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
390 struct sk_buff *build_skb(void *data, unsigned int frag_size)
392 struct sk_buff *skb = __build_skb(data, frag_size);
394 if (skb && frag_size) {
396 if (page_is_pfmemalloc(virt_to_head_page(data)))
401 EXPORT_SYMBOL(build_skb);
404 * build_skb_around - build a network buffer around provided skb
405 * @skb: sk_buff provide by caller, must be memset cleared
406 * @data: data buffer provided by caller
407 * @frag_size: size of data
409 struct sk_buff *build_skb_around(struct sk_buff *skb,
410 void *data, unsigned int frag_size)
415 __build_skb_around(skb, data, frag_size);
419 if (page_is_pfmemalloc(virt_to_head_page(data)))
424 EXPORT_SYMBOL(build_skb_around);
427 * __napi_build_skb - build a network buffer
428 * @data: data buffer provided by caller
429 * @frag_size: size of data
431 * Version of __build_skb() that uses NAPI percpu caches to obtain
432 * skbuff_head instead of inplace allocation.
434 * Returns a new &sk_buff on success, %NULL on allocation failure.
436 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
440 skb = napi_skb_cache_get();
444 memset(skb, 0, offsetof(struct sk_buff, tail));
445 __build_skb_around(skb, data, frag_size);
451 * napi_build_skb - build a network buffer
452 * @data: data buffer provided by caller
453 * @frag_size: size of data
455 * Version of __napi_build_skb() that takes care of skb->head_frag
456 * and skb->pfmemalloc when the data is a page or page fragment.
458 * Returns a new &sk_buff on success, %NULL on allocation failure.
460 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
462 struct sk_buff *skb = __napi_build_skb(data, frag_size);
464 if (likely(skb) && frag_size) {
466 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
471 EXPORT_SYMBOL(napi_build_skb);
474 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
475 * the caller if emergency pfmemalloc reserves are being used. If it is and
476 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
477 * may be used. Otherwise, the packet data may be discarded until enough
480 static void *kmalloc_reserve(size_t size, gfp_t flags, int node,
484 bool ret_pfmemalloc = false;
487 * Try a regular allocation, when that fails and we're not entitled
488 * to the reserves, fail.
490 obj = kmalloc_node_track_caller(size,
491 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
493 if (obj || !(gfp_pfmemalloc_allowed(flags)))
496 /* Try again but now we are using pfmemalloc reserves */
497 ret_pfmemalloc = true;
498 obj = kmalloc_node_track_caller(size, flags, node);
502 *pfmemalloc = ret_pfmemalloc;
507 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
508 * 'private' fields and also do memory statistics to find all the
514 * __alloc_skb - allocate a network buffer
515 * @size: size to allocate
516 * @gfp_mask: allocation mask
517 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
518 * instead of head cache and allocate a cloned (child) skb.
519 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
520 * allocations in case the data is required for writeback
521 * @node: numa node to allocate memory on
523 * Allocate a new &sk_buff. The returned buffer has no headroom and a
524 * tail room of at least size bytes. The object has a reference count
525 * of one. The return is the buffer. On a failure the return is %NULL.
527 * Buffers may only be allocated from interrupts using a @gfp_mask of
530 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
533 struct kmem_cache *cache;
539 cache = (flags & SKB_ALLOC_FCLONE)
540 ? skbuff_fclone_cache : skbuff_head_cache;
542 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
543 gfp_mask |= __GFP_MEMALLOC;
546 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
547 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
548 skb = napi_skb_cache_get();
550 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
555 /* We do our best to align skb_shared_info on a separate cache
556 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
557 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
558 * Both skb->head and skb_shared_info are cache line aligned.
560 size = SKB_DATA_ALIGN(size);
561 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
562 osize = kmalloc_size_roundup(size);
563 data = kmalloc_reserve(osize, gfp_mask, node, &pfmemalloc);
566 /* kmalloc_size_roundup() might give us more room than requested.
567 * Put skb_shared_info exactly at the end of allocated zone,
568 * to allow max possible filling before reallocation.
570 size = SKB_WITH_OVERHEAD(osize);
571 prefetchw(data + size);
574 * Only clear those fields we need to clear, not those that we will
575 * actually initialise below. Hence, don't put any more fields after
576 * the tail pointer in struct sk_buff!
578 memset(skb, 0, offsetof(struct sk_buff, tail));
579 __build_skb_around(skb, data, osize);
580 skb->pfmemalloc = pfmemalloc;
582 if (flags & SKB_ALLOC_FCLONE) {
583 struct sk_buff_fclones *fclones;
585 fclones = container_of(skb, struct sk_buff_fclones, skb1);
587 skb->fclone = SKB_FCLONE_ORIG;
588 refcount_set(&fclones->fclone_ref, 1);
594 kmem_cache_free(cache, skb);
597 EXPORT_SYMBOL(__alloc_skb);
600 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
601 * @dev: network device to receive on
602 * @len: length to allocate
603 * @gfp_mask: get_free_pages mask, passed to alloc_skb
605 * Allocate a new &sk_buff and assign it a usage count of one. The
606 * buffer has NET_SKB_PAD headroom built in. Users should allocate
607 * the headroom they think they need without accounting for the
608 * built in space. The built in space is used for optimisations.
610 * %NULL is returned if there is no free memory.
612 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
615 struct page_frag_cache *nc;
622 /* If requested length is either too small or too big,
623 * we use kmalloc() for skb->head allocation.
625 if (len <= SKB_WITH_OVERHEAD(1024) ||
626 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
627 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
628 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
634 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
635 len = SKB_DATA_ALIGN(len);
637 if (sk_memalloc_socks())
638 gfp_mask |= __GFP_MEMALLOC;
640 if (in_hardirq() || irqs_disabled()) {
641 nc = this_cpu_ptr(&netdev_alloc_cache);
642 data = page_frag_alloc(nc, len, gfp_mask);
643 pfmemalloc = nc->pfmemalloc;
646 nc = this_cpu_ptr(&napi_alloc_cache.page);
647 data = page_frag_alloc(nc, len, gfp_mask);
648 pfmemalloc = nc->pfmemalloc;
655 skb = __build_skb(data, len);
656 if (unlikely(!skb)) {
666 skb_reserve(skb, NET_SKB_PAD);
672 EXPORT_SYMBOL(__netdev_alloc_skb);
675 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
676 * @napi: napi instance this buffer was allocated for
677 * @len: length to allocate
678 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
680 * Allocate a new sk_buff for use in NAPI receive. This buffer will
681 * attempt to allocate the head from a special reserved region used
682 * only for NAPI Rx allocation. By doing this we can save several
683 * CPU cycles by avoiding having to disable and re-enable IRQs.
685 * %NULL is returned if there is no free memory.
687 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
690 struct napi_alloc_cache *nc;
695 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
696 len += NET_SKB_PAD + NET_IP_ALIGN;
698 /* If requested length is either too small or too big,
699 * we use kmalloc() for skb->head allocation.
700 * When the small frag allocator is available, prefer it over kmalloc
701 * for small fragments
703 if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
704 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
705 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
706 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
713 nc = this_cpu_ptr(&napi_alloc_cache);
715 if (sk_memalloc_socks())
716 gfp_mask |= __GFP_MEMALLOC;
718 if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
719 /* we are artificially inflating the allocation size, but
720 * that is not as bad as it may look like, as:
721 * - 'len' less than GRO_MAX_HEAD makes little sense
722 * - On most systems, larger 'len' values lead to fragment
723 * size above 512 bytes
724 * - kmalloc would use the kmalloc-1k slab for such values
725 * - Builds with smaller GRO_MAX_HEAD will very likely do
726 * little networking, as that implies no WiFi and no
727 * tunnels support, and 32 bits arches.
731 data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
732 pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
734 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
735 len = SKB_DATA_ALIGN(len);
737 data = page_frag_alloc(&nc->page, len, gfp_mask);
738 pfmemalloc = nc->page.pfmemalloc;
744 skb = __napi_build_skb(data, len);
745 if (unlikely(!skb)) {
755 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
756 skb->dev = napi->dev;
761 EXPORT_SYMBOL(__napi_alloc_skb);
763 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
764 int size, unsigned int truesize)
766 skb_fill_page_desc(skb, i, page, off, size);
768 skb->data_len += size;
769 skb->truesize += truesize;
771 EXPORT_SYMBOL(skb_add_rx_frag);
773 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
774 unsigned int truesize)
776 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
778 skb_frag_size_add(frag, size);
780 skb->data_len += size;
781 skb->truesize += truesize;
783 EXPORT_SYMBOL(skb_coalesce_rx_frag);
785 static void skb_drop_list(struct sk_buff **listp)
787 kfree_skb_list(*listp);
791 static inline void skb_drop_fraglist(struct sk_buff *skb)
793 skb_drop_list(&skb_shinfo(skb)->frag_list);
796 static void skb_clone_fraglist(struct sk_buff *skb)
798 struct sk_buff *list;
800 skb_walk_frags(skb, list)
804 static bool skb_pp_recycle(struct sk_buff *skb, void *data)
806 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
808 return page_pool_return_skb_page(virt_to_page(data));
811 static void skb_free_head(struct sk_buff *skb)
813 unsigned char *head = skb->head;
815 if (skb->head_frag) {
816 if (skb_pp_recycle(skb, head))
824 static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
826 struct skb_shared_info *shinfo = skb_shinfo(skb);
830 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
834 if (skb_zcopy(skb)) {
835 bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
837 skb_zcopy_clear(skb, true);
842 for (i = 0; i < shinfo->nr_frags; i++)
843 __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
846 if (shinfo->frag_list)
847 kfree_skb_list_reason(shinfo->frag_list, reason);
851 /* When we clone an SKB we copy the reycling bit. The pp_recycle
852 * bit is only set on the head though, so in order to avoid races
853 * while trying to recycle fragments on __skb_frag_unref() we need
854 * to make one SKB responsible for triggering the recycle path.
855 * So disable the recycling bit if an SKB is cloned and we have
856 * additional references to the fragmented part of the SKB.
857 * Eventually the last SKB will have the recycling bit set and it's
858 * dataref set to 0, which will trigger the recycling
864 * Free an skbuff by memory without cleaning the state.
866 static void kfree_skbmem(struct sk_buff *skb)
868 struct sk_buff_fclones *fclones;
870 switch (skb->fclone) {
871 case SKB_FCLONE_UNAVAILABLE:
872 kmem_cache_free(skbuff_head_cache, skb);
875 case SKB_FCLONE_ORIG:
876 fclones = container_of(skb, struct sk_buff_fclones, skb1);
878 /* We usually free the clone (TX completion) before original skb
879 * This test would have no chance to be true for the clone,
880 * while here, branch prediction will be good.
882 if (refcount_read(&fclones->fclone_ref) == 1)
886 default: /* SKB_FCLONE_CLONE */
887 fclones = container_of(skb, struct sk_buff_fclones, skb2);
890 if (!refcount_dec_and_test(&fclones->fclone_ref))
893 kmem_cache_free(skbuff_fclone_cache, fclones);
896 void skb_release_head_state(struct sk_buff *skb)
899 if (skb->destructor) {
900 DEBUG_NET_WARN_ON_ONCE(in_hardirq());
901 skb->destructor(skb);
903 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
904 nf_conntrack_put(skb_nfct(skb));
909 /* Free everything but the sk_buff shell. */
910 static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
912 skb_release_head_state(skb);
913 if (likely(skb->head))
914 skb_release_data(skb, reason);
918 * __kfree_skb - private function
921 * Free an sk_buff. Release anything attached to the buffer.
922 * Clean the state. This is an internal helper function. Users should
923 * always call kfree_skb
926 void __kfree_skb(struct sk_buff *skb)
928 skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
931 EXPORT_SYMBOL(__kfree_skb);
933 static __always_inline
934 bool __kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
936 if (unlikely(!skb_unref(skb)))
939 DEBUG_NET_WARN_ON_ONCE(reason <= 0 || reason >= SKB_DROP_REASON_MAX);
941 if (reason == SKB_CONSUMED)
942 trace_consume_skb(skb);
944 trace_kfree_skb(skb, __builtin_return_address(0), reason);
949 * kfree_skb_reason - free an sk_buff with special reason
950 * @skb: buffer to free
951 * @reason: reason why this skb is dropped
953 * Drop a reference to the buffer and free it if the usage count has
954 * hit zero. Meanwhile, pass the drop reason to 'kfree_skb'
958 kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason)
960 if (__kfree_skb_reason(skb, reason))
963 EXPORT_SYMBOL(kfree_skb_reason);
966 kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
969 struct sk_buff *next = segs->next;
971 if (__kfree_skb_reason(segs, reason))
976 EXPORT_SYMBOL(kfree_skb_list_reason);
978 /* Dump skb information and contents.
980 * Must only be called from net_ratelimit()-ed paths.
982 * Dumps whole packets if full_pkt, only headers otherwise.
984 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
986 struct skb_shared_info *sh = skb_shinfo(skb);
987 struct net_device *dev = skb->dev;
988 struct sock *sk = skb->sk;
989 struct sk_buff *list_skb;
990 bool has_mac, has_trans;
991 int headroom, tailroom;
997 len = min_t(int, skb->len, MAX_HEADER + 128);
999 headroom = skb_headroom(skb);
1000 tailroom = skb_tailroom(skb);
1002 has_mac = skb_mac_header_was_set(skb);
1003 has_trans = skb_transport_header_was_set(skb);
1005 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1006 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
1007 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1008 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1009 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
1010 level, skb->len, headroom, skb_headlen(skb), tailroom,
1011 has_mac ? skb->mac_header : -1,
1012 has_mac ? skb_mac_header_len(skb) : -1,
1013 skb->network_header,
1014 has_trans ? skb_network_header_len(skb) : -1,
1015 has_trans ? skb->transport_header : -1,
1016 sh->tx_flags, sh->nr_frags,
1017 sh->gso_size, sh->gso_type, sh->gso_segs,
1018 skb->csum, skb->ip_summed, skb->csum_complete_sw,
1019 skb->csum_valid, skb->csum_level,
1020 skb->hash, skb->sw_hash, skb->l4_hash,
1021 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
1024 printk("%sdev name=%s feat=%pNF\n",
1025 level, dev->name, &dev->features);
1027 printk("%ssk family=%hu type=%u proto=%u\n",
1028 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1030 if (full_pkt && headroom)
1031 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1032 16, 1, skb->head, headroom, false);
1034 seg_len = min_t(int, skb_headlen(skb), len);
1036 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
1037 16, 1, skb->data, seg_len, false);
1040 if (full_pkt && tailroom)
1041 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1042 16, 1, skb_tail_pointer(skb), tailroom, false);
1044 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1045 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1046 u32 p_off, p_len, copied;
1050 skb_frag_foreach_page(frag, skb_frag_off(frag),
1051 skb_frag_size(frag), p, p_off, p_len,
1053 seg_len = min_t(int, p_len, len);
1054 vaddr = kmap_atomic(p);
1055 print_hex_dump(level, "skb frag: ",
1057 16, 1, vaddr + p_off, seg_len, false);
1058 kunmap_atomic(vaddr);
1065 if (full_pkt && skb_has_frag_list(skb)) {
1066 printk("skb fraglist:\n");
1067 skb_walk_frags(skb, list_skb)
1068 skb_dump(level, list_skb, true);
1071 EXPORT_SYMBOL(skb_dump);
1074 * skb_tx_error - report an sk_buff xmit error
1075 * @skb: buffer that triggered an error
1077 * Report xmit error if a device callback is tracking this skb.
1078 * skb must be freed afterwards.
1080 void skb_tx_error(struct sk_buff *skb)
1083 skb_zcopy_downgrade_managed(skb);
1084 skb_zcopy_clear(skb, true);
1087 EXPORT_SYMBOL(skb_tx_error);
1089 #ifdef CONFIG_TRACEPOINTS
1091 * consume_skb - free an skbuff
1092 * @skb: buffer to free
1094 * Drop a ref to the buffer and free it if the usage count has hit zero
1095 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
1096 * is being dropped after a failure and notes that
1098 void consume_skb(struct sk_buff *skb)
1100 if (!skb_unref(skb))
1103 trace_consume_skb(skb);
1106 EXPORT_SYMBOL(consume_skb);
1110 * __consume_stateless_skb - free an skbuff, assuming it is stateless
1111 * @skb: buffer to free
1113 * Alike consume_skb(), but this variant assumes that this is the last
1114 * skb reference and all the head states have been already dropped
1116 void __consume_stateless_skb(struct sk_buff *skb)
1118 trace_consume_skb(skb);
1119 skb_release_data(skb, SKB_CONSUMED);
1123 static void napi_skb_cache_put(struct sk_buff *skb)
1125 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1128 kasan_poison_object_data(skbuff_head_cache, skb);
1129 nc->skb_cache[nc->skb_count++] = skb;
1131 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1132 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1133 kasan_unpoison_object_data(skbuff_head_cache,
1136 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_HALF,
1137 nc->skb_cache + NAPI_SKB_CACHE_HALF);
1138 nc->skb_count = NAPI_SKB_CACHE_HALF;
1142 void __kfree_skb_defer(struct sk_buff *skb)
1144 skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1145 napi_skb_cache_put(skb);
1148 void napi_skb_free_stolen_head(struct sk_buff *skb)
1150 if (unlikely(skb->slow_gro)) {
1157 napi_skb_cache_put(skb);
1160 void napi_consume_skb(struct sk_buff *skb, int budget)
1162 /* Zero budget indicate non-NAPI context called us, like netpoll */
1163 if (unlikely(!budget)) {
1164 dev_consume_skb_any(skb);
1168 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1170 if (!skb_unref(skb))
1173 /* if reaching here SKB is ready to free */
1174 trace_consume_skb(skb);
1176 /* if SKB is a clone, don't handle this case */
1177 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1182 skb_release_all(skb, SKB_CONSUMED);
1183 napi_skb_cache_put(skb);
1185 EXPORT_SYMBOL(napi_consume_skb);
1187 /* Make sure a field is contained by headers group */
1188 #define CHECK_SKB_FIELD(field) \
1189 BUILD_BUG_ON(offsetof(struct sk_buff, field) != \
1190 offsetof(struct sk_buff, headers.field)); \
1192 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1194 new->tstamp = old->tstamp;
1195 /* We do not copy old->sk */
1196 new->dev = old->dev;
1197 memcpy(new->cb, old->cb, sizeof(old->cb));
1198 skb_dst_copy(new, old);
1199 __skb_ext_copy(new, old);
1200 __nf_copy(new, old, false);
1202 /* Note : this field could be in the headers group.
1203 * It is not yet because we do not want to have a 16 bit hole
1205 new->queue_mapping = old->queue_mapping;
1207 memcpy(&new->headers, &old->headers, sizeof(new->headers));
1208 CHECK_SKB_FIELD(protocol);
1209 CHECK_SKB_FIELD(csum);
1210 CHECK_SKB_FIELD(hash);
1211 CHECK_SKB_FIELD(priority);
1212 CHECK_SKB_FIELD(skb_iif);
1213 CHECK_SKB_FIELD(vlan_proto);
1214 CHECK_SKB_FIELD(vlan_tci);
1215 CHECK_SKB_FIELD(transport_header);
1216 CHECK_SKB_FIELD(network_header);
1217 CHECK_SKB_FIELD(mac_header);
1218 CHECK_SKB_FIELD(inner_protocol);
1219 CHECK_SKB_FIELD(inner_transport_header);
1220 CHECK_SKB_FIELD(inner_network_header);
1221 CHECK_SKB_FIELD(inner_mac_header);
1222 CHECK_SKB_FIELD(mark);
1223 #ifdef CONFIG_NETWORK_SECMARK
1224 CHECK_SKB_FIELD(secmark);
1226 #ifdef CONFIG_NET_RX_BUSY_POLL
1227 CHECK_SKB_FIELD(napi_id);
1229 CHECK_SKB_FIELD(alloc_cpu);
1231 CHECK_SKB_FIELD(sender_cpu);
1233 #ifdef CONFIG_NET_SCHED
1234 CHECK_SKB_FIELD(tc_index);
1240 * You should not add any new code to this function. Add it to
1241 * __copy_skb_header above instead.
1243 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1245 #define C(x) n->x = skb->x
1247 n->next = n->prev = NULL;
1249 __copy_skb_header(n, skb);
1254 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1260 n->destructor = NULL;
1267 refcount_set(&n->users, 1);
1269 atomic_inc(&(skb_shinfo(skb)->dataref));
1277 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1278 * @first: first sk_buff of the msg
1280 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1284 n = alloc_skb(0, GFP_ATOMIC);
1288 n->len = first->len;
1289 n->data_len = first->len;
1290 n->truesize = first->truesize;
1292 skb_shinfo(n)->frag_list = first;
1294 __copy_skb_header(n, first);
1295 n->destructor = NULL;
1299 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1302 * skb_morph - morph one skb into another
1303 * @dst: the skb to receive the contents
1304 * @src: the skb to supply the contents
1306 * This is identical to skb_clone except that the target skb is
1307 * supplied by the user.
1309 * The target skb is returned upon exit.
1311 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1313 skb_release_all(dst, SKB_CONSUMED);
1314 return __skb_clone(dst, src);
1316 EXPORT_SYMBOL_GPL(skb_morph);
1318 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1320 unsigned long max_pg, num_pg, new_pg, old_pg;
1321 struct user_struct *user;
1323 if (capable(CAP_IPC_LOCK) || !size)
1326 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1327 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1328 user = mmp->user ? : current_user();
1330 old_pg = atomic_long_read(&user->locked_vm);
1332 new_pg = old_pg + num_pg;
1333 if (new_pg > max_pg)
1335 } while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1338 mmp->user = get_uid(user);
1339 mmp->num_pg = num_pg;
1341 mmp->num_pg += num_pg;
1346 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1348 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1351 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1352 free_uid(mmp->user);
1355 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1357 static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1359 struct ubuf_info_msgzc *uarg;
1360 struct sk_buff *skb;
1362 WARN_ON_ONCE(!in_task());
1364 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1368 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1369 uarg = (void *)skb->cb;
1370 uarg->mmp.user = NULL;
1372 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1377 uarg->ubuf.callback = msg_zerocopy_callback;
1378 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1380 uarg->bytelen = size;
1382 uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1383 refcount_set(&uarg->ubuf.refcnt, 1);
1389 static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1391 return container_of((void *)uarg, struct sk_buff, cb);
1394 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1395 struct ubuf_info *uarg)
1398 struct ubuf_info_msgzc *uarg_zc;
1399 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1402 /* there might be non MSG_ZEROCOPY users */
1403 if (uarg->callback != msg_zerocopy_callback)
1406 /* realloc only when socket is locked (TCP, UDP cork),
1407 * so uarg->len and sk_zckey access is serialized
1409 if (!sock_owned_by_user(sk)) {
1414 uarg_zc = uarg_to_msgzc(uarg);
1415 bytelen = uarg_zc->bytelen + size;
1416 if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1417 /* TCP can create new skb to attach new uarg */
1418 if (sk->sk_type == SOCK_STREAM)
1423 next = (u32)atomic_read(&sk->sk_zckey);
1424 if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1425 if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1428 uarg_zc->bytelen = bytelen;
1429 atomic_set(&sk->sk_zckey, ++next);
1431 /* no extra ref when appending to datagram (MSG_MORE) */
1432 if (sk->sk_type == SOCK_STREAM)
1433 net_zcopy_get(uarg);
1440 return msg_zerocopy_alloc(sk, size);
1442 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1444 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1446 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1450 old_lo = serr->ee.ee_info;
1451 old_hi = serr->ee.ee_data;
1452 sum_len = old_hi - old_lo + 1ULL + len;
1454 if (sum_len >= (1ULL << 32))
1457 if (lo != old_hi + 1)
1460 serr->ee.ee_data += len;
1464 static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1466 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1467 struct sock_exterr_skb *serr;
1468 struct sock *sk = skb->sk;
1469 struct sk_buff_head *q;
1470 unsigned long flags;
1475 mm_unaccount_pinned_pages(&uarg->mmp);
1477 /* if !len, there was only 1 call, and it was aborted
1478 * so do not queue a completion notification
1480 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1485 hi = uarg->id + len - 1;
1486 is_zerocopy = uarg->zerocopy;
1488 serr = SKB_EXT_ERR(skb);
1489 memset(serr, 0, sizeof(*serr));
1490 serr->ee.ee_errno = 0;
1491 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1492 serr->ee.ee_data = hi;
1493 serr->ee.ee_info = lo;
1495 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1497 q = &sk->sk_error_queue;
1498 spin_lock_irqsave(&q->lock, flags);
1499 tail = skb_peek_tail(q);
1500 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1501 !skb_zerocopy_notify_extend(tail, lo, len)) {
1502 __skb_queue_tail(q, skb);
1505 spin_unlock_irqrestore(&q->lock, flags);
1507 sk_error_report(sk);
1514 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg,
1517 struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1519 uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1521 if (refcount_dec_and_test(&uarg->refcnt))
1522 __msg_zerocopy_callback(uarg_zc);
1524 EXPORT_SYMBOL_GPL(msg_zerocopy_callback);
1526 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1528 struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1530 atomic_dec(&sk->sk_zckey);
1531 uarg_to_msgzc(uarg)->len--;
1534 msg_zerocopy_callback(NULL, uarg, true);
1536 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1538 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1539 struct msghdr *msg, int len,
1540 struct ubuf_info *uarg)
1542 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1543 int err, orig_len = skb->len;
1545 /* An skb can only point to one uarg. This edge case happens when
1546 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1548 if (orig_uarg && uarg != orig_uarg)
1551 err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1552 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1553 struct sock *save_sk = skb->sk;
1555 /* Streams do not free skb on error. Reset to prev state. */
1556 iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1558 ___pskb_trim(skb, orig_len);
1563 skb_zcopy_set(skb, uarg, NULL);
1564 return skb->len - orig_len;
1566 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1568 void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1572 skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1573 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1574 skb_frag_ref(skb, i);
1576 EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1578 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1581 if (skb_zcopy(orig)) {
1582 if (skb_zcopy(nskb)) {
1583 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1588 if (skb_uarg(nskb) == skb_uarg(orig))
1590 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1593 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1599 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1600 * @skb: the skb to modify
1601 * @gfp_mask: allocation priority
1603 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1604 * It will copy all frags into kernel and drop the reference
1605 * to userspace pages.
1607 * If this function is called from an interrupt gfp_mask() must be
1610 * Returns 0 on success or a negative error code on failure
1611 * to allocate kernel memory to copy to.
1613 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1615 int num_frags = skb_shinfo(skb)->nr_frags;
1616 struct page *page, *head = NULL;
1620 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1626 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1627 for (i = 0; i < new_frags; i++) {
1628 page = alloc_page(gfp_mask);
1631 struct page *next = (struct page *)page_private(head);
1637 set_page_private(page, (unsigned long)head);
1643 for (i = 0; i < num_frags; i++) {
1644 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1645 u32 p_off, p_len, copied;
1649 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1650 p, p_off, p_len, copied) {
1652 vaddr = kmap_atomic(p);
1654 while (done < p_len) {
1655 if (d_off == PAGE_SIZE) {
1657 page = (struct page *)page_private(page);
1659 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1660 memcpy(page_address(page) + d_off,
1661 vaddr + p_off + done, copy);
1665 kunmap_atomic(vaddr);
1669 /* skb frags release userspace buffers */
1670 for (i = 0; i < num_frags; i++)
1671 skb_frag_unref(skb, i);
1673 /* skb frags point to kernel buffers */
1674 for (i = 0; i < new_frags - 1; i++) {
1675 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1676 head = (struct page *)page_private(head);
1678 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1679 skb_shinfo(skb)->nr_frags = new_frags;
1682 skb_zcopy_clear(skb, false);
1685 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1688 * skb_clone - duplicate an sk_buff
1689 * @skb: buffer to clone
1690 * @gfp_mask: allocation priority
1692 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1693 * copies share the same packet data but not structure. The new
1694 * buffer has a reference count of 1. If the allocation fails the
1695 * function returns %NULL otherwise the new buffer is returned.
1697 * If this function is called from an interrupt gfp_mask() must be
1701 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1703 struct sk_buff_fclones *fclones = container_of(skb,
1704 struct sk_buff_fclones,
1708 if (skb_orphan_frags(skb, gfp_mask))
1711 if (skb->fclone == SKB_FCLONE_ORIG &&
1712 refcount_read(&fclones->fclone_ref) == 1) {
1714 refcount_set(&fclones->fclone_ref, 2);
1715 n->fclone = SKB_FCLONE_CLONE;
1717 if (skb_pfmemalloc(skb))
1718 gfp_mask |= __GFP_MEMALLOC;
1720 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1724 n->fclone = SKB_FCLONE_UNAVAILABLE;
1727 return __skb_clone(n, skb);
1729 EXPORT_SYMBOL(skb_clone);
1731 void skb_headers_offset_update(struct sk_buff *skb, int off)
1733 /* Only adjust this if it actually is csum_start rather than csum */
1734 if (skb->ip_summed == CHECKSUM_PARTIAL)
1735 skb->csum_start += off;
1736 /* {transport,network,mac}_header and tail are relative to skb->head */
1737 skb->transport_header += off;
1738 skb->network_header += off;
1739 if (skb_mac_header_was_set(skb))
1740 skb->mac_header += off;
1741 skb->inner_transport_header += off;
1742 skb->inner_network_header += off;
1743 skb->inner_mac_header += off;
1745 EXPORT_SYMBOL(skb_headers_offset_update);
1747 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1749 __copy_skb_header(new, old);
1751 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1752 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1753 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1755 EXPORT_SYMBOL(skb_copy_header);
1757 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1759 if (skb_pfmemalloc(skb))
1760 return SKB_ALLOC_RX;
1765 * skb_copy - create private copy of an sk_buff
1766 * @skb: buffer to copy
1767 * @gfp_mask: allocation priority
1769 * Make a copy of both an &sk_buff and its data. This is used when the
1770 * caller wishes to modify the data and needs a private copy of the
1771 * data to alter. Returns %NULL on failure or the pointer to the buffer
1772 * on success. The returned buffer has a reference count of 1.
1774 * As by-product this function converts non-linear &sk_buff to linear
1775 * one, so that &sk_buff becomes completely private and caller is allowed
1776 * to modify all the data of returned buffer. This means that this
1777 * function is not recommended for use in circumstances when only
1778 * header is going to be modified. Use pskb_copy() instead.
1781 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1783 int headerlen = skb_headroom(skb);
1784 unsigned int size = skb_end_offset(skb) + skb->data_len;
1785 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1786 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1791 /* Set the data pointer */
1792 skb_reserve(n, headerlen);
1793 /* Set the tail pointer and length */
1794 skb_put(n, skb->len);
1796 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1798 skb_copy_header(n, skb);
1801 EXPORT_SYMBOL(skb_copy);
1804 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1805 * @skb: buffer to copy
1806 * @headroom: headroom of new skb
1807 * @gfp_mask: allocation priority
1808 * @fclone: if true allocate the copy of the skb from the fclone
1809 * cache instead of the head cache; it is recommended to set this
1810 * to true for the cases where the copy will likely be cloned
1812 * Make a copy of both an &sk_buff and part of its data, located
1813 * in header. Fragmented data remain shared. This is used when
1814 * the caller wishes to modify only header of &sk_buff and needs
1815 * private copy of the header to alter. Returns %NULL on failure
1816 * or the pointer to the buffer on success.
1817 * The returned buffer has a reference count of 1.
1820 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1821 gfp_t gfp_mask, bool fclone)
1823 unsigned int size = skb_headlen(skb) + headroom;
1824 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1825 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1830 /* Set the data pointer */
1831 skb_reserve(n, headroom);
1832 /* Set the tail pointer and length */
1833 skb_put(n, skb_headlen(skb));
1834 /* Copy the bytes */
1835 skb_copy_from_linear_data(skb, n->data, n->len);
1837 n->truesize += skb->data_len;
1838 n->data_len = skb->data_len;
1841 if (skb_shinfo(skb)->nr_frags) {
1844 if (skb_orphan_frags(skb, gfp_mask) ||
1845 skb_zerocopy_clone(n, skb, gfp_mask)) {
1850 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1851 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1852 skb_frag_ref(skb, i);
1854 skb_shinfo(n)->nr_frags = i;
1857 if (skb_has_frag_list(skb)) {
1858 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1859 skb_clone_fraglist(n);
1862 skb_copy_header(n, skb);
1866 EXPORT_SYMBOL(__pskb_copy_fclone);
1869 * pskb_expand_head - reallocate header of &sk_buff
1870 * @skb: buffer to reallocate
1871 * @nhead: room to add at head
1872 * @ntail: room to add at tail
1873 * @gfp_mask: allocation priority
1875 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1876 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1877 * reference count of 1. Returns zero in the case of success or error,
1878 * if expansion failed. In the last case, &sk_buff is not changed.
1880 * All the pointers pointing into skb header may change and must be
1881 * reloaded after call to this function.
1884 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1887 unsigned int osize = skb_end_offset(skb);
1888 unsigned int size = osize + nhead + ntail;
1895 BUG_ON(skb_shared(skb));
1897 skb_zcopy_downgrade_managed(skb);
1899 if (skb_pfmemalloc(skb))
1900 gfp_mask |= __GFP_MEMALLOC;
1902 size = SKB_DATA_ALIGN(size);
1903 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1904 size = kmalloc_size_roundup(size);
1905 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
1908 size = SKB_WITH_OVERHEAD(size);
1910 /* Copy only real data... and, alas, header. This should be
1911 * optimized for the cases when header is void.
1913 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1915 memcpy((struct skb_shared_info *)(data + size),
1917 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1920 * if shinfo is shared we must drop the old head gracefully, but if it
1921 * is not we can just drop the old head and let the existing refcount
1922 * be since all we did is relocate the values
1924 if (skb_cloned(skb)) {
1925 if (skb_orphan_frags(skb, gfp_mask))
1928 refcount_inc(&skb_uarg(skb)->refcnt);
1929 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1930 skb_frag_ref(skb, i);
1932 if (skb_has_frag_list(skb))
1933 skb_clone_fraglist(skb);
1935 skb_release_data(skb, SKB_CONSUMED);
1939 off = (data + nhead) - skb->head;
1945 skb_set_end_offset(skb, size);
1946 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1950 skb_headers_offset_update(skb, nhead);
1954 atomic_set(&skb_shinfo(skb)->dataref, 1);
1956 skb_metadata_clear(skb);
1958 /* It is not generally safe to change skb->truesize.
1959 * For the moment, we really care of rx path, or
1960 * when skb is orphaned (not attached to a socket).
1962 if (!skb->sk || skb->destructor == sock_edemux)
1963 skb->truesize += size - osize;
1972 EXPORT_SYMBOL(pskb_expand_head);
1974 /* Make private copy of skb with writable head and some headroom */
1976 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1978 struct sk_buff *skb2;
1979 int delta = headroom - skb_headroom(skb);
1982 skb2 = pskb_copy(skb, GFP_ATOMIC);
1984 skb2 = skb_clone(skb, GFP_ATOMIC);
1985 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1993 EXPORT_SYMBOL(skb_realloc_headroom);
1995 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
1997 unsigned int saved_end_offset, saved_truesize;
1998 struct skb_shared_info *shinfo;
2001 saved_end_offset = skb_end_offset(skb);
2002 saved_truesize = skb->truesize;
2004 res = pskb_expand_head(skb, 0, 0, pri);
2008 skb->truesize = saved_truesize;
2010 if (likely(skb_end_offset(skb) == saved_end_offset))
2013 shinfo = skb_shinfo(skb);
2015 /* We are about to change back skb->end,
2016 * we need to move skb_shinfo() to its new location.
2018 memmove(skb->head + saved_end_offset,
2020 offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2022 skb_set_end_offset(skb, saved_end_offset);
2028 * skb_expand_head - reallocate header of &sk_buff
2029 * @skb: buffer to reallocate
2030 * @headroom: needed headroom
2032 * Unlike skb_realloc_headroom, this one does not allocate a new skb
2033 * if possible; copies skb->sk to new skb as needed
2034 * and frees original skb in case of failures.
2036 * It expect increased headroom and generates warning otherwise.
2039 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2041 int delta = headroom - skb_headroom(skb);
2042 int osize = skb_end_offset(skb);
2043 struct sock *sk = skb->sk;
2045 if (WARN_ONCE(delta <= 0,
2046 "%s is expecting an increase in the headroom", __func__))
2049 delta = SKB_DATA_ALIGN(delta);
2050 /* pskb_expand_head() might crash, if skb is shared. */
2051 if (skb_shared(skb) || !is_skb_wmem(skb)) {
2052 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2054 if (unlikely(!nskb))
2058 skb_set_owner_w(nskb, sk);
2062 if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2065 if (sk && is_skb_wmem(skb)) {
2066 delta = skb_end_offset(skb) - osize;
2067 refcount_add(delta, &sk->sk_wmem_alloc);
2068 skb->truesize += delta;
2076 EXPORT_SYMBOL(skb_expand_head);
2079 * skb_copy_expand - copy and expand sk_buff
2080 * @skb: buffer to copy
2081 * @newheadroom: new free bytes at head
2082 * @newtailroom: new free bytes at tail
2083 * @gfp_mask: allocation priority
2085 * Make a copy of both an &sk_buff and its data and while doing so
2086 * allocate additional space.
2088 * This is used when the caller wishes to modify the data and needs a
2089 * private copy of the data to alter as well as more space for new fields.
2090 * Returns %NULL on failure or the pointer to the buffer
2091 * on success. The returned buffer has a reference count of 1.
2093 * You must pass %GFP_ATOMIC as the allocation priority if this function
2094 * is called from an interrupt.
2096 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2097 int newheadroom, int newtailroom,
2101 * Allocate the copy buffer
2103 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
2104 gfp_mask, skb_alloc_rx_flag(skb),
2106 int oldheadroom = skb_headroom(skb);
2107 int head_copy_len, head_copy_off;
2112 skb_reserve(n, newheadroom);
2114 /* Set the tail pointer and length */
2115 skb_put(n, skb->len);
2117 head_copy_len = oldheadroom;
2119 if (newheadroom <= head_copy_len)
2120 head_copy_len = newheadroom;
2122 head_copy_off = newheadroom - head_copy_len;
2124 /* Copy the linear header and data. */
2125 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2126 skb->len + head_copy_len));
2128 skb_copy_header(n, skb);
2130 skb_headers_offset_update(n, newheadroom - oldheadroom);
2134 EXPORT_SYMBOL(skb_copy_expand);
2137 * __skb_pad - zero pad the tail of an skb
2138 * @skb: buffer to pad
2139 * @pad: space to pad
2140 * @free_on_error: free buffer on error
2142 * Ensure that a buffer is followed by a padding area that is zero
2143 * filled. Used by network drivers which may DMA or transfer data
2144 * beyond the buffer end onto the wire.
2146 * May return error in out of memory cases. The skb is freed on error
2147 * if @free_on_error is true.
2150 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2155 /* If the skbuff is non linear tailroom is always zero.. */
2156 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2157 memset(skb->data+skb->len, 0, pad);
2161 ntail = skb->data_len + pad - (skb->end - skb->tail);
2162 if (likely(skb_cloned(skb) || ntail > 0)) {
2163 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2168 /* FIXME: The use of this function with non-linear skb's really needs
2171 err = skb_linearize(skb);
2175 memset(skb->data + skb->len, 0, pad);
2183 EXPORT_SYMBOL(__skb_pad);
2186 * pskb_put - add data to the tail of a potentially fragmented buffer
2187 * @skb: start of the buffer to use
2188 * @tail: tail fragment of the buffer to use
2189 * @len: amount of data to add
2191 * This function extends the used data area of the potentially
2192 * fragmented buffer. @tail must be the last fragment of @skb -- or
2193 * @skb itself. If this would exceed the total buffer size the kernel
2194 * will panic. A pointer to the first byte of the extra data is
2198 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2201 skb->data_len += len;
2204 return skb_put(tail, len);
2206 EXPORT_SYMBOL_GPL(pskb_put);
2209 * skb_put - add data to a buffer
2210 * @skb: buffer to use
2211 * @len: amount of data to add
2213 * This function extends the used data area of the buffer. If this would
2214 * exceed the total buffer size the kernel will panic. A pointer to the
2215 * first byte of the extra data is returned.
2217 void *skb_put(struct sk_buff *skb, unsigned int len)
2219 void *tmp = skb_tail_pointer(skb);
2220 SKB_LINEAR_ASSERT(skb);
2223 if (unlikely(skb->tail > skb->end))
2224 skb_over_panic(skb, len, __builtin_return_address(0));
2227 EXPORT_SYMBOL(skb_put);
2230 * skb_push - add data to the start of a buffer
2231 * @skb: buffer to use
2232 * @len: amount of data to add
2234 * This function extends the used data area of the buffer at the buffer
2235 * start. If this would exceed the total buffer headroom the kernel will
2236 * panic. A pointer to the first byte of the extra data is returned.
2238 void *skb_push(struct sk_buff *skb, unsigned int len)
2242 if (unlikely(skb->data < skb->head))
2243 skb_under_panic(skb, len, __builtin_return_address(0));
2246 EXPORT_SYMBOL(skb_push);
2249 * skb_pull - remove data from the start of a buffer
2250 * @skb: buffer to use
2251 * @len: amount of data to remove
2253 * This function removes data from the start of a buffer, returning
2254 * the memory to the headroom. A pointer to the next data in the buffer
2255 * is returned. Once the data has been pulled future pushes will overwrite
2258 void *skb_pull(struct sk_buff *skb, unsigned int len)
2260 return skb_pull_inline(skb, len);
2262 EXPORT_SYMBOL(skb_pull);
2265 * skb_pull_data - remove data from the start of a buffer returning its
2266 * original position.
2267 * @skb: buffer to use
2268 * @len: amount of data to remove
2270 * This function removes data from the start of a buffer, returning
2271 * the memory to the headroom. A pointer to the original data in the buffer
2272 * is returned after checking if there is enough data to pull. Once the
2273 * data has been pulled future pushes will overwrite the old data.
2275 void *skb_pull_data(struct sk_buff *skb, size_t len)
2277 void *data = skb->data;
2286 EXPORT_SYMBOL(skb_pull_data);
2289 * skb_trim - remove end from a buffer
2290 * @skb: buffer to alter
2293 * Cut the length of a buffer down by removing data from the tail. If
2294 * the buffer is already under the length specified it is not modified.
2295 * The skb must be linear.
2297 void skb_trim(struct sk_buff *skb, unsigned int len)
2300 __skb_trim(skb, len);
2302 EXPORT_SYMBOL(skb_trim);
2304 /* Trims skb to length len. It can change skb pointers.
2307 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2309 struct sk_buff **fragp;
2310 struct sk_buff *frag;
2311 int offset = skb_headlen(skb);
2312 int nfrags = skb_shinfo(skb)->nr_frags;
2316 if (skb_cloned(skb) &&
2317 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2324 for (; i < nfrags; i++) {
2325 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2332 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2335 skb_shinfo(skb)->nr_frags = i;
2337 for (; i < nfrags; i++)
2338 skb_frag_unref(skb, i);
2340 if (skb_has_frag_list(skb))
2341 skb_drop_fraglist(skb);
2345 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2346 fragp = &frag->next) {
2347 int end = offset + frag->len;
2349 if (skb_shared(frag)) {
2350 struct sk_buff *nfrag;
2352 nfrag = skb_clone(frag, GFP_ATOMIC);
2353 if (unlikely(!nfrag))
2356 nfrag->next = frag->next;
2368 unlikely((err = pskb_trim(frag, len - offset))))
2372 skb_drop_list(&frag->next);
2377 if (len > skb_headlen(skb)) {
2378 skb->data_len -= skb->len - len;
2383 skb_set_tail_pointer(skb, len);
2386 if (!skb->sk || skb->destructor == sock_edemux)
2390 EXPORT_SYMBOL(___pskb_trim);
2392 /* Note : use pskb_trim_rcsum() instead of calling this directly
2394 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2396 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2397 int delta = skb->len - len;
2399 skb->csum = csum_block_sub(skb->csum,
2400 skb_checksum(skb, len, delta, 0),
2402 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2403 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2404 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2406 if (offset + sizeof(__sum16) > hdlen)
2409 return __pskb_trim(skb, len);
2411 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2414 * __pskb_pull_tail - advance tail of skb header
2415 * @skb: buffer to reallocate
2416 * @delta: number of bytes to advance tail
2418 * The function makes a sense only on a fragmented &sk_buff,
2419 * it expands header moving its tail forward and copying necessary
2420 * data from fragmented part.
2422 * &sk_buff MUST have reference count of 1.
2424 * Returns %NULL (and &sk_buff does not change) if pull failed
2425 * or value of new tail of skb in the case of success.
2427 * All the pointers pointing into skb header may change and must be
2428 * reloaded after call to this function.
2431 /* Moves tail of skb head forward, copying data from fragmented part,
2432 * when it is necessary.
2433 * 1. It may fail due to malloc failure.
2434 * 2. It may change skb pointers.
2436 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2438 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2440 /* If skb has not enough free space at tail, get new one
2441 * plus 128 bytes for future expansions. If we have enough
2442 * room at tail, reallocate without expansion only if skb is cloned.
2444 int i, k, eat = (skb->tail + delta) - skb->end;
2446 if (eat > 0 || skb_cloned(skb)) {
2447 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2452 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2453 skb_tail_pointer(skb), delta));
2455 /* Optimization: no fragments, no reasons to preestimate
2456 * size of pulled pages. Superb.
2458 if (!skb_has_frag_list(skb))
2461 /* Estimate size of pulled pages. */
2463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2464 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2471 /* If we need update frag list, we are in troubles.
2472 * Certainly, it is possible to add an offset to skb data,
2473 * but taking into account that pulling is expected to
2474 * be very rare operation, it is worth to fight against
2475 * further bloating skb head and crucify ourselves here instead.
2476 * Pure masohism, indeed. 8)8)
2479 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2480 struct sk_buff *clone = NULL;
2481 struct sk_buff *insp = NULL;
2484 if (list->len <= eat) {
2485 /* Eaten as whole. */
2490 /* Eaten partially. */
2491 if (skb_is_gso(skb) && !list->head_frag &&
2493 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2495 if (skb_shared(list)) {
2496 /* Sucks! We need to fork list. :-( */
2497 clone = skb_clone(list, GFP_ATOMIC);
2503 /* This may be pulled without
2507 if (!pskb_pull(list, eat)) {
2515 /* Free pulled out fragments. */
2516 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2517 skb_shinfo(skb)->frag_list = list->next;
2520 /* And insert new clone at head. */
2523 skb_shinfo(skb)->frag_list = clone;
2526 /* Success! Now we may commit changes to skb data. */
2531 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2532 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2535 skb_frag_unref(skb, i);
2538 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2540 *frag = skb_shinfo(skb)->frags[i];
2542 skb_frag_off_add(frag, eat);
2543 skb_frag_size_sub(frag, eat);
2551 skb_shinfo(skb)->nr_frags = k;
2555 skb->data_len -= delta;
2558 skb_zcopy_clear(skb, false);
2560 return skb_tail_pointer(skb);
2562 EXPORT_SYMBOL(__pskb_pull_tail);
2565 * skb_copy_bits - copy bits from skb to kernel buffer
2567 * @offset: offset in source
2568 * @to: destination buffer
2569 * @len: number of bytes to copy
2571 * Copy the specified number of bytes from the source skb to the
2572 * destination buffer.
2575 * If its prototype is ever changed,
2576 * check arch/{*}/net/{*}.S files,
2577 * since it is called from BPF assembly code.
2579 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2581 int start = skb_headlen(skb);
2582 struct sk_buff *frag_iter;
2585 if (offset > (int)skb->len - len)
2589 if ((copy = start - offset) > 0) {
2592 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2593 if ((len -= copy) == 0)
2599 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2601 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2603 WARN_ON(start > offset + len);
2605 end = start + skb_frag_size(f);
2606 if ((copy = end - offset) > 0) {
2607 u32 p_off, p_len, copied;
2614 skb_frag_foreach_page(f,
2615 skb_frag_off(f) + offset - start,
2616 copy, p, p_off, p_len, copied) {
2617 vaddr = kmap_atomic(p);
2618 memcpy(to + copied, vaddr + p_off, p_len);
2619 kunmap_atomic(vaddr);
2622 if ((len -= copy) == 0)
2630 skb_walk_frags(skb, frag_iter) {
2633 WARN_ON(start > offset + len);
2635 end = start + frag_iter->len;
2636 if ((copy = end - offset) > 0) {
2639 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2641 if ((len -= copy) == 0)
2655 EXPORT_SYMBOL(skb_copy_bits);
2658 * Callback from splice_to_pipe(), if we need to release some pages
2659 * at the end of the spd in case we error'ed out in filling the pipe.
2661 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2663 put_page(spd->pages[i]);
2666 static struct page *linear_to_page(struct page *page, unsigned int *len,
2667 unsigned int *offset,
2670 struct page_frag *pfrag = sk_page_frag(sk);
2672 if (!sk_page_frag_refill(sk, pfrag))
2675 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2677 memcpy(page_address(pfrag->page) + pfrag->offset,
2678 page_address(page) + *offset, *len);
2679 *offset = pfrag->offset;
2680 pfrag->offset += *len;
2685 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2687 unsigned int offset)
2689 return spd->nr_pages &&
2690 spd->pages[spd->nr_pages - 1] == page &&
2691 (spd->partial[spd->nr_pages - 1].offset +
2692 spd->partial[spd->nr_pages - 1].len == offset);
2696 * Fill page/offset/length into spd, if it can hold more pages.
2698 static bool spd_fill_page(struct splice_pipe_desc *spd,
2699 struct pipe_inode_info *pipe, struct page *page,
2700 unsigned int *len, unsigned int offset,
2704 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2708 page = linear_to_page(page, len, &offset, sk);
2712 if (spd_can_coalesce(spd, page, offset)) {
2713 spd->partial[spd->nr_pages - 1].len += *len;
2717 spd->pages[spd->nr_pages] = page;
2718 spd->partial[spd->nr_pages].len = *len;
2719 spd->partial[spd->nr_pages].offset = offset;
2725 static bool __splice_segment(struct page *page, unsigned int poff,
2726 unsigned int plen, unsigned int *off,
2728 struct splice_pipe_desc *spd, bool linear,
2730 struct pipe_inode_info *pipe)
2735 /* skip this segment if already processed */
2741 /* ignore any bits we already processed */
2747 unsigned int flen = min(*len, plen);
2749 if (spd_fill_page(spd, pipe, page, &flen, poff,
2755 } while (*len && plen);
2761 * Map linear and fragment data from the skb to spd. It reports true if the
2762 * pipe is full or if we already spliced the requested length.
2764 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2765 unsigned int *offset, unsigned int *len,
2766 struct splice_pipe_desc *spd, struct sock *sk)
2769 struct sk_buff *iter;
2771 /* map the linear part :
2772 * If skb->head_frag is set, this 'linear' part is backed by a
2773 * fragment, and if the head is not shared with any clones then
2774 * we can avoid a copy since we own the head portion of this page.
2776 if (__splice_segment(virt_to_page(skb->data),
2777 (unsigned long) skb->data & (PAGE_SIZE - 1),
2780 skb_head_is_locked(skb),
2785 * then map the fragments
2787 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2788 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2790 if (__splice_segment(skb_frag_page(f),
2791 skb_frag_off(f), skb_frag_size(f),
2792 offset, len, spd, false, sk, pipe))
2796 skb_walk_frags(skb, iter) {
2797 if (*offset >= iter->len) {
2798 *offset -= iter->len;
2801 /* __skb_splice_bits() only fails if the output has no room
2802 * left, so no point in going over the frag_list for the error
2805 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2813 * Map data from the skb to a pipe. Should handle both the linear part,
2814 * the fragments, and the frag list.
2816 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2817 struct pipe_inode_info *pipe, unsigned int tlen,
2820 struct partial_page partial[MAX_SKB_FRAGS];
2821 struct page *pages[MAX_SKB_FRAGS];
2822 struct splice_pipe_desc spd = {
2825 .nr_pages_max = MAX_SKB_FRAGS,
2826 .ops = &nosteal_pipe_buf_ops,
2827 .spd_release = sock_spd_release,
2831 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2834 ret = splice_to_pipe(pipe, &spd);
2838 EXPORT_SYMBOL_GPL(skb_splice_bits);
2840 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg,
2841 struct kvec *vec, size_t num, size_t size)
2843 struct socket *sock = sk->sk_socket;
2847 return kernel_sendmsg(sock, msg, vec, num, size);
2850 static int sendpage_unlocked(struct sock *sk, struct page *page, int offset,
2851 size_t size, int flags)
2853 struct socket *sock = sk->sk_socket;
2857 return kernel_sendpage(sock, page, offset, size, flags);
2860 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg,
2861 struct kvec *vec, size_t num, size_t size);
2862 typedef int (*sendpage_func)(struct sock *sk, struct page *page, int offset,
2863 size_t size, int flags);
2864 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
2865 int len, sendmsg_func sendmsg, sendpage_func sendpage)
2867 unsigned int orig_len = len;
2868 struct sk_buff *head = skb;
2869 unsigned short fragidx;
2874 /* Deal with head data */
2875 while (offset < skb_headlen(skb) && len) {
2879 slen = min_t(int, len, skb_headlen(skb) - offset);
2880 kv.iov_base = skb->data + offset;
2882 memset(&msg, 0, sizeof(msg));
2883 msg.msg_flags = MSG_DONTWAIT;
2885 ret = INDIRECT_CALL_2(sendmsg, kernel_sendmsg_locked,
2886 sendmsg_unlocked, sk, &msg, &kv, 1, slen);
2894 /* All the data was skb head? */
2898 /* Make offset relative to start of frags */
2899 offset -= skb_headlen(skb);
2901 /* Find where we are in frag list */
2902 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2903 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2905 if (offset < skb_frag_size(frag))
2908 offset -= skb_frag_size(frag);
2911 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2912 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2914 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2917 ret = INDIRECT_CALL_2(sendpage, kernel_sendpage_locked,
2918 sendpage_unlocked, sk,
2919 skb_frag_page(frag),
2920 skb_frag_off(frag) + offset,
2921 slen, MSG_DONTWAIT);
2934 /* Process any frag lists */
2937 if (skb_has_frag_list(skb)) {
2938 skb = skb_shinfo(skb)->frag_list;
2941 } else if (skb->next) {
2948 return orig_len - len;
2951 return orig_len == len ? ret : orig_len - len;
2954 /* Send skb data on a socket. Socket must be locked. */
2955 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2958 return __skb_send_sock(sk, skb, offset, len, kernel_sendmsg_locked,
2959 kernel_sendpage_locked);
2961 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2963 /* Send skb data on a socket. Socket must be unlocked. */
2964 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
2966 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked,
2971 * skb_store_bits - store bits from kernel buffer to skb
2972 * @skb: destination buffer
2973 * @offset: offset in destination
2974 * @from: source buffer
2975 * @len: number of bytes to copy
2977 * Copy the specified number of bytes from the source buffer to the
2978 * destination skb. This function handles all the messy bits of
2979 * traversing fragment lists and such.
2982 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2984 int start = skb_headlen(skb);
2985 struct sk_buff *frag_iter;
2988 if (offset > (int)skb->len - len)
2991 if ((copy = start - offset) > 0) {
2994 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2995 if ((len -= copy) == 0)
3001 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3002 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3005 WARN_ON(start > offset + len);
3007 end = start + skb_frag_size(frag);
3008 if ((copy = end - offset) > 0) {
3009 u32 p_off, p_len, copied;
3016 skb_frag_foreach_page(frag,
3017 skb_frag_off(frag) + offset - start,
3018 copy, p, p_off, p_len, copied) {
3019 vaddr = kmap_atomic(p);
3020 memcpy(vaddr + p_off, from + copied, p_len);
3021 kunmap_atomic(vaddr);
3024 if ((len -= copy) == 0)
3032 skb_walk_frags(skb, frag_iter) {
3035 WARN_ON(start > offset + len);
3037 end = start + frag_iter->len;
3038 if ((copy = end - offset) > 0) {
3041 if (skb_store_bits(frag_iter, offset - start,
3044 if ((len -= copy) == 0)
3057 EXPORT_SYMBOL(skb_store_bits);
3059 /* Checksum skb data. */
3060 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3061 __wsum csum, const struct skb_checksum_ops *ops)
3063 int start = skb_headlen(skb);
3064 int i, copy = start - offset;
3065 struct sk_buff *frag_iter;
3068 /* Checksum header. */
3072 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3073 skb->data + offset, copy, csum);
3074 if ((len -= copy) == 0)
3080 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3082 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3084 WARN_ON(start > offset + len);
3086 end = start + skb_frag_size(frag);
3087 if ((copy = end - offset) > 0) {
3088 u32 p_off, p_len, copied;
3096 skb_frag_foreach_page(frag,
3097 skb_frag_off(frag) + offset - start,
3098 copy, p, p_off, p_len, copied) {
3099 vaddr = kmap_atomic(p);
3100 csum2 = INDIRECT_CALL_1(ops->update,
3102 vaddr + p_off, p_len, 0);
3103 kunmap_atomic(vaddr);
3104 csum = INDIRECT_CALL_1(ops->combine,
3105 csum_block_add_ext, csum,
3117 skb_walk_frags(skb, frag_iter) {
3120 WARN_ON(start > offset + len);
3122 end = start + frag_iter->len;
3123 if ((copy = end - offset) > 0) {
3127 csum2 = __skb_checksum(frag_iter, offset - start,
3129 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3130 csum, csum2, pos, copy);
3131 if ((len -= copy) == 0)
3142 EXPORT_SYMBOL(__skb_checksum);
3144 __wsum skb_checksum(const struct sk_buff *skb, int offset,
3145 int len, __wsum csum)
3147 const struct skb_checksum_ops ops = {
3148 .update = csum_partial_ext,
3149 .combine = csum_block_add_ext,
3152 return __skb_checksum(skb, offset, len, csum, &ops);
3154 EXPORT_SYMBOL(skb_checksum);
3156 /* Both of above in one bottle. */
3158 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3161 int start = skb_headlen(skb);
3162 int i, copy = start - offset;
3163 struct sk_buff *frag_iter;
3171 csum = csum_partial_copy_nocheck(skb->data + offset, to,
3173 if ((len -= copy) == 0)
3180 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3183 WARN_ON(start > offset + len);
3185 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3186 if ((copy = end - offset) > 0) {
3187 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3188 u32 p_off, p_len, copied;
3196 skb_frag_foreach_page(frag,
3197 skb_frag_off(frag) + offset - start,
3198 copy, p, p_off, p_len, copied) {
3199 vaddr = kmap_atomic(p);
3200 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3203 kunmap_atomic(vaddr);
3204 csum = csum_block_add(csum, csum2, pos);
3216 skb_walk_frags(skb, frag_iter) {
3220 WARN_ON(start > offset + len);
3222 end = start + frag_iter->len;
3223 if ((copy = end - offset) > 0) {
3226 csum2 = skb_copy_and_csum_bits(frag_iter,
3229 csum = csum_block_add(csum, csum2, pos);
3230 if ((len -= copy) == 0)
3241 EXPORT_SYMBOL(skb_copy_and_csum_bits);
3243 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3247 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3248 /* See comments in __skb_checksum_complete(). */
3250 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3251 !skb->csum_complete_sw)
3252 netdev_rx_csum_fault(skb->dev, skb);
3254 if (!skb_shared(skb))
3255 skb->csum_valid = !sum;
3258 EXPORT_SYMBOL(__skb_checksum_complete_head);
3260 /* This function assumes skb->csum already holds pseudo header's checksum,
3261 * which has been changed from the hardware checksum, for example, by
3262 * __skb_checksum_validate_complete(). And, the original skb->csum must
3263 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3265 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3266 * zero. The new checksum is stored back into skb->csum unless the skb is
3269 __sum16 __skb_checksum_complete(struct sk_buff *skb)
3274 csum = skb_checksum(skb, 0, skb->len, 0);
3276 sum = csum_fold(csum_add(skb->csum, csum));
3277 /* This check is inverted, because we already knew the hardware
3278 * checksum is invalid before calling this function. So, if the
3279 * re-computed checksum is valid instead, then we have a mismatch
3280 * between the original skb->csum and skb_checksum(). This means either
3281 * the original hardware checksum is incorrect or we screw up skb->csum
3282 * when moving skb->data around.
3285 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3286 !skb->csum_complete_sw)
3287 netdev_rx_csum_fault(skb->dev, skb);
3290 if (!skb_shared(skb)) {
3291 /* Save full packet checksum */
3293 skb->ip_summed = CHECKSUM_COMPLETE;
3294 skb->csum_complete_sw = 1;
3295 skb->csum_valid = !sum;
3300 EXPORT_SYMBOL(__skb_checksum_complete);
3302 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3304 net_warn_ratelimited(
3305 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3310 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3311 int offset, int len)
3313 net_warn_ratelimited(
3314 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3319 static const struct skb_checksum_ops default_crc32c_ops = {
3320 .update = warn_crc32c_csum_update,
3321 .combine = warn_crc32c_csum_combine,
3324 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3325 &default_crc32c_ops;
3326 EXPORT_SYMBOL(crc32c_csum_stub);
3329 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3330 * @from: source buffer
3332 * Calculates the amount of linear headroom needed in the 'to' skb passed
3333 * into skb_zerocopy().
3336 skb_zerocopy_headlen(const struct sk_buff *from)
3338 unsigned int hlen = 0;
3340 if (!from->head_frag ||
3341 skb_headlen(from) < L1_CACHE_BYTES ||
3342 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3343 hlen = skb_headlen(from);
3348 if (skb_has_frag_list(from))
3353 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3356 * skb_zerocopy - Zero copy skb to skb
3357 * @to: destination buffer
3358 * @from: source buffer
3359 * @len: number of bytes to copy from source buffer
3360 * @hlen: size of linear headroom in destination buffer
3362 * Copies up to `len` bytes from `from` to `to` by creating references
3363 * to the frags in the source buffer.
3365 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3366 * headroom in the `to` buffer.
3369 * 0: everything is OK
3370 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3371 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3374 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3377 int plen = 0; /* length of skb->head fragment */
3380 unsigned int offset;
3382 BUG_ON(!from->head_frag && !hlen);
3384 /* dont bother with small payloads */
3385 if (len <= skb_tailroom(to))
3386 return skb_copy_bits(from, 0, skb_put(to, len), len);
3389 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3394 plen = min_t(int, skb_headlen(from), len);
3396 page = virt_to_head_page(from->head);
3397 offset = from->data - (unsigned char *)page_address(page);
3398 __skb_fill_page_desc(to, 0, page, offset, plen);
3405 skb_len_add(to, len + plen);
3407 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3411 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3413 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3418 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3419 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3421 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3423 skb_frag_ref(to, j);
3426 skb_shinfo(to)->nr_frags = j;
3430 EXPORT_SYMBOL_GPL(skb_zerocopy);
3432 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3437 if (skb->ip_summed == CHECKSUM_PARTIAL)
3438 csstart = skb_checksum_start_offset(skb);
3440 csstart = skb_headlen(skb);
3442 BUG_ON(csstart > skb_headlen(skb));
3444 skb_copy_from_linear_data(skb, to, csstart);
3447 if (csstart != skb->len)
3448 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3449 skb->len - csstart);
3451 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3452 long csstuff = csstart + skb->csum_offset;
3454 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3457 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3460 * skb_dequeue - remove from the head of the queue
3461 * @list: list to dequeue from
3463 * Remove the head of the list. The list lock is taken so the function
3464 * may be used safely with other locking list functions. The head item is
3465 * returned or %NULL if the list is empty.
3468 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3470 unsigned long flags;
3471 struct sk_buff *result;
3473 spin_lock_irqsave(&list->lock, flags);
3474 result = __skb_dequeue(list);
3475 spin_unlock_irqrestore(&list->lock, flags);
3478 EXPORT_SYMBOL(skb_dequeue);
3481 * skb_dequeue_tail - remove from the tail of the queue
3482 * @list: list to dequeue from
3484 * Remove the tail of the list. The list lock is taken so the function
3485 * may be used safely with other locking list functions. The tail item is
3486 * returned or %NULL if the list is empty.
3488 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3490 unsigned long flags;
3491 struct sk_buff *result;
3493 spin_lock_irqsave(&list->lock, flags);
3494 result = __skb_dequeue_tail(list);
3495 spin_unlock_irqrestore(&list->lock, flags);
3498 EXPORT_SYMBOL(skb_dequeue_tail);
3501 * skb_queue_purge - empty a list
3502 * @list: list to empty
3504 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3505 * the list and one reference dropped. This function takes the list
3506 * lock and is atomic with respect to other list locking functions.
3508 void skb_queue_purge(struct sk_buff_head *list)
3510 struct sk_buff *skb;
3511 while ((skb = skb_dequeue(list)) != NULL)
3514 EXPORT_SYMBOL(skb_queue_purge);
3517 * skb_rbtree_purge - empty a skb rbtree
3518 * @root: root of the rbtree to empty
3519 * Return value: the sum of truesizes of all purged skbs.
3521 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3522 * the list and one reference dropped. This function does not take
3523 * any lock. Synchronization should be handled by the caller (e.g., TCP
3524 * out-of-order queue is protected by the socket lock).
3526 unsigned int skb_rbtree_purge(struct rb_root *root)
3528 struct rb_node *p = rb_first(root);
3529 unsigned int sum = 0;
3532 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3535 rb_erase(&skb->rbnode, root);
3536 sum += skb->truesize;
3543 * skb_queue_head - queue a buffer at the list head
3544 * @list: list to use
3545 * @newsk: buffer to queue
3547 * Queue a buffer at the start of the list. This function takes the
3548 * list lock and can be used safely with other locking &sk_buff functions
3551 * A buffer cannot be placed on two lists at the same time.
3553 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3555 unsigned long flags;
3557 spin_lock_irqsave(&list->lock, flags);
3558 __skb_queue_head(list, newsk);
3559 spin_unlock_irqrestore(&list->lock, flags);
3561 EXPORT_SYMBOL(skb_queue_head);
3564 * skb_queue_tail - queue a buffer at the list tail
3565 * @list: list to use
3566 * @newsk: buffer to queue
3568 * Queue a buffer at the tail of the list. This function takes the
3569 * list lock and can be used safely with other locking &sk_buff functions
3572 * A buffer cannot be placed on two lists at the same time.
3574 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3576 unsigned long flags;
3578 spin_lock_irqsave(&list->lock, flags);
3579 __skb_queue_tail(list, newsk);
3580 spin_unlock_irqrestore(&list->lock, flags);
3582 EXPORT_SYMBOL(skb_queue_tail);
3585 * skb_unlink - remove a buffer from a list
3586 * @skb: buffer to remove
3587 * @list: list to use
3589 * Remove a packet from a list. The list locks are taken and this
3590 * function is atomic with respect to other list locked calls
3592 * You must know what list the SKB is on.
3594 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3596 unsigned long flags;
3598 spin_lock_irqsave(&list->lock, flags);
3599 __skb_unlink(skb, list);
3600 spin_unlock_irqrestore(&list->lock, flags);
3602 EXPORT_SYMBOL(skb_unlink);
3605 * skb_append - append a buffer
3606 * @old: buffer to insert after
3607 * @newsk: buffer to insert
3608 * @list: list to use
3610 * Place a packet after a given packet in a list. The list locks are taken
3611 * and this function is atomic with respect to other list locked calls.
3612 * A buffer cannot be placed on two lists at the same time.
3614 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3616 unsigned long flags;
3618 spin_lock_irqsave(&list->lock, flags);
3619 __skb_queue_after(list, old, newsk);
3620 spin_unlock_irqrestore(&list->lock, flags);
3622 EXPORT_SYMBOL(skb_append);
3624 static inline void skb_split_inside_header(struct sk_buff *skb,
3625 struct sk_buff* skb1,
3626 const u32 len, const int pos)
3630 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3632 /* And move data appendix as is. */
3633 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3634 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3636 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3637 skb_shinfo(skb)->nr_frags = 0;
3638 skb1->data_len = skb->data_len;
3639 skb1->len += skb1->data_len;
3642 skb_set_tail_pointer(skb, len);
3645 static inline void skb_split_no_header(struct sk_buff *skb,
3646 struct sk_buff* skb1,
3647 const u32 len, int pos)
3650 const int nfrags = skb_shinfo(skb)->nr_frags;
3652 skb_shinfo(skb)->nr_frags = 0;
3653 skb1->len = skb1->data_len = skb->len - len;
3655 skb->data_len = len - pos;
3657 for (i = 0; i < nfrags; i++) {
3658 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3660 if (pos + size > len) {
3661 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3665 * We have two variants in this case:
3666 * 1. Move all the frag to the second
3667 * part, if it is possible. F.e.
3668 * this approach is mandatory for TUX,
3669 * where splitting is expensive.
3670 * 2. Split is accurately. We make this.
3672 skb_frag_ref(skb, i);
3673 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3674 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3675 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3676 skb_shinfo(skb)->nr_frags++;
3680 skb_shinfo(skb)->nr_frags++;
3683 skb_shinfo(skb1)->nr_frags = k;
3687 * skb_split - Split fragmented skb to two parts at length len.
3688 * @skb: the buffer to split
3689 * @skb1: the buffer to receive the second part
3690 * @len: new length for skb
3692 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3694 int pos = skb_headlen(skb);
3695 const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
3697 skb_zcopy_downgrade_managed(skb);
3699 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
3700 skb_zerocopy_clone(skb1, skb, 0);
3701 if (len < pos) /* Split line is inside header. */
3702 skb_split_inside_header(skb, skb1, len, pos);
3703 else /* Second chunk has no header, nothing to copy. */
3704 skb_split_no_header(skb, skb1, len, pos);
3706 EXPORT_SYMBOL(skb_split);
3708 /* Shifting from/to a cloned skb is a no-go.
3710 * Caller cannot keep skb_shinfo related pointers past calling here!
3712 static int skb_prepare_for_shift(struct sk_buff *skb)
3714 return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
3718 * skb_shift - Shifts paged data partially from skb to another
3719 * @tgt: buffer into which tail data gets added
3720 * @skb: buffer from which the paged data comes from
3721 * @shiftlen: shift up to this many bytes
3723 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3724 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3725 * It's up to caller to free skb if everything was shifted.
3727 * If @tgt runs out of frags, the whole operation is aborted.
3729 * Skb cannot include anything else but paged data while tgt is allowed
3730 * to have non-paged data as well.
3732 * TODO: full sized shift could be optimized but that would need
3733 * specialized skb free'er to handle frags without up-to-date nr_frags.
3735 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3737 int from, to, merge, todo;
3738 skb_frag_t *fragfrom, *fragto;
3740 BUG_ON(shiftlen > skb->len);
3742 if (skb_headlen(skb))
3744 if (skb_zcopy(tgt) || skb_zcopy(skb))
3749 to = skb_shinfo(tgt)->nr_frags;
3750 fragfrom = &skb_shinfo(skb)->frags[from];
3752 /* Actual merge is delayed until the point when we know we can
3753 * commit all, so that we don't have to undo partial changes
3756 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3757 skb_frag_off(fragfrom))) {
3762 todo -= skb_frag_size(fragfrom);
3764 if (skb_prepare_for_shift(skb) ||
3765 skb_prepare_for_shift(tgt))
3768 /* All previous frag pointers might be stale! */
3769 fragfrom = &skb_shinfo(skb)->frags[from];
3770 fragto = &skb_shinfo(tgt)->frags[merge];
3772 skb_frag_size_add(fragto, shiftlen);
3773 skb_frag_size_sub(fragfrom, shiftlen);
3774 skb_frag_off_add(fragfrom, shiftlen);
3782 /* Skip full, not-fitting skb to avoid expensive operations */
3783 if ((shiftlen == skb->len) &&
3784 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3787 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3790 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3791 if (to == MAX_SKB_FRAGS)
3794 fragfrom = &skb_shinfo(skb)->frags[from];
3795 fragto = &skb_shinfo(tgt)->frags[to];
3797 if (todo >= skb_frag_size(fragfrom)) {
3798 *fragto = *fragfrom;
3799 todo -= skb_frag_size(fragfrom);
3804 __skb_frag_ref(fragfrom);
3805 skb_frag_page_copy(fragto, fragfrom);
3806 skb_frag_off_copy(fragto, fragfrom);
3807 skb_frag_size_set(fragto, todo);
3809 skb_frag_off_add(fragfrom, todo);
3810 skb_frag_size_sub(fragfrom, todo);
3818 /* Ready to "commit" this state change to tgt */
3819 skb_shinfo(tgt)->nr_frags = to;
3822 fragfrom = &skb_shinfo(skb)->frags[0];
3823 fragto = &skb_shinfo(tgt)->frags[merge];
3825 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3826 __skb_frag_unref(fragfrom, skb->pp_recycle);
3829 /* Reposition in the original skb */
3831 while (from < skb_shinfo(skb)->nr_frags)
3832 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3833 skb_shinfo(skb)->nr_frags = to;
3835 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3838 /* Most likely the tgt won't ever need its checksum anymore, skb on
3839 * the other hand might need it if it needs to be resent
3841 tgt->ip_summed = CHECKSUM_PARTIAL;
3842 skb->ip_summed = CHECKSUM_PARTIAL;
3844 skb_len_add(skb, -shiftlen);
3845 skb_len_add(tgt, shiftlen);
3851 * skb_prepare_seq_read - Prepare a sequential read of skb data
3852 * @skb: the buffer to read
3853 * @from: lower offset of data to be read
3854 * @to: upper offset of data to be read
3855 * @st: state variable
3857 * Initializes the specified state variable. Must be called before
3858 * invoking skb_seq_read() for the first time.
3860 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3861 unsigned int to, struct skb_seq_state *st)
3863 st->lower_offset = from;
3864 st->upper_offset = to;
3865 st->root_skb = st->cur_skb = skb;
3866 st->frag_idx = st->stepped_offset = 0;
3867 st->frag_data = NULL;
3870 EXPORT_SYMBOL(skb_prepare_seq_read);
3873 * skb_seq_read - Sequentially read skb data
3874 * @consumed: number of bytes consumed by the caller so far
3875 * @data: destination pointer for data to be returned
3876 * @st: state variable
3878 * Reads a block of skb data at @consumed relative to the
3879 * lower offset specified to skb_prepare_seq_read(). Assigns
3880 * the head of the data block to @data and returns the length
3881 * of the block or 0 if the end of the skb data or the upper
3882 * offset has been reached.
3884 * The caller is not required to consume all of the data
3885 * returned, i.e. @consumed is typically set to the number
3886 * of bytes already consumed and the next call to
3887 * skb_seq_read() will return the remaining part of the block.
3889 * Note 1: The size of each block of data returned can be arbitrary,
3890 * this limitation is the cost for zerocopy sequential
3891 * reads of potentially non linear data.
3893 * Note 2: Fragment lists within fragments are not implemented
3894 * at the moment, state->root_skb could be replaced with
3895 * a stack for this purpose.
3897 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3898 struct skb_seq_state *st)
3900 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3903 if (unlikely(abs_offset >= st->upper_offset)) {
3904 if (st->frag_data) {
3905 kunmap_atomic(st->frag_data);
3906 st->frag_data = NULL;
3912 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3914 if (abs_offset < block_limit && !st->frag_data) {
3915 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3916 return block_limit - abs_offset;
3919 if (st->frag_idx == 0 && !st->frag_data)
3920 st->stepped_offset += skb_headlen(st->cur_skb);
3922 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3923 unsigned int pg_idx, pg_off, pg_sz;
3925 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3928 pg_off = skb_frag_off(frag);
3929 pg_sz = skb_frag_size(frag);
3931 if (skb_frag_must_loop(skb_frag_page(frag))) {
3932 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3933 pg_off = offset_in_page(pg_off + st->frag_off);
3934 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3935 PAGE_SIZE - pg_off);
3938 block_limit = pg_sz + st->stepped_offset;
3939 if (abs_offset < block_limit) {
3941 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3943 *data = (u8 *)st->frag_data + pg_off +
3944 (abs_offset - st->stepped_offset);
3946 return block_limit - abs_offset;
3949 if (st->frag_data) {
3950 kunmap_atomic(st->frag_data);
3951 st->frag_data = NULL;
3954 st->stepped_offset += pg_sz;
3955 st->frag_off += pg_sz;
3956 if (st->frag_off == skb_frag_size(frag)) {
3962 if (st->frag_data) {
3963 kunmap_atomic(st->frag_data);
3964 st->frag_data = NULL;
3967 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3968 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3971 } else if (st->cur_skb->next) {
3972 st->cur_skb = st->cur_skb->next;
3979 EXPORT_SYMBOL(skb_seq_read);
3982 * skb_abort_seq_read - Abort a sequential read of skb data
3983 * @st: state variable
3985 * Must be called if skb_seq_read() was not called until it
3988 void skb_abort_seq_read(struct skb_seq_state *st)
3991 kunmap_atomic(st->frag_data);
3993 EXPORT_SYMBOL(skb_abort_seq_read);
3995 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3997 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3998 struct ts_config *conf,
3999 struct ts_state *state)
4001 return skb_seq_read(offset, text, TS_SKB_CB(state));
4004 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
4006 skb_abort_seq_read(TS_SKB_CB(state));
4010 * skb_find_text - Find a text pattern in skb data
4011 * @skb: the buffer to look in
4012 * @from: search offset
4014 * @config: textsearch configuration
4016 * Finds a pattern in the skb data according to the specified
4017 * textsearch configuration. Use textsearch_next() to retrieve
4018 * subsequent occurrences of the pattern. Returns the offset
4019 * to the first occurrence or UINT_MAX if no match was found.
4021 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4022 unsigned int to, struct ts_config *config)
4024 struct ts_state state;
4027 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4029 config->get_next_block = skb_ts_get_next_block;
4030 config->finish = skb_ts_finish;
4032 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4034 ret = textsearch_find(config, &state);
4035 return (ret <= to - from ? ret : UINT_MAX);
4037 EXPORT_SYMBOL(skb_find_text);
4039 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4040 int offset, size_t size)
4042 int i = skb_shinfo(skb)->nr_frags;
4044 if (skb_can_coalesce(skb, i, page, offset)) {
4045 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4046 } else if (i < MAX_SKB_FRAGS) {
4047 skb_zcopy_downgrade_managed(skb);
4049 skb_fill_page_desc_noacc(skb, i, page, offset, size);
4056 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4059 * skb_pull_rcsum - pull skb and update receive checksum
4060 * @skb: buffer to update
4061 * @len: length of data pulled
4063 * This function performs an skb_pull on the packet and updates
4064 * the CHECKSUM_COMPLETE checksum. It should be used on
4065 * receive path processing instead of skb_pull unless you know
4066 * that the checksum difference is zero (e.g., a valid IP header)
4067 * or you are setting ip_summed to CHECKSUM_NONE.
4069 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4071 unsigned char *data = skb->data;
4073 BUG_ON(len > skb->len);
4074 __skb_pull(skb, len);
4075 skb_postpull_rcsum(skb, data, len);
4078 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4080 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4082 skb_frag_t head_frag;
4085 page = virt_to_head_page(frag_skb->head);
4086 __skb_frag_set_page(&head_frag, page);
4087 skb_frag_off_set(&head_frag, frag_skb->data -
4088 (unsigned char *)page_address(page));
4089 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
4093 struct sk_buff *skb_segment_list(struct sk_buff *skb,
4094 netdev_features_t features,
4095 unsigned int offset)
4097 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4098 unsigned int tnl_hlen = skb_tnl_header_len(skb);
4099 unsigned int delta_truesize = 0;
4100 unsigned int delta_len = 0;
4101 struct sk_buff *tail = NULL;
4102 struct sk_buff *nskb, *tmp;
4105 skb_push(skb, -skb_network_offset(skb) + offset);
4107 skb_shinfo(skb)->frag_list = NULL;
4111 list_skb = list_skb->next;
4114 delta_truesize += nskb->truesize;
4115 if (skb_shared(nskb)) {
4116 tmp = skb_clone(nskb, GFP_ATOMIC);
4120 err = skb_unclone(nskb, GFP_ATOMIC);
4131 if (unlikely(err)) {
4132 nskb->next = list_skb;
4138 delta_len += nskb->len;
4140 skb_push(nskb, -skb_network_offset(nskb) + offset);
4142 skb_release_head_state(nskb);
4143 len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4144 __copy_skb_header(nskb, skb);
4146 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4147 nskb->transport_header += len_diff;
4148 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4149 nskb->data - tnl_hlen,
4152 if (skb_needs_linearize(nskb, features) &&
4153 __skb_linearize(nskb))
4158 skb->truesize = skb->truesize - delta_truesize;
4159 skb->data_len = skb->data_len - delta_len;
4160 skb->len = skb->len - delta_len;
4166 if (skb_needs_linearize(skb, features) &&
4167 __skb_linearize(skb))
4175 kfree_skb_list(skb->next);
4177 return ERR_PTR(-ENOMEM);
4179 EXPORT_SYMBOL_GPL(skb_segment_list);
4182 * skb_segment - Perform protocol segmentation on skb.
4183 * @head_skb: buffer to segment
4184 * @features: features for the output path (see dev->features)
4186 * This function performs segmentation on the given skb. It returns
4187 * a pointer to the first in a list of new skbs for the segments.
4188 * In case of error it returns ERR_PTR(err).
4190 struct sk_buff *skb_segment(struct sk_buff *head_skb,
4191 netdev_features_t features)
4193 struct sk_buff *segs = NULL;
4194 struct sk_buff *tail = NULL;
4195 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4196 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
4197 unsigned int mss = skb_shinfo(head_skb)->gso_size;
4198 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4199 struct sk_buff *frag_skb = head_skb;
4200 unsigned int offset = doffset;
4201 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4202 unsigned int partial_segs = 0;
4203 unsigned int headroom;
4204 unsigned int len = head_skb->len;
4207 int nfrags = skb_shinfo(head_skb)->nr_frags;
4212 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4213 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4214 struct sk_buff *check_skb;
4216 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4217 if (skb_headlen(check_skb) && !check_skb->head_frag) {
4218 /* gso_size is untrusted, and we have a frag_list with
4219 * a linear non head_frag item.
4221 * If head_skb's headlen does not fit requested gso_size,
4222 * it means that the frag_list members do NOT terminate
4223 * on exact gso_size boundaries. Hence we cannot perform
4224 * skb_frag_t page sharing. Therefore we must fallback to
4225 * copying the frag_list skbs; we do so by disabling SG.
4227 features &= ~NETIF_F_SG;
4233 __skb_push(head_skb, doffset);
4234 proto = skb_network_protocol(head_skb, NULL);
4235 if (unlikely(!proto))
4236 return ERR_PTR(-EINVAL);
4238 sg = !!(features & NETIF_F_SG);
4239 csum = !!can_checksum_protocol(features, proto);
4241 if (sg && csum && (mss != GSO_BY_FRAGS)) {
4242 if (!(features & NETIF_F_GSO_PARTIAL)) {
4243 struct sk_buff *iter;
4244 unsigned int frag_len;
4247 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4250 /* If we get here then all the required
4251 * GSO features except frag_list are supported.
4252 * Try to split the SKB to multiple GSO SKBs
4253 * with no frag_list.
4254 * Currently we can do that only when the buffers don't
4255 * have a linear part and all the buffers except
4256 * the last are of the same length.
4258 frag_len = list_skb->len;
4259 skb_walk_frags(head_skb, iter) {
4260 if (frag_len != iter->len && iter->next)
4262 if (skb_headlen(iter) && !iter->head_frag)
4268 if (len != frag_len)
4272 /* GSO partial only requires that we trim off any excess that
4273 * doesn't fit into an MSS sized block, so take care of that
4276 partial_segs = len / mss;
4277 if (partial_segs > 1)
4278 mss *= partial_segs;
4284 headroom = skb_headroom(head_skb);
4285 pos = skb_headlen(head_skb);
4288 struct sk_buff *nskb;
4289 skb_frag_t *nskb_frag;
4293 if (unlikely(mss == GSO_BY_FRAGS)) {
4294 len = list_skb->len;
4296 len = head_skb->len - offset;
4301 hsize = skb_headlen(head_skb) - offset;
4303 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4304 (skb_headlen(list_skb) == len || sg)) {
4305 BUG_ON(skb_headlen(list_skb) > len);
4308 nfrags = skb_shinfo(list_skb)->nr_frags;
4309 frag = skb_shinfo(list_skb)->frags;
4310 frag_skb = list_skb;
4311 pos += skb_headlen(list_skb);
4313 while (pos < offset + len) {
4314 BUG_ON(i >= nfrags);
4316 size = skb_frag_size(frag);
4317 if (pos + size > offset + len)
4325 nskb = skb_clone(list_skb, GFP_ATOMIC);
4326 list_skb = list_skb->next;
4328 if (unlikely(!nskb))
4331 if (unlikely(pskb_trim(nskb, len))) {
4336 hsize = skb_end_offset(nskb);
4337 if (skb_cow_head(nskb, doffset + headroom)) {
4342 nskb->truesize += skb_end_offset(nskb) - hsize;
4343 skb_release_head_state(nskb);
4344 __skb_push(nskb, doffset);
4348 if (hsize > len || !sg)
4351 nskb = __alloc_skb(hsize + doffset + headroom,
4352 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4355 if (unlikely(!nskb))
4358 skb_reserve(nskb, headroom);
4359 __skb_put(nskb, doffset);
4368 __copy_skb_header(nskb, head_skb);
4370 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4371 skb_reset_mac_len(nskb);
4373 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4374 nskb->data - tnl_hlen,
4375 doffset + tnl_hlen);
4377 if (nskb->len == len + doffset)
4378 goto perform_csum_check;
4382 if (!nskb->remcsum_offload)
4383 nskb->ip_summed = CHECKSUM_NONE;
4384 SKB_GSO_CB(nskb)->csum =
4385 skb_copy_and_csum_bits(head_skb, offset,
4389 SKB_GSO_CB(nskb)->csum_start =
4390 skb_headroom(nskb) + doffset;
4392 if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4398 nskb_frag = skb_shinfo(nskb)->frags;
4400 skb_copy_from_linear_data_offset(head_skb, offset,
4401 skb_put(nskb, hsize), hsize);
4403 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4406 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4407 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4410 while (pos < offset + len) {
4413 nfrags = skb_shinfo(list_skb)->nr_frags;
4414 frag = skb_shinfo(list_skb)->frags;
4415 frag_skb = list_skb;
4416 if (!skb_headlen(list_skb)) {
4419 BUG_ON(!list_skb->head_frag);
4421 /* to make room for head_frag. */
4425 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4426 skb_zerocopy_clone(nskb, frag_skb,
4430 list_skb = list_skb->next;
4433 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4435 net_warn_ratelimited(
4436 "skb_segment: too many frags: %u %u\n",
4442 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4443 __skb_frag_ref(nskb_frag);
4444 size = skb_frag_size(nskb_frag);
4447 skb_frag_off_add(nskb_frag, offset - pos);
4448 skb_frag_size_sub(nskb_frag, offset - pos);
4451 skb_shinfo(nskb)->nr_frags++;
4453 if (pos + size <= offset + len) {
4458 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4466 nskb->data_len = len - hsize;
4467 nskb->len += nskb->data_len;
4468 nskb->truesize += nskb->data_len;
4472 if (skb_has_shared_frag(nskb) &&
4473 __skb_linearize(nskb))
4476 if (!nskb->remcsum_offload)
4477 nskb->ip_summed = CHECKSUM_NONE;
4478 SKB_GSO_CB(nskb)->csum =
4479 skb_checksum(nskb, doffset,
4480 nskb->len - doffset, 0);
4481 SKB_GSO_CB(nskb)->csum_start =
4482 skb_headroom(nskb) + doffset;
4484 } while ((offset += len) < head_skb->len);
4486 /* Some callers want to get the end of the list.
4487 * Put it in segs->prev to avoid walking the list.
4488 * (see validate_xmit_skb_list() for example)
4493 struct sk_buff *iter;
4494 int type = skb_shinfo(head_skb)->gso_type;
4495 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4497 /* Update type to add partial and then remove dodgy if set */
4498 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4499 type &= ~SKB_GSO_DODGY;
4501 /* Update GSO info and prepare to start updating headers on
4502 * our way back down the stack of protocols.
4504 for (iter = segs; iter; iter = iter->next) {
4505 skb_shinfo(iter)->gso_size = gso_size;
4506 skb_shinfo(iter)->gso_segs = partial_segs;
4507 skb_shinfo(iter)->gso_type = type;
4508 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4511 if (tail->len - doffset <= gso_size)
4512 skb_shinfo(tail)->gso_size = 0;
4513 else if (tail != segs)
4514 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4517 /* Following permits correct backpressure, for protocols
4518 * using skb_set_owner_w().
4519 * Idea is to tranfert ownership from head_skb to last segment.
4521 if (head_skb->destructor == sock_wfree) {
4522 swap(tail->truesize, head_skb->truesize);
4523 swap(tail->destructor, head_skb->destructor);
4524 swap(tail->sk, head_skb->sk);
4529 kfree_skb_list(segs);
4530 return ERR_PTR(err);
4532 EXPORT_SYMBOL_GPL(skb_segment);
4534 #ifdef CONFIG_SKB_EXTENSIONS
4535 #define SKB_EXT_ALIGN_VALUE 8
4536 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4538 static const u8 skb_ext_type_len[] = {
4539 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4540 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4543 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4545 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4546 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4548 #if IS_ENABLED(CONFIG_MPTCP)
4549 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4551 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4552 [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
4556 static __always_inline unsigned int skb_ext_total_length(void)
4558 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4559 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4560 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4563 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4565 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4566 skb_ext_type_len[TC_SKB_EXT] +
4568 #if IS_ENABLED(CONFIG_MPTCP)
4569 skb_ext_type_len[SKB_EXT_MPTCP] +
4571 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
4572 skb_ext_type_len[SKB_EXT_MCTP] +
4577 static void skb_extensions_init(void)
4579 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4580 BUILD_BUG_ON(skb_ext_total_length() > 255);
4582 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4583 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4585 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4589 static void skb_extensions_init(void) {}
4592 void __init skb_init(void)
4594 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4595 sizeof(struct sk_buff),
4597 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4598 offsetof(struct sk_buff, cb),
4599 sizeof_field(struct sk_buff, cb),
4601 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4602 sizeof(struct sk_buff_fclones),
4604 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4606 skb_extensions_init();
4610 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4611 unsigned int recursion_level)
4613 int start = skb_headlen(skb);
4614 int i, copy = start - offset;
4615 struct sk_buff *frag_iter;
4618 if (unlikely(recursion_level >= 24))
4624 sg_set_buf(sg, skb->data + offset, copy);
4626 if ((len -= copy) == 0)
4631 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4634 WARN_ON(start > offset + len);
4636 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4637 if ((copy = end - offset) > 0) {
4638 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4639 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4644 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4645 skb_frag_off(frag) + offset - start);
4654 skb_walk_frags(skb, frag_iter) {
4657 WARN_ON(start > offset + len);
4659 end = start + frag_iter->len;
4660 if ((copy = end - offset) > 0) {
4661 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4666 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4667 copy, recursion_level + 1);
4668 if (unlikely(ret < 0))
4671 if ((len -= copy) == 0)
4682 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4683 * @skb: Socket buffer containing the buffers to be mapped
4684 * @sg: The scatter-gather list to map into
4685 * @offset: The offset into the buffer's contents to start mapping
4686 * @len: Length of buffer space to be mapped
4688 * Fill the specified scatter-gather list with mappings/pointers into a
4689 * region of the buffer space attached to a socket buffer. Returns either
4690 * the number of scatterlist items used, or -EMSGSIZE if the contents
4693 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4695 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4700 sg_mark_end(&sg[nsg - 1]);
4704 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4706 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4707 * sglist without mark the sg which contain last skb data as the end.
4708 * So the caller can mannipulate sg list as will when padding new data after
4709 * the first call without calling sg_unmark_end to expend sg list.
4711 * Scenario to use skb_to_sgvec_nomark:
4713 * 2. skb_to_sgvec_nomark(payload1)
4714 * 3. skb_to_sgvec_nomark(payload2)
4716 * This is equivalent to:
4718 * 2. skb_to_sgvec(payload1)
4720 * 4. skb_to_sgvec(payload2)
4722 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4723 * is more preferable.
4725 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4726 int offset, int len)
4728 return __skb_to_sgvec(skb, sg, offset, len, 0);
4730 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4735 * skb_cow_data - Check that a socket buffer's data buffers are writable
4736 * @skb: The socket buffer to check.
4737 * @tailbits: Amount of trailing space to be added
4738 * @trailer: Returned pointer to the skb where the @tailbits space begins
4740 * Make sure that the data buffers attached to a socket buffer are
4741 * writable. If they are not, private copies are made of the data buffers
4742 * and the socket buffer is set to use these instead.
4744 * If @tailbits is given, make sure that there is space to write @tailbits
4745 * bytes of data beyond current end of socket buffer. @trailer will be
4746 * set to point to the skb in which this space begins.
4748 * The number of scatterlist elements required to completely map the
4749 * COW'd and extended socket buffer will be returned.
4751 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4755 struct sk_buff *skb1, **skb_p;
4757 /* If skb is cloned or its head is paged, reallocate
4758 * head pulling out all the pages (pages are considered not writable
4759 * at the moment even if they are anonymous).
4761 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4762 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4765 /* Easy case. Most of packets will go this way. */
4766 if (!skb_has_frag_list(skb)) {
4767 /* A little of trouble, not enough of space for trailer.
4768 * This should not happen, when stack is tuned to generate
4769 * good frames. OK, on miss we reallocate and reserve even more
4770 * space, 128 bytes is fair. */
4772 if (skb_tailroom(skb) < tailbits &&
4773 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4781 /* Misery. We are in troubles, going to mincer fragments... */
4784 skb_p = &skb_shinfo(skb)->frag_list;
4787 while ((skb1 = *skb_p) != NULL) {
4790 /* The fragment is partially pulled by someone,
4791 * this can happen on input. Copy it and everything
4794 if (skb_shared(skb1))
4797 /* If the skb is the last, worry about trailer. */
4799 if (skb1->next == NULL && tailbits) {
4800 if (skb_shinfo(skb1)->nr_frags ||
4801 skb_has_frag_list(skb1) ||
4802 skb_tailroom(skb1) < tailbits)
4803 ntail = tailbits + 128;
4809 skb_shinfo(skb1)->nr_frags ||
4810 skb_has_frag_list(skb1)) {
4811 struct sk_buff *skb2;
4813 /* Fuck, we are miserable poor guys... */
4815 skb2 = skb_copy(skb1, GFP_ATOMIC);
4817 skb2 = skb_copy_expand(skb1,
4821 if (unlikely(skb2 == NULL))
4825 skb_set_owner_w(skb2, skb1->sk);
4827 /* Looking around. Are we still alive?
4828 * OK, link new skb, drop old one */
4830 skb2->next = skb1->next;
4837 skb_p = &skb1->next;
4842 EXPORT_SYMBOL_GPL(skb_cow_data);
4844 static void sock_rmem_free(struct sk_buff *skb)
4846 struct sock *sk = skb->sk;
4848 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4851 static void skb_set_err_queue(struct sk_buff *skb)
4853 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4854 * So, it is safe to (mis)use it to mark skbs on the error queue.
4856 skb->pkt_type = PACKET_OUTGOING;
4857 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4861 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4863 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4865 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4866 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4871 skb->destructor = sock_rmem_free;
4872 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4873 skb_set_err_queue(skb);
4875 /* before exiting rcu section, make sure dst is refcounted */
4878 skb_queue_tail(&sk->sk_error_queue, skb);
4879 if (!sock_flag(sk, SOCK_DEAD))
4880 sk_error_report(sk);
4883 EXPORT_SYMBOL(sock_queue_err_skb);
4885 static bool is_icmp_err_skb(const struct sk_buff *skb)
4887 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4888 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4891 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4893 struct sk_buff_head *q = &sk->sk_error_queue;
4894 struct sk_buff *skb, *skb_next = NULL;
4895 bool icmp_next = false;
4896 unsigned long flags;
4898 spin_lock_irqsave(&q->lock, flags);
4899 skb = __skb_dequeue(q);
4900 if (skb && (skb_next = skb_peek(q))) {
4901 icmp_next = is_icmp_err_skb(skb_next);
4903 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4905 spin_unlock_irqrestore(&q->lock, flags);
4907 if (is_icmp_err_skb(skb) && !icmp_next)
4911 sk_error_report(sk);
4915 EXPORT_SYMBOL(sock_dequeue_err_skb);
4918 * skb_clone_sk - create clone of skb, and take reference to socket
4919 * @skb: the skb to clone
4921 * This function creates a clone of a buffer that holds a reference on
4922 * sk_refcnt. Buffers created via this function are meant to be
4923 * returned using sock_queue_err_skb, or free via kfree_skb.
4925 * When passing buffers allocated with this function to sock_queue_err_skb
4926 * it is necessary to wrap the call with sock_hold/sock_put in order to
4927 * prevent the socket from being released prior to being enqueued on
4928 * the sk_error_queue.
4930 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4932 struct sock *sk = skb->sk;
4933 struct sk_buff *clone;
4935 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4938 clone = skb_clone(skb, GFP_ATOMIC);
4945 clone->destructor = sock_efree;
4949 EXPORT_SYMBOL(skb_clone_sk);
4951 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4956 struct sock_exterr_skb *serr;
4959 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4961 serr = SKB_EXT_ERR(skb);
4962 memset(serr, 0, sizeof(*serr));
4963 serr->ee.ee_errno = ENOMSG;
4964 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4965 serr->ee.ee_info = tstype;
4966 serr->opt_stats = opt_stats;
4967 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4968 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4969 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4971 serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
4974 err = sock_queue_err_skb(sk, skb);
4980 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4984 if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
4987 read_lock_bh(&sk->sk_callback_lock);
4988 ret = sk->sk_socket && sk->sk_socket->file &&
4989 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4990 read_unlock_bh(&sk->sk_callback_lock);
4994 void skb_complete_tx_timestamp(struct sk_buff *skb,
4995 struct skb_shared_hwtstamps *hwtstamps)
4997 struct sock *sk = skb->sk;
4999 if (!skb_may_tx_timestamp(sk, false))
5002 /* Take a reference to prevent skb_orphan() from freeing the socket,
5003 * but only if the socket refcount is not zero.
5005 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5006 *skb_hwtstamps(skb) = *hwtstamps;
5007 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5015 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5017 void __skb_tstamp_tx(struct sk_buff *orig_skb,
5018 const struct sk_buff *ack_skb,
5019 struct skb_shared_hwtstamps *hwtstamps,
5020 struct sock *sk, int tstype)
5022 struct sk_buff *skb;
5023 bool tsonly, opt_stats = false;
5028 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5029 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5032 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5033 if (!skb_may_tx_timestamp(sk, tsonly))
5038 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5040 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5045 skb = alloc_skb(0, GFP_ATOMIC);
5047 skb = skb_clone(orig_skb, GFP_ATOMIC);
5053 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5055 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5059 *skb_hwtstamps(skb) = *hwtstamps;
5061 __net_timestamp(skb);
5063 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5065 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5067 void skb_tstamp_tx(struct sk_buff *orig_skb,
5068 struct skb_shared_hwtstamps *hwtstamps)
5070 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5073 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5075 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5077 struct sock *sk = skb->sk;
5078 struct sock_exterr_skb *serr;
5081 skb->wifi_acked_valid = 1;
5082 skb->wifi_acked = acked;
5084 serr = SKB_EXT_ERR(skb);
5085 memset(serr, 0, sizeof(*serr));
5086 serr->ee.ee_errno = ENOMSG;
5087 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5089 /* Take a reference to prevent skb_orphan() from freeing the socket,
5090 * but only if the socket refcount is not zero.
5092 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5093 err = sock_queue_err_skb(sk, skb);
5099 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5102 * skb_partial_csum_set - set up and verify partial csum values for packet
5103 * @skb: the skb to set
5104 * @start: the number of bytes after skb->data to start checksumming.
5105 * @off: the offset from start to place the checksum.
5107 * For untrusted partially-checksummed packets, we need to make sure the values
5108 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5110 * This function checks and sets those values and skb->ip_summed: if this
5111 * returns false you should drop the packet.
5113 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5115 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5116 u32 csum_start = skb_headroom(skb) + (u32)start;
5118 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
5119 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5120 start, off, skb_headroom(skb), skb_headlen(skb));
5123 skb->ip_summed = CHECKSUM_PARTIAL;
5124 skb->csum_start = csum_start;
5125 skb->csum_offset = off;
5126 skb_set_transport_header(skb, start);
5129 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5131 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5134 if (skb_headlen(skb) >= len)
5137 /* If we need to pullup then pullup to the max, so we
5138 * won't need to do it again.
5143 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5146 if (skb_headlen(skb) < len)
5152 #define MAX_TCP_HDR_LEN (15 * 4)
5154 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5155 typeof(IPPROTO_IP) proto,
5162 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5163 off + MAX_TCP_HDR_LEN);
5164 if (!err && !skb_partial_csum_set(skb, off,
5165 offsetof(struct tcphdr,
5168 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5171 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5172 off + sizeof(struct udphdr));
5173 if (!err && !skb_partial_csum_set(skb, off,
5174 offsetof(struct udphdr,
5177 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5180 return ERR_PTR(-EPROTO);
5183 /* This value should be large enough to cover a tagged ethernet header plus
5184 * maximally sized IP and TCP or UDP headers.
5186 #define MAX_IP_HDR_LEN 128
5188 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5197 err = skb_maybe_pull_tail(skb,
5198 sizeof(struct iphdr),
5203 if (ip_is_fragment(ip_hdr(skb)))
5206 off = ip_hdrlen(skb);
5213 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5215 return PTR_ERR(csum);
5218 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5221 ip_hdr(skb)->protocol, 0);
5228 /* This value should be large enough to cover a tagged ethernet header plus
5229 * an IPv6 header, all options, and a maximal TCP or UDP header.
5231 #define MAX_IPV6_HDR_LEN 256
5233 #define OPT_HDR(type, skb, off) \
5234 (type *)(skb_network_header(skb) + (off))
5236 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5249 off = sizeof(struct ipv6hdr);
5251 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5255 nexthdr = ipv6_hdr(skb)->nexthdr;
5257 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5258 while (off <= len && !done) {
5260 case IPPROTO_DSTOPTS:
5261 case IPPROTO_HOPOPTS:
5262 case IPPROTO_ROUTING: {
5263 struct ipv6_opt_hdr *hp;
5265 err = skb_maybe_pull_tail(skb,
5267 sizeof(struct ipv6_opt_hdr),
5272 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5273 nexthdr = hp->nexthdr;
5274 off += ipv6_optlen(hp);
5278 struct ip_auth_hdr *hp;
5280 err = skb_maybe_pull_tail(skb,
5282 sizeof(struct ip_auth_hdr),
5287 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5288 nexthdr = hp->nexthdr;
5289 off += ipv6_authlen(hp);
5292 case IPPROTO_FRAGMENT: {
5293 struct frag_hdr *hp;
5295 err = skb_maybe_pull_tail(skb,
5297 sizeof(struct frag_hdr),
5302 hp = OPT_HDR(struct frag_hdr, skb, off);
5304 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5307 nexthdr = hp->nexthdr;
5308 off += sizeof(struct frag_hdr);
5319 if (!done || fragment)
5322 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5324 return PTR_ERR(csum);
5327 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5328 &ipv6_hdr(skb)->daddr,
5329 skb->len - off, nexthdr, 0);
5337 * skb_checksum_setup - set up partial checksum offset
5338 * @skb: the skb to set up
5339 * @recalculate: if true the pseudo-header checksum will be recalculated
5341 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5345 switch (skb->protocol) {
5346 case htons(ETH_P_IP):
5347 err = skb_checksum_setup_ipv4(skb, recalculate);
5350 case htons(ETH_P_IPV6):
5351 err = skb_checksum_setup_ipv6(skb, recalculate);
5361 EXPORT_SYMBOL(skb_checksum_setup);
5364 * skb_checksum_maybe_trim - maybe trims the given skb
5365 * @skb: the skb to check
5366 * @transport_len: the data length beyond the network header
5368 * Checks whether the given skb has data beyond the given transport length.
5369 * If so, returns a cloned skb trimmed to this transport length.
5370 * Otherwise returns the provided skb. Returns NULL in error cases
5371 * (e.g. transport_len exceeds skb length or out-of-memory).
5373 * Caller needs to set the skb transport header and free any returned skb if it
5374 * differs from the provided skb.
5376 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5377 unsigned int transport_len)
5379 struct sk_buff *skb_chk;
5380 unsigned int len = skb_transport_offset(skb) + transport_len;
5385 else if (skb->len == len)
5388 skb_chk = skb_clone(skb, GFP_ATOMIC);
5392 ret = pskb_trim_rcsum(skb_chk, len);
5402 * skb_checksum_trimmed - validate checksum of an skb
5403 * @skb: the skb to check
5404 * @transport_len: the data length beyond the network header
5405 * @skb_chkf: checksum function to use
5407 * Applies the given checksum function skb_chkf to the provided skb.
5408 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5410 * If the skb has data beyond the given transport length, then a
5411 * trimmed & cloned skb is checked and returned.
5413 * Caller needs to set the skb transport header and free any returned skb if it
5414 * differs from the provided skb.
5416 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5417 unsigned int transport_len,
5418 __sum16(*skb_chkf)(struct sk_buff *skb))
5420 struct sk_buff *skb_chk;
5421 unsigned int offset = skb_transport_offset(skb);
5424 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5428 if (!pskb_may_pull(skb_chk, offset))
5431 skb_pull_rcsum(skb_chk, offset);
5432 ret = skb_chkf(skb_chk);
5433 skb_push_rcsum(skb_chk, offset);
5441 if (skb_chk && skb_chk != skb)
5447 EXPORT_SYMBOL(skb_checksum_trimmed);
5449 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5451 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5454 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5456 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5459 skb_release_head_state(skb);
5460 kmem_cache_free(skbuff_head_cache, skb);
5465 EXPORT_SYMBOL(kfree_skb_partial);
5468 * skb_try_coalesce - try to merge skb to prior one
5470 * @from: buffer to add
5471 * @fragstolen: pointer to boolean
5472 * @delta_truesize: how much more was allocated than was requested
5474 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5475 bool *fragstolen, int *delta_truesize)
5477 struct skb_shared_info *to_shinfo, *from_shinfo;
5478 int i, delta, len = from->len;
5480 *fragstolen = false;
5485 /* In general, avoid mixing slab allocated and page_pool allocated
5486 * pages within the same SKB. However when @to is not pp_recycle and
5487 * @from is cloned, we can transition frag pages from page_pool to
5488 * reference counted.
5490 * On the other hand, don't allow coalescing two pp_recycle SKBs if
5491 * @from is cloned, in case the SKB is using page_pool fragment
5492 * references (PP_FLAG_PAGE_FRAG). Since we only take full page
5493 * references for cloned SKBs at the moment that would result in
5494 * inconsistent reference counts.
5496 if (to->pp_recycle != (from->pp_recycle && !skb_cloned(from)))
5499 if (len <= skb_tailroom(to)) {
5501 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5502 *delta_truesize = 0;
5506 to_shinfo = skb_shinfo(to);
5507 from_shinfo = skb_shinfo(from);
5508 if (to_shinfo->frag_list || from_shinfo->frag_list)
5510 if (skb_zcopy(to) || skb_zcopy(from))
5513 if (skb_headlen(from) != 0) {
5515 unsigned int offset;
5517 if (to_shinfo->nr_frags +
5518 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5521 if (skb_head_is_locked(from))
5524 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5526 page = virt_to_head_page(from->head);
5527 offset = from->data - (unsigned char *)page_address(page);
5529 skb_fill_page_desc(to, to_shinfo->nr_frags,
5530 page, offset, skb_headlen(from));
5533 if (to_shinfo->nr_frags +
5534 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5537 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5540 WARN_ON_ONCE(delta < len);
5542 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5544 from_shinfo->nr_frags * sizeof(skb_frag_t));
5545 to_shinfo->nr_frags += from_shinfo->nr_frags;
5547 if (!skb_cloned(from))
5548 from_shinfo->nr_frags = 0;
5550 /* if the skb is not cloned this does nothing
5551 * since we set nr_frags to 0.
5553 for (i = 0; i < from_shinfo->nr_frags; i++)
5554 __skb_frag_ref(&from_shinfo->frags[i]);
5556 to->truesize += delta;
5558 to->data_len += len;
5560 *delta_truesize = delta;
5563 EXPORT_SYMBOL(skb_try_coalesce);
5566 * skb_scrub_packet - scrub an skb
5568 * @skb: buffer to clean
5569 * @xnet: packet is crossing netns
5571 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5572 * into/from a tunnel. Some information have to be cleared during these
5574 * skb_scrub_packet can also be used to clean a skb before injecting it in
5575 * another namespace (@xnet == true). We have to clear all information in the
5576 * skb that could impact namespace isolation.
5578 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5580 skb->pkt_type = PACKET_HOST;
5586 nf_reset_trace(skb);
5588 #ifdef CONFIG_NET_SWITCHDEV
5589 skb->offload_fwd_mark = 0;
5590 skb->offload_l3_fwd_mark = 0;
5598 skb_clear_tstamp(skb);
5600 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5603 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5607 * skb_gso_transport_seglen is used to determine the real size of the
5608 * individual segments, including Layer4 headers (TCP/UDP).
5610 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5612 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5614 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5615 unsigned int thlen = 0;
5617 if (skb->encapsulation) {
5618 thlen = skb_inner_transport_header(skb) -
5619 skb_transport_header(skb);
5621 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5622 thlen += inner_tcp_hdrlen(skb);
5623 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5624 thlen = tcp_hdrlen(skb);
5625 } else if (unlikely(skb_is_gso_sctp(skb))) {
5626 thlen = sizeof(struct sctphdr);
5627 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5628 thlen = sizeof(struct udphdr);
5630 /* UFO sets gso_size to the size of the fragmentation
5631 * payload, i.e. the size of the L4 (UDP) header is already
5634 return thlen + shinfo->gso_size;
5638 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5642 * skb_gso_network_seglen is used to determine the real size of the
5643 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5645 * The MAC/L2 header is not accounted for.
5647 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5649 unsigned int hdr_len = skb_transport_header(skb) -
5650 skb_network_header(skb);
5652 return hdr_len + skb_gso_transport_seglen(skb);
5656 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5660 * skb_gso_mac_seglen is used to determine the real size of the
5661 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5662 * headers (TCP/UDP).
5664 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5666 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5668 return hdr_len + skb_gso_transport_seglen(skb);
5672 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5674 * There are a couple of instances where we have a GSO skb, and we
5675 * want to determine what size it would be after it is segmented.
5677 * We might want to check:
5678 * - L3+L4+payload size (e.g. IP forwarding)
5679 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5681 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5685 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5686 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5688 * @max_len: The maximum permissible length.
5690 * Returns true if the segmented length <= max length.
5692 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5693 unsigned int seg_len,
5694 unsigned int max_len) {
5695 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5696 const struct sk_buff *iter;
5698 if (shinfo->gso_size != GSO_BY_FRAGS)
5699 return seg_len <= max_len;
5701 /* Undo this so we can re-use header sizes */
5702 seg_len -= GSO_BY_FRAGS;
5704 skb_walk_frags(skb, iter) {
5705 if (seg_len + skb_headlen(iter) > max_len)
5713 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5716 * @mtu: MTU to validate against
5718 * skb_gso_validate_network_len validates if a given skb will fit a
5719 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5722 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5724 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5726 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5729 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5732 * @len: length to validate against
5734 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5735 * length once split, including L2, L3 and L4 headers and the payload.
5737 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5739 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5741 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5743 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5745 int mac_len, meta_len;
5748 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5753 mac_len = skb->data - skb_mac_header(skb);
5754 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5755 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5756 mac_len - VLAN_HLEN - ETH_TLEN);
5759 meta_len = skb_metadata_len(skb);
5761 meta = skb_metadata_end(skb) - meta_len;
5762 memmove(meta + VLAN_HLEN, meta, meta_len);
5765 skb->mac_header += VLAN_HLEN;
5769 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5771 struct vlan_hdr *vhdr;
5774 if (unlikely(skb_vlan_tag_present(skb))) {
5775 /* vlan_tci is already set-up so leave this for another time */
5779 skb = skb_share_check(skb, GFP_ATOMIC);
5782 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5783 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5786 vhdr = (struct vlan_hdr *)skb->data;
5787 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5788 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5790 skb_pull_rcsum(skb, VLAN_HLEN);
5791 vlan_set_encap_proto(skb, vhdr);
5793 skb = skb_reorder_vlan_header(skb);
5797 skb_reset_network_header(skb);
5798 if (!skb_transport_header_was_set(skb))
5799 skb_reset_transport_header(skb);
5800 skb_reset_mac_len(skb);
5808 EXPORT_SYMBOL(skb_vlan_untag);
5810 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
5812 if (!pskb_may_pull(skb, write_len))
5815 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5818 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5820 EXPORT_SYMBOL(skb_ensure_writable);
5822 /* remove VLAN header from packet and update csum accordingly.
5823 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5825 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5827 struct vlan_hdr *vhdr;
5828 int offset = skb->data - skb_mac_header(skb);
5831 if (WARN_ONCE(offset,
5832 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5837 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5841 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5843 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5844 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5846 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5847 __skb_pull(skb, VLAN_HLEN);
5849 vlan_set_encap_proto(skb, vhdr);
5850 skb->mac_header += VLAN_HLEN;
5852 if (skb_network_offset(skb) < ETH_HLEN)
5853 skb_set_network_header(skb, ETH_HLEN);
5855 skb_reset_mac_len(skb);
5859 EXPORT_SYMBOL(__skb_vlan_pop);
5861 /* Pop a vlan tag either from hwaccel or from payload.
5862 * Expects skb->data at mac header.
5864 int skb_vlan_pop(struct sk_buff *skb)
5870 if (likely(skb_vlan_tag_present(skb))) {
5871 __vlan_hwaccel_clear_tag(skb);
5873 if (unlikely(!eth_type_vlan(skb->protocol)))
5876 err = __skb_vlan_pop(skb, &vlan_tci);
5880 /* move next vlan tag to hw accel tag */
5881 if (likely(!eth_type_vlan(skb->protocol)))
5884 vlan_proto = skb->protocol;
5885 err = __skb_vlan_pop(skb, &vlan_tci);
5889 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5892 EXPORT_SYMBOL(skb_vlan_pop);
5894 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5895 * Expects skb->data at mac header.
5897 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5899 if (skb_vlan_tag_present(skb)) {
5900 int offset = skb->data - skb_mac_header(skb);
5903 if (WARN_ONCE(offset,
5904 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5909 err = __vlan_insert_tag(skb, skb->vlan_proto,
5910 skb_vlan_tag_get(skb));
5914 skb->protocol = skb->vlan_proto;
5915 skb->mac_len += VLAN_HLEN;
5917 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5919 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5922 EXPORT_SYMBOL(skb_vlan_push);
5925 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5927 * @skb: Socket buffer to modify
5929 * Drop the Ethernet header of @skb.
5931 * Expects that skb->data points to the mac header and that no VLAN tags are
5934 * Returns 0 on success, -errno otherwise.
5936 int skb_eth_pop(struct sk_buff *skb)
5938 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5939 skb_network_offset(skb) < ETH_HLEN)
5942 skb_pull_rcsum(skb, ETH_HLEN);
5943 skb_reset_mac_header(skb);
5944 skb_reset_mac_len(skb);
5948 EXPORT_SYMBOL(skb_eth_pop);
5951 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5953 * @skb: Socket buffer to modify
5954 * @dst: Destination MAC address of the new header
5955 * @src: Source MAC address of the new header
5957 * Prepend @skb with a new Ethernet header.
5959 * Expects that skb->data points to the mac header, which must be empty.
5961 * Returns 0 on success, -errno otherwise.
5963 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5964 const unsigned char *src)
5969 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5972 err = skb_cow_head(skb, sizeof(*eth));
5976 skb_push(skb, sizeof(*eth));
5977 skb_reset_mac_header(skb);
5978 skb_reset_mac_len(skb);
5981 ether_addr_copy(eth->h_dest, dst);
5982 ether_addr_copy(eth->h_source, src);
5983 eth->h_proto = skb->protocol;
5985 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5989 EXPORT_SYMBOL(skb_eth_push);
5991 /* Update the ethertype of hdr and the skb csum value if required. */
5992 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5995 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5996 __be16 diff[] = { ~hdr->h_proto, ethertype };
5998 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6001 hdr->h_proto = ethertype;
6005 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6009 * @mpls_lse: MPLS label stack entry to push
6010 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6011 * @mac_len: length of the MAC header
6012 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6015 * Expects skb->data at mac header.
6017 * Returns 0 on success, -errno otherwise.
6019 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6020 int mac_len, bool ethernet)
6022 struct mpls_shim_hdr *lse;
6025 if (unlikely(!eth_p_mpls(mpls_proto)))
6028 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6029 if (skb->encapsulation)
6032 err = skb_cow_head(skb, MPLS_HLEN);
6036 if (!skb->inner_protocol) {
6037 skb_set_inner_network_header(skb, skb_network_offset(skb));
6038 skb_set_inner_protocol(skb, skb->protocol);
6041 skb_push(skb, MPLS_HLEN);
6042 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6044 skb_reset_mac_header(skb);
6045 skb_set_network_header(skb, mac_len);
6046 skb_reset_mac_len(skb);
6048 lse = mpls_hdr(skb);
6049 lse->label_stack_entry = mpls_lse;
6050 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6052 if (ethernet && mac_len >= ETH_HLEN)
6053 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6054 skb->protocol = mpls_proto;
6058 EXPORT_SYMBOL_GPL(skb_mpls_push);
6061 * skb_mpls_pop() - pop the outermost MPLS header
6064 * @next_proto: ethertype of header after popped MPLS header
6065 * @mac_len: length of the MAC header
6066 * @ethernet: flag to indicate if the packet is ethernet
6068 * Expects skb->data at mac header.
6070 * Returns 0 on success, -errno otherwise.
6072 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6077 if (unlikely(!eth_p_mpls(skb->protocol)))
6080 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6084 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6085 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6088 __skb_pull(skb, MPLS_HLEN);
6089 skb_reset_mac_header(skb);
6090 skb_set_network_header(skb, mac_len);
6092 if (ethernet && mac_len >= ETH_HLEN) {
6095 /* use mpls_hdr() to get ethertype to account for VLANs. */
6096 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6097 skb_mod_eth_type(skb, hdr, next_proto);
6099 skb->protocol = next_proto;
6103 EXPORT_SYMBOL_GPL(skb_mpls_pop);
6106 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6109 * @mpls_lse: new MPLS label stack entry to update to
6111 * Expects skb->data at mac header.
6113 * Returns 0 on success, -errno otherwise.
6115 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6119 if (unlikely(!eth_p_mpls(skb->protocol)))
6122 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6126 if (skb->ip_summed == CHECKSUM_COMPLETE) {
6127 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6129 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6132 mpls_hdr(skb)->label_stack_entry = mpls_lse;
6136 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6139 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6143 * Expects skb->data at mac header.
6145 * Returns 0 on success, -errno otherwise.
6147 int skb_mpls_dec_ttl(struct sk_buff *skb)
6152 if (unlikely(!eth_p_mpls(skb->protocol)))
6155 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6158 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6159 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6163 lse &= ~MPLS_LS_TTL_MASK;
6164 lse |= ttl << MPLS_LS_TTL_SHIFT;
6166 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6168 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6171 * alloc_skb_with_frags - allocate skb with page frags
6173 * @header_len: size of linear part
6174 * @data_len: needed length in frags
6175 * @max_page_order: max page order desired.
6176 * @errcode: pointer to error code if any
6177 * @gfp_mask: allocation mask
6179 * This can be used to allocate a paged skb, given a maximal order for frags.
6181 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6182 unsigned long data_len,
6187 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
6188 unsigned long chunk;
6189 struct sk_buff *skb;
6193 *errcode = -EMSGSIZE;
6194 /* Note this test could be relaxed, if we succeed to allocate
6195 * high order pages...
6197 if (npages > MAX_SKB_FRAGS)
6200 *errcode = -ENOBUFS;
6201 skb = alloc_skb(header_len, gfp_mask);
6205 skb->truesize += npages << PAGE_SHIFT;
6207 for (i = 0; npages > 0; i++) {
6208 int order = max_page_order;
6211 if (npages >= 1 << order) {
6212 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6218 /* Do not retry other high order allocations */
6224 page = alloc_page(gfp_mask);
6228 chunk = min_t(unsigned long, data_len,
6229 PAGE_SIZE << order);
6230 skb_fill_page_desc(skb, i, page, 0, chunk);
6232 npages -= 1 << order;
6240 EXPORT_SYMBOL(alloc_skb_with_frags);
6242 /* carve out the first off bytes from skb when off < headlen */
6243 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6244 const int headlen, gfp_t gfp_mask)
6247 unsigned int size = skb_end_offset(skb);
6248 int new_hlen = headlen - off;
6251 if (skb_pfmemalloc(skb))
6252 gfp_mask |= __GFP_MEMALLOC;
6254 size = SKB_DATA_ALIGN(size);
6255 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
6256 size = kmalloc_size_roundup(size);
6257 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
6260 size = SKB_WITH_OVERHEAD(size);
6262 /* Copy real data, and all frags */
6263 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6266 memcpy((struct skb_shared_info *)(data + size),
6268 offsetof(struct skb_shared_info,
6269 frags[skb_shinfo(skb)->nr_frags]));
6270 if (skb_cloned(skb)) {
6271 /* drop the old head gracefully */
6272 if (skb_orphan_frags(skb, gfp_mask)) {
6276 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6277 skb_frag_ref(skb, i);
6278 if (skb_has_frag_list(skb))
6279 skb_clone_fraglist(skb);
6280 skb_release_data(skb, SKB_CONSUMED);
6282 /* we can reuse existing recount- all we did was
6291 skb_set_end_offset(skb, size);
6292 skb_set_tail_pointer(skb, skb_headlen(skb));
6293 skb_headers_offset_update(skb, 0);
6297 atomic_set(&skb_shinfo(skb)->dataref, 1);
6302 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6304 /* carve out the first eat bytes from skb's frag_list. May recurse into
6307 static int pskb_carve_frag_list(struct sk_buff *skb,
6308 struct skb_shared_info *shinfo, int eat,
6311 struct sk_buff *list = shinfo->frag_list;
6312 struct sk_buff *clone = NULL;
6313 struct sk_buff *insp = NULL;
6317 pr_err("Not enough bytes to eat. Want %d\n", eat);
6320 if (list->len <= eat) {
6321 /* Eaten as whole. */
6326 /* Eaten partially. */
6327 if (skb_shared(list)) {
6328 clone = skb_clone(list, gfp_mask);
6334 /* This may be pulled without problems. */
6337 if (pskb_carve(list, eat, gfp_mask) < 0) {
6345 /* Free pulled out fragments. */
6346 while ((list = shinfo->frag_list) != insp) {
6347 shinfo->frag_list = list->next;
6350 /* And insert new clone at head. */
6353 shinfo->frag_list = clone;
6358 /* carve off first len bytes from skb. Split line (off) is in the
6359 * non-linear part of skb
6361 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6362 int pos, gfp_t gfp_mask)
6365 unsigned int size = skb_end_offset(skb);
6367 const int nfrags = skb_shinfo(skb)->nr_frags;
6368 struct skb_shared_info *shinfo;
6370 if (skb_pfmemalloc(skb))
6371 gfp_mask |= __GFP_MEMALLOC;
6373 size = SKB_DATA_ALIGN(size);
6374 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
6375 size = kmalloc_size_roundup(size);
6376 data = kmalloc_reserve(size, gfp_mask, NUMA_NO_NODE, NULL);
6379 size = SKB_WITH_OVERHEAD(size);
6381 memcpy((struct skb_shared_info *)(data + size),
6382 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6383 if (skb_orphan_frags(skb, gfp_mask)) {
6387 shinfo = (struct skb_shared_info *)(data + size);
6388 for (i = 0; i < nfrags; i++) {
6389 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6391 if (pos + fsize > off) {
6392 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6396 * We have two variants in this case:
6397 * 1. Move all the frag to the second
6398 * part, if it is possible. F.e.
6399 * this approach is mandatory for TUX,
6400 * where splitting is expensive.
6401 * 2. Split is accurately. We make this.
6403 skb_frag_off_add(&shinfo->frags[0], off - pos);
6404 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6406 skb_frag_ref(skb, i);
6411 shinfo->nr_frags = k;
6412 if (skb_has_frag_list(skb))
6413 skb_clone_fraglist(skb);
6415 /* split line is in frag list */
6416 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6417 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6418 if (skb_has_frag_list(skb))
6419 kfree_skb_list(skb_shinfo(skb)->frag_list);
6423 skb_release_data(skb, SKB_CONSUMED);
6428 skb_set_end_offset(skb, size);
6429 skb_reset_tail_pointer(skb);
6430 skb_headers_offset_update(skb, 0);
6435 skb->data_len = skb->len;
6436 atomic_set(&skb_shinfo(skb)->dataref, 1);
6440 /* remove len bytes from the beginning of the skb */
6441 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6443 int headlen = skb_headlen(skb);
6446 return pskb_carve_inside_header(skb, len, headlen, gfp);
6448 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6451 /* Extract to_copy bytes starting at off from skb, and return this in
6454 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6455 int to_copy, gfp_t gfp)
6457 struct sk_buff *clone = skb_clone(skb, gfp);
6462 if (pskb_carve(clone, off, gfp) < 0 ||
6463 pskb_trim(clone, to_copy)) {
6469 EXPORT_SYMBOL(pskb_extract);
6472 * skb_condense - try to get rid of fragments/frag_list if possible
6475 * Can be used to save memory before skb is added to a busy queue.
6476 * If packet has bytes in frags and enough tail room in skb->head,
6477 * pull all of them, so that we can free the frags right now and adjust
6480 * We do not reallocate skb->head thus can not fail.
6481 * Caller must re-evaluate skb->truesize if needed.
6483 void skb_condense(struct sk_buff *skb)
6485 if (skb->data_len) {
6486 if (skb->data_len > skb->end - skb->tail ||
6490 /* Nice, we can free page frag(s) right now */
6491 __pskb_pull_tail(skb, skb->data_len);
6493 /* At this point, skb->truesize might be over estimated,
6494 * because skb had a fragment, and fragments do not tell
6496 * When we pulled its content into skb->head, fragment
6497 * was freed, but __pskb_pull_tail() could not possibly
6498 * adjust skb->truesize, not knowing the frag truesize.
6500 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6502 EXPORT_SYMBOL(skb_condense);
6504 #ifdef CONFIG_SKB_EXTENSIONS
6505 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6507 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6511 * __skb_ext_alloc - allocate a new skb extensions storage
6513 * @flags: See kmalloc().
6515 * Returns the newly allocated pointer. The pointer can later attached to a
6516 * skb via __skb_ext_set().
6517 * Note: caller must handle the skb_ext as an opaque data.
6519 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6521 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6524 memset(new->offset, 0, sizeof(new->offset));
6525 refcount_set(&new->refcnt, 1);
6531 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6532 unsigned int old_active)
6534 struct skb_ext *new;
6536 if (refcount_read(&old->refcnt) == 1)
6539 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6543 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6544 refcount_set(&new->refcnt, 1);
6547 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6548 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6551 for (i = 0; i < sp->len; i++)
6552 xfrm_state_hold(sp->xvec[i]);
6560 * __skb_ext_set - attach the specified extension storage to this skb
6563 * @ext: extension storage previously allocated via __skb_ext_alloc()
6565 * Existing extensions, if any, are cleared.
6567 * Returns the pointer to the extension.
6569 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6570 struct skb_ext *ext)
6572 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6575 newlen = newoff + skb_ext_type_len[id];
6576 ext->chunks = newlen;
6577 ext->offset[id] = newoff;
6578 skb->extensions = ext;
6579 skb->active_extensions = 1 << id;
6580 return skb_ext_get_ptr(ext, id);
6584 * skb_ext_add - allocate space for given extension, COW if needed
6586 * @id: extension to allocate space for
6588 * Allocates enough space for the given extension.
6589 * If the extension is already present, a pointer to that extension
6592 * If the skb was cloned, COW applies and the returned memory can be
6593 * modified without changing the extension space of clones buffers.
6595 * Returns pointer to the extension or NULL on allocation failure.
6597 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6599 struct skb_ext *new, *old = NULL;
6600 unsigned int newlen, newoff;
6602 if (skb->active_extensions) {
6603 old = skb->extensions;
6605 new = skb_ext_maybe_cow(old, skb->active_extensions);
6609 if (__skb_ext_exist(new, id))
6612 newoff = new->chunks;
6614 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6616 new = __skb_ext_alloc(GFP_ATOMIC);
6621 newlen = newoff + skb_ext_type_len[id];
6622 new->chunks = newlen;
6623 new->offset[id] = newoff;
6626 skb->extensions = new;
6627 skb->active_extensions |= 1 << id;
6628 return skb_ext_get_ptr(new, id);
6630 EXPORT_SYMBOL(skb_ext_add);
6633 static void skb_ext_put_sp(struct sec_path *sp)
6637 for (i = 0; i < sp->len; i++)
6638 xfrm_state_put(sp->xvec[i]);
6642 #ifdef CONFIG_MCTP_FLOWS
6643 static void skb_ext_put_mctp(struct mctp_flow *flow)
6646 mctp_key_unref(flow->key);
6650 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6652 struct skb_ext *ext = skb->extensions;
6654 skb->active_extensions &= ~(1 << id);
6655 if (skb->active_extensions == 0) {
6656 skb->extensions = NULL;
6659 } else if (id == SKB_EXT_SEC_PATH &&
6660 refcount_read(&ext->refcnt) == 1) {
6661 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6668 EXPORT_SYMBOL(__skb_ext_del);
6670 void __skb_ext_put(struct skb_ext *ext)
6672 /* If this is last clone, nothing can increment
6673 * it after check passes. Avoids one atomic op.
6675 if (refcount_read(&ext->refcnt) == 1)
6678 if (!refcount_dec_and_test(&ext->refcnt))
6682 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6683 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6685 #ifdef CONFIG_MCTP_FLOWS
6686 if (__skb_ext_exist(ext, SKB_EXT_MCTP))
6687 skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
6690 kmem_cache_free(skbuff_ext_cache, ext);
6692 EXPORT_SYMBOL(__skb_ext_put);
6693 #endif /* CONFIG_SKB_EXTENSIONS */
6696 * skb_attempt_defer_free - queue skb for remote freeing
6699 * Put @skb in a per-cpu list, using the cpu which
6700 * allocated the skb/pages to reduce false sharing
6701 * and memory zone spinlock contention.
6703 void skb_attempt_defer_free(struct sk_buff *skb)
6705 int cpu = skb->alloc_cpu;
6706 struct softnet_data *sd;
6707 unsigned long flags;
6708 unsigned int defer_max;
6711 if (WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
6713 cpu == raw_smp_processor_id()) {
6714 nodefer: __kfree_skb(skb);
6718 sd = &per_cpu(softnet_data, cpu);
6719 defer_max = READ_ONCE(sysctl_skb_defer_max);
6720 if (READ_ONCE(sd->defer_count) >= defer_max)
6723 spin_lock_irqsave(&sd->defer_lock, flags);
6724 /* Send an IPI every time queue reaches half capacity. */
6725 kick = sd->defer_count == (defer_max >> 1);
6726 /* Paired with the READ_ONCE() few lines above */
6727 WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
6729 skb->next = sd->defer_list;
6730 /* Paired with READ_ONCE() in skb_defer_free_flush() */
6731 WRITE_ONCE(sd->defer_list, skb);
6732 spin_unlock_irqrestore(&sd->defer_lock, flags);
6734 /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
6735 * if we are unlucky enough (this seems very unlikely).
6737 if (unlikely(kick) && !cmpxchg(&sd->defer_ipi_scheduled, 0, 1))
6738 smp_call_function_single_async(cpu, &sd->defer_csd);