Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * Routines having to do with the 'struct sk_buff' memory handlers.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 *
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
23 *
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).
29 */
30
31/*
32 * The functions in this file will not compile correctly with gcc 2.4.x
33 */
34
35#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36
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>
63
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>
72
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>
79
80#include "datagram.h"
81
82struct kmem_cache *skbuff_head_cache __ro_after_init;
83static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84#ifdef CONFIG_SKB_EXTENSIONS
85static struct kmem_cache *skbuff_ext_cache __ro_after_init;
86#endif
87int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88EXPORT_SYMBOL(sysctl_max_skb_frags);
89
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
96 *
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.
101 */
102static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103 const char msg[])
104{
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();
110}
111
112static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
118{
119 skb_panic(skb, sz, addr, __func__);
120}
121
122/*
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
128 */
129#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
131
132static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
134{
135 void *obj;
136 bool ret_pfmemalloc = false;
137
138 /*
139 * Try a regular allocation, when that fails and we're not entitled
140 * to the reserves, fail.
141 */
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;
147
148 /* Try again but now we are using pfmemalloc reserves */
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
151
152out:
153 if (pfmemalloc)
154 *pfmemalloc = ret_pfmemalloc;
155
156 return obj;
157}
158
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.
162 *
163 */
164
165/**
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
174 *
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.
178 *
179 * Buffers may only be allocated from interrupts using a @gfp_mask of
180 * %GFP_ATOMIC.
181 */
182struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183 int flags, int node)
184{
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
187 struct sk_buff *skb;
188 u8 *data;
189 bool pfmemalloc;
190
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
193
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
196
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);
202
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.
207 */
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.
216 */
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
219
220 /*
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!
224 */
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;
236
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);
241
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
244
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
246
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
249
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
251 }
252
253 skb_set_kcov_handle(skb, kcov_common_handle());
254
255out:
256 return skb;
257nodata:
258 kmem_cache_free(cache, skb);
259 skb = NULL;
260 goto out;
261}
262EXPORT_SYMBOL(__alloc_skb);
263
264/* Caller must provide SKB that is memset cleared */
265static struct sk_buff *__build_skb_around(struct sk_buff *skb,
266 void *data, unsigned int frag_size)
267{
268 struct skb_shared_info *shinfo;
269 unsigned int size = frag_size ? : ksize(data);
270
271 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
272
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;
282
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);
287
288 skb_set_kcov_handle(skb, kcov_common_handle());
289
290 return skb;
291}
292
293/**
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
297 *
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.
311 */
312struct sk_buff *__build_skb(void *data, unsigned int frag_size)
313{
314 struct sk_buff *skb;
315
316 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
317 if (unlikely(!skb))
318 return NULL;
319
320 memset(skb, 0, offsetof(struct sk_buff, tail));
321
322 return __build_skb_around(skb, data, frag_size);
323}
324
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()
329 */
330struct sk_buff *build_skb(void *data, unsigned int frag_size)
331{
332 struct sk_buff *skb = __build_skb(data, frag_size);
333
334 if (skb && frag_size) {
335 skb->head_frag = 1;
336 if (page_is_pfmemalloc(virt_to_head_page(data)))
337 skb->pfmemalloc = 1;
338 }
339 return skb;
340}
341EXPORT_SYMBOL(build_skb);
342
343/**
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
348 */
349struct sk_buff *build_skb_around(struct sk_buff *skb,
350 void *data, unsigned int frag_size)
351{
352 if (unlikely(!skb))
353 return NULL;
354
355 skb = __build_skb_around(skb, data, frag_size);
356
357 if (skb && frag_size) {
358 skb->head_frag = 1;
359 if (page_is_pfmemalloc(virt_to_head_page(data)))
360 skb->pfmemalloc = 1;
361 }
362 return skb;
363}
364EXPORT_SYMBOL(build_skb_around);
365
366#define NAPI_SKB_CACHE_SIZE 64
367
368struct napi_alloc_cache {
369 struct page_frag_cache page;
370 unsigned int skb_count;
371 void *skb_cache[NAPI_SKB_CACHE_SIZE];
372};
373
374static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
375static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
376
377static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
378{
379 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
380
381 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
382}
383
384void *napi_alloc_frag(unsigned int fragsz)
385{
386 fragsz = SKB_DATA_ALIGN(fragsz);
387
388 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
389}
390EXPORT_SYMBOL(napi_alloc_frag);
391
392/**
393 * netdev_alloc_frag - allocate a page fragment
394 * @fragsz: fragment size
395 *
396 * Allocates a frag from a page for receive buffer.
397 * Uses GFP_ATOMIC allocations.
398 */
399void *netdev_alloc_frag(unsigned int fragsz)
400{
401 struct page_frag_cache *nc;
402 void *data;
403
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();
412 }
413 return data;
414}
415EXPORT_SYMBOL(netdev_alloc_frag);
416
417/**
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
422 *
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.
427 *
428 * %NULL is returned if there is no free memory.
429 */
430struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
431 gfp_t gfp_mask)
432{
433 struct page_frag_cache *nc;
434 struct sk_buff *skb;
435 bool pfmemalloc;
436 void *data;
437
438 len += NET_SKB_PAD;
439
440 /* If requested length is either too small or too big,
441 * we use kmalloc() for skb->head allocation.
442 */
443 if (len <= SKB_WITH_OVERHEAD(1024) ||
444 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
445 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
446 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
447 if (!skb)
448 goto skb_fail;
449 goto skb_success;
450 }
451
452 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
453 len = SKB_DATA_ALIGN(len);
454
455 if (sk_memalloc_socks())
456 gfp_mask |= __GFP_MEMALLOC;
457
458 if (in_irq() || irqs_disabled()) {
459 nc = this_cpu_ptr(&netdev_alloc_cache);
460 data = page_frag_alloc(nc, len, gfp_mask);
461 pfmemalloc = nc->pfmemalloc;
462 } else {
463 local_bh_disable();
464 nc = this_cpu_ptr(&napi_alloc_cache.page);
465 data = page_frag_alloc(nc, len, gfp_mask);
466 pfmemalloc = nc->pfmemalloc;
467 local_bh_enable();
468 }
469
470 if (unlikely(!data))
471 return NULL;
472
473 skb = __build_skb(data, len);
474 if (unlikely(!skb)) {
475 skb_free_frag(data);
476 return NULL;
477 }
478
479 if (pfmemalloc)
480 skb->pfmemalloc = 1;
481 skb->head_frag = 1;
482
483skb_success:
484 skb_reserve(skb, NET_SKB_PAD);
485 skb->dev = dev;
486
487skb_fail:
488 return skb;
489}
490EXPORT_SYMBOL(__netdev_alloc_skb);
491
492/**
493 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
494 * @napi: napi instance this buffer was allocated for
495 * @len: length to allocate
496 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
497 *
498 * Allocate a new sk_buff for use in NAPI receive. This buffer will
499 * attempt to allocate the head from a special reserved region used
500 * only for NAPI Rx allocation. By doing this we can save several
501 * CPU cycles by avoiding having to disable and re-enable IRQs.
502 *
503 * %NULL is returned if there is no free memory.
504 */
505struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
506 gfp_t gfp_mask)
507{
508 struct napi_alloc_cache *nc;
509 struct sk_buff *skb;
510 void *data;
511
512 len += NET_SKB_PAD + NET_IP_ALIGN;
513
514 /* If requested length is either too small or too big,
515 * we use kmalloc() for skb->head allocation.
516 */
517 if (len <= SKB_WITH_OVERHEAD(1024) ||
518 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
519 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
520 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
521 if (!skb)
522 goto skb_fail;
523 goto skb_success;
524 }
525
526 nc = this_cpu_ptr(&napi_alloc_cache);
527 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
528 len = SKB_DATA_ALIGN(len);
529
530 if (sk_memalloc_socks())
531 gfp_mask |= __GFP_MEMALLOC;
532
533 data = page_frag_alloc(&nc->page, len, gfp_mask);
534 if (unlikely(!data))
535 return NULL;
536
537 skb = __build_skb(data, len);
538 if (unlikely(!skb)) {
539 skb_free_frag(data);
540 return NULL;
541 }
542
543 if (nc->page.pfmemalloc)
544 skb->pfmemalloc = 1;
545 skb->head_frag = 1;
546
547skb_success:
548 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
549 skb->dev = napi->dev;
550
551skb_fail:
552 return skb;
553}
554EXPORT_SYMBOL(__napi_alloc_skb);
555
556void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
557 int size, unsigned int truesize)
558{
559 skb_fill_page_desc(skb, i, page, off, size);
560 skb->len += size;
561 skb->data_len += size;
562 skb->truesize += truesize;
563}
564EXPORT_SYMBOL(skb_add_rx_frag);
565
566void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
567 unsigned int truesize)
568{
569 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
570
571 skb_frag_size_add(frag, size);
572 skb->len += size;
573 skb->data_len += size;
574 skb->truesize += truesize;
575}
576EXPORT_SYMBOL(skb_coalesce_rx_frag);
577
578static void skb_drop_list(struct sk_buff **listp)
579{
580 kfree_skb_list(*listp);
581 *listp = NULL;
582}
583
584static inline void skb_drop_fraglist(struct sk_buff *skb)
585{
586 skb_drop_list(&skb_shinfo(skb)->frag_list);
587}
588
589static void skb_clone_fraglist(struct sk_buff *skb)
590{
591 struct sk_buff *list;
592
593 skb_walk_frags(skb, list)
594 skb_get(list);
595}
596
597static void skb_free_head(struct sk_buff *skb)
598{
599 unsigned char *head = skb->head;
600
601 if (skb->head_frag)
602 skb_free_frag(head);
603 else
604 kfree(head);
605}
606
607static void skb_release_data(struct sk_buff *skb)
608{
609 struct skb_shared_info *shinfo = skb_shinfo(skb);
610 int i;
611
612 if (skb->cloned &&
613 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
614 &shinfo->dataref))
615 return;
616
617 for (i = 0; i < shinfo->nr_frags; i++)
618 __skb_frag_unref(&shinfo->frags[i]);
619
620 if (shinfo->frag_list)
621 kfree_skb_list(shinfo->frag_list);
622
623 skb_zcopy_clear(skb, true);
624 skb_free_head(skb);
625}
626
627/*
628 * Free an skbuff by memory without cleaning the state.
629 */
630static void kfree_skbmem(struct sk_buff *skb)
631{
632 struct sk_buff_fclones *fclones;
633
634 switch (skb->fclone) {
635 case SKB_FCLONE_UNAVAILABLE:
636 kmem_cache_free(skbuff_head_cache, skb);
637 return;
638
639 case SKB_FCLONE_ORIG:
640 fclones = container_of(skb, struct sk_buff_fclones, skb1);
641
642 /* We usually free the clone (TX completion) before original skb
643 * This test would have no chance to be true for the clone,
644 * while here, branch prediction will be good.
645 */
646 if (refcount_read(&fclones->fclone_ref) == 1)
647 goto fastpath;
648 break;
649
650 default: /* SKB_FCLONE_CLONE */
651 fclones = container_of(skb, struct sk_buff_fclones, skb2);
652 break;
653 }
654 if (!refcount_dec_and_test(&fclones->fclone_ref))
655 return;
656fastpath:
657 kmem_cache_free(skbuff_fclone_cache, fclones);
658}
659
660void skb_release_head_state(struct sk_buff *skb)
661{
662 skb_dst_drop(skb);
663 if (skb->destructor) {
664 WARN_ON(in_irq());
665 skb->destructor(skb);
666 }
667#if IS_ENABLED(CONFIG_NF_CONNTRACK)
668 nf_conntrack_put(skb_nfct(skb));
669#endif
670 skb_ext_put(skb);
671}
672
673/* Free everything but the sk_buff shell. */
674static void skb_release_all(struct sk_buff *skb)
675{
676 skb_release_head_state(skb);
677 if (likely(skb->head))
678 skb_release_data(skb);
679}
680
681/**
682 * __kfree_skb - private function
683 * @skb: buffer
684 *
685 * Free an sk_buff. Release anything attached to the buffer.
686 * Clean the state. This is an internal helper function. Users should
687 * always call kfree_skb
688 */
689
690void __kfree_skb(struct sk_buff *skb)
691{
692 skb_release_all(skb);
693 kfree_skbmem(skb);
694}
695EXPORT_SYMBOL(__kfree_skb);
696
697/**
698 * kfree_skb - free an sk_buff
699 * @skb: buffer to free
700 *
701 * Drop a reference to the buffer and free it if the usage count has
702 * hit zero.
703 */
704void kfree_skb(struct sk_buff *skb)
705{
706 if (!skb_unref(skb))
707 return;
708
709 trace_kfree_skb(skb, __builtin_return_address(0));
710 __kfree_skb(skb);
711}
712EXPORT_SYMBOL(kfree_skb);
713
714void kfree_skb_list(struct sk_buff *segs)
715{
716 while (segs) {
717 struct sk_buff *next = segs->next;
718
719 kfree_skb(segs);
720 segs = next;
721 }
722}
723EXPORT_SYMBOL(kfree_skb_list);
724
725/* Dump skb information and contents.
726 *
727 * Must only be called from net_ratelimit()-ed paths.
728 *
729 * Dumps whole packets if full_pkt, only headers otherwise.
730 */
731void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
732{
733 struct skb_shared_info *sh = skb_shinfo(skb);
734 struct net_device *dev = skb->dev;
735 struct sock *sk = skb->sk;
736 struct sk_buff *list_skb;
737 bool has_mac, has_trans;
738 int headroom, tailroom;
739 int i, len, seg_len;
740
741 if (full_pkt)
742 len = skb->len;
743 else
744 len = min_t(int, skb->len, MAX_HEADER + 128);
745
746 headroom = skb_headroom(skb);
747 tailroom = skb_tailroom(skb);
748
749 has_mac = skb_mac_header_was_set(skb);
750 has_trans = skb_transport_header_was_set(skb);
751
752 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
753 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
754 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
755 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
756 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
757 level, skb->len, headroom, skb_headlen(skb), tailroom,
758 has_mac ? skb->mac_header : -1,
759 has_mac ? skb_mac_header_len(skb) : -1,
760 skb->network_header,
761 has_trans ? skb_network_header_len(skb) : -1,
762 has_trans ? skb->transport_header : -1,
763 sh->tx_flags, sh->nr_frags,
764 sh->gso_size, sh->gso_type, sh->gso_segs,
765 skb->csum, skb->ip_summed, skb->csum_complete_sw,
766 skb->csum_valid, skb->csum_level,
767 skb->hash, skb->sw_hash, skb->l4_hash,
768 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
769
770 if (dev)
771 printk("%sdev name=%s feat=0x%pNF\n",
772 level, dev->name, &dev->features);
773 if (sk)
774 printk("%ssk family=%hu type=%u proto=%u\n",
775 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
776
777 if (full_pkt && headroom)
778 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb->head, headroom, false);
780
781 seg_len = min_t(int, skb_headlen(skb), len);
782 if (seg_len)
783 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
784 16, 1, skb->data, seg_len, false);
785 len -= seg_len;
786
787 if (full_pkt && tailroom)
788 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
789 16, 1, skb_tail_pointer(skb), tailroom, false);
790
791 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
792 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
793 u32 p_off, p_len, copied;
794 struct page *p;
795 u8 *vaddr;
796
797 skb_frag_foreach_page(frag, skb_frag_off(frag),
798 skb_frag_size(frag), p, p_off, p_len,
799 copied) {
800 seg_len = min_t(int, p_len, len);
801 vaddr = kmap_atomic(p);
802 print_hex_dump(level, "skb frag: ",
803 DUMP_PREFIX_OFFSET,
804 16, 1, vaddr + p_off, seg_len, false);
805 kunmap_atomic(vaddr);
806 len -= seg_len;
807 if (!len)
808 break;
809 }
810 }
811
812 if (full_pkt && skb_has_frag_list(skb)) {
813 printk("skb fraglist:\n");
814 skb_walk_frags(skb, list_skb)
815 skb_dump(level, list_skb, true);
816 }
817}
818EXPORT_SYMBOL(skb_dump);
819
820/**
821 * skb_tx_error - report an sk_buff xmit error
822 * @skb: buffer that triggered an error
823 *
824 * Report xmit error if a device callback is tracking this skb.
825 * skb must be freed afterwards.
826 */
827void skb_tx_error(struct sk_buff *skb)
828{
829 skb_zcopy_clear(skb, true);
830}
831EXPORT_SYMBOL(skb_tx_error);
832
833#ifdef CONFIG_TRACEPOINTS
834/**
835 * consume_skb - free an skbuff
836 * @skb: buffer to free
837 *
838 * Drop a ref to the buffer and free it if the usage count has hit zero
839 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
840 * is being dropped after a failure and notes that
841 */
842void consume_skb(struct sk_buff *skb)
843{
844 if (!skb_unref(skb))
845 return;
846
847 trace_consume_skb(skb);
848 __kfree_skb(skb);
849}
850EXPORT_SYMBOL(consume_skb);
851#endif
852
853/**
854 * __consume_stateless_skb - free an skbuff, assuming it is stateless
855 * @skb: buffer to free
856 *
857 * Alike consume_skb(), but this variant assumes that this is the last
858 * skb reference and all the head states have been already dropped
859 */
860void __consume_stateless_skb(struct sk_buff *skb)
861{
862 trace_consume_skb(skb);
863 skb_release_data(skb);
864 kfree_skbmem(skb);
865}
866
867void __kfree_skb_flush(void)
868{
869 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
870
871 /* flush skb_cache if containing objects */
872 if (nc->skb_count) {
873 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
874 nc->skb_cache);
875 nc->skb_count = 0;
876 }
877}
878
879static inline void _kfree_skb_defer(struct sk_buff *skb)
880{
881 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
882
883 /* drop skb->head and call any destructors for packet */
884 skb_release_all(skb);
885
886 /* record skb to CPU local list */
887 nc->skb_cache[nc->skb_count++] = skb;
888
889#ifdef CONFIG_SLUB
890 /* SLUB writes into objects when freeing */
891 prefetchw(skb);
892#endif
893
894 /* flush skb_cache if it is filled */
895 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
896 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
897 nc->skb_cache);
898 nc->skb_count = 0;
899 }
900}
901void __kfree_skb_defer(struct sk_buff *skb)
902{
903 _kfree_skb_defer(skb);
904}
905
906void napi_consume_skb(struct sk_buff *skb, int budget)
907{
908 /* Zero budget indicate non-NAPI context called us, like netpoll */
909 if (unlikely(!budget)) {
910 dev_consume_skb_any(skb);
911 return;
912 }
913
914 lockdep_assert_in_softirq();
915
916 if (!skb_unref(skb))
917 return;
918
919 /* if reaching here SKB is ready to free */
920 trace_consume_skb(skb);
921
922 /* if SKB is a clone, don't handle this case */
923 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
924 __kfree_skb(skb);
925 return;
926 }
927
928 _kfree_skb_defer(skb);
929}
930EXPORT_SYMBOL(napi_consume_skb);
931
932/* Make sure a field is enclosed inside headers_start/headers_end section */
933#define CHECK_SKB_FIELD(field) \
934 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
935 offsetof(struct sk_buff, headers_start)); \
936 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
937 offsetof(struct sk_buff, headers_end)); \
938
939static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
940{
941 new->tstamp = old->tstamp;
942 /* We do not copy old->sk */
943 new->dev = old->dev;
944 memcpy(new->cb, old->cb, sizeof(old->cb));
945 skb_dst_copy(new, old);
946 __skb_ext_copy(new, old);
947 __nf_copy(new, old, false);
948
949 /* Note : this field could be in headers_start/headers_end section
950 * It is not yet because we do not want to have a 16 bit hole
951 */
952 new->queue_mapping = old->queue_mapping;
953
954 memcpy(&new->headers_start, &old->headers_start,
955 offsetof(struct sk_buff, headers_end) -
956 offsetof(struct sk_buff, headers_start));
957 CHECK_SKB_FIELD(protocol);
958 CHECK_SKB_FIELD(csum);
959 CHECK_SKB_FIELD(hash);
960 CHECK_SKB_FIELD(priority);
961 CHECK_SKB_FIELD(skb_iif);
962 CHECK_SKB_FIELD(vlan_proto);
963 CHECK_SKB_FIELD(vlan_tci);
964 CHECK_SKB_FIELD(transport_header);
965 CHECK_SKB_FIELD(network_header);
966 CHECK_SKB_FIELD(mac_header);
967 CHECK_SKB_FIELD(inner_protocol);
968 CHECK_SKB_FIELD(inner_transport_header);
969 CHECK_SKB_FIELD(inner_network_header);
970 CHECK_SKB_FIELD(inner_mac_header);
971 CHECK_SKB_FIELD(mark);
972#ifdef CONFIG_NETWORK_SECMARK
973 CHECK_SKB_FIELD(secmark);
974#endif
975#ifdef CONFIG_NET_RX_BUSY_POLL
976 CHECK_SKB_FIELD(napi_id);
977#endif
978#ifdef CONFIG_XPS
979 CHECK_SKB_FIELD(sender_cpu);
980#endif
981#ifdef CONFIG_NET_SCHED
982 CHECK_SKB_FIELD(tc_index);
983#endif
984
985}
986
987/*
988 * You should not add any new code to this function. Add it to
989 * __copy_skb_header above instead.
990 */
991static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
992{
993#define C(x) n->x = skb->x
994
995 n->next = n->prev = NULL;
996 n->sk = NULL;
997 __copy_skb_header(n, skb);
998
999 C(len);
1000 C(data_len);
1001 C(mac_len);
1002 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1003 n->cloned = 1;
1004 n->nohdr = 0;
1005 n->peeked = 0;
1006 C(pfmemalloc);
1007 n->destructor = NULL;
1008 C(tail);
1009 C(end);
1010 C(head);
1011 C(head_frag);
1012 C(data);
1013 C(truesize);
1014 refcount_set(&n->users, 1);
1015
1016 atomic_inc(&(skb_shinfo(skb)->dataref));
1017 skb->cloned = 1;
1018
1019 return n;
1020#undef C
1021}
1022
1023/**
1024 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1025 * @first: first sk_buff of the msg
1026 */
1027struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1028{
1029 struct sk_buff *n;
1030
1031 n = alloc_skb(0, GFP_ATOMIC);
1032 if (!n)
1033 return NULL;
1034
1035 n->len = first->len;
1036 n->data_len = first->len;
1037 n->truesize = first->truesize;
1038
1039 skb_shinfo(n)->frag_list = first;
1040
1041 __copy_skb_header(n, first);
1042 n->destructor = NULL;
1043
1044 return n;
1045}
1046EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1047
1048/**
1049 * skb_morph - morph one skb into another
1050 * @dst: the skb to receive the contents
1051 * @src: the skb to supply the contents
1052 *
1053 * This is identical to skb_clone except that the target skb is
1054 * supplied by the user.
1055 *
1056 * The target skb is returned upon exit.
1057 */
1058struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1059{
1060 skb_release_all(dst);
1061 return __skb_clone(dst, src);
1062}
1063EXPORT_SYMBOL_GPL(skb_morph);
1064
1065int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1066{
1067 unsigned long max_pg, num_pg, new_pg, old_pg;
1068 struct user_struct *user;
1069
1070 if (capable(CAP_IPC_LOCK) || !size)
1071 return 0;
1072
1073 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1074 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1075 user = mmp->user ? : current_user();
1076
1077 do {
1078 old_pg = atomic_long_read(&user->locked_vm);
1079 new_pg = old_pg + num_pg;
1080 if (new_pg > max_pg)
1081 return -ENOBUFS;
1082 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1083 old_pg);
1084
1085 if (!mmp->user) {
1086 mmp->user = get_uid(user);
1087 mmp->num_pg = num_pg;
1088 } else {
1089 mmp->num_pg += num_pg;
1090 }
1091
1092 return 0;
1093}
1094EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1095
1096void mm_unaccount_pinned_pages(struct mmpin *mmp)
1097{
1098 if (mmp->user) {
1099 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1100 free_uid(mmp->user);
1101 }
1102}
1103EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1104
1105struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1106{
1107 struct ubuf_info *uarg;
1108 struct sk_buff *skb;
1109
1110 WARN_ON_ONCE(!in_task());
1111
1112 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1113 if (!skb)
1114 return NULL;
1115
1116 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1117 uarg = (void *)skb->cb;
1118 uarg->mmp.user = NULL;
1119
1120 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1121 kfree_skb(skb);
1122 return NULL;
1123 }
1124
1125 uarg->callback = sock_zerocopy_callback;
1126 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1127 uarg->len = 1;
1128 uarg->bytelen = size;
1129 uarg->zerocopy = 1;
1130 refcount_set(&uarg->refcnt, 1);
1131 sock_hold(sk);
1132
1133 return uarg;
1134}
1135EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1136
1137static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1138{
1139 return container_of((void *)uarg, struct sk_buff, cb);
1140}
1141
1142struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1143 struct ubuf_info *uarg)
1144{
1145 if (uarg) {
1146 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1147 u32 bytelen, next;
1148
1149 /* realloc only when socket is locked (TCP, UDP cork),
1150 * so uarg->len and sk_zckey access is serialized
1151 */
1152 if (!sock_owned_by_user(sk)) {
1153 WARN_ON_ONCE(1);
1154 return NULL;
1155 }
1156
1157 bytelen = uarg->bytelen + size;
1158 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1159 /* TCP can create new skb to attach new uarg */
1160 if (sk->sk_type == SOCK_STREAM)
1161 goto new_alloc;
1162 return NULL;
1163 }
1164
1165 next = (u32)atomic_read(&sk->sk_zckey);
1166 if ((u32)(uarg->id + uarg->len) == next) {
1167 if (mm_account_pinned_pages(&uarg->mmp, size))
1168 return NULL;
1169 uarg->len++;
1170 uarg->bytelen = bytelen;
1171 atomic_set(&sk->sk_zckey, ++next);
1172
1173 /* no extra ref when appending to datagram (MSG_MORE) */
1174 if (sk->sk_type == SOCK_STREAM)
1175 sock_zerocopy_get(uarg);
1176
1177 return uarg;
1178 }
1179 }
1180
1181new_alloc:
1182 return sock_zerocopy_alloc(sk, size);
1183}
1184EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1185
1186static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1187{
1188 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1189 u32 old_lo, old_hi;
1190 u64 sum_len;
1191
1192 old_lo = serr->ee.ee_info;
1193 old_hi = serr->ee.ee_data;
1194 sum_len = old_hi - old_lo + 1ULL + len;
1195
1196 if (sum_len >= (1ULL << 32))
1197 return false;
1198
1199 if (lo != old_hi + 1)
1200 return false;
1201
1202 serr->ee.ee_data += len;
1203 return true;
1204}
1205
1206void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1207{
1208 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1209 struct sock_exterr_skb *serr;
1210 struct sock *sk = skb->sk;
1211 struct sk_buff_head *q;
1212 unsigned long flags;
1213 u32 lo, hi;
1214 u16 len;
1215
1216 mm_unaccount_pinned_pages(&uarg->mmp);
1217
1218 /* if !len, there was only 1 call, and it was aborted
1219 * so do not queue a completion notification
1220 */
1221 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1222 goto release;
1223
1224 len = uarg->len;
1225 lo = uarg->id;
1226 hi = uarg->id + len - 1;
1227
1228 serr = SKB_EXT_ERR(skb);
1229 memset(serr, 0, sizeof(*serr));
1230 serr->ee.ee_errno = 0;
1231 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1232 serr->ee.ee_data = hi;
1233 serr->ee.ee_info = lo;
1234 if (!success)
1235 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1236
1237 q = &sk->sk_error_queue;
1238 spin_lock_irqsave(&q->lock, flags);
1239 tail = skb_peek_tail(q);
1240 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1241 !skb_zerocopy_notify_extend(tail, lo, len)) {
1242 __skb_queue_tail(q, skb);
1243 skb = NULL;
1244 }
1245 spin_unlock_irqrestore(&q->lock, flags);
1246
1247 sk->sk_error_report(sk);
1248
1249release:
1250 consume_skb(skb);
1251 sock_put(sk);
1252}
1253EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1254
1255void sock_zerocopy_put(struct ubuf_info *uarg)
1256{
1257 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1258 if (uarg->callback)
1259 uarg->callback(uarg, uarg->zerocopy);
1260 else
1261 consume_skb(skb_from_uarg(uarg));
1262 }
1263}
1264EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1265
1266void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1267{
1268 if (uarg) {
1269 struct sock *sk = skb_from_uarg(uarg)->sk;
1270
1271 atomic_dec(&sk->sk_zckey);
1272 uarg->len--;
1273
1274 if (have_uref)
1275 sock_zerocopy_put(uarg);
1276 }
1277}
1278EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1279
1280int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1281{
1282 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1283}
1284EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1285
1286int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1287 struct msghdr *msg, int len,
1288 struct ubuf_info *uarg)
1289{
1290 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1291 struct iov_iter orig_iter = msg->msg_iter;
1292 int err, orig_len = skb->len;
1293
1294 /* An skb can only point to one uarg. This edge case happens when
1295 * TCP appends to an skb, but zerocopy_realloc triggered a new alloc.
1296 */
1297 if (orig_uarg && uarg != orig_uarg)
1298 return -EEXIST;
1299
1300 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1301 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1302 struct sock *save_sk = skb->sk;
1303
1304 /* Streams do not free skb on error. Reset to prev state. */
1305 msg->msg_iter = orig_iter;
1306 skb->sk = sk;
1307 ___pskb_trim(skb, orig_len);
1308 skb->sk = save_sk;
1309 return err;
1310 }
1311
1312 skb_zcopy_set(skb, uarg, NULL);
1313 return skb->len - orig_len;
1314}
1315EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1316
1317static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1318 gfp_t gfp_mask)
1319{
1320 if (skb_zcopy(orig)) {
1321 if (skb_zcopy(nskb)) {
1322 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1323 if (!gfp_mask) {
1324 WARN_ON_ONCE(1);
1325 return -ENOMEM;
1326 }
1327 if (skb_uarg(nskb) == skb_uarg(orig))
1328 return 0;
1329 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1330 return -EIO;
1331 }
1332 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1333 }
1334 return 0;
1335}
1336
1337/**
1338 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1339 * @skb: the skb to modify
1340 * @gfp_mask: allocation priority
1341 *
1342 * This must be called on SKBTX_DEV_ZEROCOPY skb.
1343 * It will copy all frags into kernel and drop the reference
1344 * to userspace pages.
1345 *
1346 * If this function is called from an interrupt gfp_mask() must be
1347 * %GFP_ATOMIC.
1348 *
1349 * Returns 0 on success or a negative error code on failure
1350 * to allocate kernel memory to copy to.
1351 */
1352int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1353{
1354 int num_frags = skb_shinfo(skb)->nr_frags;
1355 struct page *page, *head = NULL;
1356 int i, new_frags;
1357 u32 d_off;
1358
1359 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1360 return -EINVAL;
1361
1362 if (!num_frags)
1363 goto release;
1364
1365 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1366 for (i = 0; i < new_frags; i++) {
1367 page = alloc_page(gfp_mask);
1368 if (!page) {
1369 while (head) {
1370 struct page *next = (struct page *)page_private(head);
1371 put_page(head);
1372 head = next;
1373 }
1374 return -ENOMEM;
1375 }
1376 set_page_private(page, (unsigned long)head);
1377 head = page;
1378 }
1379
1380 page = head;
1381 d_off = 0;
1382 for (i = 0; i < num_frags; i++) {
1383 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1384 u32 p_off, p_len, copied;
1385 struct page *p;
1386 u8 *vaddr;
1387
1388 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1389 p, p_off, p_len, copied) {
1390 u32 copy, done = 0;
1391 vaddr = kmap_atomic(p);
1392
1393 while (done < p_len) {
1394 if (d_off == PAGE_SIZE) {
1395 d_off = 0;
1396 page = (struct page *)page_private(page);
1397 }
1398 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1399 memcpy(page_address(page) + d_off,
1400 vaddr + p_off + done, copy);
1401 done += copy;
1402 d_off += copy;
1403 }
1404 kunmap_atomic(vaddr);
1405 }
1406 }
1407
1408 /* skb frags release userspace buffers */
1409 for (i = 0; i < num_frags; i++)
1410 skb_frag_unref(skb, i);
1411
1412 /* skb frags point to kernel buffers */
1413 for (i = 0; i < new_frags - 1; i++) {
1414 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1415 head = (struct page *)page_private(head);
1416 }
1417 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1418 skb_shinfo(skb)->nr_frags = new_frags;
1419
1420release:
1421 skb_zcopy_clear(skb, false);
1422 return 0;
1423}
1424EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1425
1426/**
1427 * skb_clone - duplicate an sk_buff
1428 * @skb: buffer to clone
1429 * @gfp_mask: allocation priority
1430 *
1431 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1432 * copies share the same packet data but not structure. The new
1433 * buffer has a reference count of 1. If the allocation fails the
1434 * function returns %NULL otherwise the new buffer is returned.
1435 *
1436 * If this function is called from an interrupt gfp_mask() must be
1437 * %GFP_ATOMIC.
1438 */
1439
1440struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1441{
1442 struct sk_buff_fclones *fclones = container_of(skb,
1443 struct sk_buff_fclones,
1444 skb1);
1445 struct sk_buff *n;
1446
1447 if (skb_orphan_frags(skb, gfp_mask))
1448 return NULL;
1449
1450 if (skb->fclone == SKB_FCLONE_ORIG &&
1451 refcount_read(&fclones->fclone_ref) == 1) {
1452 n = &fclones->skb2;
1453 refcount_set(&fclones->fclone_ref, 2);
1454 } else {
1455 if (skb_pfmemalloc(skb))
1456 gfp_mask |= __GFP_MEMALLOC;
1457
1458 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1459 if (!n)
1460 return NULL;
1461
1462 n->fclone = SKB_FCLONE_UNAVAILABLE;
1463 }
1464
1465 return __skb_clone(n, skb);
1466}
1467EXPORT_SYMBOL(skb_clone);
1468
1469void skb_headers_offset_update(struct sk_buff *skb, int off)
1470{
1471 /* Only adjust this if it actually is csum_start rather than csum */
1472 if (skb->ip_summed == CHECKSUM_PARTIAL)
1473 skb->csum_start += off;
1474 /* {transport,network,mac}_header and tail are relative to skb->head */
1475 skb->transport_header += off;
1476 skb->network_header += off;
1477 if (skb_mac_header_was_set(skb))
1478 skb->mac_header += off;
1479 skb->inner_transport_header += off;
1480 skb->inner_network_header += off;
1481 skb->inner_mac_header += off;
1482}
1483EXPORT_SYMBOL(skb_headers_offset_update);
1484
1485void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1486{
1487 __copy_skb_header(new, old);
1488
1489 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1490 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1491 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1492}
1493EXPORT_SYMBOL(skb_copy_header);
1494
1495static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1496{
1497 if (skb_pfmemalloc(skb))
1498 return SKB_ALLOC_RX;
1499 return 0;
1500}
1501
1502/**
1503 * skb_copy - create private copy of an sk_buff
1504 * @skb: buffer to copy
1505 * @gfp_mask: allocation priority
1506 *
1507 * Make a copy of both an &sk_buff and its data. This is used when the
1508 * caller wishes to modify the data and needs a private copy of the
1509 * data to alter. Returns %NULL on failure or the pointer to the buffer
1510 * on success. The returned buffer has a reference count of 1.
1511 *
1512 * As by-product this function converts non-linear &sk_buff to linear
1513 * one, so that &sk_buff becomes completely private and caller is allowed
1514 * to modify all the data of returned buffer. This means that this
1515 * function is not recommended for use in circumstances when only
1516 * header is going to be modified. Use pskb_copy() instead.
1517 */
1518
1519struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1520{
1521 int headerlen = skb_headroom(skb);
1522 unsigned int size = skb_end_offset(skb) + skb->data_len;
1523 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1524 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1525
1526 if (!n)
1527 return NULL;
1528
1529 /* Set the data pointer */
1530 skb_reserve(n, headerlen);
1531 /* Set the tail pointer and length */
1532 skb_put(n, skb->len);
1533
1534 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1535
1536 skb_copy_header(n, skb);
1537 return n;
1538}
1539EXPORT_SYMBOL(skb_copy);
1540
1541/**
1542 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1543 * @skb: buffer to copy
1544 * @headroom: headroom of new skb
1545 * @gfp_mask: allocation priority
1546 * @fclone: if true allocate the copy of the skb from the fclone
1547 * cache instead of the head cache; it is recommended to set this
1548 * to true for the cases where the copy will likely be cloned
1549 *
1550 * Make a copy of both an &sk_buff and part of its data, located
1551 * in header. Fragmented data remain shared. This is used when
1552 * the caller wishes to modify only header of &sk_buff and needs
1553 * private copy of the header to alter. Returns %NULL on failure
1554 * or the pointer to the buffer on success.
1555 * The returned buffer has a reference count of 1.
1556 */
1557
1558struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1559 gfp_t gfp_mask, bool fclone)
1560{
1561 unsigned int size = skb_headlen(skb) + headroom;
1562 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1563 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1564
1565 if (!n)
1566 goto out;
1567
1568 /* Set the data pointer */
1569 skb_reserve(n, headroom);
1570 /* Set the tail pointer and length */
1571 skb_put(n, skb_headlen(skb));
1572 /* Copy the bytes */
1573 skb_copy_from_linear_data(skb, n->data, n->len);
1574
1575 n->truesize += skb->data_len;
1576 n->data_len = skb->data_len;
1577 n->len = skb->len;
1578
1579 if (skb_shinfo(skb)->nr_frags) {
1580 int i;
1581
1582 if (skb_orphan_frags(skb, gfp_mask) ||
1583 skb_zerocopy_clone(n, skb, gfp_mask)) {
1584 kfree_skb(n);
1585 n = NULL;
1586 goto out;
1587 }
1588 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1589 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1590 skb_frag_ref(skb, i);
1591 }
1592 skb_shinfo(n)->nr_frags = i;
1593 }
1594
1595 if (skb_has_frag_list(skb)) {
1596 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1597 skb_clone_fraglist(n);
1598 }
1599
1600 skb_copy_header(n, skb);
1601out:
1602 return n;
1603}
1604EXPORT_SYMBOL(__pskb_copy_fclone);
1605
1606/**
1607 * pskb_expand_head - reallocate header of &sk_buff
1608 * @skb: buffer to reallocate
1609 * @nhead: room to add at head
1610 * @ntail: room to add at tail
1611 * @gfp_mask: allocation priority
1612 *
1613 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1614 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1615 * reference count of 1. Returns zero in the case of success or error,
1616 * if expansion failed. In the last case, &sk_buff is not changed.
1617 *
1618 * All the pointers pointing into skb header may change and must be
1619 * reloaded after call to this function.
1620 */
1621
1622int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1623 gfp_t gfp_mask)
1624{
1625 int i, osize = skb_end_offset(skb);
1626 int size = osize + nhead + ntail;
1627 long off;
1628 u8 *data;
1629
1630 BUG_ON(nhead < 0);
1631
1632 BUG_ON(skb_shared(skb));
1633
1634 size = SKB_DATA_ALIGN(size);
1635
1636 if (skb_pfmemalloc(skb))
1637 gfp_mask |= __GFP_MEMALLOC;
1638 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1639 gfp_mask, NUMA_NO_NODE, NULL);
1640 if (!data)
1641 goto nodata;
1642 size = SKB_WITH_OVERHEAD(ksize(data));
1643
1644 /* Copy only real data... and, alas, header. This should be
1645 * optimized for the cases when header is void.
1646 */
1647 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1648
1649 memcpy((struct skb_shared_info *)(data + size),
1650 skb_shinfo(skb),
1651 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1652
1653 /*
1654 * if shinfo is shared we must drop the old head gracefully, but if it
1655 * is not we can just drop the old head and let the existing refcount
1656 * be since all we did is relocate the values
1657 */
1658 if (skb_cloned(skb)) {
1659 if (skb_orphan_frags(skb, gfp_mask))
1660 goto nofrags;
1661 if (skb_zcopy(skb))
1662 refcount_inc(&skb_uarg(skb)->refcnt);
1663 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1664 skb_frag_ref(skb, i);
1665
1666 if (skb_has_frag_list(skb))
1667 skb_clone_fraglist(skb);
1668
1669 skb_release_data(skb);
1670 } else {
1671 skb_free_head(skb);
1672 }
1673 off = (data + nhead) - skb->head;
1674
1675 skb->head = data;
1676 skb->head_frag = 0;
1677 skb->data += off;
1678#ifdef NET_SKBUFF_DATA_USES_OFFSET
1679 skb->end = size;
1680 off = nhead;
1681#else
1682 skb->end = skb->head + size;
1683#endif
1684 skb->tail += off;
1685 skb_headers_offset_update(skb, nhead);
1686 skb->cloned = 0;
1687 skb->hdr_len = 0;
1688 skb->nohdr = 0;
1689 atomic_set(&skb_shinfo(skb)->dataref, 1);
1690
1691 skb_metadata_clear(skb);
1692
1693 /* It is not generally safe to change skb->truesize.
1694 * For the moment, we really care of rx path, or
1695 * when skb is orphaned (not attached to a socket).
1696 */
1697 if (!skb->sk || skb->destructor == sock_edemux)
1698 skb->truesize += size - osize;
1699
1700 return 0;
1701
1702nofrags:
1703 kfree(data);
1704nodata:
1705 return -ENOMEM;
1706}
1707EXPORT_SYMBOL(pskb_expand_head);
1708
1709/* Make private copy of skb with writable head and some headroom */
1710
1711struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1712{
1713 struct sk_buff *skb2;
1714 int delta = headroom - skb_headroom(skb);
1715
1716 if (delta <= 0)
1717 skb2 = pskb_copy(skb, GFP_ATOMIC);
1718 else {
1719 skb2 = skb_clone(skb, GFP_ATOMIC);
1720 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1721 GFP_ATOMIC)) {
1722 kfree_skb(skb2);
1723 skb2 = NULL;
1724 }
1725 }
1726 return skb2;
1727}
1728EXPORT_SYMBOL(skb_realloc_headroom);
1729
1730/**
1731 * skb_copy_expand - copy and expand sk_buff
1732 * @skb: buffer to copy
1733 * @newheadroom: new free bytes at head
1734 * @newtailroom: new free bytes at tail
1735 * @gfp_mask: allocation priority
1736 *
1737 * Make a copy of both an &sk_buff and its data and while doing so
1738 * allocate additional space.
1739 *
1740 * This is used when the caller wishes to modify the data and needs a
1741 * private copy of the data to alter as well as more space for new fields.
1742 * Returns %NULL on failure or the pointer to the buffer
1743 * on success. The returned buffer has a reference count of 1.
1744 *
1745 * You must pass %GFP_ATOMIC as the allocation priority if this function
1746 * is called from an interrupt.
1747 */
1748struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1749 int newheadroom, int newtailroom,
1750 gfp_t gfp_mask)
1751{
1752 /*
1753 * Allocate the copy buffer
1754 */
1755 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1756 gfp_mask, skb_alloc_rx_flag(skb),
1757 NUMA_NO_NODE);
1758 int oldheadroom = skb_headroom(skb);
1759 int head_copy_len, head_copy_off;
1760
1761 if (!n)
1762 return NULL;
1763
1764 skb_reserve(n, newheadroom);
1765
1766 /* Set the tail pointer and length */
1767 skb_put(n, skb->len);
1768
1769 head_copy_len = oldheadroom;
1770 head_copy_off = 0;
1771 if (newheadroom <= head_copy_len)
1772 head_copy_len = newheadroom;
1773 else
1774 head_copy_off = newheadroom - head_copy_len;
1775
1776 /* Copy the linear header and data. */
1777 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1778 skb->len + head_copy_len));
1779
1780 skb_copy_header(n, skb);
1781
1782 skb_headers_offset_update(n, newheadroom - oldheadroom);
1783
1784 return n;
1785}
1786EXPORT_SYMBOL(skb_copy_expand);
1787
1788/**
1789 * __skb_pad - zero pad the tail of an skb
1790 * @skb: buffer to pad
1791 * @pad: space to pad
1792 * @free_on_error: free buffer on error
1793 *
1794 * Ensure that a buffer is followed by a padding area that is zero
1795 * filled. Used by network drivers which may DMA or transfer data
1796 * beyond the buffer end onto the wire.
1797 *
1798 * May return error in out of memory cases. The skb is freed on error
1799 * if @free_on_error is true.
1800 */
1801
1802int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1803{
1804 int err;
1805 int ntail;
1806
1807 /* If the skbuff is non linear tailroom is always zero.. */
1808 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1809 memset(skb->data+skb->len, 0, pad);
1810 return 0;
1811 }
1812
1813 ntail = skb->data_len + pad - (skb->end - skb->tail);
1814 if (likely(skb_cloned(skb) || ntail > 0)) {
1815 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1816 if (unlikely(err))
1817 goto free_skb;
1818 }
1819
1820 /* FIXME: The use of this function with non-linear skb's really needs
1821 * to be audited.
1822 */
1823 err = skb_linearize(skb);
1824 if (unlikely(err))
1825 goto free_skb;
1826
1827 memset(skb->data + skb->len, 0, pad);
1828 return 0;
1829
1830free_skb:
1831 if (free_on_error)
1832 kfree_skb(skb);
1833 return err;
1834}
1835EXPORT_SYMBOL(__skb_pad);
1836
1837/**
1838 * pskb_put - add data to the tail of a potentially fragmented buffer
1839 * @skb: start of the buffer to use
1840 * @tail: tail fragment of the buffer to use
1841 * @len: amount of data to add
1842 *
1843 * This function extends the used data area of the potentially
1844 * fragmented buffer. @tail must be the last fragment of @skb -- or
1845 * @skb itself. If this would exceed the total buffer size the kernel
1846 * will panic. A pointer to the first byte of the extra data is
1847 * returned.
1848 */
1849
1850void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1851{
1852 if (tail != skb) {
1853 skb->data_len += len;
1854 skb->len += len;
1855 }
1856 return skb_put(tail, len);
1857}
1858EXPORT_SYMBOL_GPL(pskb_put);
1859
1860/**
1861 * skb_put - add data to a buffer
1862 * @skb: buffer to use
1863 * @len: amount of data to add
1864 *
1865 * This function extends the used data area of the buffer. If this would
1866 * exceed the total buffer size the kernel will panic. A pointer to the
1867 * first byte of the extra data is returned.
1868 */
1869void *skb_put(struct sk_buff *skb, unsigned int len)
1870{
1871 void *tmp = skb_tail_pointer(skb);
1872 SKB_LINEAR_ASSERT(skb);
1873 skb->tail += len;
1874 skb->len += len;
1875 if (unlikely(skb->tail > skb->end))
1876 skb_over_panic(skb, len, __builtin_return_address(0));
1877 return tmp;
1878}
1879EXPORT_SYMBOL(skb_put);
1880
1881/**
1882 * skb_push - add data to the start of a buffer
1883 * @skb: buffer to use
1884 * @len: amount of data to add
1885 *
1886 * This function extends the used data area of the buffer at the buffer
1887 * start. If this would exceed the total buffer headroom the kernel will
1888 * panic. A pointer to the first byte of the extra data is returned.
1889 */
1890void *skb_push(struct sk_buff *skb, unsigned int len)
1891{
1892 skb->data -= len;
1893 skb->len += len;
1894 if (unlikely(skb->data < skb->head))
1895 skb_under_panic(skb, len, __builtin_return_address(0));
1896 return skb->data;
1897}
1898EXPORT_SYMBOL(skb_push);
1899
1900/**
1901 * skb_pull - remove data from the start of a buffer
1902 * @skb: buffer to use
1903 * @len: amount of data to remove
1904 *
1905 * This function removes data from the start of a buffer, returning
1906 * the memory to the headroom. A pointer to the next data in the buffer
1907 * is returned. Once the data has been pulled future pushes will overwrite
1908 * the old data.
1909 */
1910void *skb_pull(struct sk_buff *skb, unsigned int len)
1911{
1912 return skb_pull_inline(skb, len);
1913}
1914EXPORT_SYMBOL(skb_pull);
1915
1916/**
1917 * skb_trim - remove end from a buffer
1918 * @skb: buffer to alter
1919 * @len: new length
1920 *
1921 * Cut the length of a buffer down by removing data from the tail. If
1922 * the buffer is already under the length specified it is not modified.
1923 * The skb must be linear.
1924 */
1925void skb_trim(struct sk_buff *skb, unsigned int len)
1926{
1927 if (skb->len > len)
1928 __skb_trim(skb, len);
1929}
1930EXPORT_SYMBOL(skb_trim);
1931
1932/* Trims skb to length len. It can change skb pointers.
1933 */
1934
1935int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1936{
1937 struct sk_buff **fragp;
1938 struct sk_buff *frag;
1939 int offset = skb_headlen(skb);
1940 int nfrags = skb_shinfo(skb)->nr_frags;
1941 int i;
1942 int err;
1943
1944 if (skb_cloned(skb) &&
1945 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1946 return err;
1947
1948 i = 0;
1949 if (offset >= len)
1950 goto drop_pages;
1951
1952 for (; i < nfrags; i++) {
1953 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1954
1955 if (end < len) {
1956 offset = end;
1957 continue;
1958 }
1959
1960 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1961
1962drop_pages:
1963 skb_shinfo(skb)->nr_frags = i;
1964
1965 for (; i < nfrags; i++)
1966 skb_frag_unref(skb, i);
1967
1968 if (skb_has_frag_list(skb))
1969 skb_drop_fraglist(skb);
1970 goto done;
1971 }
1972
1973 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1974 fragp = &frag->next) {
1975 int end = offset + frag->len;
1976
1977 if (skb_shared(frag)) {
1978 struct sk_buff *nfrag;
1979
1980 nfrag = skb_clone(frag, GFP_ATOMIC);
1981 if (unlikely(!nfrag))
1982 return -ENOMEM;
1983
1984 nfrag->next = frag->next;
1985 consume_skb(frag);
1986 frag = nfrag;
1987 *fragp = frag;
1988 }
1989
1990 if (end < len) {
1991 offset = end;
1992 continue;
1993 }
1994
1995 if (end > len &&
1996 unlikely((err = pskb_trim(frag, len - offset))))
1997 return err;
1998
1999 if (frag->next)
2000 skb_drop_list(&frag->next);
2001 break;
2002 }
2003
2004done:
2005 if (len > skb_headlen(skb)) {
2006 skb->data_len -= skb->len - len;
2007 skb->len = len;
2008 } else {
2009 skb->len = len;
2010 skb->data_len = 0;
2011 skb_set_tail_pointer(skb, len);
2012 }
2013
2014 if (!skb->sk || skb->destructor == sock_edemux)
2015 skb_condense(skb);
2016 return 0;
2017}
2018EXPORT_SYMBOL(___pskb_trim);
2019
2020/* Note : use pskb_trim_rcsum() instead of calling this directly
2021 */
2022int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2023{
2024 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2025 int delta = skb->len - len;
2026
2027 skb->csum = csum_block_sub(skb->csum,
2028 skb_checksum(skb, len, delta, 0),
2029 len);
2030 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2031 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2032 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2033
2034 if (offset + sizeof(__sum16) > hdlen)
2035 return -EINVAL;
2036 }
2037 return __pskb_trim(skb, len);
2038}
2039EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2040
2041/**
2042 * __pskb_pull_tail - advance tail of skb header
2043 * @skb: buffer to reallocate
2044 * @delta: number of bytes to advance tail
2045 *
2046 * The function makes a sense only on a fragmented &sk_buff,
2047 * it expands header moving its tail forward and copying necessary
2048 * data from fragmented part.
2049 *
2050 * &sk_buff MUST have reference count of 1.
2051 *
2052 * Returns %NULL (and &sk_buff does not change) if pull failed
2053 * or value of new tail of skb in the case of success.
2054 *
2055 * All the pointers pointing into skb header may change and must be
2056 * reloaded after call to this function.
2057 */
2058
2059/* Moves tail of skb head forward, copying data from fragmented part,
2060 * when it is necessary.
2061 * 1. It may fail due to malloc failure.
2062 * 2. It may change skb pointers.
2063 *
2064 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2065 */
2066void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2067{
2068 /* If skb has not enough free space at tail, get new one
2069 * plus 128 bytes for future expansions. If we have enough
2070 * room at tail, reallocate without expansion only if skb is cloned.
2071 */
2072 int i, k, eat = (skb->tail + delta) - skb->end;
2073
2074 if (eat > 0 || skb_cloned(skb)) {
2075 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2076 GFP_ATOMIC))
2077 return NULL;
2078 }
2079
2080 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2081 skb_tail_pointer(skb), delta));
2082
2083 /* Optimization: no fragments, no reasons to preestimate
2084 * size of pulled pages. Superb.
2085 */
2086 if (!skb_has_frag_list(skb))
2087 goto pull_pages;
2088
2089 /* Estimate size of pulled pages. */
2090 eat = delta;
2091 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2092 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2093
2094 if (size >= eat)
2095 goto pull_pages;
2096 eat -= size;
2097 }
2098
2099 /* If we need update frag list, we are in troubles.
2100 * Certainly, it is possible to add an offset to skb data,
2101 * but taking into account that pulling is expected to
2102 * be very rare operation, it is worth to fight against
2103 * further bloating skb head and crucify ourselves here instead.
2104 * Pure masohism, indeed. 8)8)
2105 */
2106 if (eat) {
2107 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2108 struct sk_buff *clone = NULL;
2109 struct sk_buff *insp = NULL;
2110
2111 do {
2112 if (list->len <= eat) {
2113 /* Eaten as whole. */
2114 eat -= list->len;
2115 list = list->next;
2116 insp = list;
2117 } else {
2118 /* Eaten partially. */
2119
2120 if (skb_shared(list)) {
2121 /* Sucks! We need to fork list. :-( */
2122 clone = skb_clone(list, GFP_ATOMIC);
2123 if (!clone)
2124 return NULL;
2125 insp = list->next;
2126 list = clone;
2127 } else {
2128 /* This may be pulled without
2129 * problems. */
2130 insp = list;
2131 }
2132 if (!pskb_pull(list, eat)) {
2133 kfree_skb(clone);
2134 return NULL;
2135 }
2136 break;
2137 }
2138 } while (eat);
2139
2140 /* Free pulled out fragments. */
2141 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2142 skb_shinfo(skb)->frag_list = list->next;
2143 kfree_skb(list);
2144 }
2145 /* And insert new clone at head. */
2146 if (clone) {
2147 clone->next = list;
2148 skb_shinfo(skb)->frag_list = clone;
2149 }
2150 }
2151 /* Success! Now we may commit changes to skb data. */
2152
2153pull_pages:
2154 eat = delta;
2155 k = 0;
2156 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2157 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2158
2159 if (size <= eat) {
2160 skb_frag_unref(skb, i);
2161 eat -= size;
2162 } else {
2163 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2164
2165 *frag = skb_shinfo(skb)->frags[i];
2166 if (eat) {
2167 skb_frag_off_add(frag, eat);
2168 skb_frag_size_sub(frag, eat);
2169 if (!i)
2170 goto end;
2171 eat = 0;
2172 }
2173 k++;
2174 }
2175 }
2176 skb_shinfo(skb)->nr_frags = k;
2177
2178end:
2179 skb->tail += delta;
2180 skb->data_len -= delta;
2181
2182 if (!skb->data_len)
2183 skb_zcopy_clear(skb, false);
2184
2185 return skb_tail_pointer(skb);
2186}
2187EXPORT_SYMBOL(__pskb_pull_tail);
2188
2189/**
2190 * skb_copy_bits - copy bits from skb to kernel buffer
2191 * @skb: source skb
2192 * @offset: offset in source
2193 * @to: destination buffer
2194 * @len: number of bytes to copy
2195 *
2196 * Copy the specified number of bytes from the source skb to the
2197 * destination buffer.
2198 *
2199 * CAUTION ! :
2200 * If its prototype is ever changed,
2201 * check arch/{*}/net/{*}.S files,
2202 * since it is called from BPF assembly code.
2203 */
2204int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2205{
2206 int start = skb_headlen(skb);
2207 struct sk_buff *frag_iter;
2208 int i, copy;
2209
2210 if (offset > (int)skb->len - len)
2211 goto fault;
2212
2213 /* Copy header. */
2214 if ((copy = start - offset) > 0) {
2215 if (copy > len)
2216 copy = len;
2217 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2218 if ((len -= copy) == 0)
2219 return 0;
2220 offset += copy;
2221 to += copy;
2222 }
2223
2224 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2225 int end;
2226 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2227
2228 WARN_ON(start > offset + len);
2229
2230 end = start + skb_frag_size(f);
2231 if ((copy = end - offset) > 0) {
2232 u32 p_off, p_len, copied;
2233 struct page *p;
2234 u8 *vaddr;
2235
2236 if (copy > len)
2237 copy = len;
2238
2239 skb_frag_foreach_page(f,
2240 skb_frag_off(f) + offset - start,
2241 copy, p, p_off, p_len, copied) {
2242 vaddr = kmap_atomic(p);
2243 memcpy(to + copied, vaddr + p_off, p_len);
2244 kunmap_atomic(vaddr);
2245 }
2246
2247 if ((len -= copy) == 0)
2248 return 0;
2249 offset += copy;
2250 to += copy;
2251 }
2252 start = end;
2253 }
2254
2255 skb_walk_frags(skb, frag_iter) {
2256 int end;
2257
2258 WARN_ON(start > offset + len);
2259
2260 end = start + frag_iter->len;
2261 if ((copy = end - offset) > 0) {
2262 if (copy > len)
2263 copy = len;
2264 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2265 goto fault;
2266 if ((len -= copy) == 0)
2267 return 0;
2268 offset += copy;
2269 to += copy;
2270 }
2271 start = end;
2272 }
2273
2274 if (!len)
2275 return 0;
2276
2277fault:
2278 return -EFAULT;
2279}
2280EXPORT_SYMBOL(skb_copy_bits);
2281
2282/*
2283 * Callback from splice_to_pipe(), if we need to release some pages
2284 * at the end of the spd in case we error'ed out in filling the pipe.
2285 */
2286static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2287{
2288 put_page(spd->pages[i]);
2289}
2290
2291static struct page *linear_to_page(struct page *page, unsigned int *len,
2292 unsigned int *offset,
2293 struct sock *sk)
2294{
2295 struct page_frag *pfrag = sk_page_frag(sk);
2296
2297 if (!sk_page_frag_refill(sk, pfrag))
2298 return NULL;
2299
2300 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2301
2302 memcpy(page_address(pfrag->page) + pfrag->offset,
2303 page_address(page) + *offset, *len);
2304 *offset = pfrag->offset;
2305 pfrag->offset += *len;
2306
2307 return pfrag->page;
2308}
2309
2310static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2311 struct page *page,
2312 unsigned int offset)
2313{
2314 return spd->nr_pages &&
2315 spd->pages[spd->nr_pages - 1] == page &&
2316 (spd->partial[spd->nr_pages - 1].offset +
2317 spd->partial[spd->nr_pages - 1].len == offset);
2318}
2319
2320/*
2321 * Fill page/offset/length into spd, if it can hold more pages.
2322 */
2323static bool spd_fill_page(struct splice_pipe_desc *spd,
2324 struct pipe_inode_info *pipe, struct page *page,
2325 unsigned int *len, unsigned int offset,
2326 bool linear,
2327 struct sock *sk)
2328{
2329 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2330 return true;
2331
2332 if (linear) {
2333 page = linear_to_page(page, len, &offset, sk);
2334 if (!page)
2335 return true;
2336 }
2337 if (spd_can_coalesce(spd, page, offset)) {
2338 spd->partial[spd->nr_pages - 1].len += *len;
2339 return false;
2340 }
2341 get_page(page);
2342 spd->pages[spd->nr_pages] = page;
2343 spd->partial[spd->nr_pages].len = *len;
2344 spd->partial[spd->nr_pages].offset = offset;
2345 spd->nr_pages++;
2346
2347 return false;
2348}
2349
2350static bool __splice_segment(struct page *page, unsigned int poff,
2351 unsigned int plen, unsigned int *off,
2352 unsigned int *len,
2353 struct splice_pipe_desc *spd, bool linear,
2354 struct sock *sk,
2355 struct pipe_inode_info *pipe)
2356{
2357 if (!*len)
2358 return true;
2359
2360 /* skip this segment if already processed */
2361 if (*off >= plen) {
2362 *off -= plen;
2363 return false;
2364 }
2365
2366 /* ignore any bits we already processed */
2367 poff += *off;
2368 plen -= *off;
2369 *off = 0;
2370
2371 do {
2372 unsigned int flen = min(*len, plen);
2373
2374 if (spd_fill_page(spd, pipe, page, &flen, poff,
2375 linear, sk))
2376 return true;
2377 poff += flen;
2378 plen -= flen;
2379 *len -= flen;
2380 } while (*len && plen);
2381
2382 return false;
2383}
2384
2385/*
2386 * Map linear and fragment data from the skb to spd. It reports true if the
2387 * pipe is full or if we already spliced the requested length.
2388 */
2389static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2390 unsigned int *offset, unsigned int *len,
2391 struct splice_pipe_desc *spd, struct sock *sk)
2392{
2393 int seg;
2394 struct sk_buff *iter;
2395
2396 /* map the linear part :
2397 * If skb->head_frag is set, this 'linear' part is backed by a
2398 * fragment, and if the head is not shared with any clones then
2399 * we can avoid a copy since we own the head portion of this page.
2400 */
2401 if (__splice_segment(virt_to_page(skb->data),
2402 (unsigned long) skb->data & (PAGE_SIZE - 1),
2403 skb_headlen(skb),
2404 offset, len, spd,
2405 skb_head_is_locked(skb),
2406 sk, pipe))
2407 return true;
2408
2409 /*
2410 * then map the fragments
2411 */
2412 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2413 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2414
2415 if (__splice_segment(skb_frag_page(f),
2416 skb_frag_off(f), skb_frag_size(f),
2417 offset, len, spd, false, sk, pipe))
2418 return true;
2419 }
2420
2421 skb_walk_frags(skb, iter) {
2422 if (*offset >= iter->len) {
2423 *offset -= iter->len;
2424 continue;
2425 }
2426 /* __skb_splice_bits() only fails if the output has no room
2427 * left, so no point in going over the frag_list for the error
2428 * case.
2429 */
2430 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2431 return true;
2432 }
2433
2434 return false;
2435}
2436
2437/*
2438 * Map data from the skb to a pipe. Should handle both the linear part,
2439 * the fragments, and the frag list.
2440 */
2441int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2442 struct pipe_inode_info *pipe, unsigned int tlen,
2443 unsigned int flags)
2444{
2445 struct partial_page partial[MAX_SKB_FRAGS];
2446 struct page *pages[MAX_SKB_FRAGS];
2447 struct splice_pipe_desc spd = {
2448 .pages = pages,
2449 .partial = partial,
2450 .nr_pages_max = MAX_SKB_FRAGS,
2451 .ops = &nosteal_pipe_buf_ops,
2452 .spd_release = sock_spd_release,
2453 };
2454 int ret = 0;
2455
2456 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2457
2458 if (spd.nr_pages)
2459 ret = splice_to_pipe(pipe, &spd);
2460
2461 return ret;
2462}
2463EXPORT_SYMBOL_GPL(skb_splice_bits);
2464
2465/* Send skb data on a socket. Socket must be locked. */
2466int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2467 int len)
2468{
2469 unsigned int orig_len = len;
2470 struct sk_buff *head = skb;
2471 unsigned short fragidx;
2472 int slen, ret;
2473
2474do_frag_list:
2475
2476 /* Deal with head data */
2477 while (offset < skb_headlen(skb) && len) {
2478 struct kvec kv;
2479 struct msghdr msg;
2480
2481 slen = min_t(int, len, skb_headlen(skb) - offset);
2482 kv.iov_base = skb->data + offset;
2483 kv.iov_len = slen;
2484 memset(&msg, 0, sizeof(msg));
2485 msg.msg_flags = MSG_DONTWAIT;
2486
2487 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2488 if (ret <= 0)
2489 goto error;
2490
2491 offset += ret;
2492 len -= ret;
2493 }
2494
2495 /* All the data was skb head? */
2496 if (!len)
2497 goto out;
2498
2499 /* Make offset relative to start of frags */
2500 offset -= skb_headlen(skb);
2501
2502 /* Find where we are in frag list */
2503 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2504 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2505
2506 if (offset < skb_frag_size(frag))
2507 break;
2508
2509 offset -= skb_frag_size(frag);
2510 }
2511
2512 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2513 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2514
2515 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2516
2517 while (slen) {
2518 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2519 skb_frag_off(frag) + offset,
2520 slen, MSG_DONTWAIT);
2521 if (ret <= 0)
2522 goto error;
2523
2524 len -= ret;
2525 offset += ret;
2526 slen -= ret;
2527 }
2528
2529 offset = 0;
2530 }
2531
2532 if (len) {
2533 /* Process any frag lists */
2534
2535 if (skb == head) {
2536 if (skb_has_frag_list(skb)) {
2537 skb = skb_shinfo(skb)->frag_list;
2538 goto do_frag_list;
2539 }
2540 } else if (skb->next) {
2541 skb = skb->next;
2542 goto do_frag_list;
2543 }
2544 }
2545
2546out:
2547 return orig_len - len;
2548
2549error:
2550 return orig_len == len ? ret : orig_len - len;
2551}
2552EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2553
2554/**
2555 * skb_store_bits - store bits from kernel buffer to skb
2556 * @skb: destination buffer
2557 * @offset: offset in destination
2558 * @from: source buffer
2559 * @len: number of bytes to copy
2560 *
2561 * Copy the specified number of bytes from the source buffer to the
2562 * destination skb. This function handles all the messy bits of
2563 * traversing fragment lists and such.
2564 */
2565
2566int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2567{
2568 int start = skb_headlen(skb);
2569 struct sk_buff *frag_iter;
2570 int i, copy;
2571
2572 if (offset > (int)skb->len - len)
2573 goto fault;
2574
2575 if ((copy = start - offset) > 0) {
2576 if (copy > len)
2577 copy = len;
2578 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2579 if ((len -= copy) == 0)
2580 return 0;
2581 offset += copy;
2582 from += copy;
2583 }
2584
2585 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2586 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2587 int end;
2588
2589 WARN_ON(start > offset + len);
2590
2591 end = start + skb_frag_size(frag);
2592 if ((copy = end - offset) > 0) {
2593 u32 p_off, p_len, copied;
2594 struct page *p;
2595 u8 *vaddr;
2596
2597 if (copy > len)
2598 copy = len;
2599
2600 skb_frag_foreach_page(frag,
2601 skb_frag_off(frag) + offset - start,
2602 copy, p, p_off, p_len, copied) {
2603 vaddr = kmap_atomic(p);
2604 memcpy(vaddr + p_off, from + copied, p_len);
2605 kunmap_atomic(vaddr);
2606 }
2607
2608 if ((len -= copy) == 0)
2609 return 0;
2610 offset += copy;
2611 from += copy;
2612 }
2613 start = end;
2614 }
2615
2616 skb_walk_frags(skb, frag_iter) {
2617 int end;
2618
2619 WARN_ON(start > offset + len);
2620
2621 end = start + frag_iter->len;
2622 if ((copy = end - offset) > 0) {
2623 if (copy > len)
2624 copy = len;
2625 if (skb_store_bits(frag_iter, offset - start,
2626 from, copy))
2627 goto fault;
2628 if ((len -= copy) == 0)
2629 return 0;
2630 offset += copy;
2631 from += copy;
2632 }
2633 start = end;
2634 }
2635 if (!len)
2636 return 0;
2637
2638fault:
2639 return -EFAULT;
2640}
2641EXPORT_SYMBOL(skb_store_bits);
2642
2643/* Checksum skb data. */
2644__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2645 __wsum csum, const struct skb_checksum_ops *ops)
2646{
2647 int start = skb_headlen(skb);
2648 int i, copy = start - offset;
2649 struct sk_buff *frag_iter;
2650 int pos = 0;
2651
2652 /* Checksum header. */
2653 if (copy > 0) {
2654 if (copy > len)
2655 copy = len;
2656 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2657 skb->data + offset, copy, csum);
2658 if ((len -= copy) == 0)
2659 return csum;
2660 offset += copy;
2661 pos = copy;
2662 }
2663
2664 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2665 int end;
2666 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2667
2668 WARN_ON(start > offset + len);
2669
2670 end = start + skb_frag_size(frag);
2671 if ((copy = end - offset) > 0) {
2672 u32 p_off, p_len, copied;
2673 struct page *p;
2674 __wsum csum2;
2675 u8 *vaddr;
2676
2677 if (copy > len)
2678 copy = len;
2679
2680 skb_frag_foreach_page(frag,
2681 skb_frag_off(frag) + offset - start,
2682 copy, p, p_off, p_len, copied) {
2683 vaddr = kmap_atomic(p);
2684 csum2 = INDIRECT_CALL_1(ops->update,
2685 csum_partial_ext,
2686 vaddr + p_off, p_len, 0);
2687 kunmap_atomic(vaddr);
2688 csum = INDIRECT_CALL_1(ops->combine,
2689 csum_block_add_ext, csum,
2690 csum2, pos, p_len);
2691 pos += p_len;
2692 }
2693
2694 if (!(len -= copy))
2695 return csum;
2696 offset += copy;
2697 }
2698 start = end;
2699 }
2700
2701 skb_walk_frags(skb, frag_iter) {
2702 int end;
2703
2704 WARN_ON(start > offset + len);
2705
2706 end = start + frag_iter->len;
2707 if ((copy = end - offset) > 0) {
2708 __wsum csum2;
2709 if (copy > len)
2710 copy = len;
2711 csum2 = __skb_checksum(frag_iter, offset - start,
2712 copy, 0, ops);
2713 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2714 csum, csum2, pos, copy);
2715 if ((len -= copy) == 0)
2716 return csum;
2717 offset += copy;
2718 pos += copy;
2719 }
2720 start = end;
2721 }
2722 BUG_ON(len);
2723
2724 return csum;
2725}
2726EXPORT_SYMBOL(__skb_checksum);
2727
2728__wsum skb_checksum(const struct sk_buff *skb, int offset,
2729 int len, __wsum csum)
2730{
2731 const struct skb_checksum_ops ops = {
2732 .update = csum_partial_ext,
2733 .combine = csum_block_add_ext,
2734 };
2735
2736 return __skb_checksum(skb, offset, len, csum, &ops);
2737}
2738EXPORT_SYMBOL(skb_checksum);
2739
2740/* Both of above in one bottle. */
2741
2742__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2743 u8 *to, int len)
2744{
2745 int start = skb_headlen(skb);
2746 int i, copy = start - offset;
2747 struct sk_buff *frag_iter;
2748 int pos = 0;
2749 __wsum csum = 0;
2750
2751 /* Copy header. */
2752 if (copy > 0) {
2753 if (copy > len)
2754 copy = len;
2755 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2756 copy);
2757 if ((len -= copy) == 0)
2758 return csum;
2759 offset += copy;
2760 to += copy;
2761 pos = copy;
2762 }
2763
2764 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2765 int end;
2766
2767 WARN_ON(start > offset + len);
2768
2769 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2770 if ((copy = end - offset) > 0) {
2771 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2772 u32 p_off, p_len, copied;
2773 struct page *p;
2774 __wsum csum2;
2775 u8 *vaddr;
2776
2777 if (copy > len)
2778 copy = len;
2779
2780 skb_frag_foreach_page(frag,
2781 skb_frag_off(frag) + offset - start,
2782 copy, p, p_off, p_len, copied) {
2783 vaddr = kmap_atomic(p);
2784 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2785 to + copied,
2786 p_len);
2787 kunmap_atomic(vaddr);
2788 csum = csum_block_add(csum, csum2, pos);
2789 pos += p_len;
2790 }
2791
2792 if (!(len -= copy))
2793 return csum;
2794 offset += copy;
2795 to += copy;
2796 }
2797 start = end;
2798 }
2799
2800 skb_walk_frags(skb, frag_iter) {
2801 __wsum csum2;
2802 int end;
2803
2804 WARN_ON(start > offset + len);
2805
2806 end = start + frag_iter->len;
2807 if ((copy = end - offset) > 0) {
2808 if (copy > len)
2809 copy = len;
2810 csum2 = skb_copy_and_csum_bits(frag_iter,
2811 offset - start,
2812 to, copy);
2813 csum = csum_block_add(csum, csum2, pos);
2814 if ((len -= copy) == 0)
2815 return csum;
2816 offset += copy;
2817 to += copy;
2818 pos += copy;
2819 }
2820 start = end;
2821 }
2822 BUG_ON(len);
2823 return csum;
2824}
2825EXPORT_SYMBOL(skb_copy_and_csum_bits);
2826
2827__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2828{
2829 __sum16 sum;
2830
2831 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2832 /* See comments in __skb_checksum_complete(). */
2833 if (likely(!sum)) {
2834 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2835 !skb->csum_complete_sw)
2836 netdev_rx_csum_fault(skb->dev, skb);
2837 }
2838 if (!skb_shared(skb))
2839 skb->csum_valid = !sum;
2840 return sum;
2841}
2842EXPORT_SYMBOL(__skb_checksum_complete_head);
2843
2844/* This function assumes skb->csum already holds pseudo header's checksum,
2845 * which has been changed from the hardware checksum, for example, by
2846 * __skb_checksum_validate_complete(). And, the original skb->csum must
2847 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
2848 *
2849 * It returns non-zero if the recomputed checksum is still invalid, otherwise
2850 * zero. The new checksum is stored back into skb->csum unless the skb is
2851 * shared.
2852 */
2853__sum16 __skb_checksum_complete(struct sk_buff *skb)
2854{
2855 __wsum csum;
2856 __sum16 sum;
2857
2858 csum = skb_checksum(skb, 0, skb->len, 0);
2859
2860 sum = csum_fold(csum_add(skb->csum, csum));
2861 /* This check is inverted, because we already knew the hardware
2862 * checksum is invalid before calling this function. So, if the
2863 * re-computed checksum is valid instead, then we have a mismatch
2864 * between the original skb->csum and skb_checksum(). This means either
2865 * the original hardware checksum is incorrect or we screw up skb->csum
2866 * when moving skb->data around.
2867 */
2868 if (likely(!sum)) {
2869 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2870 !skb->csum_complete_sw)
2871 netdev_rx_csum_fault(skb->dev, skb);
2872 }
2873
2874 if (!skb_shared(skb)) {
2875 /* Save full packet checksum */
2876 skb->csum = csum;
2877 skb->ip_summed = CHECKSUM_COMPLETE;
2878 skb->csum_complete_sw = 1;
2879 skb->csum_valid = !sum;
2880 }
2881
2882 return sum;
2883}
2884EXPORT_SYMBOL(__skb_checksum_complete);
2885
2886static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2887{
2888 net_warn_ratelimited(
2889 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2890 __func__);
2891 return 0;
2892}
2893
2894static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2895 int offset, int len)
2896{
2897 net_warn_ratelimited(
2898 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2899 __func__);
2900 return 0;
2901}
2902
2903static const struct skb_checksum_ops default_crc32c_ops = {
2904 .update = warn_crc32c_csum_update,
2905 .combine = warn_crc32c_csum_combine,
2906};
2907
2908const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2909 &default_crc32c_ops;
2910EXPORT_SYMBOL(crc32c_csum_stub);
2911
2912 /**
2913 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2914 * @from: source buffer
2915 *
2916 * Calculates the amount of linear headroom needed in the 'to' skb passed
2917 * into skb_zerocopy().
2918 */
2919unsigned int
2920skb_zerocopy_headlen(const struct sk_buff *from)
2921{
2922 unsigned int hlen = 0;
2923
2924 if (!from->head_frag ||
2925 skb_headlen(from) < L1_CACHE_BYTES ||
2926 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2927 hlen = skb_headlen(from);
2928
2929 if (skb_has_frag_list(from))
2930 hlen = from->len;
2931
2932 return hlen;
2933}
2934EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2935
2936/**
2937 * skb_zerocopy - Zero copy skb to skb
2938 * @to: destination buffer
2939 * @from: source buffer
2940 * @len: number of bytes to copy from source buffer
2941 * @hlen: size of linear headroom in destination buffer
2942 *
2943 * Copies up to `len` bytes from `from` to `to` by creating references
2944 * to the frags in the source buffer.
2945 *
2946 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2947 * headroom in the `to` buffer.
2948 *
2949 * Return value:
2950 * 0: everything is OK
2951 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2952 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2953 */
2954int
2955skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2956{
2957 int i, j = 0;
2958 int plen = 0; /* length of skb->head fragment */
2959 int ret;
2960 struct page *page;
2961 unsigned int offset;
2962
2963 BUG_ON(!from->head_frag && !hlen);
2964
2965 /* dont bother with small payloads */
2966 if (len <= skb_tailroom(to))
2967 return skb_copy_bits(from, 0, skb_put(to, len), len);
2968
2969 if (hlen) {
2970 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2971 if (unlikely(ret))
2972 return ret;
2973 len -= hlen;
2974 } else {
2975 plen = min_t(int, skb_headlen(from), len);
2976 if (plen) {
2977 page = virt_to_head_page(from->head);
2978 offset = from->data - (unsigned char *)page_address(page);
2979 __skb_fill_page_desc(to, 0, page, offset, plen);
2980 get_page(page);
2981 j = 1;
2982 len -= plen;
2983 }
2984 }
2985
2986 to->truesize += len + plen;
2987 to->len += len + plen;
2988 to->data_len += len + plen;
2989
2990 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2991 skb_tx_error(from);
2992 return -ENOMEM;
2993 }
2994 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2995
2996 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2997 int size;
2998
2999 if (!len)
3000 break;
3001 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3002 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3003 len);
3004 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3005 len -= size;
3006 skb_frag_ref(to, j);
3007 j++;
3008 }
3009 skb_shinfo(to)->nr_frags = j;
3010
3011 return 0;
3012}
3013EXPORT_SYMBOL_GPL(skb_zerocopy);
3014
3015void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3016{
3017 __wsum csum;
3018 long csstart;
3019
3020 if (skb->ip_summed == CHECKSUM_PARTIAL)
3021 csstart = skb_checksum_start_offset(skb);
3022 else
3023 csstart = skb_headlen(skb);
3024
3025 BUG_ON(csstart > skb_headlen(skb));
3026
3027 skb_copy_from_linear_data(skb, to, csstart);
3028
3029 csum = 0;
3030 if (csstart != skb->len)
3031 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3032 skb->len - csstart);
3033
3034 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3035 long csstuff = csstart + skb->csum_offset;
3036
3037 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3038 }
3039}
3040EXPORT_SYMBOL(skb_copy_and_csum_dev);
3041
3042/**
3043 * skb_dequeue - remove from the head of the queue
3044 * @list: list to dequeue from
3045 *
3046 * Remove the head of the list. The list lock is taken so the function
3047 * may be used safely with other locking list functions. The head item is
3048 * returned or %NULL if the list is empty.
3049 */
3050
3051struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3052{
3053 unsigned long flags;
3054 struct sk_buff *result;
3055
3056 spin_lock_irqsave(&list->lock, flags);
3057 result = __skb_dequeue(list);
3058 spin_unlock_irqrestore(&list->lock, flags);
3059 return result;
3060}
3061EXPORT_SYMBOL(skb_dequeue);
3062
3063/**
3064 * skb_dequeue_tail - remove from the tail of the queue
3065 * @list: list to dequeue from
3066 *
3067 * Remove the tail of the list. The list lock is taken so the function
3068 * may be used safely with other locking list functions. The tail item is
3069 * returned or %NULL if the list is empty.
3070 */
3071struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3072{
3073 unsigned long flags;
3074 struct sk_buff *result;
3075
3076 spin_lock_irqsave(&list->lock, flags);
3077 result = __skb_dequeue_tail(list);
3078 spin_unlock_irqrestore(&list->lock, flags);
3079 return result;
3080}
3081EXPORT_SYMBOL(skb_dequeue_tail);
3082
3083/**
3084 * skb_queue_purge - empty a list
3085 * @list: list to empty
3086 *
3087 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3088 * the list and one reference dropped. This function takes the list
3089 * lock and is atomic with respect to other list locking functions.
3090 */
3091void skb_queue_purge(struct sk_buff_head *list)
3092{
3093 struct sk_buff *skb;
3094 while ((skb = skb_dequeue(list)) != NULL)
3095 kfree_skb(skb);
3096}
3097EXPORT_SYMBOL(skb_queue_purge);
3098
3099/**
3100 * skb_rbtree_purge - empty a skb rbtree
3101 * @root: root of the rbtree to empty
3102 * Return value: the sum of truesizes of all purged skbs.
3103 *
3104 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3105 * the list and one reference dropped. This function does not take
3106 * any lock. Synchronization should be handled by the caller (e.g., TCP
3107 * out-of-order queue is protected by the socket lock).
3108 */
3109unsigned int skb_rbtree_purge(struct rb_root *root)
3110{
3111 struct rb_node *p = rb_first(root);
3112 unsigned int sum = 0;
3113
3114 while (p) {
3115 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3116
3117 p = rb_next(p);
3118 rb_erase(&skb->rbnode, root);
3119 sum += skb->truesize;
3120 kfree_skb(skb);
3121 }
3122 return sum;
3123}
3124
3125/**
3126 * skb_queue_head - queue a buffer at the list head
3127 * @list: list to use
3128 * @newsk: buffer to queue
3129 *
3130 * Queue a buffer at the start of the list. This function takes the
3131 * list lock and can be used safely with other locking &sk_buff functions
3132 * safely.
3133 *
3134 * A buffer cannot be placed on two lists at the same time.
3135 */
3136void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3137{
3138 unsigned long flags;
3139
3140 spin_lock_irqsave(&list->lock, flags);
3141 __skb_queue_head(list, newsk);
3142 spin_unlock_irqrestore(&list->lock, flags);
3143}
3144EXPORT_SYMBOL(skb_queue_head);
3145
3146/**
3147 * skb_queue_tail - queue a buffer at the list tail
3148 * @list: list to use
3149 * @newsk: buffer to queue
3150 *
3151 * Queue a buffer at the tail of the list. This function takes the
3152 * list lock and can be used safely with other locking &sk_buff functions
3153 * safely.
3154 *
3155 * A buffer cannot be placed on two lists at the same time.
3156 */
3157void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3158{
3159 unsigned long flags;
3160
3161 spin_lock_irqsave(&list->lock, flags);
3162 __skb_queue_tail(list, newsk);
3163 spin_unlock_irqrestore(&list->lock, flags);
3164}
3165EXPORT_SYMBOL(skb_queue_tail);
3166
3167/**
3168 * skb_unlink - remove a buffer from a list
3169 * @skb: buffer to remove
3170 * @list: list to use
3171 *
3172 * Remove a packet from a list. The list locks are taken and this
3173 * function is atomic with respect to other list locked calls
3174 *
3175 * You must know what list the SKB is on.
3176 */
3177void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3178{
3179 unsigned long flags;
3180
3181 spin_lock_irqsave(&list->lock, flags);
3182 __skb_unlink(skb, list);
3183 spin_unlock_irqrestore(&list->lock, flags);
3184}
3185EXPORT_SYMBOL(skb_unlink);
3186
3187/**
3188 * skb_append - append a buffer
3189 * @old: buffer to insert after
3190 * @newsk: buffer to insert
3191 * @list: list to use
3192 *
3193 * Place a packet after a given packet in a list. The list locks are taken
3194 * and this function is atomic with respect to other list locked calls.
3195 * A buffer cannot be placed on two lists at the same time.
3196 */
3197void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3198{
3199 unsigned long flags;
3200
3201 spin_lock_irqsave(&list->lock, flags);
3202 __skb_queue_after(list, old, newsk);
3203 spin_unlock_irqrestore(&list->lock, flags);
3204}
3205EXPORT_SYMBOL(skb_append);
3206
3207static inline void skb_split_inside_header(struct sk_buff *skb,
3208 struct sk_buff* skb1,
3209 const u32 len, const int pos)
3210{
3211 int i;
3212
3213 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3214 pos - len);
3215 /* And move data appendix as is. */
3216 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3217 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3218
3219 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3220 skb_shinfo(skb)->nr_frags = 0;
3221 skb1->data_len = skb->data_len;
3222 skb1->len += skb1->data_len;
3223 skb->data_len = 0;
3224 skb->len = len;
3225 skb_set_tail_pointer(skb, len);
3226}
3227
3228static inline void skb_split_no_header(struct sk_buff *skb,
3229 struct sk_buff* skb1,
3230 const u32 len, int pos)
3231{
3232 int i, k = 0;
3233 const int nfrags = skb_shinfo(skb)->nr_frags;
3234
3235 skb_shinfo(skb)->nr_frags = 0;
3236 skb1->len = skb1->data_len = skb->len - len;
3237 skb->len = len;
3238 skb->data_len = len - pos;
3239
3240 for (i = 0; i < nfrags; i++) {
3241 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3242
3243 if (pos + size > len) {
3244 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3245
3246 if (pos < len) {
3247 /* Split frag.
3248 * We have two variants in this case:
3249 * 1. Move all the frag to the second
3250 * part, if it is possible. F.e.
3251 * this approach is mandatory for TUX,
3252 * where splitting is expensive.
3253 * 2. Split is accurately. We make this.
3254 */
3255 skb_frag_ref(skb, i);
3256 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3257 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3258 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3259 skb_shinfo(skb)->nr_frags++;
3260 }
3261 k++;
3262 } else
3263 skb_shinfo(skb)->nr_frags++;
3264 pos += size;
3265 }
3266 skb_shinfo(skb1)->nr_frags = k;
3267}
3268
3269/**
3270 * skb_split - Split fragmented skb to two parts at length len.
3271 * @skb: the buffer to split
3272 * @skb1: the buffer to receive the second part
3273 * @len: new length for skb
3274 */
3275void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3276{
3277 int pos = skb_headlen(skb);
3278
3279 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3280 SKBTX_SHARED_FRAG;
3281 skb_zerocopy_clone(skb1, skb, 0);
3282 if (len < pos) /* Split line is inside header. */
3283 skb_split_inside_header(skb, skb1, len, pos);
3284 else /* Second chunk has no header, nothing to copy. */
3285 skb_split_no_header(skb, skb1, len, pos);
3286}
3287EXPORT_SYMBOL(skb_split);
3288
3289/* Shifting from/to a cloned skb is a no-go.
3290 *
3291 * Caller cannot keep skb_shinfo related pointers past calling here!
3292 */
3293static int skb_prepare_for_shift(struct sk_buff *skb)
3294{
3295 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3296}
3297
3298/**
3299 * skb_shift - Shifts paged data partially from skb to another
3300 * @tgt: buffer into which tail data gets added
3301 * @skb: buffer from which the paged data comes from
3302 * @shiftlen: shift up to this many bytes
3303 *
3304 * Attempts to shift up to shiftlen worth of bytes, which may be less than
3305 * the length of the skb, from skb to tgt. Returns number bytes shifted.
3306 * It's up to caller to free skb if everything was shifted.
3307 *
3308 * If @tgt runs out of frags, the whole operation is aborted.
3309 *
3310 * Skb cannot include anything else but paged data while tgt is allowed
3311 * to have non-paged data as well.
3312 *
3313 * TODO: full sized shift could be optimized but that would need
3314 * specialized skb free'er to handle frags without up-to-date nr_frags.
3315 */
3316int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3317{
3318 int from, to, merge, todo;
3319 skb_frag_t *fragfrom, *fragto;
3320
3321 BUG_ON(shiftlen > skb->len);
3322
3323 if (skb_headlen(skb))
3324 return 0;
3325 if (skb_zcopy(tgt) || skb_zcopy(skb))
3326 return 0;
3327
3328 todo = shiftlen;
3329 from = 0;
3330 to = skb_shinfo(tgt)->nr_frags;
3331 fragfrom = &skb_shinfo(skb)->frags[from];
3332
3333 /* Actual merge is delayed until the point when we know we can
3334 * commit all, so that we don't have to undo partial changes
3335 */
3336 if (!to ||
3337 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3338 skb_frag_off(fragfrom))) {
3339 merge = -1;
3340 } else {
3341 merge = to - 1;
3342
3343 todo -= skb_frag_size(fragfrom);
3344 if (todo < 0) {
3345 if (skb_prepare_for_shift(skb) ||
3346 skb_prepare_for_shift(tgt))
3347 return 0;
3348
3349 /* All previous frag pointers might be stale! */
3350 fragfrom = &skb_shinfo(skb)->frags[from];
3351 fragto = &skb_shinfo(tgt)->frags[merge];
3352
3353 skb_frag_size_add(fragto, shiftlen);
3354 skb_frag_size_sub(fragfrom, shiftlen);
3355 skb_frag_off_add(fragfrom, shiftlen);
3356
3357 goto onlymerged;
3358 }
3359
3360 from++;
3361 }
3362
3363 /* Skip full, not-fitting skb to avoid expensive operations */
3364 if ((shiftlen == skb->len) &&
3365 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3366 return 0;
3367
3368 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3369 return 0;
3370
3371 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3372 if (to == MAX_SKB_FRAGS)
3373 return 0;
3374
3375 fragfrom = &skb_shinfo(skb)->frags[from];
3376 fragto = &skb_shinfo(tgt)->frags[to];
3377
3378 if (todo >= skb_frag_size(fragfrom)) {
3379 *fragto = *fragfrom;
3380 todo -= skb_frag_size(fragfrom);
3381 from++;
3382 to++;
3383
3384 } else {
3385 __skb_frag_ref(fragfrom);
3386 skb_frag_page_copy(fragto, fragfrom);
3387 skb_frag_off_copy(fragto, fragfrom);
3388 skb_frag_size_set(fragto, todo);
3389
3390 skb_frag_off_add(fragfrom, todo);
3391 skb_frag_size_sub(fragfrom, todo);
3392 todo = 0;
3393
3394 to++;
3395 break;
3396 }
3397 }
3398
3399 /* Ready to "commit" this state change to tgt */
3400 skb_shinfo(tgt)->nr_frags = to;
3401
3402 if (merge >= 0) {
3403 fragfrom = &skb_shinfo(skb)->frags[0];
3404 fragto = &skb_shinfo(tgt)->frags[merge];
3405
3406 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3407 __skb_frag_unref(fragfrom);
3408 }
3409
3410 /* Reposition in the original skb */
3411 to = 0;
3412 while (from < skb_shinfo(skb)->nr_frags)
3413 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3414 skb_shinfo(skb)->nr_frags = to;
3415
3416 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3417
3418onlymerged:
3419 /* Most likely the tgt won't ever need its checksum anymore, skb on
3420 * the other hand might need it if it needs to be resent
3421 */
3422 tgt->ip_summed = CHECKSUM_PARTIAL;
3423 skb->ip_summed = CHECKSUM_PARTIAL;
3424
3425 /* Yak, is it really working this way? Some helper please? */
3426 skb->len -= shiftlen;
3427 skb->data_len -= shiftlen;
3428 skb->truesize -= shiftlen;
3429 tgt->len += shiftlen;
3430 tgt->data_len += shiftlen;
3431 tgt->truesize += shiftlen;
3432
3433 return shiftlen;
3434}
3435
3436/**
3437 * skb_prepare_seq_read - Prepare a sequential read of skb data
3438 * @skb: the buffer to read
3439 * @from: lower offset of data to be read
3440 * @to: upper offset of data to be read
3441 * @st: state variable
3442 *
3443 * Initializes the specified state variable. Must be called before
3444 * invoking skb_seq_read() for the first time.
3445 */
3446void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3447 unsigned int to, struct skb_seq_state *st)
3448{
3449 st->lower_offset = from;
3450 st->upper_offset = to;
3451 st->root_skb = st->cur_skb = skb;
3452 st->frag_idx = st->stepped_offset = 0;
3453 st->frag_data = NULL;
3454 st->frag_off = 0;
3455}
3456EXPORT_SYMBOL(skb_prepare_seq_read);
3457
3458/**
3459 * skb_seq_read - Sequentially read skb data
3460 * @consumed: number of bytes consumed by the caller so far
3461 * @data: destination pointer for data to be returned
3462 * @st: state variable
3463 *
3464 * Reads a block of skb data at @consumed relative to the
3465 * lower offset specified to skb_prepare_seq_read(). Assigns
3466 * the head of the data block to @data and returns the length
3467 * of the block or 0 if the end of the skb data or the upper
3468 * offset has been reached.
3469 *
3470 * The caller is not required to consume all of the data
3471 * returned, i.e. @consumed is typically set to the number
3472 * of bytes already consumed and the next call to
3473 * skb_seq_read() will return the remaining part of the block.
3474 *
3475 * Note 1: The size of each block of data returned can be arbitrary,
3476 * this limitation is the cost for zerocopy sequential
3477 * reads of potentially non linear data.
3478 *
3479 * Note 2: Fragment lists within fragments are not implemented
3480 * at the moment, state->root_skb could be replaced with
3481 * a stack for this purpose.
3482 */
3483unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3484 struct skb_seq_state *st)
3485{
3486 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3487 skb_frag_t *frag;
3488
3489 if (unlikely(abs_offset >= st->upper_offset)) {
3490 if (st->frag_data) {
3491 kunmap_atomic(st->frag_data);
3492 st->frag_data = NULL;
3493 }
3494 return 0;
3495 }
3496
3497next_skb:
3498 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3499
3500 if (abs_offset < block_limit && !st->frag_data) {
3501 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3502 return block_limit - abs_offset;
3503 }
3504
3505 if (st->frag_idx == 0 && !st->frag_data)
3506 st->stepped_offset += skb_headlen(st->cur_skb);
3507
3508 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3509 unsigned int pg_idx, pg_off, pg_sz;
3510
3511 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3512
3513 pg_idx = 0;
3514 pg_off = skb_frag_off(frag);
3515 pg_sz = skb_frag_size(frag);
3516
3517 if (skb_frag_must_loop(skb_frag_page(frag))) {
3518 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
3519 pg_off = offset_in_page(pg_off + st->frag_off);
3520 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
3521 PAGE_SIZE - pg_off);
3522 }
3523
3524 block_limit = pg_sz + st->stepped_offset;
3525 if (abs_offset < block_limit) {
3526 if (!st->frag_data)
3527 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
3528
3529 *data = (u8 *)st->frag_data + pg_off +
3530 (abs_offset - st->stepped_offset);
3531
3532 return block_limit - abs_offset;
3533 }
3534
3535 if (st->frag_data) {
3536 kunmap_atomic(st->frag_data);
3537 st->frag_data = NULL;
3538 }
3539
3540 st->stepped_offset += pg_sz;
3541 st->frag_off += pg_sz;
3542 if (st->frag_off == skb_frag_size(frag)) {
3543 st->frag_off = 0;
3544 st->frag_idx++;
3545 }
3546 }
3547
3548 if (st->frag_data) {
3549 kunmap_atomic(st->frag_data);
3550 st->frag_data = NULL;
3551 }
3552
3553 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3554 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3555 st->frag_idx = 0;
3556 goto next_skb;
3557 } else if (st->cur_skb->next) {
3558 st->cur_skb = st->cur_skb->next;
3559 st->frag_idx = 0;
3560 goto next_skb;
3561 }
3562
3563 return 0;
3564}
3565EXPORT_SYMBOL(skb_seq_read);
3566
3567/**
3568 * skb_abort_seq_read - Abort a sequential read of skb data
3569 * @st: state variable
3570 *
3571 * Must be called if skb_seq_read() was not called until it
3572 * returned 0.
3573 */
3574void skb_abort_seq_read(struct skb_seq_state *st)
3575{
3576 if (st->frag_data)
3577 kunmap_atomic(st->frag_data);
3578}
3579EXPORT_SYMBOL(skb_abort_seq_read);
3580
3581#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3582
3583static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3584 struct ts_config *conf,
3585 struct ts_state *state)
3586{
3587 return skb_seq_read(offset, text, TS_SKB_CB(state));
3588}
3589
3590static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3591{
3592 skb_abort_seq_read(TS_SKB_CB(state));
3593}
3594
3595/**
3596 * skb_find_text - Find a text pattern in skb data
3597 * @skb: the buffer to look in
3598 * @from: search offset
3599 * @to: search limit
3600 * @config: textsearch configuration
3601 *
3602 * Finds a pattern in the skb data according to the specified
3603 * textsearch configuration. Use textsearch_next() to retrieve
3604 * subsequent occurrences of the pattern. Returns the offset
3605 * to the first occurrence or UINT_MAX if no match was found.
3606 */
3607unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3608 unsigned int to, struct ts_config *config)
3609{
3610 struct ts_state state;
3611 unsigned int ret;
3612
3613 config->get_next_block = skb_ts_get_next_block;
3614 config->finish = skb_ts_finish;
3615
3616 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3617
3618 ret = textsearch_find(config, &state);
3619 return (ret <= to - from ? ret : UINT_MAX);
3620}
3621EXPORT_SYMBOL(skb_find_text);
3622
3623int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3624 int offset, size_t size)
3625{
3626 int i = skb_shinfo(skb)->nr_frags;
3627
3628 if (skb_can_coalesce(skb, i, page, offset)) {
3629 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3630 } else if (i < MAX_SKB_FRAGS) {
3631 get_page(page);
3632 skb_fill_page_desc(skb, i, page, offset, size);
3633 } else {
3634 return -EMSGSIZE;
3635 }
3636
3637 return 0;
3638}
3639EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3640
3641/**
3642 * skb_pull_rcsum - pull skb and update receive checksum
3643 * @skb: buffer to update
3644 * @len: length of data pulled
3645 *
3646 * This function performs an skb_pull on the packet and updates
3647 * the CHECKSUM_COMPLETE checksum. It should be used on
3648 * receive path processing instead of skb_pull unless you know
3649 * that the checksum difference is zero (e.g., a valid IP header)
3650 * or you are setting ip_summed to CHECKSUM_NONE.
3651 */
3652void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3653{
3654 unsigned char *data = skb->data;
3655
3656 BUG_ON(len > skb->len);
3657 __skb_pull(skb, len);
3658 skb_postpull_rcsum(skb, data, len);
3659 return skb->data;
3660}
3661EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3662
3663static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3664{
3665 skb_frag_t head_frag;
3666 struct page *page;
3667
3668 page = virt_to_head_page(frag_skb->head);
3669 __skb_frag_set_page(&head_frag, page);
3670 skb_frag_off_set(&head_frag, frag_skb->data -
3671 (unsigned char *)page_address(page));
3672 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3673 return head_frag;
3674}
3675
3676struct sk_buff *skb_segment_list(struct sk_buff *skb,
3677 netdev_features_t features,
3678 unsigned int offset)
3679{
3680 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3681 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3682 unsigned int delta_truesize = 0;
3683 unsigned int delta_len = 0;
3684 struct sk_buff *tail = NULL;
3685 struct sk_buff *nskb, *tmp;
3686 int err;
3687
3688 skb_push(skb, -skb_network_offset(skb) + offset);
3689
3690 skb_shinfo(skb)->frag_list = NULL;
3691
3692 do {
3693 nskb = list_skb;
3694 list_skb = list_skb->next;
3695
3696 err = 0;
3697 if (skb_shared(nskb)) {
3698 tmp = skb_clone(nskb, GFP_ATOMIC);
3699 if (tmp) {
3700 consume_skb(nskb);
3701 nskb = tmp;
3702 err = skb_unclone(nskb, GFP_ATOMIC);
3703 } else {
3704 err = -ENOMEM;
3705 }
3706 }
3707
3708 if (!tail)
3709 skb->next = nskb;
3710 else
3711 tail->next = nskb;
3712
3713 if (unlikely(err)) {
3714 nskb->next = list_skb;
3715 goto err_linearize;
3716 }
3717
3718 tail = nskb;
3719
3720 delta_len += nskb->len;
3721 delta_truesize += nskb->truesize;
3722
3723 skb_push(nskb, -skb_network_offset(nskb) + offset);
3724
3725 skb_release_head_state(nskb);
3726 __copy_skb_header(nskb, skb);
3727
3728 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3729 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3730 nskb->data - tnl_hlen,
3731 offset + tnl_hlen);
3732
3733 if (skb_needs_linearize(nskb, features) &&
3734 __skb_linearize(nskb))
3735 goto err_linearize;
3736
3737 } while (list_skb);
3738
3739 skb->truesize = skb->truesize - delta_truesize;
3740 skb->data_len = skb->data_len - delta_len;
3741 skb->len = skb->len - delta_len;
3742
3743 skb_gso_reset(skb);
3744
3745 skb->prev = tail;
3746
3747 if (skb_needs_linearize(skb, features) &&
3748 __skb_linearize(skb))
3749 goto err_linearize;
3750
3751 skb_get(skb);
3752
3753 return skb;
3754
3755err_linearize:
3756 kfree_skb_list(skb->next);
3757 skb->next = NULL;
3758 return ERR_PTR(-ENOMEM);
3759}
3760EXPORT_SYMBOL_GPL(skb_segment_list);
3761
3762int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3763{
3764 if (unlikely(p->len + skb->len >= 65536))
3765 return -E2BIG;
3766
3767 if (NAPI_GRO_CB(p)->last == p)
3768 skb_shinfo(p)->frag_list = skb;
3769 else
3770 NAPI_GRO_CB(p)->last->next = skb;
3771
3772 skb_pull(skb, skb_gro_offset(skb));
3773
3774 NAPI_GRO_CB(p)->last = skb;
3775 NAPI_GRO_CB(p)->count++;
3776 p->data_len += skb->len;
3777 p->truesize += skb->truesize;
3778 p->len += skb->len;
3779
3780 NAPI_GRO_CB(skb)->same_flow = 1;
3781
3782 return 0;
3783}
3784
3785/**
3786 * skb_segment - Perform protocol segmentation on skb.
3787 * @head_skb: buffer to segment
3788 * @features: features for the output path (see dev->features)
3789 *
3790 * This function performs segmentation on the given skb. It returns
3791 * a pointer to the first in a list of new skbs for the segments.
3792 * In case of error it returns ERR_PTR(err).
3793 */
3794struct sk_buff *skb_segment(struct sk_buff *head_skb,
3795 netdev_features_t features)
3796{
3797 struct sk_buff *segs = NULL;
3798 struct sk_buff *tail = NULL;
3799 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3800 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3801 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3802 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3803 struct sk_buff *frag_skb = head_skb;
3804 unsigned int offset = doffset;
3805 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3806 unsigned int partial_segs = 0;
3807 unsigned int headroom;
3808 unsigned int len = head_skb->len;
3809 __be16 proto;
3810 bool csum, sg;
3811 int nfrags = skb_shinfo(head_skb)->nr_frags;
3812 int err = -ENOMEM;
3813 int i = 0;
3814 int pos;
3815
3816 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3817 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3818 /* gso_size is untrusted, and we have a frag_list with a linear
3819 * non head_frag head.
3820 *
3821 * (we assume checking the first list_skb member suffices;
3822 * i.e if either of the list_skb members have non head_frag
3823 * head, then the first one has too).
3824 *
3825 * If head_skb's headlen does not fit requested gso_size, it
3826 * means that the frag_list members do NOT terminate on exact
3827 * gso_size boundaries. Hence we cannot perform skb_frag_t page
3828 * sharing. Therefore we must fallback to copying the frag_list
3829 * skbs; we do so by disabling SG.
3830 */
3831 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3832 features &= ~NETIF_F_SG;
3833 }
3834
3835 __skb_push(head_skb, doffset);
3836 proto = skb_network_protocol(head_skb, NULL);
3837 if (unlikely(!proto))
3838 return ERR_PTR(-EINVAL);
3839
3840 sg = !!(features & NETIF_F_SG);
3841 csum = !!can_checksum_protocol(features, proto);
3842
3843 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3844 if (!(features & NETIF_F_GSO_PARTIAL)) {
3845 struct sk_buff *iter;
3846 unsigned int frag_len;
3847
3848 if (!list_skb ||
3849 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3850 goto normal;
3851
3852 /* If we get here then all the required
3853 * GSO features except frag_list are supported.
3854 * Try to split the SKB to multiple GSO SKBs
3855 * with no frag_list.
3856 * Currently we can do that only when the buffers don't
3857 * have a linear part and all the buffers except
3858 * the last are of the same length.
3859 */
3860 frag_len = list_skb->len;
3861 skb_walk_frags(head_skb, iter) {
3862 if (frag_len != iter->len && iter->next)
3863 goto normal;
3864 if (skb_headlen(iter) && !iter->head_frag)
3865 goto normal;
3866
3867 len -= iter->len;
3868 }
3869
3870 if (len != frag_len)
3871 goto normal;
3872 }
3873
3874 /* GSO partial only requires that we trim off any excess that
3875 * doesn't fit into an MSS sized block, so take care of that
3876 * now.
3877 */
3878 partial_segs = len / mss;
3879 if (partial_segs > 1)
3880 mss *= partial_segs;
3881 else
3882 partial_segs = 0;
3883 }
3884
3885normal:
3886 headroom = skb_headroom(head_skb);
3887 pos = skb_headlen(head_skb);
3888
3889 do {
3890 struct sk_buff *nskb;
3891 skb_frag_t *nskb_frag;
3892 int hsize;
3893 int size;
3894
3895 if (unlikely(mss == GSO_BY_FRAGS)) {
3896 len = list_skb->len;
3897 } else {
3898 len = head_skb->len - offset;
3899 if (len > mss)
3900 len = mss;
3901 }
3902
3903 hsize = skb_headlen(head_skb) - offset;
3904 if (hsize < 0)
3905 hsize = 0;
3906 if (hsize > len || !sg)
3907 hsize = len;
3908
3909 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3910 (skb_headlen(list_skb) == len || sg)) {
3911 BUG_ON(skb_headlen(list_skb) > len);
3912
3913 i = 0;
3914 nfrags = skb_shinfo(list_skb)->nr_frags;
3915 frag = skb_shinfo(list_skb)->frags;
3916 frag_skb = list_skb;
3917 pos += skb_headlen(list_skb);
3918
3919 while (pos < offset + len) {
3920 BUG_ON(i >= nfrags);
3921
3922 size = skb_frag_size(frag);
3923 if (pos + size > offset + len)
3924 break;
3925
3926 i++;
3927 pos += size;
3928 frag++;
3929 }
3930
3931 nskb = skb_clone(list_skb, GFP_ATOMIC);
3932 list_skb = list_skb->next;
3933
3934 if (unlikely(!nskb))
3935 goto err;
3936
3937 if (unlikely(pskb_trim(nskb, len))) {
3938 kfree_skb(nskb);
3939 goto err;
3940 }
3941
3942 hsize = skb_end_offset(nskb);
3943 if (skb_cow_head(nskb, doffset + headroom)) {
3944 kfree_skb(nskb);
3945 goto err;
3946 }
3947
3948 nskb->truesize += skb_end_offset(nskb) - hsize;
3949 skb_release_head_state(nskb);
3950 __skb_push(nskb, doffset);
3951 } else {
3952 nskb = __alloc_skb(hsize + doffset + headroom,
3953 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3954 NUMA_NO_NODE);
3955
3956 if (unlikely(!nskb))
3957 goto err;
3958
3959 skb_reserve(nskb, headroom);
3960 __skb_put(nskb, doffset);
3961 }
3962
3963 if (segs)
3964 tail->next = nskb;
3965 else
3966 segs = nskb;
3967 tail = nskb;
3968
3969 __copy_skb_header(nskb, head_skb);
3970
3971 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3972 skb_reset_mac_len(nskb);
3973
3974 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3975 nskb->data - tnl_hlen,
3976 doffset + tnl_hlen);
3977
3978 if (nskb->len == len + doffset)
3979 goto perform_csum_check;
3980
3981 if (!sg) {
3982 if (!csum) {
3983 if (!nskb->remcsum_offload)
3984 nskb->ip_summed = CHECKSUM_NONE;
3985 SKB_GSO_CB(nskb)->csum =
3986 skb_copy_and_csum_bits(head_skb, offset,
3987 skb_put(nskb,
3988 len),
3989 len);
3990 SKB_GSO_CB(nskb)->csum_start =
3991 skb_headroom(nskb) + doffset;
3992 } else {
3993 skb_copy_bits(head_skb, offset,
3994 skb_put(nskb, len),
3995 len);
3996 }
3997 continue;
3998 }
3999
4000 nskb_frag = skb_shinfo(nskb)->frags;
4001
4002 skb_copy_from_linear_data_offset(head_skb, offset,
4003 skb_put(nskb, hsize), hsize);
4004
4005 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
4006 SKBTX_SHARED_FRAG;
4007
4008 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4009 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4010 goto err;
4011
4012 while (pos < offset + len) {
4013 if (i >= nfrags) {
4014 i = 0;
4015 nfrags = skb_shinfo(list_skb)->nr_frags;
4016 frag = skb_shinfo(list_skb)->frags;
4017 frag_skb = list_skb;
4018 if (!skb_headlen(list_skb)) {
4019 BUG_ON(!nfrags);
4020 } else {
4021 BUG_ON(!list_skb->head_frag);
4022
4023 /* to make room for head_frag. */
4024 i--;
4025 frag--;
4026 }
4027 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
4028 skb_zerocopy_clone(nskb, frag_skb,
4029 GFP_ATOMIC))
4030 goto err;
4031
4032 list_skb = list_skb->next;
4033 }
4034
4035 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4036 MAX_SKB_FRAGS)) {
4037 net_warn_ratelimited(
4038 "skb_segment: too many frags: %u %u\n",
4039 pos, mss);
4040 err = -EINVAL;
4041 goto err;
4042 }
4043
4044 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4045 __skb_frag_ref(nskb_frag);
4046 size = skb_frag_size(nskb_frag);
4047
4048 if (pos < offset) {
4049 skb_frag_off_add(nskb_frag, offset - pos);
4050 skb_frag_size_sub(nskb_frag, offset - pos);
4051 }
4052
4053 skb_shinfo(nskb)->nr_frags++;
4054
4055 if (pos + size <= offset + len) {
4056 i++;
4057 frag++;
4058 pos += size;
4059 } else {
4060 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4061 goto skip_fraglist;
4062 }
4063
4064 nskb_frag++;
4065 }
4066
4067skip_fraglist:
4068 nskb->data_len = len - hsize;
4069 nskb->len += nskb->data_len;
4070 nskb->truesize += nskb->data_len;
4071
4072perform_csum_check:
4073 if (!csum) {
4074 if (skb_has_shared_frag(nskb) &&
4075 __skb_linearize(nskb))
4076 goto err;
4077
4078 if (!nskb->remcsum_offload)
4079 nskb->ip_summed = CHECKSUM_NONE;
4080 SKB_GSO_CB(nskb)->csum =
4081 skb_checksum(nskb, doffset,
4082 nskb->len - doffset, 0);
4083 SKB_GSO_CB(nskb)->csum_start =
4084 skb_headroom(nskb) + doffset;
4085 }
4086 } while ((offset += len) < head_skb->len);
4087
4088 /* Some callers want to get the end of the list.
4089 * Put it in segs->prev to avoid walking the list.
4090 * (see validate_xmit_skb_list() for example)
4091 */
4092 segs->prev = tail;
4093
4094 if (partial_segs) {
4095 struct sk_buff *iter;
4096 int type = skb_shinfo(head_skb)->gso_type;
4097 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4098
4099 /* Update type to add partial and then remove dodgy if set */
4100 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4101 type &= ~SKB_GSO_DODGY;
4102
4103 /* Update GSO info and prepare to start updating headers on
4104 * our way back down the stack of protocols.
4105 */
4106 for (iter = segs; iter; iter = iter->next) {
4107 skb_shinfo(iter)->gso_size = gso_size;
4108 skb_shinfo(iter)->gso_segs = partial_segs;
4109 skb_shinfo(iter)->gso_type = type;
4110 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4111 }
4112
4113 if (tail->len - doffset <= gso_size)
4114 skb_shinfo(tail)->gso_size = 0;
4115 else if (tail != segs)
4116 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4117 }
4118
4119 /* Following permits correct backpressure, for protocols
4120 * using skb_set_owner_w().
4121 * Idea is to tranfert ownership from head_skb to last segment.
4122 */
4123 if (head_skb->destructor == sock_wfree) {
4124 swap(tail->truesize, head_skb->truesize);
4125 swap(tail->destructor, head_skb->destructor);
4126 swap(tail->sk, head_skb->sk);
4127 }
4128 return segs;
4129
4130err:
4131 kfree_skb_list(segs);
4132 return ERR_PTR(err);
4133}
4134EXPORT_SYMBOL_GPL(skb_segment);
4135
4136int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4137{
4138 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4139 unsigned int offset = skb_gro_offset(skb);
4140 unsigned int headlen = skb_headlen(skb);
4141 unsigned int len = skb_gro_len(skb);
4142 unsigned int delta_truesize;
4143 struct sk_buff *lp;
4144
4145 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4146 return -E2BIG;
4147
4148 lp = NAPI_GRO_CB(p)->last;
4149 pinfo = skb_shinfo(lp);
4150
4151 if (headlen <= offset) {
4152 skb_frag_t *frag;
4153 skb_frag_t *frag2;
4154 int i = skbinfo->nr_frags;
4155 int nr_frags = pinfo->nr_frags + i;
4156
4157 if (nr_frags > MAX_SKB_FRAGS)
4158 goto merge;
4159
4160 offset -= headlen;
4161 pinfo->nr_frags = nr_frags;
4162 skbinfo->nr_frags = 0;
4163
4164 frag = pinfo->frags + nr_frags;
4165 frag2 = skbinfo->frags + i;
4166 do {
4167 *--frag = *--frag2;
4168 } while (--i);
4169
4170 skb_frag_off_add(frag, offset);
4171 skb_frag_size_sub(frag, offset);
4172
4173 /* all fragments truesize : remove (head size + sk_buff) */
4174 delta_truesize = skb->truesize -
4175 SKB_TRUESIZE(skb_end_offset(skb));
4176
4177 skb->truesize -= skb->data_len;
4178 skb->len -= skb->data_len;
4179 skb->data_len = 0;
4180
4181 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4182 goto done;
4183 } else if (skb->head_frag) {
4184 int nr_frags = pinfo->nr_frags;
4185 skb_frag_t *frag = pinfo->frags + nr_frags;
4186 struct page *page = virt_to_head_page(skb->head);
4187 unsigned int first_size = headlen - offset;
4188 unsigned int first_offset;
4189
4190 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4191 goto merge;
4192
4193 first_offset = skb->data -
4194 (unsigned char *)page_address(page) +
4195 offset;
4196
4197 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4198
4199 __skb_frag_set_page(frag, page);
4200 skb_frag_off_set(frag, first_offset);
4201 skb_frag_size_set(frag, first_size);
4202
4203 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4204 /* We dont need to clear skbinfo->nr_frags here */
4205
4206 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4207 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4208 goto done;
4209 }
4210
4211merge:
4212 delta_truesize = skb->truesize;
4213 if (offset > headlen) {
4214 unsigned int eat = offset - headlen;
4215
4216 skb_frag_off_add(&skbinfo->frags[0], eat);
4217 skb_frag_size_sub(&skbinfo->frags[0], eat);
4218 skb->data_len -= eat;
4219 skb->len -= eat;
4220 offset = headlen;
4221 }
4222
4223 __skb_pull(skb, offset);
4224
4225 if (NAPI_GRO_CB(p)->last == p)
4226 skb_shinfo(p)->frag_list = skb;
4227 else
4228 NAPI_GRO_CB(p)->last->next = skb;
4229 NAPI_GRO_CB(p)->last = skb;
4230 __skb_header_release(skb);
4231 lp = p;
4232
4233done:
4234 NAPI_GRO_CB(p)->count++;
4235 p->data_len += len;
4236 p->truesize += delta_truesize;
4237 p->len += len;
4238 if (lp != p) {
4239 lp->data_len += len;
4240 lp->truesize += delta_truesize;
4241 lp->len += len;
4242 }
4243 NAPI_GRO_CB(skb)->same_flow = 1;
4244 return 0;
4245}
4246
4247#ifdef CONFIG_SKB_EXTENSIONS
4248#define SKB_EXT_ALIGN_VALUE 8
4249#define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4250
4251static const u8 skb_ext_type_len[] = {
4252#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4253 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4254#endif
4255#ifdef CONFIG_XFRM
4256 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4257#endif
4258#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4259 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4260#endif
4261#if IS_ENABLED(CONFIG_MPTCP)
4262 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4263#endif
4264};
4265
4266static __always_inline unsigned int skb_ext_total_length(void)
4267{
4268 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4269#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4270 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4271#endif
4272#ifdef CONFIG_XFRM
4273 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4274#endif
4275#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4276 skb_ext_type_len[TC_SKB_EXT] +
4277#endif
4278#if IS_ENABLED(CONFIG_MPTCP)
4279 skb_ext_type_len[SKB_EXT_MPTCP] +
4280#endif
4281 0;
4282}
4283
4284static void skb_extensions_init(void)
4285{
4286 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4287 BUILD_BUG_ON(skb_ext_total_length() > 255);
4288
4289 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4290 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4291 0,
4292 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4293 NULL);
4294}
4295#else
4296static void skb_extensions_init(void) {}
4297#endif
4298
4299void __init skb_init(void)
4300{
4301 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4302 sizeof(struct sk_buff),
4303 0,
4304 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4305 offsetof(struct sk_buff, cb),
4306 sizeof_field(struct sk_buff, cb),
4307 NULL);
4308 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4309 sizeof(struct sk_buff_fclones),
4310 0,
4311 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4312 NULL);
4313 skb_extensions_init();
4314}
4315
4316static int
4317__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4318 unsigned int recursion_level)
4319{
4320 int start = skb_headlen(skb);
4321 int i, copy = start - offset;
4322 struct sk_buff *frag_iter;
4323 int elt = 0;
4324
4325 if (unlikely(recursion_level >= 24))
4326 return -EMSGSIZE;
4327
4328 if (copy > 0) {
4329 if (copy > len)
4330 copy = len;
4331 sg_set_buf(sg, skb->data + offset, copy);
4332 elt++;
4333 if ((len -= copy) == 0)
4334 return elt;
4335 offset += copy;
4336 }
4337
4338 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4339 int end;
4340
4341 WARN_ON(start > offset + len);
4342
4343 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4344 if ((copy = end - offset) > 0) {
4345 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4346 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4347 return -EMSGSIZE;
4348
4349 if (copy > len)
4350 copy = len;
4351 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4352 skb_frag_off(frag) + offset - start);
4353 elt++;
4354 if (!(len -= copy))
4355 return elt;
4356 offset += copy;
4357 }
4358 start = end;
4359 }
4360
4361 skb_walk_frags(skb, frag_iter) {
4362 int end, ret;
4363
4364 WARN_ON(start > offset + len);
4365
4366 end = start + frag_iter->len;
4367 if ((copy = end - offset) > 0) {
4368 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4369 return -EMSGSIZE;
4370
4371 if (copy > len)
4372 copy = len;
4373 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4374 copy, recursion_level + 1);
4375 if (unlikely(ret < 0))
4376 return ret;
4377 elt += ret;
4378 if ((len -= copy) == 0)
4379 return elt;
4380 offset += copy;
4381 }
4382 start = end;
4383 }
4384 BUG_ON(len);
4385 return elt;
4386}
4387
4388/**
4389 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
4390 * @skb: Socket buffer containing the buffers to be mapped
4391 * @sg: The scatter-gather list to map into
4392 * @offset: The offset into the buffer's contents to start mapping
4393 * @len: Length of buffer space to be mapped
4394 *
4395 * Fill the specified scatter-gather list with mappings/pointers into a
4396 * region of the buffer space attached to a socket buffer. Returns either
4397 * the number of scatterlist items used, or -EMSGSIZE if the contents
4398 * could not fit.
4399 */
4400int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4401{
4402 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4403
4404 if (nsg <= 0)
4405 return nsg;
4406
4407 sg_mark_end(&sg[nsg - 1]);
4408
4409 return nsg;
4410}
4411EXPORT_SYMBOL_GPL(skb_to_sgvec);
4412
4413/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
4414 * sglist without mark the sg which contain last skb data as the end.
4415 * So the caller can mannipulate sg list as will when padding new data after
4416 * the first call without calling sg_unmark_end to expend sg list.
4417 *
4418 * Scenario to use skb_to_sgvec_nomark:
4419 * 1. sg_init_table
4420 * 2. skb_to_sgvec_nomark(payload1)
4421 * 3. skb_to_sgvec_nomark(payload2)
4422 *
4423 * This is equivalent to:
4424 * 1. sg_init_table
4425 * 2. skb_to_sgvec(payload1)
4426 * 3. sg_unmark_end
4427 * 4. skb_to_sgvec(payload2)
4428 *
4429 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
4430 * is more preferable.
4431 */
4432int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4433 int offset, int len)
4434{
4435 return __skb_to_sgvec(skb, sg, offset, len, 0);
4436}
4437EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4438
4439
4440
4441/**
4442 * skb_cow_data - Check that a socket buffer's data buffers are writable
4443 * @skb: The socket buffer to check.
4444 * @tailbits: Amount of trailing space to be added
4445 * @trailer: Returned pointer to the skb where the @tailbits space begins
4446 *
4447 * Make sure that the data buffers attached to a socket buffer are
4448 * writable. If they are not, private copies are made of the data buffers
4449 * and the socket buffer is set to use these instead.
4450 *
4451 * If @tailbits is given, make sure that there is space to write @tailbits
4452 * bytes of data beyond current end of socket buffer. @trailer will be
4453 * set to point to the skb in which this space begins.
4454 *
4455 * The number of scatterlist elements required to completely map the
4456 * COW'd and extended socket buffer will be returned.
4457 */
4458int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4459{
4460 int copyflag;
4461 int elt;
4462 struct sk_buff *skb1, **skb_p;
4463
4464 /* If skb is cloned or its head is paged, reallocate
4465 * head pulling out all the pages (pages are considered not writable
4466 * at the moment even if they are anonymous).
4467 */
4468 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4469 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
4470 return -ENOMEM;
4471
4472 /* Easy case. Most of packets will go this way. */
4473 if (!skb_has_frag_list(skb)) {
4474 /* A little of trouble, not enough of space for trailer.
4475 * This should not happen, when stack is tuned to generate
4476 * good frames. OK, on miss we reallocate and reserve even more
4477 * space, 128 bytes is fair. */
4478
4479 if (skb_tailroom(skb) < tailbits &&
4480 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4481 return -ENOMEM;
4482
4483 /* Voila! */
4484 *trailer = skb;
4485 return 1;
4486 }
4487
4488 /* Misery. We are in troubles, going to mincer fragments... */
4489
4490 elt = 1;
4491 skb_p = &skb_shinfo(skb)->frag_list;
4492 copyflag = 0;
4493
4494 while ((skb1 = *skb_p) != NULL) {
4495 int ntail = 0;
4496
4497 /* The fragment is partially pulled by someone,
4498 * this can happen on input. Copy it and everything
4499 * after it. */
4500
4501 if (skb_shared(skb1))
4502 copyflag = 1;
4503
4504 /* If the skb is the last, worry about trailer. */
4505
4506 if (skb1->next == NULL && tailbits) {
4507 if (skb_shinfo(skb1)->nr_frags ||
4508 skb_has_frag_list(skb1) ||
4509 skb_tailroom(skb1) < tailbits)
4510 ntail = tailbits + 128;
4511 }
4512
4513 if (copyflag ||
4514 skb_cloned(skb1) ||
4515 ntail ||
4516 skb_shinfo(skb1)->nr_frags ||
4517 skb_has_frag_list(skb1)) {
4518 struct sk_buff *skb2;
4519
4520 /* Fuck, we are miserable poor guys... */
4521 if (ntail == 0)
4522 skb2 = skb_copy(skb1, GFP_ATOMIC);
4523 else
4524 skb2 = skb_copy_expand(skb1,
4525 skb_headroom(skb1),
4526 ntail,
4527 GFP_ATOMIC);
4528 if (unlikely(skb2 == NULL))
4529 return -ENOMEM;
4530
4531 if (skb1->sk)
4532 skb_set_owner_w(skb2, skb1->sk);
4533
4534 /* Looking around. Are we still alive?
4535 * OK, link new skb, drop old one */
4536
4537 skb2->next = skb1->next;
4538 *skb_p = skb2;
4539 kfree_skb(skb1);
4540 skb1 = skb2;
4541 }
4542 elt++;
4543 *trailer = skb1;
4544 skb_p = &skb1->next;
4545 }
4546
4547 return elt;
4548}
4549EXPORT_SYMBOL_GPL(skb_cow_data);
4550
4551static void sock_rmem_free(struct sk_buff *skb)
4552{
4553 struct sock *sk = skb->sk;
4554
4555 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4556}
4557
4558static void skb_set_err_queue(struct sk_buff *skb)
4559{
4560 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
4561 * So, it is safe to (mis)use it to mark skbs on the error queue.
4562 */
4563 skb->pkt_type = PACKET_OUTGOING;
4564 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4565}
4566
4567/*
4568 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
4569 */
4570int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4571{
4572 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4573 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4574 return -ENOMEM;
4575
4576 skb_orphan(skb);
4577 skb->sk = sk;
4578 skb->destructor = sock_rmem_free;
4579 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4580 skb_set_err_queue(skb);
4581
4582 /* before exiting rcu section, make sure dst is refcounted */
4583 skb_dst_force(skb);
4584
4585 skb_queue_tail(&sk->sk_error_queue, skb);
4586 if (!sock_flag(sk, SOCK_DEAD))
4587 sk->sk_error_report(sk);
4588 return 0;
4589}
4590EXPORT_SYMBOL(sock_queue_err_skb);
4591
4592static bool is_icmp_err_skb(const struct sk_buff *skb)
4593{
4594 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4595 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4596}
4597
4598struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4599{
4600 struct sk_buff_head *q = &sk->sk_error_queue;
4601 struct sk_buff *skb, *skb_next = NULL;
4602 bool icmp_next = false;
4603 unsigned long flags;
4604
4605 spin_lock_irqsave(&q->lock, flags);
4606 skb = __skb_dequeue(q);
4607 if (skb && (skb_next = skb_peek(q))) {
4608 icmp_next = is_icmp_err_skb(skb_next);
4609 if (icmp_next)
4610 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
4611 }
4612 spin_unlock_irqrestore(&q->lock, flags);
4613
4614 if (is_icmp_err_skb(skb) && !icmp_next)
4615 sk->sk_err = 0;
4616
4617 if (skb_next)
4618 sk->sk_error_report(sk);
4619
4620 return skb;
4621}
4622EXPORT_SYMBOL(sock_dequeue_err_skb);
4623
4624/**
4625 * skb_clone_sk - create clone of skb, and take reference to socket
4626 * @skb: the skb to clone
4627 *
4628 * This function creates a clone of a buffer that holds a reference on
4629 * sk_refcnt. Buffers created via this function are meant to be
4630 * returned using sock_queue_err_skb, or free via kfree_skb.
4631 *
4632 * When passing buffers allocated with this function to sock_queue_err_skb
4633 * it is necessary to wrap the call with sock_hold/sock_put in order to
4634 * prevent the socket from being released prior to being enqueued on
4635 * the sk_error_queue.
4636 */
4637struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4638{
4639 struct sock *sk = skb->sk;
4640 struct sk_buff *clone;
4641
4642 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4643 return NULL;
4644
4645 clone = skb_clone(skb, GFP_ATOMIC);
4646 if (!clone) {
4647 sock_put(sk);
4648 return NULL;
4649 }
4650
4651 clone->sk = sk;
4652 clone->destructor = sock_efree;
4653
4654 return clone;
4655}
4656EXPORT_SYMBOL(skb_clone_sk);
4657
4658static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4659 struct sock *sk,
4660 int tstype,
4661 bool opt_stats)
4662{
4663 struct sock_exterr_skb *serr;
4664 int err;
4665
4666 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4667
4668 serr = SKB_EXT_ERR(skb);
4669 memset(serr, 0, sizeof(*serr));
4670 serr->ee.ee_errno = ENOMSG;
4671 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4672 serr->ee.ee_info = tstype;
4673 serr->opt_stats = opt_stats;
4674 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4675 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4676 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4677 if (sk->sk_protocol == IPPROTO_TCP &&
4678 sk->sk_type == SOCK_STREAM)
4679 serr->ee.ee_data -= sk->sk_tskey;
4680 }
4681
4682 err = sock_queue_err_skb(sk, skb);
4683
4684 if (err)
4685 kfree_skb(skb);
4686}
4687
4688static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4689{
4690 bool ret;
4691
4692 if (likely(sysctl_tstamp_allow_data || tsonly))
4693 return true;
4694
4695 read_lock_bh(&sk->sk_callback_lock);
4696 ret = sk->sk_socket && sk->sk_socket->file &&
4697 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4698 read_unlock_bh(&sk->sk_callback_lock);
4699 return ret;
4700}
4701
4702void skb_complete_tx_timestamp(struct sk_buff *skb,
4703 struct skb_shared_hwtstamps *hwtstamps)
4704{
4705 struct sock *sk = skb->sk;
4706
4707 if (!skb_may_tx_timestamp(sk, false))
4708 goto err;
4709
4710 /* Take a reference to prevent skb_orphan() from freeing the socket,
4711 * but only if the socket refcount is not zero.
4712 */
4713 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4714 *skb_hwtstamps(skb) = *hwtstamps;
4715 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4716 sock_put(sk);
4717 return;
4718 }
4719
4720err:
4721 kfree_skb(skb);
4722}
4723EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4724
4725void __skb_tstamp_tx(struct sk_buff *orig_skb,
4726 struct skb_shared_hwtstamps *hwtstamps,
4727 struct sock *sk, int tstype)
4728{
4729 struct sk_buff *skb;
4730 bool tsonly, opt_stats = false;
4731
4732 if (!sk)
4733 return;
4734
4735 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4736 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4737 return;
4738
4739 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4740 if (!skb_may_tx_timestamp(sk, tsonly))
4741 return;
4742
4743 if (tsonly) {
4744#ifdef CONFIG_INET
4745 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4746 sk->sk_protocol == IPPROTO_TCP &&
4747 sk->sk_type == SOCK_STREAM) {
4748 skb = tcp_get_timestamping_opt_stats(sk, orig_skb);
4749 opt_stats = true;
4750 } else
4751#endif
4752 skb = alloc_skb(0, GFP_ATOMIC);
4753 } else {
4754 skb = skb_clone(orig_skb, GFP_ATOMIC);
4755 }
4756 if (!skb)
4757 return;
4758
4759 if (tsonly) {
4760 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4761 SKBTX_ANY_TSTAMP;
4762 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4763 }
4764
4765 if (hwtstamps)
4766 *skb_hwtstamps(skb) = *hwtstamps;
4767 else
4768 skb->tstamp = ktime_get_real();
4769
4770 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4771}
4772EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4773
4774void skb_tstamp_tx(struct sk_buff *orig_skb,
4775 struct skb_shared_hwtstamps *hwtstamps)
4776{
4777 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4778 SCM_TSTAMP_SND);
4779}
4780EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4781
4782void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4783{
4784 struct sock *sk = skb->sk;
4785 struct sock_exterr_skb *serr;
4786 int err = 1;
4787
4788 skb->wifi_acked_valid = 1;
4789 skb->wifi_acked = acked;
4790
4791 serr = SKB_EXT_ERR(skb);
4792 memset(serr, 0, sizeof(*serr));
4793 serr->ee.ee_errno = ENOMSG;
4794 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4795
4796 /* Take a reference to prevent skb_orphan() from freeing the socket,
4797 * but only if the socket refcount is not zero.
4798 */
4799 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4800 err = sock_queue_err_skb(sk, skb);
4801 sock_put(sk);
4802 }
4803 if (err)
4804 kfree_skb(skb);
4805}
4806EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4807
4808/**
4809 * skb_partial_csum_set - set up and verify partial csum values for packet
4810 * @skb: the skb to set
4811 * @start: the number of bytes after skb->data to start checksumming.
4812 * @off: the offset from start to place the checksum.
4813 *
4814 * For untrusted partially-checksummed packets, we need to make sure the values
4815 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
4816 *
4817 * This function checks and sets those values and skb->ip_summed: if this
4818 * returns false you should drop the packet.
4819 */
4820bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4821{
4822 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4823 u32 csum_start = skb_headroom(skb) + (u32)start;
4824
4825 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4826 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4827 start, off, skb_headroom(skb), skb_headlen(skb));
4828 return false;
4829 }
4830 skb->ip_summed = CHECKSUM_PARTIAL;
4831 skb->csum_start = csum_start;
4832 skb->csum_offset = off;
4833 skb_set_transport_header(skb, start);
4834 return true;
4835}
4836EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4837
4838static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4839 unsigned int max)
4840{
4841 if (skb_headlen(skb) >= len)
4842 return 0;
4843
4844 /* If we need to pullup then pullup to the max, so we
4845 * won't need to do it again.
4846 */
4847 if (max > skb->len)
4848 max = skb->len;
4849
4850 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4851 return -ENOMEM;
4852
4853 if (skb_headlen(skb) < len)
4854 return -EPROTO;
4855
4856 return 0;
4857}
4858
4859#define MAX_TCP_HDR_LEN (15 * 4)
4860
4861static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4862 typeof(IPPROTO_IP) proto,
4863 unsigned int off)
4864{
4865 int err;
4866
4867 switch (proto) {
4868 case IPPROTO_TCP:
4869 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4870 off + MAX_TCP_HDR_LEN);
4871 if (!err && !skb_partial_csum_set(skb, off,
4872 offsetof(struct tcphdr,
4873 check)))
4874 err = -EPROTO;
4875 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4876
4877 case IPPROTO_UDP:
4878 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4879 off + sizeof(struct udphdr));
4880 if (!err && !skb_partial_csum_set(skb, off,
4881 offsetof(struct udphdr,
4882 check)))
4883 err = -EPROTO;
4884 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4885 }
4886
4887 return ERR_PTR(-EPROTO);
4888}
4889
4890/* This value should be large enough to cover a tagged ethernet header plus
4891 * maximally sized IP and TCP or UDP headers.
4892 */
4893#define MAX_IP_HDR_LEN 128
4894
4895static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4896{
4897 unsigned int off;
4898 bool fragment;
4899 __sum16 *csum;
4900 int err;
4901
4902 fragment = false;
4903
4904 err = skb_maybe_pull_tail(skb,
4905 sizeof(struct iphdr),
4906 MAX_IP_HDR_LEN);
4907 if (err < 0)
4908 goto out;
4909
4910 if (ip_is_fragment(ip_hdr(skb)))
4911 fragment = true;
4912
4913 off = ip_hdrlen(skb);
4914
4915 err = -EPROTO;
4916
4917 if (fragment)
4918 goto out;
4919
4920 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4921 if (IS_ERR(csum))
4922 return PTR_ERR(csum);
4923
4924 if (recalculate)
4925 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4926 ip_hdr(skb)->daddr,
4927 skb->len - off,
4928 ip_hdr(skb)->protocol, 0);
4929 err = 0;
4930
4931out:
4932 return err;
4933}
4934
4935/* This value should be large enough to cover a tagged ethernet header plus
4936 * an IPv6 header, all options, and a maximal TCP or UDP header.
4937 */
4938#define MAX_IPV6_HDR_LEN 256
4939
4940#define OPT_HDR(type, skb, off) \
4941 (type *)(skb_network_header(skb) + (off))
4942
4943static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4944{
4945 int err;
4946 u8 nexthdr;
4947 unsigned int off;
4948 unsigned int len;
4949 bool fragment;
4950 bool done;
4951 __sum16 *csum;
4952
4953 fragment = false;
4954 done = false;
4955
4956 off = sizeof(struct ipv6hdr);
4957
4958 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4959 if (err < 0)
4960 goto out;
4961
4962 nexthdr = ipv6_hdr(skb)->nexthdr;
4963
4964 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4965 while (off <= len && !done) {
4966 switch (nexthdr) {
4967 case IPPROTO_DSTOPTS:
4968 case IPPROTO_HOPOPTS:
4969 case IPPROTO_ROUTING: {
4970 struct ipv6_opt_hdr *hp;
4971
4972 err = skb_maybe_pull_tail(skb,
4973 off +
4974 sizeof(struct ipv6_opt_hdr),
4975 MAX_IPV6_HDR_LEN);
4976 if (err < 0)
4977 goto out;
4978
4979 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4980 nexthdr = hp->nexthdr;
4981 off += ipv6_optlen(hp);
4982 break;
4983 }
4984 case IPPROTO_AH: {
4985 struct ip_auth_hdr *hp;
4986
4987 err = skb_maybe_pull_tail(skb,
4988 off +
4989 sizeof(struct ip_auth_hdr),
4990 MAX_IPV6_HDR_LEN);
4991 if (err < 0)
4992 goto out;
4993
4994 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4995 nexthdr = hp->nexthdr;
4996 off += ipv6_authlen(hp);
4997 break;
4998 }
4999 case IPPROTO_FRAGMENT: {
5000 struct frag_hdr *hp;
5001
5002 err = skb_maybe_pull_tail(skb,
5003 off +
5004 sizeof(struct frag_hdr),
5005 MAX_IPV6_HDR_LEN);
5006 if (err < 0)
5007 goto out;
5008
5009 hp = OPT_HDR(struct frag_hdr, skb, off);
5010
5011 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5012 fragment = true;
5013
5014 nexthdr = hp->nexthdr;
5015 off += sizeof(struct frag_hdr);
5016 break;
5017 }
5018 default:
5019 done = true;
5020 break;
5021 }
5022 }
5023
5024 err = -EPROTO;
5025
5026 if (!done || fragment)
5027 goto out;
5028
5029 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5030 if (IS_ERR(csum))
5031 return PTR_ERR(csum);
5032
5033 if (recalculate)
5034 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5035 &ipv6_hdr(skb)->daddr,
5036 skb->len - off, nexthdr, 0);
5037 err = 0;
5038
5039out:
5040 return err;
5041}
5042
5043/**
5044 * skb_checksum_setup - set up partial checksum offset
5045 * @skb: the skb to set up
5046 * @recalculate: if true the pseudo-header checksum will be recalculated
5047 */
5048int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5049{
5050 int err;
5051
5052 switch (skb->protocol) {
5053 case htons(ETH_P_IP):
5054 err = skb_checksum_setup_ipv4(skb, recalculate);
5055 break;
5056
5057 case htons(ETH_P_IPV6):
5058 err = skb_checksum_setup_ipv6(skb, recalculate);
5059 break;
5060
5061 default:
5062 err = -EPROTO;
5063 break;
5064 }
5065
5066 return err;
5067}
5068EXPORT_SYMBOL(skb_checksum_setup);
5069
5070/**
5071 * skb_checksum_maybe_trim - maybe trims the given skb
5072 * @skb: the skb to check
5073 * @transport_len: the data length beyond the network header
5074 *
5075 * Checks whether the given skb has data beyond the given transport length.
5076 * If so, returns a cloned skb trimmed to this transport length.
5077 * Otherwise returns the provided skb. Returns NULL in error cases
5078 * (e.g. transport_len exceeds skb length or out-of-memory).
5079 *
5080 * Caller needs to set the skb transport header and free any returned skb if it
5081 * differs from the provided skb.
5082 */
5083static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5084 unsigned int transport_len)
5085{
5086 struct sk_buff *skb_chk;
5087 unsigned int len = skb_transport_offset(skb) + transport_len;
5088 int ret;
5089
5090 if (skb->len < len)
5091 return NULL;
5092 else if (skb->len == len)
5093 return skb;
5094
5095 skb_chk = skb_clone(skb, GFP_ATOMIC);
5096 if (!skb_chk)
5097 return NULL;
5098
5099 ret = pskb_trim_rcsum(skb_chk, len);
5100 if (ret) {
5101 kfree_skb(skb_chk);
5102 return NULL;
5103 }
5104
5105 return skb_chk;
5106}
5107
5108/**
5109 * skb_checksum_trimmed - validate checksum of an skb
5110 * @skb: the skb to check
5111 * @transport_len: the data length beyond the network header
5112 * @skb_chkf: checksum function to use
5113 *
5114 * Applies the given checksum function skb_chkf to the provided skb.
5115 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5116 *
5117 * If the skb has data beyond the given transport length, then a
5118 * trimmed & cloned skb is checked and returned.
5119 *
5120 * Caller needs to set the skb transport header and free any returned skb if it
5121 * differs from the provided skb.
5122 */
5123struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5124 unsigned int transport_len,
5125 __sum16(*skb_chkf)(struct sk_buff *skb))
5126{
5127 struct sk_buff *skb_chk;
5128 unsigned int offset = skb_transport_offset(skb);
5129 __sum16 ret;
5130
5131 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5132 if (!skb_chk)
5133 goto err;
5134
5135 if (!pskb_may_pull(skb_chk, offset))
5136 goto err;
5137
5138 skb_pull_rcsum(skb_chk, offset);
5139 ret = skb_chkf(skb_chk);
5140 skb_push_rcsum(skb_chk, offset);
5141
5142 if (ret)
5143 goto err;
5144
5145 return skb_chk;
5146
5147err:
5148 if (skb_chk && skb_chk != skb)
5149 kfree_skb(skb_chk);
5150
5151 return NULL;
5152
5153}
5154EXPORT_SYMBOL(skb_checksum_trimmed);
5155
5156void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5157{
5158 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5159 skb->dev->name);
5160}
5161EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5162
5163void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5164{
5165 if (head_stolen) {
5166 skb_release_head_state(skb);
5167 kmem_cache_free(skbuff_head_cache, skb);
5168 } else {
5169 __kfree_skb(skb);
5170 }
5171}
5172EXPORT_SYMBOL(kfree_skb_partial);
5173
5174/**
5175 * skb_try_coalesce - try to merge skb to prior one
5176 * @to: prior buffer
5177 * @from: buffer to add
5178 * @fragstolen: pointer to boolean
5179 * @delta_truesize: how much more was allocated than was requested
5180 */
5181bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5182 bool *fragstolen, int *delta_truesize)
5183{
5184 struct skb_shared_info *to_shinfo, *from_shinfo;
5185 int i, delta, len = from->len;
5186
5187 *fragstolen = false;
5188
5189 if (skb_cloned(to))
5190 return false;
5191
5192 if (len <= skb_tailroom(to)) {
5193 if (len)
5194 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5195 *delta_truesize = 0;
5196 return true;
5197 }
5198
5199 to_shinfo = skb_shinfo(to);
5200 from_shinfo = skb_shinfo(from);
5201 if (to_shinfo->frag_list || from_shinfo->frag_list)
5202 return false;
5203 if (skb_zcopy(to) || skb_zcopy(from))
5204 return false;
5205
5206 if (skb_headlen(from) != 0) {
5207 struct page *page;
5208 unsigned int offset;
5209
5210 if (to_shinfo->nr_frags +
5211 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5212 return false;
5213
5214 if (skb_head_is_locked(from))
5215 return false;
5216
5217 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5218
5219 page = virt_to_head_page(from->head);
5220 offset = from->data - (unsigned char *)page_address(page);
5221
5222 skb_fill_page_desc(to, to_shinfo->nr_frags,
5223 page, offset, skb_headlen(from));
5224 *fragstolen = true;
5225 } else {
5226 if (to_shinfo->nr_frags +
5227 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5228 return false;
5229
5230 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5231 }
5232
5233 WARN_ON_ONCE(delta < len);
5234
5235 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5236 from_shinfo->frags,
5237 from_shinfo->nr_frags * sizeof(skb_frag_t));
5238 to_shinfo->nr_frags += from_shinfo->nr_frags;
5239
5240 if (!skb_cloned(from))
5241 from_shinfo->nr_frags = 0;
5242
5243 /* if the skb is not cloned this does nothing
5244 * since we set nr_frags to 0.
5245 */
5246 for (i = 0; i < from_shinfo->nr_frags; i++)
5247 __skb_frag_ref(&from_shinfo->frags[i]);
5248
5249 to->truesize += delta;
5250 to->len += len;
5251 to->data_len += len;
5252
5253 *delta_truesize = delta;
5254 return true;
5255}
5256EXPORT_SYMBOL(skb_try_coalesce);
5257
5258/**
5259 * skb_scrub_packet - scrub an skb
5260 *
5261 * @skb: buffer to clean
5262 * @xnet: packet is crossing netns
5263 *
5264 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
5265 * into/from a tunnel. Some information have to be cleared during these
5266 * operations.
5267 * skb_scrub_packet can also be used to clean a skb before injecting it in
5268 * another namespace (@xnet == true). We have to clear all information in the
5269 * skb that could impact namespace isolation.
5270 */
5271void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5272{
5273 skb->pkt_type = PACKET_HOST;
5274 skb->skb_iif = 0;
5275 skb->ignore_df = 0;
5276 skb_dst_drop(skb);
5277 skb_ext_reset(skb);
5278 nf_reset_ct(skb);
5279 nf_reset_trace(skb);
5280
5281#ifdef CONFIG_NET_SWITCHDEV
5282 skb->offload_fwd_mark = 0;
5283 skb->offload_l3_fwd_mark = 0;
5284#endif
5285
5286 if (!xnet)
5287 return;
5288
5289 ipvs_reset(skb);
5290 skb->mark = 0;
5291 skb->tstamp = 0;
5292}
5293EXPORT_SYMBOL_GPL(skb_scrub_packet);
5294
5295/**
5296 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
5297 *
5298 * @skb: GSO skb
5299 *
5300 * skb_gso_transport_seglen is used to determine the real size of the
5301 * individual segments, including Layer4 headers (TCP/UDP).
5302 *
5303 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
5304 */
5305static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5306{
5307 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5308 unsigned int thlen = 0;
5309
5310 if (skb->encapsulation) {
5311 thlen = skb_inner_transport_header(skb) -
5312 skb_transport_header(skb);
5313
5314 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5315 thlen += inner_tcp_hdrlen(skb);
5316 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5317 thlen = tcp_hdrlen(skb);
5318 } else if (unlikely(skb_is_gso_sctp(skb))) {
5319 thlen = sizeof(struct sctphdr);
5320 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5321 thlen = sizeof(struct udphdr);
5322 }
5323 /* UFO sets gso_size to the size of the fragmentation
5324 * payload, i.e. the size of the L4 (UDP) header is already
5325 * accounted for.
5326 */
5327 return thlen + shinfo->gso_size;
5328}
5329
5330/**
5331 * skb_gso_network_seglen - Return length of individual segments of a gso packet
5332 *
5333 * @skb: GSO skb
5334 *
5335 * skb_gso_network_seglen is used to determine the real size of the
5336 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
5337 *
5338 * The MAC/L2 header is not accounted for.
5339 */
5340static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5341{
5342 unsigned int hdr_len = skb_transport_header(skb) -
5343 skb_network_header(skb);
5344
5345 return hdr_len + skb_gso_transport_seglen(skb);
5346}
5347
5348/**
5349 * skb_gso_mac_seglen - Return length of individual segments of a gso packet
5350 *
5351 * @skb: GSO skb
5352 *
5353 * skb_gso_mac_seglen is used to determine the real size of the
5354 * individual segments, including MAC/L2, Layer3 (IP, IPv6) and L4
5355 * headers (TCP/UDP).
5356 */
5357static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5358{
5359 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5360
5361 return hdr_len + skb_gso_transport_seglen(skb);
5362}
5363
5364/**
5365 * skb_gso_size_check - check the skb size, considering GSO_BY_FRAGS
5366 *
5367 * There are a couple of instances where we have a GSO skb, and we
5368 * want to determine what size it would be after it is segmented.
5369 *
5370 * We might want to check:
5371 * - L3+L4+payload size (e.g. IP forwarding)
5372 * - L2+L3+L4+payload size (e.g. sanity check before passing to driver)
5373 *
5374 * This is a helper to do that correctly considering GSO_BY_FRAGS.
5375 *
5376 * @skb: GSO skb
5377 *
5378 * @seg_len: The segmented length (from skb_gso_*_seglen). In the
5379 * GSO_BY_FRAGS case this will be [header sizes + GSO_BY_FRAGS].
5380 *
5381 * @max_len: The maximum permissible length.
5382 *
5383 * Returns true if the segmented length <= max length.
5384 */
5385static inline bool skb_gso_size_check(const struct sk_buff *skb,
5386 unsigned int seg_len,
5387 unsigned int max_len) {
5388 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5389 const struct sk_buff *iter;
5390
5391 if (shinfo->gso_size != GSO_BY_FRAGS)
5392 return seg_len <= max_len;
5393
5394 /* Undo this so we can re-use header sizes */
5395 seg_len -= GSO_BY_FRAGS;
5396
5397 skb_walk_frags(skb, iter) {
5398 if (seg_len + skb_headlen(iter) > max_len)
5399 return false;
5400 }
5401
5402 return true;
5403}
5404
5405/**
5406 * skb_gso_validate_network_len - Will a split GSO skb fit into a given MTU?
5407 *
5408 * @skb: GSO skb
5409 * @mtu: MTU to validate against
5410 *
5411 * skb_gso_validate_network_len validates if a given skb will fit a
5412 * wanted MTU once split. It considers L3 headers, L4 headers, and the
5413 * payload.
5414 */
5415bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5416{
5417 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5418}
5419EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5420
5421/**
5422 * skb_gso_validate_mac_len - Will a split GSO skb fit in a given length?
5423 *
5424 * @skb: GSO skb
5425 * @len: length to validate against
5426 *
5427 * skb_gso_validate_mac_len validates if a given skb will fit a wanted
5428 * length once split, including L2, L3 and L4 headers and the payload.
5429 */
5430bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5431{
5432 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5433}
5434EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5435
5436static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5437{
5438 int mac_len, meta_len;
5439 void *meta;
5440
5441 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5442 kfree_skb(skb);
5443 return NULL;
5444 }
5445
5446 mac_len = skb->data - skb_mac_header(skb);
5447 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5448 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5449 mac_len - VLAN_HLEN - ETH_TLEN);
5450 }
5451
5452 meta_len = skb_metadata_len(skb);
5453 if (meta_len) {
5454 meta = skb_metadata_end(skb) - meta_len;
5455 memmove(meta + VLAN_HLEN, meta, meta_len);
5456 }
5457
5458 skb->mac_header += VLAN_HLEN;
5459 return skb;
5460}
5461
5462struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5463{
5464 struct vlan_hdr *vhdr;
5465 u16 vlan_tci;
5466
5467 if (unlikely(skb_vlan_tag_present(skb))) {
5468 /* vlan_tci is already set-up so leave this for another time */
5469 return skb;
5470 }
5471
5472 skb = skb_share_check(skb, GFP_ATOMIC);
5473 if (unlikely(!skb))
5474 goto err_free;
5475 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
5476 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
5477 goto err_free;
5478
5479 vhdr = (struct vlan_hdr *)skb->data;
5480 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5481 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5482
5483 skb_pull_rcsum(skb, VLAN_HLEN);
5484 vlan_set_encap_proto(skb, vhdr);
5485
5486 skb = skb_reorder_vlan_header(skb);
5487 if (unlikely(!skb))
5488 goto err_free;
5489
5490 skb_reset_network_header(skb);
5491 if (!skb_transport_header_was_set(skb))
5492 skb_reset_transport_header(skb);
5493 skb_reset_mac_len(skb);
5494
5495 return skb;
5496
5497err_free:
5498 kfree_skb(skb);
5499 return NULL;
5500}
5501EXPORT_SYMBOL(skb_vlan_untag);
5502
5503int skb_ensure_writable(struct sk_buff *skb, int write_len)
5504{
5505 if (!pskb_may_pull(skb, write_len))
5506 return -ENOMEM;
5507
5508 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5509 return 0;
5510
5511 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5512}
5513EXPORT_SYMBOL(skb_ensure_writable);
5514
5515/* remove VLAN header from packet and update csum accordingly.
5516 * expects a non skb_vlan_tag_present skb with a vlan tag payload
5517 */
5518int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5519{
5520 struct vlan_hdr *vhdr;
5521 int offset = skb->data - skb_mac_header(skb);
5522 int err;
5523
5524 if (WARN_ONCE(offset,
5525 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5526 offset)) {
5527 return -EINVAL;
5528 }
5529
5530 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5531 if (unlikely(err))
5532 return err;
5533
5534 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5535
5536 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5537 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5538
5539 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5540 __skb_pull(skb, VLAN_HLEN);
5541
5542 vlan_set_encap_proto(skb, vhdr);
5543 skb->mac_header += VLAN_HLEN;
5544
5545 if (skb_network_offset(skb) < ETH_HLEN)
5546 skb_set_network_header(skb, ETH_HLEN);
5547
5548 skb_reset_mac_len(skb);
5549
5550 return err;
5551}
5552EXPORT_SYMBOL(__skb_vlan_pop);
5553
5554/* Pop a vlan tag either from hwaccel or from payload.
5555 * Expects skb->data at mac header.
5556 */
5557int skb_vlan_pop(struct sk_buff *skb)
5558{
5559 u16 vlan_tci;
5560 __be16 vlan_proto;
5561 int err;
5562
5563 if (likely(skb_vlan_tag_present(skb))) {
5564 __vlan_hwaccel_clear_tag(skb);
5565 } else {
5566 if (unlikely(!eth_type_vlan(skb->protocol)))
5567 return 0;
5568
5569 err = __skb_vlan_pop(skb, &vlan_tci);
5570 if (err)
5571 return err;
5572 }
5573 /* move next vlan tag to hw accel tag */
5574 if (likely(!eth_type_vlan(skb->protocol)))
5575 return 0;
5576
5577 vlan_proto = skb->protocol;
5578 err = __skb_vlan_pop(skb, &vlan_tci);
5579 if (unlikely(err))
5580 return err;
5581
5582 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5583 return 0;
5584}
5585EXPORT_SYMBOL(skb_vlan_pop);
5586
5587/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
5588 * Expects skb->data at mac header.
5589 */
5590int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5591{
5592 if (skb_vlan_tag_present(skb)) {
5593 int offset = skb->data - skb_mac_header(skb);
5594 int err;
5595
5596 if (WARN_ONCE(offset,
5597 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5598 offset)) {
5599 return -EINVAL;
5600 }
5601
5602 err = __vlan_insert_tag(skb, skb->vlan_proto,
5603 skb_vlan_tag_get(skb));
5604 if (err)
5605 return err;
5606
5607 skb->protocol = skb->vlan_proto;
5608 skb->mac_len += VLAN_HLEN;
5609
5610 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5611 }
5612 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5613 return 0;
5614}
5615EXPORT_SYMBOL(skb_vlan_push);
5616
5617/**
5618 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
5619 *
5620 * @skb: Socket buffer to modify
5621 *
5622 * Drop the Ethernet header of @skb.
5623 *
5624 * Expects that skb->data points to the mac header and that no VLAN tags are
5625 * present.
5626 *
5627 * Returns 0 on success, -errno otherwise.
5628 */
5629int skb_eth_pop(struct sk_buff *skb)
5630{
5631 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
5632 skb_network_offset(skb) < ETH_HLEN)
5633 return -EPROTO;
5634
5635 skb_pull_rcsum(skb, ETH_HLEN);
5636 skb_reset_mac_header(skb);
5637 skb_reset_mac_len(skb);
5638
5639 return 0;
5640}
5641EXPORT_SYMBOL(skb_eth_pop);
5642
5643/**
5644 * skb_eth_push() - Add a new Ethernet header at the head of a packet
5645 *
5646 * @skb: Socket buffer to modify
5647 * @dst: Destination MAC address of the new header
5648 * @src: Source MAC address of the new header
5649 *
5650 * Prepend @skb with a new Ethernet header.
5651 *
5652 * Expects that skb->data points to the mac header, which must be empty.
5653 *
5654 * Returns 0 on success, -errno otherwise.
5655 */
5656int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
5657 const unsigned char *src)
5658{
5659 struct ethhdr *eth;
5660 int err;
5661
5662 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
5663 return -EPROTO;
5664
5665 err = skb_cow_head(skb, sizeof(*eth));
5666 if (err < 0)
5667 return err;
5668
5669 skb_push(skb, sizeof(*eth));
5670 skb_reset_mac_header(skb);
5671 skb_reset_mac_len(skb);
5672
5673 eth = eth_hdr(skb);
5674 ether_addr_copy(eth->h_dest, dst);
5675 ether_addr_copy(eth->h_source, src);
5676 eth->h_proto = skb->protocol;
5677
5678 skb_postpush_rcsum(skb, eth, sizeof(*eth));
5679
5680 return 0;
5681}
5682EXPORT_SYMBOL(skb_eth_push);
5683
5684/* Update the ethertype of hdr and the skb csum value if required. */
5685static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5686 __be16 ethertype)
5687{
5688 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5689 __be16 diff[] = { ~hdr->h_proto, ethertype };
5690
5691 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5692 }
5693
5694 hdr->h_proto = ethertype;
5695}
5696
5697/**
5698 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
5699 * the packet
5700 *
5701 * @skb: buffer
5702 * @mpls_lse: MPLS label stack entry to push
5703 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
5704 * @mac_len: length of the MAC header
5705 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
5706 * ethernet
5707 *
5708 * Expects skb->data at mac header.
5709 *
5710 * Returns 0 on success, -errno otherwise.
5711 */
5712int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5713 int mac_len, bool ethernet)
5714{
5715 struct mpls_shim_hdr *lse;
5716 int err;
5717
5718 if (unlikely(!eth_p_mpls(mpls_proto)))
5719 return -EINVAL;
5720
5721 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
5722 if (skb->encapsulation)
5723 return -EINVAL;
5724
5725 err = skb_cow_head(skb, MPLS_HLEN);
5726 if (unlikely(err))
5727 return err;
5728
5729 if (!skb->inner_protocol) {
5730 skb_set_inner_network_header(skb, skb_network_offset(skb));
5731 skb_set_inner_protocol(skb, skb->protocol);
5732 }
5733
5734 skb_push(skb, MPLS_HLEN);
5735 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5736 mac_len);
5737 skb_reset_mac_header(skb);
5738 skb_set_network_header(skb, mac_len);
5739 skb_reset_mac_len(skb);
5740
5741 lse = mpls_hdr(skb);
5742 lse->label_stack_entry = mpls_lse;
5743 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5744
5745 if (ethernet && mac_len >= ETH_HLEN)
5746 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5747 skb->protocol = mpls_proto;
5748
5749 return 0;
5750}
5751EXPORT_SYMBOL_GPL(skb_mpls_push);
5752
5753/**
5754 * skb_mpls_pop() - pop the outermost MPLS header
5755 *
5756 * @skb: buffer
5757 * @next_proto: ethertype of header after popped MPLS header
5758 * @mac_len: length of the MAC header
5759 * @ethernet: flag to indicate if the packet is ethernet
5760 *
5761 * Expects skb->data at mac header.
5762 *
5763 * Returns 0 on success, -errno otherwise.
5764 */
5765int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5766 bool ethernet)
5767{
5768 int err;
5769
5770 if (unlikely(!eth_p_mpls(skb->protocol)))
5771 return 0;
5772
5773 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5774 if (unlikely(err))
5775 return err;
5776
5777 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5778 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5779 mac_len);
5780
5781 __skb_pull(skb, MPLS_HLEN);
5782 skb_reset_mac_header(skb);
5783 skb_set_network_header(skb, mac_len);
5784
5785 if (ethernet && mac_len >= ETH_HLEN) {
5786 struct ethhdr *hdr;
5787
5788 /* use mpls_hdr() to get ethertype to account for VLANs. */
5789 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5790 skb_mod_eth_type(skb, hdr, next_proto);
5791 }
5792 skb->protocol = next_proto;
5793
5794 return 0;
5795}
5796EXPORT_SYMBOL_GPL(skb_mpls_pop);
5797
5798/**
5799 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
5800 *
5801 * @skb: buffer
5802 * @mpls_lse: new MPLS label stack entry to update to
5803 *
5804 * Expects skb->data at mac header.
5805 *
5806 * Returns 0 on success, -errno otherwise.
5807 */
5808int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5809{
5810 int err;
5811
5812 if (unlikely(!eth_p_mpls(skb->protocol)))
5813 return -EINVAL;
5814
5815 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5816 if (unlikely(err))
5817 return err;
5818
5819 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5820 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5821
5822 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5823 }
5824
5825 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5826
5827 return 0;
5828}
5829EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5830
5831/**
5832 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
5833 *
5834 * @skb: buffer
5835 *
5836 * Expects skb->data at mac header.
5837 *
5838 * Returns 0 on success, -errno otherwise.
5839 */
5840int skb_mpls_dec_ttl(struct sk_buff *skb)
5841{
5842 u32 lse;
5843 u8 ttl;
5844
5845 if (unlikely(!eth_p_mpls(skb->protocol)))
5846 return -EINVAL;
5847
5848 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
5849 return -ENOMEM;
5850
5851 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5852 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5853 if (!--ttl)
5854 return -EINVAL;
5855
5856 lse &= ~MPLS_LS_TTL_MASK;
5857 lse |= ttl << MPLS_LS_TTL_SHIFT;
5858
5859 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5860}
5861EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5862
5863/**
5864 * alloc_skb_with_frags - allocate skb with page frags
5865 *
5866 * @header_len: size of linear part
5867 * @data_len: needed length in frags
5868 * @max_page_order: max page order desired.
5869 * @errcode: pointer to error code if any
5870 * @gfp_mask: allocation mask
5871 *
5872 * This can be used to allocate a paged skb, given a maximal order for frags.
5873 */
5874struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5875 unsigned long data_len,
5876 int max_page_order,
5877 int *errcode,
5878 gfp_t gfp_mask)
5879{
5880 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5881 unsigned long chunk;
5882 struct sk_buff *skb;
5883 struct page *page;
5884 int i;
5885
5886 *errcode = -EMSGSIZE;
5887 /* Note this test could be relaxed, if we succeed to allocate
5888 * high order pages...
5889 */
5890 if (npages > MAX_SKB_FRAGS)
5891 return NULL;
5892
5893 *errcode = -ENOBUFS;
5894 skb = alloc_skb(header_len, gfp_mask);
5895 if (!skb)
5896 return NULL;
5897
5898 skb->truesize += npages << PAGE_SHIFT;
5899
5900 for (i = 0; npages > 0; i++) {
5901 int order = max_page_order;
5902
5903 while (order) {
5904 if (npages >= 1 << order) {
5905 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5906 __GFP_COMP |
5907 __GFP_NOWARN,
5908 order);
5909 if (page)
5910 goto fill_page;
5911 /* Do not retry other high order allocations */
5912 order = 1;
5913 max_page_order = 0;
5914 }
5915 order--;
5916 }
5917 page = alloc_page(gfp_mask);
5918 if (!page)
5919 goto failure;
5920fill_page:
5921 chunk = min_t(unsigned long, data_len,
5922 PAGE_SIZE << order);
5923 skb_fill_page_desc(skb, i, page, 0, chunk);
5924 data_len -= chunk;
5925 npages -= 1 << order;
5926 }
5927 return skb;
5928
5929failure:
5930 kfree_skb(skb);
5931 return NULL;
5932}
5933EXPORT_SYMBOL(alloc_skb_with_frags);
5934
5935/* carve out the first off bytes from skb when off < headlen */
5936static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5937 const int headlen, gfp_t gfp_mask)
5938{
5939 int i;
5940 int size = skb_end_offset(skb);
5941 int new_hlen = headlen - off;
5942 u8 *data;
5943
5944 size = SKB_DATA_ALIGN(size);
5945
5946 if (skb_pfmemalloc(skb))
5947 gfp_mask |= __GFP_MEMALLOC;
5948 data = kmalloc_reserve(size +
5949 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5950 gfp_mask, NUMA_NO_NODE, NULL);
5951 if (!data)
5952 return -ENOMEM;
5953
5954 size = SKB_WITH_OVERHEAD(ksize(data));
5955
5956 /* Copy real data, and all frags */
5957 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5958 skb->len -= off;
5959
5960 memcpy((struct skb_shared_info *)(data + size),
5961 skb_shinfo(skb),
5962 offsetof(struct skb_shared_info,
5963 frags[skb_shinfo(skb)->nr_frags]));
5964 if (skb_cloned(skb)) {
5965 /* drop the old head gracefully */
5966 if (skb_orphan_frags(skb, gfp_mask)) {
5967 kfree(data);
5968 return -ENOMEM;
5969 }
5970 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5971 skb_frag_ref(skb, i);
5972 if (skb_has_frag_list(skb))
5973 skb_clone_fraglist(skb);
5974 skb_release_data(skb);
5975 } else {
5976 /* we can reuse existing recount- all we did was
5977 * relocate values
5978 */
5979 skb_free_head(skb);
5980 }
5981
5982 skb->head = data;
5983 skb->data = data;
5984 skb->head_frag = 0;
5985#ifdef NET_SKBUFF_DATA_USES_OFFSET
5986 skb->end = size;
5987#else
5988 skb->end = skb->head + size;
5989#endif
5990 skb_set_tail_pointer(skb, skb_headlen(skb));
5991 skb_headers_offset_update(skb, 0);
5992 skb->cloned = 0;
5993 skb->hdr_len = 0;
5994 skb->nohdr = 0;
5995 atomic_set(&skb_shinfo(skb)->dataref, 1);
5996
5997 return 0;
5998}
5999
6000static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6001
6002/* carve out the first eat bytes from skb's frag_list. May recurse into
6003 * pskb_carve()
6004 */
6005static int pskb_carve_frag_list(struct sk_buff *skb,
6006 struct skb_shared_info *shinfo, int eat,
6007 gfp_t gfp_mask)
6008{
6009 struct sk_buff *list = shinfo->frag_list;
6010 struct sk_buff *clone = NULL;
6011 struct sk_buff *insp = NULL;
6012
6013 do {
6014 if (!list) {
6015 pr_err("Not enough bytes to eat. Want %d\n", eat);
6016 return -EFAULT;
6017 }
6018 if (list->len <= eat) {
6019 /* Eaten as whole. */
6020 eat -= list->len;
6021 list = list->next;
6022 insp = list;
6023 } else {
6024 /* Eaten partially. */
6025 if (skb_shared(list)) {
6026 clone = skb_clone(list, gfp_mask);
6027 if (!clone)
6028 return -ENOMEM;
6029 insp = list->next;
6030 list = clone;
6031 } else {
6032 /* This may be pulled without problems. */
6033 insp = list;
6034 }
6035 if (pskb_carve(list, eat, gfp_mask) < 0) {
6036 kfree_skb(clone);
6037 return -ENOMEM;
6038 }
6039 break;
6040 }
6041 } while (eat);
6042
6043 /* Free pulled out fragments. */
6044 while ((list = shinfo->frag_list) != insp) {
6045 shinfo->frag_list = list->next;
6046 kfree_skb(list);
6047 }
6048 /* And insert new clone at head. */
6049 if (clone) {
6050 clone->next = list;
6051 shinfo->frag_list = clone;
6052 }
6053 return 0;
6054}
6055
6056/* carve off first len bytes from skb. Split line (off) is in the
6057 * non-linear part of skb
6058 */
6059static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6060 int pos, gfp_t gfp_mask)
6061{
6062 int i, k = 0;
6063 int size = skb_end_offset(skb);
6064 u8 *data;
6065 const int nfrags = skb_shinfo(skb)->nr_frags;
6066 struct skb_shared_info *shinfo;
6067
6068 size = SKB_DATA_ALIGN(size);
6069
6070 if (skb_pfmemalloc(skb))
6071 gfp_mask |= __GFP_MEMALLOC;
6072 data = kmalloc_reserve(size +
6073 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
6074 gfp_mask, NUMA_NO_NODE, NULL);
6075 if (!data)
6076 return -ENOMEM;
6077
6078 size = SKB_WITH_OVERHEAD(ksize(data));
6079
6080 memcpy((struct skb_shared_info *)(data + size),
6081 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6082 if (skb_orphan_frags(skb, gfp_mask)) {
6083 kfree(data);
6084 return -ENOMEM;
6085 }
6086 shinfo = (struct skb_shared_info *)(data + size);
6087 for (i = 0; i < nfrags; i++) {
6088 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6089
6090 if (pos + fsize > off) {
6091 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6092
6093 if (pos < off) {
6094 /* Split frag.
6095 * We have two variants in this case:
6096 * 1. Move all the frag to the second
6097 * part, if it is possible. F.e.
6098 * this approach is mandatory for TUX,
6099 * where splitting is expensive.
6100 * 2. Split is accurately. We make this.
6101 */
6102 skb_frag_off_add(&shinfo->frags[0], off - pos);
6103 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6104 }
6105 skb_frag_ref(skb, i);
6106 k++;
6107 }
6108 pos += fsize;
6109 }
6110 shinfo->nr_frags = k;
6111 if (skb_has_frag_list(skb))
6112 skb_clone_fraglist(skb);
6113
6114 /* split line is in frag list */
6115 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6116 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6117 if (skb_has_frag_list(skb))
6118 kfree_skb_list(skb_shinfo(skb)->frag_list);
6119 kfree(data);
6120 return -ENOMEM;
6121 }
6122 skb_release_data(skb);
6123
6124 skb->head = data;
6125 skb->head_frag = 0;
6126 skb->data = data;
6127#ifdef NET_SKBUFF_DATA_USES_OFFSET
6128 skb->end = size;
6129#else
6130 skb->end = skb->head + size;
6131#endif
6132 skb_reset_tail_pointer(skb);
6133 skb_headers_offset_update(skb, 0);
6134 skb->cloned = 0;
6135 skb->hdr_len = 0;
6136 skb->nohdr = 0;
6137 skb->len -= off;
6138 skb->data_len = skb->len;
6139 atomic_set(&skb_shinfo(skb)->dataref, 1);
6140 return 0;
6141}
6142
6143/* remove len bytes from the beginning of the skb */
6144static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6145{
6146 int headlen = skb_headlen(skb);
6147
6148 if (len < headlen)
6149 return pskb_carve_inside_header(skb, len, headlen, gfp);
6150 else
6151 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6152}
6153
6154/* Extract to_copy bytes starting at off from skb, and return this in
6155 * a new skb
6156 */
6157struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6158 int to_copy, gfp_t gfp)
6159{
6160 struct sk_buff *clone = skb_clone(skb, gfp);
6161
6162 if (!clone)
6163 return NULL;
6164
6165 if (pskb_carve(clone, off, gfp) < 0 ||
6166 pskb_trim(clone, to_copy)) {
6167 kfree_skb(clone);
6168 return NULL;
6169 }
6170 return clone;
6171}
6172EXPORT_SYMBOL(pskb_extract);
6173
6174/**
6175 * skb_condense - try to get rid of fragments/frag_list if possible
6176 * @skb: buffer
6177 *
6178 * Can be used to save memory before skb is added to a busy queue.
6179 * If packet has bytes in frags and enough tail room in skb->head,
6180 * pull all of them, so that we can free the frags right now and adjust
6181 * truesize.
6182 * Notes:
6183 * We do not reallocate skb->head thus can not fail.
6184 * Caller must re-evaluate skb->truesize if needed.
6185 */
6186void skb_condense(struct sk_buff *skb)
6187{
6188 if (skb->data_len) {
6189 if (skb->data_len > skb->end - skb->tail ||
6190 skb_cloned(skb))
6191 return;
6192
6193 /* Nice, we can free page frag(s) right now */
6194 __pskb_pull_tail(skb, skb->data_len);
6195 }
6196 /* At this point, skb->truesize might be over estimated,
6197 * because skb had a fragment, and fragments do not tell
6198 * their truesize.
6199 * When we pulled its content into skb->head, fragment
6200 * was freed, but __pskb_pull_tail() could not possibly
6201 * adjust skb->truesize, not knowing the frag truesize.
6202 */
6203 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6204}
6205
6206#ifdef CONFIG_SKB_EXTENSIONS
6207static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6208{
6209 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6210}
6211
6212/**
6213 * __skb_ext_alloc - allocate a new skb extensions storage
6214 *
6215 * @flags: See kmalloc().
6216 *
6217 * Returns the newly allocated pointer. The pointer can later attached to a
6218 * skb via __skb_ext_set().
6219 * Note: caller must handle the skb_ext as an opaque data.
6220 */
6221struct skb_ext *__skb_ext_alloc(gfp_t flags)
6222{
6223 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6224
6225 if (new) {
6226 memset(new->offset, 0, sizeof(new->offset));
6227 refcount_set(&new->refcnt, 1);
6228 }
6229
6230 return new;
6231}
6232
6233static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6234 unsigned int old_active)
6235{
6236 struct skb_ext *new;
6237
6238 if (refcount_read(&old->refcnt) == 1)
6239 return old;
6240
6241 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6242 if (!new)
6243 return NULL;
6244
6245 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6246 refcount_set(&new->refcnt, 1);
6247
6248#ifdef CONFIG_XFRM
6249 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6250 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6251 unsigned int i;
6252
6253 for (i = 0; i < sp->len; i++)
6254 xfrm_state_hold(sp->xvec[i]);
6255 }
6256#endif
6257 __skb_ext_put(old);
6258 return new;
6259}
6260
6261/**
6262 * __skb_ext_set - attach the specified extension storage to this skb
6263 * @skb: buffer
6264 * @id: extension id
6265 * @ext: extension storage previously allocated via __skb_ext_alloc()
6266 *
6267 * Existing extensions, if any, are cleared.
6268 *
6269 * Returns the pointer to the extension.
6270 */
6271void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6272 struct skb_ext *ext)
6273{
6274 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6275
6276 skb_ext_put(skb);
6277 newlen = newoff + skb_ext_type_len[id];
6278 ext->chunks = newlen;
6279 ext->offset[id] = newoff;
6280 skb->extensions = ext;
6281 skb->active_extensions = 1 << id;
6282 return skb_ext_get_ptr(ext, id);
6283}
6284
6285/**
6286 * skb_ext_add - allocate space for given extension, COW if needed
6287 * @skb: buffer
6288 * @id: extension to allocate space for
6289 *
6290 * Allocates enough space for the given extension.
6291 * If the extension is already present, a pointer to that extension
6292 * is returned.
6293 *
6294 * If the skb was cloned, COW applies and the returned memory can be
6295 * modified without changing the extension space of clones buffers.
6296 *
6297 * Returns pointer to the extension or NULL on allocation failure.
6298 */
6299void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6300{
6301 struct skb_ext *new, *old = NULL;
6302 unsigned int newlen, newoff;
6303
6304 if (skb->active_extensions) {
6305 old = skb->extensions;
6306
6307 new = skb_ext_maybe_cow(old, skb->active_extensions);
6308 if (!new)
6309 return NULL;
6310
6311 if (__skb_ext_exist(new, id))
6312 goto set_active;
6313
6314 newoff = new->chunks;
6315 } else {
6316 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6317
6318 new = __skb_ext_alloc(GFP_ATOMIC);
6319 if (!new)
6320 return NULL;
6321 }
6322
6323 newlen = newoff + skb_ext_type_len[id];
6324 new->chunks = newlen;
6325 new->offset[id] = newoff;
6326set_active:
6327 skb->extensions = new;
6328 skb->active_extensions |= 1 << id;
6329 return skb_ext_get_ptr(new, id);
6330}
6331EXPORT_SYMBOL(skb_ext_add);
6332
6333#ifdef CONFIG_XFRM
6334static void skb_ext_put_sp(struct sec_path *sp)
6335{
6336 unsigned int i;
6337
6338 for (i = 0; i < sp->len; i++)
6339 xfrm_state_put(sp->xvec[i]);
6340}
6341#endif
6342
6343void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6344{
6345 struct skb_ext *ext = skb->extensions;
6346
6347 skb->active_extensions &= ~(1 << id);
6348 if (skb->active_extensions == 0) {
6349 skb->extensions = NULL;
6350 __skb_ext_put(ext);
6351#ifdef CONFIG_XFRM
6352 } else if (id == SKB_EXT_SEC_PATH &&
6353 refcount_read(&ext->refcnt) == 1) {
6354 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6355
6356 skb_ext_put_sp(sp);
6357 sp->len = 0;
6358#endif
6359 }
6360}
6361EXPORT_SYMBOL(__skb_ext_del);
6362
6363void __skb_ext_put(struct skb_ext *ext)
6364{
6365 /* If this is last clone, nothing can increment
6366 * it after check passes. Avoids one atomic op.
6367 */
6368 if (refcount_read(&ext->refcnt) == 1)
6369 goto free_now;
6370
6371 if (!refcount_dec_and_test(&ext->refcnt))
6372 return;
6373free_now:
6374#ifdef CONFIG_XFRM
6375 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6376 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6377#endif
6378
6379 kmem_cache_free(skbuff_ext_cache, ext);
6380}
6381EXPORT_SYMBOL(__skb_ext_put);
6382#endif /* CONFIG_SKB_EXTENSIONS */