Merge tag 'iommu-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/arm64/linux
[linux/fpc-iii.git] / net / core / skbuff.c
blobf62cae3f75d877adec1efbf9726c42515f3912e8
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
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>
8 * Fixes:
9 * Alan Cox : Fixed the worst of the load
10 * balancer bugs.
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
24 * NOTE:
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>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/in.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>
51 #endif
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/splice.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 #include <linux/errqueue.h>
60 #include <linux/prefetch.h>
61 #include <linux/if_vlan.h>
62 #include <linux/mpls.h>
64 #include <net/protocol.h>
65 #include <net/dst.h>
66 #include <net/sock.h>
67 #include <net/checksum.h>
68 #include <net/ip6_checksum.h>
69 #include <net/xfrm.h>
70 #include <net/mpls.h>
71 #include <net/mptcp.h>
73 #include <linux/uaccess.h>
74 #include <trace/events/skb.h>
75 #include <linux/highmem.h>
76 #include <linux/capability.h>
77 #include <linux/user_namespace.h>
78 #include <linux/indirect_call_wrapper.h>
80 #include "datagram.h"
82 struct kmem_cache *skbuff_head_cache __ro_after_init;
83 static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84 #ifdef CONFIG_SKB_EXTENSIONS
85 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
86 #endif
87 int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88 EXPORT_SYMBOL(sysctl_max_skb_frags);
90 /**
91 * skb_panic - private function for out-of-line support
92 * @skb: buffer
93 * @sz: size
94 * @addr: address
95 * @msg: skb_over_panic or skb_under_panic
97 * Out-of-line support for skb_put() and skb_push().
98 * Called via the wrapper skb_over_panic() or skb_under_panic().
99 * Keep out of line to prevent kernel bloat.
100 * __builtin_return_address is not used because it is not always reliable.
102 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103 const char msg[])
105 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
109 BUG();
112 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114 skb_panic(skb, sz, addr, __func__);
117 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
119 skb_panic(skb, sz, addr, __func__);
123 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
124 * the caller if emergency pfmemalloc reserves are being used. If it is and
125 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
126 * may be used. Otherwise, the packet data may be discarded until enough
127 * memory is free
129 #define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
132 static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
135 void *obj;
136 bool ret_pfmemalloc = false;
139 * Try a regular allocation, when that fails and we're not entitled
140 * to the reserves, fail.
142 obj = kmalloc_node_track_caller(size,
143 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
144 node);
145 if (obj || !(gfp_pfmemalloc_allowed(flags)))
146 goto out;
148 /* Try again but now we are using pfmemalloc reserves */
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
152 out:
153 if (pfmemalloc)
154 *pfmemalloc = ret_pfmemalloc;
156 return obj;
159 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
160 * 'private' fields and also do memory statistics to find all the
161 * [BEEP] leaks.
166 * __alloc_skb - allocate a network buffer
167 * @size: size to allocate
168 * @gfp_mask: allocation mask
169 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
170 * instead of head cache and allocate a cloned (child) skb.
171 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
172 * allocations in case the data is required for writeback
173 * @node: numa node to allocate memory on
175 * Allocate a new &sk_buff. The returned buffer has no headroom and a
176 * tail room of at least size bytes. The object has a reference count
177 * of one. The return is the buffer. On a failure the return is %NULL.
179 * Buffers may only be allocated from interrupts using a @gfp_mask of
180 * %GFP_ATOMIC.
182 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183 int flags, int node)
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
187 struct sk_buff *skb;
188 u8 *data;
189 bool pfmemalloc;
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
197 /* Get the HEAD */
198 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
199 if (!skb)
200 goto out;
201 prefetchw(skb);
203 /* We do our best to align skb_shared_info on a separate cache
204 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
205 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
206 * Both skb->head and skb_shared_info are cache line aligned.
208 size = SKB_DATA_ALIGN(size);
209 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
211 if (!data)
212 goto nodata;
213 /* kmalloc(size) might give us more room than requested.
214 * Put skb_shared_info exactly at the end of allocated zone,
215 * to allow max possible filling before reallocation.
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
221 * Only clear those fields we need to clear, not those that we will
222 * actually initialise below. Hence, don't put any more fields after
223 * the tail pointer in struct sk_buff!
225 memset(skb, 0, offsetof(struct sk_buff, tail));
226 /* Account for allocated memory : skb + skb->head */
227 skb->truesize = SKB_TRUESIZE(size);
228 skb->pfmemalloc = pfmemalloc;
229 refcount_set(&skb->users, 1);
230 skb->head = data;
231 skb->data = data;
232 skb_reset_tail_pointer(skb);
233 skb->end = skb->tail + size;
234 skb->mac_header = (typeof(skb->mac_header))~0U;
235 skb->transport_header = (typeof(skb->transport_header))~0U;
237 /* make sure we initialize shinfo sequentially */
238 shinfo = skb_shinfo(skb);
239 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240 atomic_set(&shinfo->dataref, 1);
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
253 skb_set_kcov_handle(skb, kcov_common_handle());
255 out:
256 return skb;
257 nodata:
258 kmem_cache_free(cache, skb);
259 skb = NULL;
260 goto out;
262 EXPORT_SYMBOL(__alloc_skb);
264 /* Caller must provide SKB that is memset cleared */
265 static struct sk_buff *__build_skb_around(struct sk_buff *skb,
266 void *data, unsigned int frag_size)
268 struct skb_shared_info *shinfo;
269 unsigned int size = frag_size ? : ksize(data);
271 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
273 /* Assumes caller memset cleared SKB */
274 skb->truesize = SKB_TRUESIZE(size);
275 refcount_set(&skb->users, 1);
276 skb->head = data;
277 skb->data = data;
278 skb_reset_tail_pointer(skb);
279 skb->end = skb->tail + size;
280 skb->mac_header = (typeof(skb->mac_header))~0U;
281 skb->transport_header = (typeof(skb->transport_header))~0U;
283 /* make sure we initialize shinfo sequentially */
284 shinfo = skb_shinfo(skb);
285 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
286 atomic_set(&shinfo->dataref, 1);
288 skb_set_kcov_handle(skb, kcov_common_handle());
290 return skb;
294 * __build_skb - build a network buffer
295 * @data: data buffer provided by caller
296 * @frag_size: size of data, or 0 if head was kmalloced
298 * Allocate a new &sk_buff. Caller provides space holding head and
299 * skb_shared_info. @data must have been allocated by kmalloc() only if
300 * @frag_size is 0, otherwise data should come from the page allocator
301 * or vmalloc()
302 * The return is the new skb buffer.
303 * On a failure the return is %NULL, and @data is not freed.
304 * Notes :
305 * Before IO, driver allocates only data buffer where NIC put incoming frame
306 * Driver should add room at head (NET_SKB_PAD) and
307 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
308 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
309 * before giving packet to stack.
310 * RX rings only contains data buffers, not full skbs.
312 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
314 struct sk_buff *skb;
316 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
317 if (unlikely(!skb))
318 return NULL;
320 memset(skb, 0, offsetof(struct sk_buff, tail));
322 return __build_skb_around(skb, data, frag_size);
325 /* build_skb() is wrapper over __build_skb(), that specifically
326 * takes care of skb->head and skb->pfmemalloc
327 * This means that if @frag_size is not zero, then @data must be backed
328 * by a page fragment, not kmalloc() or vmalloc()
330 struct sk_buff *build_skb(void *data, unsigned int frag_size)
332 struct sk_buff *skb = __build_skb(data, frag_size);
334 if (skb && frag_size) {
335 skb->head_frag = 1;
336 if (page_is_pfmemalloc(virt_to_head_page(data)))
337 skb->pfmemalloc = 1;
339 return skb;
341 EXPORT_SYMBOL(build_skb);
344 * build_skb_around - build a network buffer around provided skb
345 * @skb: sk_buff provide by caller, must be memset cleared
346 * @data: data buffer provided by caller
347 * @frag_size: size of data, or 0 if head was kmalloced
349 struct sk_buff *build_skb_around(struct sk_buff *skb,
350 void *data, unsigned int frag_size)
352 if (unlikely(!skb))
353 return NULL;
355 skb = __build_skb_around(skb, data, frag_size);
357 if (skb && frag_size) {
358 skb->head_frag = 1;
359 if (page_is_pfmemalloc(virt_to_head_page(data)))
360 skb->pfmemalloc = 1;
362 return skb;
364 EXPORT_SYMBOL(build_skb_around);
366 #define NAPI_SKB_CACHE_SIZE 64
368 struct napi_alloc_cache {
369 struct page_frag_cache page;
370 unsigned int skb_count;
371 void *skb_cache[NAPI_SKB_CACHE_SIZE];
374 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
375 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
377 static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
379 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
381 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
384 void *napi_alloc_frag(unsigned int fragsz)
386 fragsz = SKB_DATA_ALIGN(fragsz);
388 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
390 EXPORT_SYMBOL(napi_alloc_frag);
393 * netdev_alloc_frag - allocate a page fragment
394 * @fragsz: fragment size
396 * Allocates a frag from a page for receive buffer.
397 * Uses GFP_ATOMIC allocations.
399 void *netdev_alloc_frag(unsigned int fragsz)
401 struct page_frag_cache *nc;
402 void *data;
404 fragsz = SKB_DATA_ALIGN(fragsz);
405 if (in_irq() || irqs_disabled()) {
406 nc = this_cpu_ptr(&netdev_alloc_cache);
407 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
408 } else {
409 local_bh_disable();
410 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
411 local_bh_enable();
413 return data;
415 EXPORT_SYMBOL(netdev_alloc_frag);
418 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
419 * @dev: network device to receive on
420 * @len: length to allocate
421 * @gfp_mask: get_free_pages mask, passed to alloc_skb
423 * Allocate a new &sk_buff and assign it a usage count of one. The
424 * buffer has NET_SKB_PAD headroom built in. Users should allocate
425 * the headroom they think they need without accounting for the
426 * built in space. The built in space is used for optimisations.
428 * %NULL is returned if there is no free memory.
430 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
431 gfp_t gfp_mask)
433 struct page_frag_cache *nc;
434 struct sk_buff *skb;
435 bool pfmemalloc;
436 void *data;
438 len += NET_SKB_PAD;
440 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
441 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
442 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
443 if (!skb)
444 goto skb_fail;
445 goto skb_success;
448 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
449 len = SKB_DATA_ALIGN(len);
451 if (sk_memalloc_socks())
452 gfp_mask |= __GFP_MEMALLOC;
454 if (in_irq() || irqs_disabled()) {
455 nc = this_cpu_ptr(&netdev_alloc_cache);
456 data = page_frag_alloc(nc, len, gfp_mask);
457 pfmemalloc = nc->pfmemalloc;
458 } else {
459 local_bh_disable();
460 nc = this_cpu_ptr(&napi_alloc_cache.page);
461 data = page_frag_alloc(nc, len, gfp_mask);
462 pfmemalloc = nc->pfmemalloc;
463 local_bh_enable();
466 if (unlikely(!data))
467 return NULL;
469 skb = __build_skb(data, len);
470 if (unlikely(!skb)) {
471 skb_free_frag(data);
472 return NULL;
475 if (pfmemalloc)
476 skb->pfmemalloc = 1;
477 skb->head_frag = 1;
479 skb_success:
480 skb_reserve(skb, NET_SKB_PAD);
481 skb->dev = dev;
483 skb_fail:
484 return skb;
486 EXPORT_SYMBOL(__netdev_alloc_skb);
489 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
490 * @napi: napi instance this buffer was allocated for
491 * @len: length to allocate
492 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
494 * Allocate a new sk_buff for use in NAPI receive. This buffer will
495 * attempt to allocate the head from a special reserved region used
496 * only for NAPI Rx allocation. By doing this we can save several
497 * CPU cycles by avoiding having to disable and re-enable IRQs.
499 * %NULL is returned if there is no free memory.
501 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
502 gfp_t gfp_mask)
504 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
505 struct sk_buff *skb;
506 void *data;
508 len += NET_SKB_PAD + NET_IP_ALIGN;
510 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
511 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
512 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
513 if (!skb)
514 goto skb_fail;
515 goto skb_success;
518 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
519 len = SKB_DATA_ALIGN(len);
521 if (sk_memalloc_socks())
522 gfp_mask |= __GFP_MEMALLOC;
524 data = page_frag_alloc(&nc->page, len, gfp_mask);
525 if (unlikely(!data))
526 return NULL;
528 skb = __build_skb(data, len);
529 if (unlikely(!skb)) {
530 skb_free_frag(data);
531 return NULL;
534 if (nc->page.pfmemalloc)
535 skb->pfmemalloc = 1;
536 skb->head_frag = 1;
538 skb_success:
539 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
540 skb->dev = napi->dev;
542 skb_fail:
543 return skb;
545 EXPORT_SYMBOL(__napi_alloc_skb);
547 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
548 int size, unsigned int truesize)
550 skb_fill_page_desc(skb, i, page, off, size);
551 skb->len += size;
552 skb->data_len += size;
553 skb->truesize += truesize;
555 EXPORT_SYMBOL(skb_add_rx_frag);
557 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
558 unsigned int truesize)
560 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
562 skb_frag_size_add(frag, size);
563 skb->len += size;
564 skb->data_len += size;
565 skb->truesize += truesize;
567 EXPORT_SYMBOL(skb_coalesce_rx_frag);
569 static void skb_drop_list(struct sk_buff **listp)
571 kfree_skb_list(*listp);
572 *listp = NULL;
575 static inline void skb_drop_fraglist(struct sk_buff *skb)
577 skb_drop_list(&skb_shinfo(skb)->frag_list);
580 static void skb_clone_fraglist(struct sk_buff *skb)
582 struct sk_buff *list;
584 skb_walk_frags(skb, list)
585 skb_get(list);
588 static void skb_free_head(struct sk_buff *skb)
590 unsigned char *head = skb->head;
592 if (skb->head_frag)
593 skb_free_frag(head);
594 else
595 kfree(head);
598 static void skb_release_data(struct sk_buff *skb)
600 struct skb_shared_info *shinfo = skb_shinfo(skb);
601 int i;
603 if (skb->cloned &&
604 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
605 &shinfo->dataref))
606 return;
608 for (i = 0; i < shinfo->nr_frags; i++)
609 __skb_frag_unref(&shinfo->frags[i]);
611 if (shinfo->frag_list)
612 kfree_skb_list(shinfo->frag_list);
614 skb_zcopy_clear(skb, true);
615 skb_free_head(skb);
619 * Free an skbuff by memory without cleaning the state.
621 static void kfree_skbmem(struct sk_buff *skb)
623 struct sk_buff_fclones *fclones;
625 switch (skb->fclone) {
626 case SKB_FCLONE_UNAVAILABLE:
627 kmem_cache_free(skbuff_head_cache, skb);
628 return;
630 case SKB_FCLONE_ORIG:
631 fclones = container_of(skb, struct sk_buff_fclones, skb1);
633 /* We usually free the clone (TX completion) before original skb
634 * This test would have no chance to be true for the clone,
635 * while here, branch prediction will be good.
637 if (refcount_read(&fclones->fclone_ref) == 1)
638 goto fastpath;
639 break;
641 default: /* SKB_FCLONE_CLONE */
642 fclones = container_of(skb, struct sk_buff_fclones, skb2);
643 break;
645 if (!refcount_dec_and_test(&fclones->fclone_ref))
646 return;
647 fastpath:
648 kmem_cache_free(skbuff_fclone_cache, fclones);
651 void skb_release_head_state(struct sk_buff *skb)
653 skb_dst_drop(skb);
654 if (skb->destructor) {
655 WARN_ON(in_irq());
656 skb->destructor(skb);
658 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
659 nf_conntrack_put(skb_nfct(skb));
660 #endif
661 skb_ext_put(skb);
664 /* Free everything but the sk_buff shell. */
665 static void skb_release_all(struct sk_buff *skb)
667 skb_release_head_state(skb);
668 if (likely(skb->head))
669 skb_release_data(skb);
673 * __kfree_skb - private function
674 * @skb: buffer
676 * Free an sk_buff. Release anything attached to the buffer.
677 * Clean the state. This is an internal helper function. Users should
678 * always call kfree_skb
681 void __kfree_skb(struct sk_buff *skb)
683 skb_release_all(skb);
684 kfree_skbmem(skb);
686 EXPORT_SYMBOL(__kfree_skb);
689 * kfree_skb - free an sk_buff
690 * @skb: buffer to free
692 * Drop a reference to the buffer and free it if the usage count has
693 * hit zero.
695 void kfree_skb(struct sk_buff *skb)
697 if (!skb_unref(skb))
698 return;
700 trace_kfree_skb(skb, __builtin_return_address(0));
701 __kfree_skb(skb);
703 EXPORT_SYMBOL(kfree_skb);
705 void kfree_skb_list(struct sk_buff *segs)
707 while (segs) {
708 struct sk_buff *next = segs->next;
710 kfree_skb(segs);
711 segs = next;
714 EXPORT_SYMBOL(kfree_skb_list);
716 /* Dump skb information and contents.
718 * Must only be called from net_ratelimit()-ed paths.
720 * Dumps whole packets if full_pkt, only headers otherwise.
722 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
724 struct skb_shared_info *sh = skb_shinfo(skb);
725 struct net_device *dev = skb->dev;
726 struct sock *sk = skb->sk;
727 struct sk_buff *list_skb;
728 bool has_mac, has_trans;
729 int headroom, tailroom;
730 int i, len, seg_len;
732 if (full_pkt)
733 len = skb->len;
734 else
735 len = min_t(int, skb->len, MAX_HEADER + 128);
737 headroom = skb_headroom(skb);
738 tailroom = skb_tailroom(skb);
740 has_mac = skb_mac_header_was_set(skb);
741 has_trans = skb_transport_header_was_set(skb);
743 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
744 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
745 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
746 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
747 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
748 level, skb->len, headroom, skb_headlen(skb), tailroom,
749 has_mac ? skb->mac_header : -1,
750 has_mac ? skb_mac_header_len(skb) : -1,
751 skb->network_header,
752 has_trans ? skb_network_header_len(skb) : -1,
753 has_trans ? skb->transport_header : -1,
754 sh->tx_flags, sh->nr_frags,
755 sh->gso_size, sh->gso_type, sh->gso_segs,
756 skb->csum, skb->ip_summed, skb->csum_complete_sw,
757 skb->csum_valid, skb->csum_level,
758 skb->hash, skb->sw_hash, skb->l4_hash,
759 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
761 if (dev)
762 printk("%sdev name=%s feat=0x%pNF\n",
763 level, dev->name, &dev->features);
764 if (sk)
765 printk("%ssk family=%hu type=%u proto=%u\n",
766 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
768 if (full_pkt && headroom)
769 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
770 16, 1, skb->head, headroom, false);
772 seg_len = min_t(int, skb_headlen(skb), len);
773 if (seg_len)
774 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
775 16, 1, skb->data, seg_len, false);
776 len -= seg_len;
778 if (full_pkt && tailroom)
779 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
780 16, 1, skb_tail_pointer(skb), tailroom, false);
782 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
783 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
784 u32 p_off, p_len, copied;
785 struct page *p;
786 u8 *vaddr;
788 skb_frag_foreach_page(frag, skb_frag_off(frag),
789 skb_frag_size(frag), p, p_off, p_len,
790 copied) {
791 seg_len = min_t(int, p_len, len);
792 vaddr = kmap_atomic(p);
793 print_hex_dump(level, "skb frag: ",
794 DUMP_PREFIX_OFFSET,
795 16, 1, vaddr + p_off, seg_len, false);
796 kunmap_atomic(vaddr);
797 len -= seg_len;
798 if (!len)
799 break;
803 if (full_pkt && skb_has_frag_list(skb)) {
804 printk("skb fraglist:\n");
805 skb_walk_frags(skb, list_skb)
806 skb_dump(level, list_skb, true);
809 EXPORT_SYMBOL(skb_dump);
812 * skb_tx_error - report an sk_buff xmit error
813 * @skb: buffer that triggered an error
815 * Report xmit error if a device callback is tracking this skb.
816 * skb must be freed afterwards.
818 void skb_tx_error(struct sk_buff *skb)
820 skb_zcopy_clear(skb, true);
822 EXPORT_SYMBOL(skb_tx_error);
824 #ifdef CONFIG_TRACEPOINTS
826 * consume_skb - free an skbuff
827 * @skb: buffer to free
829 * Drop a ref to the buffer and free it if the usage count has hit zero
830 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
831 * is being dropped after a failure and notes that
833 void consume_skb(struct sk_buff *skb)
835 if (!skb_unref(skb))
836 return;
838 trace_consume_skb(skb);
839 __kfree_skb(skb);
841 EXPORT_SYMBOL(consume_skb);
842 #endif
845 * __consume_stateless_skb - free an skbuff, assuming it is stateless
846 * @skb: buffer to free
848 * Alike consume_skb(), but this variant assumes that this is the last
849 * skb reference and all the head states have been already dropped
851 void __consume_stateless_skb(struct sk_buff *skb)
853 trace_consume_skb(skb);
854 skb_release_data(skb);
855 kfree_skbmem(skb);
858 void __kfree_skb_flush(void)
860 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
862 /* flush skb_cache if containing objects */
863 if (nc->skb_count) {
864 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
865 nc->skb_cache);
866 nc->skb_count = 0;
870 static inline void _kfree_skb_defer(struct sk_buff *skb)
872 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
874 /* drop skb->head and call any destructors for packet */
875 skb_release_all(skb);
877 /* record skb to CPU local list */
878 nc->skb_cache[nc->skb_count++] = skb;
880 #ifdef CONFIG_SLUB
881 /* SLUB writes into objects when freeing */
882 prefetchw(skb);
883 #endif
885 /* flush skb_cache if it is filled */
886 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
887 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
888 nc->skb_cache);
889 nc->skb_count = 0;
892 void __kfree_skb_defer(struct sk_buff *skb)
894 _kfree_skb_defer(skb);
897 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);
902 return;
905 lockdep_assert_in_softirq();
907 if (!skb_unref(skb))
908 return;
910 /* if reaching here SKB is ready to free */
911 trace_consume_skb(skb);
913 /* if SKB is a clone, don't handle this case */
914 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
915 __kfree_skb(skb);
916 return;
919 _kfree_skb_defer(skb);
921 EXPORT_SYMBOL(napi_consume_skb);
923 /* Make sure a field is enclosed inside headers_start/headers_end section */
924 #define CHECK_SKB_FIELD(field) \
925 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
926 offsetof(struct sk_buff, headers_start)); \
927 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
928 offsetof(struct sk_buff, headers_end)); \
930 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
932 new->tstamp = old->tstamp;
933 /* We do not copy old->sk */
934 new->dev = old->dev;
935 memcpy(new->cb, old->cb, sizeof(old->cb));
936 skb_dst_copy(new, old);
937 __skb_ext_copy(new, old);
938 __nf_copy(new, old, false);
940 /* Note : this field could be in headers_start/headers_end section
941 * It is not yet because we do not want to have a 16 bit hole
943 new->queue_mapping = old->queue_mapping;
945 memcpy(&new->headers_start, &old->headers_start,
946 offsetof(struct sk_buff, headers_end) -
947 offsetof(struct sk_buff, headers_start));
948 CHECK_SKB_FIELD(protocol);
949 CHECK_SKB_FIELD(csum);
950 CHECK_SKB_FIELD(hash);
951 CHECK_SKB_FIELD(priority);
952 CHECK_SKB_FIELD(skb_iif);
953 CHECK_SKB_FIELD(vlan_proto);
954 CHECK_SKB_FIELD(vlan_tci);
955 CHECK_SKB_FIELD(transport_header);
956 CHECK_SKB_FIELD(network_header);
957 CHECK_SKB_FIELD(mac_header);
958 CHECK_SKB_FIELD(inner_protocol);
959 CHECK_SKB_FIELD(inner_transport_header);
960 CHECK_SKB_FIELD(inner_network_header);
961 CHECK_SKB_FIELD(inner_mac_header);
962 CHECK_SKB_FIELD(mark);
963 #ifdef CONFIG_NETWORK_SECMARK
964 CHECK_SKB_FIELD(secmark);
965 #endif
966 #ifdef CONFIG_NET_RX_BUSY_POLL
967 CHECK_SKB_FIELD(napi_id);
968 #endif
969 #ifdef CONFIG_XPS
970 CHECK_SKB_FIELD(sender_cpu);
971 #endif
972 #ifdef CONFIG_NET_SCHED
973 CHECK_SKB_FIELD(tc_index);
974 #endif
979 * You should not add any new code to this function. Add it to
980 * __copy_skb_header above instead.
982 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
984 #define C(x) n->x = skb->x
986 n->next = n->prev = NULL;
987 n->sk = NULL;
988 __copy_skb_header(n, skb);
990 C(len);
991 C(data_len);
992 C(mac_len);
993 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
994 n->cloned = 1;
995 n->nohdr = 0;
996 n->peeked = 0;
997 C(pfmemalloc);
998 n->destructor = NULL;
999 C(tail);
1000 C(end);
1001 C(head);
1002 C(head_frag);
1003 C(data);
1004 C(truesize);
1005 refcount_set(&n->users, 1);
1007 atomic_inc(&(skb_shinfo(skb)->dataref));
1008 skb->cloned = 1;
1010 return n;
1011 #undef C
1015 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1016 * @first: first sk_buff of the msg
1018 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1020 struct sk_buff *n;
1022 n = alloc_skb(0, GFP_ATOMIC);
1023 if (!n)
1024 return NULL;
1026 n->len = first->len;
1027 n->data_len = first->len;
1028 n->truesize = first->truesize;
1030 skb_shinfo(n)->frag_list = first;
1032 __copy_skb_header(n, first);
1033 n->destructor = NULL;
1035 return n;
1037 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1040 * skb_morph - morph one skb into another
1041 * @dst: the skb to receive the contents
1042 * @src: the skb to supply the contents
1044 * This is identical to skb_clone except that the target skb is
1045 * supplied by the user.
1047 * The target skb is returned upon exit.
1049 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1051 skb_release_all(dst);
1052 return __skb_clone(dst, src);
1054 EXPORT_SYMBOL_GPL(skb_morph);
1056 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1058 unsigned long max_pg, num_pg, new_pg, old_pg;
1059 struct user_struct *user;
1061 if (capable(CAP_IPC_LOCK) || !size)
1062 return 0;
1064 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1065 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1066 user = mmp->user ? : current_user();
1068 do {
1069 old_pg = atomic_long_read(&user->locked_vm);
1070 new_pg = old_pg + num_pg;
1071 if (new_pg > max_pg)
1072 return -ENOBUFS;
1073 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1074 old_pg);
1076 if (!mmp->user) {
1077 mmp->user = get_uid(user);
1078 mmp->num_pg = num_pg;
1079 } else {
1080 mmp->num_pg += num_pg;
1083 return 0;
1085 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1087 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1089 if (mmp->user) {
1090 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1091 free_uid(mmp->user);
1094 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1096 struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1098 struct ubuf_info *uarg;
1099 struct sk_buff *skb;
1101 WARN_ON_ONCE(!in_task());
1103 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1104 if (!skb)
1105 return NULL;
1107 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1108 uarg = (void *)skb->cb;
1109 uarg->mmp.user = NULL;
1111 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1112 kfree_skb(skb);
1113 return NULL;
1116 uarg->callback = sock_zerocopy_callback;
1117 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1118 uarg->len = 1;
1119 uarg->bytelen = size;
1120 uarg->zerocopy = 1;
1121 refcount_set(&uarg->refcnt, 1);
1122 sock_hold(sk);
1124 return uarg;
1126 EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1128 static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1130 return container_of((void *)uarg, struct sk_buff, cb);
1133 struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1134 struct ubuf_info *uarg)
1136 if (uarg) {
1137 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1138 u32 bytelen, next;
1140 /* realloc only when socket is locked (TCP, UDP cork),
1141 * so uarg->len and sk_zckey access is serialized
1143 if (!sock_owned_by_user(sk)) {
1144 WARN_ON_ONCE(1);
1145 return NULL;
1148 bytelen = uarg->bytelen + size;
1149 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1150 /* TCP can create new skb to attach new uarg */
1151 if (sk->sk_type == SOCK_STREAM)
1152 goto new_alloc;
1153 return NULL;
1156 next = (u32)atomic_read(&sk->sk_zckey);
1157 if ((u32)(uarg->id + uarg->len) == next) {
1158 if (mm_account_pinned_pages(&uarg->mmp, size))
1159 return NULL;
1160 uarg->len++;
1161 uarg->bytelen = bytelen;
1162 atomic_set(&sk->sk_zckey, ++next);
1164 /* no extra ref when appending to datagram (MSG_MORE) */
1165 if (sk->sk_type == SOCK_STREAM)
1166 sock_zerocopy_get(uarg);
1168 return uarg;
1172 new_alloc:
1173 return sock_zerocopy_alloc(sk, size);
1175 EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1177 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1179 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1180 u32 old_lo, old_hi;
1181 u64 sum_len;
1183 old_lo = serr->ee.ee_info;
1184 old_hi = serr->ee.ee_data;
1185 sum_len = old_hi - old_lo + 1ULL + len;
1187 if (sum_len >= (1ULL << 32))
1188 return false;
1190 if (lo != old_hi + 1)
1191 return false;
1193 serr->ee.ee_data += len;
1194 return true;
1197 void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1199 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1200 struct sock_exterr_skb *serr;
1201 struct sock *sk = skb->sk;
1202 struct sk_buff_head *q;
1203 unsigned long flags;
1204 u32 lo, hi;
1205 u16 len;
1207 mm_unaccount_pinned_pages(&uarg->mmp);
1209 /* if !len, there was only 1 call, and it was aborted
1210 * so do not queue a completion notification
1212 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1213 goto release;
1215 len = uarg->len;
1216 lo = uarg->id;
1217 hi = uarg->id + len - 1;
1219 serr = SKB_EXT_ERR(skb);
1220 memset(serr, 0, sizeof(*serr));
1221 serr->ee.ee_errno = 0;
1222 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1223 serr->ee.ee_data = hi;
1224 serr->ee.ee_info = lo;
1225 if (!success)
1226 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1228 q = &sk->sk_error_queue;
1229 spin_lock_irqsave(&q->lock, flags);
1230 tail = skb_peek_tail(q);
1231 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1232 !skb_zerocopy_notify_extend(tail, lo, len)) {
1233 __skb_queue_tail(q, skb);
1234 skb = NULL;
1236 spin_unlock_irqrestore(&q->lock, flags);
1238 sk->sk_error_report(sk);
1240 release:
1241 consume_skb(skb);
1242 sock_put(sk);
1244 EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1246 void sock_zerocopy_put(struct ubuf_info *uarg)
1248 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1249 if (uarg->callback)
1250 uarg->callback(uarg, uarg->zerocopy);
1251 else
1252 consume_skb(skb_from_uarg(uarg));
1255 EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1257 void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1259 if (uarg) {
1260 struct sock *sk = skb_from_uarg(uarg)->sk;
1262 atomic_dec(&sk->sk_zckey);
1263 uarg->len--;
1265 if (have_uref)
1266 sock_zerocopy_put(uarg);
1269 EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1271 int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1273 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1275 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1277 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1278 struct msghdr *msg, int len,
1279 struct ubuf_info *uarg)
1281 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1282 struct iov_iter orig_iter = msg->msg_iter;
1283 int err, orig_len = skb->len;
1285 /* An skb can only point to one uarg. This edge case happens when
1286 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1288 if (orig_uarg && uarg != orig_uarg)
1289 return -EEXIST;
1291 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1292 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1293 struct sock *save_sk = skb->sk;
1295 /* Streams do not free skb on error. Reset to prev state. */
1296 msg->msg_iter = orig_iter;
1297 skb->sk = sk;
1298 ___pskb_trim(skb, orig_len);
1299 skb->sk = save_sk;
1300 return err;
1303 skb_zcopy_set(skb, uarg, NULL);
1304 return skb->len - orig_len;
1306 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1308 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1309 gfp_t gfp_mask)
1311 if (skb_zcopy(orig)) {
1312 if (skb_zcopy(nskb)) {
1313 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1314 if (!gfp_mask) {
1315 WARN_ON_ONCE(1);
1316 return -ENOMEM;
1318 if (skb_uarg(nskb) == skb_uarg(orig))
1319 return 0;
1320 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1321 return -EIO;
1323 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1325 return 0;
1329 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1330 * @skb: the skb to modify
1331 * @gfp_mask: allocation priority
1333 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1334 * It will copy all frags into kernel and drop the reference
1335 * to userspace pages.
1337 * If this function is called from an interrupt gfp_mask() must be
1338 * %GFP_ATOMIC.
1340 * Returns 0 on success or a negative error code on failure
1341 * to allocate kernel memory to copy to.
1343 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1345 int num_frags = skb_shinfo(skb)->nr_frags;
1346 struct page *page, *head = NULL;
1347 int i, new_frags;
1348 u32 d_off;
1350 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1351 return -EINVAL;
1353 if (!num_frags)
1354 goto release;
1356 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1357 for (i = 0; i < new_frags; i++) {
1358 page = alloc_page(gfp_mask);
1359 if (!page) {
1360 while (head) {
1361 struct page *next = (struct page *)page_private(head);
1362 put_page(head);
1363 head = next;
1365 return -ENOMEM;
1367 set_page_private(page, (unsigned long)head);
1368 head = page;
1371 page = head;
1372 d_off = 0;
1373 for (i = 0; i < num_frags; i++) {
1374 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1375 u32 p_off, p_len, copied;
1376 struct page *p;
1377 u8 *vaddr;
1379 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1380 p, p_off, p_len, copied) {
1381 u32 copy, done = 0;
1382 vaddr = kmap_atomic(p);
1384 while (done < p_len) {
1385 if (d_off == PAGE_SIZE) {
1386 d_off = 0;
1387 page = (struct page *)page_private(page);
1389 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1390 memcpy(page_address(page) + d_off,
1391 vaddr + p_off + done, copy);
1392 done += copy;
1393 d_off += copy;
1395 kunmap_atomic(vaddr);
1399 /* skb frags release userspace buffers */
1400 for (i = 0; i < num_frags; i++)
1401 skb_frag_unref(skb, i);
1403 /* skb frags point to kernel buffers */
1404 for (i = 0; i < new_frags - 1; i++) {
1405 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1406 head = (struct page *)page_private(head);
1408 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1409 skb_shinfo(skb)->nr_frags = new_frags;
1411 release:
1412 skb_zcopy_clear(skb, false);
1413 return 0;
1415 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1418 * skb_clone - duplicate an sk_buff
1419 * @skb: buffer to clone
1420 * @gfp_mask: allocation priority
1422 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1423 * copies share the same packet data but not structure. The new
1424 * buffer has a reference count of 1. If the allocation fails the
1425 * function returns %NULL otherwise the new buffer is returned.
1427 * If this function is called from an interrupt gfp_mask() must be
1428 * %GFP_ATOMIC.
1431 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1433 struct sk_buff_fclones *fclones = container_of(skb,
1434 struct sk_buff_fclones,
1435 skb1);
1436 struct sk_buff *n;
1438 if (skb_orphan_frags(skb, gfp_mask))
1439 return NULL;
1441 if (skb->fclone == SKB_FCLONE_ORIG &&
1442 refcount_read(&fclones->fclone_ref) == 1) {
1443 n = &fclones->skb2;
1444 refcount_set(&fclones->fclone_ref, 2);
1445 } else {
1446 if (skb_pfmemalloc(skb))
1447 gfp_mask |= __GFP_MEMALLOC;
1449 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1450 if (!n)
1451 return NULL;
1453 n->fclone = SKB_FCLONE_UNAVAILABLE;
1456 return __skb_clone(n, skb);
1458 EXPORT_SYMBOL(skb_clone);
1460 void skb_headers_offset_update(struct sk_buff *skb, int off)
1462 /* Only adjust this if it actually is csum_start rather than csum */
1463 if (skb->ip_summed == CHECKSUM_PARTIAL)
1464 skb->csum_start += off;
1465 /* {transport,network,mac}_header and tail are relative to skb->head */
1466 skb->transport_header += off;
1467 skb->network_header += off;
1468 if (skb_mac_header_was_set(skb))
1469 skb->mac_header += off;
1470 skb->inner_transport_header += off;
1471 skb->inner_network_header += off;
1472 skb->inner_mac_header += off;
1474 EXPORT_SYMBOL(skb_headers_offset_update);
1476 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1478 __copy_skb_header(new, old);
1480 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1481 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1482 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1484 EXPORT_SYMBOL(skb_copy_header);
1486 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1488 if (skb_pfmemalloc(skb))
1489 return SKB_ALLOC_RX;
1490 return 0;
1494 * skb_copy - create private copy of an sk_buff
1495 * @skb: buffer to copy
1496 * @gfp_mask: allocation priority
1498 * Make a copy of both an &sk_buff and its data. This is used when the
1499 * caller wishes to modify the data and needs a private copy of the
1500 * data to alter. Returns %NULL on failure or the pointer to the buffer
1501 * on success. The returned buffer has a reference count of 1.
1503 * As by-product this function converts non-linear &sk_buff to linear
1504 * one, so that &sk_buff becomes completely private and caller is allowed
1505 * to modify all the data of returned buffer. This means that this
1506 * function is not recommended for use in circumstances when only
1507 * header is going to be modified. Use pskb_copy() instead.
1510 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1512 int headerlen = skb_headroom(skb);
1513 unsigned int size = skb_end_offset(skb) + skb->data_len;
1514 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1515 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1517 if (!n)
1518 return NULL;
1520 /* Set the data pointer */
1521 skb_reserve(n, headerlen);
1522 /* Set the tail pointer and length */
1523 skb_put(n, skb->len);
1525 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1527 skb_copy_header(n, skb);
1528 return n;
1530 EXPORT_SYMBOL(skb_copy);
1533 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1534 * @skb: buffer to copy
1535 * @headroom: headroom of new skb
1536 * @gfp_mask: allocation priority
1537 * @fclone: if true allocate the copy of the skb from the fclone
1538 * cache instead of the head cache; it is recommended to set this
1539 * to true for the cases where the copy will likely be cloned
1541 * Make a copy of both an &sk_buff and part of its data, located
1542 * in header. Fragmented data remain shared. This is used when
1543 * the caller wishes to modify only header of &sk_buff and needs
1544 * private copy of the header to alter. Returns %NULL on failure
1545 * or the pointer to the buffer on success.
1546 * The returned buffer has a reference count of 1.
1549 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1550 gfp_t gfp_mask, bool fclone)
1552 unsigned int size = skb_headlen(skb) + headroom;
1553 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1554 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1556 if (!n)
1557 goto out;
1559 /* Set the data pointer */
1560 skb_reserve(n, headroom);
1561 /* Set the tail pointer and length */
1562 skb_put(n, skb_headlen(skb));
1563 /* Copy the bytes */
1564 skb_copy_from_linear_data(skb, n->data, n->len);
1566 n->truesize += skb->data_len;
1567 n->data_len = skb->data_len;
1568 n->len = skb->len;
1570 if (skb_shinfo(skb)->nr_frags) {
1571 int i;
1573 if (skb_orphan_frags(skb, gfp_mask) ||
1574 skb_zerocopy_clone(n, skb, gfp_mask)) {
1575 kfree_skb(n);
1576 n = NULL;
1577 goto out;
1579 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1580 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1581 skb_frag_ref(skb, i);
1583 skb_shinfo(n)->nr_frags = i;
1586 if (skb_has_frag_list(skb)) {
1587 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1588 skb_clone_fraglist(n);
1591 skb_copy_header(n, skb);
1592 out:
1593 return n;
1595 EXPORT_SYMBOL(__pskb_copy_fclone);
1598 * pskb_expand_head - reallocate header of &sk_buff
1599 * @skb: buffer to reallocate
1600 * @nhead: room to add at head
1601 * @ntail: room to add at tail
1602 * @gfp_mask: allocation priority
1604 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1605 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1606 * reference count of 1. Returns zero in the case of success or error,
1607 * if expansion failed. In the last case, &sk_buff is not changed.
1609 * All the pointers pointing into skb header may change and must be
1610 * reloaded after call to this function.
1613 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1614 gfp_t gfp_mask)
1616 int i, osize = skb_end_offset(skb);
1617 int size = osize + nhead + ntail;
1618 long off;
1619 u8 *data;
1621 BUG_ON(nhead < 0);
1623 BUG_ON(skb_shared(skb));
1625 size = SKB_DATA_ALIGN(size);
1627 if (skb_pfmemalloc(skb))
1628 gfp_mask |= __GFP_MEMALLOC;
1629 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1630 gfp_mask, NUMA_NO_NODE, NULL);
1631 if (!data)
1632 goto nodata;
1633 size = SKB_WITH_OVERHEAD(ksize(data));
1635 /* Copy only real data... and, alas, header. This should be
1636 * optimized for the cases when header is void.
1638 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1640 memcpy((struct skb_shared_info *)(data + size),
1641 skb_shinfo(skb),
1642 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1645 * if shinfo is shared we must drop the old head gracefully, but if it
1646 * is not we can just drop the old head and let the existing refcount
1647 * be since all we did is relocate the values
1649 if (skb_cloned(skb)) {
1650 if (skb_orphan_frags(skb, gfp_mask))
1651 goto nofrags;
1652 if (skb_zcopy(skb))
1653 refcount_inc(&skb_uarg(skb)->refcnt);
1654 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1655 skb_frag_ref(skb, i);
1657 if (skb_has_frag_list(skb))
1658 skb_clone_fraglist(skb);
1660 skb_release_data(skb);
1661 } else {
1662 skb_free_head(skb);
1664 off = (data + nhead) - skb->head;
1666 skb->head = data;
1667 skb->head_frag = 0;
1668 skb->data += off;
1669 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1670 skb->end = size;
1671 off = nhead;
1672 #else
1673 skb->end = skb->head + size;
1674 #endif
1675 skb->tail += off;
1676 skb_headers_offset_update(skb, nhead);
1677 skb->cloned = 0;
1678 skb->hdr_len = 0;
1679 skb->nohdr = 0;
1680 atomic_set(&skb_shinfo(skb)->dataref, 1);
1682 skb_metadata_clear(skb);
1684 /* It is not generally safe to change skb->truesize.
1685 * For the moment, we really care of rx path, or
1686 * when skb is orphaned (not attached to a socket).
1688 if (!skb->sk || skb->destructor == sock_edemux)
1689 skb->truesize += size - osize;
1691 return 0;
1693 nofrags:
1694 kfree(data);
1695 nodata:
1696 return -ENOMEM;
1698 EXPORT_SYMBOL(pskb_expand_head);
1700 /* Make private copy of skb with writable head and some headroom */
1702 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1704 struct sk_buff *skb2;
1705 int delta = headroom - skb_headroom(skb);
1707 if (delta <= 0)
1708 skb2 = pskb_copy(skb, GFP_ATOMIC);
1709 else {
1710 skb2 = skb_clone(skb, GFP_ATOMIC);
1711 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1712 GFP_ATOMIC)) {
1713 kfree_skb(skb2);
1714 skb2 = NULL;
1717 return skb2;
1719 EXPORT_SYMBOL(skb_realloc_headroom);
1722 * skb_copy_expand - copy and expand sk_buff
1723 * @skb: buffer to copy
1724 * @newheadroom: new free bytes at head
1725 * @newtailroom: new free bytes at tail
1726 * @gfp_mask: allocation priority
1728 * Make a copy of both an &sk_buff and its data and while doing so
1729 * allocate additional space.
1731 * This is used when the caller wishes to modify the data and needs a
1732 * private copy of the data to alter as well as more space for new fields.
1733 * Returns %NULL on failure or the pointer to the buffer
1734 * on success. The returned buffer has a reference count of 1.
1736 * You must pass %GFP_ATOMIC as the allocation priority if this function
1737 * is called from an interrupt.
1739 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1740 int newheadroom, int newtailroom,
1741 gfp_t gfp_mask)
1744 * Allocate the copy buffer
1746 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1747 gfp_mask, skb_alloc_rx_flag(skb),
1748 NUMA_NO_NODE);
1749 int oldheadroom = skb_headroom(skb);
1750 int head_copy_len, head_copy_off;
1752 if (!n)
1753 return NULL;
1755 skb_reserve(n, newheadroom);
1757 /* Set the tail pointer and length */
1758 skb_put(n, skb->len);
1760 head_copy_len = oldheadroom;
1761 head_copy_off = 0;
1762 if (newheadroom <= head_copy_len)
1763 head_copy_len = newheadroom;
1764 else
1765 head_copy_off = newheadroom - head_copy_len;
1767 /* Copy the linear header and data. */
1768 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1769 skb->len + head_copy_len));
1771 skb_copy_header(n, skb);
1773 skb_headers_offset_update(n, newheadroom - oldheadroom);
1775 return n;
1777 EXPORT_SYMBOL(skb_copy_expand);
1780 * __skb_pad - zero pad the tail of an skb
1781 * @skb: buffer to pad
1782 * @pad: space to pad
1783 * @free_on_error: free buffer on error
1785 * Ensure that a buffer is followed by a padding area that is zero
1786 * filled. Used by network drivers which may DMA or transfer data
1787 * beyond the buffer end onto the wire.
1789 * May return error in out of memory cases. The skb is freed on error
1790 * if @free_on_error is true.
1793 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1795 int err;
1796 int ntail;
1798 /* If the skbuff is non linear tailroom is always zero.. */
1799 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1800 memset(skb->data+skb->len, 0, pad);
1801 return 0;
1804 ntail = skb->data_len + pad - (skb->end - skb->tail);
1805 if (likely(skb_cloned(skb) || ntail > 0)) {
1806 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1807 if (unlikely(err))
1808 goto free_skb;
1811 /* FIXME: The use of this function with non-linear skb's really needs
1812 * to be audited.
1814 err = skb_linearize(skb);
1815 if (unlikely(err))
1816 goto free_skb;
1818 memset(skb->data + skb->len, 0, pad);
1819 return 0;
1821 free_skb:
1822 if (free_on_error)
1823 kfree_skb(skb);
1824 return err;
1826 EXPORT_SYMBOL(__skb_pad);
1829 * pskb_put - add data to the tail of a potentially fragmented buffer
1830 * @skb: start of the buffer to use
1831 * @tail: tail fragment of the buffer to use
1832 * @len: amount of data to add
1834 * This function extends the used data area of the potentially
1835 * fragmented buffer. @tail must be the last fragment of @skb -- or
1836 * @skb itself. If this would exceed the total buffer size the kernel
1837 * will panic. A pointer to the first byte of the extra data is
1838 * returned.
1841 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1843 if (tail != skb) {
1844 skb->data_len += len;
1845 skb->len += len;
1847 return skb_put(tail, len);
1849 EXPORT_SYMBOL_GPL(pskb_put);
1852 * skb_put - add data to a buffer
1853 * @skb: buffer to use
1854 * @len: amount of data to add
1856 * This function extends the used data area of the buffer. If this would
1857 * exceed the total buffer size the kernel will panic. A pointer to the
1858 * first byte of the extra data is returned.
1860 void *skb_put(struct sk_buff *skb, unsigned int len)
1862 void *tmp = skb_tail_pointer(skb);
1863 SKB_LINEAR_ASSERT(skb);
1864 skb->tail += len;
1865 skb->len += len;
1866 if (unlikely(skb->tail > skb->end))
1867 skb_over_panic(skb, len, __builtin_return_address(0));
1868 return tmp;
1870 EXPORT_SYMBOL(skb_put);
1873 * skb_push - add data to the start of a buffer
1874 * @skb: buffer to use
1875 * @len: amount of data to add
1877 * This function extends the used data area of the buffer at the buffer
1878 * start. If this would exceed the total buffer headroom the kernel will
1879 * panic. A pointer to the first byte of the extra data is returned.
1881 void *skb_push(struct sk_buff *skb, unsigned int len)
1883 skb->data -= len;
1884 skb->len += len;
1885 if (unlikely(skb->data < skb->head))
1886 skb_under_panic(skb, len, __builtin_return_address(0));
1887 return skb->data;
1889 EXPORT_SYMBOL(skb_push);
1892 * skb_pull - remove data from the start of a buffer
1893 * @skb: buffer to use
1894 * @len: amount of data to remove
1896 * This function removes data from the start of a buffer, returning
1897 * the memory to the headroom. A pointer to the next data in the buffer
1898 * is returned. Once the data has been pulled future pushes will overwrite
1899 * the old data.
1901 void *skb_pull(struct sk_buff *skb, unsigned int len)
1903 return skb_pull_inline(skb, len);
1905 EXPORT_SYMBOL(skb_pull);
1908 * skb_trim - remove end from a buffer
1909 * @skb: buffer to alter
1910 * @len: new length
1912 * Cut the length of a buffer down by removing data from the tail. If
1913 * the buffer is already under the length specified it is not modified.
1914 * The skb must be linear.
1916 void skb_trim(struct sk_buff *skb, unsigned int len)
1918 if (skb->len > len)
1919 __skb_trim(skb, len);
1921 EXPORT_SYMBOL(skb_trim);
1923 /* Trims skb to length len. It can change skb pointers.
1926 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1928 struct sk_buff **fragp;
1929 struct sk_buff *frag;
1930 int offset = skb_headlen(skb);
1931 int nfrags = skb_shinfo(skb)->nr_frags;
1932 int i;
1933 int err;
1935 if (skb_cloned(skb) &&
1936 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1937 return err;
1939 i = 0;
1940 if (offset >= len)
1941 goto drop_pages;
1943 for (; i < nfrags; i++) {
1944 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1946 if (end < len) {
1947 offset = end;
1948 continue;
1951 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1953 drop_pages:
1954 skb_shinfo(skb)->nr_frags = i;
1956 for (; i < nfrags; i++)
1957 skb_frag_unref(skb, i);
1959 if (skb_has_frag_list(skb))
1960 skb_drop_fraglist(skb);
1961 goto done;
1964 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1965 fragp = &frag->next) {
1966 int end = offset + frag->len;
1968 if (skb_shared(frag)) {
1969 struct sk_buff *nfrag;
1971 nfrag = skb_clone(frag, GFP_ATOMIC);
1972 if (unlikely(!nfrag))
1973 return -ENOMEM;
1975 nfrag->next = frag->next;
1976 consume_skb(frag);
1977 frag = nfrag;
1978 *fragp = frag;
1981 if (end < len) {
1982 offset = end;
1983 continue;
1986 if (end > len &&
1987 unlikely((err = pskb_trim(frag, len - offset))))
1988 return err;
1990 if (frag->next)
1991 skb_drop_list(&frag->next);
1992 break;
1995 done:
1996 if (len > skb_headlen(skb)) {
1997 skb->data_len -= skb->len - len;
1998 skb->len = len;
1999 } else {
2000 skb->len = len;
2001 skb->data_len = 0;
2002 skb_set_tail_pointer(skb, len);
2005 if (!skb->sk || skb->destructor == sock_edemux)
2006 skb_condense(skb);
2007 return 0;
2009 EXPORT_SYMBOL(___pskb_trim);
2011 /* Note : use pskb_trim_rcsum() instead of calling this directly
2013 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2015 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2016 int delta = skb->len - len;
2018 skb->csum = csum_block_sub(skb->csum,
2019 skb_checksum(skb, len, delta, 0),
2020 len);
2021 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2022 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2023 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2025 if (offset + sizeof(__sum16) > hdlen)
2026 return -EINVAL;
2028 return __pskb_trim(skb, len);
2030 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2033 * __pskb_pull_tail - advance tail of skb header
2034 * @skb: buffer to reallocate
2035 * @delta: number of bytes to advance tail
2037 * The function makes a sense only on a fragmented &sk_buff,
2038 * it expands header moving its tail forward and copying necessary
2039 * data from fragmented part.
2041 * &sk_buff MUST have reference count of 1.
2043 * Returns %NULL (and &sk_buff does not change) if pull failed
2044 * or value of new tail of skb in the case of success.
2046 * All the pointers pointing into skb header may change and must be
2047 * reloaded after call to this function.
2050 /* Moves tail of skb head forward, copying data from fragmented part,
2051 * when it is necessary.
2052 * 1. It may fail due to malloc failure.
2053 * 2. It may change skb pointers.
2055 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2057 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2059 /* If skb has not enough free space at tail, get new one
2060 * plus 128 bytes for future expansions. If we have enough
2061 * room at tail, reallocate without expansion only if skb is cloned.
2063 int i, k, eat = (skb->tail + delta) - skb->end;
2065 if (eat > 0 || skb_cloned(skb)) {
2066 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2067 GFP_ATOMIC))
2068 return NULL;
2071 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2072 skb_tail_pointer(skb), delta));
2074 /* Optimization: no fragments, no reasons to preestimate
2075 * size of pulled pages. Superb.
2077 if (!skb_has_frag_list(skb))
2078 goto pull_pages;
2080 /* Estimate size of pulled pages. */
2081 eat = delta;
2082 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2083 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2085 if (size >= eat)
2086 goto pull_pages;
2087 eat -= size;
2090 /* If we need update frag list, we are in troubles.
2091 * Certainly, it is possible to add an offset to skb data,
2092 * but taking into account that pulling is expected to
2093 * be very rare operation, it is worth to fight against
2094 * further bloating skb head and crucify ourselves here instead.
2095 * Pure masohism, indeed. 8)8)
2097 if (eat) {
2098 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2099 struct sk_buff *clone = NULL;
2100 struct sk_buff *insp = NULL;
2102 do {
2103 if (list->len <= eat) {
2104 /* Eaten as whole. */
2105 eat -= list->len;
2106 list = list->next;
2107 insp = list;
2108 } else {
2109 /* Eaten partially. */
2111 if (skb_shared(list)) {
2112 /* Sucks! We need to fork list. :-( */
2113 clone = skb_clone(list, GFP_ATOMIC);
2114 if (!clone)
2115 return NULL;
2116 insp = list->next;
2117 list = clone;
2118 } else {
2119 /* This may be pulled without
2120 * problems. */
2121 insp = list;
2123 if (!pskb_pull(list, eat)) {
2124 kfree_skb(clone);
2125 return NULL;
2127 break;
2129 } while (eat);
2131 /* Free pulled out fragments. */
2132 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2133 skb_shinfo(skb)->frag_list = list->next;
2134 kfree_skb(list);
2136 /* And insert new clone at head. */
2137 if (clone) {
2138 clone->next = list;
2139 skb_shinfo(skb)->frag_list = clone;
2142 /* Success! Now we may commit changes to skb data. */
2144 pull_pages:
2145 eat = delta;
2146 k = 0;
2147 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2148 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2150 if (size <= eat) {
2151 skb_frag_unref(skb, i);
2152 eat -= size;
2153 } else {
2154 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2156 *frag = skb_shinfo(skb)->frags[i];
2157 if (eat) {
2158 skb_frag_off_add(frag, eat);
2159 skb_frag_size_sub(frag, eat);
2160 if (!i)
2161 goto end;
2162 eat = 0;
2164 k++;
2167 skb_shinfo(skb)->nr_frags = k;
2169 end:
2170 skb->tail += delta;
2171 skb->data_len -= delta;
2173 if (!skb->data_len)
2174 skb_zcopy_clear(skb, false);
2176 return skb_tail_pointer(skb);
2178 EXPORT_SYMBOL(__pskb_pull_tail);
2181 * skb_copy_bits - copy bits from skb to kernel buffer
2182 * @skb: source skb
2183 * @offset: offset in source
2184 * @to: destination buffer
2185 * @len: number of bytes to copy
2187 * Copy the specified number of bytes from the source skb to the
2188 * destination buffer.
2190 * CAUTION ! :
2191 * If its prototype is ever changed,
2192 * check arch/{*}/net/{*}.S files,
2193 * since it is called from BPF assembly code.
2195 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2197 int start = skb_headlen(skb);
2198 struct sk_buff *frag_iter;
2199 int i, copy;
2201 if (offset > (int)skb->len - len)
2202 goto fault;
2204 /* Copy header. */
2205 if ((copy = start - offset) > 0) {
2206 if (copy > len)
2207 copy = len;
2208 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2209 if ((len -= copy) == 0)
2210 return 0;
2211 offset += copy;
2212 to += copy;
2215 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2216 int end;
2217 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2219 WARN_ON(start > offset + len);
2221 end = start + skb_frag_size(f);
2222 if ((copy = end - offset) > 0) {
2223 u32 p_off, p_len, copied;
2224 struct page *p;
2225 u8 *vaddr;
2227 if (copy > len)
2228 copy = len;
2230 skb_frag_foreach_page(f,
2231 skb_frag_off(f) + offset - start,
2232 copy, p, p_off, p_len, copied) {
2233 vaddr = kmap_atomic(p);
2234 memcpy(to + copied, vaddr + p_off, p_len);
2235 kunmap_atomic(vaddr);
2238 if ((len -= copy) == 0)
2239 return 0;
2240 offset += copy;
2241 to += copy;
2243 start = end;
2246 skb_walk_frags(skb, frag_iter) {
2247 int end;
2249 WARN_ON(start > offset + len);
2251 end = start + frag_iter->len;
2252 if ((copy = end - offset) > 0) {
2253 if (copy > len)
2254 copy = len;
2255 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2256 goto fault;
2257 if ((len -= copy) == 0)
2258 return 0;
2259 offset += copy;
2260 to += copy;
2262 start = end;
2265 if (!len)
2266 return 0;
2268 fault:
2269 return -EFAULT;
2271 EXPORT_SYMBOL(skb_copy_bits);
2274 * Callback from splice_to_pipe(), if we need to release some pages
2275 * at the end of the spd in case we error'ed out in filling the pipe.
2277 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2279 put_page(spd->pages[i]);
2282 static struct page *linear_to_page(struct page *page, unsigned int *len,
2283 unsigned int *offset,
2284 struct sock *sk)
2286 struct page_frag *pfrag = sk_page_frag(sk);
2288 if (!sk_page_frag_refill(sk, pfrag))
2289 return NULL;
2291 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2293 memcpy(page_address(pfrag->page) + pfrag->offset,
2294 page_address(page) + *offset, *len);
2295 *offset = pfrag->offset;
2296 pfrag->offset += *len;
2298 return pfrag->page;
2301 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2302 struct page *page,
2303 unsigned int offset)
2305 return spd->nr_pages &&
2306 spd->pages[spd->nr_pages - 1] == page &&
2307 (spd->partial[spd->nr_pages - 1].offset +
2308 spd->partial[spd->nr_pages - 1].len == offset);
2312 * Fill page/offset/length into spd, if it can hold more pages.
2314 static bool spd_fill_page(struct splice_pipe_desc *spd,
2315 struct pipe_inode_info *pipe, struct page *page,
2316 unsigned int *len, unsigned int offset,
2317 bool linear,
2318 struct sock *sk)
2320 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2321 return true;
2323 if (linear) {
2324 page = linear_to_page(page, len, &offset, sk);
2325 if (!page)
2326 return true;
2328 if (spd_can_coalesce(spd, page, offset)) {
2329 spd->partial[spd->nr_pages - 1].len += *len;
2330 return false;
2332 get_page(page);
2333 spd->pages[spd->nr_pages] = page;
2334 spd->partial[spd->nr_pages].len = *len;
2335 spd->partial[spd->nr_pages].offset = offset;
2336 spd->nr_pages++;
2338 return false;
2341 static bool __splice_segment(struct page *page, unsigned int poff,
2342 unsigned int plen, unsigned int *off,
2343 unsigned int *len,
2344 struct splice_pipe_desc *spd, bool linear,
2345 struct sock *sk,
2346 struct pipe_inode_info *pipe)
2348 if (!*len)
2349 return true;
2351 /* skip this segment if already processed */
2352 if (*off >= plen) {
2353 *off -= plen;
2354 return false;
2357 /* ignore any bits we already processed */
2358 poff += *off;
2359 plen -= *off;
2360 *off = 0;
2362 do {
2363 unsigned int flen = min(*len, plen);
2365 if (spd_fill_page(spd, pipe, page, &flen, poff,
2366 linear, sk))
2367 return true;
2368 poff += flen;
2369 plen -= flen;
2370 *len -= flen;
2371 } while (*len && plen);
2373 return false;
2377 * Map linear and fragment data from the skb to spd. It reports true if the
2378 * pipe is full or if we already spliced the requested length.
2380 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2381 unsigned int *offset, unsigned int *len,
2382 struct splice_pipe_desc *spd, struct sock *sk)
2384 int seg;
2385 struct sk_buff *iter;
2387 /* map the linear part :
2388 * If skb->head_frag is set, this 'linear' part is backed by a
2389 * fragment, and if the head is not shared with any clones then
2390 * we can avoid a copy since we own the head portion of this page.
2392 if (__splice_segment(virt_to_page(skb->data),
2393 (unsigned long) skb->data & (PAGE_SIZE - 1),
2394 skb_headlen(skb),
2395 offset, len, spd,
2396 skb_head_is_locked(skb),
2397 sk, pipe))
2398 return true;
2401 * then map the fragments
2403 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2404 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2406 if (__splice_segment(skb_frag_page(f),
2407 skb_frag_off(f), skb_frag_size(f),
2408 offset, len, spd, false, sk, pipe))
2409 return true;
2412 skb_walk_frags(skb, iter) {
2413 if (*offset >= iter->len) {
2414 *offset -= iter->len;
2415 continue;
2417 /* __skb_splice_bits() only fails if the output has no room
2418 * left, so no point in going over the frag_list for the error
2419 * case.
2421 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2422 return true;
2425 return false;
2429 * Map data from the skb to a pipe. Should handle both the linear part,
2430 * the fragments, and the frag list.
2432 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2433 struct pipe_inode_info *pipe, unsigned int tlen,
2434 unsigned int flags)
2436 struct partial_page partial[MAX_SKB_FRAGS];
2437 struct page *pages[MAX_SKB_FRAGS];
2438 struct splice_pipe_desc spd = {
2439 .pages = pages,
2440 .partial = partial,
2441 .nr_pages_max = MAX_SKB_FRAGS,
2442 .ops = &nosteal_pipe_buf_ops,
2443 .spd_release = sock_spd_release,
2445 int ret = 0;
2447 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2449 if (spd.nr_pages)
2450 ret = splice_to_pipe(pipe, &spd);
2452 return ret;
2454 EXPORT_SYMBOL_GPL(skb_splice_bits);
2456 /* Send skb data on a socket. Socket must be locked. */
2457 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2458 int len)
2460 unsigned int orig_len = len;
2461 struct sk_buff *head = skb;
2462 unsigned short fragidx;
2463 int slen, ret;
2465 do_frag_list:
2467 /* Deal with head data */
2468 while (offset < skb_headlen(skb) && len) {
2469 struct kvec kv;
2470 struct msghdr msg;
2472 slen = min_t(int, len, skb_headlen(skb) - offset);
2473 kv.iov_base = skb->data + offset;
2474 kv.iov_len = slen;
2475 memset(&msg, 0, sizeof(msg));
2476 msg.msg_flags = MSG_DONTWAIT;
2478 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2479 if (ret <= 0)
2480 goto error;
2482 offset += ret;
2483 len -= ret;
2486 /* All the data was skb head? */
2487 if (!len)
2488 goto out;
2490 /* Make offset relative to start of frags */
2491 offset -= skb_headlen(skb);
2493 /* Find where we are in frag list */
2494 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2495 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2497 if (offset < skb_frag_size(frag))
2498 break;
2500 offset -= skb_frag_size(frag);
2503 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2504 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2506 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2508 while (slen) {
2509 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2510 skb_frag_off(frag) + offset,
2511 slen, MSG_DONTWAIT);
2512 if (ret <= 0)
2513 goto error;
2515 len -= ret;
2516 offset += ret;
2517 slen -= ret;
2520 offset = 0;
2523 if (len) {
2524 /* Process any frag lists */
2526 if (skb == head) {
2527 if (skb_has_frag_list(skb)) {
2528 skb = skb_shinfo(skb)->frag_list;
2529 goto do_frag_list;
2531 } else if (skb->next) {
2532 skb = skb->next;
2533 goto do_frag_list;
2537 out:
2538 return orig_len - len;
2540 error:
2541 return orig_len == len ? ret : orig_len - len;
2543 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2546 * skb_store_bits - store bits from kernel buffer to skb
2547 * @skb: destination buffer
2548 * @offset: offset in destination
2549 * @from: source buffer
2550 * @len: number of bytes to copy
2552 * Copy the specified number of bytes from the source buffer to the
2553 * destination skb. This function handles all the messy bits of
2554 * traversing fragment lists and such.
2557 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2559 int start = skb_headlen(skb);
2560 struct sk_buff *frag_iter;
2561 int i, copy;
2563 if (offset > (int)skb->len - len)
2564 goto fault;
2566 if ((copy = start - offset) > 0) {
2567 if (copy > len)
2568 copy = len;
2569 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2570 if ((len -= copy) == 0)
2571 return 0;
2572 offset += copy;
2573 from += copy;
2576 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2577 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2578 int end;
2580 WARN_ON(start > offset + len);
2582 end = start + skb_frag_size(frag);
2583 if ((copy = end - offset) > 0) {
2584 u32 p_off, p_len, copied;
2585 struct page *p;
2586 u8 *vaddr;
2588 if (copy > len)
2589 copy = len;
2591 skb_frag_foreach_page(frag,
2592 skb_frag_off(frag) + offset - start,
2593 copy, p, p_off, p_len, copied) {
2594 vaddr = kmap_atomic(p);
2595 memcpy(vaddr + p_off, from + copied, p_len);
2596 kunmap_atomic(vaddr);
2599 if ((len -= copy) == 0)
2600 return 0;
2601 offset += copy;
2602 from += copy;
2604 start = end;
2607 skb_walk_frags(skb, frag_iter) {
2608 int end;
2610 WARN_ON(start > offset + len);
2612 end = start + frag_iter->len;
2613 if ((copy = end - offset) > 0) {
2614 if (copy > len)
2615 copy = len;
2616 if (skb_store_bits(frag_iter, offset - start,
2617 from, copy))
2618 goto fault;
2619 if ((len -= copy) == 0)
2620 return 0;
2621 offset += copy;
2622 from += copy;
2624 start = end;
2626 if (!len)
2627 return 0;
2629 fault:
2630 return -EFAULT;
2632 EXPORT_SYMBOL(skb_store_bits);
2634 /* Checksum skb data. */
2635 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2636 __wsum csum, const struct skb_checksum_ops *ops)
2638 int start = skb_headlen(skb);
2639 int i, copy = start - offset;
2640 struct sk_buff *frag_iter;
2641 int pos = 0;
2643 /* Checksum header. */
2644 if (copy > 0) {
2645 if (copy > len)
2646 copy = len;
2647 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2648 skb->data + offset, copy, csum);
2649 if ((len -= copy) == 0)
2650 return csum;
2651 offset += copy;
2652 pos = copy;
2655 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2656 int end;
2657 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2659 WARN_ON(start > offset + len);
2661 end = start + skb_frag_size(frag);
2662 if ((copy = end - offset) > 0) {
2663 u32 p_off, p_len, copied;
2664 struct page *p;
2665 __wsum csum2;
2666 u8 *vaddr;
2668 if (copy > len)
2669 copy = len;
2671 skb_frag_foreach_page(frag,
2672 skb_frag_off(frag) + offset - start,
2673 copy, p, p_off, p_len, copied) {
2674 vaddr = kmap_atomic(p);
2675 csum2 = INDIRECT_CALL_1(ops->update,
2676 csum_partial_ext,
2677 vaddr + p_off, p_len, 0);
2678 kunmap_atomic(vaddr);
2679 csum = INDIRECT_CALL_1(ops->combine,
2680 csum_block_add_ext, csum,
2681 csum2, pos, p_len);
2682 pos += p_len;
2685 if (!(len -= copy))
2686 return csum;
2687 offset += copy;
2689 start = end;
2692 skb_walk_frags(skb, frag_iter) {
2693 int end;
2695 WARN_ON(start > offset + len);
2697 end = start + frag_iter->len;
2698 if ((copy = end - offset) > 0) {
2699 __wsum csum2;
2700 if (copy > len)
2701 copy = len;
2702 csum2 = __skb_checksum(frag_iter, offset - start,
2703 copy, 0, ops);
2704 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2705 csum, csum2, pos, copy);
2706 if ((len -= copy) == 0)
2707 return csum;
2708 offset += copy;
2709 pos += copy;
2711 start = end;
2713 BUG_ON(len);
2715 return csum;
2717 EXPORT_SYMBOL(__skb_checksum);
2719 __wsum skb_checksum(const struct sk_buff *skb, int offset,
2720 int len, __wsum csum)
2722 const struct skb_checksum_ops ops = {
2723 .update = csum_partial_ext,
2724 .combine = csum_block_add_ext,
2727 return __skb_checksum(skb, offset, len, csum, &ops);
2729 EXPORT_SYMBOL(skb_checksum);
2731 /* Both of above in one bottle. */
2733 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2734 u8 *to, int len)
2736 int start = skb_headlen(skb);
2737 int i, copy = start - offset;
2738 struct sk_buff *frag_iter;
2739 int pos = 0;
2740 __wsum csum = 0;
2742 /* Copy header. */
2743 if (copy > 0) {
2744 if (copy > len)
2745 copy = len;
2746 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2747 copy);
2748 if ((len -= copy) == 0)
2749 return csum;
2750 offset += copy;
2751 to += copy;
2752 pos = copy;
2755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2756 int end;
2758 WARN_ON(start > offset + len);
2760 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2761 if ((copy = end - offset) > 0) {
2762 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2763 u32 p_off, p_len, copied;
2764 struct page *p;
2765 __wsum csum2;
2766 u8 *vaddr;
2768 if (copy > len)
2769 copy = len;
2771 skb_frag_foreach_page(frag,
2772 skb_frag_off(frag) + offset - start,
2773 copy, p, p_off, p_len, copied) {
2774 vaddr = kmap_atomic(p);
2775 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2776 to + copied,
2777 p_len);
2778 kunmap_atomic(vaddr);
2779 csum = csum_block_add(csum, csum2, pos);
2780 pos += p_len;
2783 if (!(len -= copy))
2784 return csum;
2785 offset += copy;
2786 to += copy;
2788 start = end;
2791 skb_walk_frags(skb, frag_iter) {
2792 __wsum csum2;
2793 int end;
2795 WARN_ON(start > offset + len);
2797 end = start + frag_iter->len;
2798 if ((copy = end - offset) > 0) {
2799 if (copy > len)
2800 copy = len;
2801 csum2 = skb_copy_and_csum_bits(frag_iter,
2802 offset - start,
2803 to, copy);
2804 csum = csum_block_add(csum, csum2, pos);
2805 if ((len -= copy) == 0)
2806 return csum;
2807 offset += copy;
2808 to += copy;
2809 pos += copy;
2811 start = end;
2813 BUG_ON(len);
2814 return csum;
2816 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2818 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2820 __sum16 sum;
2822 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2823 /* See comments in __skb_checksum_complete(). */
2824 if (likely(!sum)) {
2825 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2826 !skb->csum_complete_sw)
2827 netdev_rx_csum_fault(skb->dev, skb);
2829 if (!skb_shared(skb))
2830 skb->csum_valid = !sum;
2831 return sum;
2833 EXPORT_SYMBOL(__skb_checksum_complete_head);
2835 /* This function assumes skb->csum already holds pseudo header's checksum,
2836 * which has been changed from the hardware checksum, for example, by
2837 * __skb_checksum_validate_complete(). And, the original skb->csum must
2838 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2840 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2841 * zero. The new checksum is stored back into skb->csum unless the skb is
2842 * shared.
2844 __sum16 __skb_checksum_complete(struct sk_buff *skb)
2846 __wsum csum;
2847 __sum16 sum;
2849 csum = skb_checksum(skb, 0, skb->len, 0);
2851 sum = csum_fold(csum_add(skb->csum, csum));
2852 /* This check is inverted, because we already knew the hardware
2853 * checksum is invalid before calling this function. So, if the
2854 * re-computed checksum is valid instead, then we have a mismatch
2855 * between the original skb->csum and skb_checksum(). This means either
2856 * the original hardware checksum is incorrect or we screw up skb->csum
2857 * when moving skb->data around.
2859 if (likely(!sum)) {
2860 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2861 !skb->csum_complete_sw)
2862 netdev_rx_csum_fault(skb->dev, skb);
2865 if (!skb_shared(skb)) {
2866 /* Save full packet checksum */
2867 skb->csum = csum;
2868 skb->ip_summed = CHECKSUM_COMPLETE;
2869 skb->csum_complete_sw = 1;
2870 skb->csum_valid = !sum;
2873 return sum;
2875 EXPORT_SYMBOL(__skb_checksum_complete);
2877 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2879 net_warn_ratelimited(
2880 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2881 __func__);
2882 return 0;
2885 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2886 int offset, int len)
2888 net_warn_ratelimited(
2889 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2890 __func__);
2891 return 0;
2894 static const struct skb_checksum_ops default_crc32c_ops = {
2895 .update = warn_crc32c_csum_update,
2896 .combine = warn_crc32c_csum_combine,
2899 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2900 &default_crc32c_ops;
2901 EXPORT_SYMBOL(crc32c_csum_stub);
2904 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2905 * @from: source buffer
2907 * Calculates the amount of linear headroom needed in the 'to' skb passed
2908 * into skb_zerocopy().
2910 unsigned int
2911 skb_zerocopy_headlen(const struct sk_buff *from)
2913 unsigned int hlen = 0;
2915 if (!from->head_frag ||
2916 skb_headlen(from) < L1_CACHE_BYTES ||
2917 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2918 hlen = skb_headlen(from);
2920 if (skb_has_frag_list(from))
2921 hlen = from->len;
2923 return hlen;
2925 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2928 * skb_zerocopy - Zero copy skb to skb
2929 * @to: destination buffer
2930 * @from: source buffer
2931 * @len: number of bytes to copy from source buffer
2932 * @hlen: size of linear headroom in destination buffer
2934 * Copies up to `len` bytes from `from` to `to` by creating references
2935 * to the frags in the source buffer.
2937 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2938 * headroom in the `to` buffer.
2940 * Return value:
2941 * 0: everything is OK
2942 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2943 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2946 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2948 int i, j = 0;
2949 int plen = 0; /* length of skb->head fragment */
2950 int ret;
2951 struct page *page;
2952 unsigned int offset;
2954 BUG_ON(!from->head_frag && !hlen);
2956 /* dont bother with small payloads */
2957 if (len <= skb_tailroom(to))
2958 return skb_copy_bits(from, 0, skb_put(to, len), len);
2960 if (hlen) {
2961 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2962 if (unlikely(ret))
2963 return ret;
2964 len -= hlen;
2965 } else {
2966 plen = min_t(int, skb_headlen(from), len);
2967 if (plen) {
2968 page = virt_to_head_page(from->head);
2969 offset = from->data - (unsigned char *)page_address(page);
2970 __skb_fill_page_desc(to, 0, page, offset, plen);
2971 get_page(page);
2972 j = 1;
2973 len -= plen;
2977 to->truesize += len + plen;
2978 to->len += len + plen;
2979 to->data_len += len + plen;
2981 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2982 skb_tx_error(from);
2983 return -ENOMEM;
2985 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2987 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2988 int size;
2990 if (!len)
2991 break;
2992 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2993 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2994 len);
2995 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2996 len -= size;
2997 skb_frag_ref(to, j);
2998 j++;
3000 skb_shinfo(to)->nr_frags = j;
3002 return 0;
3004 EXPORT_SYMBOL_GPL(skb_zerocopy);
3006 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3008 __wsum csum;
3009 long csstart;
3011 if (skb->ip_summed == CHECKSUM_PARTIAL)
3012 csstart = skb_checksum_start_offset(skb);
3013 else
3014 csstart = skb_headlen(skb);
3016 BUG_ON(csstart > skb_headlen(skb));
3018 skb_copy_from_linear_data(skb, to, csstart);
3020 csum = 0;
3021 if (csstart != skb->len)
3022 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3023 skb->len - csstart);
3025 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3026 long csstuff = csstart + skb->csum_offset;
3028 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3031 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3034 * skb_dequeue - remove from the head of the queue
3035 * @list: list to dequeue from
3037 * Remove the head of the list. The list lock is taken so the function
3038 * may be used safely with other locking list functions. The head item is
3039 * returned or %NULL if the list is empty.
3042 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3044 unsigned long flags;
3045 struct sk_buff *result;
3047 spin_lock_irqsave(&list->lock, flags);
3048 result = __skb_dequeue(list);
3049 spin_unlock_irqrestore(&list->lock, flags);
3050 return result;
3052 EXPORT_SYMBOL(skb_dequeue);
3055 * skb_dequeue_tail - remove from the tail of the queue
3056 * @list: list to dequeue from
3058 * Remove the tail of the list. The list lock is taken so the function
3059 * may be used safely with other locking list functions. The tail item is
3060 * returned or %NULL if the list is empty.
3062 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3064 unsigned long flags;
3065 struct sk_buff *result;
3067 spin_lock_irqsave(&list->lock, flags);
3068 result = __skb_dequeue_tail(list);
3069 spin_unlock_irqrestore(&list->lock, flags);
3070 return result;
3072 EXPORT_SYMBOL(skb_dequeue_tail);
3075 * skb_queue_purge - empty a list
3076 * @list: list to empty
3078 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3079 * the list and one reference dropped. This function takes the list
3080 * lock and is atomic with respect to other list locking functions.
3082 void skb_queue_purge(struct sk_buff_head *list)
3084 struct sk_buff *skb;
3085 while ((skb = skb_dequeue(list)) != NULL)
3086 kfree_skb(skb);
3088 EXPORT_SYMBOL(skb_queue_purge);
3091 * skb_rbtree_purge - empty a skb rbtree
3092 * @root: root of the rbtree to empty
3093 * Return value: the sum of truesizes of all purged skbs.
3095 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3096 * the list and one reference dropped. This function does not take
3097 * any lock. Synchronization should be handled by the caller (e.g., TCP
3098 * out-of-order queue is protected by the socket lock).
3100 unsigned int skb_rbtree_purge(struct rb_root *root)
3102 struct rb_node *p = rb_first(root);
3103 unsigned int sum = 0;
3105 while (p) {
3106 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3108 p = rb_next(p);
3109 rb_erase(&skb->rbnode, root);
3110 sum += skb->truesize;
3111 kfree_skb(skb);
3113 return sum;
3117 * skb_queue_head - queue a buffer at the list head
3118 * @list: list to use
3119 * @newsk: buffer to queue
3121 * Queue a buffer at the start of the list. This function takes the
3122 * list lock and can be used safely with other locking &sk_buff functions
3123 * safely.
3125 * A buffer cannot be placed on two lists at the same time.
3127 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3129 unsigned long flags;
3131 spin_lock_irqsave(&list->lock, flags);
3132 __skb_queue_head(list, newsk);
3133 spin_unlock_irqrestore(&list->lock, flags);
3135 EXPORT_SYMBOL(skb_queue_head);
3138 * skb_queue_tail - queue a buffer at the list tail
3139 * @list: list to use
3140 * @newsk: buffer to queue
3142 * Queue a buffer at the tail of the list. This function takes the
3143 * list lock and can be used safely with other locking &sk_buff functions
3144 * safely.
3146 * A buffer cannot be placed on two lists at the same time.
3148 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3150 unsigned long flags;
3152 spin_lock_irqsave(&list->lock, flags);
3153 __skb_queue_tail(list, newsk);
3154 spin_unlock_irqrestore(&list->lock, flags);
3156 EXPORT_SYMBOL(skb_queue_tail);
3159 * skb_unlink - remove a buffer from a list
3160 * @skb: buffer to remove
3161 * @list: list to use
3163 * Remove a packet from a list. The list locks are taken and this
3164 * function is atomic with respect to other list locked calls
3166 * You must know what list the SKB is on.
3168 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3170 unsigned long flags;
3172 spin_lock_irqsave(&list->lock, flags);
3173 __skb_unlink(skb, list);
3174 spin_unlock_irqrestore(&list->lock, flags);
3176 EXPORT_SYMBOL(skb_unlink);
3179 * skb_append - append a buffer
3180 * @old: buffer to insert after
3181 * @newsk: buffer to insert
3182 * @list: list to use
3184 * Place a packet after a given packet in a list. The list locks are taken
3185 * and this function is atomic with respect to other list locked calls.
3186 * A buffer cannot be placed on two lists at the same time.
3188 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3190 unsigned long flags;
3192 spin_lock_irqsave(&list->lock, flags);
3193 __skb_queue_after(list, old, newsk);
3194 spin_unlock_irqrestore(&list->lock, flags);
3196 EXPORT_SYMBOL(skb_append);
3198 static inline void skb_split_inside_header(struct sk_buff *skb,
3199 struct sk_buff* skb1,
3200 const u32 len, const int pos)
3202 int i;
3204 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3205 pos - len);
3206 /* And move data appendix as is. */
3207 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3208 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3210 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3211 skb_shinfo(skb)->nr_frags = 0;
3212 skb1->data_len = skb->data_len;
3213 skb1->len += skb1->data_len;
3214 skb->data_len = 0;
3215 skb->len = len;
3216 skb_set_tail_pointer(skb, len);
3219 static inline void skb_split_no_header(struct sk_buff *skb,
3220 struct sk_buff* skb1,
3221 const u32 len, int pos)
3223 int i, k = 0;
3224 const int nfrags = skb_shinfo(skb)->nr_frags;
3226 skb_shinfo(skb)->nr_frags = 0;
3227 skb1->len = skb1->data_len = skb->len - len;
3228 skb->len = len;
3229 skb->data_len = len - pos;
3231 for (i = 0; i < nfrags; i++) {
3232 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3234 if (pos + size > len) {
3235 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3237 if (pos < len) {
3238 /* Split frag.
3239 * We have two variants in this case:
3240 * 1. Move all the frag to the second
3241 * part, if it is possible. F.e.
3242 * this approach is mandatory for TUX,
3243 * where splitting is expensive.
3244 * 2. Split is accurately. We make this.
3246 skb_frag_ref(skb, i);
3247 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3248 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3249 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3250 skb_shinfo(skb)->nr_frags++;
3252 k++;
3253 } else
3254 skb_shinfo(skb)->nr_frags++;
3255 pos += size;
3257 skb_shinfo(skb1)->nr_frags = k;
3261 * skb_split - Split fragmented skb to two parts at length len.
3262 * @skb: the buffer to split
3263 * @skb1: the buffer to receive the second part
3264 * @len: new length for skb
3266 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3268 int pos = skb_headlen(skb);
3270 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3271 SKBTX_SHARED_FRAG;
3272 skb_zerocopy_clone(skb1, skb, 0);
3273 if (len < pos) /* Split line is inside header. */
3274 skb_split_inside_header(skb, skb1, len, pos);
3275 else /* Second chunk has no header, nothing to copy. */
3276 skb_split_no_header(skb, skb1, len, pos);
3278 EXPORT_SYMBOL(skb_split);
3280 /* Shifting from/to a cloned skb is a no-go.
3282 * Caller cannot keep skb_shinfo related pointers past calling here!
3284 static int skb_prepare_for_shift(struct sk_buff *skb)
3286 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3290 * skb_shift - Shifts paged data partially from skb to another
3291 * @tgt: buffer into which tail data gets added
3292 * @skb: buffer from which the paged data comes from
3293 * @shiftlen: shift up to this many bytes
3295 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3296 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3297 * It's up to caller to free skb if everything was shifted.
3299 * If @tgt runs out of frags, the whole operation is aborted.
3301 * Skb cannot include anything else but paged data while tgt is allowed
3302 * to have non-paged data as well.
3304 * TODO: full sized shift could be optimized but that would need
3305 * specialized skb free'er to handle frags without up-to-date nr_frags.
3307 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3309 int from, to, merge, todo;
3310 skb_frag_t *fragfrom, *fragto;
3312 BUG_ON(shiftlen > skb->len);
3314 if (skb_headlen(skb))
3315 return 0;
3316 if (skb_zcopy(tgt) || skb_zcopy(skb))
3317 return 0;
3319 todo = shiftlen;
3320 from = 0;
3321 to = skb_shinfo(tgt)->nr_frags;
3322 fragfrom = &skb_shinfo(skb)->frags[from];
3324 /* Actual merge is delayed until the point when we know we can
3325 * commit all, so that we don't have to undo partial changes
3327 if (!to ||
3328 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3329 skb_frag_off(fragfrom))) {
3330 merge = -1;
3331 } else {
3332 merge = to - 1;
3334 todo -= skb_frag_size(fragfrom);
3335 if (todo < 0) {
3336 if (skb_prepare_for_shift(skb) ||
3337 skb_prepare_for_shift(tgt))
3338 return 0;
3340 /* All previous frag pointers might be stale! */
3341 fragfrom = &skb_shinfo(skb)->frags[from];
3342 fragto = &skb_shinfo(tgt)->frags[merge];
3344 skb_frag_size_add(fragto, shiftlen);
3345 skb_frag_size_sub(fragfrom, shiftlen);
3346 skb_frag_off_add(fragfrom, shiftlen);
3348 goto onlymerged;
3351 from++;
3354 /* Skip full, not-fitting skb to avoid expensive operations */
3355 if ((shiftlen == skb->len) &&
3356 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3357 return 0;
3359 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3360 return 0;
3362 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3363 if (to == MAX_SKB_FRAGS)
3364 return 0;
3366 fragfrom = &skb_shinfo(skb)->frags[from];
3367 fragto = &skb_shinfo(tgt)->frags[to];
3369 if (todo >= skb_frag_size(fragfrom)) {
3370 *fragto = *fragfrom;
3371 todo -= skb_frag_size(fragfrom);
3372 from++;
3373 to++;
3375 } else {
3376 __skb_frag_ref(fragfrom);
3377 skb_frag_page_copy(fragto, fragfrom);
3378 skb_frag_off_copy(fragto, fragfrom);
3379 skb_frag_size_set(fragto, todo);
3381 skb_frag_off_add(fragfrom, todo);
3382 skb_frag_size_sub(fragfrom, todo);
3383 todo = 0;
3385 to++;
3386 break;
3390 /* Ready to "commit" this state change to tgt */
3391 skb_shinfo(tgt)->nr_frags = to;
3393 if (merge >= 0) {
3394 fragfrom = &skb_shinfo(skb)->frags[0];
3395 fragto = &skb_shinfo(tgt)->frags[merge];
3397 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3398 __skb_frag_unref(fragfrom);
3401 /* Reposition in the original skb */
3402 to = 0;
3403 while (from < skb_shinfo(skb)->nr_frags)
3404 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3405 skb_shinfo(skb)->nr_frags = to;
3407 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3409 onlymerged:
3410 /* Most likely the tgt won't ever need its checksum anymore, skb on
3411 * the other hand might need it if it needs to be resent
3413 tgt->ip_summed = CHECKSUM_PARTIAL;
3414 skb->ip_summed = CHECKSUM_PARTIAL;
3416 /* Yak, is it really working this way? Some helper please? */
3417 skb->len -= shiftlen;
3418 skb->data_len -= shiftlen;
3419 skb->truesize -= shiftlen;
3420 tgt->len += shiftlen;
3421 tgt->data_len += shiftlen;
3422 tgt->truesize += shiftlen;
3424 return shiftlen;
3428 * skb_prepare_seq_read - Prepare a sequential read of skb data
3429 * @skb: the buffer to read
3430 * @from: lower offset of data to be read
3431 * @to: upper offset of data to be read
3432 * @st: state variable
3434 * Initializes the specified state variable. Must be called before
3435 * invoking skb_seq_read() for the first time.
3437 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3438 unsigned int to, struct skb_seq_state *st)
3440 st->lower_offset = from;
3441 st->upper_offset = to;
3442 st->root_skb = st->cur_skb = skb;
3443 st->frag_idx = st->stepped_offset = 0;
3444 st->frag_data = NULL;
3446 EXPORT_SYMBOL(skb_prepare_seq_read);
3449 * skb_seq_read - Sequentially read skb data
3450 * @consumed: number of bytes consumed by the caller so far
3451 * @data: destination pointer for data to be returned
3452 * @st: state variable
3454 * Reads a block of skb data at @consumed relative to the
3455 * lower offset specified to skb_prepare_seq_read(). Assigns
3456 * the head of the data block to @data and returns the length
3457 * of the block or 0 if the end of the skb data or the upper
3458 * offset has been reached.
3460 * The caller is not required to consume all of the data
3461 * returned, i.e. @consumed is typically set to the number
3462 * of bytes already consumed and the next call to
3463 * skb_seq_read() will return the remaining part of the block.
3465 * Note 1: The size of each block of data returned can be arbitrary,
3466 * this limitation is the cost for zerocopy sequential
3467 * reads of potentially non linear data.
3469 * Note 2: Fragment lists within fragments are not implemented
3470 * at the moment, state->root_skb could be replaced with
3471 * a stack for this purpose.
3473 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3474 struct skb_seq_state *st)
3476 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3477 skb_frag_t *frag;
3479 if (unlikely(abs_offset >= st->upper_offset)) {
3480 if (st->frag_data) {
3481 kunmap_atomic(st->frag_data);
3482 st->frag_data = NULL;
3484 return 0;
3487 next_skb:
3488 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3490 if (abs_offset < block_limit && !st->frag_data) {
3491 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3492 return block_limit - abs_offset;
3495 if (st->frag_idx == 0 && !st->frag_data)
3496 st->stepped_offset += skb_headlen(st->cur_skb);
3498 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3499 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3500 block_limit = skb_frag_size(frag) + st->stepped_offset;
3502 if (abs_offset < block_limit) {
3503 if (!st->frag_data)
3504 st->frag_data = kmap_atomic(skb_frag_page(frag));
3506 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3507 (abs_offset - st->stepped_offset);
3509 return block_limit - abs_offset;
3512 if (st->frag_data) {
3513 kunmap_atomic(st->frag_data);
3514 st->frag_data = NULL;
3517 st->frag_idx++;
3518 st->stepped_offset += skb_frag_size(frag);
3521 if (st->frag_data) {
3522 kunmap_atomic(st->frag_data);
3523 st->frag_data = NULL;
3526 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3527 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3528 st->frag_idx = 0;
3529 goto next_skb;
3530 } else if (st->cur_skb->next) {
3531 st->cur_skb = st->cur_skb->next;
3532 st->frag_idx = 0;
3533 goto next_skb;
3536 return 0;
3538 EXPORT_SYMBOL(skb_seq_read);
3541 * skb_abort_seq_read - Abort a sequential read of skb data
3542 * @st: state variable
3544 * Must be called if skb_seq_read() was not called until it
3545 * returned 0.
3547 void skb_abort_seq_read(struct skb_seq_state *st)
3549 if (st->frag_data)
3550 kunmap_atomic(st->frag_data);
3552 EXPORT_SYMBOL(skb_abort_seq_read);
3554 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3556 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3557 struct ts_config *conf,
3558 struct ts_state *state)
3560 return skb_seq_read(offset, text, TS_SKB_CB(state));
3563 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3565 skb_abort_seq_read(TS_SKB_CB(state));
3569 * skb_find_text - Find a text pattern in skb data
3570 * @skb: the buffer to look in
3571 * @from: search offset
3572 * @to: search limit
3573 * @config: textsearch configuration
3575 * Finds a pattern in the skb data according to the specified
3576 * textsearch configuration. Use textsearch_next() to retrieve
3577 * subsequent occurrences of the pattern. Returns the offset
3578 * to the first occurrence or UINT_MAX if no match was found.
3580 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3581 unsigned int to, struct ts_config *config)
3583 struct ts_state state;
3584 unsigned int ret;
3586 config->get_next_block = skb_ts_get_next_block;
3587 config->finish = skb_ts_finish;
3589 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3591 ret = textsearch_find(config, &state);
3592 return (ret <= to - from ? ret : UINT_MAX);
3594 EXPORT_SYMBOL(skb_find_text);
3596 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3597 int offset, size_t size)
3599 int i = skb_shinfo(skb)->nr_frags;
3601 if (skb_can_coalesce(skb, i, page, offset)) {
3602 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3603 } else if (i < MAX_SKB_FRAGS) {
3604 get_page(page);
3605 skb_fill_page_desc(skb, i, page, offset, size);
3606 } else {
3607 return -EMSGSIZE;
3610 return 0;
3612 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3615 * skb_pull_rcsum - pull skb and update receive checksum
3616 * @skb: buffer to update
3617 * @len: length of data pulled
3619 * This function performs an skb_pull on the packet and updates
3620 * the CHECKSUM_COMPLETE checksum. It should be used on
3621 * receive path processing instead of skb_pull unless you know
3622 * that the checksum difference is zero (e.g., a valid IP header)
3623 * or you are setting ip_summed to CHECKSUM_NONE.
3625 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3627 unsigned char *data = skb->data;
3629 BUG_ON(len > skb->len);
3630 __skb_pull(skb, len);
3631 skb_postpull_rcsum(skb, data, len);
3632 return skb->data;
3634 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3636 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3638 skb_frag_t head_frag;
3639 struct page *page;
3641 page = virt_to_head_page(frag_skb->head);
3642 __skb_frag_set_page(&head_frag, page);
3643 skb_frag_off_set(&head_frag, frag_skb->data -
3644 (unsigned char *)page_address(page));
3645 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3646 return head_frag;
3649 struct sk_buff *skb_segment_list(struct sk_buff *skb,
3650 netdev_features_t features,
3651 unsigned int offset)
3653 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3654 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3655 unsigned int delta_truesize = 0;
3656 unsigned int delta_len = 0;
3657 struct sk_buff *tail = NULL;
3658 struct sk_buff *nskb;
3660 skb_push(skb, -skb_network_offset(skb) + offset);
3662 skb_shinfo(skb)->frag_list = NULL;
3664 do {
3665 nskb = list_skb;
3666 list_skb = list_skb->next;
3668 if (!tail)
3669 skb->next = nskb;
3670 else
3671 tail->next = nskb;
3673 tail = nskb;
3675 delta_len += nskb->len;
3676 delta_truesize += nskb->truesize;
3678 skb_push(nskb, -skb_network_offset(nskb) + offset);
3680 skb_release_head_state(nskb);
3681 __copy_skb_header(nskb, skb);
3683 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3684 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3685 nskb->data - tnl_hlen,
3686 offset + tnl_hlen);
3688 if (skb_needs_linearize(nskb, features) &&
3689 __skb_linearize(nskb))
3690 goto err_linearize;
3692 } while (list_skb);
3694 skb->truesize = skb->truesize - delta_truesize;
3695 skb->data_len = skb->data_len - delta_len;
3696 skb->len = skb->len - delta_len;
3698 skb_gso_reset(skb);
3700 skb->prev = tail;
3702 if (skb_needs_linearize(skb, features) &&
3703 __skb_linearize(skb))
3704 goto err_linearize;
3706 skb_get(skb);
3708 return skb;
3710 err_linearize:
3711 kfree_skb_list(skb->next);
3712 skb->next = NULL;
3713 return ERR_PTR(-ENOMEM);
3715 EXPORT_SYMBOL_GPL(skb_segment_list);
3717 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3719 if (unlikely(p->len + skb->len >= 65536))
3720 return -E2BIG;
3722 if (NAPI_GRO_CB(p)->last == p)
3723 skb_shinfo(p)->frag_list = skb;
3724 else
3725 NAPI_GRO_CB(p)->last->next = skb;
3727 skb_pull(skb, skb_gro_offset(skb));
3729 NAPI_GRO_CB(p)->last = skb;
3730 NAPI_GRO_CB(p)->count++;
3731 p->data_len += skb->len;
3732 p->truesize += skb->truesize;
3733 p->len += skb->len;
3735 NAPI_GRO_CB(skb)->same_flow = 1;
3737 return 0;
3741 * skb_segment - Perform protocol segmentation on skb.
3742 * @head_skb: buffer to segment
3743 * @features: features for the output path (see dev->features)
3745 * This function performs segmentation on the given skb. It returns
3746 * a pointer to the first in a list of new skbs for the segments.
3747 * In case of error it returns ERR_PTR(err).
3749 struct sk_buff *skb_segment(struct sk_buff *head_skb,
3750 netdev_features_t features)
3752 struct sk_buff *segs = NULL;
3753 struct sk_buff *tail = NULL;
3754 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3755 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3756 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3757 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3758 struct sk_buff *frag_skb = head_skb;
3759 unsigned int offset = doffset;
3760 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3761 unsigned int partial_segs = 0;
3762 unsigned int headroom;
3763 unsigned int len = head_skb->len;
3764 __be16 proto;
3765 bool csum, sg;
3766 int nfrags = skb_shinfo(head_skb)->nr_frags;
3767 int err = -ENOMEM;
3768 int i = 0;
3769 int pos;
3771 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3772 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3773 /* gso_size is untrusted, and we have a frag_list with a linear
3774 * non head_frag head.
3776 * (we assume checking the first list_skb member suffices;
3777 * i.e if either of the list_skb members have non head_frag
3778 * head, then the first one has too).
3780 * If head_skb's headlen does not fit requested gso_size, it
3781 * means that the frag_list members do NOT terminate on exact
3782 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3783 * sharing. Therefore we must fallback to copying the frag_list
3784 * skbs; we do so by disabling SG.
3786 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3787 features &= ~NETIF_F_SG;
3790 __skb_push(head_skb, doffset);
3791 proto = skb_network_protocol(head_skb, NULL);
3792 if (unlikely(!proto))
3793 return ERR_PTR(-EINVAL);
3795 sg = !!(features & NETIF_F_SG);
3796 csum = !!can_checksum_protocol(features, proto);
3798 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3799 if (!(features & NETIF_F_GSO_PARTIAL)) {
3800 struct sk_buff *iter;
3801 unsigned int frag_len;
3803 if (!list_skb ||
3804 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3805 goto normal;
3807 /* If we get here then all the required
3808 * GSO features except frag_list are supported.
3809 * Try to split the SKB to multiple GSO SKBs
3810 * with no frag_list.
3811 * Currently we can do that only when the buffers don't
3812 * have a linear part and all the buffers except
3813 * the last are of the same length.
3815 frag_len = list_skb->len;
3816 skb_walk_frags(head_skb, iter) {
3817 if (frag_len != iter->len && iter->next)
3818 goto normal;
3819 if (skb_headlen(iter) && !iter->head_frag)
3820 goto normal;
3822 len -= iter->len;
3825 if (len != frag_len)
3826 goto normal;
3829 /* GSO partial only requires that we trim off any excess that
3830 * doesn't fit into an MSS sized block, so take care of that
3831 * now.
3833 partial_segs = len / mss;
3834 if (partial_segs > 1)
3835 mss *= partial_segs;
3836 else
3837 partial_segs = 0;
3840 normal:
3841 headroom = skb_headroom(head_skb);
3842 pos = skb_headlen(head_skb);
3844 do {
3845 struct sk_buff *nskb;
3846 skb_frag_t *nskb_frag;
3847 int hsize;
3848 int size;
3850 if (unlikely(mss == GSO_BY_FRAGS)) {
3851 len = list_skb->len;
3852 } else {
3853 len = head_skb->len - offset;
3854 if (len > mss)
3855 len = mss;
3858 hsize = skb_headlen(head_skb) - offset;
3859 if (hsize < 0)
3860 hsize = 0;
3861 if (hsize > len || !sg)
3862 hsize = len;
3864 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3865 (skb_headlen(list_skb) == len || sg)) {
3866 BUG_ON(skb_headlen(list_skb) > len);
3868 i = 0;
3869 nfrags = skb_shinfo(list_skb)->nr_frags;
3870 frag = skb_shinfo(list_skb)->frags;
3871 frag_skb = list_skb;
3872 pos += skb_headlen(list_skb);
3874 while (pos < offset + len) {
3875 BUG_ON(i >= nfrags);
3877 size = skb_frag_size(frag);
3878 if (pos + size > offset + len)
3879 break;
3881 i++;
3882 pos += size;
3883 frag++;
3886 nskb = skb_clone(list_skb, GFP_ATOMIC);
3887 list_skb = list_skb->next;
3889 if (unlikely(!nskb))
3890 goto err;
3892 if (unlikely(pskb_trim(nskb, len))) {
3893 kfree_skb(nskb);
3894 goto err;
3897 hsize = skb_end_offset(nskb);
3898 if (skb_cow_head(nskb, doffset + headroom)) {
3899 kfree_skb(nskb);
3900 goto err;
3903 nskb->truesize += skb_end_offset(nskb) - hsize;
3904 skb_release_head_state(nskb);
3905 __skb_push(nskb, doffset);
3906 } else {
3907 nskb = __alloc_skb(hsize + doffset + headroom,
3908 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3909 NUMA_NO_NODE);
3911 if (unlikely(!nskb))
3912 goto err;
3914 skb_reserve(nskb, headroom);
3915 __skb_put(nskb, doffset);
3918 if (segs)
3919 tail->next = nskb;
3920 else
3921 segs = nskb;
3922 tail = nskb;
3924 __copy_skb_header(nskb, head_skb);
3926 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3927 skb_reset_mac_len(nskb);
3929 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3930 nskb->data - tnl_hlen,
3931 doffset + tnl_hlen);
3933 if (nskb->len == len + doffset)
3934 goto perform_csum_check;
3936 if (!sg) {
3937 if (!csum) {
3938 if (!nskb->remcsum_offload)
3939 nskb->ip_summed = CHECKSUM_NONE;
3940 SKB_GSO_CB(nskb)->csum =
3941 skb_copy_and_csum_bits(head_skb, offset,
3942 skb_put(nskb,
3943 len),
3944 len);
3945 SKB_GSO_CB(nskb)->csum_start =
3946 skb_headroom(nskb) + doffset;
3947 } else {
3948 skb_copy_bits(head_skb, offset,
3949 skb_put(nskb, len),
3950 len);
3952 continue;
3955 nskb_frag = skb_shinfo(nskb)->frags;
3957 skb_copy_from_linear_data_offset(head_skb, offset,
3958 skb_put(nskb, hsize), hsize);
3960 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3961 SKBTX_SHARED_FRAG;
3963 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3964 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3965 goto err;
3967 while (pos < offset + len) {
3968 if (i >= nfrags) {
3969 i = 0;
3970 nfrags = skb_shinfo(list_skb)->nr_frags;
3971 frag = skb_shinfo(list_skb)->frags;
3972 frag_skb = list_skb;
3973 if (!skb_headlen(list_skb)) {
3974 BUG_ON(!nfrags);
3975 } else {
3976 BUG_ON(!list_skb->head_frag);
3978 /* to make room for head_frag. */
3979 i--;
3980 frag--;
3982 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3983 skb_zerocopy_clone(nskb, frag_skb,
3984 GFP_ATOMIC))
3985 goto err;
3987 list_skb = list_skb->next;
3990 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3991 MAX_SKB_FRAGS)) {
3992 net_warn_ratelimited(
3993 "skb_segment: too many frags: %u %u\n",
3994 pos, mss);
3995 err = -EINVAL;
3996 goto err;
3999 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4000 __skb_frag_ref(nskb_frag);
4001 size = skb_frag_size(nskb_frag);
4003 if (pos < offset) {
4004 skb_frag_off_add(nskb_frag, offset - pos);
4005 skb_frag_size_sub(nskb_frag, offset - pos);
4008 skb_shinfo(nskb)->nr_frags++;
4010 if (pos + size <= offset + len) {
4011 i++;
4012 frag++;
4013 pos += size;
4014 } else {
4015 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4016 goto skip_fraglist;
4019 nskb_frag++;
4022 skip_fraglist:
4023 nskb->data_len = len - hsize;
4024 nskb->len += nskb->data_len;
4025 nskb->truesize += nskb->data_len;
4027 perform_csum_check:
4028 if (!csum) {
4029 if (skb_has_shared_frag(nskb) &&
4030 __skb_linearize(nskb))
4031 goto err;
4033 if (!nskb->remcsum_offload)
4034 nskb->ip_summed = CHECKSUM_NONE;
4035 SKB_GSO_CB(nskb)->csum =
4036 skb_checksum(nskb, doffset,
4037 nskb->len - doffset, 0);
4038 SKB_GSO_CB(nskb)->csum_start =
4039 skb_headroom(nskb) + doffset;
4041 } while ((offset += len) < head_skb->len);
4043 /* Some callers want to get the end of the list.
4044 * Put it in segs->prev to avoid walking the list.
4045 * (see validate_xmit_skb_list() for example)
4047 segs->prev = tail;
4049 if (partial_segs) {
4050 struct sk_buff *iter;
4051 int type = skb_shinfo(head_skb)->gso_type;
4052 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4054 /* Update type to add partial and then remove dodgy if set */
4055 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4056 type &= ~SKB_GSO_DODGY;
4058 /* Update GSO info and prepare to start updating headers on
4059 * our way back down the stack of protocols.
4061 for (iter = segs; iter; iter = iter->next) {
4062 skb_shinfo(iter)->gso_size = gso_size;
4063 skb_shinfo(iter)->gso_segs = partial_segs;
4064 skb_shinfo(iter)->gso_type = type;
4065 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4068 if (tail->len - doffset <= gso_size)
4069 skb_shinfo(tail)->gso_size = 0;
4070 else if (tail != segs)
4071 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4074 /* Following permits correct backpressure, for protocols
4075 * using skb_set_owner_w().
4076 * Idea is to tranfert ownership from head_skb to last segment.
4078 if (head_skb->destructor == sock_wfree) {
4079 swap(tail->truesize, head_skb->truesize);
4080 swap(tail->destructor, head_skb->destructor);
4081 swap(tail->sk, head_skb->sk);
4083 return segs;
4085 err:
4086 kfree_skb_list(segs);
4087 return ERR_PTR(err);
4089 EXPORT_SYMBOL_GPL(skb_segment);
4091 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4093 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4094 unsigned int offset = skb_gro_offset(skb);
4095 unsigned int headlen = skb_headlen(skb);
4096 unsigned int len = skb_gro_len(skb);
4097 unsigned int delta_truesize;
4098 struct sk_buff *lp;
4100 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4101 return -E2BIG;
4103 lp = NAPI_GRO_CB(p)->last;
4104 pinfo = skb_shinfo(lp);
4106 if (headlen <= offset) {
4107 skb_frag_t *frag;
4108 skb_frag_t *frag2;
4109 int i = skbinfo->nr_frags;
4110 int nr_frags = pinfo->nr_frags + i;
4112 if (nr_frags > MAX_SKB_FRAGS)
4113 goto merge;
4115 offset -= headlen;
4116 pinfo->nr_frags = nr_frags;
4117 skbinfo->nr_frags = 0;
4119 frag = pinfo->frags + nr_frags;
4120 frag2 = skbinfo->frags + i;
4121 do {
4122 *--frag = *--frag2;
4123 } while (--i);
4125 skb_frag_off_add(frag, offset);
4126 skb_frag_size_sub(frag, offset);
4128 /* all fragments truesize : remove (head size + sk_buff) */
4129 delta_truesize = skb->truesize -
4130 SKB_TRUESIZE(skb_end_offset(skb));
4132 skb->truesize -= skb->data_len;
4133 skb->len -= skb->data_len;
4134 skb->data_len = 0;
4136 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4137 goto done;
4138 } else if (skb->head_frag) {
4139 int nr_frags = pinfo->nr_frags;
4140 skb_frag_t *frag = pinfo->frags + nr_frags;
4141 struct page *page = virt_to_head_page(skb->head);
4142 unsigned int first_size = headlen - offset;
4143 unsigned int first_offset;
4145 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4146 goto merge;
4148 first_offset = skb->data -
4149 (unsigned char *)page_address(page) +
4150 offset;
4152 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4154 __skb_frag_set_page(frag, page);
4155 skb_frag_off_set(frag, first_offset);
4156 skb_frag_size_set(frag, first_size);
4158 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4159 /* We dont need to clear skbinfo->nr_frags here */
4161 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4162 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4163 goto done;
4166 merge:
4167 delta_truesize = skb->truesize;
4168 if (offset > headlen) {
4169 unsigned int eat = offset - headlen;
4171 skb_frag_off_add(&skbinfo->frags[0], eat);
4172 skb_frag_size_sub(&skbinfo->frags[0], eat);
4173 skb->data_len -= eat;
4174 skb->len -= eat;
4175 offset = headlen;
4178 __skb_pull(skb, offset);
4180 if (NAPI_GRO_CB(p)->last == p)
4181 skb_shinfo(p)->frag_list = skb;
4182 else
4183 NAPI_GRO_CB(p)->last->next = skb;
4184 NAPI_GRO_CB(p)->last = skb;
4185 __skb_header_release(skb);
4186 lp = p;
4188 done:
4189 NAPI_GRO_CB(p)->count++;
4190 p->data_len += len;
4191 p->truesize += delta_truesize;
4192 p->len += len;
4193 if (lp != p) {
4194 lp->data_len += len;
4195 lp->truesize += delta_truesize;
4196 lp->len += len;
4198 NAPI_GRO_CB(skb)->same_flow = 1;
4199 return 0;
4202 #ifdef CONFIG_SKB_EXTENSIONS
4203 #define SKB_EXT_ALIGN_VALUE 8
4204 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4206 static const u8 skb_ext_type_len[] = {
4207 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4208 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4209 #endif
4210 #ifdef CONFIG_XFRM
4211 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4212 #endif
4213 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4214 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4215 #endif
4216 #if IS_ENABLED(CONFIG_MPTCP)
4217 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4218 #endif
4221 static __always_inline unsigned int skb_ext_total_length(void)
4223 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4224 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4225 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4226 #endif
4227 #ifdef CONFIG_XFRM
4228 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4229 #endif
4230 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4231 skb_ext_type_len[TC_SKB_EXT] +
4232 #endif
4233 #if IS_ENABLED(CONFIG_MPTCP)
4234 skb_ext_type_len[SKB_EXT_MPTCP] +
4235 #endif
4239 static void skb_extensions_init(void)
4241 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4242 BUILD_BUG_ON(skb_ext_total_length() > 255);
4244 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4245 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4247 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4248 NULL);
4250 #else
4251 static void skb_extensions_init(void) {}
4252 #endif
4254 void __init skb_init(void)
4256 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4257 sizeof(struct sk_buff),
4259 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4260 offsetof(struct sk_buff, cb),
4261 sizeof_field(struct sk_buff, cb),
4262 NULL);
4263 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4264 sizeof(struct sk_buff_fclones),
4266 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4267 NULL);
4268 skb_extensions_init();
4271 static int
4272 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4273 unsigned int recursion_level)
4275 int start = skb_headlen(skb);
4276 int i, copy = start - offset;
4277 struct sk_buff *frag_iter;
4278 int elt = 0;
4280 if (unlikely(recursion_level >= 24))
4281 return -EMSGSIZE;
4283 if (copy > 0) {
4284 if (copy > len)
4285 copy = len;
4286 sg_set_buf(sg, skb->data + offset, copy);
4287 elt++;
4288 if ((len -= copy) == 0)
4289 return elt;
4290 offset += copy;
4293 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4294 int end;
4296 WARN_ON(start > offset + len);
4298 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4299 if ((copy = end - offset) > 0) {
4300 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4301 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4302 return -EMSGSIZE;
4304 if (copy > len)
4305 copy = len;
4306 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4307 skb_frag_off(frag) + offset - start);
4308 elt++;
4309 if (!(len -= copy))
4310 return elt;
4311 offset += copy;
4313 start = end;
4316 skb_walk_frags(skb, frag_iter) {
4317 int end, ret;
4319 WARN_ON(start > offset + len);
4321 end = start + frag_iter->len;
4322 if ((copy = end - offset) > 0) {
4323 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4324 return -EMSGSIZE;
4326 if (copy > len)
4327 copy = len;
4328 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4329 copy, recursion_level + 1);
4330 if (unlikely(ret < 0))
4331 return ret;
4332 elt += ret;
4333 if ((len -= copy) == 0)
4334 return elt;
4335 offset += copy;
4337 start = end;
4339 BUG_ON(len);
4340 return elt;
4344 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4345 * @skb: Socket buffer containing the buffers to be mapped
4346 * @sg: The scatter-gather list to map into
4347 * @offset: The offset into the buffer's contents to start mapping
4348 * @len: Length of buffer space to be mapped
4350 * Fill the specified scatter-gather list with mappings/pointers into a
4351 * region of the buffer space attached to a socket buffer. Returns either
4352 * the number of scatterlist items used, or -EMSGSIZE if the contents
4353 * could not fit.
4355 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4357 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4359 if (nsg <= 0)
4360 return nsg;
4362 sg_mark_end(&sg[nsg - 1]);
4364 return nsg;
4366 EXPORT_SYMBOL_GPL(skb_to_sgvec);
4368 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4369 * sglist without mark the sg which contain last skb data as the end.
4370 * So the caller can mannipulate sg list as will when padding new data after
4371 * the first call without calling sg_unmark_end to expend sg list.
4373 * Scenario to use skb_to_sgvec_nomark:
4374 * 1. sg_init_table
4375 * 2. skb_to_sgvec_nomark(payload1)
4376 * 3. skb_to_sgvec_nomark(payload2)
4378 * This is equivalent to:
4379 * 1. sg_init_table
4380 * 2. skb_to_sgvec(payload1)
4381 * 3. sg_unmark_end
4382 * 4. skb_to_sgvec(payload2)
4384 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4385 * is more preferable.
4387 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4388 int offset, int len)
4390 return __skb_to_sgvec(skb, sg, offset, len, 0);
4392 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4397 * skb_cow_data - Check that a socket buffer's data buffers are writable
4398 * @skb: The socket buffer to check.
4399 * @tailbits: Amount of trailing space to be added
4400 * @trailer: Returned pointer to the skb where the @tailbits space begins
4402 * Make sure that the data buffers attached to a socket buffer are
4403 * writable. If they are not, private copies are made of the data buffers
4404 * and the socket buffer is set to use these instead.
4406 * If @tailbits is given, make sure that there is space to write @tailbits
4407 * bytes of data beyond current end of socket buffer. @trailer will be
4408 * set to point to the skb in which this space begins.
4410 * The number of scatterlist elements required to completely map the
4411 * COW'd and extended socket buffer will be returned.
4413 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4415 int copyflag;
4416 int elt;
4417 struct sk_buff *skb1, **skb_p;
4419 /* If skb is cloned or its head is paged, reallocate
4420 * head pulling out all the pages (pages are considered not writable
4421 * at the moment even if they are anonymous).
4423 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4424 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4425 return -ENOMEM;
4427 /* Easy case. Most of packets will go this way. */
4428 if (!skb_has_frag_list(skb)) {
4429 /* A little of trouble, not enough of space for trailer.
4430 * This should not happen, when stack is tuned to generate
4431 * good frames. OK, on miss we reallocate and reserve even more
4432 * space, 128 bytes is fair. */
4434 if (skb_tailroom(skb) < tailbits &&
4435 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4436 return -ENOMEM;
4438 /* Voila! */
4439 *trailer = skb;
4440 return 1;
4443 /* Misery. We are in troubles, going to mincer fragments... */
4445 elt = 1;
4446 skb_p = &skb_shinfo(skb)->frag_list;
4447 copyflag = 0;
4449 while ((skb1 = *skb_p) != NULL) {
4450 int ntail = 0;
4452 /* The fragment is partially pulled by someone,
4453 * this can happen on input. Copy it and everything
4454 * after it. */
4456 if (skb_shared(skb1))
4457 copyflag = 1;
4459 /* If the skb is the last, worry about trailer. */
4461 if (skb1->next == NULL && tailbits) {
4462 if (skb_shinfo(skb1)->nr_frags ||
4463 skb_has_frag_list(skb1) ||
4464 skb_tailroom(skb1) < tailbits)
4465 ntail = tailbits + 128;
4468 if (copyflag ||
4469 skb_cloned(skb1) ||
4470 ntail ||
4471 skb_shinfo(skb1)->nr_frags ||
4472 skb_has_frag_list(skb1)) {
4473 struct sk_buff *skb2;
4475 /* Fuck, we are miserable poor guys... */
4476 if (ntail == 0)
4477 skb2 = skb_copy(skb1, GFP_ATOMIC);
4478 else
4479 skb2 = skb_copy_expand(skb1,
4480 skb_headroom(skb1),
4481 ntail,
4482 GFP_ATOMIC);
4483 if (unlikely(skb2 == NULL))
4484 return -ENOMEM;
4486 if (skb1->sk)
4487 skb_set_owner_w(skb2, skb1->sk);
4489 /* Looking around. Are we still alive?
4490 * OK, link new skb, drop old one */
4492 skb2->next = skb1->next;
4493 *skb_p = skb2;
4494 kfree_skb(skb1);
4495 skb1 = skb2;
4497 elt++;
4498 *trailer = skb1;
4499 skb_p = &skb1->next;
4502 return elt;
4504 EXPORT_SYMBOL_GPL(skb_cow_data);
4506 static void sock_rmem_free(struct sk_buff *skb)
4508 struct sock *sk = skb->sk;
4510 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4513 static void skb_set_err_queue(struct sk_buff *skb)
4515 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4516 * So, it is safe to (mis)use it to mark skbs on the error queue.
4518 skb->pkt_type = PACKET_OUTGOING;
4519 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4523 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4525 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4527 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4528 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4529 return -ENOMEM;
4531 skb_orphan(skb);
4532 skb->sk = sk;
4533 skb->destructor = sock_rmem_free;
4534 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4535 skb_set_err_queue(skb);
4537 /* before exiting rcu section, make sure dst is refcounted */
4538 skb_dst_force(skb);
4540 skb_queue_tail(&sk->sk_error_queue, skb);
4541 if (!sock_flag(sk, SOCK_DEAD))
4542 sk->sk_error_report(sk);
4543 return 0;
4545 EXPORT_SYMBOL(sock_queue_err_skb);
4547 static bool is_icmp_err_skb(const struct sk_buff *skb)
4549 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4550 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4553 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4555 struct sk_buff_head *q = &sk->sk_error_queue;
4556 struct sk_buff *skb, *skb_next = NULL;
4557 bool icmp_next = false;
4558 unsigned long flags;
4560 spin_lock_irqsave(&q->lock, flags);
4561 skb = __skb_dequeue(q);
4562 if (skb && (skb_next = skb_peek(q))) {
4563 icmp_next = is_icmp_err_skb(skb_next);
4564 if (icmp_next)
4565 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4567 spin_unlock_irqrestore(&q->lock, flags);
4569 if (is_icmp_err_skb(skb) && !icmp_next)
4570 sk->sk_err = 0;
4572 if (skb_next)
4573 sk->sk_error_report(sk);
4575 return skb;
4577 EXPORT_SYMBOL(sock_dequeue_err_skb);
4580 * skb_clone_sk - create clone of skb, and take reference to socket
4581 * @skb: the skb to clone
4583 * This function creates a clone of a buffer that holds a reference on
4584 * sk_refcnt. Buffers created via this function are meant to be
4585 * returned using sock_queue_err_skb, or free via kfree_skb.
4587 * When passing buffers allocated with this function to sock_queue_err_skb
4588 * it is necessary to wrap the call with sock_hold/sock_put in order to
4589 * prevent the socket from being released prior to being enqueued on
4590 * the sk_error_queue.
4592 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4594 struct sock *sk = skb->sk;
4595 struct sk_buff *clone;
4597 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4598 return NULL;
4600 clone = skb_clone(skb, GFP_ATOMIC);
4601 if (!clone) {
4602 sock_put(sk);
4603 return NULL;
4606 clone->sk = sk;
4607 clone->destructor = sock_efree;
4609 return clone;
4611 EXPORT_SYMBOL(skb_clone_sk);
4613 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4614 struct sock *sk,
4615 int tstype,
4616 bool opt_stats)
4618 struct sock_exterr_skb *serr;
4619 int err;
4621 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4623 serr = SKB_EXT_ERR(skb);
4624 memset(serr, 0, sizeof(*serr));
4625 serr->ee.ee_errno = ENOMSG;
4626 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4627 serr->ee.ee_info = tstype;
4628 serr->opt_stats = opt_stats;
4629 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4630 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4631 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4632 if (sk->sk_protocol == IPPROTO_TCP &&
4633 sk->sk_type == SOCK_STREAM)
4634 serr->ee.ee_data -= sk->sk_tskey;
4637 err = sock_queue_err_skb(sk, skb);
4639 if (err)
4640 kfree_skb(skb);
4643 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4645 bool ret;
4647 if (likely(sysctl_tstamp_allow_data || tsonly))
4648 return true;
4650 read_lock_bh(&sk->sk_callback_lock);
4651 ret = sk->sk_socket && sk->sk_socket->file &&
4652 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4653 read_unlock_bh(&sk->sk_callback_lock);
4654 return ret;
4657 void skb_complete_tx_timestamp(struct sk_buff *skb,
4658 struct skb_shared_hwtstamps *hwtstamps)
4660 struct sock *sk = skb->sk;
4662 if (!skb_may_tx_timestamp(sk, false))
4663 goto err;
4665 /* Take a reference to prevent skb_orphan() from freeing the socket,
4666 * but only if the socket refcount is not zero.
4668 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4669 *skb_hwtstamps(skb) = *hwtstamps;
4670 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4671 sock_put(sk);
4672 return;
4675 err:
4676 kfree_skb(skb);
4678 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4680 void __skb_tstamp_tx(struct sk_buff *orig_skb,
4681 struct skb_shared_hwtstamps *hwtstamps,
4682 struct sock *sk, int tstype)
4684 struct sk_buff *skb;
4685 bool tsonly, opt_stats = false;
4687 if (!sk)
4688 return;
4690 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4691 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4692 return;
4694 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4695 if (!skb_may_tx_timestamp(sk, tsonly))
4696 return;
4698 if (tsonly) {
4699 #ifdef CONFIG_INET
4700 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4701 sk->sk_protocol == IPPROTO_TCP &&
4702 sk->sk_type == SOCK_STREAM) {
4703 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4704 opt_stats = true;
4705 } else
4706 #endif
4707 skb = alloc_skb(0, GFP_ATOMIC);
4708 } else {
4709 skb = skb_clone(orig_skb, GFP_ATOMIC);
4711 if (!skb)
4712 return;
4714 if (tsonly) {
4715 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4716 SKBTX_ANY_TSTAMP;
4717 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4720 if (hwtstamps)
4721 *skb_hwtstamps(skb) = *hwtstamps;
4722 else
4723 skb->tstamp = ktime_get_real();
4725 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4727 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4729 void skb_tstamp_tx(struct sk_buff *orig_skb,
4730 struct skb_shared_hwtstamps *hwtstamps)
4732 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4733 SCM_TSTAMP_SND);
4735 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4737 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4739 struct sock *sk = skb->sk;
4740 struct sock_exterr_skb *serr;
4741 int err = 1;
4743 skb->wifi_acked_valid = 1;
4744 skb->wifi_acked = acked;
4746 serr = SKB_EXT_ERR(skb);
4747 memset(serr, 0, sizeof(*serr));
4748 serr->ee.ee_errno = ENOMSG;
4749 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4751 /* Take a reference to prevent skb_orphan() from freeing the socket,
4752 * but only if the socket refcount is not zero.
4754 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4755 err = sock_queue_err_skb(sk, skb);
4756 sock_put(sk);
4758 if (err)
4759 kfree_skb(skb);
4761 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4764 * skb_partial_csum_set - set up and verify partial csum values for packet
4765 * @skb: the skb to set
4766 * @start: the number of bytes after skb->data to start checksumming.
4767 * @off: the offset from start to place the checksum.
4769 * For untrusted partially-checksummed packets, we need to make sure the values
4770 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4772 * This function checks and sets those values and skb->ip_summed: if this
4773 * returns false you should drop the packet.
4775 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4777 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4778 u32 csum_start = skb_headroom(skb) + (u32)start;
4780 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4781 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4782 start, off, skb_headroom(skb), skb_headlen(skb));
4783 return false;
4785 skb->ip_summed = CHECKSUM_PARTIAL;
4786 skb->csum_start = csum_start;
4787 skb->csum_offset = off;
4788 skb_set_transport_header(skb, start);
4789 return true;
4791 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4793 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4794 unsigned int max)
4796 if (skb_headlen(skb) >= len)
4797 return 0;
4799 /* If we need to pullup then pullup to the max, so we
4800 * won't need to do it again.
4802 if (max > skb->len)
4803 max = skb->len;
4805 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4806 return -ENOMEM;
4808 if (skb_headlen(skb) < len)
4809 return -EPROTO;
4811 return 0;
4814 #define MAX_TCP_HDR_LEN (15 * 4)
4816 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4817 typeof(IPPROTO_IP) proto,
4818 unsigned int off)
4820 int err;
4822 switch (proto) {
4823 case IPPROTO_TCP:
4824 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4825 off + MAX_TCP_HDR_LEN);
4826 if (!err && !skb_partial_csum_set(skb, off,
4827 offsetof(struct tcphdr,
4828 check)))
4829 err = -EPROTO;
4830 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4832 case IPPROTO_UDP:
4833 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4834 off + sizeof(struct udphdr));
4835 if (!err && !skb_partial_csum_set(skb, off,
4836 offsetof(struct udphdr,
4837 check)))
4838 err = -EPROTO;
4839 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4842 return ERR_PTR(-EPROTO);
4845 /* This value should be large enough to cover a tagged ethernet header plus
4846 * maximally sized IP and TCP or UDP headers.
4848 #define MAX_IP_HDR_LEN 128
4850 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4852 unsigned int off;
4853 bool fragment;
4854 __sum16 *csum;
4855 int err;
4857 fragment = false;
4859 err = skb_maybe_pull_tail(skb,
4860 sizeof(struct iphdr),
4861 MAX_IP_HDR_LEN);
4862 if (err < 0)
4863 goto out;
4865 if (ip_is_fragment(ip_hdr(skb)))
4866 fragment = true;
4868 off = ip_hdrlen(skb);
4870 err = -EPROTO;
4872 if (fragment)
4873 goto out;
4875 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4876 if (IS_ERR(csum))
4877 return PTR_ERR(csum);
4879 if (recalculate)
4880 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4881 ip_hdr(skb)->daddr,
4882 skb->len - off,
4883 ip_hdr(skb)->protocol, 0);
4884 err = 0;
4886 out:
4887 return err;
4890 /* This value should be large enough to cover a tagged ethernet header plus
4891 * an IPv6 header, all options, and a maximal TCP or UDP header.
4893 #define MAX_IPV6_HDR_LEN 256
4895 #define OPT_HDR(type, skb, off) \
4896 (type *)(skb_network_header(skb) + (off))
4898 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4900 int err;
4901 u8 nexthdr;
4902 unsigned int off;
4903 unsigned int len;
4904 bool fragment;
4905 bool done;
4906 __sum16 *csum;
4908 fragment = false;
4909 done = false;
4911 off = sizeof(struct ipv6hdr);
4913 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4914 if (err < 0)
4915 goto out;
4917 nexthdr = ipv6_hdr(skb)->nexthdr;
4919 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4920 while (off <= len && !done) {
4921 switch (nexthdr) {
4922 case IPPROTO_DSTOPTS:
4923 case IPPROTO_HOPOPTS:
4924 case IPPROTO_ROUTING: {
4925 struct ipv6_opt_hdr *hp;
4927 err = skb_maybe_pull_tail(skb,
4928 off +
4929 sizeof(struct ipv6_opt_hdr),
4930 MAX_IPV6_HDR_LEN);
4931 if (err < 0)
4932 goto out;
4934 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4935 nexthdr = hp->nexthdr;
4936 off += ipv6_optlen(hp);
4937 break;
4939 case IPPROTO_AH: {
4940 struct ip_auth_hdr *hp;
4942 err = skb_maybe_pull_tail(skb,
4943 off +
4944 sizeof(struct ip_auth_hdr),
4945 MAX_IPV6_HDR_LEN);
4946 if (err < 0)
4947 goto out;
4949 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4950 nexthdr = hp->nexthdr;
4951 off += ipv6_authlen(hp);
4952 break;
4954 case IPPROTO_FRAGMENT: {
4955 struct frag_hdr *hp;
4957 err = skb_maybe_pull_tail(skb,
4958 off +
4959 sizeof(struct frag_hdr),
4960 MAX_IPV6_HDR_LEN);
4961 if (err < 0)
4962 goto out;
4964 hp = OPT_HDR(struct frag_hdr, skb, off);
4966 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4967 fragment = true;
4969 nexthdr = hp->nexthdr;
4970 off += sizeof(struct frag_hdr);
4971 break;
4973 default:
4974 done = true;
4975 break;
4979 err = -EPROTO;
4981 if (!done || fragment)
4982 goto out;
4984 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4985 if (IS_ERR(csum))
4986 return PTR_ERR(csum);
4988 if (recalculate)
4989 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4990 &ipv6_hdr(skb)->daddr,
4991 skb->len - off, nexthdr, 0);
4992 err = 0;
4994 out:
4995 return err;
4999 * skb_checksum_setup - set up partial checksum offset
5000 * @skb: the skb to set up
5001 * @recalculate: if true the pseudo-header checksum will be recalculated
5003 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5005 int err;
5007 switch (skb->protocol) {
5008 case htons(ETH_P_IP):
5009 err = skb_checksum_setup_ipv4(skb, recalculate);
5010 break;
5012 case htons(ETH_P_IPV6):
5013 err = skb_checksum_setup_ipv6(skb, recalculate);
5014 break;
5016 default:
5017 err = -EPROTO;
5018 break;
5021 return err;
5023 EXPORT_SYMBOL(skb_checksum_setup);
5026 * skb_checksum_maybe_trim - maybe trims the given skb
5027 * @skb: the skb to check
5028 * @transport_len: the data length beyond the network header
5030 * Checks whether the given skb has data beyond the given transport length.
5031 * If so, returns a cloned skb trimmed to this transport length.
5032 * Otherwise returns the provided skb. Returns NULL in error cases
5033 * (e.g. transport_len exceeds skb length or out-of-memory).
5035 * Caller needs to set the skb transport header and free any returned skb if it
5036 * differs from the provided skb.
5038 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5039 unsigned int transport_len)
5041 struct sk_buff *skb_chk;
5042 unsigned int len = skb_transport_offset(skb) + transport_len;
5043 int ret;
5045 if (skb->len < len)
5046 return NULL;
5047 else if (skb->len == len)
5048 return skb;
5050 skb_chk = skb_clone(skb, GFP_ATOMIC);
5051 if (!skb_chk)
5052 return NULL;
5054 ret = pskb_trim_rcsum(skb_chk, len);
5055 if (ret) {
5056 kfree_skb(skb_chk);
5057 return NULL;
5060 return skb_chk;
5064 * skb_checksum_trimmed - validate checksum of an skb
5065 * @skb: the skb to check
5066 * @transport_len: the data length beyond the network header
5067 * @skb_chkf: checksum function to use
5069 * Applies the given checksum function skb_chkf to the provided skb.
5070 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5072 * If the skb has data beyond the given transport length, then a
5073 * trimmed & cloned skb is checked and returned.
5075 * Caller needs to set the skb transport header and free any returned skb if it
5076 * differs from the provided skb.
5078 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5079 unsigned int transport_len,
5080 __sum16(*skb_chkf)(struct sk_buff *skb))
5082 struct sk_buff *skb_chk;
5083 unsigned int offset = skb_transport_offset(skb);
5084 __sum16 ret;
5086 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5087 if (!skb_chk)
5088 goto err;
5090 if (!pskb_may_pull(skb_chk, offset))
5091 goto err;
5093 skb_pull_rcsum(skb_chk, offset);
5094 ret = skb_chkf(skb_chk);
5095 skb_push_rcsum(skb_chk, offset);
5097 if (ret)
5098 goto err;
5100 return skb_chk;
5102 err:
5103 if (skb_chk && skb_chk != skb)
5104 kfree_skb(skb_chk);
5106 return NULL;
5109 EXPORT_SYMBOL(skb_checksum_trimmed);
5111 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5113 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5114 skb->dev->name);
5116 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5118 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5120 if (head_stolen) {
5121 skb_release_head_state(skb);
5122 kmem_cache_free(skbuff_head_cache, skb);
5123 } else {
5124 __kfree_skb(skb);
5127 EXPORT_SYMBOL(kfree_skb_partial);
5130 * skb_try_coalesce - try to merge skb to prior one
5131 * @to: prior buffer
5132 * @from: buffer to add
5133 * @fragstolen: pointer to boolean
5134 * @delta_truesize: how much more was allocated than was requested
5136 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5137 bool *fragstolen, int *delta_truesize)
5139 struct skb_shared_info *to_shinfo, *from_shinfo;
5140 int i, delta, len = from->len;
5142 *fragstolen = false;
5144 if (skb_cloned(to))
5145 return false;
5147 if (len <= skb_tailroom(to)) {
5148 if (len)
5149 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5150 *delta_truesize = 0;
5151 return true;
5154 to_shinfo = skb_shinfo(to);
5155 from_shinfo = skb_shinfo(from);
5156 if (to_shinfo->frag_list || from_shinfo->frag_list)
5157 return false;
5158 if (skb_zcopy(to) || skb_zcopy(from))
5159 return false;
5161 if (skb_headlen(from) != 0) {
5162 struct page *page;
5163 unsigned int offset;
5165 if (to_shinfo->nr_frags +
5166 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5167 return false;
5169 if (skb_head_is_locked(from))
5170 return false;
5172 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5174 page = virt_to_head_page(from->head);
5175 offset = from->data - (unsigned char *)page_address(page);
5177 skb_fill_page_desc(to, to_shinfo->nr_frags,
5178 page, offset, skb_headlen(from));
5179 *fragstolen = true;
5180 } else {
5181 if (to_shinfo->nr_frags +
5182 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5183 return false;
5185 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5188 WARN_ON_ONCE(delta < len);
5190 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5191 from_shinfo->frags,
5192 from_shinfo->nr_frags * sizeof(skb_frag_t));
5193 to_shinfo->nr_frags += from_shinfo->nr_frags;
5195 if (!skb_cloned(from))
5196 from_shinfo->nr_frags = 0;
5198 /* if the skb is not cloned this does nothing
5199 * since we set nr_frags to 0.
5201 for (i = 0; i < from_shinfo->nr_frags; i++)
5202 __skb_frag_ref(&from_shinfo->frags[i]);
5204 to->truesize += delta;
5205 to->len += len;
5206 to->data_len += len;
5208 *delta_truesize = delta;
5209 return true;
5211 EXPORT_SYMBOL(skb_try_coalesce);
5214 * skb_scrub_packet - scrub an skb
5216 * @skb: buffer to clean
5217 * @xnet: packet is crossing netns
5219 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5220 * into/from a tunnel. Some information have to be cleared during these
5221 * operations.
5222 * skb_scrub_packet can also be used to clean a skb before injecting it in
5223 * another namespace (@xnet == true). We have to clear all information in the
5224 * skb that could impact namespace isolation.
5226 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5228 skb->pkt_type = PACKET_HOST;
5229 skb->skb_iif = 0;
5230 skb->ignore_df = 0;
5231 skb_dst_drop(skb);
5232 skb_ext_reset(skb);
5233 nf_reset_ct(skb);
5234 nf_reset_trace(skb);
5236 #ifdef CONFIG_NET_SWITCHDEV
5237 skb->offload_fwd_mark = 0;
5238 skb->offload_l3_fwd_mark = 0;
5239 #endif
5241 if (!xnet)
5242 return;
5244 ipvs_reset(skb);
5245 skb->mark = 0;
5246 skb->tstamp = 0;
5248 EXPORT_SYMBOL_GPL(skb_scrub_packet);
5251 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5253 * @skb: GSO skb
5255 * skb_gso_transport_seglen is used to determine the real size of the
5256 * individual segments, including Layer4 headers (TCP/UDP).
5258 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5260 static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5262 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5263 unsigned int thlen = 0;
5265 if (skb->encapsulation) {
5266 thlen = skb_inner_transport_header(skb) -
5267 skb_transport_header(skb);
5269 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5270 thlen += inner_tcp_hdrlen(skb);
5271 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5272 thlen = tcp_hdrlen(skb);
5273 } else if (unlikely(skb_is_gso_sctp(skb))) {
5274 thlen = sizeof(struct sctphdr);
5275 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5276 thlen = sizeof(struct udphdr);
5278 /* UFO sets gso_size to the size of the fragmentation
5279 * payload, i.e. the size of the L4 (UDP) header is already
5280 * accounted for.
5282 return thlen + shinfo->gso_size;
5286 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5288 * @skb: GSO skb
5290 * skb_gso_network_seglen is used to determine the real size of the
5291 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5293 * The MAC/L2 header is not accounted for.
5295 static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5297 unsigned int hdr_len = skb_transport_header(skb) -
5298 skb_network_header(skb);
5300 return hdr_len + skb_gso_transport_seglen(skb);
5304 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5306 * @skb: GSO skb
5308 * skb_gso_mac_seglen is used to determine the real size of the
5309 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5310 * headers (TCP/UDP).
5312 static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5314 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5316 return hdr_len + skb_gso_transport_seglen(skb);
5320 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5322 * There are a couple of instances where we have a GSO skb, and we
5323 * want to determine what size it would be after it is segmented.
5325 * We might want to check:
5326 * - L3+L4+payload size (e.g. IP forwarding)
5327 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5329 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5331 * @skb: GSO skb
5333 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5334 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5336 * @max_len: The maximum permissible length.
5338 * Returns true if the segmented length <= max length.
5340 static inline bool skb_gso_size_check(const struct sk_buff *skb,
5341 unsigned int seg_len,
5342 unsigned int max_len) {
5343 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5344 const struct sk_buff *iter;
5346 if (shinfo->gso_size != GSO_BY_FRAGS)
5347 return seg_len <= max_len;
5349 /* Undo this so we can re-use header sizes */
5350 seg_len -= GSO_BY_FRAGS;
5352 skb_walk_frags(skb, iter) {
5353 if (seg_len + skb_headlen(iter) > max_len)
5354 return false;
5357 return true;
5361 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5363 * @skb: GSO skb
5364 * @mtu: MTU to validate against
5366 * skb_gso_validate_network_len validates if a given skb will fit a
5367 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5368 * payload.
5370 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5372 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5374 EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5377 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5379 * @skb: GSO skb
5380 * @len: length to validate against
5382 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5383 * length once split, including L2, L3 and L4 headers and the payload.
5385 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5387 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5389 EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5391 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5393 int mac_len, meta_len;
5394 void *meta;
5396 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5397 kfree_skb(skb);
5398 return NULL;
5401 mac_len = skb->data - skb_mac_header(skb);
5402 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5403 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5404 mac_len - VLAN_HLEN - ETH_TLEN);
5407 meta_len = skb_metadata_len(skb);
5408 if (meta_len) {
5409 meta = skb_metadata_end(skb) - meta_len;
5410 memmove(meta + VLAN_HLEN, meta, meta_len);
5413 skb->mac_header += VLAN_HLEN;
5414 return skb;
5417 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5419 struct vlan_hdr *vhdr;
5420 u16 vlan_tci;
5422 if (unlikely(skb_vlan_tag_present(skb))) {
5423 /* vlan_tci is already set-up so leave this for another time */
5424 return skb;
5427 skb = skb_share_check(skb, GFP_ATOMIC);
5428 if (unlikely(!skb))
5429 goto err_free;
5430 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5431 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5432 goto err_free;
5434 vhdr = (struct vlan_hdr *)skb->data;
5435 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5436 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5438 skb_pull_rcsum(skb, VLAN_HLEN);
5439 vlan_set_encap_proto(skb, vhdr);
5441 skb = skb_reorder_vlan_header(skb);
5442 if (unlikely(!skb))
5443 goto err_free;
5445 skb_reset_network_header(skb);
5446 if (!skb_transport_header_was_set(skb))
5447 skb_reset_transport_header(skb);
5448 skb_reset_mac_len(skb);
5450 return skb;
5452 err_free:
5453 kfree_skb(skb);
5454 return NULL;
5456 EXPORT_SYMBOL(skb_vlan_untag);
5458 int skb_ensure_writable(struct sk_buff *skb, int write_len)
5460 if (!pskb_may_pull(skb, write_len))
5461 return -ENOMEM;
5463 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5464 return 0;
5466 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5468 EXPORT_SYMBOL(skb_ensure_writable);
5470 /* remove VLAN header from packet and update csum accordingly.
5471 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5473 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5475 struct vlan_hdr *vhdr;
5476 int offset = skb->data - skb_mac_header(skb);
5477 int err;
5479 if (WARN_ONCE(offset,
5480 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5481 offset)) {
5482 return -EINVAL;
5485 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5486 if (unlikely(err))
5487 return err;
5489 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5491 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5492 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5494 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5495 __skb_pull(skb, VLAN_HLEN);
5497 vlan_set_encap_proto(skb, vhdr);
5498 skb->mac_header += VLAN_HLEN;
5500 if (skb_network_offset(skb) < ETH_HLEN)
5501 skb_set_network_header(skb, ETH_HLEN);
5503 skb_reset_mac_len(skb);
5505 return err;
5507 EXPORT_SYMBOL(__skb_vlan_pop);
5509 /* Pop a vlan tag either from hwaccel or from payload.
5510 * Expects skb->data at mac header.
5512 int skb_vlan_pop(struct sk_buff *skb)
5514 u16 vlan_tci;
5515 __be16 vlan_proto;
5516 int err;
5518 if (likely(skb_vlan_tag_present(skb))) {
5519 __vlan_hwaccel_clear_tag(skb);
5520 } else {
5521 if (unlikely(!eth_type_vlan(skb->protocol)))
5522 return 0;
5524 err = __skb_vlan_pop(skb, &vlan_tci);
5525 if (err)
5526 return err;
5528 /* move next vlan tag to hw accel tag */
5529 if (likely(!eth_type_vlan(skb->protocol)))
5530 return 0;
5532 vlan_proto = skb->protocol;
5533 err = __skb_vlan_pop(skb, &vlan_tci);
5534 if (unlikely(err))
5535 return err;
5537 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5538 return 0;
5540 EXPORT_SYMBOL(skb_vlan_pop);
5542 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5543 * Expects skb->data at mac header.
5545 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5547 if (skb_vlan_tag_present(skb)) {
5548 int offset = skb->data - skb_mac_header(skb);
5549 int err;
5551 if (WARN_ONCE(offset,
5552 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5553 offset)) {
5554 return -EINVAL;
5557 err = __vlan_insert_tag(skb, skb->vlan_proto,
5558 skb_vlan_tag_get(skb));
5559 if (err)
5560 return err;
5562 skb->protocol = skb->vlan_proto;
5563 skb->mac_len += VLAN_HLEN;
5565 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5567 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5568 return 0;
5570 EXPORT_SYMBOL(skb_vlan_push);
5573 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5575 * @skb: Socket buffer to modify
5577 * Drop the Ethernet header of @skb.
5579 * Expects that skb->data points to the mac header and that no VLAN tags are
5580 * present.
5582 * Returns 0 on success, -errno otherwise.
5584 int skb_eth_pop(struct sk_buff *skb)
5586 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5587 skb_network_offset(skb) < ETH_HLEN)
5588 return -EPROTO;
5590 skb_pull_rcsum(skb, ETH_HLEN);
5591 skb_reset_mac_header(skb);
5592 skb_reset_mac_len(skb);
5594 return 0;
5596 EXPORT_SYMBOL(skb_eth_pop);
5599 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5601 * @skb: Socket buffer to modify
5602 * @dst: Destination MAC address of the new header
5603 * @src: Source MAC address of the new header
5605 * Prepend @skb with a new Ethernet header.
5607 * Expects that skb->data points to the mac header, which must be empty.
5609 * Returns 0 on success, -errno otherwise.
5611 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5612 const unsigned char *src)
5614 struct ethhdr *eth;
5615 int err;
5617 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5618 return -EPROTO;
5620 err = skb_cow_head(skb, sizeof(*eth));
5621 if (err < 0)
5622 return err;
5624 skb_push(skb, sizeof(*eth));
5625 skb_reset_mac_header(skb);
5626 skb_reset_mac_len(skb);
5628 eth = eth_hdr(skb);
5629 ether_addr_copy(eth->h_dest, dst);
5630 ether_addr_copy(eth->h_source, src);
5631 eth->h_proto = skb->protocol;
5633 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5635 return 0;
5637 EXPORT_SYMBOL(skb_eth_push);
5639 /* Update the ethertype of hdr and the skb csum value if required. */
5640 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5641 __be16 ethertype)
5643 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5644 __be16 diff[] = { ~hdr->h_proto, ethertype };
5646 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5649 hdr->h_proto = ethertype;
5653 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5654 * the packet
5656 * @skb: buffer
5657 * @mpls_lse: MPLS label stack entry to push
5658 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5659 * @mac_len: length of the MAC header
5660 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5661 * ethernet
5663 * Expects skb->data at mac header.
5665 * Returns 0 on success, -errno otherwise.
5667 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5668 int mac_len, bool ethernet)
5670 struct mpls_shim_hdr *lse;
5671 int err;
5673 if (unlikely(!eth_p_mpls(mpls_proto)))
5674 return -EINVAL;
5676 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5677 if (skb->encapsulation)
5678 return -EINVAL;
5680 err = skb_cow_head(skb, MPLS_HLEN);
5681 if (unlikely(err))
5682 return err;
5684 if (!skb->inner_protocol) {
5685 skb_set_inner_network_header(skb, skb_network_offset(skb));
5686 skb_set_inner_protocol(skb, skb->protocol);
5689 skb_push(skb, MPLS_HLEN);
5690 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5691 mac_len);
5692 skb_reset_mac_header(skb);
5693 skb_set_network_header(skb, mac_len);
5694 skb_reset_mac_len(skb);
5696 lse = mpls_hdr(skb);
5697 lse->label_stack_entry = mpls_lse;
5698 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5700 if (ethernet && mac_len >= ETH_HLEN)
5701 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5702 skb->protocol = mpls_proto;
5704 return 0;
5706 EXPORT_SYMBOL_GPL(skb_mpls_push);
5709 * skb_mpls_pop() - pop the outermost MPLS header
5711 * @skb: buffer
5712 * @next_proto: ethertype of header after popped MPLS header
5713 * @mac_len: length of the MAC header
5714 * @ethernet: flag to indicate if the packet is ethernet
5716 * Expects skb->data at mac header.
5718 * Returns 0 on success, -errno otherwise.
5720 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5721 bool ethernet)
5723 int err;
5725 if (unlikely(!eth_p_mpls(skb->protocol)))
5726 return 0;
5728 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5729 if (unlikely(err))
5730 return err;
5732 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5733 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5734 mac_len);
5736 __skb_pull(skb, MPLS_HLEN);
5737 skb_reset_mac_header(skb);
5738 skb_set_network_header(skb, mac_len);
5740 if (ethernet && mac_len >= ETH_HLEN) {
5741 struct ethhdr *hdr;
5743 /* use mpls_hdr() to get ethertype to account for VLANs. */
5744 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5745 skb_mod_eth_type(skb, hdr, next_proto);
5747 skb->protocol = next_proto;
5749 return 0;
5751 EXPORT_SYMBOL_GPL(skb_mpls_pop);
5754 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5756 * @skb: buffer
5757 * @mpls_lse: new MPLS label stack entry to update to
5759 * Expects skb->data at mac header.
5761 * Returns 0 on success, -errno otherwise.
5763 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5765 int err;
5767 if (unlikely(!eth_p_mpls(skb->protocol)))
5768 return -EINVAL;
5770 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5771 if (unlikely(err))
5772 return err;
5774 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5775 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5777 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5780 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5782 return 0;
5784 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5787 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5789 * @skb: buffer
5791 * Expects skb->data at mac header.
5793 * Returns 0 on success, -errno otherwise.
5795 int skb_mpls_dec_ttl(struct sk_buff *skb)
5797 u32 lse;
5798 u8 ttl;
5800 if (unlikely(!eth_p_mpls(skb->protocol)))
5801 return -EINVAL;
5803 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5804 return -ENOMEM;
5806 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5807 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5808 if (!--ttl)
5809 return -EINVAL;
5811 lse &= ~MPLS_LS_TTL_MASK;
5812 lse |= ttl << MPLS_LS_TTL_SHIFT;
5814 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5816 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5819 * alloc_skb_with_frags - allocate skb with page frags
5821 * @header_len: size of linear part
5822 * @data_len: needed length in frags
5823 * @max_page_order: max page order desired.
5824 * @errcode: pointer to error code if any
5825 * @gfp_mask: allocation mask
5827 * This can be used to allocate a paged skb, given a maximal order for frags.
5829 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5830 unsigned long data_len,
5831 int max_page_order,
5832 int *errcode,
5833 gfp_t gfp_mask)
5835 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5836 unsigned long chunk;
5837 struct sk_buff *skb;
5838 struct page *page;
5839 int i;
5841 *errcode = -EMSGSIZE;
5842 /* Note this test could be relaxed, if we succeed to allocate
5843 * high order pages...
5845 if (npages > MAX_SKB_FRAGS)
5846 return NULL;
5848 *errcode = -ENOBUFS;
5849 skb = alloc_skb(header_len, gfp_mask);
5850 if (!skb)
5851 return NULL;
5853 skb->truesize += npages << PAGE_SHIFT;
5855 for (i = 0; npages > 0; i++) {
5856 int order = max_page_order;
5858 while (order) {
5859 if (npages >= 1 << order) {
5860 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5861 __GFP_COMP |
5862 __GFP_NOWARN,
5863 order);
5864 if (page)
5865 goto fill_page;
5866 /* Do not retry other high order allocations */
5867 order = 1;
5868 max_page_order = 0;
5870 order--;
5872 page = alloc_page(gfp_mask);
5873 if (!page)
5874 goto failure;
5875 fill_page:
5876 chunk = min_t(unsigned long, data_len,
5877 PAGE_SIZE << order);
5878 skb_fill_page_desc(skb, i, page, 0, chunk);
5879 data_len -= chunk;
5880 npages -= 1 << order;
5882 return skb;
5884 failure:
5885 kfree_skb(skb);
5886 return NULL;
5888 EXPORT_SYMBOL(alloc_skb_with_frags);
5890 /* carve out the first off bytes from skb when off < headlen */
5891 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5892 const int headlen, gfp_t gfp_mask)
5894 int i;
5895 int size = skb_end_offset(skb);
5896 int new_hlen = headlen - off;
5897 u8 *data;
5899 size = SKB_DATA_ALIGN(size);
5901 if (skb_pfmemalloc(skb))
5902 gfp_mask |= __GFP_MEMALLOC;
5903 data = kmalloc_reserve(size +
5904 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5905 gfp_mask, NUMA_NO_NODE, NULL);
5906 if (!data)
5907 return -ENOMEM;
5909 size = SKB_WITH_OVERHEAD(ksize(data));
5911 /* Copy real data, and all frags */
5912 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5913 skb->len -= off;
5915 memcpy((struct skb_shared_info *)(data + size),
5916 skb_shinfo(skb),
5917 offsetof(struct skb_shared_info,
5918 frags[skb_shinfo(skb)->nr_frags]));
5919 if (skb_cloned(skb)) {
5920 /* drop the old head gracefully */
5921 if (skb_orphan_frags(skb, gfp_mask)) {
5922 kfree(data);
5923 return -ENOMEM;
5925 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5926 skb_frag_ref(skb, i);
5927 if (skb_has_frag_list(skb))
5928 skb_clone_fraglist(skb);
5929 skb_release_data(skb);
5930 } else {
5931 /* we can reuse existing recount- all we did was
5932 * relocate values
5934 skb_free_head(skb);
5937 skb->head = data;
5938 skb->data = data;
5939 skb->head_frag = 0;
5940 #ifdef NET_SKBUFF_DATA_USES_OFFSET
5941 skb->end = size;
5942 #else
5943 skb->end = skb->head + size;
5944 #endif
5945 skb_set_tail_pointer(skb, skb_headlen(skb));
5946 skb_headers_offset_update(skb, 0);
5947 skb->cloned = 0;
5948 skb->hdr_len = 0;
5949 skb->nohdr = 0;
5950 atomic_set(&skb_shinfo(skb)->dataref, 1);
5952 return 0;
5955 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5957 /* carve out the first eat bytes from skb's frag_list. May recurse into
5958 * pskb_carve()
5960 static int pskb_carve_frag_list(struct sk_buff *skb,
5961 struct skb_shared_info *shinfo, int eat,
5962 gfp_t gfp_mask)
5964 struct sk_buff *list = shinfo->frag_list;
5965 struct sk_buff *clone = NULL;
5966 struct sk_buff *insp = NULL;
5968 do {
5969 if (!list) {
5970 pr_err("Not enough bytes to eat. Want %d\n", eat);
5971 return -EFAULT;
5973 if (list->len <= eat) {
5974 /* Eaten as whole. */
5975 eat -= list->len;
5976 list = list->next;
5977 insp = list;
5978 } else {
5979 /* Eaten partially. */
5980 if (skb_shared(list)) {
5981 clone = skb_clone(list, gfp_mask);
5982 if (!clone)
5983 return -ENOMEM;
5984 insp = list->next;
5985 list = clone;
5986 } else {
5987 /* This may be pulled without problems. */
5988 insp = list;
5990 if (pskb_carve(list, eat, gfp_mask) < 0) {
5991 kfree_skb(clone);
5992 return -ENOMEM;
5994 break;
5996 } while (eat);
5998 /* Free pulled out fragments. */
5999 while ((list = shinfo->frag_list) != insp) {
6000 shinfo->frag_list = list->next;
6001 kfree_skb(list);
6003 /* And insert new clone at head. */
6004 if (clone) {
6005 clone->next = list;
6006 shinfo->frag_list = clone;
6008 return 0;
6011 /* carve off first len bytes from skb. Split line (off) is in the
6012 * non-linear part of skb
6014 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6015 int pos, gfp_t gfp_mask)
6017 int i, k = 0;
6018 int size = skb_end_offset(skb);
6019 u8 *data;
6020 const int nfrags = skb_shinfo(skb)->nr_frags;
6021 struct skb_shared_info *shinfo;
6023 size = SKB_DATA_ALIGN(size);
6025 if (skb_pfmemalloc(skb))
6026 gfp_mask |= __GFP_MEMALLOC;
6027 data = kmalloc_reserve(size +
6028 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6029 gfp_mask, NUMA_NO_NODE, NULL);
6030 if (!data)
6031 return -ENOMEM;
6033 size = SKB_WITH_OVERHEAD(ksize(data));
6035 memcpy((struct skb_shared_info *)(data + size),
6036 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6037 if (skb_orphan_frags(skb, gfp_mask)) {
6038 kfree(data);
6039 return -ENOMEM;
6041 shinfo = (struct skb_shared_info *)(data + size);
6042 for (i = 0; i < nfrags; i++) {
6043 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6045 if (pos + fsize > off) {
6046 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6048 if (pos < off) {
6049 /* Split frag.
6050 * We have two variants in this case:
6051 * 1. Move all the frag to the second
6052 * part, if it is possible. F.e.
6053 * this approach is mandatory for TUX,
6054 * where splitting is expensive.
6055 * 2. Split is accurately. We make this.
6057 skb_frag_off_add(&shinfo->frags[0], off - pos);
6058 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6060 skb_frag_ref(skb, i);
6061 k++;
6063 pos += fsize;
6065 shinfo->nr_frags = k;
6066 if (skb_has_frag_list(skb))
6067 skb_clone_fraglist(skb);
6069 /* split line is in frag list */
6070 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6071 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6072 if (skb_has_frag_list(skb))
6073 kfree_skb_list(skb_shinfo(skb)->frag_list);
6074 kfree(data);
6075 return -ENOMEM;
6077 skb_release_data(skb);
6079 skb->head = data;
6080 skb->head_frag = 0;
6081 skb->data = data;
6082 #ifdef NET_SKBUFF_DATA_USES_OFFSET
6083 skb->end = size;
6084 #else
6085 skb->end = skb->head + size;
6086 #endif
6087 skb_reset_tail_pointer(skb);
6088 skb_headers_offset_update(skb, 0);
6089 skb->cloned = 0;
6090 skb->hdr_len = 0;
6091 skb->nohdr = 0;
6092 skb->len -= off;
6093 skb->data_len = skb->len;
6094 atomic_set(&skb_shinfo(skb)->dataref, 1);
6095 return 0;
6098 /* remove len bytes from the beginning of the skb */
6099 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6101 int headlen = skb_headlen(skb);
6103 if (len < headlen)
6104 return pskb_carve_inside_header(skb, len, headlen, gfp);
6105 else
6106 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6109 /* Extract to_copy bytes starting at off from skb, and return this in
6110 * a new skb
6112 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6113 int to_copy, gfp_t gfp)
6115 struct sk_buff *clone = skb_clone(skb, gfp);
6117 if (!clone)
6118 return NULL;
6120 if (pskb_carve(clone, off, gfp) < 0 ||
6121 pskb_trim(clone, to_copy)) {
6122 kfree_skb(clone);
6123 return NULL;
6125 return clone;
6127 EXPORT_SYMBOL(pskb_extract);
6130 * skb_condense - try to get rid of fragments/frag_list if possible
6131 * @skb: buffer
6133 * Can be used to save memory before skb is added to a busy queue.
6134 * If packet has bytes in frags and enough tail room in skb->head,
6135 * pull all of them, so that we can free the frags right now and adjust
6136 * truesize.
6137 * Notes:
6138 * We do not reallocate skb->head thus can not fail.
6139 * Caller must re-evaluate skb->truesize if needed.
6141 void skb_condense(struct sk_buff *skb)
6143 if (skb->data_len) {
6144 if (skb->data_len > skb->end - skb->tail ||
6145 skb_cloned(skb))
6146 return;
6148 /* Nice, we can free page frag(s) right now */
6149 __pskb_pull_tail(skb, skb->data_len);
6151 /* At this point, skb->truesize might be over estimated,
6152 * because skb had a fragment, and fragments do not tell
6153 * their truesize.
6154 * When we pulled its content into skb->head, fragment
6155 * was freed, but __pskb_pull_tail() could not possibly
6156 * adjust skb->truesize, not knowing the frag truesize.
6158 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6161 #ifdef CONFIG_SKB_EXTENSIONS
6162 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6164 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6168 * __skb_ext_alloc - allocate a new skb extensions storage
6170 * @flags: See kmalloc().
6172 * Returns the newly allocated pointer. The pointer can later attached to a
6173 * skb via __skb_ext_set().
6174 * Note: caller must handle the skb_ext as an opaque data.
6176 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6178 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6180 if (new) {
6181 memset(new->offset, 0, sizeof(new->offset));
6182 refcount_set(&new->refcnt, 1);
6185 return new;
6188 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6189 unsigned int old_active)
6191 struct skb_ext *new;
6193 if (refcount_read(&old->refcnt) == 1)
6194 return old;
6196 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6197 if (!new)
6198 return NULL;
6200 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6201 refcount_set(&new->refcnt, 1);
6203 #ifdef CONFIG_XFRM
6204 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6205 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6206 unsigned int i;
6208 for (i = 0; i < sp->len; i++)
6209 xfrm_state_hold(sp->xvec[i]);
6211 #endif
6212 __skb_ext_put(old);
6213 return new;
6217 * __skb_ext_set - attach the specified extension storage to this skb
6218 * @skb: buffer
6219 * @id: extension id
6220 * @ext: extension storage previously allocated via __skb_ext_alloc()
6222 * Existing extensions, if any, are cleared.
6224 * Returns the pointer to the extension.
6226 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6227 struct skb_ext *ext)
6229 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6231 skb_ext_put(skb);
6232 newlen = newoff + skb_ext_type_len[id];
6233 ext->chunks = newlen;
6234 ext->offset[id] = newoff;
6235 skb->extensions = ext;
6236 skb->active_extensions = 1 << id;
6237 return skb_ext_get_ptr(ext, id);
6241 * skb_ext_add - allocate space for given extension, COW if needed
6242 * @skb: buffer
6243 * @id: extension to allocate space for
6245 * Allocates enough space for the given extension.
6246 * If the extension is already present, a pointer to that extension
6247 * is returned.
6249 * If the skb was cloned, COW applies and the returned memory can be
6250 * modified without changing the extension space of clones buffers.
6252 * Returns pointer to the extension or NULL on allocation failure.
6254 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6256 struct skb_ext *new, *old = NULL;
6257 unsigned int newlen, newoff;
6259 if (skb->active_extensions) {
6260 old = skb->extensions;
6262 new = skb_ext_maybe_cow(old, skb->active_extensions);
6263 if (!new)
6264 return NULL;
6266 if (__skb_ext_exist(new, id))
6267 goto set_active;
6269 newoff = new->chunks;
6270 } else {
6271 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6273 new = __skb_ext_alloc(GFP_ATOMIC);
6274 if (!new)
6275 return NULL;
6278 newlen = newoff + skb_ext_type_len[id];
6279 new->chunks = newlen;
6280 new->offset[id] = newoff;
6281 set_active:
6282 skb->extensions = new;
6283 skb->active_extensions |= 1 << id;
6284 return skb_ext_get_ptr(new, id);
6286 EXPORT_SYMBOL(skb_ext_add);
6288 #ifdef CONFIG_XFRM
6289 static void skb_ext_put_sp(struct sec_path *sp)
6291 unsigned int i;
6293 for (i = 0; i < sp->len; i++)
6294 xfrm_state_put(sp->xvec[i]);
6296 #endif
6298 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6300 struct skb_ext *ext = skb->extensions;
6302 skb->active_extensions &= ~(1 << id);
6303 if (skb->active_extensions == 0) {
6304 skb->extensions = NULL;
6305 __skb_ext_put(ext);
6306 #ifdef CONFIG_XFRM
6307 } else if (id == SKB_EXT_SEC_PATH &&
6308 refcount_read(&ext->refcnt) == 1) {
6309 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6311 skb_ext_put_sp(sp);
6312 sp->len = 0;
6313 #endif
6316 EXPORT_SYMBOL(__skb_ext_del);
6318 void __skb_ext_put(struct skb_ext *ext)
6320 /* If this is last clone, nothing can increment
6321 * it after check passes. Avoids one atomic op.
6323 if (refcount_read(&ext->refcnt) == 1)
6324 goto free_now;
6326 if (!refcount_dec_and_test(&ext->refcnt))
6327 return;
6328 free_now:
6329 #ifdef CONFIG_XFRM
6330 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6331 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6332 #endif
6334 kmem_cache_free(skbuff_ext_cache, ext);
6336 EXPORT_SYMBOL(__skb_ext_put);
6337 #endif /* CONFIG_SKB_EXTENSIONS */