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>
63 #include <net/protocol.h>
66 #include <net/checksum.h>
67 #include <net/ip6_checksum.h>
71 #include <linux/uaccess.h>
72 #include <trace/events/skb.h>
73 #include <linux/highmem.h>
74 #include <linux/capability.h>
75 #include <linux/user_namespace.h>
76 #include <linux/indirect_call_wrapper.h>
80 struct kmem_cache *skbuff_head_cache __ro_after_init;
81 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
82 #ifdef CONFIG_SKB_EXTENSIONS
83 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
85 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
86 EXPORT_SYMBOL(sysctl_max_skb_frags);
89 * skb_panic - private function for out-of-line support
93 * @msg: skb_over_panic or skb_under_panic
95 * Out-of-line support for skb_put() and skb_push().
96 * Called via the wrapper skb_over_panic() or skb_under_panic().
97 * Keep out of line to prevent kernel bloat.
98 * __builtin_return_address is not used because it is not always reliable.
100 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
104 msg, addr, skb->len, sz, skb->head, skb->data,
105 (unsigned long)skb->tail, (unsigned long)skb->end,
106 skb->dev ? skb->dev->name : "<NULL>");
110 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
112 skb_panic(skb, sz, addr, __func__);
115 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
117 skb_panic(skb, sz, addr, __func__);
121 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
122 * the caller if emergency pfmemalloc reserves are being used. If it is and
123 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
124 * may be used. Otherwise, the packet data may be discarded until enough
127 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
128 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
130 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
131 unsigned long ip, bool *pfmemalloc)
134 bool ret_pfmemalloc = false;
137 * Try a regular allocation, when that fails and we're not entitled
138 * to the reserves, fail.
140 obj = kmalloc_node_track_caller(size,
141 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
143 if (obj || !(gfp_pfmemalloc_allowed(flags)))
146 /* Try again but now we are using pfmemalloc reserves */
147 ret_pfmemalloc = true;
148 obj = kmalloc_node_track_caller(size, flags, node);
152 *pfmemalloc = ret_pfmemalloc;
157 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
158 * 'private' fields and also do memory statistics to find all the
164 * __alloc_skb - allocate a network buffer
165 * @size: size to allocate
166 * @gfp_mask: allocation mask
167 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
168 * instead of head cache and allocate a cloned (child) skb.
169 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
170 * allocations in case the data is required for writeback
171 * @node: numa node to allocate memory on
173 * Allocate a new &sk_buff. The returned buffer has no headroom and a
174 * tail room of at least size bytes. The object has a reference count
175 * of one. The return is the buffer. On a failure the return is %NULL.
177 * Buffers may only be allocated from interrupts using a @gfp_mask of
180 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183 struct kmem_cache *cache;
184 struct skb_shared_info *shinfo;
189 cache = (flags & SKB_ALLOC_FCLONE)
190 ? skbuff_fclone_cache : skbuff_head_cache;
192 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
193 gfp_mask |= __GFP_MEMALLOC;
196 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
201 /* We do our best to align skb_shared_info on a separate cache
202 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
203 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
204 * Both skb->head and skb_shared_info are cache line aligned.
206 size = SKB_DATA_ALIGN(size);
207 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
208 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
211 /* kmalloc(size) might give us more room than requested.
212 * Put skb_shared_info exactly at the end of allocated zone,
213 * to allow max possible filling before reallocation.
215 size = SKB_WITH_OVERHEAD(ksize(data));
216 prefetchw(data + size);
219 * Only clear those fields we need to clear, not those that we will
220 * actually initialise below. Hence, don't put any more fields after
221 * the tail pointer in struct sk_buff!
223 memset(skb, 0, offsetof(struct sk_buff, tail));
224 /* Account for allocated memory : skb + skb->head */
225 skb->truesize = SKB_TRUESIZE(size);
226 skb->pfmemalloc = pfmemalloc;
227 refcount_set(&skb->users, 1);
230 skb_reset_tail_pointer(skb);
231 skb->end = skb->tail + size;
232 skb->mac_header = (typeof(skb->mac_header))~0U;
233 skb->transport_header = (typeof(skb->transport_header))~0U;
235 /* make sure we initialize shinfo sequentially */
236 shinfo = skb_shinfo(skb);
237 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
238 atomic_set(&shinfo->dataref, 1);
240 if (flags & SKB_ALLOC_FCLONE) {
241 struct sk_buff_fclones *fclones;
243 fclones = container_of(skb, struct sk_buff_fclones, skb1);
245 skb->fclone = SKB_FCLONE_ORIG;
246 refcount_set(&fclones->fclone_ref, 1);
248 fclones->skb2.fclone = SKB_FCLONE_CLONE;
253 kmem_cache_free(cache, skb);
257 EXPORT_SYMBOL(__alloc_skb);
259 /* Caller must provide SKB that is memset cleared */
260 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
261 void *data, unsigned int frag_size)
263 struct skb_shared_info *shinfo;
264 unsigned int size = frag_size ? : ksize(data);
266 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
268 /* Assumes caller memset cleared SKB */
269 skb->truesize = SKB_TRUESIZE(size);
270 refcount_set(&skb->users, 1);
273 skb_reset_tail_pointer(skb);
274 skb->end = skb->tail + size;
275 skb->mac_header = (typeof(skb->mac_header))~0U;
276 skb->transport_header = (typeof(skb->transport_header))~0U;
278 /* make sure we initialize shinfo sequentially */
279 shinfo = skb_shinfo(skb);
280 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
281 atomic_set(&shinfo->dataref, 1);
287 * __build_skb - build a network buffer
288 * @data: data buffer provided by caller
289 * @frag_size: size of data, or 0 if head was kmalloced
291 * Allocate a new &sk_buff. Caller provides space holding head and
292 * skb_shared_info. @data must have been allocated by kmalloc() only if
293 * @frag_size is 0, otherwise data should come from the page allocator
295 * The return is the new skb buffer.
296 * On a failure the return is %NULL, and @data is not freed.
298 * Before IO, driver allocates only data buffer where NIC put incoming frame
299 * Driver should add room at head (NET_SKB_PAD) and
300 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
301 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
302 * before giving packet to stack.
303 * RX rings only contains data buffers, not full skbs.
305 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
309 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
313 memset(skb, 0, offsetof(struct sk_buff, tail));
315 return __build_skb_around(skb, data, frag_size);
318 /* build_skb() is wrapper over __build_skb(), that specifically
319 * takes care of skb->head and skb->pfmemalloc
320 * This means that if @frag_size is not zero, then @data must be backed
321 * by a page fragment, not kmalloc() or vmalloc()
323 struct sk_buff *build_skb(void *data, unsigned int frag_size)
325 struct sk_buff *skb = __build_skb(data, frag_size);
327 if (skb && frag_size) {
329 if (page_is_pfmemalloc(virt_to_head_page(data)))
334 EXPORT_SYMBOL(build_skb);
337 * build_skb_around - build a network buffer around provided skb
338 * @skb: sk_buff provide by caller, must be memset cleared
339 * @data: data buffer provided by caller
340 * @frag_size: size of data, or 0 if head was kmalloced
342 struct sk_buff *build_skb_around(struct sk_buff *skb,
343 void *data, unsigned int frag_size)
348 skb = __build_skb_around(skb, data, frag_size);
350 if (skb && frag_size) {
352 if (page_is_pfmemalloc(virt_to_head_page(data)))
357 EXPORT_SYMBOL(build_skb_around);
359 #define NAPI_SKB_CACHE_SIZE 64
361 struct napi_alloc_cache {
362 struct page_frag_cache page;
363 unsigned int skb_count;
364 void *skb_cache[NAPI_SKB_CACHE_SIZE];
367 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
368 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
370 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
372 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
374 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
377 void *napi_alloc_frag(unsigned int fragsz)
379 fragsz = SKB_DATA_ALIGN(fragsz);
381 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
383 EXPORT_SYMBOL(napi_alloc_frag);
386 * netdev_alloc_frag - allocate a page fragment
387 * @fragsz: fragment size
389 * Allocates a frag from a page for receive buffer.
390 * Uses GFP_ATOMIC allocations.
392 void *netdev_alloc_frag(unsigned int fragsz)
394 struct page_frag_cache *nc;
397 fragsz = SKB_DATA_ALIGN(fragsz);
398 if (in_irq() || irqs_disabled()) {
399 nc = this_cpu_ptr(&netdev_alloc_cache);
400 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
403 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
408 EXPORT_SYMBOL(netdev_alloc_frag);
411 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
412 * @dev: network device to receive on
413 * @len: length to allocate
414 * @gfp_mask: get_free_pages mask, passed to alloc_skb
416 * Allocate a new &sk_buff and assign it a usage count of one. The
417 * buffer has NET_SKB_PAD headroom built in. Users should allocate
418 * the headroom they think they need without accounting for the
419 * built in space. The built in space is used for optimisations.
421 * %NULL is returned if there is no free memory.
423 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
426 struct page_frag_cache *nc;
433 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
434 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
435 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
441 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
442 len = SKB_DATA_ALIGN(len);
444 if (sk_memalloc_socks())
445 gfp_mask |= __GFP_MEMALLOC;
447 if (in_irq() || irqs_disabled()) {
448 nc = this_cpu_ptr(&netdev_alloc_cache);
449 data = page_frag_alloc(nc, len, gfp_mask);
450 pfmemalloc = nc->pfmemalloc;
453 nc = this_cpu_ptr(&napi_alloc_cache.page);
454 data = page_frag_alloc(nc, len, gfp_mask);
455 pfmemalloc = nc->pfmemalloc;
462 skb = __build_skb(data, len);
463 if (unlikely(!skb)) {
468 /* use OR instead of assignment to avoid clearing of bits in mask */
474 skb_reserve(skb, NET_SKB_PAD);
480 EXPORT_SYMBOL(__netdev_alloc_skb);
483 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
484 * @napi: napi instance this buffer was allocated for
485 * @len: length to allocate
486 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
488 * Allocate a new sk_buff for use in NAPI receive. This buffer will
489 * attempt to allocate the head from a special reserved region used
490 * only for NAPI Rx allocation. By doing this we can save several
491 * CPU cycles by avoiding having to disable and re-enable IRQs.
493 * %NULL is returned if there is no free memory.
495 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
498 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
502 len += NET_SKB_PAD + NET_IP_ALIGN;
504 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
505 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
506 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
512 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
513 len = SKB_DATA_ALIGN(len);
515 if (sk_memalloc_socks())
516 gfp_mask |= __GFP_MEMALLOC;
518 data = page_frag_alloc(&nc->page, len, gfp_mask);
522 skb = __build_skb(data, len);
523 if (unlikely(!skb)) {
528 /* use OR instead of assignment to avoid clearing of bits in mask */
529 if (nc->page.pfmemalloc)
534 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
535 skb->dev = napi->dev;
540 EXPORT_SYMBOL(__napi_alloc_skb);
542 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
543 int size, unsigned int truesize)
545 skb_fill_page_desc(skb, i, page, off, size);
547 skb->data_len += size;
548 skb->truesize += truesize;
550 EXPORT_SYMBOL(skb_add_rx_frag);
552 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
553 unsigned int truesize)
555 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
557 skb_frag_size_add(frag, size);
559 skb->data_len += size;
560 skb->truesize += truesize;
562 EXPORT_SYMBOL(skb_coalesce_rx_frag);
564 static void skb_drop_list(struct sk_buff **listp)
566 kfree_skb_list(*listp);
570 static inline void skb_drop_fraglist(struct sk_buff *skb)
572 skb_drop_list(&skb_shinfo(skb)->frag_list);
575 static void skb_clone_fraglist(struct sk_buff *skb)
577 struct sk_buff *list;
579 skb_walk_frags(skb, list)
583 static void skb_free_head(struct sk_buff *skb)
585 unsigned char *head = skb->head;
593 static void skb_release_data(struct sk_buff *skb)
595 struct skb_shared_info *shinfo = skb_shinfo(skb);
599 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
603 for (i = 0; i < shinfo->nr_frags; i++)
604 __skb_frag_unref(&shinfo->frags[i]);
606 if (shinfo->frag_list)
607 kfree_skb_list(shinfo->frag_list);
609 skb_zcopy_clear(skb, true);
614 * Free an skbuff by memory without cleaning the state.
616 static void kfree_skbmem(struct sk_buff *skb)
618 struct sk_buff_fclones *fclones;
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
632 if (refcount_read(&fclones->fclone_ref) == 1)
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
640 if (!refcount_dec_and_test(&fclones->fclone_ref))
643 kmem_cache_free(skbuff_fclone_cache, fclones);
646 void skb_release_head_state(struct sk_buff *skb)
649 if (skb->destructor) {
651 skb->destructor(skb);
653 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
654 nf_conntrack_put(skb_nfct(skb));
659 /* Free everything but the sk_buff shell. */
660 static void skb_release_all(struct sk_buff *skb)
662 skb_release_head_state(skb);
663 if (likely(skb->head))
664 skb_release_data(skb);
668 * __kfree_skb - private function
671 * Free an sk_buff. Release anything attached to the buffer.
672 * Clean the state. This is an internal helper function. Users should
673 * always call kfree_skb
676 void __kfree_skb(struct sk_buff *skb)
678 skb_release_all(skb);
681 EXPORT_SYMBOL(__kfree_skb);
684 * kfree_skb - free an sk_buff
685 * @skb: buffer to free
687 * Drop a reference to the buffer and free it if the usage count has
690 void kfree_skb(struct sk_buff *skb)
695 trace_kfree_skb(skb, __builtin_return_address(0));
698 EXPORT_SYMBOL(kfree_skb);
700 void kfree_skb_list(struct sk_buff *segs)
703 struct sk_buff *next = segs->next;
709 EXPORT_SYMBOL(kfree_skb_list);
711 /* Dump skb information and contents.
713 * Must only be called from net_ratelimit()-ed paths.
715 * Dumps up to can_dump_full whole packets if full_pkt, headers otherwise.
717 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
719 static atomic_t can_dump_full = ATOMIC_INIT(5);
720 struct skb_shared_info *sh = skb_shinfo(skb);
721 struct net_device *dev = skb->dev;
722 struct sock *sk = skb->sk;
723 struct sk_buff *list_skb;
724 bool has_mac, has_trans;
725 int headroom, tailroom;
729 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
734 len = min_t(int, skb->len, MAX_HEADER + 128);
736 headroom = skb_headroom(skb);
737 tailroom = skb_tailroom(skb);
739 has_mac = skb_mac_header_was_set(skb);
740 has_trans = skb_transport_header_was_set(skb);
742 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
743 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
744 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
745 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
746 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
747 level, skb->len, headroom, skb_headlen(skb), tailroom,
748 has_mac ? skb->mac_header : -1,
749 has_mac ? skb_mac_header_len(skb) : -1,
751 has_trans ? skb_network_header_len(skb) : -1,
752 has_trans ? skb->transport_header : -1,
753 sh->tx_flags, sh->nr_frags,
754 sh->gso_size, sh->gso_type, sh->gso_segs,
755 skb->csum, skb->ip_summed, skb->csum_complete_sw,
756 skb->csum_valid, skb->csum_level,
757 skb->hash, skb->sw_hash, skb->l4_hash,
758 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
761 printk("%sdev name=%s feat=0x%pNF\n",
762 level, dev->name, &dev->features);
764 printk("%ssk family=%hu type=%hu proto=%hu\n",
765 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
767 if (full_pkt && headroom)
768 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
769 16, 1, skb->head, headroom, false);
771 seg_len = min_t(int, skb_headlen(skb), len);
773 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
774 16, 1, skb->data, seg_len, false);
777 if (full_pkt && tailroom)
778 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb_tail_pointer(skb), tailroom, false);
781 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
782 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
783 u32 p_off, p_len, copied;
787 skb_frag_foreach_page(frag, frag->page_offset,
788 skb_frag_size(frag), p, p_off, p_len,
790 seg_len = min_t(int, p_len, len);
791 vaddr = kmap_atomic(p);
792 print_hex_dump(level, "skb frag: ",
794 16, 1, vaddr + p_off, seg_len, false);
795 kunmap_atomic(vaddr);
802 if (full_pkt && skb_has_frag_list(skb)) {
803 printk("skb fraglist:\n");
804 skb_walk_frags(skb, list_skb)
805 skb_dump(level, list_skb, true);
808 EXPORT_SYMBOL(skb_dump);
811 * skb_tx_error - report an sk_buff xmit error
812 * @skb: buffer that triggered an error
814 * Report xmit error if a device callback is tracking this skb.
815 * skb must be freed afterwards.
817 void skb_tx_error(struct sk_buff *skb)
819 skb_zcopy_clear(skb, true);
821 EXPORT_SYMBOL(skb_tx_error);
824 * consume_skb - free an skbuff
825 * @skb: buffer to free
827 * Drop a ref to the buffer and free it if the usage count has hit zero
828 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
829 * is being dropped after a failure and notes that
831 void consume_skb(struct sk_buff *skb)
836 trace_consume_skb(skb);
839 EXPORT_SYMBOL(consume_skb);
842 * consume_stateless_skb - free an skbuff, assuming it is stateless
843 * @skb: buffer to free
845 * Alike consume_skb(), but this variant assumes that this is the last
846 * skb reference and all the head states have been already dropped
848 void __consume_stateless_skb(struct sk_buff *skb)
850 trace_consume_skb(skb);
851 skb_release_data(skb);
855 void __kfree_skb_flush(void)
857 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
859 /* flush skb_cache if containing objects */
861 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
867 static inline void _kfree_skb_defer(struct sk_buff *skb)
869 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
871 /* drop skb->head and call any destructors for packet */
872 skb_release_all(skb);
874 /* record skb to CPU local list */
875 nc->skb_cache[nc->skb_count++] = skb;
878 /* SLUB writes into objects when freeing */
882 /* flush skb_cache if it is filled */
883 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
884 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
889 void __kfree_skb_defer(struct sk_buff *skb)
891 _kfree_skb_defer(skb);
894 void napi_consume_skb(struct sk_buff *skb, int budget)
899 /* Zero budget indicate non-NAPI context called us, like netpoll */
900 if (unlikely(!budget)) {
901 dev_consume_skb_any(skb);
908 /* if reaching here SKB is ready to free */
909 trace_consume_skb(skb);
911 /* if SKB is a clone, don't handle this case */
912 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
917 _kfree_skb_defer(skb);
919 EXPORT_SYMBOL(napi_consume_skb);
921 /* Make sure a field is enclosed inside headers_start/headers_end section */
922 #define CHECK_SKB_FIELD(field) \
923 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
924 offsetof(struct sk_buff, headers_start)); \
925 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
926 offsetof(struct sk_buff, headers_end)); \
928 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
930 new->tstamp = old->tstamp;
931 /* We do not copy old->sk */
933 memcpy(new->cb, old->cb, sizeof(old->cb));
934 skb_dst_copy(new, old);
935 __skb_ext_copy(new, old);
936 __nf_copy(new, old, false);
938 /* Note : this field could be in headers_start/headers_end section
939 * It is not yet because we do not want to have a 16 bit hole
941 new->queue_mapping = old->queue_mapping;
943 memcpy(&new->headers_start, &old->headers_start,
944 offsetof(struct sk_buff, headers_end) -
945 offsetof(struct sk_buff, headers_start));
946 CHECK_SKB_FIELD(protocol);
947 CHECK_SKB_FIELD(csum);
948 CHECK_SKB_FIELD(hash);
949 CHECK_SKB_FIELD(priority);
950 CHECK_SKB_FIELD(skb_iif);
951 CHECK_SKB_FIELD(vlan_proto);
952 CHECK_SKB_FIELD(vlan_tci);
953 CHECK_SKB_FIELD(transport_header);
954 CHECK_SKB_FIELD(network_header);
955 CHECK_SKB_FIELD(mac_header);
956 CHECK_SKB_FIELD(inner_protocol);
957 CHECK_SKB_FIELD(inner_transport_header);
958 CHECK_SKB_FIELD(inner_network_header);
959 CHECK_SKB_FIELD(inner_mac_header);
960 CHECK_SKB_FIELD(mark);
961 #ifdef CONFIG_NETWORK_SECMARK
962 CHECK_SKB_FIELD(secmark);
964 #ifdef CONFIG_NET_RX_BUSY_POLL
965 CHECK_SKB_FIELD(napi_id);
968 CHECK_SKB_FIELD(sender_cpu);
970 #ifdef CONFIG_NET_SCHED
971 CHECK_SKB_FIELD(tc_index);
977 * You should not add any new code to this function. Add it to
978 * __copy_skb_header above instead.
980 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
982 #define C(x) n->x = skb->x
984 n->next = n->prev = NULL;
986 __copy_skb_header(n, skb);
991 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
996 n->destructor = NULL;
1003 refcount_set(&n->users, 1);
1005 atomic_inc(&(skb_shinfo(skb)->dataref));
1013 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1014 * @first: first sk_buff of the msg
1016 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1020 n = alloc_skb(0, GFP_ATOMIC);
1024 n->len = first->len;
1025 n->data_len = first->len;
1026 n->truesize = first->truesize;
1028 skb_shinfo(n)->frag_list = first;
1030 __copy_skb_header(n, first);
1031 n->destructor = NULL;
1035 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1038 * skb_morph - morph one skb into another
1039 * @dst: the skb to receive the contents
1040 * @src: the skb to supply the contents
1042 * This is identical to skb_clone except that the target skb is
1043 * supplied by the user.
1045 * The target skb is returned upon exit.
1047 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1049 skb_release_all(dst);
1050 return __skb_clone(dst, src);
1052 EXPORT_SYMBOL_GPL(skb_morph);
1054 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1056 unsigned long max_pg, num_pg, new_pg, old_pg;
1057 struct user_struct *user;
1059 if (capable(CAP_IPC_LOCK) || !size)
1062 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1063 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1064 user = mmp->user ? : current_user();
1067 old_pg = atomic_long_read(&user->locked_vm);
1068 new_pg = old_pg + num_pg;
1069 if (new_pg > max_pg)
1071 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1075 mmp->user = get_uid(user);
1076 mmp->num_pg = num_pg;
1078 mmp->num_pg += num_pg;
1083 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1085 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1088 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1089 free_uid(mmp->user);
1092 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1094 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1096 struct ubuf_info *uarg;
1097 struct sk_buff *skb;
1099 WARN_ON_ONCE(!in_task());
1101 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1105 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1106 uarg = (void *)skb->cb;
1107 uarg->mmp.user = NULL;
1109 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1114 uarg->callback = sock_zerocopy_callback;
1115 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1117 uarg->bytelen = size;
1119 refcount_set(&uarg->refcnt, 1);
1124 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1126 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1128 return container_of((void *)uarg, struct sk_buff, cb);
1131 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1132 struct ubuf_info *uarg)
1135 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1138 /* realloc only when socket is locked (TCP, UDP cork),
1139 * so uarg->len and sk_zckey access is serialized
1141 if (!sock_owned_by_user(sk)) {
1146 bytelen = uarg->bytelen + size;
1147 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1148 /* TCP can create new skb to attach new uarg */
1149 if (sk->sk_type == SOCK_STREAM)
1154 next = (u32)atomic_read(&sk->sk_zckey);
1155 if ((u32)(uarg->id + uarg->len) == next) {
1156 if (mm_account_pinned_pages(&uarg->mmp, size))
1159 uarg->bytelen = bytelen;
1160 atomic_set(&sk->sk_zckey, ++next);
1162 /* no extra ref when appending to datagram (MSG_MORE) */
1163 if (sk->sk_type == SOCK_STREAM)
1164 sock_zerocopy_get(uarg);
1171 return sock_zerocopy_alloc(sk, size);
1173 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1175 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1177 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1181 old_lo = serr->ee.ee_info;
1182 old_hi = serr->ee.ee_data;
1183 sum_len = old_hi - old_lo + 1ULL + len;
1185 if (sum_len >= (1ULL << 32))
1188 if (lo != old_hi + 1)
1191 serr->ee.ee_data += len;
1195 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1197 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1198 struct sock_exterr_skb *serr;
1199 struct sock *sk = skb->sk;
1200 struct sk_buff_head *q;
1201 unsigned long flags;
1205 mm_unaccount_pinned_pages(&uarg->mmp);
1207 /* if !len, there was only 1 call, and it was aborted
1208 * so do not queue a completion notification
1210 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1215 hi = uarg->id + len - 1;
1217 serr = SKB_EXT_ERR(skb);
1218 memset(serr, 0, sizeof(*serr));
1219 serr->ee.ee_errno = 0;
1220 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1221 serr->ee.ee_data = hi;
1222 serr->ee.ee_info = lo;
1224 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1226 q = &sk->sk_error_queue;
1227 spin_lock_irqsave(&q->lock, flags);
1228 tail = skb_peek_tail(q);
1229 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1230 !skb_zerocopy_notify_extend(tail, lo, len)) {
1231 __skb_queue_tail(q, skb);
1234 spin_unlock_irqrestore(&q->lock, flags);
1236 sk->sk_error_report(sk);
1242 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1244 void sock_zerocopy_put(struct ubuf_info *uarg)
1246 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1248 uarg->callback(uarg, uarg->zerocopy);
1250 consume_skb(skb_from_uarg(uarg));
1253 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1255 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1258 struct sock *sk = skb_from_uarg(uarg)->sk;
1260 atomic_dec(&sk->sk_zckey);
1264 sock_zerocopy_put(uarg);
1267 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1269 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1271 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1273 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1275 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1276 struct msghdr *msg, int len,
1277 struct ubuf_info *uarg)
1279 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1280 struct iov_iter orig_iter = msg->msg_iter;
1281 int err, orig_len = skb->len;
1283 /* An skb can only point to one uarg. This edge case happens when
1284 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1286 if (orig_uarg && uarg != orig_uarg)
1289 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1290 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1291 struct sock *save_sk = skb->sk;
1293 /* Streams do not free skb on error. Reset to prev state. */
1294 msg->msg_iter = orig_iter;
1296 ___pskb_trim(skb, orig_len);
1301 skb_zcopy_set(skb, uarg, NULL);
1302 return skb->len - orig_len;
1304 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1306 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1309 if (skb_zcopy(orig)) {
1310 if (skb_zcopy(nskb)) {
1311 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1316 if (skb_uarg(nskb) == skb_uarg(orig))
1318 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1321 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1327 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1328 * @skb: the skb to modify
1329 * @gfp_mask: allocation priority
1331 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1332 * It will copy all frags into kernel and drop the reference
1333 * to userspace pages.
1335 * If this function is called from an interrupt gfp_mask() must be
1338 * Returns 0 on success or a negative error code on failure
1339 * to allocate kernel memory to copy to.
1341 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1343 int num_frags = skb_shinfo(skb)->nr_frags;
1344 struct page *page, *head = NULL;
1348 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1354 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1355 for (i = 0; i < new_frags; i++) {
1356 page = alloc_page(gfp_mask);
1359 struct page *next = (struct page *)page_private(head);
1365 set_page_private(page, (unsigned long)head);
1371 for (i = 0; i < num_frags; i++) {
1372 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1373 u32 p_off, p_len, copied;
1377 skb_frag_foreach_page(f, f->page_offset, skb_frag_size(f),
1378 p, p_off, p_len, copied) {
1380 vaddr = kmap_atomic(p);
1382 while (done < p_len) {
1383 if (d_off == PAGE_SIZE) {
1385 page = (struct page *)page_private(page);
1387 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1388 memcpy(page_address(page) + d_off,
1389 vaddr + p_off + done, copy);
1393 kunmap_atomic(vaddr);
1397 /* skb frags release userspace buffers */
1398 for (i = 0; i < num_frags; i++)
1399 skb_frag_unref(skb, i);
1401 /* skb frags point to kernel buffers */
1402 for (i = 0; i < new_frags - 1; i++) {
1403 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1404 head = (struct page *)page_private(head);
1406 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1407 skb_shinfo(skb)->nr_frags = new_frags;
1410 skb_zcopy_clear(skb, false);
1413 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1416 * skb_clone - duplicate an sk_buff
1417 * @skb: buffer to clone
1418 * @gfp_mask: allocation priority
1420 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1421 * copies share the same packet data but not structure. The new
1422 * buffer has a reference count of 1. If the allocation fails the
1423 * function returns %NULL otherwise the new buffer is returned.
1425 * If this function is called from an interrupt gfp_mask() must be
1429 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1431 struct sk_buff_fclones *fclones = container_of(skb,
1432 struct sk_buff_fclones,
1436 if (skb_orphan_frags(skb, gfp_mask))
1439 if (skb->fclone == SKB_FCLONE_ORIG &&
1440 refcount_read(&fclones->fclone_ref) == 1) {
1442 refcount_set(&fclones->fclone_ref, 2);
1444 if (skb_pfmemalloc(skb))
1445 gfp_mask |= __GFP_MEMALLOC;
1447 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1451 n->fclone = SKB_FCLONE_UNAVAILABLE;
1454 return __skb_clone(n, skb);
1456 EXPORT_SYMBOL(skb_clone);
1458 void skb_headers_offset_update(struct sk_buff *skb, int off)
1460 /* Only adjust this if it actually is csum_start rather than csum */
1461 if (skb->ip_summed == CHECKSUM_PARTIAL)
1462 skb->csum_start += off;
1463 /* {transport,network,mac}_header and tail are relative to skb->head */
1464 skb->transport_header += off;
1465 skb->network_header += off;
1466 if (skb_mac_header_was_set(skb))
1467 skb->mac_header += off;
1468 skb->inner_transport_header += off;
1469 skb->inner_network_header += off;
1470 skb->inner_mac_header += off;
1472 EXPORT_SYMBOL(skb_headers_offset_update);
1474 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1476 __copy_skb_header(new, old);
1478 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1479 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1480 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1482 EXPORT_SYMBOL(skb_copy_header);
1484 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1486 if (skb_pfmemalloc(skb))
1487 return SKB_ALLOC_RX;
1492 * skb_copy - create private copy of an sk_buff
1493 * @skb: buffer to copy
1494 * @gfp_mask: allocation priority
1496 * Make a copy of both an &sk_buff and its data. This is used when the
1497 * caller wishes to modify the data and needs a private copy of the
1498 * data to alter. Returns %NULL on failure or the pointer to the buffer
1499 * on success. The returned buffer has a reference count of 1.
1501 * As by-product this function converts non-linear &sk_buff to linear
1502 * one, so that &sk_buff becomes completely private and caller is allowed
1503 * to modify all the data of returned buffer. This means that this
1504 * function is not recommended for use in circumstances when only
1505 * header is going to be modified. Use pskb_copy() instead.
1508 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1510 int headerlen = skb_headroom(skb);
1511 unsigned int size = skb_end_offset(skb) + skb->data_len;
1512 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1513 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1518 /* Set the data pointer */
1519 skb_reserve(n, headerlen);
1520 /* Set the tail pointer and length */
1521 skb_put(n, skb->len);
1523 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1525 skb_copy_header(n, skb);
1528 EXPORT_SYMBOL(skb_copy);
1531 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1532 * @skb: buffer to copy
1533 * @headroom: headroom of new skb
1534 * @gfp_mask: allocation priority
1535 * @fclone: if true allocate the copy of the skb from the fclone
1536 * cache instead of the head cache; it is recommended to set this
1537 * to true for the cases where the copy will likely be cloned
1539 * Make a copy of both an &sk_buff and part of its data, located
1540 * in header. Fragmented data remain shared. This is used when
1541 * the caller wishes to modify only header of &sk_buff and needs
1542 * private copy of the header to alter. Returns %NULL on failure
1543 * or the pointer to the buffer on success.
1544 * The returned buffer has a reference count of 1.
1547 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1548 gfp_t gfp_mask, bool fclone)
1550 unsigned int size = skb_headlen(skb) + headroom;
1551 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1552 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1557 /* Set the data pointer */
1558 skb_reserve(n, headroom);
1559 /* Set the tail pointer and length */
1560 skb_put(n, skb_headlen(skb));
1561 /* Copy the bytes */
1562 skb_copy_from_linear_data(skb, n->data, n->len);
1564 n->truesize += skb->data_len;
1565 n->data_len = skb->data_len;
1568 if (skb_shinfo(skb)->nr_frags) {
1571 if (skb_orphan_frags(skb, gfp_mask) ||
1572 skb_zerocopy_clone(n, skb, gfp_mask)) {
1577 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1578 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1579 skb_frag_ref(skb, i);
1581 skb_shinfo(n)->nr_frags = i;
1584 if (skb_has_frag_list(skb)) {
1585 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1586 skb_clone_fraglist(n);
1589 skb_copy_header(n, skb);
1593 EXPORT_SYMBOL(__pskb_copy_fclone);
1596 * pskb_expand_head - reallocate header of &sk_buff
1597 * @skb: buffer to reallocate
1598 * @nhead: room to add at head
1599 * @ntail: room to add at tail
1600 * @gfp_mask: allocation priority
1602 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1603 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1604 * reference count of 1. Returns zero in the case of success or error,
1605 * if expansion failed. In the last case, &sk_buff is not changed.
1607 * All the pointers pointing into skb header may change and must be
1608 * reloaded after call to this function.
1611 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1614 int i, osize = skb_end_offset(skb);
1615 int size = osize + nhead + ntail;
1621 BUG_ON(skb_shared(skb));
1623 size = SKB_DATA_ALIGN(size);
1625 if (skb_pfmemalloc(skb))
1626 gfp_mask |= __GFP_MEMALLOC;
1627 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1628 gfp_mask, NUMA_NO_NODE, NULL);
1631 size = SKB_WITH_OVERHEAD(ksize(data));
1633 /* Copy only real data... and, alas, header. This should be
1634 * optimized for the cases when header is void.
1636 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1638 memcpy((struct skb_shared_info *)(data + size),
1640 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1643 * if shinfo is shared we must drop the old head gracefully, but if it
1644 * is not we can just drop the old head and let the existing refcount
1645 * be since all we did is relocate the values
1647 if (skb_cloned(skb)) {
1648 if (skb_orphan_frags(skb, gfp_mask))
1651 refcount_inc(&skb_uarg(skb)->refcnt);
1652 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1653 skb_frag_ref(skb, i);
1655 if (skb_has_frag_list(skb))
1656 skb_clone_fraglist(skb);
1658 skb_release_data(skb);
1662 off = (data + nhead) - skb->head;
1667 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1671 skb->end = skb->head + size;
1674 skb_headers_offset_update(skb, nhead);
1678 atomic_set(&skb_shinfo(skb)->dataref, 1);
1680 skb_metadata_clear(skb);
1682 /* It is not generally safe to change skb->truesize.
1683 * For the moment, we really care of rx path, or
1684 * when skb is orphaned (not attached to a socket).
1686 if (!skb->sk || skb->destructor == sock_edemux)
1687 skb->truesize += size - osize;
1696 EXPORT_SYMBOL(pskb_expand_head);
1698 /* Make private copy of skb with writable head and some headroom */
1700 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1702 struct sk_buff *skb2;
1703 int delta = headroom - skb_headroom(skb);
1706 skb2 = pskb_copy(skb, GFP_ATOMIC);
1708 skb2 = skb_clone(skb, GFP_ATOMIC);
1709 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1717 EXPORT_SYMBOL(skb_realloc_headroom);
1720 * skb_copy_expand - copy and expand sk_buff
1721 * @skb: buffer to copy
1722 * @newheadroom: new free bytes at head
1723 * @newtailroom: new free bytes at tail
1724 * @gfp_mask: allocation priority
1726 * Make a copy of both an &sk_buff and its data and while doing so
1727 * allocate additional space.
1729 * This is used when the caller wishes to modify the data and needs a
1730 * private copy of the data to alter as well as more space for new fields.
1731 * Returns %NULL on failure or the pointer to the buffer
1732 * on success. The returned buffer has a reference count of 1.
1734 * You must pass %GFP_ATOMIC as the allocation priority if this function
1735 * is called from an interrupt.
1737 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1738 int newheadroom, int newtailroom,
1742 * Allocate the copy buffer
1744 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1745 gfp_mask, skb_alloc_rx_flag(skb),
1747 int oldheadroom = skb_headroom(skb);
1748 int head_copy_len, head_copy_off;
1753 skb_reserve(n, newheadroom);
1755 /* Set the tail pointer and length */
1756 skb_put(n, skb->len);
1758 head_copy_len = oldheadroom;
1760 if (newheadroom <= head_copy_len)
1761 head_copy_len = newheadroom;
1763 head_copy_off = newheadroom - head_copy_len;
1765 /* Copy the linear header and data. */
1766 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1767 skb->len + head_copy_len));
1769 skb_copy_header(n, skb);
1771 skb_headers_offset_update(n, newheadroom - oldheadroom);
1775 EXPORT_SYMBOL(skb_copy_expand);
1778 * __skb_pad - zero pad the tail of an skb
1779 * @skb: buffer to pad
1780 * @pad: space to pad
1781 * @free_on_error: free buffer on error
1783 * Ensure that a buffer is followed by a padding area that is zero
1784 * filled. Used by network drivers which may DMA or transfer data
1785 * beyond the buffer end onto the wire.
1787 * May return error in out of memory cases. The skb is freed on error
1788 * if @free_on_error is true.
1791 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1796 /* If the skbuff is non linear tailroom is always zero.. */
1797 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1798 memset(skb->data+skb->len, 0, pad);
1802 ntail = skb->data_len + pad - (skb->end - skb->tail);
1803 if (likely(skb_cloned(skb) || ntail > 0)) {
1804 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1809 /* FIXME: The use of this function with non-linear skb's really needs
1812 err = skb_linearize(skb);
1816 memset(skb->data + skb->len, 0, pad);
1824 EXPORT_SYMBOL(__skb_pad);
1827 * pskb_put - add data to the tail of a potentially fragmented buffer
1828 * @skb: start of the buffer to use
1829 * @tail: tail fragment of the buffer to use
1830 * @len: amount of data to add
1832 * This function extends the used data area of the potentially
1833 * fragmented buffer. @tail must be the last fragment of @skb -- or
1834 * @skb itself. If this would exceed the total buffer size the kernel
1835 * will panic. A pointer to the first byte of the extra data is
1839 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1842 skb->data_len += len;
1845 return skb_put(tail, len);
1847 EXPORT_SYMBOL_GPL(pskb_put);
1850 * skb_put - add data to a buffer
1851 * @skb: buffer to use
1852 * @len: amount of data to add
1854 * This function extends the used data area of the buffer. If this would
1855 * exceed the total buffer size the kernel will panic. A pointer to the
1856 * first byte of the extra data is returned.
1858 void *skb_put(struct sk_buff *skb, unsigned int len)
1860 void *tmp = skb_tail_pointer(skb);
1861 SKB_LINEAR_ASSERT(skb);
1864 if (unlikely(skb->tail > skb->end))
1865 skb_over_panic(skb, len, __builtin_return_address(0));
1868 EXPORT_SYMBOL(skb_put);
1871 * skb_push - add data to the start of a buffer
1872 * @skb: buffer to use
1873 * @len: amount of data to add
1875 * This function extends the used data area of the buffer at the buffer
1876 * start. If this would exceed the total buffer headroom the kernel will
1877 * panic. A pointer to the first byte of the extra data is returned.
1879 void *skb_push(struct sk_buff *skb, unsigned int len)
1883 if (unlikely(skb->data < skb->head))
1884 skb_under_panic(skb, len, __builtin_return_address(0));
1887 EXPORT_SYMBOL(skb_push);
1890 * skb_pull - remove data from the start of a buffer
1891 * @skb: buffer to use
1892 * @len: amount of data to remove
1894 * This function removes data from the start of a buffer, returning
1895 * the memory to the headroom. A pointer to the next data in the buffer
1896 * is returned. Once the data has been pulled future pushes will overwrite
1899 void *skb_pull(struct sk_buff *skb, unsigned int len)
1901 return skb_pull_inline(skb, len);
1903 EXPORT_SYMBOL(skb_pull);
1906 * skb_trim - remove end from a buffer
1907 * @skb: buffer to alter
1910 * Cut the length of a buffer down by removing data from the tail. If
1911 * the buffer is already under the length specified it is not modified.
1912 * The skb must be linear.
1914 void skb_trim(struct sk_buff *skb, unsigned int len)
1917 __skb_trim(skb, len);
1919 EXPORT_SYMBOL(skb_trim);
1921 /* Trims skb to length len. It can change skb pointers.
1924 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1926 struct sk_buff **fragp;
1927 struct sk_buff *frag;
1928 int offset = skb_headlen(skb);
1929 int nfrags = skb_shinfo(skb)->nr_frags;
1933 if (skb_cloned(skb) &&
1934 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1941 for (; i < nfrags; i++) {
1942 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1949 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1952 skb_shinfo(skb)->nr_frags = i;
1954 for (; i < nfrags; i++)
1955 skb_frag_unref(skb, i);
1957 if (skb_has_frag_list(skb))
1958 skb_drop_fraglist(skb);
1962 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1963 fragp = &frag->next) {
1964 int end = offset + frag->len;
1966 if (skb_shared(frag)) {
1967 struct sk_buff *nfrag;
1969 nfrag = skb_clone(frag, GFP_ATOMIC);
1970 if (unlikely(!nfrag))
1973 nfrag->next = frag->next;
1985 unlikely((err = pskb_trim(frag, len - offset))))
1989 skb_drop_list(&frag->next);
1994 if (len > skb_headlen(skb)) {
1995 skb->data_len -= skb->len - len;
2000 skb_set_tail_pointer(skb, len);
2003 if (!skb->sk || skb->destructor == sock_edemux)
2007 EXPORT_SYMBOL(___pskb_trim);
2009 /* Note : use pskb_trim_rcsum() instead of calling this directly
2011 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2013 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2014 int delta = skb->len - len;
2016 skb->csum = csum_block_sub(skb->csum,
2017 skb_checksum(skb, len, delta, 0),
2020 return __pskb_trim(skb, len);
2022 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2025 * __pskb_pull_tail - advance tail of skb header
2026 * @skb: buffer to reallocate
2027 * @delta: number of bytes to advance tail
2029 * The function makes a sense only on a fragmented &sk_buff,
2030 * it expands header moving its tail forward and copying necessary
2031 * data from fragmented part.
2033 * &sk_buff MUST have reference count of 1.
2035 * Returns %NULL (and &sk_buff does not change) if pull failed
2036 * or value of new tail of skb in the case of success.
2038 * All the pointers pointing into skb header may change and must be
2039 * reloaded after call to this function.
2042 /* Moves tail of skb head forward, copying data from fragmented part,
2043 * when it is necessary.
2044 * 1. It may fail due to malloc failure.
2045 * 2. It may change skb pointers.
2047 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2049 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2051 /* If skb has not enough free space at tail, get new one
2052 * plus 128 bytes for future expansions. If we have enough
2053 * room at tail, reallocate without expansion only if skb is cloned.
2055 int i, k, eat = (skb->tail + delta) - skb->end;
2057 if (eat > 0 || skb_cloned(skb)) {
2058 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2063 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2064 skb_tail_pointer(skb), delta));
2066 /* Optimization: no fragments, no reasons to preestimate
2067 * size of pulled pages. Superb.
2069 if (!skb_has_frag_list(skb))
2072 /* Estimate size of pulled pages. */
2074 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2075 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2082 /* If we need update frag list, we are in troubles.
2083 * Certainly, it is possible to add an offset to skb data,
2084 * but taking into account that pulling is expected to
2085 * be very rare operation, it is worth to fight against
2086 * further bloating skb head and crucify ourselves here instead.
2087 * Pure masohism, indeed. 8)8)
2090 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2091 struct sk_buff *clone = NULL;
2092 struct sk_buff *insp = NULL;
2095 if (list->len <= eat) {
2096 /* Eaten as whole. */
2101 /* Eaten partially. */
2103 if (skb_shared(list)) {
2104 /* Sucks! We need to fork list. :-( */
2105 clone = skb_clone(list, GFP_ATOMIC);
2111 /* This may be pulled without
2115 if (!pskb_pull(list, eat)) {
2123 /* Free pulled out fragments. */
2124 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2125 skb_shinfo(skb)->frag_list = list->next;
2128 /* And insert new clone at head. */
2131 skb_shinfo(skb)->frag_list = clone;
2134 /* Success! Now we may commit changes to skb data. */
2139 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2140 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2143 skb_frag_unref(skb, i);
2146 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
2148 skb_shinfo(skb)->frags[k].page_offset += eat;
2149 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
2157 skb_shinfo(skb)->nr_frags = k;
2161 skb->data_len -= delta;
2164 skb_zcopy_clear(skb, false);
2166 return skb_tail_pointer(skb);
2168 EXPORT_SYMBOL(__pskb_pull_tail);
2171 * skb_copy_bits - copy bits from skb to kernel buffer
2173 * @offset: offset in source
2174 * @to: destination buffer
2175 * @len: number of bytes to copy
2177 * Copy the specified number of bytes from the source skb to the
2178 * destination buffer.
2181 * If its prototype is ever changed,
2182 * check arch/{*}/net/{*}.S files,
2183 * since it is called from BPF assembly code.
2185 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2187 int start = skb_headlen(skb);
2188 struct sk_buff *frag_iter;
2191 if (offset > (int)skb->len - len)
2195 if ((copy = start - offset) > 0) {
2198 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2199 if ((len -= copy) == 0)
2205 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2207 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2209 WARN_ON(start > offset + len);
2211 end = start + skb_frag_size(f);
2212 if ((copy = end - offset) > 0) {
2213 u32 p_off, p_len, copied;
2220 skb_frag_foreach_page(f,
2221 f->page_offset + offset - start,
2222 copy, p, p_off, p_len, copied) {
2223 vaddr = kmap_atomic(p);
2224 memcpy(to + copied, vaddr + p_off, p_len);
2225 kunmap_atomic(vaddr);
2228 if ((len -= copy) == 0)
2236 skb_walk_frags(skb, frag_iter) {
2239 WARN_ON(start > offset + len);
2241 end = start + frag_iter->len;
2242 if ((copy = end - offset) > 0) {
2245 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2247 if ((len -= copy) == 0)
2261 EXPORT_SYMBOL(skb_copy_bits);
2264 * Callback from splice_to_pipe(), if we need to release some pages
2265 * at the end of the spd in case we error'ed out in filling the pipe.
2267 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2269 put_page(spd->pages[i]);
2272 static struct page *linear_to_page(struct page *page, unsigned int *len,
2273 unsigned int *offset,
2276 struct page_frag *pfrag = sk_page_frag(sk);
2278 if (!sk_page_frag_refill(sk, pfrag))
2281 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2283 memcpy(page_address(pfrag->page) + pfrag->offset,
2284 page_address(page) + *offset, *len);
2285 *offset = pfrag->offset;
2286 pfrag->offset += *len;
2291 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2293 unsigned int offset)
2295 return spd->nr_pages &&
2296 spd->pages[spd->nr_pages - 1] == page &&
2297 (spd->partial[spd->nr_pages - 1].offset +
2298 spd->partial[spd->nr_pages - 1].len == offset);
2302 * Fill page/offset/length into spd, if it can hold more pages.
2304 static bool spd_fill_page(struct splice_pipe_desc *spd,
2305 struct pipe_inode_info *pipe, struct page *page,
2306 unsigned int *len, unsigned int offset,
2310 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2314 page = linear_to_page(page, len, &offset, sk);
2318 if (spd_can_coalesce(spd, page, offset)) {
2319 spd->partial[spd->nr_pages - 1].len += *len;
2323 spd->pages[spd->nr_pages] = page;
2324 spd->partial[spd->nr_pages].len = *len;
2325 spd->partial[spd->nr_pages].offset = offset;
2331 static bool __splice_segment(struct page *page, unsigned int poff,
2332 unsigned int plen, unsigned int *off,
2334 struct splice_pipe_desc *spd, bool linear,
2336 struct pipe_inode_info *pipe)
2341 /* skip this segment if already processed */
2347 /* ignore any bits we already processed */
2353 unsigned int flen = min(*len, plen);
2355 if (spd_fill_page(spd, pipe, page, &flen, poff,
2361 } while (*len && plen);
2367 * Map linear and fragment data from the skb to spd. It reports true if the
2368 * pipe is full or if we already spliced the requested length.
2370 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2371 unsigned int *offset, unsigned int *len,
2372 struct splice_pipe_desc *spd, struct sock *sk)
2375 struct sk_buff *iter;
2377 /* map the linear part :
2378 * If skb->head_frag is set, this 'linear' part is backed by a
2379 * fragment, and if the head is not shared with any clones then
2380 * we can avoid a copy since we own the head portion of this page.
2382 if (__splice_segment(virt_to_page(skb->data),
2383 (unsigned long) skb->data & (PAGE_SIZE - 1),
2386 skb_head_is_locked(skb),
2391 * then map the fragments
2393 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2394 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2396 if (__splice_segment(skb_frag_page(f),
2397 f->page_offset, skb_frag_size(f),
2398 offset, len, spd, false, sk, pipe))
2402 skb_walk_frags(skb, iter) {
2403 if (*offset >= iter->len) {
2404 *offset -= iter->len;
2407 /* __skb_splice_bits() only fails if the output has no room
2408 * left, so no point in going over the frag_list for the error
2411 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2419 * Map data from the skb to a pipe. Should handle both the linear part,
2420 * the fragments, and the frag list.
2422 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2423 struct pipe_inode_info *pipe, unsigned int tlen,
2426 struct partial_page partial[MAX_SKB_FRAGS];
2427 struct page *pages[MAX_SKB_FRAGS];
2428 struct splice_pipe_desc spd = {
2431 .nr_pages_max = MAX_SKB_FRAGS,
2432 .ops = &nosteal_pipe_buf_ops,
2433 .spd_release = sock_spd_release,
2437 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2440 ret = splice_to_pipe(pipe, &spd);
2444 EXPORT_SYMBOL_GPL(skb_splice_bits);
2446 /* Send skb data on a socket. Socket must be locked. */
2447 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2450 unsigned int orig_len = len;
2451 struct sk_buff *head = skb;
2452 unsigned short fragidx;
2457 /* Deal with head data */
2458 while (offset < skb_headlen(skb) && len) {
2462 slen = min_t(int, len, skb_headlen(skb) - offset);
2463 kv.iov_base = skb->data + offset;
2465 memset(&msg, 0, sizeof(msg));
2466 msg.msg_flags = MSG_DONTWAIT;
2468 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2476 /* All the data was skb head? */
2480 /* Make offset relative to start of frags */
2481 offset -= skb_headlen(skb);
2483 /* Find where we are in frag list */
2484 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2485 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2487 if (offset < frag->size)
2490 offset -= frag->size;
2493 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2494 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2496 slen = min_t(size_t, len, frag->size - offset);
2499 ret = kernel_sendpage_locked(sk, frag->page.p,
2500 frag->page_offset + offset,
2501 slen, MSG_DONTWAIT);
2514 /* Process any frag lists */
2517 if (skb_has_frag_list(skb)) {
2518 skb = skb_shinfo(skb)->frag_list;
2521 } else if (skb->next) {
2528 return orig_len - len;
2531 return orig_len == len ? ret : orig_len - len;
2533 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2536 * skb_store_bits - store bits from kernel buffer to skb
2537 * @skb: destination buffer
2538 * @offset: offset in destination
2539 * @from: source buffer
2540 * @len: number of bytes to copy
2542 * Copy the specified number of bytes from the source buffer to the
2543 * destination skb. This function handles all the messy bits of
2544 * traversing fragment lists and such.
2547 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2549 int start = skb_headlen(skb);
2550 struct sk_buff *frag_iter;
2553 if (offset > (int)skb->len - len)
2556 if ((copy = start - offset) > 0) {
2559 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2560 if ((len -= copy) == 0)
2566 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2567 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2570 WARN_ON(start > offset + len);
2572 end = start + skb_frag_size(frag);
2573 if ((copy = end - offset) > 0) {
2574 u32 p_off, p_len, copied;
2581 skb_frag_foreach_page(frag,
2582 frag->page_offset + offset - start,
2583 copy, p, p_off, p_len, copied) {
2584 vaddr = kmap_atomic(p);
2585 memcpy(vaddr + p_off, from + copied, p_len);
2586 kunmap_atomic(vaddr);
2589 if ((len -= copy) == 0)
2597 skb_walk_frags(skb, frag_iter) {
2600 WARN_ON(start > offset + len);
2602 end = start + frag_iter->len;
2603 if ((copy = end - offset) > 0) {
2606 if (skb_store_bits(frag_iter, offset - start,
2609 if ((len -= copy) == 0)
2622 EXPORT_SYMBOL(skb_store_bits);
2624 /* Checksum skb data. */
2625 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2626 __wsum csum, const struct skb_checksum_ops *ops)
2628 int start = skb_headlen(skb);
2629 int i, copy = start - offset;
2630 struct sk_buff *frag_iter;
2633 /* Checksum header. */
2637 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2638 skb->data + offset, copy, csum);
2639 if ((len -= copy) == 0)
2645 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2647 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2649 WARN_ON(start > offset + len);
2651 end = start + skb_frag_size(frag);
2652 if ((copy = end - offset) > 0) {
2653 u32 p_off, p_len, copied;
2661 skb_frag_foreach_page(frag,
2662 frag->page_offset + offset - start,
2663 copy, p, p_off, p_len, copied) {
2664 vaddr = kmap_atomic(p);
2665 csum2 = INDIRECT_CALL_1(ops->update,
2667 vaddr + p_off, p_len, 0);
2668 kunmap_atomic(vaddr);
2669 csum = INDIRECT_CALL_1(ops->combine,
2670 csum_block_add_ext, csum,
2682 skb_walk_frags(skb, frag_iter) {
2685 WARN_ON(start > offset + len);
2687 end = start + frag_iter->len;
2688 if ((copy = end - offset) > 0) {
2692 csum2 = __skb_checksum(frag_iter, offset - start,
2694 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2695 csum, csum2, pos, copy);
2696 if ((len -= copy) == 0)
2707 EXPORT_SYMBOL(__skb_checksum);
2709 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2710 int len, __wsum csum)
2712 const struct skb_checksum_ops ops = {
2713 .update = csum_partial_ext,
2714 .combine = csum_block_add_ext,
2717 return __skb_checksum(skb, offset, len, csum, &ops);
2719 EXPORT_SYMBOL(skb_checksum);
2721 /* Both of above in one bottle. */
2723 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2724 u8 *to, int len, __wsum csum)
2726 int start = skb_headlen(skb);
2727 int i, copy = start - offset;
2728 struct sk_buff *frag_iter;
2735 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2737 if ((len -= copy) == 0)
2744 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2747 WARN_ON(start > offset + len);
2749 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2750 if ((copy = end - offset) > 0) {
2751 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2752 u32 p_off, p_len, copied;
2760 skb_frag_foreach_page(frag,
2761 frag->page_offset + offset - start,
2762 copy, p, p_off, p_len, copied) {
2763 vaddr = kmap_atomic(p);
2764 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2767 kunmap_atomic(vaddr);
2768 csum = csum_block_add(csum, csum2, pos);
2780 skb_walk_frags(skb, frag_iter) {
2784 WARN_ON(start > offset + len);
2786 end = start + frag_iter->len;
2787 if ((copy = end - offset) > 0) {
2790 csum2 = skb_copy_and_csum_bits(frag_iter,
2793 csum = csum_block_add(csum, csum2, pos);
2794 if ((len -= copy) == 0)
2805 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2807 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2811 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2812 /* See comments in __skb_checksum_complete(). */
2814 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2815 !skb->csum_complete_sw)
2816 netdev_rx_csum_fault(skb->dev, skb);
2818 if (!skb_shared(skb))
2819 skb->csum_valid = !sum;
2822 EXPORT_SYMBOL(__skb_checksum_complete_head);
2824 /* This function assumes skb->csum already holds pseudo header's checksum,
2825 * which has been changed from the hardware checksum, for example, by
2826 * __skb_checksum_validate_complete(). And, the original skb->csum must
2827 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2829 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2830 * zero. The new checksum is stored back into skb->csum unless the skb is
2833 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2838 csum = skb_checksum(skb, 0, skb->len, 0);
2840 sum = csum_fold(csum_add(skb->csum, csum));
2841 /* This check is inverted, because we already knew the hardware
2842 * checksum is invalid before calling this function. So, if the
2843 * re-computed checksum is valid instead, then we have a mismatch
2844 * between the original skb->csum and skb_checksum(). This means either
2845 * the original hardware checksum is incorrect or we screw up skb->csum
2846 * when moving skb->data around.
2849 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2850 !skb->csum_complete_sw)
2851 netdev_rx_csum_fault(skb->dev, skb);
2854 if (!skb_shared(skb)) {
2855 /* Save full packet checksum */
2857 skb->ip_summed = CHECKSUM_COMPLETE;
2858 skb->csum_complete_sw = 1;
2859 skb->csum_valid = !sum;
2864 EXPORT_SYMBOL(__skb_checksum_complete);
2866 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2868 net_warn_ratelimited(
2869 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2874 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2875 int offset, int len)
2877 net_warn_ratelimited(
2878 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2883 static const struct skb_checksum_ops default_crc32c_ops = {
2884 .update = warn_crc32c_csum_update,
2885 .combine = warn_crc32c_csum_combine,
2888 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2889 &default_crc32c_ops;
2890 EXPORT_SYMBOL(crc32c_csum_stub);
2893 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2894 * @from: source buffer
2896 * Calculates the amount of linear headroom needed in the 'to' skb passed
2897 * into skb_zerocopy().
2900 skb_zerocopy_headlen(const struct sk_buff *from)
2902 unsigned int hlen = 0;
2904 if (!from->head_frag ||
2905 skb_headlen(from) < L1_CACHE_BYTES ||
2906 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2907 hlen = skb_headlen(from);
2909 if (skb_has_frag_list(from))
2914 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2917 * skb_zerocopy - Zero copy skb to skb
2918 * @to: destination buffer
2919 * @from: source buffer
2920 * @len: number of bytes to copy from source buffer
2921 * @hlen: size of linear headroom in destination buffer
2923 * Copies up to `len` bytes from `from` to `to` by creating references
2924 * to the frags in the source buffer.
2926 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2927 * headroom in the `to` buffer.
2930 * 0: everything is OK
2931 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2932 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2935 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2938 int plen = 0; /* length of skb->head fragment */
2941 unsigned int offset;
2943 BUG_ON(!from->head_frag && !hlen);
2945 /* dont bother with small payloads */
2946 if (len <= skb_tailroom(to))
2947 return skb_copy_bits(from, 0, skb_put(to, len), len);
2950 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2955 plen = min_t(int, skb_headlen(from), len);
2957 page = virt_to_head_page(from->head);
2958 offset = from->data - (unsigned char *)page_address(page);
2959 __skb_fill_page_desc(to, 0, page, offset, plen);
2966 to->truesize += len + plen;
2967 to->len += len + plen;
2968 to->data_len += len + plen;
2970 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2974 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2976 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2979 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2980 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2981 len -= skb_shinfo(to)->frags[j].size;
2982 skb_frag_ref(to, j);
2985 skb_shinfo(to)->nr_frags = j;
2989 EXPORT_SYMBOL_GPL(skb_zerocopy);
2991 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2996 if (skb->ip_summed == CHECKSUM_PARTIAL)
2997 csstart = skb_checksum_start_offset(skb);
2999 csstart = skb_headlen(skb);
3001 BUG_ON(csstart > skb_headlen(skb));
3003 skb_copy_from_linear_data(skb, to, csstart);
3006 if (csstart != skb->len)
3007 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3008 skb->len - csstart, 0);
3010 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3011 long csstuff = csstart + skb->csum_offset;
3013 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3016 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3019 * skb_dequeue - remove from the head of the queue
3020 * @list: list to dequeue from
3022 * Remove the head of the list. The list lock is taken so the function
3023 * may be used safely with other locking list functions. The head item is
3024 * returned or %NULL if the list is empty.
3027 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3029 unsigned long flags;
3030 struct sk_buff *result;
3032 spin_lock_irqsave(&list->lock, flags);
3033 result = __skb_dequeue(list);
3034 spin_unlock_irqrestore(&list->lock, flags);
3037 EXPORT_SYMBOL(skb_dequeue);
3040 * skb_dequeue_tail - remove from the tail of the queue
3041 * @list: list to dequeue from
3043 * Remove the tail of the list. The list lock is taken so the function
3044 * may be used safely with other locking list functions. The tail item is
3045 * returned or %NULL if the list is empty.
3047 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3049 unsigned long flags;
3050 struct sk_buff *result;
3052 spin_lock_irqsave(&list->lock, flags);
3053 result = __skb_dequeue_tail(list);
3054 spin_unlock_irqrestore(&list->lock, flags);
3057 EXPORT_SYMBOL(skb_dequeue_tail);
3060 * skb_queue_purge - empty a list
3061 * @list: list to empty
3063 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3064 * the list and one reference dropped. This function takes the list
3065 * lock and is atomic with respect to other list locking functions.
3067 void skb_queue_purge(struct sk_buff_head *list)
3069 struct sk_buff *skb;
3070 while ((skb = skb_dequeue(list)) != NULL)
3073 EXPORT_SYMBOL(skb_queue_purge);
3076 * skb_rbtree_purge - empty a skb rbtree
3077 * @root: root of the rbtree to empty
3078 * Return value: the sum of truesizes of all purged skbs.
3080 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3081 * the list and one reference dropped. This function does not take
3082 * any lock. Synchronization should be handled by the caller (e.g., TCP
3083 * out-of-order queue is protected by the socket lock).
3085 unsigned int skb_rbtree_purge(struct rb_root *root)
3087 struct rb_node *p = rb_first(root);
3088 unsigned int sum = 0;
3091 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3094 rb_erase(&skb->rbnode, root);
3095 sum += skb->truesize;
3102 * skb_queue_head - queue a buffer at the list head
3103 * @list: list to use
3104 * @newsk: buffer to queue
3106 * Queue a buffer at the start of the list. This function takes the
3107 * list lock and can be used safely with other locking &sk_buff functions
3110 * A buffer cannot be placed on two lists at the same time.
3112 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3114 unsigned long flags;
3116 spin_lock_irqsave(&list->lock, flags);
3117 __skb_queue_head(list, newsk);
3118 spin_unlock_irqrestore(&list->lock, flags);
3120 EXPORT_SYMBOL(skb_queue_head);
3123 * skb_queue_tail - queue a buffer at the list tail
3124 * @list: list to use
3125 * @newsk: buffer to queue
3127 * Queue a buffer at the tail of the list. This function takes the
3128 * list lock and can be used safely with other locking &sk_buff functions
3131 * A buffer cannot be placed on two lists at the same time.
3133 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3135 unsigned long flags;
3137 spin_lock_irqsave(&list->lock, flags);
3138 __skb_queue_tail(list, newsk);
3139 spin_unlock_irqrestore(&list->lock, flags);
3141 EXPORT_SYMBOL(skb_queue_tail);
3144 * skb_unlink - remove a buffer from a list
3145 * @skb: buffer to remove
3146 * @list: list to use
3148 * Remove a packet from a list. The list locks are taken and this
3149 * function is atomic with respect to other list locked calls
3151 * You must know what list the SKB is on.
3153 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3155 unsigned long flags;
3157 spin_lock_irqsave(&list->lock, flags);
3158 __skb_unlink(skb, list);
3159 spin_unlock_irqrestore(&list->lock, flags);
3161 EXPORT_SYMBOL(skb_unlink);
3164 * skb_append - append a buffer
3165 * @old: buffer to insert after
3166 * @newsk: buffer to insert
3167 * @list: list to use
3169 * Place a packet after a given packet in a list. The list locks are taken
3170 * and this function is atomic with respect to other list locked calls.
3171 * A buffer cannot be placed on two lists at the same time.
3173 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3175 unsigned long flags;
3177 spin_lock_irqsave(&list->lock, flags);
3178 __skb_queue_after(list, old, newsk);
3179 spin_unlock_irqrestore(&list->lock, flags);
3181 EXPORT_SYMBOL(skb_append);
3183 static inline void skb_split_inside_header(struct sk_buff *skb,
3184 struct sk_buff* skb1,
3185 const u32 len, const int pos)
3189 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3191 /* And move data appendix as is. */
3192 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3193 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3195 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3196 skb_shinfo(skb)->nr_frags = 0;
3197 skb1->data_len = skb->data_len;
3198 skb1->len += skb1->data_len;
3201 skb_set_tail_pointer(skb, len);
3204 static inline void skb_split_no_header(struct sk_buff *skb,
3205 struct sk_buff* skb1,
3206 const u32 len, int pos)
3209 const int nfrags = skb_shinfo(skb)->nr_frags;
3211 skb_shinfo(skb)->nr_frags = 0;
3212 skb1->len = skb1->data_len = skb->len - len;
3214 skb->data_len = len - pos;
3216 for (i = 0; i < nfrags; i++) {
3217 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3219 if (pos + size > len) {
3220 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3224 * We have two variants in this case:
3225 * 1. Move all the frag to the second
3226 * part, if it is possible. F.e.
3227 * this approach is mandatory for TUX,
3228 * where splitting is expensive.
3229 * 2. Split is accurately. We make this.
3231 skb_frag_ref(skb, i);
3232 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
3233 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3234 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3235 skb_shinfo(skb)->nr_frags++;
3239 skb_shinfo(skb)->nr_frags++;
3242 skb_shinfo(skb1)->nr_frags = k;
3246 * skb_split - Split fragmented skb to two parts at length len.
3247 * @skb: the buffer to split
3248 * @skb1: the buffer to receive the second part
3249 * @len: new length for skb
3251 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3253 int pos = skb_headlen(skb);
3255 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3257 skb_zerocopy_clone(skb1, skb, 0);
3258 if (len < pos) /* Split line is inside header. */
3259 skb_split_inside_header(skb, skb1, len, pos);
3260 else /* Second chunk has no header, nothing to copy. */
3261 skb_split_no_header(skb, skb1, len, pos);
3263 EXPORT_SYMBOL(skb_split);
3265 /* Shifting from/to a cloned skb is a no-go.
3267 * Caller cannot keep skb_shinfo related pointers past calling here!
3269 static int skb_prepare_for_shift(struct sk_buff *skb)
3271 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3275 * skb_shift - Shifts paged data partially from skb to another
3276 * @tgt: buffer into which tail data gets added
3277 * @skb: buffer from which the paged data comes from
3278 * @shiftlen: shift up to this many bytes
3280 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3281 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3282 * It's up to caller to free skb if everything was shifted.
3284 * If @tgt runs out of frags, the whole operation is aborted.
3286 * Skb cannot include anything else but paged data while tgt is allowed
3287 * to have non-paged data as well.
3289 * TODO: full sized shift could be optimized but that would need
3290 * specialized skb free'er to handle frags without up-to-date nr_frags.
3292 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3294 int from, to, merge, todo;
3295 struct skb_frag_struct *fragfrom, *fragto;
3297 BUG_ON(shiftlen > skb->len);
3299 if (skb_headlen(skb))
3301 if (skb_zcopy(tgt) || skb_zcopy(skb))
3306 to = skb_shinfo(tgt)->nr_frags;
3307 fragfrom = &skb_shinfo(skb)->frags[from];
3309 /* Actual merge is delayed until the point when we know we can
3310 * commit all, so that we don't have to undo partial changes
3313 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3314 fragfrom->page_offset)) {
3319 todo -= skb_frag_size(fragfrom);
3321 if (skb_prepare_for_shift(skb) ||
3322 skb_prepare_for_shift(tgt))
3325 /* All previous frag pointers might be stale! */
3326 fragfrom = &skb_shinfo(skb)->frags[from];
3327 fragto = &skb_shinfo(tgt)->frags[merge];
3329 skb_frag_size_add(fragto, shiftlen);
3330 skb_frag_size_sub(fragfrom, shiftlen);
3331 fragfrom->page_offset += shiftlen;
3339 /* Skip full, not-fitting skb to avoid expensive operations */
3340 if ((shiftlen == skb->len) &&
3341 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3344 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3347 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3348 if (to == MAX_SKB_FRAGS)
3351 fragfrom = &skb_shinfo(skb)->frags[from];
3352 fragto = &skb_shinfo(tgt)->frags[to];
3354 if (todo >= skb_frag_size(fragfrom)) {
3355 *fragto = *fragfrom;
3356 todo -= skb_frag_size(fragfrom);
3361 __skb_frag_ref(fragfrom);
3362 fragto->page = fragfrom->page;
3363 fragto->page_offset = fragfrom->page_offset;
3364 skb_frag_size_set(fragto, todo);
3366 fragfrom->page_offset += todo;
3367 skb_frag_size_sub(fragfrom, todo);
3375 /* Ready to "commit" this state change to tgt */
3376 skb_shinfo(tgt)->nr_frags = to;
3379 fragfrom = &skb_shinfo(skb)->frags[0];
3380 fragto = &skb_shinfo(tgt)->frags[merge];
3382 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3383 __skb_frag_unref(fragfrom);
3386 /* Reposition in the original skb */
3388 while (from < skb_shinfo(skb)->nr_frags)
3389 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3390 skb_shinfo(skb)->nr_frags = to;
3392 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3395 /* Most likely the tgt won't ever need its checksum anymore, skb on
3396 * the other hand might need it if it needs to be resent
3398 tgt->ip_summed = CHECKSUM_PARTIAL;
3399 skb->ip_summed = CHECKSUM_PARTIAL;
3401 /* Yak, is it really working this way? Some helper please? */
3402 skb->len -= shiftlen;
3403 skb->data_len -= shiftlen;
3404 skb->truesize -= shiftlen;
3405 tgt->len += shiftlen;
3406 tgt->data_len += shiftlen;
3407 tgt->truesize += shiftlen;
3413 * skb_prepare_seq_read - Prepare a sequential read of skb data
3414 * @skb: the buffer to read
3415 * @from: lower offset of data to be read
3416 * @to: upper offset of data to be read
3417 * @st: state variable
3419 * Initializes the specified state variable. Must be called before
3420 * invoking skb_seq_read() for the first time.
3422 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3423 unsigned int to, struct skb_seq_state *st)
3425 st->lower_offset = from;
3426 st->upper_offset = to;
3427 st->root_skb = st->cur_skb = skb;
3428 st->frag_idx = st->stepped_offset = 0;
3429 st->frag_data = NULL;
3431 EXPORT_SYMBOL(skb_prepare_seq_read);
3434 * skb_seq_read - Sequentially read skb data
3435 * @consumed: number of bytes consumed by the caller so far
3436 * @data: destination pointer for data to be returned
3437 * @st: state variable
3439 * Reads a block of skb data at @consumed relative to the
3440 * lower offset specified to skb_prepare_seq_read(). Assigns
3441 * the head of the data block to @data and returns the length
3442 * of the block or 0 if the end of the skb data or the upper
3443 * offset has been reached.
3445 * The caller is not required to consume all of the data
3446 * returned, i.e. @consumed is typically set to the number
3447 * of bytes already consumed and the next call to
3448 * skb_seq_read() will return the remaining part of the block.
3450 * Note 1: The size of each block of data returned can be arbitrary,
3451 * this limitation is the cost for zerocopy sequential
3452 * reads of potentially non linear data.
3454 * Note 2: Fragment lists within fragments are not implemented
3455 * at the moment, state->root_skb could be replaced with
3456 * a stack for this purpose.
3458 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3459 struct skb_seq_state *st)
3461 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3464 if (unlikely(abs_offset >= st->upper_offset)) {
3465 if (st->frag_data) {
3466 kunmap_atomic(st->frag_data);
3467 st->frag_data = NULL;
3473 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3475 if (abs_offset < block_limit && !st->frag_data) {
3476 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3477 return block_limit - abs_offset;
3480 if (st->frag_idx == 0 && !st->frag_data)
3481 st->stepped_offset += skb_headlen(st->cur_skb);
3483 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3484 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3485 block_limit = skb_frag_size(frag) + st->stepped_offset;
3487 if (abs_offset < block_limit) {
3489 st->frag_data = kmap_atomic(skb_frag_page(frag));
3491 *data = (u8 *) st->frag_data + frag->page_offset +
3492 (abs_offset - st->stepped_offset);
3494 return block_limit - abs_offset;
3497 if (st->frag_data) {
3498 kunmap_atomic(st->frag_data);
3499 st->frag_data = NULL;
3503 st->stepped_offset += skb_frag_size(frag);
3506 if (st->frag_data) {
3507 kunmap_atomic(st->frag_data);
3508 st->frag_data = NULL;
3511 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3512 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3515 } else if (st->cur_skb->next) {
3516 st->cur_skb = st->cur_skb->next;
3523 EXPORT_SYMBOL(skb_seq_read);
3526 * skb_abort_seq_read - Abort a sequential read of skb data
3527 * @st: state variable
3529 * Must be called if skb_seq_read() was not called until it
3532 void skb_abort_seq_read(struct skb_seq_state *st)
3535 kunmap_atomic(st->frag_data);
3537 EXPORT_SYMBOL(skb_abort_seq_read);
3539 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3541 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3542 struct ts_config *conf,
3543 struct ts_state *state)
3545 return skb_seq_read(offset, text, TS_SKB_CB(state));
3548 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3550 skb_abort_seq_read(TS_SKB_CB(state));
3554 * skb_find_text - Find a text pattern in skb data
3555 * @skb: the buffer to look in
3556 * @from: search offset
3558 * @config: textsearch configuration
3560 * Finds a pattern in the skb data according to the specified
3561 * textsearch configuration. Use textsearch_next() to retrieve
3562 * subsequent occurrences of the pattern. Returns the offset
3563 * to the first occurrence or UINT_MAX if no match was found.
3565 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3566 unsigned int to, struct ts_config *config)
3568 struct ts_state state;
3571 config->get_next_block = skb_ts_get_next_block;
3572 config->finish = skb_ts_finish;
3574 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3576 ret = textsearch_find(config, &state);
3577 return (ret <= to - from ? ret : UINT_MAX);
3579 EXPORT_SYMBOL(skb_find_text);
3581 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3582 int offset, size_t size)
3584 int i = skb_shinfo(skb)->nr_frags;
3586 if (skb_can_coalesce(skb, i, page, offset)) {
3587 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3588 } else if (i < MAX_SKB_FRAGS) {
3590 skb_fill_page_desc(skb, i, page, offset, size);
3597 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3600 * skb_pull_rcsum - pull skb and update receive checksum
3601 * @skb: buffer to update
3602 * @len: length of data pulled
3604 * This function performs an skb_pull on the packet and updates
3605 * the CHECKSUM_COMPLETE checksum. It should be used on
3606 * receive path processing instead of skb_pull unless you know
3607 * that the checksum difference is zero (e.g., a valid IP header)
3608 * or you are setting ip_summed to CHECKSUM_NONE.
3610 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3612 unsigned char *data = skb->data;
3614 BUG_ON(len > skb->len);
3615 __skb_pull(skb, len);
3616 skb_postpull_rcsum(skb, data, len);
3619 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3621 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3623 skb_frag_t head_frag;
3626 page = virt_to_head_page(frag_skb->head);
3627 head_frag.page.p = page;
3628 head_frag.page_offset = frag_skb->data -
3629 (unsigned char *)page_address(page);
3630 head_frag.size = skb_headlen(frag_skb);
3635 * skb_segment - Perform protocol segmentation on skb.
3636 * @head_skb: buffer to segment
3637 * @features: features for the output path (see dev->features)
3639 * This function performs segmentation on the given skb. It returns
3640 * a pointer to the first in a list of new skbs for the segments.
3641 * In case of error it returns ERR_PTR(err).
3643 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3644 netdev_features_t features)
3646 struct sk_buff *segs = NULL;
3647 struct sk_buff *tail = NULL;
3648 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3649 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3650 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3651 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3652 struct sk_buff *frag_skb = head_skb;
3653 unsigned int offset = doffset;
3654 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3655 unsigned int partial_segs = 0;
3656 unsigned int headroom;
3657 unsigned int len = head_skb->len;
3660 int nfrags = skb_shinfo(head_skb)->nr_frags;
3666 __skb_push(head_skb, doffset);
3667 proto = skb_network_protocol(head_skb, &dummy);
3668 if (unlikely(!proto))
3669 return ERR_PTR(-EINVAL);
3671 sg = !!(features & NETIF_F_SG);
3672 csum = !!can_checksum_protocol(features, proto);
3674 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3675 if (!(features & NETIF_F_GSO_PARTIAL)) {
3676 struct sk_buff *iter;
3677 unsigned int frag_len;
3680 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3683 /* If we get here then all the required
3684 * GSO features except frag_list are supported.
3685 * Try to split the SKB to multiple GSO SKBs
3686 * with no frag_list.
3687 * Currently we can do that only when the buffers don't
3688 * have a linear part and all the buffers except
3689 * the last are of the same length.
3691 frag_len = list_skb->len;
3692 skb_walk_frags(head_skb, iter) {
3693 if (frag_len != iter->len && iter->next)
3695 if (skb_headlen(iter) && !iter->head_frag)
3701 if (len != frag_len)
3705 /* GSO partial only requires that we trim off any excess that
3706 * doesn't fit into an MSS sized block, so take care of that
3709 partial_segs = len / mss;
3710 if (partial_segs > 1)
3711 mss *= partial_segs;
3717 headroom = skb_headroom(head_skb);
3718 pos = skb_headlen(head_skb);
3721 struct sk_buff *nskb;
3722 skb_frag_t *nskb_frag;
3726 if (unlikely(mss == GSO_BY_FRAGS)) {
3727 len = list_skb->len;
3729 len = head_skb->len - offset;
3734 hsize = skb_headlen(head_skb) - offset;
3737 if (hsize > len || !sg)
3740 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3741 (skb_headlen(list_skb) == len || sg)) {
3742 BUG_ON(skb_headlen(list_skb) > len);
3745 nfrags = skb_shinfo(list_skb)->nr_frags;
3746 frag = skb_shinfo(list_skb)->frags;
3747 frag_skb = list_skb;
3748 pos += skb_headlen(list_skb);
3750 while (pos < offset + len) {
3751 BUG_ON(i >= nfrags);
3753 size = skb_frag_size(frag);
3754 if (pos + size > offset + len)
3762 nskb = skb_clone(list_skb, GFP_ATOMIC);
3763 list_skb = list_skb->next;
3765 if (unlikely(!nskb))
3768 if (unlikely(pskb_trim(nskb, len))) {
3773 hsize = skb_end_offset(nskb);
3774 if (skb_cow_head(nskb, doffset + headroom)) {
3779 nskb->truesize += skb_end_offset(nskb) - hsize;
3780 skb_release_head_state(nskb);
3781 __skb_push(nskb, doffset);
3783 nskb = __alloc_skb(hsize + doffset + headroom,
3784 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3787 if (unlikely(!nskb))
3790 skb_reserve(nskb, headroom);
3791 __skb_put(nskb, doffset);
3800 __copy_skb_header(nskb, head_skb);
3802 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3803 skb_reset_mac_len(nskb);
3805 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3806 nskb->data - tnl_hlen,
3807 doffset + tnl_hlen);
3809 if (nskb->len == len + doffset)
3810 goto perform_csum_check;
3813 if (!nskb->remcsum_offload)
3814 nskb->ip_summed = CHECKSUM_NONE;
3815 SKB_GSO_CB(nskb)->csum =
3816 skb_copy_and_csum_bits(head_skb, offset,
3819 SKB_GSO_CB(nskb)->csum_start =
3820 skb_headroom(nskb) + doffset;
3824 nskb_frag = skb_shinfo(nskb)->frags;
3826 skb_copy_from_linear_data_offset(head_skb, offset,
3827 skb_put(nskb, hsize), hsize);
3829 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3832 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3833 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3836 while (pos < offset + len) {
3839 nfrags = skb_shinfo(list_skb)->nr_frags;
3840 frag = skb_shinfo(list_skb)->frags;
3841 frag_skb = list_skb;
3842 if (!skb_headlen(list_skb)) {
3845 BUG_ON(!list_skb->head_frag);
3847 /* to make room for head_frag. */
3851 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3852 skb_zerocopy_clone(nskb, frag_skb,
3856 list_skb = list_skb->next;
3859 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3861 net_warn_ratelimited(
3862 "skb_segment: too many frags: %u %u\n",
3868 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3869 __skb_frag_ref(nskb_frag);
3870 size = skb_frag_size(nskb_frag);
3873 nskb_frag->page_offset += offset - pos;
3874 skb_frag_size_sub(nskb_frag, offset - pos);
3877 skb_shinfo(nskb)->nr_frags++;
3879 if (pos + size <= offset + len) {
3884 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3892 nskb->data_len = len - hsize;
3893 nskb->len += nskb->data_len;
3894 nskb->truesize += nskb->data_len;
3898 if (skb_has_shared_frag(nskb) &&
3899 __skb_linearize(nskb))
3902 if (!nskb->remcsum_offload)
3903 nskb->ip_summed = CHECKSUM_NONE;
3904 SKB_GSO_CB(nskb)->csum =
3905 skb_checksum(nskb, doffset,
3906 nskb->len - doffset, 0);
3907 SKB_GSO_CB(nskb)->csum_start =
3908 skb_headroom(nskb) + doffset;
3910 } while ((offset += len) < head_skb->len);
3912 /* Some callers want to get the end of the list.
3913 * Put it in segs->prev to avoid walking the list.
3914 * (see validate_xmit_skb_list() for example)
3919 struct sk_buff *iter;
3920 int type = skb_shinfo(head_skb)->gso_type;
3921 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3923 /* Update type to add partial and then remove dodgy if set */
3924 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3925 type &= ~SKB_GSO_DODGY;
3927 /* Update GSO info and prepare to start updating headers on
3928 * our way back down the stack of protocols.
3930 for (iter = segs; iter; iter = iter->next) {
3931 skb_shinfo(iter)->gso_size = gso_size;
3932 skb_shinfo(iter)->gso_segs = partial_segs;
3933 skb_shinfo(iter)->gso_type = type;
3934 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3937 if (tail->len - doffset <= gso_size)
3938 skb_shinfo(tail)->gso_size = 0;
3939 else if (tail != segs)
3940 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3943 /* Following permits correct backpressure, for protocols
3944 * using skb_set_owner_w().
3945 * Idea is to tranfert ownership from head_skb to last segment.
3947 if (head_skb->destructor == sock_wfree) {
3948 swap(tail->truesize, head_skb->truesize);
3949 swap(tail->destructor, head_skb->destructor);
3950 swap(tail->sk, head_skb->sk);
3955 kfree_skb_list(segs);
3956 return ERR_PTR(err);
3958 EXPORT_SYMBOL_GPL(skb_segment);
3960 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
3962 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3963 unsigned int offset = skb_gro_offset(skb);
3964 unsigned int headlen = skb_headlen(skb);
3965 unsigned int len = skb_gro_len(skb);
3966 unsigned int delta_truesize;
3969 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
3972 lp = NAPI_GRO_CB(p)->last;
3973 pinfo = skb_shinfo(lp);
3975 if (headlen <= offset) {
3978 int i = skbinfo->nr_frags;
3979 int nr_frags = pinfo->nr_frags + i;
3981 if (nr_frags > MAX_SKB_FRAGS)
3985 pinfo->nr_frags = nr_frags;
3986 skbinfo->nr_frags = 0;
3988 frag = pinfo->frags + nr_frags;
3989 frag2 = skbinfo->frags + i;
3994 frag->page_offset += offset;
3995 skb_frag_size_sub(frag, offset);
3997 /* all fragments truesize : remove (head size + sk_buff) */
3998 delta_truesize = skb->truesize -
3999 SKB_TRUESIZE(skb_end_offset(skb));
4001 skb->truesize -= skb->data_len;
4002 skb->len -= skb->data_len;
4005 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4007 } else if (skb->head_frag) {
4008 int nr_frags = pinfo->nr_frags;
4009 skb_frag_t *frag = pinfo->frags + nr_frags;
4010 struct page *page = virt_to_head_page(skb->head);
4011 unsigned int first_size = headlen - offset;
4012 unsigned int first_offset;
4014 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4017 first_offset = skb->data -
4018 (unsigned char *)page_address(page) +
4021 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4023 frag->page.p = page;
4024 frag->page_offset = first_offset;
4025 skb_frag_size_set(frag, first_size);
4027 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4028 /* We dont need to clear skbinfo->nr_frags here */
4030 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4031 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4036 delta_truesize = skb->truesize;
4037 if (offset > headlen) {
4038 unsigned int eat = offset - headlen;
4040 skbinfo->frags[0].page_offset += eat;
4041 skb_frag_size_sub(&skbinfo->frags[0], eat);
4042 skb->data_len -= eat;
4047 __skb_pull(skb, offset);
4049 if (NAPI_GRO_CB(p)->last == p)
4050 skb_shinfo(p)->frag_list = skb;
4052 NAPI_GRO_CB(p)->last->next = skb;
4053 NAPI_GRO_CB(p)->last = skb;
4054 __skb_header_release(skb);
4058 NAPI_GRO_CB(p)->count++;
4060 p->truesize += delta_truesize;
4063 lp->data_len += len;
4064 lp->truesize += delta_truesize;
4067 NAPI_GRO_CB(skb)->same_flow = 1;
4070 EXPORT_SYMBOL_GPL(skb_gro_receive);
4072 #ifdef CONFIG_SKB_EXTENSIONS
4073 #define SKB_EXT_ALIGN_VALUE 8
4074 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4076 static const u8 skb_ext_type_len[] = {
4077 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4078 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4081 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4085 static __always_inline unsigned int skb_ext_total_length(void)
4087 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4088 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4089 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4092 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4097 static void skb_extensions_init(void)
4099 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4100 BUILD_BUG_ON(skb_ext_total_length() > 255);
4102 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4103 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4105 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4109 static void skb_extensions_init(void) {}
4112 void __init skb_init(void)
4114 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4115 sizeof(struct sk_buff),
4117 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4118 offsetof(struct sk_buff, cb),
4119 sizeof_field(struct sk_buff, cb),
4121 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4122 sizeof(struct sk_buff_fclones),
4124 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4126 skb_extensions_init();
4130 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4131 unsigned int recursion_level)
4133 int start = skb_headlen(skb);
4134 int i, copy = start - offset;
4135 struct sk_buff *frag_iter;
4138 if (unlikely(recursion_level >= 24))
4144 sg_set_buf(sg, skb->data + offset, copy);
4146 if ((len -= copy) == 0)
4151 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4154 WARN_ON(start > offset + len);
4156 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4157 if ((copy = end - offset) > 0) {
4158 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4159 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4164 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4165 frag->page_offset+offset-start);
4174 skb_walk_frags(skb, frag_iter) {
4177 WARN_ON(start > offset + len);
4179 end = start + frag_iter->len;
4180 if ((copy = end - offset) > 0) {
4181 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4186 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4187 copy, recursion_level + 1);
4188 if (unlikely(ret < 0))
4191 if ((len -= copy) == 0)
4202 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4203 * @skb: Socket buffer containing the buffers to be mapped
4204 * @sg: The scatter-gather list to map into
4205 * @offset: The offset into the buffer's contents to start mapping
4206 * @len: Length of buffer space to be mapped
4208 * Fill the specified scatter-gather list with mappings/pointers into a
4209 * region of the buffer space attached to a socket buffer. Returns either
4210 * the number of scatterlist items used, or -EMSGSIZE if the contents
4213 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4215 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4220 sg_mark_end(&sg[nsg - 1]);
4224 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4226 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4227 * sglist without mark the sg which contain last skb data as the end.
4228 * So the caller can mannipulate sg list as will when padding new data after
4229 * the first call without calling sg_unmark_end to expend sg list.
4231 * Scenario to use skb_to_sgvec_nomark:
4233 * 2. skb_to_sgvec_nomark(payload1)
4234 * 3. skb_to_sgvec_nomark(payload2)
4236 * This is equivalent to:
4238 * 2. skb_to_sgvec(payload1)
4240 * 4. skb_to_sgvec(payload2)
4242 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4243 * is more preferable.
4245 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4246 int offset, int len)
4248 return __skb_to_sgvec(skb, sg, offset, len, 0);
4250 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4255 * skb_cow_data - Check that a socket buffer's data buffers are writable
4256 * @skb: The socket buffer to check.
4257 * @tailbits: Amount of trailing space to be added
4258 * @trailer: Returned pointer to the skb where the @tailbits space begins
4260 * Make sure that the data buffers attached to a socket buffer are
4261 * writable. If they are not, private copies are made of the data buffers
4262 * and the socket buffer is set to use these instead.
4264 * If @tailbits is given, make sure that there is space to write @tailbits
4265 * bytes of data beyond current end of socket buffer. @trailer will be
4266 * set to point to the skb in which this space begins.
4268 * The number of scatterlist elements required to completely map the
4269 * COW'd and extended socket buffer will be returned.
4271 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4275 struct sk_buff *skb1, **skb_p;
4277 /* If skb is cloned or its head is paged, reallocate
4278 * head pulling out all the pages (pages are considered not writable
4279 * at the moment even if they are anonymous).
4281 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4282 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4285 /* Easy case. Most of packets will go this way. */
4286 if (!skb_has_frag_list(skb)) {
4287 /* A little of trouble, not enough of space for trailer.
4288 * This should not happen, when stack is tuned to generate
4289 * good frames. OK, on miss we reallocate and reserve even more
4290 * space, 128 bytes is fair. */
4292 if (skb_tailroom(skb) < tailbits &&
4293 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4301 /* Misery. We are in troubles, going to mincer fragments... */
4304 skb_p = &skb_shinfo(skb)->frag_list;
4307 while ((skb1 = *skb_p) != NULL) {
4310 /* The fragment is partially pulled by someone,
4311 * this can happen on input. Copy it and everything
4314 if (skb_shared(skb1))
4317 /* If the skb is the last, worry about trailer. */
4319 if (skb1->next == NULL && tailbits) {
4320 if (skb_shinfo(skb1)->nr_frags ||
4321 skb_has_frag_list(skb1) ||
4322 skb_tailroom(skb1) < tailbits)
4323 ntail = tailbits + 128;
4329 skb_shinfo(skb1)->nr_frags ||
4330 skb_has_frag_list(skb1)) {
4331 struct sk_buff *skb2;
4333 /* Fuck, we are miserable poor guys... */
4335 skb2 = skb_copy(skb1, GFP_ATOMIC);
4337 skb2 = skb_copy_expand(skb1,
4341 if (unlikely(skb2 == NULL))
4345 skb_set_owner_w(skb2, skb1->sk);
4347 /* Looking around. Are we still alive?
4348 * OK, link new skb, drop old one */
4350 skb2->next = skb1->next;
4357 skb_p = &skb1->next;
4362 EXPORT_SYMBOL_GPL(skb_cow_data);
4364 static void sock_rmem_free(struct sk_buff *skb)
4366 struct sock *sk = skb->sk;
4368 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4371 static void skb_set_err_queue(struct sk_buff *skb)
4373 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4374 * So, it is safe to (mis)use it to mark skbs on the error queue.
4376 skb->pkt_type = PACKET_OUTGOING;
4377 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4381 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4383 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4385 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4386 (unsigned int)sk->sk_rcvbuf)
4391 skb->destructor = sock_rmem_free;
4392 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4393 skb_set_err_queue(skb);
4395 /* before exiting rcu section, make sure dst is refcounted */
4398 skb_queue_tail(&sk->sk_error_queue, skb);
4399 if (!sock_flag(sk, SOCK_DEAD))
4400 sk->sk_error_report(sk);
4403 EXPORT_SYMBOL(sock_queue_err_skb);
4405 static bool is_icmp_err_skb(const struct sk_buff *skb)
4407 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4408 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4411 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4413 struct sk_buff_head *q = &sk->sk_error_queue;
4414 struct sk_buff *skb, *skb_next = NULL;
4415 bool icmp_next = false;
4416 unsigned long flags;
4418 spin_lock_irqsave(&q->lock, flags);
4419 skb = __skb_dequeue(q);
4420 if (skb && (skb_next = skb_peek(q))) {
4421 icmp_next = is_icmp_err_skb(skb_next);
4423 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4425 spin_unlock_irqrestore(&q->lock, flags);
4427 if (is_icmp_err_skb(skb) && !icmp_next)
4431 sk->sk_error_report(sk);
4435 EXPORT_SYMBOL(sock_dequeue_err_skb);
4438 * skb_clone_sk - create clone of skb, and take reference to socket
4439 * @skb: the skb to clone
4441 * This function creates a clone of a buffer that holds a reference on
4442 * sk_refcnt. Buffers created via this function are meant to be
4443 * returned using sock_queue_err_skb, or free via kfree_skb.
4445 * When passing buffers allocated with this function to sock_queue_err_skb
4446 * it is necessary to wrap the call with sock_hold/sock_put in order to
4447 * prevent the socket from being released prior to being enqueued on
4448 * the sk_error_queue.
4450 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4452 struct sock *sk = skb->sk;
4453 struct sk_buff *clone;
4455 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4458 clone = skb_clone(skb, GFP_ATOMIC);
4465 clone->destructor = sock_efree;
4469 EXPORT_SYMBOL(skb_clone_sk);
4471 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4476 struct sock_exterr_skb *serr;
4479 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4481 serr = SKB_EXT_ERR(skb);
4482 memset(serr, 0, sizeof(*serr));
4483 serr->ee.ee_errno = ENOMSG;
4484 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4485 serr->ee.ee_info = tstype;
4486 serr->opt_stats = opt_stats;
4487 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4488 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4489 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4490 if (sk->sk_protocol == IPPROTO_TCP &&
4491 sk->sk_type == SOCK_STREAM)
4492 serr->ee.ee_data -= sk->sk_tskey;
4495 err = sock_queue_err_skb(sk, skb);
4501 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4505 if (likely(sysctl_tstamp_allow_data || tsonly))
4508 read_lock_bh(&sk->sk_callback_lock);
4509 ret = sk->sk_socket && sk->sk_socket->file &&
4510 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4511 read_unlock_bh(&sk->sk_callback_lock);
4515 void skb_complete_tx_timestamp(struct sk_buff *skb,
4516 struct skb_shared_hwtstamps *hwtstamps)
4518 struct sock *sk = skb->sk;
4520 if (!skb_may_tx_timestamp(sk, false))
4523 /* Take a reference to prevent skb_orphan() from freeing the socket,
4524 * but only if the socket refcount is not zero.
4526 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4527 *skb_hwtstamps(skb) = *hwtstamps;
4528 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4536 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4538 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4539 struct skb_shared_hwtstamps *hwtstamps,
4540 struct sock *sk, int tstype)
4542 struct sk_buff *skb;
4543 bool tsonly, opt_stats = false;
4548 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4549 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4552 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4553 if (!skb_may_tx_timestamp(sk, tsonly))
4558 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4559 sk->sk_protocol == IPPROTO_TCP &&
4560 sk->sk_type == SOCK_STREAM) {
4561 skb = tcp_get_timestamping_opt_stats(sk);
4565 skb = alloc_skb(0, GFP_ATOMIC);
4567 skb = skb_clone(orig_skb, GFP_ATOMIC);
4573 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4575 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4579 *skb_hwtstamps(skb) = *hwtstamps;
4581 skb->tstamp = ktime_get_real();
4583 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4585 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4587 void skb_tstamp_tx(struct sk_buff *orig_skb,
4588 struct skb_shared_hwtstamps *hwtstamps)
4590 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4593 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4595 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4597 struct sock *sk = skb->sk;
4598 struct sock_exterr_skb *serr;
4601 skb->wifi_acked_valid = 1;
4602 skb->wifi_acked = acked;
4604 serr = SKB_EXT_ERR(skb);
4605 memset(serr, 0, sizeof(*serr));
4606 serr->ee.ee_errno = ENOMSG;
4607 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4609 /* Take a reference to prevent skb_orphan() from freeing the socket,
4610 * but only if the socket refcount is not zero.
4612 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4613 err = sock_queue_err_skb(sk, skb);
4619 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4622 * skb_partial_csum_set - set up and verify partial csum values for packet
4623 * @skb: the skb to set
4624 * @start: the number of bytes after skb->data to start checksumming.
4625 * @off: the offset from start to place the checksum.
4627 * For untrusted partially-checksummed packets, we need to make sure the values
4628 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4630 * This function checks and sets those values and skb->ip_summed: if this
4631 * returns false you should drop the packet.
4633 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4635 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4636 u32 csum_start = skb_headroom(skb) + (u32)start;
4638 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4639 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4640 start, off, skb_headroom(skb), skb_headlen(skb));
4643 skb->ip_summed = CHECKSUM_PARTIAL;
4644 skb->csum_start = csum_start;
4645 skb->csum_offset = off;
4646 skb_set_transport_header(skb, start);
4649 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4651 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4654 if (skb_headlen(skb) >= len)
4657 /* If we need to pullup then pullup to the max, so we
4658 * won't need to do it again.
4663 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4666 if (skb_headlen(skb) < len)
4672 #define MAX_TCP_HDR_LEN (15 * 4)
4674 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4675 typeof(IPPROTO_IP) proto,
4682 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4683 off + MAX_TCP_HDR_LEN);
4684 if (!err && !skb_partial_csum_set(skb, off,
4685 offsetof(struct tcphdr,
4688 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4691 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4692 off + sizeof(struct udphdr));
4693 if (!err && !skb_partial_csum_set(skb, off,
4694 offsetof(struct udphdr,
4697 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4700 return ERR_PTR(-EPROTO);
4703 /* This value should be large enough to cover a tagged ethernet header plus
4704 * maximally sized IP and TCP or UDP headers.
4706 #define MAX_IP_HDR_LEN 128
4708 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4717 err = skb_maybe_pull_tail(skb,
4718 sizeof(struct iphdr),
4723 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4726 off = ip_hdrlen(skb);
4733 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4735 return PTR_ERR(csum);
4738 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4741 ip_hdr(skb)->protocol, 0);
4748 /* This value should be large enough to cover a tagged ethernet header plus
4749 * an IPv6 header, all options, and a maximal TCP or UDP header.
4751 #define MAX_IPV6_HDR_LEN 256
4753 #define OPT_HDR(type, skb, off) \
4754 (type *)(skb_network_header(skb) + (off))
4756 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4769 off = sizeof(struct ipv6hdr);
4771 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4775 nexthdr = ipv6_hdr(skb)->nexthdr;
4777 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4778 while (off <= len && !done) {
4780 case IPPROTO_DSTOPTS:
4781 case IPPROTO_HOPOPTS:
4782 case IPPROTO_ROUTING: {
4783 struct ipv6_opt_hdr *hp;
4785 err = skb_maybe_pull_tail(skb,
4787 sizeof(struct ipv6_opt_hdr),
4792 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4793 nexthdr = hp->nexthdr;
4794 off += ipv6_optlen(hp);
4798 struct ip_auth_hdr *hp;
4800 err = skb_maybe_pull_tail(skb,
4802 sizeof(struct ip_auth_hdr),
4807 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4808 nexthdr = hp->nexthdr;
4809 off += ipv6_authlen(hp);
4812 case IPPROTO_FRAGMENT: {
4813 struct frag_hdr *hp;
4815 err = skb_maybe_pull_tail(skb,
4817 sizeof(struct frag_hdr),
4822 hp = OPT_HDR(struct frag_hdr, skb, off);
4824 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4827 nexthdr = hp->nexthdr;
4828 off += sizeof(struct frag_hdr);
4839 if (!done || fragment)
4842 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4844 return PTR_ERR(csum);
4847 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4848 &ipv6_hdr(skb)->daddr,
4849 skb->len - off, nexthdr, 0);
4857 * skb_checksum_setup - set up partial checksum offset
4858 * @skb: the skb to set up
4859 * @recalculate: if true the pseudo-header checksum will be recalculated
4861 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4865 switch (skb->protocol) {
4866 case htons(ETH_P_IP):
4867 err = skb_checksum_setup_ipv4(skb, recalculate);
4870 case htons(ETH_P_IPV6):
4871 err = skb_checksum_setup_ipv6(skb, recalculate);
4881 EXPORT_SYMBOL(skb_checksum_setup);
4884 * skb_checksum_maybe_trim - maybe trims the given skb
4885 * @skb: the skb to check
4886 * @transport_len: the data length beyond the network header
4888 * Checks whether the given skb has data beyond the given transport length.
4889 * If so, returns a cloned skb trimmed to this transport length.
4890 * Otherwise returns the provided skb. Returns NULL in error cases
4891 * (e.g. transport_len exceeds skb length or out-of-memory).
4893 * Caller needs to set the skb transport header and free any returned skb if it
4894 * differs from the provided skb.
4896 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4897 unsigned int transport_len)
4899 struct sk_buff *skb_chk;
4900 unsigned int len = skb_transport_offset(skb) + transport_len;
4905 else if (skb->len == len)
4908 skb_chk = skb_clone(skb, GFP_ATOMIC);
4912 ret = pskb_trim_rcsum(skb_chk, len);
4922 * skb_checksum_trimmed - validate checksum of an skb
4923 * @skb: the skb to check
4924 * @transport_len: the data length beyond the network header
4925 * @skb_chkf: checksum function to use
4927 * Applies the given checksum function skb_chkf to the provided skb.
4928 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4930 * If the skb has data beyond the given transport length, then a
4931 * trimmed & cloned skb is checked and returned.
4933 * Caller needs to set the skb transport header and free any returned skb if it
4934 * differs from the provided skb.
4936 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4937 unsigned int transport_len,
4938 __sum16(*skb_chkf)(struct sk_buff *skb))
4940 struct sk_buff *skb_chk;
4941 unsigned int offset = skb_transport_offset(skb);
4944 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4948 if (!pskb_may_pull(skb_chk, offset))
4951 skb_pull_rcsum(skb_chk, offset);
4952 ret = skb_chkf(skb_chk);
4953 skb_push_rcsum(skb_chk, offset);
4961 if (skb_chk && skb_chk != skb)
4967 EXPORT_SYMBOL(skb_checksum_trimmed);
4969 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4971 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4974 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4976 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4979 skb_release_head_state(skb);
4980 kmem_cache_free(skbuff_head_cache, skb);
4985 EXPORT_SYMBOL(kfree_skb_partial);
4988 * skb_try_coalesce - try to merge skb to prior one
4990 * @from: buffer to add
4991 * @fragstolen: pointer to boolean
4992 * @delta_truesize: how much more was allocated than was requested
4994 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4995 bool *fragstolen, int *delta_truesize)
4997 struct skb_shared_info *to_shinfo, *from_shinfo;
4998 int i, delta, len = from->len;
5000 *fragstolen = false;
5005 if (len <= skb_tailroom(to)) {
5007 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5008 *delta_truesize = 0;
5012 to_shinfo = skb_shinfo(to);
5013 from_shinfo = skb_shinfo(from);
5014 if (to_shinfo->frag_list || from_shinfo->frag_list)
5016 if (skb_zcopy(to) || skb_zcopy(from))
5019 if (skb_headlen(from) != 0) {
5021 unsigned int offset;
5023 if (to_shinfo->nr_frags +
5024 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5027 if (skb_head_is_locked(from))
5030 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5032 page = virt_to_head_page(from->head);
5033 offset = from->data - (unsigned char *)page_address(page);
5035 skb_fill_page_desc(to, to_shinfo->nr_frags,
5036 page, offset, skb_headlen(from));
5039 if (to_shinfo->nr_frags +
5040 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5043 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5046 WARN_ON_ONCE(delta < len);
5048 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5050 from_shinfo->nr_frags * sizeof(skb_frag_t));
5051 to_shinfo->nr_frags += from_shinfo->nr_frags;
5053 if (!skb_cloned(from))
5054 from_shinfo->nr_frags = 0;
5056 /* if the skb is not cloned this does nothing
5057 * since we set nr_frags to 0.
5059 for (i = 0; i < from_shinfo->nr_frags; i++)
5060 __skb_frag_ref(&from_shinfo->frags[i]);
5062 to->truesize += delta;
5064 to->data_len += len;
5066 *delta_truesize = delta;
5069 EXPORT_SYMBOL(skb_try_coalesce);
5072 * skb_scrub_packet - scrub an skb
5074 * @skb: buffer to clean
5075 * @xnet: packet is crossing netns
5077 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5078 * into/from a tunnel. Some information have to be cleared during these
5080 * skb_scrub_packet can also be used to clean a skb before injecting it in
5081 * another namespace (@xnet == true). We have to clear all information in the
5082 * skb that could impact namespace isolation.
5084 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5086 skb->pkt_type = PACKET_HOST;
5092 nf_reset_trace(skb);
5094 #ifdef CONFIG_NET_SWITCHDEV
5095 skb->offload_fwd_mark = 0;
5096 skb->offload_l3_fwd_mark = 0;
5106 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5109 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5113 * skb_gso_transport_seglen is used to determine the real size of the
5114 * individual segments, including Layer4 headers (TCP/UDP).
5116 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5118 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5120 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5121 unsigned int thlen = 0;
5123 if (skb->encapsulation) {
5124 thlen = skb_inner_transport_header(skb) -
5125 skb_transport_header(skb);
5127 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5128 thlen += inner_tcp_hdrlen(skb);
5129 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5130 thlen = tcp_hdrlen(skb);
5131 } else if (unlikely(skb_is_gso_sctp(skb))) {
5132 thlen = sizeof(struct sctphdr);
5133 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5134 thlen = sizeof(struct udphdr);
5136 /* UFO sets gso_size to the size of the fragmentation
5137 * payload, i.e. the size of the L4 (UDP) header is already
5140 return thlen + shinfo->gso_size;
5144 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5148 * skb_gso_network_seglen is used to determine the real size of the
5149 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5151 * The MAC/L2 header is not accounted for.
5153 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5155 unsigned int hdr_len = skb_transport_header(skb) -
5156 skb_network_header(skb);
5158 return hdr_len + skb_gso_transport_seglen(skb);
5162 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5166 * skb_gso_mac_seglen is used to determine the real size of the
5167 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5168 * headers (TCP/UDP).
5170 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5172 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5174 return hdr_len + skb_gso_transport_seglen(skb);
5178 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5180 * There are a couple of instances where we have a GSO skb, and we
5181 * want to determine what size it would be after it is segmented.
5183 * We might want to check:
5184 * - L3+L4+payload size (e.g. IP forwarding)
5185 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5187 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5191 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5192 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5194 * @max_len: The maximum permissible length.
5196 * Returns true if the segmented length <= max length.
5198 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5199 unsigned int seg_len,
5200 unsigned int max_len) {
5201 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5202 const struct sk_buff *iter;
5204 if (shinfo->gso_size != GSO_BY_FRAGS)
5205 return seg_len <= max_len;
5207 /* Undo this so we can re-use header sizes */
5208 seg_len -= GSO_BY_FRAGS;
5210 skb_walk_frags(skb, iter) {
5211 if (seg_len + skb_headlen(iter) > max_len)
5219 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5222 * @mtu: MTU to validate against
5224 * skb_gso_validate_network_len validates if a given skb will fit a
5225 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5228 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5230 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5232 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5235 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5238 * @len: length to validate against
5240 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5241 * length once split, including L2, L3 and L4 headers and the payload.
5243 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5245 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5247 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5249 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5251 int mac_len, meta_len;
5254 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5259 mac_len = skb->data - skb_mac_header(skb);
5260 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5261 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5262 mac_len - VLAN_HLEN - ETH_TLEN);
5265 meta_len = skb_metadata_len(skb);
5267 meta = skb_metadata_end(skb) - meta_len;
5268 memmove(meta + VLAN_HLEN, meta, meta_len);
5271 skb->mac_header += VLAN_HLEN;
5275 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5277 struct vlan_hdr *vhdr;
5280 if (unlikely(skb_vlan_tag_present(skb))) {
5281 /* vlan_tci is already set-up so leave this for another time */
5285 skb = skb_share_check(skb, GFP_ATOMIC);
5289 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5292 vhdr = (struct vlan_hdr *)skb->data;
5293 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5294 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5296 skb_pull_rcsum(skb, VLAN_HLEN);
5297 vlan_set_encap_proto(skb, vhdr);
5299 skb = skb_reorder_vlan_header(skb);
5303 skb_reset_network_header(skb);
5304 skb_reset_transport_header(skb);
5305 skb_reset_mac_len(skb);
5313 EXPORT_SYMBOL(skb_vlan_untag);
5315 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5317 if (!pskb_may_pull(skb, write_len))
5320 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5323 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5325 EXPORT_SYMBOL(skb_ensure_writable);
5327 /* remove VLAN header from packet and update csum accordingly.
5328 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5330 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5332 struct vlan_hdr *vhdr;
5333 int offset = skb->data - skb_mac_header(skb);
5336 if (WARN_ONCE(offset,
5337 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5342 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5346 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5348 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5349 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5351 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5352 __skb_pull(skb, VLAN_HLEN);
5354 vlan_set_encap_proto(skb, vhdr);
5355 skb->mac_header += VLAN_HLEN;
5357 if (skb_network_offset(skb) < ETH_HLEN)
5358 skb_set_network_header(skb, ETH_HLEN);
5360 skb_reset_mac_len(skb);
5364 EXPORT_SYMBOL(__skb_vlan_pop);
5366 /* Pop a vlan tag either from hwaccel or from payload.
5367 * Expects skb->data at mac header.
5369 int skb_vlan_pop(struct sk_buff *skb)
5375 if (likely(skb_vlan_tag_present(skb))) {
5376 __vlan_hwaccel_clear_tag(skb);
5378 if (unlikely(!eth_type_vlan(skb->protocol)))
5381 err = __skb_vlan_pop(skb, &vlan_tci);
5385 /* move next vlan tag to hw accel tag */
5386 if (likely(!eth_type_vlan(skb->protocol)))
5389 vlan_proto = skb->protocol;
5390 err = __skb_vlan_pop(skb, &vlan_tci);
5394 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5397 EXPORT_SYMBOL(skb_vlan_pop);
5399 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5400 * Expects skb->data at mac header.
5402 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5404 if (skb_vlan_tag_present(skb)) {
5405 int offset = skb->data - skb_mac_header(skb);
5408 if (WARN_ONCE(offset,
5409 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5414 err = __vlan_insert_tag(skb, skb->vlan_proto,
5415 skb_vlan_tag_get(skb));
5419 skb->protocol = skb->vlan_proto;
5420 skb->mac_len += VLAN_HLEN;
5422 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5424 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5427 EXPORT_SYMBOL(skb_vlan_push);
5429 /* Update the ethertype of hdr and the skb csum value if required. */
5430 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5433 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5434 __be16 diff[] = { ~hdr->h_proto, ethertype };
5436 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5439 hdr->h_proto = ethertype;
5443 * skb_mpls_push() - push a new MPLS header after the mac header
5446 * @mpls_lse: MPLS label stack entry to push
5447 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5449 * Expects skb->data at mac header.
5451 * Returns 0 on success, -errno otherwise.
5453 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto)
5455 struct mpls_shim_hdr *lse;
5458 if (unlikely(!eth_p_mpls(mpls_proto)))
5461 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5462 if (skb->encapsulation)
5465 err = skb_cow_head(skb, MPLS_HLEN);
5469 if (!skb->inner_protocol) {
5470 skb_set_inner_network_header(skb, skb->mac_len);
5471 skb_set_inner_protocol(skb, skb->protocol);
5474 skb_push(skb, MPLS_HLEN);
5475 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5477 skb_reset_mac_header(skb);
5478 skb_set_network_header(skb, skb->mac_len);
5480 lse = mpls_hdr(skb);
5481 lse->label_stack_entry = mpls_lse;
5482 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5484 if (skb->dev && skb->dev->type == ARPHRD_ETHER)
5485 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5486 skb->protocol = mpls_proto;
5490 EXPORT_SYMBOL_GPL(skb_mpls_push);
5493 * skb_mpls_pop() - pop the outermost MPLS header
5496 * @next_proto: ethertype of header after popped MPLS header
5498 * Expects skb->data at mac header.
5500 * Returns 0 on success, -errno otherwise.
5502 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto)
5506 if (unlikely(!eth_p_mpls(skb->protocol)))
5509 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5513 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5514 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5517 __skb_pull(skb, MPLS_HLEN);
5518 skb_reset_mac_header(skb);
5519 skb_set_network_header(skb, skb->mac_len);
5521 if (skb->dev && skb->dev->type == ARPHRD_ETHER) {
5524 /* use mpls_hdr() to get ethertype to account for VLANs. */
5525 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5526 skb_mod_eth_type(skb, hdr, next_proto);
5528 skb->protocol = next_proto;
5532 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5535 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5538 * @mpls_lse: new MPLS label stack entry to update to
5540 * Expects skb->data at mac header.
5542 * Returns 0 on success, -errno otherwise.
5544 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5548 if (unlikely(!eth_p_mpls(skb->protocol)))
5551 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5555 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5556 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5558 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5561 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5565 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5568 * alloc_skb_with_frags - allocate skb with page frags
5570 * @header_len: size of linear part
5571 * @data_len: needed length in frags
5572 * @max_page_order: max page order desired.
5573 * @errcode: pointer to error code if any
5574 * @gfp_mask: allocation mask
5576 * This can be used to allocate a paged skb, given a maximal order for frags.
5578 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5579 unsigned long data_len,
5584 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5585 unsigned long chunk;
5586 struct sk_buff *skb;
5590 *errcode = -EMSGSIZE;
5591 /* Note this test could be relaxed, if we succeed to allocate
5592 * high order pages...
5594 if (npages > MAX_SKB_FRAGS)
5597 *errcode = -ENOBUFS;
5598 skb = alloc_skb(header_len, gfp_mask);
5602 skb->truesize += npages << PAGE_SHIFT;
5604 for (i = 0; npages > 0; i++) {
5605 int order = max_page_order;
5608 if (npages >= 1 << order) {
5609 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5615 /* Do not retry other high order allocations */
5621 page = alloc_page(gfp_mask);
5625 chunk = min_t(unsigned long, data_len,
5626 PAGE_SIZE << order);
5627 skb_fill_page_desc(skb, i, page, 0, chunk);
5629 npages -= 1 << order;
5637 EXPORT_SYMBOL(alloc_skb_with_frags);
5639 /* carve out the first off bytes from skb when off < headlen */
5640 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5641 const int headlen, gfp_t gfp_mask)
5644 int size = skb_end_offset(skb);
5645 int new_hlen = headlen - off;
5648 size = SKB_DATA_ALIGN(size);
5650 if (skb_pfmemalloc(skb))
5651 gfp_mask |= __GFP_MEMALLOC;
5652 data = kmalloc_reserve(size +
5653 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5654 gfp_mask, NUMA_NO_NODE, NULL);
5658 size = SKB_WITH_OVERHEAD(ksize(data));
5660 /* Copy real data, and all frags */
5661 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5664 memcpy((struct skb_shared_info *)(data + size),
5666 offsetof(struct skb_shared_info,
5667 frags[skb_shinfo(skb)->nr_frags]));
5668 if (skb_cloned(skb)) {
5669 /* drop the old head gracefully */
5670 if (skb_orphan_frags(skb, gfp_mask)) {
5674 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5675 skb_frag_ref(skb, i);
5676 if (skb_has_frag_list(skb))
5677 skb_clone_fraglist(skb);
5678 skb_release_data(skb);
5680 /* we can reuse existing recount- all we did was
5689 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5692 skb->end = skb->head + size;
5694 skb_set_tail_pointer(skb, skb_headlen(skb));
5695 skb_headers_offset_update(skb, 0);
5699 atomic_set(&skb_shinfo(skb)->dataref, 1);
5704 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5706 /* carve out the first eat bytes from skb's frag_list. May recurse into
5709 static int pskb_carve_frag_list(struct sk_buff *skb,
5710 struct skb_shared_info *shinfo, int eat,
5713 struct sk_buff *list = shinfo->frag_list;
5714 struct sk_buff *clone = NULL;
5715 struct sk_buff *insp = NULL;
5719 pr_err("Not enough bytes to eat. Want %d\n", eat);
5722 if (list->len <= eat) {
5723 /* Eaten as whole. */
5728 /* Eaten partially. */
5729 if (skb_shared(list)) {
5730 clone = skb_clone(list, gfp_mask);
5736 /* This may be pulled without problems. */
5739 if (pskb_carve(list, eat, gfp_mask) < 0) {
5747 /* Free pulled out fragments. */
5748 while ((list = shinfo->frag_list) != insp) {
5749 shinfo->frag_list = list->next;
5752 /* And insert new clone at head. */
5755 shinfo->frag_list = clone;
5760 /* carve off first len bytes from skb. Split line (off) is in the
5761 * non-linear part of skb
5763 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5764 int pos, gfp_t gfp_mask)
5767 int size = skb_end_offset(skb);
5769 const int nfrags = skb_shinfo(skb)->nr_frags;
5770 struct skb_shared_info *shinfo;
5772 size = SKB_DATA_ALIGN(size);
5774 if (skb_pfmemalloc(skb))
5775 gfp_mask |= __GFP_MEMALLOC;
5776 data = kmalloc_reserve(size +
5777 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5778 gfp_mask, NUMA_NO_NODE, NULL);
5782 size = SKB_WITH_OVERHEAD(ksize(data));
5784 memcpy((struct skb_shared_info *)(data + size),
5785 skb_shinfo(skb), offsetof(struct skb_shared_info,
5786 frags[skb_shinfo(skb)->nr_frags]));
5787 if (skb_orphan_frags(skb, gfp_mask)) {
5791 shinfo = (struct skb_shared_info *)(data + size);
5792 for (i = 0; i < nfrags; i++) {
5793 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5795 if (pos + fsize > off) {
5796 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5800 * We have two variants in this case:
5801 * 1. Move all the frag to the second
5802 * part, if it is possible. F.e.
5803 * this approach is mandatory for TUX,
5804 * where splitting is expensive.
5805 * 2. Split is accurately. We make this.
5807 shinfo->frags[0].page_offset += off - pos;
5808 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5810 skb_frag_ref(skb, i);
5815 shinfo->nr_frags = k;
5816 if (skb_has_frag_list(skb))
5817 skb_clone_fraglist(skb);
5820 /* split line is in frag list */
5821 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5823 skb_release_data(skb);
5828 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5831 skb->end = skb->head + size;
5833 skb_reset_tail_pointer(skb);
5834 skb_headers_offset_update(skb, 0);
5839 skb->data_len = skb->len;
5840 atomic_set(&skb_shinfo(skb)->dataref, 1);
5844 /* remove len bytes from the beginning of the skb */
5845 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
5847 int headlen = skb_headlen(skb);
5850 return pskb_carve_inside_header(skb, len, headlen, gfp);
5852 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
5855 /* Extract to_copy bytes starting at off from skb, and return this in
5858 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
5859 int to_copy, gfp_t gfp)
5861 struct sk_buff *clone = skb_clone(skb, gfp);
5866 if (pskb_carve(clone, off, gfp) < 0 ||
5867 pskb_trim(clone, to_copy)) {
5873 EXPORT_SYMBOL(pskb_extract);
5876 * skb_condense - try to get rid of fragments/frag_list if possible
5879 * Can be used to save memory before skb is added to a busy queue.
5880 * If packet has bytes in frags and enough tail room in skb->head,
5881 * pull all of them, so that we can free the frags right now and adjust
5884 * We do not reallocate skb->head thus can not fail.
5885 * Caller must re-evaluate skb->truesize if needed.
5887 void skb_condense(struct sk_buff *skb)
5889 if (skb->data_len) {
5890 if (skb->data_len > skb->end - skb->tail ||
5894 /* Nice, we can free page frag(s) right now */
5895 __pskb_pull_tail(skb, skb->data_len);
5897 /* At this point, skb->truesize might be over estimated,
5898 * because skb had a fragment, and fragments do not tell
5900 * When we pulled its content into skb->head, fragment
5901 * was freed, but __pskb_pull_tail() could not possibly
5902 * adjust skb->truesize, not knowing the frag truesize.
5904 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
5907 #ifdef CONFIG_SKB_EXTENSIONS
5908 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
5910 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
5913 static struct skb_ext *skb_ext_alloc(void)
5915 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5918 memset(new->offset, 0, sizeof(new->offset));
5919 refcount_set(&new->refcnt, 1);
5925 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
5926 unsigned int old_active)
5928 struct skb_ext *new;
5930 if (refcount_read(&old->refcnt) == 1)
5933 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
5937 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
5938 refcount_set(&new->refcnt, 1);
5941 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
5942 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
5945 for (i = 0; i < sp->len; i++)
5946 xfrm_state_hold(sp->xvec[i]);
5954 * skb_ext_add - allocate space for given extension, COW if needed
5956 * @id: extension to allocate space for
5958 * Allocates enough space for the given extension.
5959 * If the extension is already present, a pointer to that extension
5962 * If the skb was cloned, COW applies and the returned memory can be
5963 * modified without changing the extension space of clones buffers.
5965 * Returns pointer to the extension or NULL on allocation failure.
5967 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
5969 struct skb_ext *new, *old = NULL;
5970 unsigned int newlen, newoff;
5972 if (skb->active_extensions) {
5973 old = skb->extensions;
5975 new = skb_ext_maybe_cow(old, skb->active_extensions);
5979 if (__skb_ext_exist(new, id))
5982 newoff = new->chunks;
5984 newoff = SKB_EXT_CHUNKSIZEOF(*new);
5986 new = skb_ext_alloc();
5991 newlen = newoff + skb_ext_type_len[id];
5992 new->chunks = newlen;
5993 new->offset[id] = newoff;
5995 skb->extensions = new;
5996 skb->active_extensions |= 1 << id;
5997 return skb_ext_get_ptr(new, id);
5999 EXPORT_SYMBOL(skb_ext_add);
6002 static void skb_ext_put_sp(struct sec_path *sp)
6006 for (i = 0; i < sp->len; i++)
6007 xfrm_state_put(sp->xvec[i]);
6011 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6013 struct skb_ext *ext = skb->extensions;
6015 skb->active_extensions &= ~(1 << id);
6016 if (skb->active_extensions == 0) {
6017 skb->extensions = NULL;
6020 } else if (id == SKB_EXT_SEC_PATH &&
6021 refcount_read(&ext->refcnt) == 1) {
6022 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6029 EXPORT_SYMBOL(__skb_ext_del);
6031 void __skb_ext_put(struct skb_ext *ext)
6033 /* If this is last clone, nothing can increment
6034 * it after check passes. Avoids one atomic op.
6036 if (refcount_read(&ext->refcnt) == 1)
6039 if (!refcount_dec_and_test(&ext->refcnt))
6043 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6044 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6047 kmem_cache_free(skbuff_ext_cache, ext);
6049 EXPORT_SYMBOL(__skb_ext_put);
6050 #endif /* CONFIG_SKB_EXTENSIONS */