to alter
2287 * Cut the length of a buffer down by removing data from the tail. If
2288 * the buffer is already under the length specified it is not modified.
2289 * The skb must be linear.
2291 void skb_trim(struct sk_buff *skb, unsigned int len)
2294 __skb_trim(skb, len);
2296 EXPORT_SYMBOL(skb_trim);
2298 /* Trims skb to length len. It can change skb pointers.
2301 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2303 struct sk_buff **fragp;
2304 struct sk_buff *frag;
2305 int offset = skb_headlen(skb);
2306 int nfrags = skb_shinfo(skb)->nr_frags;
2310 if (skb_cloned(skb) &&
2311 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2318 for (; i < nfrags; i++) {
2319 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2326 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2329 skb_shinfo(skb)->nr_frags = i;
2331 for (; i < nfrags; i++)
2332 skb_frag_unref(skb, i);
2334 if (skb_has_frag_list(skb))
2335 skb_drop_fraglist(skb);
2339 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2340 fragp = &frag->next) {
2341 int end = offset + frag->len;
2343 if (skb_shared(frag)) {
2344 struct sk_buff *nfrag;
2346 nfrag = skb_clone(frag, GFP_ATOMIC);
2347 if (unlikely(!nfrag))
2350 nfrag->next = frag->next;
2362 unlikely((err = pskb_trim(frag, len - offset))))
2366 skb_drop_list(&frag->next);
2371 if (len > skb_headlen(skb)) {
2372 skb->data_len -= skb->len - len;
2377 skb_set_tail_pointer(skb, len);
2380 if (!skb->sk || skb->destructor == sock_edemux)
2384 EXPORT_SYMBOL(___pskb_trim);
2386 /* Note : use pskb_trim_rcsum() instead of calling this directly
2388 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2390 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2391 int delta = skb->len - len;
2393 skb->csum = csum_block_sub(skb->csum,
2394 skb_checksum(skb, len, delta, 0),
2396 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2397 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2398 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2400 if (offset + sizeof(__sum16) > hdlen)
2403 return __pskb_trim(skb, len);
2405 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2408 * __pskb_pull_tail - advance tail of skb header
2409 * @skb: buffer to reallocate
2410 * @delta: number of bytes to advance tail
2412 * The function makes a sense only on a fragmented &sk_buff,
2413 * it expands header moving its tail forward and copying necessary
2414 * data from fragmented part.
2416 * &sk_buff MUST have reference count of 1.
2418 * Returns %NULL (and &sk_buff does not change) if pull failed
2419 * or value of new tail of skb in the case of success.
2421 * All the pointers pointing into skb header may change and must be
2422 * reloaded after call to this function.
2425 /* Moves tail of skb head forward, copying data from fragmented part,
2426 * when it is necessary.
2427 * 1. It may fail due to malloc failure.
2428 * 2. It may change skb pointers.
2430 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2432 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2434 /* If skb has not enough free space at tail, get new one
2435 * plus 128 bytes for future expansions. If we have enough
2436 * room at tail, reallocate without expansion only if skb is cloned.
2438 int i, k, eat = (skb->tail + delta) - skb->end;
2440 if (eat > 0 || skb_cloned(skb)) {
2441 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2446 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2447 skb_tail_pointer(skb), delta));
2449 /* Optimization: no fragments, no reasons to preestimate
2450 * size of pulled pages. Superb.
2452 if (!skb_has_frag_list(skb))
2455 /* Estimate size of pulled pages. */
2457 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2458 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2465 /* If we need update frag list, we are in troubles.
2466 * Certainly, it is possible to add an offset to skb data,
2467 * but taking into account that pulling is expected to
2468 * be very rare operation, it is worth to fight against
2469 * further bloating skb head and crucify ourselves here instead.
2470 * Pure masohism, indeed. 8)8)
2473 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2474 struct sk_buff *clone = NULL;
2475 struct sk_buff *insp = NULL;
2478 if (list->len <= eat) {
2479 /* Eaten as whole. */
2484 /* Eaten partially. */
2486 if (skb_shared(list)) {
2487 /* Sucks! We need to fork list. :-( */
2488 clone = skb_clone(list, GFP_ATOMIC);
2494 /* This may be pulled without
2498 if (!pskb_pull(list, eat)) {
2506 /* Free pulled out fragments. */
2507 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2508 skb_shinfo(skb)->frag_list = list->next;
2511 /* And insert new clone at head. */
2514 skb_shinfo(skb)->frag_list = clone;
2517 /* Success! Now we may commit changes to skb data. */
2522 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2523 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2526 skb_frag_unref(skb, i);
2529 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2531 *frag = skb_shinfo(skb)->frags[i];
2533 skb_frag_off_add(frag, eat);
2534 skb_frag_size_sub(frag, eat);
2542 skb_shinfo(skb)->nr_frags = k;
2546 skb->data_len -= delta;
2549 skb_zcopy_clear(skb, false);
2551 return skb_tail_pointer(skb);
2553 EXPORT_SYMBOL(__pskb_pull_tail);
2556 * skb_copy_bits - copy bits from skb to kernel buffer
2558 * @offset: offset in source
2559 * @to: destination buffer
2560 * @len: number of bytes to copy
2562 * Copy the specified number of bytes from the source skb to the
2563 * destination buffer.
2566 * If its prototype is ever changed,
2567 * check arch/{*}/net/{*}.S files,
2568 * since it is called from BPF assembly code.
2570 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2572 int start = skb_headlen(skb);
2573 struct sk_buff *frag_iter;
2576 if (offset > (int)skb->len - len)
2580 if ((copy = start - offset) > 0) {
2583 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2584 if ((len -= copy) == 0)
2590 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2592 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2594 WARN_ON(start > offset + len);
2596 end = start + skb_frag_size(f);
2597 if ((copy = end - offset) > 0) {
2598 u32 p_off, p_len, copied;
2605 skb_frag_foreach_page(f,
2606 skb_frag_off(f) + offset - start,
2607 copy, p, p_off, p_len, copied) {
2608 vaddr = kmap_atomic(p);
2609 memcpy(to + copied, vaddr + p_off, p_len);
2610 kunmap_atomic(vaddr);
2613 if ((len -= copy) == 0)
2621 skb_walk_frags(skb, frag_iter) {
2624 WARN_ON(start > offset + len);
2626 end = start + frag_iter->len;
2627 if ((copy = end - offset) > 0) {
2630 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2632 if ((len -= copy) == 0)
2646 EXPORT_SYMBOL(skb_copy_bits);
2649 * Callback from splice_to_pipe(), if we need to release some pages
2650 * at the end of the spd in case we error'ed out in filling the pipe.
2652 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2654 put_page(spd->pages[i]);
2657 static struct page *linear_to_page(struct page *page, unsigned int *len,
2658 unsigned int *offset,
2661 struct page_frag *pfrag = sk_page_frag(sk);
2663 if (!sk_page_frag_refill(sk, pfrag))
2666 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2668 memcpy(page_address(pfrag->page) + pfrag->offset,
2669 page_address(page) + *offset, *len);
2670 *offset = pfrag->offset;
2671 pfrag->offset += *len;
2676 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2678 unsigned int offset)
2680 return spd->nr_pages &&
2681 spd->pages[spd->nr_pages - 1] == page &&
2682 (spd->partial[spd->nr_pages - 1].offset +
2683 spd->partial[spd->nr_pages - 1].len == offset);
2687 * Fill page/offset/length into spd, if it can hold more pages.
2689 static bool spd_fill_page(struct splice_pipe_desc *spd,
2690 struct pipe_inode_info *pipe, struct page *page,
2691 unsigned int *len, unsigned int offset,
2695 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2699 page = linear_to_page(page, len, &offset, sk);
2703 if (spd_can_coalesce(spd, page, offset)) {
2704 spd->partial[spd->nr_pages - 1].len += *len;
2708 spd->pages[spd->nr_pages] = page;
2709 spd->partial[spd->nr_pages].len = *len;
2710 spd->partial[spd->nr_pages].offset = offset;
2716 static bool __splice_segment(struct page *page, unsigned int poff,
2717 unsigned int plen, unsigned int *off,
2719 struct splice_pipe_desc *spd, bool linear,
2721 struct pipe_inode_info *pipe)
2726 /* skip this segment if already processed */
2732 /* ignore any bits we already processed */
2738 unsigned int flen = min(*len, plen);
2740 if (spd_fill_page(spd, pipe, page, &flen, poff,
2746 } while (*len && plen);
2752 * Map linear and fragment data from the skb to spd. It reports true if the
2753 * pipe is full or if we already spliced the requested length.
2755 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2756 unsigned int *offset, unsigned int *len,
2757 struct splice_pipe_desc *spd, struct sock *sk)
2760 struct sk_buff *iter;
2762 /* map the linear part :
2763 * If skb->head_frag is set, this 'linear' part is backed by a
2764 * fragment, and if the head is not shared with any clones then
2765 * we can avoid a copy since we own the head portion of this page.
2767 if (__splice_segment(virt_to_page(skb->data),
2768 (unsigned long) skb->data & (PAGE_SIZE - 1),
2771 skb_head_is_locked(skb),
2776 * then map the fragments
2778 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2779 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2781 if (__splice_segment(skb_frag_page(f),
2782 skb_frag_off(f), skb_frag_size(f),
2783 offset, len, spd, false, sk, pipe))
2787 skb_walk_frags(skb, iter) {
2788 if (*offset >= iter->len) {
2789 *offset -= iter->len;
2792 /* __skb_splice_bits() only fails if the output has no room
2793 * left, so no point in going over the frag_list for the error
2796 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2804 * Map data from the skb to a pipe. Should handle both the linear part,
2805 * the fragments, and the frag list.
2807 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2808 struct pipe_inode_info *pipe, unsigned int tlen,
2811 struct partial_page partial[MAX_SKB_FRAGS];
2812 struct page *pages[MAX_SKB_FRAGS];
2813 struct splice_pipe_desc spd = {
2816 .nr_pages_max = MAX_SKB_FRAGS,
2817 .ops = &nosteal_pipe_buf_ops,
2818 .spd_release = sock_spd_release,
2822 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2825 ret = splice_to_pipe(pipe, &spd);
2829 EXPORT_SYMBOL_GPL(skb_splice_bits);
2831 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2832 struct kvec *vec, size_t num, size_t size)
2834 struct socket *sock = sk->sk_socket;
2838 return kernel_sendmsg(sock, msg, vec, num, size);
2841 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2842 size_t size, int flags)
2844 struct socket *sock = sk->sk_socket;
2848 return kernel_sendpage(sock, page, offset, size, flags);
2851 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2852 struct kvec *vec, size_t num, size_t size);
2853 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2854 size_t size, int flags);
2855 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2856 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2858 unsigned int orig_len = len;
2859 struct sk_buff *head = skb;
2860 unsigned short fragidx;
2865 /* Deal with head data */
2866 while (offset < skb_headlen(skb) && len) {
2870 slen = min_t(int, len, skb_headlen(skb) - offset);
2871 kv.iov_base = skb->data + offset;
2873 memset(&msg, 0, sizeof(msg));
2874 msg.msg_flags = MSG_DONTWAIT;
2876 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2877 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2885 /* All the data was skb head? */
2889 /* Make offset relative to start of frags */
2890 offset -= skb_headlen(skb);
2892 /* Find where we are in frag list */
2893 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2894 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2896 if (offset < skb_frag_size(frag))
2899 offset -= skb_frag_size(frag);
2902 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2903 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2905 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2908 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2909 sendpage_unlocked, sk,
2910 skb_frag_page(frag),
2911 skb_frag_off(frag) + offset,
2912 slen, MSG_DONTWAIT);
2925 /* Process any frag lists */
2928 if (skb_has_frag_list(skb)) {
2929 skb = skb_shinfo(skb)->frag_list;
2932 } else if (skb->next) {
2939 return orig_len - len;
2942 return orig_len == len ? ret : orig_len - len;
2945 /* Send skb data on a socket. Socket must be locked. */
2946 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2949 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2950 kernel_sendpage_locked);
2952 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2954 /* Send skb data on a socket. Socket must be unlocked. */
2955 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2957 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2962 * skb_store_bits - store bits from kernel buffer to skb
2963 * @skb: destination buffer
2964 * @offset: offset in destination
2965 * @from: source buffer
2966 * @len: number of bytes to copy
2968 * Copy the specified number of bytes from the source buffer to the
2969 * destination skb. This function handles all the messy bits of
2970 * traversing fragment lists and such.
2973 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2975 int start = skb_headlen(skb);
2976 struct sk_buff *frag_iter;
2979 if (offset > (int)skb->len - len)
2982 if ((copy = start - offset) > 0) {
2985 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2986 if ((len -= copy) == 0)
2992 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2993 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2996 WARN_ON(start > offset + len);
2998 end = start + skb_frag_size(frag);
2999 if ((copy = end - offset) > 0) {
3000 u32 p_off, p_len, copied;
3007 skb_frag_foreach_page(frag,
3008 skb_frag_off(frag) + offset - start,
3009 copy, p, p_off, p_len, copied) {
3010 vaddr = kmap_atomic(p);
3011 memcpy(vaddr + p_off, from + copied, p_len);
3012 kunmap_atomic(vaddr);
3015 if ((len -= copy) == 0)
3023 skb_walk_frags(skb, frag_iter) {
3026 WARN_ON(start > offset + len);
3028 end = start + frag_iter->len;
3029 if ((copy = end - offset) > 0) {
3032 if (skb_store_bits(frag_iter, offset - start,
3035 if ((len -= copy) == 0)
3048 EXPORT_SYMBOL(skb_store_bits);
3050 /* Checksum skb data. */
3051 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3052 __wsum csum, const struct skb_checksum_ops *ops)
3054 int start = skb_headlen(skb);
3055 int i, copy = start - offset;
3056 struct sk_buff *frag_iter;
3059 /* Checksum header. */
3063 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3064 skb->data + offset, copy, csum);
3065 if ((len -= copy) == 0)
3071 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3073 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3075 WARN_ON(start > offset + len);
3077 end = start + skb_frag_size(frag);
3078 if ((copy = end - offset) > 0) {
3079 u32 p_off, p_len, copied;
3087 skb_frag_foreach_page(frag,
3088 skb_frag_off(frag) + offset - start,
3089 copy, p, p_off, p_len, copied) {
3090 vaddr = kmap_atomic(p);
3091 csum2 = INDIRECT_CALL_1(ops->update,
3093 vaddr + p_off, p_len, 0);
3094 kunmap_atomic(vaddr);
3095 csum = INDIRECT_CALL_1(ops->combine,
3096 csum_block_add_ext, csum,
3108 skb_walk_frags(skb, frag_iter) {
3111 WARN_ON(start > offset + len);
3113 end = start + frag_iter->len;
3114 if ((copy = end - offset) > 0) {
3118 csum2 = __skb_checksum(frag_iter, offset - start,
3120 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3121 csum, csum2, pos, copy);
3122 if ((len -= copy) == 0)
3133 EXPORT_SYMBOL(__skb_checksum);
3135 __wsum skb_checksum(const struct sk_buff *skb, int offset,
3136 int len, __wsum csum)
3138 const struct skb_checksum_ops ops = {
3139 .update = csum_partial_ext,
3140 .combine = csum_block_add_ext,
3143 return __skb_checksum(skb, offset, len, csum, &ops);
3145 EXPORT_SYMBOL(skb_checksum);
3147 /* Both of above in one bottle. */
3149 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3152 int start = skb_headlen(skb);
3153 int i, copy = start - offset;
3154 struct sk_buff *frag_iter;
3162 csum = csum_partial_copy_nocheck(skb->data + offset, to,
3164 if ((len -= copy) == 0)
3171 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3174 WARN_ON(start > offset + len);
3176 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3177 if ((copy = end - offset) > 0) {
3178 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3179 u32 p_off, p_len, copied;
3187 skb_frag_foreach_page(frag,
3188 skb_frag_off(frag) + offset - start,
3189 copy, p, p_off, p_len, copied) {
3190 vaddr = kmap_atomic(p);
3191 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3194 kunmap_atomic(vaddr);
3195 csum = csum_block_add(csum, csum2, pos);
3207 skb_walk_frags(skb, frag_iter) {
3211 WARN_ON(start > offset + len);
3213 end = start + frag_iter->len;
3214 if ((copy = end - offset) > 0) {
3217 csum2 = skb_copy_and_csum_bits(frag_iter,
3220 csum = csum_block_add(csum, csum2, pos);
3221 if ((len -= copy) == 0)
3232 EXPORT_SYMBOL(skb_copy_and_csum_bits);
3234 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3238 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3239 /* See comments in __skb_checksum_complete(). */
3241 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3242 !skb->csum_complete_sw)
3243 netdev_rx_csum_fault(skb->dev, skb);
3245 if (!skb_shared(skb))
3246 skb->csum_valid = !sum;
3249 EXPORT_SYMBOL(__skb_checksum_complete_head);
3251 /* This function assumes skb->csum already holds pseudo header's checksum,
3252 * which has been changed from the hardware checksum, for example, by
3253 * __skb_checksum_validate_complete(). And, the original skb->csum must
3254 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3256 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3257 * zero. The new checksum is stored back into skb->csum unless the skb is
3260 __sum16 __skb_checksum_complete(struct sk_buff *skb)
3265 csum = skb_checksum(skb, 0, skb->len, 0);
3267 sum = csum_fold(csum_add(skb->csum, csum));
3268 /* This check is inverted, because we already knew the hardware
3269 * checksum is invalid before calling this function. So, if the
3270 * re-computed checksum is valid instead, then we have a mismatch
3271 * between the original skb->csum and skb_checksum(). This means either
3272 * the original hardware checksum is incorrect or we screw up skb->csum
3273 * when moving skb->data around.
3276 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3277 !skb->csum_complete_sw)
3278 netdev_rx_csum_fault(skb->dev, skb);
3281 if (!skb_shared(skb)) {
3282 /* Save full packet checksum */
3284 skb->ip_summed = CHECKSUM_COMPLETE;
3285 skb->csum_complete_sw = 1;
3286 skb->csum_valid = !sum;
3291 EXPORT_SYMBOL(__skb_checksum_complete);
3293 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3295 net_warn_ratelimited(
3296 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3301 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3302 int offset, int len)
3304 net_warn_ratelimited(
3305 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3310 static const struct skb_checksum_ops default_crc32c_ops = {
3311 .update = warn_crc32c_csum_update,
3312 .combine = warn_crc32c_csum_combine,
3315 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3316 &default_crc32c_ops;
3317 EXPORT_SYMBOL(crc32c_csum_stub);
3320 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3321 * @from: source buffer
3323 * Calculates the amount of linear headroom needed in the 'to' skb passed
3324 * into skb_zerocopy().
3327 skb_zerocopy_headlen(const struct sk_buff *from)
3329 unsigned int hlen = 0;
3331 if (!from->head_frag ||
3332 skb_headlen(from) < L1_CACHE_BYTES ||
3333 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3334 hlen = skb_headlen(from);
3339 if (skb_has_frag_list(from))
3344 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3347 * skb_zerocopy - Zero copy skb to skb
3348 * @to: destination buffer
3349 * @from: source buffer
3350 * @len: number of bytes to copy from source buffer
3351 * @hlen: size of linear headroom in destination buffer
3353 * Copies up to `len` bytes from `from` to `to` by creating references
3354 * to the frags in the source buffer.
3356 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3357 * headroom in the `to` buffer.
3360 * 0: everything is OK
3361 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3362 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3365 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3368 int plen = 0; /* length of skb->head fragment */
3371 unsigned int offset;
3373 BUG_ON(!from->head_frag && !hlen);
3375 /* dont bother with small payloads */
3376 if (len <= skb_tailroom(to))
3377 return skb_copy_bits(from, 0, skb_put(to, len), len);
3380 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3385 plen = min_t(int, skb_headlen(from), len);
3387 page = virt_to_head_page(from->head);
3388 offset = from->data - (unsigned char *)page_address(page);
3389 __skb_fill_page_desc(to, 0, page, offset, plen);
3396 skb_len_add(to, len + plen);
3398 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3402 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3404 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3409 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3410 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3412 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3414 skb_frag_ref(to, j);
3417 skb_shinfo(to)->nr_frags = j;
3421 EXPORT_SYMBOL_GPL(skb_zerocopy);
3423 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3428 if (skb->ip_summed == CHECKSUM_PARTIAL)
3429 csstart = skb_checksum_start_offset(skb);
3431 csstart = skb_headlen(skb);
3433 BUG_ON(csstart > skb_headlen(skb));
3435 skb_copy_from_linear_data(skb, to, csstart);
3438 if (csstart != skb->len)
3439 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3440 skb->len - csstart);
3442 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3443 long csstuff = csstart + skb->csum_offset;
3445 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3448 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3451 * skb_dequeue - remove from the head of the queue
3452 * @list: list to dequeue from
3454 * Remove the head of the list. The list lock is taken so the function
3455 * may be used safely with other locking list functions. The head item is
3456 * returned or %NULL if the list is empty.
3459 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3461 unsigned long flags;
3462 struct sk_buff *result;
3464 spin_lock_irqsave(&list->lock, flags);
3465 result = __skb_dequeue(list);
3466 spin_unlock_irqrestore(&list->lock, flags);
3469 EXPORT_SYMBOL(skb_dequeue);
3472 * skb_dequeue_tail - remove from the tail of the queue
3473 * @list: list to dequeue from
3475 * Remove the tail of the list. The list lock is taken so the function
3476 * may be used safely with other locking list functions. The tail item is
3477 * returned or %NULL if the list is empty.
3479 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3481 unsigned long flags;
3482 struct sk_buff *result;
3484 spin_lock_irqsave(&list->lock, flags);
3485 result = __skb_dequeue_tail(list);
3486 spin_unlock_irqrestore(&list->lock, flags);
3489 EXPORT_SYMBOL(skb_dequeue_tail);
3492 * skb_queue_purge - empty a list
3493 * @list: list to empty
3495 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3496 * the list and one reference dropped. This function takes the list
3497 * lock and is atomic with respect to other list locking functions.
3499 void skb_queue_purge(struct sk_buff_head *list)
3501 struct sk_buff *skb;
3502 while ((skb = skb_dequeue(list)) != NULL)
3505 EXPORT_SYMBOL(skb_queue_purge);
3508 * skb_rbtree_purge - empty a skb rbtree
3509 * @root: root of the rbtree to empty
3510 * Return value: the sum of truesizes of all purged skbs.
3512 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3513 * the list and one reference dropped. This function does not take
3514 * any lock. Synchronization should be handled by the caller (e.g., TCP
3515 * out-of-order queue is protected by the socket lock).
3517 unsigned int skb_rbtree_purge(struct rb_root *root)
3519 struct rb_node *p = rb_first(root);
3520 unsigned int sum = 0;
3523 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3526 rb_erase(&skb->rbnode, root);
3527 sum += skb->truesize;
3534 * skb_queue_head - queue a buffer at the list head
3535 * @list: list to use
3536 * @newsk: buffer to queue
3538 * Queue a buffer at the start of the list. This function takes the
3539 * list lock and can be used safely with other locking &sk_buff functions
3542 * A buffer cannot be placed on two lists at the same time.
3544 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3546 unsigned long flags;
3548 spin_lock_irqsave(&list->lock, flags);
3549 __skb_queue_head(list, newsk);
3550 spin_unlock_irqrestore(&list->lock, flags);
3552 EXPORT_SYMBOL(skb_queue_head);
3555 * skb_queue_tail - queue a buffer at the list tail
3556 * @list: list to use
3557 * @newsk: buffer to queue
3559 * Queue a buffer at the tail of the list. This function takes the
3560 * list lock and can be used safely with other locking &sk_buff functions
3563 * A buffer cannot be placed on two lists at the same time.
3565 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3567 unsigned long flags;
3569 spin_lock_irqsave(&list->lock, flags);
3570 __skb_queue_tail(list, newsk);
3571 spin_unlock_irqrestore(&list->lock, flags);
3573 EXPORT_SYMBOL(skb_queue_tail);
3576 * skb_unlink - remove a buffer from a list
3577 * @skb: buffer to remove
3578 * @list: list to use
3580 * Remove a packet from a list. The list locks are taken and this
3581 * function is atomic with respect to other list locked calls
3583 * You must know what list the SKB is on.
3585 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3587 unsigned long flags;
3589 spin_lock_irqsave(&list->lock, flags);
3590 __skb_unlink(skb, list);
3591 spin_unlock_irqrestore(&list->lock, flags);
3593 EXPORT_SYMBOL(skb_unlink);
3596 * skb_append - append a buffer
3597 * @old: buffer to insert after
3598 * @newsk: buffer to insert
3599 * @list: list to use
3601 * Place a packet after a given packet in a list. The list locks are taken
3602 * and this function is atomic with respect to other list locked calls.
3603 * A buffer cannot be placed on two lists at the same time.
3605 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3607 unsigned long flags;
3609 spin_lock_irqsave(&list->lock, flags);
3610 __skb_queue_after(list, old, newsk);
3611 spin_unlock_irqrestore(&list->lock, flags);
3613 EXPORT_SYMBOL(skb_append);
3615 static inline void skb_split_inside_header(struct sk_buff *skb,
3616 struct sk_buff* skb1,
3617 const u32 len, const int pos)
3621 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3623 /* And move data appendix as is. */
3624 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3625 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3627 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3628 skb_shinfo(skb)->nr_frags = 0;
3629 skb1->data_len = skb->data_len;
3630 skb1->len += skb1->data_len;
3633 skb_set_tail_pointer(skb, len);
3636 static inline void skb_split_no_header(struct sk_buff *skb,
3637 struct sk_buff* skb1,
3638 const u32 len, int pos)
3641 const int nfrags = skb_shinfo(skb)->nr_frags;
3643 skb_shinfo(skb)->nr_frags = 0;
3644 skb1->len = skb1->data_len = skb->len - len;
3646 skb->data_len = len - pos;
3648 for (i = 0; i < nfrags; i++) {
3649 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3651 if (pos + size > len) {
3652 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3656 * We have two variants in this case:
3657 * 1. Move all the frag to the second
3658 * part, if it is possible. F.e.
3659 * this approach is mandatory for TUX,
3660 * where splitting is expensive.
3661 * 2. Split is accurately. We make this.
3663 skb_frag_ref(skb, i);
3664 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3665 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3666 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3667 skb_shinfo(skb)->nr_frags++;
3671 skb_shinfo(skb)->nr_frags++;
3674 skb_shinfo(skb1)->nr_frags = k;
3678 * skb_split - Split fragmented skb to two parts at length len.
3679 * @skb: the buffer to split
3680 * @skb1: the buffer to receive the second part
3681 * @len: new length for skb
3683 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3685 int pos = skb_headlen(skb);
3686 const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
3688 skb_zcopy_downgrade_managed(skb);
3690 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
3691 skb_zerocopy_clone(skb1, skb, 0);
3692 if (len < pos) /* Split line is inside header. */
3693 skb_split_inside_header(skb, skb1, len, pos);
3694 else /* Second chunk has no header, nothing to copy. */
3695 skb_split_no_header(skb, skb1, len, pos);
3697 EXPORT_SYMBOL(skb_split);
3699 /* Shifting from/to a cloned skb is a no-go.
3701 * Caller cannot keep skb_shinfo related pointers past calling here!
3703 static int skb_prepare_for_shift(struct sk_buff *skb)
3705 return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
3709 * skb_shift - Shifts paged data partially from skb to another
3710 * @tgt: buffer into which tail data gets added
3711 * @skb: buffer from which the paged data comes from
3712 * @shiftlen: shift up to this many bytes
3714 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3715 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3716 * It's up to caller to free skb if everything was shifted.
3718 * If @tgt runs out of frags, the whole operation is aborted.
3720 * Skb cannot include anything else but paged data while tgt is allowed
3721 * to have non-paged data as well.
3723 * TODO: full sized shift could be optimized but that would need
3724 * specialized skb free'er to handle frags without up-to-date nr_frags.
3726 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3728 int from, to, merge, todo;
3729 skb_frag_t *fragfrom, *fragto;
3731 BUG_ON(shiftlen > skb->len);
3733 if (skb_headlen(skb))
3735 if (skb_zcopy(tgt) || skb_zcopy(skb))
3740 to = skb_shinfo(tgt)->nr_frags;
3741 fragfrom = &skb_shinfo(skb)->frags[from];
3743 /* Actual merge is delayed until the point when we know we can
3744 * commit all, so that we don't have to undo partial changes
3747 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3748 skb_frag_off(fragfrom))) {
3753 todo -= skb_frag_size(fragfrom);
3755 if (skb_prepare_for_shift(skb) ||
3756 skb_prepare_for_shift(tgt))
3759 /* All previous frag pointers might be stale! */
3760 fragfrom = &skb_shinfo(skb)->frags[from];
3761 fragto = &skb_shinfo(tgt)->frags[merge];
3763 skb_frag_size_add(fragto, shiftlen);
3764 skb_frag_size_sub(fragfrom, shiftlen);
3765 skb_frag_off_add(fragfrom, shiftlen);
3773 /* Skip full, not-fitting skb to avoid expensive operations */
3774 if ((shiftlen == skb->len) &&
3775 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3778 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3781 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3782 if (to == MAX_SKB_FRAGS)
3785 fragfrom = &skb_shinfo(skb)->frags[from];
3786 fragto = &skb_shinfo(tgt)->frags[to];
3788 if (todo >= skb_frag_size(fragfrom)) {
3789 *fragto = *fragfrom;
3790 todo -= skb_frag_size(fragfrom);
3795 __skb_frag_ref(fragfrom);
3796 skb_frag_page_copy(fragto, fragfrom);
3797 skb_frag_off_copy(fragto, fragfrom);
3798 skb_frag_size_set(fragto, todo);
3800 skb_frag_off_add(fragfrom, todo);
3801 skb_frag_size_sub(fragfrom, todo);
3809 /* Ready to "commit" this state change to tgt */
3810 skb_shinfo(tgt)->nr_frags = to;
3813 fragfrom = &skb_shinfo(skb)->frags[0];
3814 fragto = &skb_shinfo(tgt)->frags[merge];
3816 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3817 __skb_frag_unref(fragfrom, skb->pp_recycle);
3820 /* Reposition in the original skb */
3822 while (from < skb_shinfo(skb)->nr_frags)
3823 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3824 skb_shinfo(skb)->nr_frags = to;
3826 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3829 /* Most likely the tgt won't ever need its checksum anymore, skb on
3830 * the other hand might need it if it needs to be resent
3832 tgt->ip_summed = CHECKSUM_PARTIAL;
3833 skb->ip_summed = CHECKSUM_PARTIAL;
3835 skb_len_add(skb, -shiftlen);
3836 skb_len_add(tgt, shiftlen);
3842 * skb_prepare_seq_read - Prepare a sequential read of skb data
3843 * @skb: the buffer to read
3844 * @from: lower offset of data to be read
3845 * @to: upper offset of data to be read
3846 * @st: state variable
3848 * Initializes the specified state variable. Must be called before
3849 * invoking skb_seq_read() for the first time.
3851 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3852 unsigned int to, struct skb_seq_state *st)
3854 st->lower_offset = from;
3855 st->upper_offset = to;
3856 st->root_skb = st->cur_skb = skb;
3857 st->frag_idx = st->stepped_offset = 0;
3858 st->frag_data = NULL;
3861 EXPORT_SYMBOL(skb_prepare_seq_read);
3864 * skb_seq_read - Sequentially read skb data
3865 * @consumed: number of bytes consumed by the caller so far
3866 * @data: destination pointer for data to be returned
3867 * @st: state variable
3869 * Reads a block of skb data at @consumed relative to the
3870 * lower offset specified to skb_prepare_seq_read(). Assigns
3871 * the head of the data block to @data and returns the length
3872 * of the block or 0 if the end of the skb data or the upper
3873 * offset has been reached.
3875 * The caller is not required to consume all of the data
3876 * returned, i.e. @consumed is typically set to the number
3877 * of bytes already consumed and the next call to
3878 * skb_seq_read() will return the remaining part of the block.
3880 * Note 1: The size of each block of data returned can be arbitrary,
3881 * this limitation is the cost for zerocopy sequential
3882 * reads of potentially non linear data.
3884 * Note 2: Fragment lists within fragments are not implemented
3885 * at the moment, state->root_skb could be replaced with
3886 * a stack for this purpose.
3888 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3889 struct skb_seq_state *st)
3891 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3894 if (unlikely(abs_offset >= st->upper_offset)) {
3895 if (st->frag_data) {
3896 kunmap_atomic(st->frag_data);
3897 st->frag_data = NULL;
3903 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3905 if (abs_offset < block_limit && !st->frag_data) {
3906 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3907 return block_limit - abs_offset;
3910 if (st->frag_idx == 0 && !st->frag_data)
3911 st->stepped_offset += skb_headlen(st->cur_skb);
3913 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3914 unsigned int pg_idx, pg_off, pg_sz;
3916 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3919 pg_off = skb_frag_off(frag);
3920 pg_sz = skb_frag_size(frag);
3922 if (skb_frag_must_loop(skb_frag_page(frag))) {
3923 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3924 pg_off = offset_in_page(pg_off + st->frag_off);
3925 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3926 PAGE_SIZE - pg_off);
3929 block_limit = pg_sz + st->stepped_offset;
3930 if (abs_offset < block_limit) {
3932 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3934 *data = (u8 *)st->frag_data + pg_off +
3935 (abs_offset - st->stepped_offset);
3937 return block_limit - abs_offset;
3940 if (st->frag_data) {
3941 kunmap_atomic(st->frag_data);
3942 st->frag_data = NULL;
3945 st->stepped_offset += pg_sz;
3946 st->frag_off += pg_sz;
3947 if (st->frag_off == skb_frag_size(frag)) {
3953 if (st->frag_data) {
3954 kunmap_atomic(st->frag_data);
3955 st->frag_data = NULL;
3958 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3959 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3962 } else if (st->cur_skb->next) {
3963 st->cur_skb = st->cur_skb->next;
3970 EXPORT_SYMBOL(skb_seq_read);
3973 * skb_abort_seq_read - Abort a sequential read of skb data
3974 * @st: state variable
3976 * Must be called if skb_seq_read() was not called until it
3979 void skb_abort_seq_read(struct skb_seq_state *st)
3982 kunmap_atomic(st->frag_data);
3984 EXPORT_SYMBOL(skb_abort_seq_read);
3986 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3988 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3989 struct ts_config *conf,
3990 struct ts_state *state)
3992 return skb_seq_read(offset, text, TS_SKB_CB(state));
3995 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3997 skb_abort_seq_read(TS_SKB_CB(state));
4001 * skb_find_text - Find a text pattern in skb data
4002 * @skb: the buffer to look in
4003 * @from: search offset
4005 * @config: textsearch configuration
4007 * Finds a pattern in the skb data according to the specified
4008 * textsearch configuration. Use textsearch_next() to retrieve
4009 * subsequent occurrences of the pattern. Returns the offset
4010 * to the first occurrence or UINT_MAX if no match was found.
4012 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4013 unsigned int to, struct ts_config *config)
4015 struct ts_state state;
4018 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4020 config->get_next_block = skb_ts_get_next_block;
4021 config->finish = skb_ts_finish;
4023 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4025 ret = textsearch_find(config, &state);
4026 return (ret <= to - from ? ret : UINT_MAX);
4028 EXPORT_SYMBOL(skb_find_text);
4030 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4031 int offset, size_t size)
4033 int i = skb_shinfo(skb)->nr_frags;
4035 if (skb_can_coalesce(skb, i, page, offset)) {
4036 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4037 } else if (i < MAX_SKB_FRAGS) {
4038 skb_zcopy_downgrade_managed(skb);
4040 skb_fill_page_desc_noacc(skb, i, page, offset, size);
4047 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4050 * skb_pull_rcsum - pull skb and update receive checksum
4051 * @skb: buffer to update
4052 * @len: length of data pulled
4054 * This function performs an skb_pull on the packet and updates
4055 * the CHECKSUM_COMPLETE checksum. It should be used on
4056 * receive path processing instead of skb_pull unless you know
4057 * that the checksum difference is zero (e.g., a valid IP header)
4058 * or you are setting ip_summed to CHECKSUM_NONE.
4060 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4062 unsigned char *data = skb->data;
4064 BUG_ON(len > skb->len);
4065 __skb_pull(skb, len);
4066 skb_postpull_rcsum(skb, data, len);
4069 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4071 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4073 skb_frag_t head_frag;
4076 page = virt_to_head_page(frag_skb->head);
4077 __skb_frag_set_page(&head_frag, page);
4078 skb_frag_off_set(&head_frag, frag_skb->data -
4079 (unsigned char *)page_address(page));
4080 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
4084 struct sk_buff *skb_segment_list(struct sk_buff *skb,
4085 netdev_features_t features,
4086 unsigned int offset)
4088 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4089 unsigned int tnl_hlen = skb_tnl_header_len(skb);
4090 unsigned int delta_truesize = 0;
4091 unsigned int delta_len = 0;
4092 struct sk_buff *tail = NULL;
4093 struct sk_buff *nskb, *tmp;
4096 skb_push(skb, -skb_network_offset(skb) + offset);
4098 skb_shinfo(skb)->frag_list = NULL;
4102 list_skb = list_skb->next;
4105 delta_truesize += nskb->truesize;
4106 if (skb_shared(nskb)) {
4107 tmp = skb_clone(nskb, GFP_ATOMIC);
4111 err = skb_unclone(nskb, GFP_ATOMIC);
4122 if (unlikely(err)) {
4123 nskb->next = list_skb;
4129 delta_len += nskb->len;
4131 skb_push(nskb, -skb_network_offset(nskb) + offset);
4133 skb_release_head_state(nskb);
4134 len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4135 __copy_skb_header(nskb, skb);
4137 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4138 nskb->transport_header += len_diff;
4139 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4140 nskb->data - tnl_hlen,
4143 if (skb_needs_linearize(nskb, features) &&
4144 __skb_linearize(nskb))
4149 skb->truesize = skb->truesize - delta_truesize;
4150 skb->data_len = skb->data_len - delta_len;
4151 skb->len = skb->len - delta_len;
4157 if (skb_needs_linearize(skb, features) &&
4158 __skb_linearize(skb))
4166 kfree_skb_list(skb->next);
4168 return ERR_PTR(-ENOMEM);
4170 EXPORT_SYMBOL_GPL(skb_segment_list);
4173 * skb_segment - Perform protocol segmentation on skb.
4174 * @head_skb: buffer to segment
4175 * @features: features for the output path (see dev->features)
4177 * This function performs segmentation on the given skb. It returns
4178 * a pointer to the first in a list of new skbs for the segments.
4179 * In case of error it returns ERR_PTR(err).
4181 struct sk_buff *skb_segment(struct sk_buff *head_skb,
4182 netdev_features_t features)
4184 struct sk_buff *segs = NULL;
4185 struct sk_buff *tail = NULL;
4186 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4187 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
4188 unsigned int mss = skb_shinfo(head_skb)->gso_size;
4189 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4190 struct sk_buff *frag_skb = head_skb;
4191 unsigned int offset = doffset;
4192 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4193 unsigned int partial_segs = 0;
4194 unsigned int headroom;
4195 unsigned int len = head_skb->len;
4198 int nfrags = skb_shinfo(head_skb)->nr_frags;
4203 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4204 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4205 struct sk_buff *check_skb;
4207 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4208 if (skb_headlen(check_skb) && !check_skb->head_frag) {
4209 /* gso_size is untrusted, and we have a frag_list with
4210 * a linear non head_frag item.
4212 * If head_skb's headlen does not fit requested gso_size,
4213 * it means that the frag_list members do NOT terminate
4214 * on exact gso_size boundaries. Hence we cannot perform
4215 * skb_frag_t page sharing. Therefore we must fallback to
4216 * copying the frag_list skbs; we do so by disabling SG.
4218 features &= ~NETIF_F_SG;
4224 __skb_push(head_skb, doffset);
4225 proto = skb_network_protocol(head_skb, NULL);
4226 if (unlikely(!proto))
4227 return ERR_PTR(-EINVAL);
4229 sg = !!(features & NETIF_F_SG);
4230 csum = !!can_checksum_protocol(features, proto);
4232 if (sg && csum && (mss != GSO_BY_FRAGS)) {
4233 if (!(features & NETIF_F_GSO_PARTIAL)) {
4234 struct sk_buff *iter;
4235 unsigned int frag_len;
4238 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4241 /* If we get here then all the required
4242 * GSO features except frag_list are supported.
4243 * Try to split the SKB to multiple GSO SKBs
4244 * with no frag_list.
4245 * Currently we can do that only when the buffers don't
4246 * have a linear part and all the buffers except
4247 * the last are of the same length.
4249 frag_len = list_skb->len;
4250 skb_walk_frags(head_skb, iter) {
4251 if (frag_len != iter->len && iter->next)
4253 if (skb_headlen(iter) && !iter->head_frag)
4259 if (len != frag_len)
4263 /* GSO partial only requires that we trim off any excess that
4264 * doesn't fit into an MSS sized block, so take care of that
4267 partial_segs = len / mss;
4268 if (partial_segs > 1)
4269 mss *= partial_segs;
4275 headroom = skb_headroom(head_skb);
4276 pos = skb_headlen(head_skb);
4279 struct sk_buff *nskb;
4280 skb_frag_t *nskb_frag;
4284 if (unlikely(mss == GSO_BY_FRAGS)) {
4285 len = list_skb->len;
4287 len = head_skb->len - offset;
4292 hsize = skb_headlen(head_skb) - offset;
4294 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4295 (skb_headlen(list_skb) == len || sg)) {
4296 BUG_ON(skb_headlen(list_skb) > len);
4299 nfrags = skb_shinfo(list_skb)->nr_frags;
4300 frag = skb_shinfo(list_skb)->frags;
4301 frag_skb = list_skb;
4302 pos += skb_headlen(list_skb);
4304 while (pos < offset + len) {
4305 BUG_ON(i >= nfrags);
4307 size = skb_frag_size(frag);
4308 if (pos + size > offset + len)
4316 nskb = skb_clone(list_skb, GFP_ATOMIC);
4317 list_skb = list_skb->next;
4319 if (unlikely(!nskb))
4322 if (unlikely(pskb_trim(nskb, len))) {
4327 hsize = skb_end_offset(nskb);
4328 if (skb_cow_head(nskb, doffset + headroom)) {
4333 nskb->truesize += skb_end_offset(nskb) - hsize;
4334 skb_release_head_state(nskb);
4335 __skb_push(nskb, doffset);
4339 if (hsize > len || !sg)
4342 nskb = __alloc_skb(hsize + doffset + headroom,
4343 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4346 if (unlikely(!nskb))
4349 skb_reserve(nskb, headroom);
4350 __skb_put(nskb, doffset);
4359 __copy_skb_header(nskb, head_skb);
4361 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4362 skb_reset_mac_len(nskb);
4364 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4365 nskb->data - tnl_hlen,
4366 doffset + tnl_hlen);
4368 if (nskb->len == len + doffset)
4369 goto perform_csum_check;
4373 if (!nskb->remcsum_offload)
4374 nskb->ip_summed = CHECKSUM_NONE;
4375 SKB_GSO_CB(nskb)->csum =
4376 skb_copy_and_csum_bits(head_skb, offset,
4380 SKB_GSO_CB(nskb)->csum_start =
4381 skb_headroom(nskb) + doffset;
4383 if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4389 nskb_frag = skb_shinfo(nskb)->frags;
4391 skb_copy_from_linear_data_offset(head_skb, offset,
4392 skb_put(nskb, hsize), hsize);
4394 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4397 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4398 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4401 while (pos < offset + len) {
4404 nfrags = skb_shinfo(list_skb)->nr_frags;
4405 frag = skb_shinfo(list_skb)->frags;
4406 frag_skb = list_skb;
4407 if (!skb_headlen(list_skb)) {
4410 BUG_ON(!list_skb->head_frag);
4412 /* to make room for head_frag. */
4416 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4417 skb_zerocopy_clone(nskb, frag_skb,
4421 list_skb = list_skb->next;
4424 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4426 net_warn_ratelimited(
4427 "skb_segment: too many frags: %u %u\n",
4433 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4434 __skb_frag_ref(nskb_frag);
4435 size = skb_frag_size(nskb_frag);
4438 skb_frag_off_add(nskb_frag, offset - pos);
4439 skb_frag_size_sub(nskb_frag, offset - pos);
4442 skb_shinfo(nskb)->nr_frags++;
4444 if (pos + size <= offset + len) {
4449 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4457 nskb->data_len = len - hsize;
4458 nskb->len += nskb->data_len;
4459 nskb->truesize += nskb->data_len;
4463 if (skb_has_shared_frag(nskb) &&
4464 __skb_linearize(nskb))
4467 if (!nskb->remcsum_offload)
4468 nskb->ip_summed = CHECKSUM_NONE;
4469 SKB_GSO_CB(nskb)->csum =
4470 skb_checksum(nskb, doffset,
4471 nskb->len - doffset, 0);
4472 SKB_GSO_CB(nskb)->csum_start =
4473 skb_headroom(nskb) + doffset;
4475 } while ((offset += len) < head_skb->len);
4477 /* Some callers want to get the end of the list.
4478 * Put it in segs->prev to avoid walking the list.
4479 * (see validate_xmit_skb_list() for example)
4484 struct sk_buff *iter;
4485 int type = skb_shinfo(head_skb)->gso_type;
4486 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4488 /* Update type to add partial and then remove dodgy if set */
4489 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4490 type &= ~SKB_GSO_DODGY;
4492 /* Update GSO info and prepare to start updating headers on
4493 * our way back down the stack of protocols.
4495 for (iter = segs; iter; iter = iter->next) {
4496 skb_shinfo(iter)->gso_size = gso_size;
4497 skb_shinfo(iter)->gso_segs = partial_segs;
4498 skb_shinfo(iter)->gso_type = type;
4499 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4502 if (tail->len - doffset <= gso_size)
4503 skb_shinfo(tail)->gso_size = 0;
4504 else if (tail != segs)
4505 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4508 /* Following permits correct backpressure, for protocols
4509 * using skb_set_owner_w().
4510 * Idea is to tranfert ownership from head_skb to last segment.
4512 if (head_skb->destructor == sock_wfree) {
4513 swap(tail->truesize, head_skb->truesize);
4514 swap(tail->destructor, head_skb->destructor);
4515 swap(tail->sk, head_skb->sk);
4520 kfree_skb_list(segs);
4521 return ERR_PTR(err);
4523 EXPORT_SYMBOL_GPL(skb_segment);
4525 #ifdef CONFIG_SKB_EXTENSIONS
4526 #define SKB_EXT_ALIGN_VALUE 8
4527 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4529 static const u8 skb_ext_type_len[] = {
4530 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4531 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4534 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4536 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4537 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4539 #if IS_ENABLED(CONFIG_MPTCP)
4540 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4542 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4543 [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
4547 static __always_inline unsigned int skb_ext_total_length(void)
4549 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4550 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4551 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4554 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4556 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4557 skb_ext_type_len[TC_SKB_EXT] +
4559 #if IS_ENABLED(CONFIG_MPTCP)
4560 skb_ext_type_len[SKB_EXT_MPTCP] +
4562 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4563 skb_ext_type_len[SKB_EXT_MCTP] +
4568 static void skb_extensions_init(void)
4570 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4571 BUILD_BUG_ON(skb_ext_total_length() > 255);
4573 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4574 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4576 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4580 static void skb_extensions_init(void) {}
4583 void __init skb_init(void)
4585 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4586 sizeof(struct sk_buff),
4588 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4589 offsetof(struct sk_buff, cb),
4590 sizeof_field(struct sk_buff, cb),
4592 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4593 sizeof(struct sk_buff_fclones),
4595 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4597 skb_extensions_init();
4601 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4602 unsigned int recursion_level)
4604 int start = skb_headlen(skb);
4605 int i, copy = start - offset;
4606 struct sk_buff *frag_iter;
4609 if (unlikely(recursion_level >= 24))
4615 sg_set_buf(sg, skb->data + offset, copy);
4617 if ((len -= copy) == 0)
4622 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4625 WARN_ON(start > offset + len);
4627 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4628 if ((copy = end - offset) > 0) {
4629 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4630 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4635 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4636 skb_frag_off(frag) + offset - start);
4645 skb_walk_frags(skb, frag_iter) {
4648 WARN_ON(start > offset + len);
4650 end = start + frag_iter->len;
4651 if ((copy = end - offset) > 0) {
4652 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4657 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4658 copy, recursion_level + 1);
4659 if (unlikely(ret < 0))
4662 if ((len -= copy) == 0)
4673 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4674 * @skb: Socket buffer containing the buffers to be mapped
4675 * @sg: The scatter-gather list to map into
4676 * @offset: The offset into the buffer's contents to start mapping
4677 * @len: Length of buffer space to be mapped
4679 * Fill the specified scatter-gather list with mappings/pointers into a
4680 * region of the buffer space attached to a socket buffer. Returns either
4681 * the number of scatterlist items used, or -EMSGSIZE if the contents
4684 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4686 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4691 sg_mark_end(&sg[nsg - 1]);
4695 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4697 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4698 * sglist without mark the sg which contain last skb data as the end.
4699 * So the caller can mannipulate sg list as will when padding new data after
4700 * the first call without calling sg_unmark_end to expend sg list.
4702 * Scenario to use skb_to_sgvec_nomark:
4704 * 2. skb_to_sgvec_nomark(payload1)
4705 * 3. skb_to_sgvec_nomark(payload2)
4707 * This is equivalent to:
4709 * 2. skb_to_sgvec(payload1)
4711 * 4. skb_to_sgvec(payload2)
4713 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4714 * is more preferable.
4716 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4717 int offset, int len)
4719 return __skb_to_sgvec(skb, sg, offset, len, 0);
4721 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4726 * skb_cow_data - Check that a socket buffer's data buffers are writable
4727 * @skb: The socket buffer to check.
4728 * @tailbits: Amount of trailing space to be added
4729 * @trailer: Returned pointer to the skb where the @tailbits space begins
4731 * Make sure that the data buffers attached to a socket buffer are
4732 * writable. If they are not, private copies are made of the data buffers
4733 * and the socket buffer is set to use these instead.
4735 * If @tailbits is given, make sure that there is space to write @tailbits
4736 * bytes of data beyond current end of socket buffer. @trailer will be
4737 * set to point to the skb in which this space begins.
4739 * The number of scatterlist elements required to completely map the
4740 * COW'd and extended socket buffer will be returned.
4742 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4746 struct sk_buff *skb1, **skb_p;
4748 /* If skb is cloned or its head is paged, reallocate
4749 * head pulling out all the pages (pages are considered not writable
4750 * at the moment even if they are anonymous).
4752 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4753 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4756 /* Easy case. Most of packets will go this way. */
4757 if (!skb_has_frag_list(skb)) {
4758 /* A little of trouble, not enough of space for trailer.
4759 * This should not happen, when stack is tuned to generate
4760 * good frames. OK, on miss we reallocate and reserve even more
4761 * space, 128 bytes is fair. */
4763 if (skb_tailroom(skb) < tailbits &&
4764 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4772 /* Misery. We are in troubles, going to mincer fragments... */
4775 skb_p = &skb_shinfo(skb)->frag_list;
4778 while ((skb1 = *skb_p) != NULL) {
4781 /* The fragment is partially pulled by someone,
4782 * this can happen on input. Copy it and everything
4785 if (skb_shared(skb1))
4788 /* If the skb is the last, worry about trailer. */
4790 if (skb1->next == NULL && tailbits) {
4791 if (skb_shinfo(skb1)->nr_frags ||
4792 skb_has_frag_list(skb1) ||
4793 skb_tailroom(skb1) < tailbits)
4794 ntail = tailbits + 128;
4800 skb_shinfo(skb1)->nr_frags ||
4801 skb_has_frag_list(skb1)) {
4802 struct sk_buff *skb2;
4804 /* Fuck, we are miserable poor guys... */
4806 skb2 = skb_copy(skb1, GFP_ATOMIC);
4808 skb2 = skb_copy_expand(skb1,
4812 if (unlikely(skb2 == NULL))
4816 skb_set_owner_w(skb2, skb1->sk);
4818 /* Looking around. Are we still alive?
4819 * OK, link new skb, drop old one */
4821 skb2->next = skb1->next;
4828 skb_p = &skb1->next;
4833 EXPORT_SYMBOL_GPL(skb_cow_data);
4835 static void sock_rmem_free(struct sk_buff *skb)
4837 struct sock *sk = skb->sk;
4839 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4842 static void skb_set_err_queue(struct sk_buff *skb)
4844 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4845 * So, it is safe to (mis)use it to mark skbs on the error queue.
4847 skb->pkt_type = PACKET_OUTGOING;
4848 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4852 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4854 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4856 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4857 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4862 skb->destructor = sock_rmem_free;
4863 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4864 skb_set_err_queue(skb);
4866 /* before exiting rcu section, make sure dst is refcounted */
4869 skb_queue_tail(&sk->sk_error_queue, skb);
4870 if (!sock_flag(sk, SOCK_DEAD))
4871 sk_error_report(sk);
4874 EXPORT_SYMBOL(sock_queue_err_skb);
4876 static bool is_icmp_err_skb(const struct sk_buff *skb)
4878 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4879 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4882 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4884 struct sk_buff_head *q = &sk->sk_error_queue;
4885 struct sk_buff *skb, *skb_next = NULL;
4886 bool icmp_next = false;
4887 unsigned long flags;
4889 spin_lock_irqsave(&q->lock, flags);
4890 skb = __skb_dequeue(q);
4891 if (skb && (skb_next = skb_peek(q))) {
4892 icmp_next = is_icmp_err_skb(skb_next);
4894 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4896 spin_unlock_irqrestore(&q->lock, flags);
4898 if (is_icmp_err_skb(skb) && !icmp_next)
4902 sk_error_report(sk);
4906 EXPORT_SYMBOL(sock_dequeue_err_skb);
4909 * skb_clone_sk - create clone of skb, and take reference to socket
4910 * @skb: the skb to clone
4912 * This function creates a clone of a buffer that holds a reference on
4913 * sk_refcnt. Buffers created via this function are meant to be
4914 * returned using sock_queue_err_skb, or free via kfree_skb.
4916 * When passing buffers allocated with this function to sock_queue_err_skb
4917 * it is necessary to wrap the call with sock_hold/sock_put in order to
4918 * prevent the socket from being released prior to being enqueued on
4919 * the sk_error_queue.
4921 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4923 struct sock *sk = skb->sk;
4924 struct sk_buff *clone;
4926 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4929 clone = skb_clone(skb, GFP_ATOMIC);
4936 clone->destructor = sock_efree;
4940 EXPORT_SYMBOL(skb_clone_sk);
4942 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4947 struct sock_exterr_skb *serr;
4950 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4952 serr = SKB_EXT_ERR(skb);
4953 memset(serr, 0, sizeof(*serr));
4954 serr->ee.ee_errno = ENOMSG;
4955 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4956 serr->ee.ee_info = tstype;
4957 serr->opt_stats = opt_stats;
4958 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4959 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4960 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4962 serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
4965 err = sock_queue_err_skb(sk, skb);
4971 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4975 if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
4978 read_lock_bh(&sk->sk_callback_lock);
4979 ret = sk->sk_socket && sk->sk_socket->file &&
4980 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4981 read_unlock_bh(&sk->sk_callback_lock);
4985 void skb_complete_tx_timestamp(struct sk_buff *skb,
4986 struct skb_shared_hwtstamps *hwtstamps)
4988 struct sock *sk = skb->sk;
4990 if (!skb_may_tx_timestamp(sk, false))
4993 /* Take a reference to prevent skb_orphan() from freeing the socket,
4994 * but only if the socket refcount is not zero.
4996 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4997 *skb_hwtstamps(skb) = *hwtstamps;
4998 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5006 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5008 void __skb_tstamp_tx(struct sk_buff *orig_skb,
5009 const struct sk_buff *ack_skb,
5010 struct skb_shared_hwtstamps *hwtstamps,
5011 struct sock *sk, int tstype)
5013 struct sk_buff *skb;
5014 bool tsonly, opt_stats = false;
5019 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5020 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5023 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5024 if (!skb_may_tx_timestamp(sk, tsonly))
5029 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5031 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5036 skb = alloc_skb(0, GFP_ATOMIC);
5038 skb = skb_clone(orig_skb, GFP_ATOMIC);
5044 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5046 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5050 *skb_hwtstamps(skb) = *hwtstamps;
5052 __net_timestamp(skb);
5054 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5056 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5058 void skb_tstamp_tx(struct sk_buff *orig_skb,
5059 struct skb_shared_hwtstamps *hwtstamps)
5061 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5064 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5066 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5068 struct sock *sk = skb->sk;
5069 struct sock_exterr_skb *serr;
5072 skb->wifi_acked_valid = 1;
5073 skb->wifi_acked = acked;
5075 serr = SKB_EXT_ERR(skb);
5076 memset(serr, 0, sizeof(*serr));
5077 serr->ee.ee_errno = ENOMSG;
5078 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5080 /* Take a reference to prevent skb_orphan() from freeing the socket,
5081 * but only if the socket refcount is not zero.
5083 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5084 err = sock_queue_err_skb(sk, skb);
5090 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5093 * skb_partial_csum_set - set up and verify partial csum values for packet
5094 * @skb: the skb to set
5095 * @start: the number of bytes after skb->data to start checksumming.
5096 * @off: the offset from start to place the checksum.
5098 * For untrusted partially-checksummed packets, we need to make sure the values
5099 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5101 * This function checks and sets those values and skb->ip_summed: if this
5102 * returns false you should drop the packet.
5104 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5106 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5107 u32 csum_start = skb_headroom(skb) + (u32)start;
5109 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
5110 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5111 start, off, skb_headroom(skb), skb_headlen(skb));
5114 skb->ip_summed = CHECKSUM_PARTIAL;
5115 skb->csum_start = csum_start;
5116 skb->csum_offset = off;
5117 skb_set_transport_header(skb, start);
5120 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5122 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5125 if (skb_headlen(skb) >= len)
5128 /* If we need to pullup then pullup to the max, so we
5129 * won't need to do it again.
5134 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5137 if (skb_headlen(skb) < len)
5143 #define MAX_TCP_HDR_LEN (15 * 4)
5145 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5146 typeof(IPPROTO_IP) proto,
5153 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5154 off + MAX_TCP_HDR_LEN);
5155 if (!err && !skb_partial_csum_set(skb, off,
5156 offsetof(struct tcphdr,
5159 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5162 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5163 off + sizeof(struct udphdr));
5164 if (!err && !skb_partial_csum_set(skb, off,
5165 offsetof(struct udphdr,
5168 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5171 return ERR_PTR(-EPROTO);
5174 /* This value should be large enough to cover a tagged ethernet header plus
5175 * maximally sized IP and TCP or UDP headers.
5177 #define MAX_IP_HDR_LEN 128
5179 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5188 err = skb_maybe_pull_tail(skb,
5189 sizeof(struct iphdr),
5194 if (ip_is_fragment(ip_hdr(skb)))
5197 off = ip_hdrlen(skb);
5204 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5206 return PTR_ERR(csum);
5209 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5212 ip_hdr(skb)->protocol, 0);
5219 /* This value should be large enough to cover a tagged ethernet header plus
5220 * an IPv6 header, all options, and a maximal TCP or UDP header.
5222 #define MAX_IPV6_HDR_LEN 256
5224 #define OPT_HDR(type, skb, off) \
5225 (type *)(skb_network_header(skb) + (off))
5227 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5240 off = sizeof(struct ipv6hdr);
5242 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5246 nexthdr = ipv6_hdr(skb)->nexthdr;
5248 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5249 while (off <= len && !done) {
5251 case IPPROTO_DSTOPTS:
5252 case IPPROTO_HOPOPTS:
5253 case IPPROTO_ROUTING: {
5254 struct ipv6_opt_hdr *hp;
5256 err = skb_maybe_pull_tail(skb,
5258 sizeof(struct ipv6_opt_hdr),
5263 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5264 nexthdr = hp->nexthdr;
5265 off += ipv6_optlen(hp);
5269 struct ip_auth_hdr *hp;
5271 err = skb_maybe_pull_tail(skb,
5273 sizeof(struct ip_auth_hdr),
5278 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5279 nexthdr = hp->nexthdr;
5280 off += ipv6_authlen(hp);
5283 case IPPROTO_FRAGMENT: {
5284 struct frag_hdr *hp;
5286 err = skb_maybe_pull_tail(skb,
5288 sizeof(struct frag_hdr),
5293 hp = OPT_HDR(struct frag_hdr, skb, off);
5295 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5298 nexthdr = hp->nexthdr;
5299 off += sizeof(struct frag_hdr);
5310 if (!done || fragment)
5313 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5315 return PTR_ERR(csum);
5318 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5319 &ipv6_hdr(skb)->daddr,
5320 skb->len - off, nexthdr, 0);
5328 * skb_checksum_setup - set up partial checksum offset
5329 * @skb: the skb to set up
5330 * @recalculate: if true the pseudo-header checksum will be recalculated
5332 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5336 switch (skb->protocol) {
5337 case htons(ETH_P_IP):
5338 err = skb_checksum_setup_ipv4(skb, recalculate);
5341 case htons(ETH_P_IPV6):
5342 err = skb_checksum_setup_ipv6(skb, recalculate);
5352 EXPORT_SYMBOL(skb_checksum_setup);
5355 * skb_checksum_maybe_trim - maybe trims the given skb
5356 * @skb: the skb to check
5357 * @transport_len: the data length beyond the network header
5359 * Checks whether the given skb has data beyond the given transport length.
5360 * If so, returns a cloned skb trimmed to this transport length.
5361 * Otherwise returns the provided skb. Returns NULL in error cases
5362 * (e.g. transport_len exceeds skb length or out-of-memory).
5364 * Caller needs to set the skb transport header and free any returned skb if it
5365 * differs from the provided skb.
5367 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5368 unsigned int transport_len)
5370 struct sk_buff *skb_chk;
5371 unsigned int len = skb_transport_offset(skb) + transport_len;
5376 else if (skb->len == len)
5379 skb_chk = skb_clone(skb, GFP_ATOMIC);
5383 ret = pskb_trim_rcsum(skb_chk, len);
5393 * skb_checksum_trimmed - validate checksum of an skb
5394 * @skb: the skb to check
5395 * @transport_len: the data length beyond the network header
5396 * @skb_chkf: checksum function to use
5398 * Applies the given checksum function skb_chkf to the provided skb.
5399 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5401 * If the skb has data beyond the given transport length, then a
5402 * trimmed & cloned skb is checked and returned.
5404 * Caller needs to set the skb transport header and free any returned skb if it
5405 * differs from the provided skb.
5407 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5408 unsigned int transport_len,
5409 __sum16(*skb_chkf)(struct sk_buff *skb))
5411 struct sk_buff *skb_chk;
5412 unsigned int offset = skb_transport_offset(skb);
5415 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5419 if (!pskb_may_pull(skb_chk, offset))
5422 skb_pull_rcsum(skb_chk, offset);
5423 ret = skb_chkf(skb_chk);
5424 skb_push_rcsum(skb_chk, offset);
5432 if (skb_chk && skb_chk != skb)
5438 EXPORT_SYMBOL(skb_checksum_trimmed);
5440 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5442 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5445 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5447 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5450 skb_release_head_state(skb);
5451 kmem_cache_free(skbuff_head_cache, skb);
5456 EXPORT_SYMBOL(kfree_skb_partial);
5459 * skb_try_coalesce - try to merge skb to prior one
5461 * @from: buffer to add
5462 * @fragstolen: pointer to boolean
5463 * @delta_truesize: how much more was allocated than was requested
5465 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5466 bool *fragstolen, int *delta_truesize)
5468 struct skb_shared_info *to_shinfo, *from_shinfo;
5469 int i, delta, len = from->len;
5471 *fragstolen = false;
5476 /* In general, avoid mixing slab allocated and page_pool allocated
5477 * pages within the same SKB. However when @to is not pp_recycle and
5478 * @from is cloned, we can transition frag pages from page_pool to
5479 * reference counted.
5481 * On the other hand, don't allow coalescing two pp_recycle SKBs if
5482 * @from is cloned, in case the SKB is using page_pool fragment
5483 * references (PP_FLAG_PAGE_FRAG). Since we only take full page
5484 * references for cloned SKBs at the moment that would result in
5485 * inconsistent reference counts.
5487 if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
5490 if (len <= skb_tailroom(to)) {
5492 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5493 *delta_truesize = 0;
5497 to_shinfo = skb_shinfo(to);
5498 from_shinfo = skb_shinfo(from);
5499 if (to_shinfo->frag_list || from_shinfo->frag_list)
5501 if (skb_zcopy(to) || skb_zcopy(from))
5504 if (skb_headlen(from) != 0) {
5506 unsigned int offset;
5508 if (to_shinfo->nr_frags +
5509 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5512 if (skb_head_is_locked(from))
5515 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5517 page = virt_to_head_page(from->head);
5518 offset = from->data - (unsigned char *)page_address(page);
5520 skb_fill_page_desc(to, to_shinfo->nr_frags,
5521 page, offset, skb_headlen(from));
5524 if (to_shinfo->nr_frags +
5525 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5528 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5531 WARN_ON_ONCE(delta < len);
5533 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5535 from_shinfo->nr_frags * sizeof(skb_frag_t));
5536 to_shinfo->nr_frags += from_shinfo->nr_frags;
5538 if (!skb_cloned(from))
5539 from_shinfo->nr_frags = 0;
5541 /* if the skb is not cloned this does nothing
5542 * since we set nr_frags to 0.
5544 for (i = 0; i < from_shinfo->nr_frags; i++)
5545 __skb_frag_ref(&from_shinfo->frags[i]);
5547 to->truesize += delta;
5549 to->data_len += len;
5551 *delta_truesize = delta;
5554 EXPORT_SYMBOL(skb_try_coalesce);
5557 * skb_scrub_packet - scrub an skb
5559 * @skb: buffer to clean
5560 * @xnet: packet is crossing netns
5562 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5563 * into/from a tunnel. Some information have to be cleared during these
5565 * skb_scrub_packet can also be used to clean a skb before injecting it in
5566 * another namespace (@xnet == true). We have to clear all information in the
5567 * skb that could impact namespace isolation.
5569 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5571 skb->pkt_type = PACKET_HOST;
5577 nf_reset_trace(skb);
5579 #ifdef CONFIG_NET_SWITCHDEV
5580 skb->offload_fwd_mark = 0;
5581 skb->offload_l3_fwd_mark = 0;
5589 skb_clear_tstamp(skb);
5591 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5594 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5598 * skb_gso_transport_seglen is used to determine the real size of the
5599 * individual segments, including Layer4 headers (TCP/UDP).
5601 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5603 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5605 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5606 unsigned int thlen = 0;
5608 if (skb->encapsulation) {
5609 thlen = skb_inner_transport_header(skb) -
5610 skb_transport_header(skb);
5612 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5613 thlen += inner_tcp_hdrlen(skb);
5614 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5615 thlen = tcp_hdrlen(skb);
5616 } else if (unlikely(skb_is_gso_sctp(skb))) {
5617 thlen = sizeof(struct sctphdr);
5618 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5619 thlen = sizeof(struct udphdr);
5621 /* UFO sets gso_size to the size of the fragmentation
5622 * payload, i.e. the size of the L4 (UDP) header is already
5625 return thlen + shinfo->gso_size;
5629 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5633 * skb_gso_network_seglen is used to determine the real size of the
5634 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5636 * The MAC/L2 header is not accounted for.
5638 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5640 unsigned int hdr_len = skb_transport_header(skb) -
5641 skb_network_header(skb);
5643 return hdr_len + skb_gso_transport_seglen(skb);
5647 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5651 * skb_gso_mac_seglen is used to determine the real size of the
5652 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5653 * headers (TCP/UDP).
5655 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5657 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5659 return hdr_len + skb_gso_transport_seglen(skb);
5663 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5665 * There are a couple of instances where we have a GSO skb, and we
5666 * want to determine what size it would be after it is segmented.
5668 * We might want to check:
5669 * - L3+L4+payload size (e.g. IP forwarding)
5670 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5672 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5676 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5677 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5679 * @max_len: The maximum permissible length.
5681 * Returns true if the segmented length <= max length.
5683 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5684 unsigned int seg_len,
5685 unsigned int max_len) {
5686 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5687 const struct sk_buff *iter;
5689 if (shinfo->gso_size != GSO_BY_FRAGS)
5690 return seg_len <= max_len;
5692 /* Undo this so we can re-use header sizes */
5693 seg_len -= GSO_BY_FRAGS;
5695 skb_walk_frags(skb, iter) {
5696 if (seg_len + skb_headlen(iter) > max_len)
5704 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5707 * @mtu: MTU to validate against
5709 * skb_gso_validate_network_len validates if a given skb will fit a
5710 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5713 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5715 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5717 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5720 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5723 * @len: length to validate against
5725 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5726 * length once split, including L2, L3 and L4 headers and the payload.
5728 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5730 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5732 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5734 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5736 int mac_len, meta_len;
5739 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5744 mac_len = skb->data - skb_mac_header(skb);
5745 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5746 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5747 mac_len - VLAN_HLEN - ETH_TLEN);
5750 meta_len = skb_metadata_len(skb);
5752 meta = skb_metadata_end(skb) - meta_len;
5753 memmove(meta + VLAN_HLEN, meta, meta_len);
5756 skb->mac_header += VLAN_HLEN;
5760 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5762 struct vlan_hdr *vhdr;
5765 if (unlikely(skb_vlan_tag_present(skb))) {
5766 /* vlan_tci is already set-up so leave this for another time */
5770 skb = skb_share_check(skb, GFP_ATOMIC);
5773 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5774 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5777 vhdr = (struct vlan_hdr *)skb->data;
5778 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5779 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5781 skb_pull_rcsum(skb, VLAN_HLEN);
5782 vlan_set_encap_proto(skb, vhdr);
5784 skb = skb_reorder_vlan_header(skb);
5788 skb_reset_network_header(skb);
5789 if (!skb_transport_header_was_set(skb))
5790 skb_reset_transport_header(skb);
5791 skb_reset_mac_len(skb);
5799 EXPORT_SYMBOL(skb_vlan_untag);
5801 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
5803 if (!pskb_may_pull(skb, write_len))
5806 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5809 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5811 EXPORT_SYMBOL(skb_ensure_writable);
5813 /* remove VLAN header from packet and update csum accordingly.
5814 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5816 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5818 struct vlan_hdr *vhdr;
5819 int offset = skb->data - skb_mac_header(skb);
5822 if (WARN_ONCE(offset,
5823 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5828 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5832 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5834 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5835 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5837 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5838 __skb_pull(skb, VLAN_HLEN);
5840 vlan_set_encap_proto(skb, vhdr);
5841 skb->mac_header += VLAN_HLEN;
5843 if (skb_network_offset(skb) < ETH_HLEN)
5844 skb_set_network_header(skb, ETH_HLEN);
5846 skb_reset_mac_len(skb);
5850 EXPORT_SYMBOL(__skb_vlan_pop);
5852 /* Pop a vlan tag either from hwaccel or from payload.
5853 * Expects skb->data at mac header.
5855 int skb_vlan_pop(struct sk_buff *skb)
5861 if (likely(skb_vlan_tag_present(skb))) {
5862 __vlan_hwaccel_clear_tag(skb);
5864 if (unlikely(!eth_type_vlan(skb->protocol)))
5867 err = __skb_vlan_pop(skb, &vlan_tci);
5871 /* move next vlan tag to hw accel tag */
5872 if (likely(!eth_type_vlan(skb->protocol)))
5875 vlan_proto = skb->protocol;
5876 err = __skb_vlan_pop(skb, &vlan_tci);
5880 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5883 EXPORT_SYMBOL(skb_vlan_pop);
5885 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5886 * Expects skb->data at mac header.
5888 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5890 if (skb_vlan_tag_present(skb)) {
5891 int offset = skb->data - skb_mac_header(skb);
5894 if (WARN_ONCE(offset,
5895 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5900 err = __vlan_insert_tag(skb, skb->vlan_proto,
5901 skb_vlan_tag_get(skb));
5905 skb->protocol = skb->vlan_proto;
5906 skb->mac_len += VLAN_HLEN;
5908 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5910 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5913 EXPORT_SYMBOL(skb_vlan_push);
5916 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5918 * @skb: Socket buffer to modify
5920 * Drop the Ethernet header of @skb.
5922 * Expects that skb->data points to the mac header and that no VLAN tags are
5925 * Returns 0 on success, -errno otherwise.
5927 int skb_eth_pop(struct sk_buff *skb)
5929 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5930 skb_network_offset(skb) < ETH_HLEN)
5933 skb_pull_rcsum(skb, ETH_HLEN);
5934 skb_reset_mac_header(skb);
5935 skb_reset_mac_len(skb);
5939 EXPORT_SYMBOL(skb_eth_pop);
5942 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5944 * @skb: Socket buffer to modify
5945 * @dst: Destination MAC address of the new header
5946 * @src: Source MAC address of the new header
5948 * Prepend @skb with a new Ethernet header.
5950 * Expects that skb->data points to the mac header, which must be empty.
5952 * Returns 0 on success, -errno otherwise.
5954 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5955 const unsigned char *src)
5960 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5963 err = skb_cow_head(skb, sizeof(*eth));
5967 skb_push(skb, sizeof(*eth));
5968 skb_reset_mac_header(skb);
5969 skb_reset_mac_len(skb);
5972 ether_addr_copy(eth->h_dest, dst);
5973 ether_addr_copy(eth->h_source, src);
5974 eth->h_proto = skb->protocol;
5976 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5980 EXPORT_SYMBOL(skb_eth_push);
5982 /* Update the ethertype of hdr and the skb csum value if required. */
5983 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5986 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5987 __be16 diff[] = { ~hdr->h_proto, ethertype };
5989 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5992 hdr->h_proto = ethertype;
5996 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6000 * @mpls_lse: MPLS label stack entry to push
6001 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6002 * @mac_len: length of the MAC header
6003 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6006 * Expects skb->data at mac header.
6008 * Returns 0 on success, -errno otherwise.
6010 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6011 int mac_len, bool ethernet)
6013 struct mpls_shim_hdr *lse;
6016 if (unlikely(!eth_p_mpls(mpls_proto)))
6019 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6020 if (skb->encapsulation)
6023 err = skb_cow_head(skb, MPLS_HLEN);
6027 if (!skb->inner_protocol) {
6028 skb_set_inner_network_header(skb, skb_network_offset(skb));
6029 skb_set_inner_protocol(skb, skb->protocol);
6032 skb_push(skb, MPLS_HLEN);
6033 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6035 skb_reset_mac_header(skb);
6036 skb_set_network_header(skb, mac_len);
6037 skb_reset_mac_len(skb);
6039 lse = mpls_hdr(skb);
6040 lse->label_stack_entry = mpls_lse;
6041 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6043 if (ethernet && mac_len >= ETH_HLEN)
6044 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6045 skb->protocol = mpls_proto;
6049 EXPORT_SYMBOL_GPL(skb_mpls_push);
6052 * skb_mpls_pop() - pop the outermost MPLS header
6055 * @next_proto: ethertype of header after popped MPLS header
6056 * @mac_len: length of the MAC header
6057 * @ethernet: flag to indicate if the packet is ethernet
6059 * Expects skb->data at mac header.
6061 * Returns 0 on success, -errno otherwise.
6063 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6068 if (unlikely(!eth_p_mpls(skb->protocol)))
6071 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6075 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6076 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6079 __skb_pull(skb, MPLS_HLEN);
6080 skb_reset_mac_header(skb);
6081 skb_set_network_header(skb, mac_len);
6083 if (ethernet && mac_len >= ETH_HLEN) {
6086 /* use mpls_hdr() to get ethertype to account for VLANs. */
6087 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6088 skb_mod_eth_type(skb, hdr, next_proto);
6090 skb->protocol = next_proto;
6094 EXPORT_SYMBOL_GPL(skb_mpls_pop);
6097 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6100 * @mpls_lse: new MPLS label stack entry to update to
6102 * Expects skb->data at mac header.
6104 * Returns 0 on success, -errno otherwise.
6106 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6110 if (unlikely(!eth_p_mpls(skb->protocol)))
6113 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6117 if (skb->ip_summed == CHECKSUM_COMPLETE) {
6118 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6120 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6123 mpls_hdr(skb)->label_stack_entry = mpls_lse;
6127 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6130 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6134 * Expects skb->data at mac header.
6136 * Returns 0 on success, -errno otherwise.
6138 int skb_mpls_dec_ttl(struct sk_buff *skb)
6143 if (unlikely(!eth_p_mpls(skb->protocol)))
6146 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6149 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6150 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6154 lse &= ~MPLS_LS_TTL_MASK;
6155 lse |= ttl << MPLS_LS_TTL_SHIFT;
6157 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6159 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6162 * alloc_skb_with_frags - allocate skb with page frags
6164 * @header_len: size of linear part
6165 * @data_len: needed length in frags
6166 * @max_page_order: max page order desired.
6167 * @errcode: pointer to error code if any
6168 * @gfp_mask: allocation mask
6170 * This can be used to allocate a paged skb, given a maximal order for frags.
6172 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6173 unsigned long data_len,
6178 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
6179 unsigned long chunk;
6180 struct sk_buff *skb;
6184 *errcode = -EMSGSIZE;
6185 /* Note this test could be relaxed, if we succeed to allocate
6186 * high order pages...
6188 if (npages > MAX_SKB_FRAGS)
6191 *errcode = -ENOBUFS;
6192 skb = alloc_skb(header_len, gfp_mask);
6196 skb->truesize += npages << PAGE_SHIFT;
6198 for (i = 0; npages > 0; i++) {
6199 int order = max_page_order;
6202 if (npages >= 1 << order) {
6203 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6209 /* Do not retry other high order allocations */
6215 page = alloc_page(gfp_mask);
6219 chunk = min_t(unsigned long, data_len,
6220 PAGE_SIZE << order);
6221 skb_fill_page_desc(skb, i, page, 0, chunk);
6223 npages -= 1 << order;
6231 EXPORT_SYMBOL(alloc_skb_with_frags);
6233 /* carve out the first off bytes from skb when off < headlen */
6234 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6235 const int headlen, gfp_t gfp_mask)
6238 unsigned int size = skb_end_offset(skb);
6239 int new_hlen = headlen - off;
6242 if (skb_pfmemalloc(skb))
6243 gfp_mask |= __GFP_MEMALLOC;
6245 size = SKB_DATA_ALIGN(size);
6246 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
6247 size = kmalloc_size_roundup(size);
6248 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
6251 size = SKB_WITH_OVERHEAD(size);
6253 /* Copy real data, and all frags */
6254 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6257 memcpy((struct skb_shared_info *)(data + size),
6259 offsetof(struct skb_shared_info,
6260 frags[skb_shinfo(skb)->nr_frags]));
6261 if (skb_cloned(skb)) {
6262 /* drop the old head gracefully */
6263 if (skb_orphan_frags(skb, gfp_mask)) {
6267 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6268 skb_frag_ref(skb, i);
6269 if (skb_has_frag_list(skb))
6270 skb_clone_fraglist(skb);
6271 skb_release_data(skb, SKB_CONSUMED);
6273 /* we can reuse existing recount- all we did was
6282 skb_set_end_offset(skb, size);
6283 skb_set_tail_pointer(skb, skb_headlen(skb));
6284 skb_headers_offset_update(skb, 0);
6288 atomic_set(&skb_shinfo(skb)->dataref, 1);
6293 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6295 /* carve out the first eat bytes from skb's frag_list. May recurse into
6298 static int pskb_carve_frag_list(struct sk_buff *skb,
6299 struct skb_shared_info *shinfo, int eat,
6302 struct sk_buff *list = shinfo->frag_list;
6303 struct sk_buff *clone = NULL;
6304 struct sk_buff *insp = NULL;
6308 pr_err("Not enough bytes to eat. Want %d\n", eat);
6311 if (list->len <= eat) {
6312 /* Eaten as whole. */
6317 /* Eaten partially. */
6318 if (skb_shared(list)) {
6319 clone = skb_clone(list, gfp_mask);
6325 /* This may be pulled without problems. */
6328 if (pskb_carve(list, eat, gfp_mask) < 0) {
6336 /* Free pulled out fragments. */
6337 while ((list = shinfo->frag_list) != insp) {
6338 shinfo->frag_list = list->next;
6341 /* And insert new clone at head. */
6344 shinfo->frag_list = clone;
6349 /* carve off first len bytes from skb. Split line (off) is in the
6350 * non-linear part of skb
6352 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6353 int pos, gfp_t gfp_mask)
6356 unsigned int size = skb_end_offset(skb);
6358 const int nfrags = skb_shinfo(skb)->nr_frags;
6359 struct skb_shared_info *shinfo;
6361 if (skb_pfmemalloc(skb))
6362 gfp_mask |= __GFP_MEMALLOC;
6364 size = SKB_DATA_ALIGN(size);
6365 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
6366 size = kmalloc_size_roundup(size);
6367 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
6370 size = SKB_WITH_OVERHEAD(size);
6372 memcpy((struct skb_shared_info *)(data + size),
6373 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6374 if (skb_orphan_frags(skb, gfp_mask)) {
6378 shinfo = (struct skb_shared_info *)(data + size);
6379 for (i = 0; i < nfrags; i++) {
6380 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6382 if (pos + fsize > off) {
6383 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6387 * We have two variants in this case:
6388 * 1. Move all the frag to the second
6389 * part, if it is possible. F.e.
6390 * this approach is mandatory for TUX,
6391 * where splitting is expensive.
6392 * 2. Split is accurately. We make this.
6394 skb_frag_off_add(&shinfo->frags[0], off - pos);
6395 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6397 skb_frag_ref(skb, i);
6402 shinfo->nr_frags = k;
6403 if (skb_has_frag_list(skb))
6404 skb_clone_fraglist(skb);
6406 /* split line is in frag list */
6407 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6408 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6409 if (skb_has_frag_list(skb))
6410 kfree_skb_list(skb_shinfo(skb)->frag_list);
6414 skb_release_data(skb, SKB_CONSUMED);
6419 skb_set_end_offset(skb, size);
6420 skb_reset_tail_pointer(skb);
6421 skb_headers_offset_update(skb, 0);
6426 skb->data_len = skb->len;
6427 atomic_set(&skb_shinfo(skb)->dataref, 1);
6431 /* remove len bytes from the beginning of the skb */
6432 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6434 int headlen = skb_headlen(skb);
6437 return pskb_carve_inside_header(skb, len, headlen, gfp);
6439 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6442 /* Extract to_copy bytes starting at off from skb, and return this in
6445 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6446 int to_copy, gfp_t gfp)
6448 struct sk_buff *clone = skb_clone(skb, gfp);
6453 if (pskb_carve(clone, off, gfp) < 0 ||
6454 pskb_trim(clone, to_copy)) {
6460 EXPORT_SYMBOL(pskb_extract);
6463 * skb_condense - try to get rid of fragments/frag_list if possible
6466 * Can be used to save memory before skb is added to a busy queue.
6467 * If packet has bytes in frags and enough tail room in skb->head,
6468 * pull all of them, so that we can free the frags right now and adjust
6471 * We do not reallocate skb->head thus can not fail.
6472 * Caller must re-evaluate skb->truesize if needed.
6474 void skb_condense(struct sk_buff *skb)
6476 if (skb->data_len) {
6477 if (skb->data_len > skb->end - skb->tail ||
6481 /* Nice, we can free page frag(s) right now */
6482 __pskb_pull_tail(skb, skb->data_len);
6484 /* At this point, skb->truesize might be over estimated,
6485 * because skb had a fragment, and fragments do not tell
6487 * When we pulled its content into skb->head, fragment
6488 * was freed, but __pskb_pull_tail() could not possibly
6489 * adjust skb->truesize, not knowing the frag truesize.
6491 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6493 EXPORT_SYMBOL(skb_condense);
6495 #ifdef CONFIG_SKB_EXTENSIONS
6496 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6498 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6502 * __skb_ext_alloc - allocate a new skb extensions storage
6504 * @flags: See kmalloc().
6506 * Returns the newly allocated pointer. The pointer can later attached to a
6507 * skb via __skb_ext_set().
6508 * Note: caller must handle the skb_ext as an opaque data.
6510 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6512 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6515 memset(new->offset, 0, sizeof(new->offset));
6516 refcount_set(&new->refcnt, 1);
6522 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6523 unsigned int old_active)
6525 struct skb_ext *new;
6527 if (refcount_read(&old->refcnt) == 1)
6530 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6534 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6535 refcount_set(&new->refcnt, 1);
6538 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6539 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6542 for (i = 0; i < sp->len; i++)
6543 xfrm_state_hold(sp->xvec[i]);
6551 * __skb_ext_set - attach the specified extension storage to this skb
6554 * @ext: extension storage previously allocated via __skb_ext_alloc()
6556 * Existing extensions, if any, are cleared.
6558 * Returns the pointer to the extension.
6560 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6561 struct skb_ext *ext)
6563 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6566 newlen = newoff + skb_ext_type_len[id];
6567 ext->chunks = newlen;
6568 ext->offset[id] = newoff;
6569 skb->extensions = ext;
6570 skb->active_extensions = 1 << id;
6571 return skb_ext_get_ptr(ext, id);
6575 * skb_ext_add - allocate space for given extension, COW if needed
6577 * @id: extension to allocate space for
6579 * Allocates enough space for the given extension.
6580 * If the extension is already present, a pointer to that extension
6583 * If the skb was cloned, COW applies and the returned memory can be
6584 * modified without changing the extension space of clones buffers.
6586 * Returns pointer to the extension or NULL on allocation failure.
6588 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6590 struct skb_ext *new, *old = NULL;
6591 unsigned int newlen, newoff;
6593 if (skb->active_extensions) {
6594 old = skb->extensions;
6596 new = skb_ext_maybe_cow(old, skb->active_extensions);
6600 if (__skb_ext_exist(new, id))
6603 newoff = new->chunks;
6605 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6607 new = __skb_ext_alloc(GFP_ATOMIC);
6612 newlen = newoff + skb_ext_type_len[id];
6613 new->chunks = newlen;
6614 new->offset[id] = newoff;
6617 skb->extensions = new;
6618 skb->active_extensions |= 1 << id;
6619 return skb_ext_get_ptr(new, id);
6621 EXPORT_SYMBOL(skb_ext_add);
6624 static void skb_ext_put_sp(struct sec_path *sp)
6628 for (i = 0; i < sp->len; i++)
6629 xfrm_state_put(sp->xvec[i]);
6633 #ifdef CONFIG_MCTP_FLOWS
6634 static void skb_ext_put_mctp(struct mctp_flow *flow)
6637 mctp_key_unref(flow->key);
6641 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6643 struct skb_ext *ext = skb->extensions;
6645 skb->active_extensions &= ~(1 << id);
6646 if (skb->active_extensions == 0) {
6647 skb->extensions = NULL;
6650 } else if (id == SKB_EXT_SEC_PATH &&
6651 refcount_read(&ext->refcnt) == 1) {
6652 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6659 EXPORT_SYMBOL(__skb_ext_del);
6661 void __skb_ext_put(struct skb_ext *ext)
6663 /* If this is last clone, nothing can increment
6664 * it after check passes. Avoids one atomic op.
6666 if (refcount_read(&ext->refcnt) == 1)
6669 if (!refcount_dec_and_test(&ext->refcnt))
6673 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6674 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6676 #ifdef CONFIG_MCTP_FLOWS
6677 if (__skb_ext_exist(ext, SKB_EXT_MCTP))
6678 skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
6681 kmem_cache_free(skbuff_ext_cache, ext);
6683 EXPORT_SYMBOL(__skb_ext_put);
6684 #endif /* CONFIG_SKB_EXTENSIONS */
6687 * skb_attempt_defer_free - queue skb for remote freeing
6690 * Put @skb in a per-cpu list, using the cpu which
6691 * allocated the skb/pages to reduce false sharing
6692 * and memory zone spinlock contention.
6694 void skb_attempt_defer_free(struct sk_buff *skb)
6696 int cpu = skb->alloc_cpu;
6697 struct softnet_data *sd;
6698 unsigned long flags;
6699 unsigned int defer_max;
6702 if (WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
6704 cpu == raw_smp_processor_id()) {
6705 nodefer: __kfree_skb(skb);
6709 sd = &per_cpu(softnet_data, cpu);
6710 defer_max = READ_ONCE(sysctl_skb_defer_max);
6711 if (READ_ONCE(sd->defer_count) >= defer_max)
6714 spin_lock_irqsave(&sd->defer_lock, flags);
6715 /* Send an IPI every time queue reaches half capacity. */
6716 kick = sd->defer_count == (defer_max >> 1);
6717 /* Paired with the READ_ONCE() few lines above */
6718 WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
6720 skb->next = sd->defer_list;
6721 /* Paired with READ_ONCE() in skb_defer_free_flush() */
6722 WRITE_ONCE(sd->defer_list, skb);
6723 spin_unlock_irqrestore(&sd->defer_lock, flags);
6725 /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
6726 * if we are unlucky enough (this seems very unlikely).
6728 if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1))
6729 smp_call_function_single_async(cpu, &sd->defer_csd);