Loading...
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/kmemcheck.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/tcp.h>
51#include <linux/udp.h>
52#include <linux/netdevice.h>
53#ifdef CONFIG_NET_CLS_ACT
54#include <net/pkt_sched.h>
55#endif
56#include <linux/string.h>
57#include <linux/skbuff.h>
58#include <linux/splice.h>
59#include <linux/cache.h>
60#include <linux/rtnetlink.h>
61#include <linux/init.h>
62#include <linux/scatterlist.h>
63#include <linux/errqueue.h>
64#include <linux/prefetch.h>
65#include <linux/if_vlan.h>
66
67#include <net/protocol.h>
68#include <net/dst.h>
69#include <net/sock.h>
70#include <net/checksum.h>
71#include <net/ip6_checksum.h>
72#include <net/xfrm.h>
73
74#include <asm/uaccess.h>
75#include <trace/events/skb.h>
76#include <linux/highmem.h>
77#include <linux/capability.h>
78#include <linux/user_namespace.h>
79
80struct kmem_cache *skbuff_head_cache __read_mostly;
81static struct kmem_cache *skbuff_fclone_cache __read_mostly;
82int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83EXPORT_SYMBOL(sysctl_max_skb_frags);
84
85/**
86 * skb_panic - private function for out-of-line support
87 * @skb: buffer
88 * @sz: size
89 * @addr: address
90 * @msg: skb_over_panic or skb_under_panic
91 *
92 * Out-of-line support for skb_put() and skb_push().
93 * Called via the wrapper skb_over_panic() or skb_under_panic().
94 * Keep out of line to prevent kernel bloat.
95 * __builtin_return_address is not used because it is not always reliable.
96 */
97static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
98 const char msg[])
99{
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
104 BUG();
105}
106
107static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
111
112static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117/*
118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
119 * the caller if emergency pfmemalloc reserves are being used. If it is and
120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
121 * may be used. Otherwise, the packet data may be discarded until enough
122 * memory is free
123 */
124#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
126
127static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
129{
130 void *obj;
131 bool ret_pfmemalloc = false;
132
133 /*
134 * Try a regular allocation, when that fails and we're not entitled
135 * to the reserves, fail.
136 */
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
139 node);
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
141 goto out;
142
143 /* Try again but now we are using pfmemalloc reserves */
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
146
147out:
148 if (pfmemalloc)
149 *pfmemalloc = ret_pfmemalloc;
150
151 return obj;
152}
153
154/* Allocate a new skbuff. We do this ourselves so we can fill in a few
155 * 'private' fields and also do memory statistics to find all the
156 * [BEEP] leaks.
157 *
158 */
159
160struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
161{
162 struct sk_buff *skb;
163
164 /* Get the HEAD */
165 skb = kmem_cache_alloc_node(skbuff_head_cache,
166 gfp_mask & ~__GFP_DMA, node);
167 if (!skb)
168 goto out;
169
170 /*
171 * Only clear those fields we need to clear, not those that we will
172 * actually initialise below. Hence, don't put any more fields after
173 * the tail pointer in struct sk_buff!
174 */
175 memset(skb, 0, offsetof(struct sk_buff, tail));
176 skb->head = NULL;
177 skb->truesize = sizeof(struct sk_buff);
178 atomic_set(&skb->users, 1);
179
180 skb->mac_header = (typeof(skb->mac_header))~0U;
181out:
182 return skb;
183}
184
185/**
186 * __alloc_skb - allocate a network buffer
187 * @size: size to allocate
188 * @gfp_mask: allocation mask
189 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
190 * instead of head cache and allocate a cloned (child) skb.
191 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
192 * allocations in case the data is required for writeback
193 * @node: numa node to allocate memory on
194 *
195 * Allocate a new &sk_buff. The returned buffer has no headroom and a
196 * tail room of at least size bytes. The object has a reference count
197 * of one. The return is the buffer. On a failure the return is %NULL.
198 *
199 * Buffers may only be allocated from interrupts using a @gfp_mask of
200 * %GFP_ATOMIC.
201 */
202struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
203 int flags, int node)
204{
205 struct kmem_cache *cache;
206 struct skb_shared_info *shinfo;
207 struct sk_buff *skb;
208 u8 *data;
209 bool pfmemalloc;
210
211 cache = (flags & SKB_ALLOC_FCLONE)
212 ? skbuff_fclone_cache : skbuff_head_cache;
213
214 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
215 gfp_mask |= __GFP_MEMALLOC;
216
217 /* Get the HEAD */
218 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
219 if (!skb)
220 goto out;
221 prefetchw(skb);
222
223 /* We do our best to align skb_shared_info on a separate cache
224 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
225 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
226 * Both skb->head and skb_shared_info are cache line aligned.
227 */
228 size = SKB_DATA_ALIGN(size);
229 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
230 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
231 if (!data)
232 goto nodata;
233 /* kmalloc(size) might give us more room than requested.
234 * Put skb_shared_info exactly at the end of allocated zone,
235 * to allow max possible filling before reallocation.
236 */
237 size = SKB_WITH_OVERHEAD(ksize(data));
238 prefetchw(data + size);
239
240 /*
241 * Only clear those fields we need to clear, not those that we will
242 * actually initialise below. Hence, don't put any more fields after
243 * the tail pointer in struct sk_buff!
244 */
245 memset(skb, 0, offsetof(struct sk_buff, tail));
246 /* Account for allocated memory : skb + skb->head */
247 skb->truesize = SKB_TRUESIZE(size);
248 skb->pfmemalloc = pfmemalloc;
249 atomic_set(&skb->users, 1);
250 skb->head = data;
251 skb->data = data;
252 skb_reset_tail_pointer(skb);
253 skb->end = skb->tail + size;
254 skb->mac_header = (typeof(skb->mac_header))~0U;
255 skb->transport_header = (typeof(skb->transport_header))~0U;
256
257 /* make sure we initialize shinfo sequentially */
258 shinfo = skb_shinfo(skb);
259 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
260 atomic_set(&shinfo->dataref, 1);
261 kmemcheck_annotate_variable(shinfo->destructor_arg);
262
263 if (flags & SKB_ALLOC_FCLONE) {
264 struct sk_buff_fclones *fclones;
265
266 fclones = container_of(skb, struct sk_buff_fclones, skb1);
267
268 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
269 skb->fclone = SKB_FCLONE_ORIG;
270 atomic_set(&fclones->fclone_ref, 1);
271
272 fclones->skb2.fclone = SKB_FCLONE_CLONE;
273 fclones->skb2.pfmemalloc = pfmemalloc;
274 }
275out:
276 return skb;
277nodata:
278 kmem_cache_free(cache, skb);
279 skb = NULL;
280 goto out;
281}
282EXPORT_SYMBOL(__alloc_skb);
283
284/**
285 * __build_skb - build a network buffer
286 * @data: data buffer provided by caller
287 * @frag_size: size of data, or 0 if head was kmalloced
288 *
289 * Allocate a new &sk_buff. Caller provides space holding head and
290 * skb_shared_info. @data must have been allocated by kmalloc() only if
291 * @frag_size is 0, otherwise data should come from the page allocator
292 * or vmalloc()
293 * The return is the new skb buffer.
294 * On a failure the return is %NULL, and @data is not freed.
295 * Notes :
296 * Before IO, driver allocates only data buffer where NIC put incoming frame
297 * Driver should add room at head (NET_SKB_PAD) and
298 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
299 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
300 * before giving packet to stack.
301 * RX rings only contains data buffers, not full skbs.
302 */
303struct sk_buff *__build_skb(void *data, unsigned int frag_size)
304{
305 struct skb_shared_info *shinfo;
306 struct sk_buff *skb;
307 unsigned int size = frag_size ? : ksize(data);
308
309 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
310 if (!skb)
311 return NULL;
312
313 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
314
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316 skb->truesize = SKB_TRUESIZE(size);
317 atomic_set(&skb->users, 1);
318 skb->head = data;
319 skb->data = data;
320 skb_reset_tail_pointer(skb);
321 skb->end = skb->tail + size;
322 skb->mac_header = (typeof(skb->mac_header))~0U;
323 skb->transport_header = (typeof(skb->transport_header))~0U;
324
325 /* make sure we initialize shinfo sequentially */
326 shinfo = skb_shinfo(skb);
327 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
328 atomic_set(&shinfo->dataref, 1);
329 kmemcheck_annotate_variable(shinfo->destructor_arg);
330
331 return skb;
332}
333
334/* build_skb() is wrapper over __build_skb(), that specifically
335 * takes care of skb->head and skb->pfmemalloc
336 * This means that if @frag_size is not zero, then @data must be backed
337 * by a page fragment, not kmalloc() or vmalloc()
338 */
339struct sk_buff *build_skb(void *data, unsigned int frag_size)
340{
341 struct sk_buff *skb = __build_skb(data, frag_size);
342
343 if (skb && frag_size) {
344 skb->head_frag = 1;
345 if (page_is_pfmemalloc(virt_to_head_page(data)))
346 skb->pfmemalloc = 1;
347 }
348 return skb;
349}
350EXPORT_SYMBOL(build_skb);
351
352#define NAPI_SKB_CACHE_SIZE 64
353
354struct napi_alloc_cache {
355 struct page_frag_cache page;
356 size_t skb_count;
357 void *skb_cache[NAPI_SKB_CACHE_SIZE];
358};
359
360static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
361static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
362
363static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
364{
365 struct page_frag_cache *nc;
366 unsigned long flags;
367 void *data;
368
369 local_irq_save(flags);
370 nc = this_cpu_ptr(&netdev_alloc_cache);
371 data = __alloc_page_frag(nc, fragsz, gfp_mask);
372 local_irq_restore(flags);
373 return data;
374}
375
376/**
377 * netdev_alloc_frag - allocate a page fragment
378 * @fragsz: fragment size
379 *
380 * Allocates a frag from a page for receive buffer.
381 * Uses GFP_ATOMIC allocations.
382 */
383void *netdev_alloc_frag(unsigned int fragsz)
384{
385 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
386}
387EXPORT_SYMBOL(netdev_alloc_frag);
388
389static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
390{
391 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
392
393 return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
394}
395
396void *napi_alloc_frag(unsigned int fragsz)
397{
398 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
399}
400EXPORT_SYMBOL(napi_alloc_frag);
401
402/**
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @len: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
407 *
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has NET_SKB_PAD headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
412 *
413 * %NULL is returned if there is no free memory.
414 */
415struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
416 gfp_t gfp_mask)
417{
418 struct page_frag_cache *nc;
419 unsigned long flags;
420 struct sk_buff *skb;
421 bool pfmemalloc;
422 void *data;
423
424 len += NET_SKB_PAD;
425
426 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
427 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
428 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
429 if (!skb)
430 goto skb_fail;
431 goto skb_success;
432 }
433
434 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
435 len = SKB_DATA_ALIGN(len);
436
437 if (sk_memalloc_socks())
438 gfp_mask |= __GFP_MEMALLOC;
439
440 local_irq_save(flags);
441
442 nc = this_cpu_ptr(&netdev_alloc_cache);
443 data = __alloc_page_frag(nc, len, gfp_mask);
444 pfmemalloc = nc->pfmemalloc;
445
446 local_irq_restore(flags);
447
448 if (unlikely(!data))
449 return NULL;
450
451 skb = __build_skb(data, len);
452 if (unlikely(!skb)) {
453 skb_free_frag(data);
454 return NULL;
455 }
456
457 /* use OR instead of assignment to avoid clearing of bits in mask */
458 if (pfmemalloc)
459 skb->pfmemalloc = 1;
460 skb->head_frag = 1;
461
462skb_success:
463 skb_reserve(skb, NET_SKB_PAD);
464 skb->dev = dev;
465
466skb_fail:
467 return skb;
468}
469EXPORT_SYMBOL(__netdev_alloc_skb);
470
471/**
472 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
473 * @napi: napi instance this buffer was allocated for
474 * @len: length to allocate
475 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
476 *
477 * Allocate a new sk_buff for use in NAPI receive. This buffer will
478 * attempt to allocate the head from a special reserved region used
479 * only for NAPI Rx allocation. By doing this we can save several
480 * CPU cycles by avoiding having to disable and re-enable IRQs.
481 *
482 * %NULL is returned if there is no free memory.
483 */
484struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
485 gfp_t gfp_mask)
486{
487 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
488 struct sk_buff *skb;
489 void *data;
490
491 len += NET_SKB_PAD + NET_IP_ALIGN;
492
493 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
494 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
495 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
496 if (!skb)
497 goto skb_fail;
498 goto skb_success;
499 }
500
501 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
502 len = SKB_DATA_ALIGN(len);
503
504 if (sk_memalloc_socks())
505 gfp_mask |= __GFP_MEMALLOC;
506
507 data = __alloc_page_frag(&nc->page, len, gfp_mask);
508 if (unlikely(!data))
509 return NULL;
510
511 skb = __build_skb(data, len);
512 if (unlikely(!skb)) {
513 skb_free_frag(data);
514 return NULL;
515 }
516
517 /* use OR instead of assignment to avoid clearing of bits in mask */
518 if (nc->page.pfmemalloc)
519 skb->pfmemalloc = 1;
520 skb->head_frag = 1;
521
522skb_success:
523 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
524 skb->dev = napi->dev;
525
526skb_fail:
527 return skb;
528}
529EXPORT_SYMBOL(__napi_alloc_skb);
530
531void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
532 int size, unsigned int truesize)
533{
534 skb_fill_page_desc(skb, i, page, off, size);
535 skb->len += size;
536 skb->data_len += size;
537 skb->truesize += truesize;
538}
539EXPORT_SYMBOL(skb_add_rx_frag);
540
541void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
542 unsigned int truesize)
543{
544 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
545
546 skb_frag_size_add(frag, size);
547 skb->len += size;
548 skb->data_len += size;
549 skb->truesize += truesize;
550}
551EXPORT_SYMBOL(skb_coalesce_rx_frag);
552
553static void skb_drop_list(struct sk_buff **listp)
554{
555 kfree_skb_list(*listp);
556 *listp = NULL;
557}
558
559static inline void skb_drop_fraglist(struct sk_buff *skb)
560{
561 skb_drop_list(&skb_shinfo(skb)->frag_list);
562}
563
564static void skb_clone_fraglist(struct sk_buff *skb)
565{
566 struct sk_buff *list;
567
568 skb_walk_frags(skb, list)
569 skb_get(list);
570}
571
572static void skb_free_head(struct sk_buff *skb)
573{
574 unsigned char *head = skb->head;
575
576 if (skb->head_frag)
577 skb_free_frag(head);
578 else
579 kfree(head);
580}
581
582static void skb_release_data(struct sk_buff *skb)
583{
584 struct skb_shared_info *shinfo = skb_shinfo(skb);
585 int i;
586
587 if (skb->cloned &&
588 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
589 &shinfo->dataref))
590 return;
591
592 for (i = 0; i < shinfo->nr_frags; i++)
593 __skb_frag_unref(&shinfo->frags[i]);
594
595 /*
596 * If skb buf is from userspace, we need to notify the caller
597 * the lower device DMA has done;
598 */
599 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
600 struct ubuf_info *uarg;
601
602 uarg = shinfo->destructor_arg;
603 if (uarg->callback)
604 uarg->callback(uarg, true);
605 }
606
607 if (shinfo->frag_list)
608 kfree_skb_list(shinfo->frag_list);
609
610 skb_free_head(skb);
611}
612
613/*
614 * Free an skbuff by memory without cleaning the state.
615 */
616static void kfree_skbmem(struct sk_buff *skb)
617{
618 struct sk_buff_fclones *fclones;
619
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
623 return;
624
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
627
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
631 */
632 if (atomic_read(&fclones->fclone_ref) == 1)
633 goto fastpath;
634 break;
635
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
638 break;
639 }
640 if (!atomic_dec_and_test(&fclones->fclone_ref))
641 return;
642fastpath:
643 kmem_cache_free(skbuff_fclone_cache, fclones);
644}
645
646static void skb_release_head_state(struct sk_buff *skb)
647{
648 skb_dst_drop(skb);
649#ifdef CONFIG_XFRM
650 secpath_put(skb->sp);
651#endif
652 if (skb->destructor) {
653 WARN_ON(in_irq());
654 skb->destructor(skb);
655 }
656#if IS_ENABLED(CONFIG_NF_CONNTRACK)
657 nf_conntrack_put(skb->nfct);
658#endif
659#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
660 nf_bridge_put(skb->nf_bridge);
661#endif
662}
663
664/* Free everything but the sk_buff shell. */
665static void skb_release_all(struct sk_buff *skb)
666{
667 skb_release_head_state(skb);
668 if (likely(skb->head))
669 skb_release_data(skb);
670}
671
672/**
673 * __kfree_skb - private function
674 * @skb: buffer
675 *
676 * Free an sk_buff. Release anything attached to the buffer.
677 * Clean the state. This is an internal helper function. Users should
678 * always call kfree_skb
679 */
680
681void __kfree_skb(struct sk_buff *skb)
682{
683 skb_release_all(skb);
684 kfree_skbmem(skb);
685}
686EXPORT_SYMBOL(__kfree_skb);
687
688/**
689 * kfree_skb - free an sk_buff
690 * @skb: buffer to free
691 *
692 * Drop a reference to the buffer and free it if the usage count has
693 * hit zero.
694 */
695void kfree_skb(struct sk_buff *skb)
696{
697 if (unlikely(!skb))
698 return;
699 if (likely(atomic_read(&skb->users) == 1))
700 smp_rmb();
701 else if (likely(!atomic_dec_and_test(&skb->users)))
702 return;
703 trace_kfree_skb(skb, __builtin_return_address(0));
704 __kfree_skb(skb);
705}
706EXPORT_SYMBOL(kfree_skb);
707
708void kfree_skb_list(struct sk_buff *segs)
709{
710 while (segs) {
711 struct sk_buff *next = segs->next;
712
713 kfree_skb(segs);
714 segs = next;
715 }
716}
717EXPORT_SYMBOL(kfree_skb_list);
718
719/**
720 * skb_tx_error - report an sk_buff xmit error
721 * @skb: buffer that triggered an error
722 *
723 * Report xmit error if a device callback is tracking this skb.
724 * skb must be freed afterwards.
725 */
726void skb_tx_error(struct sk_buff *skb)
727{
728 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
729 struct ubuf_info *uarg;
730
731 uarg = skb_shinfo(skb)->destructor_arg;
732 if (uarg->callback)
733 uarg->callback(uarg, false);
734 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
735 }
736}
737EXPORT_SYMBOL(skb_tx_error);
738
739/**
740 * consume_skb - free an skbuff
741 * @skb: buffer to free
742 *
743 * Drop a ref to the buffer and free it if the usage count has hit zero
744 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
745 * is being dropped after a failure and notes that
746 */
747void consume_skb(struct sk_buff *skb)
748{
749 if (unlikely(!skb))
750 return;
751 if (likely(atomic_read(&skb->users) == 1))
752 smp_rmb();
753 else if (likely(!atomic_dec_and_test(&skb->users)))
754 return;
755 trace_consume_skb(skb);
756 __kfree_skb(skb);
757}
758EXPORT_SYMBOL(consume_skb);
759
760void __kfree_skb_flush(void)
761{
762 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
763
764 /* flush skb_cache if containing objects */
765 if (nc->skb_count) {
766 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
767 nc->skb_cache);
768 nc->skb_count = 0;
769 }
770}
771
772static inline void _kfree_skb_defer(struct sk_buff *skb)
773{
774 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
775
776 /* drop skb->head and call any destructors for packet */
777 skb_release_all(skb);
778
779 /* record skb to CPU local list */
780 nc->skb_cache[nc->skb_count++] = skb;
781
782#ifdef CONFIG_SLUB
783 /* SLUB writes into objects when freeing */
784 prefetchw(skb);
785#endif
786
787 /* flush skb_cache if it is filled */
788 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
789 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
790 nc->skb_cache);
791 nc->skb_count = 0;
792 }
793}
794void __kfree_skb_defer(struct sk_buff *skb)
795{
796 _kfree_skb_defer(skb);
797}
798
799void napi_consume_skb(struct sk_buff *skb, int budget)
800{
801 if (unlikely(!skb))
802 return;
803
804 /* Zero budget indicate non-NAPI context called us, like netpoll */
805 if (unlikely(!budget)) {
806 dev_consume_skb_any(skb);
807 return;
808 }
809
810 if (likely(atomic_read(&skb->users) == 1))
811 smp_rmb();
812 else if (likely(!atomic_dec_and_test(&skb->users)))
813 return;
814 /* if reaching here SKB is ready to free */
815 trace_consume_skb(skb);
816
817 /* if SKB is a clone, don't handle this case */
818 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
819 __kfree_skb(skb);
820 return;
821 }
822
823 _kfree_skb_defer(skb);
824}
825EXPORT_SYMBOL(napi_consume_skb);
826
827/* Make sure a field is enclosed inside headers_start/headers_end section */
828#define CHECK_SKB_FIELD(field) \
829 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
830 offsetof(struct sk_buff, headers_start)); \
831 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
832 offsetof(struct sk_buff, headers_end)); \
833
834static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
835{
836 new->tstamp = old->tstamp;
837 /* We do not copy old->sk */
838 new->dev = old->dev;
839 memcpy(new->cb, old->cb, sizeof(old->cb));
840 skb_dst_copy(new, old);
841#ifdef CONFIG_XFRM
842 new->sp = secpath_get(old->sp);
843#endif
844 __nf_copy(new, old, false);
845
846 /* Note : this field could be in headers_start/headers_end section
847 * It is not yet because we do not want to have a 16 bit hole
848 */
849 new->queue_mapping = old->queue_mapping;
850
851 memcpy(&new->headers_start, &old->headers_start,
852 offsetof(struct sk_buff, headers_end) -
853 offsetof(struct sk_buff, headers_start));
854 CHECK_SKB_FIELD(protocol);
855 CHECK_SKB_FIELD(csum);
856 CHECK_SKB_FIELD(hash);
857 CHECK_SKB_FIELD(priority);
858 CHECK_SKB_FIELD(skb_iif);
859 CHECK_SKB_FIELD(vlan_proto);
860 CHECK_SKB_FIELD(vlan_tci);
861 CHECK_SKB_FIELD(transport_header);
862 CHECK_SKB_FIELD(network_header);
863 CHECK_SKB_FIELD(mac_header);
864 CHECK_SKB_FIELD(inner_protocol);
865 CHECK_SKB_FIELD(inner_transport_header);
866 CHECK_SKB_FIELD(inner_network_header);
867 CHECK_SKB_FIELD(inner_mac_header);
868 CHECK_SKB_FIELD(mark);
869#ifdef CONFIG_NETWORK_SECMARK
870 CHECK_SKB_FIELD(secmark);
871#endif
872#ifdef CONFIG_NET_RX_BUSY_POLL
873 CHECK_SKB_FIELD(napi_id);
874#endif
875#ifdef CONFIG_XPS
876 CHECK_SKB_FIELD(sender_cpu);
877#endif
878#ifdef CONFIG_NET_SCHED
879 CHECK_SKB_FIELD(tc_index);
880#ifdef CONFIG_NET_CLS_ACT
881 CHECK_SKB_FIELD(tc_verd);
882#endif
883#endif
884
885}
886
887/*
888 * You should not add any new code to this function. Add it to
889 * __copy_skb_header above instead.
890 */
891static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
892{
893#define C(x) n->x = skb->x
894
895 n->next = n->prev = NULL;
896 n->sk = NULL;
897 __copy_skb_header(n, skb);
898
899 C(len);
900 C(data_len);
901 C(mac_len);
902 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
903 n->cloned = 1;
904 n->nohdr = 0;
905 n->destructor = NULL;
906 C(tail);
907 C(end);
908 C(head);
909 C(head_frag);
910 C(data);
911 C(truesize);
912 atomic_set(&n->users, 1);
913
914 atomic_inc(&(skb_shinfo(skb)->dataref));
915 skb->cloned = 1;
916
917 return n;
918#undef C
919}
920
921/**
922 * skb_morph - morph one skb into another
923 * @dst: the skb to receive the contents
924 * @src: the skb to supply the contents
925 *
926 * This is identical to skb_clone except that the target skb is
927 * supplied by the user.
928 *
929 * The target skb is returned upon exit.
930 */
931struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
932{
933 skb_release_all(dst);
934 return __skb_clone(dst, src);
935}
936EXPORT_SYMBOL_GPL(skb_morph);
937
938/**
939 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
940 * @skb: the skb to modify
941 * @gfp_mask: allocation priority
942 *
943 * This must be called on SKBTX_DEV_ZEROCOPY skb.
944 * It will copy all frags into kernel and drop the reference
945 * to userspace pages.
946 *
947 * If this function is called from an interrupt gfp_mask() must be
948 * %GFP_ATOMIC.
949 *
950 * Returns 0 on success or a negative error code on failure
951 * to allocate kernel memory to copy to.
952 */
953int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
954{
955 int i;
956 int num_frags = skb_shinfo(skb)->nr_frags;
957 struct page *page, *head = NULL;
958 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
959
960 for (i = 0; i < num_frags; i++) {
961 u8 *vaddr;
962 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
963
964 page = alloc_page(gfp_mask);
965 if (!page) {
966 while (head) {
967 struct page *next = (struct page *)page_private(head);
968 put_page(head);
969 head = next;
970 }
971 return -ENOMEM;
972 }
973 vaddr = kmap_atomic(skb_frag_page(f));
974 memcpy(page_address(page),
975 vaddr + f->page_offset, skb_frag_size(f));
976 kunmap_atomic(vaddr);
977 set_page_private(page, (unsigned long)head);
978 head = page;
979 }
980
981 /* skb frags release userspace buffers */
982 for (i = 0; i < num_frags; i++)
983 skb_frag_unref(skb, i);
984
985 uarg->callback(uarg, false);
986
987 /* skb frags point to kernel buffers */
988 for (i = num_frags - 1; i >= 0; i--) {
989 __skb_fill_page_desc(skb, i, head, 0,
990 skb_shinfo(skb)->frags[i].size);
991 head = (struct page *)page_private(head);
992 }
993
994 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
995 return 0;
996}
997EXPORT_SYMBOL_GPL(skb_copy_ubufs);
998
999/**
1000 * skb_clone - duplicate an sk_buff
1001 * @skb: buffer to clone
1002 * @gfp_mask: allocation priority
1003 *
1004 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1005 * copies share the same packet data but not structure. The new
1006 * buffer has a reference count of 1. If the allocation fails the
1007 * function returns %NULL otherwise the new buffer is returned.
1008 *
1009 * If this function is called from an interrupt gfp_mask() must be
1010 * %GFP_ATOMIC.
1011 */
1012
1013struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1014{
1015 struct sk_buff_fclones *fclones = container_of(skb,
1016 struct sk_buff_fclones,
1017 skb1);
1018 struct sk_buff *n;
1019
1020 if (skb_orphan_frags(skb, gfp_mask))
1021 return NULL;
1022
1023 if (skb->fclone == SKB_FCLONE_ORIG &&
1024 atomic_read(&fclones->fclone_ref) == 1) {
1025 n = &fclones->skb2;
1026 atomic_set(&fclones->fclone_ref, 2);
1027 } else {
1028 if (skb_pfmemalloc(skb))
1029 gfp_mask |= __GFP_MEMALLOC;
1030
1031 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1032 if (!n)
1033 return NULL;
1034
1035 kmemcheck_annotate_bitfield(n, flags1);
1036 n->fclone = SKB_FCLONE_UNAVAILABLE;
1037 }
1038
1039 return __skb_clone(n, skb);
1040}
1041EXPORT_SYMBOL(skb_clone);
1042
1043static void skb_headers_offset_update(struct sk_buff *skb, int off)
1044{
1045 /* Only adjust this if it actually is csum_start rather than csum */
1046 if (skb->ip_summed == CHECKSUM_PARTIAL)
1047 skb->csum_start += off;
1048 /* {transport,network,mac}_header and tail are relative to skb->head */
1049 skb->transport_header += off;
1050 skb->network_header += off;
1051 if (skb_mac_header_was_set(skb))
1052 skb->mac_header += off;
1053 skb->inner_transport_header += off;
1054 skb->inner_network_header += off;
1055 skb->inner_mac_header += off;
1056}
1057
1058static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1059{
1060 __copy_skb_header(new, old);
1061
1062 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1063 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1064 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1065}
1066
1067static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1068{
1069 if (skb_pfmemalloc(skb))
1070 return SKB_ALLOC_RX;
1071 return 0;
1072}
1073
1074/**
1075 * skb_copy - create private copy of an sk_buff
1076 * @skb: buffer to copy
1077 * @gfp_mask: allocation priority
1078 *
1079 * Make a copy of both an &sk_buff and its data. This is used when the
1080 * caller wishes to modify the data and needs a private copy of the
1081 * data to alter. Returns %NULL on failure or the pointer to the buffer
1082 * on success. The returned buffer has a reference count of 1.
1083 *
1084 * As by-product this function converts non-linear &sk_buff to linear
1085 * one, so that &sk_buff becomes completely private and caller is allowed
1086 * to modify all the data of returned buffer. This means that this
1087 * function is not recommended for use in circumstances when only
1088 * header is going to be modified. Use pskb_copy() instead.
1089 */
1090
1091struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1092{
1093 int headerlen = skb_headroom(skb);
1094 unsigned int size = skb_end_offset(skb) + skb->data_len;
1095 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1096 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1097
1098 if (!n)
1099 return NULL;
1100
1101 /* Set the data pointer */
1102 skb_reserve(n, headerlen);
1103 /* Set the tail pointer and length */
1104 skb_put(n, skb->len);
1105
1106 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1107 BUG();
1108
1109 copy_skb_header(n, skb);
1110 return n;
1111}
1112EXPORT_SYMBOL(skb_copy);
1113
1114/**
1115 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1116 * @skb: buffer to copy
1117 * @headroom: headroom of new skb
1118 * @gfp_mask: allocation priority
1119 * @fclone: if true allocate the copy of the skb from the fclone
1120 * cache instead of the head cache; it is recommended to set this
1121 * to true for the cases where the copy will likely be cloned
1122 *
1123 * Make a copy of both an &sk_buff and part of its data, located
1124 * in header. Fragmented data remain shared. This is used when
1125 * the caller wishes to modify only header of &sk_buff and needs
1126 * private copy of the header to alter. Returns %NULL on failure
1127 * or the pointer to the buffer on success.
1128 * The returned buffer has a reference count of 1.
1129 */
1130
1131struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1132 gfp_t gfp_mask, bool fclone)
1133{
1134 unsigned int size = skb_headlen(skb) + headroom;
1135 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1136 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1137
1138 if (!n)
1139 goto out;
1140
1141 /* Set the data pointer */
1142 skb_reserve(n, headroom);
1143 /* Set the tail pointer and length */
1144 skb_put(n, skb_headlen(skb));
1145 /* Copy the bytes */
1146 skb_copy_from_linear_data(skb, n->data, n->len);
1147
1148 n->truesize += skb->data_len;
1149 n->data_len = skb->data_len;
1150 n->len = skb->len;
1151
1152 if (skb_shinfo(skb)->nr_frags) {
1153 int i;
1154
1155 if (skb_orphan_frags(skb, gfp_mask)) {
1156 kfree_skb(n);
1157 n = NULL;
1158 goto out;
1159 }
1160 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1162 skb_frag_ref(skb, i);
1163 }
1164 skb_shinfo(n)->nr_frags = i;
1165 }
1166
1167 if (skb_has_frag_list(skb)) {
1168 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1169 skb_clone_fraglist(n);
1170 }
1171
1172 copy_skb_header(n, skb);
1173out:
1174 return n;
1175}
1176EXPORT_SYMBOL(__pskb_copy_fclone);
1177
1178/**
1179 * pskb_expand_head - reallocate header of &sk_buff
1180 * @skb: buffer to reallocate
1181 * @nhead: room to add at head
1182 * @ntail: room to add at tail
1183 * @gfp_mask: allocation priority
1184 *
1185 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1186 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1187 * reference count of 1. Returns zero in the case of success or error,
1188 * if expansion failed. In the last case, &sk_buff is not changed.
1189 *
1190 * All the pointers pointing into skb header may change and must be
1191 * reloaded after call to this function.
1192 */
1193
1194int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1195 gfp_t gfp_mask)
1196{
1197 int i;
1198 u8 *data;
1199 int size = nhead + skb_end_offset(skb) + ntail;
1200 long off;
1201
1202 BUG_ON(nhead < 0);
1203
1204 if (skb_shared(skb))
1205 BUG();
1206
1207 size = SKB_DATA_ALIGN(size);
1208
1209 if (skb_pfmemalloc(skb))
1210 gfp_mask |= __GFP_MEMALLOC;
1211 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1212 gfp_mask, NUMA_NO_NODE, NULL);
1213 if (!data)
1214 goto nodata;
1215 size = SKB_WITH_OVERHEAD(ksize(data));
1216
1217 /* Copy only real data... and, alas, header. This should be
1218 * optimized for the cases when header is void.
1219 */
1220 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1221
1222 memcpy((struct skb_shared_info *)(data + size),
1223 skb_shinfo(skb),
1224 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1225
1226 /*
1227 * if shinfo is shared we must drop the old head gracefully, but if it
1228 * is not we can just drop the old head and let the existing refcount
1229 * be since all we did is relocate the values
1230 */
1231 if (skb_cloned(skb)) {
1232 /* copy this zero copy skb frags */
1233 if (skb_orphan_frags(skb, gfp_mask))
1234 goto nofrags;
1235 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1236 skb_frag_ref(skb, i);
1237
1238 if (skb_has_frag_list(skb))
1239 skb_clone_fraglist(skb);
1240
1241 skb_release_data(skb);
1242 } else {
1243 skb_free_head(skb);
1244 }
1245 off = (data + nhead) - skb->head;
1246
1247 skb->head = data;
1248 skb->head_frag = 0;
1249 skb->data += off;
1250#ifdef NET_SKBUFF_DATA_USES_OFFSET
1251 skb->end = size;
1252 off = nhead;
1253#else
1254 skb->end = skb->head + size;
1255#endif
1256 skb->tail += off;
1257 skb_headers_offset_update(skb, nhead);
1258 skb->cloned = 0;
1259 skb->hdr_len = 0;
1260 skb->nohdr = 0;
1261 atomic_set(&skb_shinfo(skb)->dataref, 1);
1262 return 0;
1263
1264nofrags:
1265 kfree(data);
1266nodata:
1267 return -ENOMEM;
1268}
1269EXPORT_SYMBOL(pskb_expand_head);
1270
1271/* Make private copy of skb with writable head and some headroom */
1272
1273struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1274{
1275 struct sk_buff *skb2;
1276 int delta = headroom - skb_headroom(skb);
1277
1278 if (delta <= 0)
1279 skb2 = pskb_copy(skb, GFP_ATOMIC);
1280 else {
1281 skb2 = skb_clone(skb, GFP_ATOMIC);
1282 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1283 GFP_ATOMIC)) {
1284 kfree_skb(skb2);
1285 skb2 = NULL;
1286 }
1287 }
1288 return skb2;
1289}
1290EXPORT_SYMBOL(skb_realloc_headroom);
1291
1292/**
1293 * skb_copy_expand - copy and expand sk_buff
1294 * @skb: buffer to copy
1295 * @newheadroom: new free bytes at head
1296 * @newtailroom: new free bytes at tail
1297 * @gfp_mask: allocation priority
1298 *
1299 * Make a copy of both an &sk_buff and its data and while doing so
1300 * allocate additional space.
1301 *
1302 * This is used when the caller wishes to modify the data and needs a
1303 * private copy of the data to alter as well as more space for new fields.
1304 * Returns %NULL on failure or the pointer to the buffer
1305 * on success. The returned buffer has a reference count of 1.
1306 *
1307 * You must pass %GFP_ATOMIC as the allocation priority if this function
1308 * is called from an interrupt.
1309 */
1310struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1311 int newheadroom, int newtailroom,
1312 gfp_t gfp_mask)
1313{
1314 /*
1315 * Allocate the copy buffer
1316 */
1317 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1318 gfp_mask, skb_alloc_rx_flag(skb),
1319 NUMA_NO_NODE);
1320 int oldheadroom = skb_headroom(skb);
1321 int head_copy_len, head_copy_off;
1322
1323 if (!n)
1324 return NULL;
1325
1326 skb_reserve(n, newheadroom);
1327
1328 /* Set the tail pointer and length */
1329 skb_put(n, skb->len);
1330
1331 head_copy_len = oldheadroom;
1332 head_copy_off = 0;
1333 if (newheadroom <= head_copy_len)
1334 head_copy_len = newheadroom;
1335 else
1336 head_copy_off = newheadroom - head_copy_len;
1337
1338 /* Copy the linear header and data. */
1339 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1340 skb->len + head_copy_len))
1341 BUG();
1342
1343 copy_skb_header(n, skb);
1344
1345 skb_headers_offset_update(n, newheadroom - oldheadroom);
1346
1347 return n;
1348}
1349EXPORT_SYMBOL(skb_copy_expand);
1350
1351/**
1352 * skb_pad - zero pad the tail of an skb
1353 * @skb: buffer to pad
1354 * @pad: space to pad
1355 *
1356 * Ensure that a buffer is followed by a padding area that is zero
1357 * filled. Used by network drivers which may DMA or transfer data
1358 * beyond the buffer end onto the wire.
1359 *
1360 * May return error in out of memory cases. The skb is freed on error.
1361 */
1362
1363int skb_pad(struct sk_buff *skb, int pad)
1364{
1365 int err;
1366 int ntail;
1367
1368 /* If the skbuff is non linear tailroom is always zero.. */
1369 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1370 memset(skb->data+skb->len, 0, pad);
1371 return 0;
1372 }
1373
1374 ntail = skb->data_len + pad - (skb->end - skb->tail);
1375 if (likely(skb_cloned(skb) || ntail > 0)) {
1376 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1377 if (unlikely(err))
1378 goto free_skb;
1379 }
1380
1381 /* FIXME: The use of this function with non-linear skb's really needs
1382 * to be audited.
1383 */
1384 err = skb_linearize(skb);
1385 if (unlikely(err))
1386 goto free_skb;
1387
1388 memset(skb->data + skb->len, 0, pad);
1389 return 0;
1390
1391free_skb:
1392 kfree_skb(skb);
1393 return err;
1394}
1395EXPORT_SYMBOL(skb_pad);
1396
1397/**
1398 * pskb_put - add data to the tail of a potentially fragmented buffer
1399 * @skb: start of the buffer to use
1400 * @tail: tail fragment of the buffer to use
1401 * @len: amount of data to add
1402 *
1403 * This function extends the used data area of the potentially
1404 * fragmented buffer. @tail must be the last fragment of @skb -- or
1405 * @skb itself. If this would exceed the total buffer size the kernel
1406 * will panic. A pointer to the first byte of the extra data is
1407 * returned.
1408 */
1409
1410unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1411{
1412 if (tail != skb) {
1413 skb->data_len += len;
1414 skb->len += len;
1415 }
1416 return skb_put(tail, len);
1417}
1418EXPORT_SYMBOL_GPL(pskb_put);
1419
1420/**
1421 * skb_put - add data to a buffer
1422 * @skb: buffer to use
1423 * @len: amount of data to add
1424 *
1425 * This function extends the used data area of the buffer. If this would
1426 * exceed the total buffer size the kernel will panic. A pointer to the
1427 * first byte of the extra data is returned.
1428 */
1429unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1430{
1431 unsigned char *tmp = skb_tail_pointer(skb);
1432 SKB_LINEAR_ASSERT(skb);
1433 skb->tail += len;
1434 skb->len += len;
1435 if (unlikely(skb->tail > skb->end))
1436 skb_over_panic(skb, len, __builtin_return_address(0));
1437 return tmp;
1438}
1439EXPORT_SYMBOL(skb_put);
1440
1441/**
1442 * skb_push - add data to the start of a buffer
1443 * @skb: buffer to use
1444 * @len: amount of data to add
1445 *
1446 * This function extends the used data area of the buffer at the buffer
1447 * start. If this would exceed the total buffer headroom the kernel will
1448 * panic. A pointer to the first byte of the extra data is returned.
1449 */
1450unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1451{
1452 skb->data -= len;
1453 skb->len += len;
1454 if (unlikely(skb->data<skb->head))
1455 skb_under_panic(skb, len, __builtin_return_address(0));
1456 return skb->data;
1457}
1458EXPORT_SYMBOL(skb_push);
1459
1460/**
1461 * skb_pull - remove data from the start of a buffer
1462 * @skb: buffer to use
1463 * @len: amount of data to remove
1464 *
1465 * This function removes data from the start of a buffer, returning
1466 * the memory to the headroom. A pointer to the next data in the buffer
1467 * is returned. Once the data has been pulled future pushes will overwrite
1468 * the old data.
1469 */
1470unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1471{
1472 return skb_pull_inline(skb, len);
1473}
1474EXPORT_SYMBOL(skb_pull);
1475
1476/**
1477 * skb_trim - remove end from a buffer
1478 * @skb: buffer to alter
1479 * @len: new length
1480 *
1481 * Cut the length of a buffer down by removing data from the tail. If
1482 * the buffer is already under the length specified it is not modified.
1483 * The skb must be linear.
1484 */
1485void skb_trim(struct sk_buff *skb, unsigned int len)
1486{
1487 if (skb->len > len)
1488 __skb_trim(skb, len);
1489}
1490EXPORT_SYMBOL(skb_trim);
1491
1492/* Trims skb to length len. It can change skb pointers.
1493 */
1494
1495int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1496{
1497 struct sk_buff **fragp;
1498 struct sk_buff *frag;
1499 int offset = skb_headlen(skb);
1500 int nfrags = skb_shinfo(skb)->nr_frags;
1501 int i;
1502 int err;
1503
1504 if (skb_cloned(skb) &&
1505 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1506 return err;
1507
1508 i = 0;
1509 if (offset >= len)
1510 goto drop_pages;
1511
1512 for (; i < nfrags; i++) {
1513 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1514
1515 if (end < len) {
1516 offset = end;
1517 continue;
1518 }
1519
1520 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1521
1522drop_pages:
1523 skb_shinfo(skb)->nr_frags = i;
1524
1525 for (; i < nfrags; i++)
1526 skb_frag_unref(skb, i);
1527
1528 if (skb_has_frag_list(skb))
1529 skb_drop_fraglist(skb);
1530 goto done;
1531 }
1532
1533 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1534 fragp = &frag->next) {
1535 int end = offset + frag->len;
1536
1537 if (skb_shared(frag)) {
1538 struct sk_buff *nfrag;
1539
1540 nfrag = skb_clone(frag, GFP_ATOMIC);
1541 if (unlikely(!nfrag))
1542 return -ENOMEM;
1543
1544 nfrag->next = frag->next;
1545 consume_skb(frag);
1546 frag = nfrag;
1547 *fragp = frag;
1548 }
1549
1550 if (end < len) {
1551 offset = end;
1552 continue;
1553 }
1554
1555 if (end > len &&
1556 unlikely((err = pskb_trim(frag, len - offset))))
1557 return err;
1558
1559 if (frag->next)
1560 skb_drop_list(&frag->next);
1561 break;
1562 }
1563
1564done:
1565 if (len > skb_headlen(skb)) {
1566 skb->data_len -= skb->len - len;
1567 skb->len = len;
1568 } else {
1569 skb->len = len;
1570 skb->data_len = 0;
1571 skb_set_tail_pointer(skb, len);
1572 }
1573
1574 return 0;
1575}
1576EXPORT_SYMBOL(___pskb_trim);
1577
1578/**
1579 * __pskb_pull_tail - advance tail of skb header
1580 * @skb: buffer to reallocate
1581 * @delta: number of bytes to advance tail
1582 *
1583 * The function makes a sense only on a fragmented &sk_buff,
1584 * it expands header moving its tail forward and copying necessary
1585 * data from fragmented part.
1586 *
1587 * &sk_buff MUST have reference count of 1.
1588 *
1589 * Returns %NULL (and &sk_buff does not change) if pull failed
1590 * or value of new tail of skb in the case of success.
1591 *
1592 * All the pointers pointing into skb header may change and must be
1593 * reloaded after call to this function.
1594 */
1595
1596/* Moves tail of skb head forward, copying data from fragmented part,
1597 * when it is necessary.
1598 * 1. It may fail due to malloc failure.
1599 * 2. It may change skb pointers.
1600 *
1601 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1602 */
1603unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1604{
1605 /* If skb has not enough free space at tail, get new one
1606 * plus 128 bytes for future expansions. If we have enough
1607 * room at tail, reallocate without expansion only if skb is cloned.
1608 */
1609 int i, k, eat = (skb->tail + delta) - skb->end;
1610
1611 if (eat > 0 || skb_cloned(skb)) {
1612 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1613 GFP_ATOMIC))
1614 return NULL;
1615 }
1616
1617 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1618 BUG();
1619
1620 /* Optimization: no fragments, no reasons to preestimate
1621 * size of pulled pages. Superb.
1622 */
1623 if (!skb_has_frag_list(skb))
1624 goto pull_pages;
1625
1626 /* Estimate size of pulled pages. */
1627 eat = delta;
1628 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1629 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1630
1631 if (size >= eat)
1632 goto pull_pages;
1633 eat -= size;
1634 }
1635
1636 /* If we need update frag list, we are in troubles.
1637 * Certainly, it possible to add an offset to skb data,
1638 * but taking into account that pulling is expected to
1639 * be very rare operation, it is worth to fight against
1640 * further bloating skb head and crucify ourselves here instead.
1641 * Pure masohism, indeed. 8)8)
1642 */
1643 if (eat) {
1644 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1645 struct sk_buff *clone = NULL;
1646 struct sk_buff *insp = NULL;
1647
1648 do {
1649 BUG_ON(!list);
1650
1651 if (list->len <= eat) {
1652 /* Eaten as whole. */
1653 eat -= list->len;
1654 list = list->next;
1655 insp = list;
1656 } else {
1657 /* Eaten partially. */
1658
1659 if (skb_shared(list)) {
1660 /* Sucks! We need to fork list. :-( */
1661 clone = skb_clone(list, GFP_ATOMIC);
1662 if (!clone)
1663 return NULL;
1664 insp = list->next;
1665 list = clone;
1666 } else {
1667 /* This may be pulled without
1668 * problems. */
1669 insp = list;
1670 }
1671 if (!pskb_pull(list, eat)) {
1672 kfree_skb(clone);
1673 return NULL;
1674 }
1675 break;
1676 }
1677 } while (eat);
1678
1679 /* Free pulled out fragments. */
1680 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1681 skb_shinfo(skb)->frag_list = list->next;
1682 kfree_skb(list);
1683 }
1684 /* And insert new clone at head. */
1685 if (clone) {
1686 clone->next = list;
1687 skb_shinfo(skb)->frag_list = clone;
1688 }
1689 }
1690 /* Success! Now we may commit changes to skb data. */
1691
1692pull_pages:
1693 eat = delta;
1694 k = 0;
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1696 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1697
1698 if (size <= eat) {
1699 skb_frag_unref(skb, i);
1700 eat -= size;
1701 } else {
1702 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1703 if (eat) {
1704 skb_shinfo(skb)->frags[k].page_offset += eat;
1705 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1706 eat = 0;
1707 }
1708 k++;
1709 }
1710 }
1711 skb_shinfo(skb)->nr_frags = k;
1712
1713 skb->tail += delta;
1714 skb->data_len -= delta;
1715
1716 return skb_tail_pointer(skb);
1717}
1718EXPORT_SYMBOL(__pskb_pull_tail);
1719
1720/**
1721 * skb_copy_bits - copy bits from skb to kernel buffer
1722 * @skb: source skb
1723 * @offset: offset in source
1724 * @to: destination buffer
1725 * @len: number of bytes to copy
1726 *
1727 * Copy the specified number of bytes from the source skb to the
1728 * destination buffer.
1729 *
1730 * CAUTION ! :
1731 * If its prototype is ever changed,
1732 * check arch/{*}/net/{*}.S files,
1733 * since it is called from BPF assembly code.
1734 */
1735int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1736{
1737 int start = skb_headlen(skb);
1738 struct sk_buff *frag_iter;
1739 int i, copy;
1740
1741 if (offset > (int)skb->len - len)
1742 goto fault;
1743
1744 /* Copy header. */
1745 if ((copy = start - offset) > 0) {
1746 if (copy > len)
1747 copy = len;
1748 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1749 if ((len -= copy) == 0)
1750 return 0;
1751 offset += copy;
1752 to += copy;
1753 }
1754
1755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1756 int end;
1757 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1758
1759 WARN_ON(start > offset + len);
1760
1761 end = start + skb_frag_size(f);
1762 if ((copy = end - offset) > 0) {
1763 u8 *vaddr;
1764
1765 if (copy > len)
1766 copy = len;
1767
1768 vaddr = kmap_atomic(skb_frag_page(f));
1769 memcpy(to,
1770 vaddr + f->page_offset + offset - start,
1771 copy);
1772 kunmap_atomic(vaddr);
1773
1774 if ((len -= copy) == 0)
1775 return 0;
1776 offset += copy;
1777 to += copy;
1778 }
1779 start = end;
1780 }
1781
1782 skb_walk_frags(skb, frag_iter) {
1783 int end;
1784
1785 WARN_ON(start > offset + len);
1786
1787 end = start + frag_iter->len;
1788 if ((copy = end - offset) > 0) {
1789 if (copy > len)
1790 copy = len;
1791 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1792 goto fault;
1793 if ((len -= copy) == 0)
1794 return 0;
1795 offset += copy;
1796 to += copy;
1797 }
1798 start = end;
1799 }
1800
1801 if (!len)
1802 return 0;
1803
1804fault:
1805 return -EFAULT;
1806}
1807EXPORT_SYMBOL(skb_copy_bits);
1808
1809/*
1810 * Callback from splice_to_pipe(), if we need to release some pages
1811 * at the end of the spd in case we error'ed out in filling the pipe.
1812 */
1813static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1814{
1815 put_page(spd->pages[i]);
1816}
1817
1818static struct page *linear_to_page(struct page *page, unsigned int *len,
1819 unsigned int *offset,
1820 struct sock *sk)
1821{
1822 struct page_frag *pfrag = sk_page_frag(sk);
1823
1824 if (!sk_page_frag_refill(sk, pfrag))
1825 return NULL;
1826
1827 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1828
1829 memcpy(page_address(pfrag->page) + pfrag->offset,
1830 page_address(page) + *offset, *len);
1831 *offset = pfrag->offset;
1832 pfrag->offset += *len;
1833
1834 return pfrag->page;
1835}
1836
1837static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1838 struct page *page,
1839 unsigned int offset)
1840{
1841 return spd->nr_pages &&
1842 spd->pages[spd->nr_pages - 1] == page &&
1843 (spd->partial[spd->nr_pages - 1].offset +
1844 spd->partial[spd->nr_pages - 1].len == offset);
1845}
1846
1847/*
1848 * Fill page/offset/length into spd, if it can hold more pages.
1849 */
1850static bool spd_fill_page(struct splice_pipe_desc *spd,
1851 struct pipe_inode_info *pipe, struct page *page,
1852 unsigned int *len, unsigned int offset,
1853 bool linear,
1854 struct sock *sk)
1855{
1856 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1857 return true;
1858
1859 if (linear) {
1860 page = linear_to_page(page, len, &offset, sk);
1861 if (!page)
1862 return true;
1863 }
1864 if (spd_can_coalesce(spd, page, offset)) {
1865 spd->partial[spd->nr_pages - 1].len += *len;
1866 return false;
1867 }
1868 get_page(page);
1869 spd->pages[spd->nr_pages] = page;
1870 spd->partial[spd->nr_pages].len = *len;
1871 spd->partial[spd->nr_pages].offset = offset;
1872 spd->nr_pages++;
1873
1874 return false;
1875}
1876
1877static bool __splice_segment(struct page *page, unsigned int poff,
1878 unsigned int plen, unsigned int *off,
1879 unsigned int *len,
1880 struct splice_pipe_desc *spd, bool linear,
1881 struct sock *sk,
1882 struct pipe_inode_info *pipe)
1883{
1884 if (!*len)
1885 return true;
1886
1887 /* skip this segment if already processed */
1888 if (*off >= plen) {
1889 *off -= plen;
1890 return false;
1891 }
1892
1893 /* ignore any bits we already processed */
1894 poff += *off;
1895 plen -= *off;
1896 *off = 0;
1897
1898 do {
1899 unsigned int flen = min(*len, plen);
1900
1901 if (spd_fill_page(spd, pipe, page, &flen, poff,
1902 linear, sk))
1903 return true;
1904 poff += flen;
1905 plen -= flen;
1906 *len -= flen;
1907 } while (*len && plen);
1908
1909 return false;
1910}
1911
1912/*
1913 * Map linear and fragment data from the skb to spd. It reports true if the
1914 * pipe is full or if we already spliced the requested length.
1915 */
1916static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1917 unsigned int *offset, unsigned int *len,
1918 struct splice_pipe_desc *spd, struct sock *sk)
1919{
1920 int seg;
1921 struct sk_buff *iter;
1922
1923 /* map the linear part :
1924 * If skb->head_frag is set, this 'linear' part is backed by a
1925 * fragment, and if the head is not shared with any clones then
1926 * we can avoid a copy since we own the head portion of this page.
1927 */
1928 if (__splice_segment(virt_to_page(skb->data),
1929 (unsigned long) skb->data & (PAGE_SIZE - 1),
1930 skb_headlen(skb),
1931 offset, len, spd,
1932 skb_head_is_locked(skb),
1933 sk, pipe))
1934 return true;
1935
1936 /*
1937 * then map the fragments
1938 */
1939 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1940 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1941
1942 if (__splice_segment(skb_frag_page(f),
1943 f->page_offset, skb_frag_size(f),
1944 offset, len, spd, false, sk, pipe))
1945 return true;
1946 }
1947
1948 skb_walk_frags(skb, iter) {
1949 if (*offset >= iter->len) {
1950 *offset -= iter->len;
1951 continue;
1952 }
1953 /* __skb_splice_bits() only fails if the output has no room
1954 * left, so no point in going over the frag_list for the error
1955 * case.
1956 */
1957 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1958 return true;
1959 }
1960
1961 return false;
1962}
1963
1964ssize_t skb_socket_splice(struct sock *sk,
1965 struct pipe_inode_info *pipe,
1966 struct splice_pipe_desc *spd)
1967{
1968 int ret;
1969
1970 /* Drop the socket lock, otherwise we have reverse
1971 * locking dependencies between sk_lock and i_mutex
1972 * here as compared to sendfile(). We enter here
1973 * with the socket lock held, and splice_to_pipe() will
1974 * grab the pipe inode lock. For sendfile() emulation,
1975 * we call into ->sendpage() with the i_mutex lock held
1976 * and networking will grab the socket lock.
1977 */
1978 release_sock(sk);
1979 ret = splice_to_pipe(pipe, spd);
1980 lock_sock(sk);
1981
1982 return ret;
1983}
1984
1985/*
1986 * Map data from the skb to a pipe. Should handle both the linear part,
1987 * the fragments, and the frag list.
1988 */
1989int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1990 struct pipe_inode_info *pipe, unsigned int tlen,
1991 unsigned int flags,
1992 ssize_t (*splice_cb)(struct sock *,
1993 struct pipe_inode_info *,
1994 struct splice_pipe_desc *))
1995{
1996 struct partial_page partial[MAX_SKB_FRAGS];
1997 struct page *pages[MAX_SKB_FRAGS];
1998 struct splice_pipe_desc spd = {
1999 .pages = pages,
2000 .partial = partial,
2001 .nr_pages_max = MAX_SKB_FRAGS,
2002 .flags = flags,
2003 .ops = &nosteal_pipe_buf_ops,
2004 .spd_release = sock_spd_release,
2005 };
2006 int ret = 0;
2007
2008 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2009
2010 if (spd.nr_pages)
2011 ret = splice_cb(sk, pipe, &spd);
2012
2013 return ret;
2014}
2015EXPORT_SYMBOL_GPL(skb_splice_bits);
2016
2017/**
2018 * skb_store_bits - store bits from kernel buffer to skb
2019 * @skb: destination buffer
2020 * @offset: offset in destination
2021 * @from: source buffer
2022 * @len: number of bytes to copy
2023 *
2024 * Copy the specified number of bytes from the source buffer to the
2025 * destination skb. This function handles all the messy bits of
2026 * traversing fragment lists and such.
2027 */
2028
2029int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2030{
2031 int start = skb_headlen(skb);
2032 struct sk_buff *frag_iter;
2033 int i, copy;
2034
2035 if (offset > (int)skb->len - len)
2036 goto fault;
2037
2038 if ((copy = start - offset) > 0) {
2039 if (copy > len)
2040 copy = len;
2041 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2042 if ((len -= copy) == 0)
2043 return 0;
2044 offset += copy;
2045 from += copy;
2046 }
2047
2048 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2049 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2050 int end;
2051
2052 WARN_ON(start > offset + len);
2053
2054 end = start + skb_frag_size(frag);
2055 if ((copy = end - offset) > 0) {
2056 u8 *vaddr;
2057
2058 if (copy > len)
2059 copy = len;
2060
2061 vaddr = kmap_atomic(skb_frag_page(frag));
2062 memcpy(vaddr + frag->page_offset + offset - start,
2063 from, copy);
2064 kunmap_atomic(vaddr);
2065
2066 if ((len -= copy) == 0)
2067 return 0;
2068 offset += copy;
2069 from += copy;
2070 }
2071 start = end;
2072 }
2073
2074 skb_walk_frags(skb, frag_iter) {
2075 int end;
2076
2077 WARN_ON(start > offset + len);
2078
2079 end = start + frag_iter->len;
2080 if ((copy = end - offset) > 0) {
2081 if (copy > len)
2082 copy = len;
2083 if (skb_store_bits(frag_iter, offset - start,
2084 from, copy))
2085 goto fault;
2086 if ((len -= copy) == 0)
2087 return 0;
2088 offset += copy;
2089 from += copy;
2090 }
2091 start = end;
2092 }
2093 if (!len)
2094 return 0;
2095
2096fault:
2097 return -EFAULT;
2098}
2099EXPORT_SYMBOL(skb_store_bits);
2100
2101/* Checksum skb data. */
2102__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2103 __wsum csum, const struct skb_checksum_ops *ops)
2104{
2105 int start = skb_headlen(skb);
2106 int i, copy = start - offset;
2107 struct sk_buff *frag_iter;
2108 int pos = 0;
2109
2110 /* Checksum header. */
2111 if (copy > 0) {
2112 if (copy > len)
2113 copy = len;
2114 csum = ops->update(skb->data + offset, copy, csum);
2115 if ((len -= copy) == 0)
2116 return csum;
2117 offset += copy;
2118 pos = copy;
2119 }
2120
2121 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2122 int end;
2123 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2124
2125 WARN_ON(start > offset + len);
2126
2127 end = start + skb_frag_size(frag);
2128 if ((copy = end - offset) > 0) {
2129 __wsum csum2;
2130 u8 *vaddr;
2131
2132 if (copy > len)
2133 copy = len;
2134 vaddr = kmap_atomic(skb_frag_page(frag));
2135 csum2 = ops->update(vaddr + frag->page_offset +
2136 offset - start, copy, 0);
2137 kunmap_atomic(vaddr);
2138 csum = ops->combine(csum, csum2, pos, copy);
2139 if (!(len -= copy))
2140 return csum;
2141 offset += copy;
2142 pos += copy;
2143 }
2144 start = end;
2145 }
2146
2147 skb_walk_frags(skb, frag_iter) {
2148 int end;
2149
2150 WARN_ON(start > offset + len);
2151
2152 end = start + frag_iter->len;
2153 if ((copy = end - offset) > 0) {
2154 __wsum csum2;
2155 if (copy > len)
2156 copy = len;
2157 csum2 = __skb_checksum(frag_iter, offset - start,
2158 copy, 0, ops);
2159 csum = ops->combine(csum, csum2, pos, copy);
2160 if ((len -= copy) == 0)
2161 return csum;
2162 offset += copy;
2163 pos += copy;
2164 }
2165 start = end;
2166 }
2167 BUG_ON(len);
2168
2169 return csum;
2170}
2171EXPORT_SYMBOL(__skb_checksum);
2172
2173__wsum skb_checksum(const struct sk_buff *skb, int offset,
2174 int len, __wsum csum)
2175{
2176 const struct skb_checksum_ops ops = {
2177 .update = csum_partial_ext,
2178 .combine = csum_block_add_ext,
2179 };
2180
2181 return __skb_checksum(skb, offset, len, csum, &ops);
2182}
2183EXPORT_SYMBOL(skb_checksum);
2184
2185/* Both of above in one bottle. */
2186
2187__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2188 u8 *to, int len, __wsum csum)
2189{
2190 int start = skb_headlen(skb);
2191 int i, copy = start - offset;
2192 struct sk_buff *frag_iter;
2193 int pos = 0;
2194
2195 /* Copy header. */
2196 if (copy > 0) {
2197 if (copy > len)
2198 copy = len;
2199 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2200 copy, csum);
2201 if ((len -= copy) == 0)
2202 return csum;
2203 offset += copy;
2204 to += copy;
2205 pos = copy;
2206 }
2207
2208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209 int end;
2210
2211 WARN_ON(start > offset + len);
2212
2213 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2214 if ((copy = end - offset) > 0) {
2215 __wsum csum2;
2216 u8 *vaddr;
2217 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2218
2219 if (copy > len)
2220 copy = len;
2221 vaddr = kmap_atomic(skb_frag_page(frag));
2222 csum2 = csum_partial_copy_nocheck(vaddr +
2223 frag->page_offset +
2224 offset - start, to,
2225 copy, 0);
2226 kunmap_atomic(vaddr);
2227 csum = csum_block_add(csum, csum2, pos);
2228 if (!(len -= copy))
2229 return csum;
2230 offset += copy;
2231 to += copy;
2232 pos += copy;
2233 }
2234 start = end;
2235 }
2236
2237 skb_walk_frags(skb, frag_iter) {
2238 __wsum csum2;
2239 int end;
2240
2241 WARN_ON(start > offset + len);
2242
2243 end = start + frag_iter->len;
2244 if ((copy = end - offset) > 0) {
2245 if (copy > len)
2246 copy = len;
2247 csum2 = skb_copy_and_csum_bits(frag_iter,
2248 offset - start,
2249 to, copy, 0);
2250 csum = csum_block_add(csum, csum2, pos);
2251 if ((len -= copy) == 0)
2252 return csum;
2253 offset += copy;
2254 to += copy;
2255 pos += copy;
2256 }
2257 start = end;
2258 }
2259 BUG_ON(len);
2260 return csum;
2261}
2262EXPORT_SYMBOL(skb_copy_and_csum_bits);
2263
2264 /**
2265 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2266 * @from: source buffer
2267 *
2268 * Calculates the amount of linear headroom needed in the 'to' skb passed
2269 * into skb_zerocopy().
2270 */
2271unsigned int
2272skb_zerocopy_headlen(const struct sk_buff *from)
2273{
2274 unsigned int hlen = 0;
2275
2276 if (!from->head_frag ||
2277 skb_headlen(from) < L1_CACHE_BYTES ||
2278 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2279 hlen = skb_headlen(from);
2280
2281 if (skb_has_frag_list(from))
2282 hlen = from->len;
2283
2284 return hlen;
2285}
2286EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2287
2288/**
2289 * skb_zerocopy - Zero copy skb to skb
2290 * @to: destination buffer
2291 * @from: source buffer
2292 * @len: number of bytes to copy from source buffer
2293 * @hlen: size of linear headroom in destination buffer
2294 *
2295 * Copies up to `len` bytes from `from` to `to` by creating references
2296 * to the frags in the source buffer.
2297 *
2298 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2299 * headroom in the `to` buffer.
2300 *
2301 * Return value:
2302 * 0: everything is OK
2303 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2304 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2305 */
2306int
2307skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2308{
2309 int i, j = 0;
2310 int plen = 0; /* length of skb->head fragment */
2311 int ret;
2312 struct page *page;
2313 unsigned int offset;
2314
2315 BUG_ON(!from->head_frag && !hlen);
2316
2317 /* dont bother with small payloads */
2318 if (len <= skb_tailroom(to))
2319 return skb_copy_bits(from, 0, skb_put(to, len), len);
2320
2321 if (hlen) {
2322 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2323 if (unlikely(ret))
2324 return ret;
2325 len -= hlen;
2326 } else {
2327 plen = min_t(int, skb_headlen(from), len);
2328 if (plen) {
2329 page = virt_to_head_page(from->head);
2330 offset = from->data - (unsigned char *)page_address(page);
2331 __skb_fill_page_desc(to, 0, page, offset, plen);
2332 get_page(page);
2333 j = 1;
2334 len -= plen;
2335 }
2336 }
2337
2338 to->truesize += len + plen;
2339 to->len += len + plen;
2340 to->data_len += len + plen;
2341
2342 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2343 skb_tx_error(from);
2344 return -ENOMEM;
2345 }
2346
2347 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2348 if (!len)
2349 break;
2350 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2351 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2352 len -= skb_shinfo(to)->frags[j].size;
2353 skb_frag_ref(to, j);
2354 j++;
2355 }
2356 skb_shinfo(to)->nr_frags = j;
2357
2358 return 0;
2359}
2360EXPORT_SYMBOL_GPL(skb_zerocopy);
2361
2362void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2363{
2364 __wsum csum;
2365 long csstart;
2366
2367 if (skb->ip_summed == CHECKSUM_PARTIAL)
2368 csstart = skb_checksum_start_offset(skb);
2369 else
2370 csstart = skb_headlen(skb);
2371
2372 BUG_ON(csstart > skb_headlen(skb));
2373
2374 skb_copy_from_linear_data(skb, to, csstart);
2375
2376 csum = 0;
2377 if (csstart != skb->len)
2378 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2379 skb->len - csstart, 0);
2380
2381 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2382 long csstuff = csstart + skb->csum_offset;
2383
2384 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2385 }
2386}
2387EXPORT_SYMBOL(skb_copy_and_csum_dev);
2388
2389/**
2390 * skb_dequeue - remove from the head of the queue
2391 * @list: list to dequeue from
2392 *
2393 * Remove the head of the list. The list lock is taken so the function
2394 * may be used safely with other locking list functions. The head item is
2395 * returned or %NULL if the list is empty.
2396 */
2397
2398struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2399{
2400 unsigned long flags;
2401 struct sk_buff *result;
2402
2403 spin_lock_irqsave(&list->lock, flags);
2404 result = __skb_dequeue(list);
2405 spin_unlock_irqrestore(&list->lock, flags);
2406 return result;
2407}
2408EXPORT_SYMBOL(skb_dequeue);
2409
2410/**
2411 * skb_dequeue_tail - remove from the tail of the queue
2412 * @list: list to dequeue from
2413 *
2414 * Remove the tail of the list. The list lock is taken so the function
2415 * may be used safely with other locking list functions. The tail item is
2416 * returned or %NULL if the list is empty.
2417 */
2418struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2419{
2420 unsigned long flags;
2421 struct sk_buff *result;
2422
2423 spin_lock_irqsave(&list->lock, flags);
2424 result = __skb_dequeue_tail(list);
2425 spin_unlock_irqrestore(&list->lock, flags);
2426 return result;
2427}
2428EXPORT_SYMBOL(skb_dequeue_tail);
2429
2430/**
2431 * skb_queue_purge - empty a list
2432 * @list: list to empty
2433 *
2434 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2435 * the list and one reference dropped. This function takes the list
2436 * lock and is atomic with respect to other list locking functions.
2437 */
2438void skb_queue_purge(struct sk_buff_head *list)
2439{
2440 struct sk_buff *skb;
2441 while ((skb = skb_dequeue(list)) != NULL)
2442 kfree_skb(skb);
2443}
2444EXPORT_SYMBOL(skb_queue_purge);
2445
2446/**
2447 * skb_queue_head - queue a buffer at the list head
2448 * @list: list to use
2449 * @newsk: buffer to queue
2450 *
2451 * Queue a buffer at the start of the list. This function takes the
2452 * list lock and can be used safely with other locking &sk_buff functions
2453 * safely.
2454 *
2455 * A buffer cannot be placed on two lists at the same time.
2456 */
2457void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2458{
2459 unsigned long flags;
2460
2461 spin_lock_irqsave(&list->lock, flags);
2462 __skb_queue_head(list, newsk);
2463 spin_unlock_irqrestore(&list->lock, flags);
2464}
2465EXPORT_SYMBOL(skb_queue_head);
2466
2467/**
2468 * skb_queue_tail - queue a buffer at the list tail
2469 * @list: list to use
2470 * @newsk: buffer to queue
2471 *
2472 * Queue a buffer at the tail of the list. This function takes the
2473 * list lock and can be used safely with other locking &sk_buff functions
2474 * safely.
2475 *
2476 * A buffer cannot be placed on two lists at the same time.
2477 */
2478void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2479{
2480 unsigned long flags;
2481
2482 spin_lock_irqsave(&list->lock, flags);
2483 __skb_queue_tail(list, newsk);
2484 spin_unlock_irqrestore(&list->lock, flags);
2485}
2486EXPORT_SYMBOL(skb_queue_tail);
2487
2488/**
2489 * skb_unlink - remove a buffer from a list
2490 * @skb: buffer to remove
2491 * @list: list to use
2492 *
2493 * Remove a packet from a list. The list locks are taken and this
2494 * function is atomic with respect to other list locked calls
2495 *
2496 * You must know what list the SKB is on.
2497 */
2498void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2499{
2500 unsigned long flags;
2501
2502 spin_lock_irqsave(&list->lock, flags);
2503 __skb_unlink(skb, list);
2504 spin_unlock_irqrestore(&list->lock, flags);
2505}
2506EXPORT_SYMBOL(skb_unlink);
2507
2508/**
2509 * skb_append - append a buffer
2510 * @old: buffer to insert after
2511 * @newsk: buffer to insert
2512 * @list: list to use
2513 *
2514 * Place a packet after a given packet in a list. The list locks are taken
2515 * and this function is atomic with respect to other list locked calls.
2516 * A buffer cannot be placed on two lists at the same time.
2517 */
2518void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2519{
2520 unsigned long flags;
2521
2522 spin_lock_irqsave(&list->lock, flags);
2523 __skb_queue_after(list, old, newsk);
2524 spin_unlock_irqrestore(&list->lock, flags);
2525}
2526EXPORT_SYMBOL(skb_append);
2527
2528/**
2529 * skb_insert - insert a buffer
2530 * @old: buffer to insert before
2531 * @newsk: buffer to insert
2532 * @list: list to use
2533 *
2534 * Place a packet before a given packet in a list. The list locks are
2535 * taken and this function is atomic with respect to other list locked
2536 * calls.
2537 *
2538 * A buffer cannot be placed on two lists at the same time.
2539 */
2540void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2541{
2542 unsigned long flags;
2543
2544 spin_lock_irqsave(&list->lock, flags);
2545 __skb_insert(newsk, old->prev, old, list);
2546 spin_unlock_irqrestore(&list->lock, flags);
2547}
2548EXPORT_SYMBOL(skb_insert);
2549
2550static inline void skb_split_inside_header(struct sk_buff *skb,
2551 struct sk_buff* skb1,
2552 const u32 len, const int pos)
2553{
2554 int i;
2555
2556 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2557 pos - len);
2558 /* And move data appendix as is. */
2559 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2560 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2561
2562 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2563 skb_shinfo(skb)->nr_frags = 0;
2564 skb1->data_len = skb->data_len;
2565 skb1->len += skb1->data_len;
2566 skb->data_len = 0;
2567 skb->len = len;
2568 skb_set_tail_pointer(skb, len);
2569}
2570
2571static inline void skb_split_no_header(struct sk_buff *skb,
2572 struct sk_buff* skb1,
2573 const u32 len, int pos)
2574{
2575 int i, k = 0;
2576 const int nfrags = skb_shinfo(skb)->nr_frags;
2577
2578 skb_shinfo(skb)->nr_frags = 0;
2579 skb1->len = skb1->data_len = skb->len - len;
2580 skb->len = len;
2581 skb->data_len = len - pos;
2582
2583 for (i = 0; i < nfrags; i++) {
2584 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2585
2586 if (pos + size > len) {
2587 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2588
2589 if (pos < len) {
2590 /* Split frag.
2591 * We have two variants in this case:
2592 * 1. Move all the frag to the second
2593 * part, if it is possible. F.e.
2594 * this approach is mandatory for TUX,
2595 * where splitting is expensive.
2596 * 2. Split is accurately. We make this.
2597 */
2598 skb_frag_ref(skb, i);
2599 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2600 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2601 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2602 skb_shinfo(skb)->nr_frags++;
2603 }
2604 k++;
2605 } else
2606 skb_shinfo(skb)->nr_frags++;
2607 pos += size;
2608 }
2609 skb_shinfo(skb1)->nr_frags = k;
2610}
2611
2612/**
2613 * skb_split - Split fragmented skb to two parts at length len.
2614 * @skb: the buffer to split
2615 * @skb1: the buffer to receive the second part
2616 * @len: new length for skb
2617 */
2618void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2619{
2620 int pos = skb_headlen(skb);
2621
2622 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2623 if (len < pos) /* Split line is inside header. */
2624 skb_split_inside_header(skb, skb1, len, pos);
2625 else /* Second chunk has no header, nothing to copy. */
2626 skb_split_no_header(skb, skb1, len, pos);
2627}
2628EXPORT_SYMBOL(skb_split);
2629
2630/* Shifting from/to a cloned skb is a no-go.
2631 *
2632 * Caller cannot keep skb_shinfo related pointers past calling here!
2633 */
2634static int skb_prepare_for_shift(struct sk_buff *skb)
2635{
2636 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2637}
2638
2639/**
2640 * skb_shift - Shifts paged data partially from skb to another
2641 * @tgt: buffer into which tail data gets added
2642 * @skb: buffer from which the paged data comes from
2643 * @shiftlen: shift up to this many bytes
2644 *
2645 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2646 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2647 * It's up to caller to free skb if everything was shifted.
2648 *
2649 * If @tgt runs out of frags, the whole operation is aborted.
2650 *
2651 * Skb cannot include anything else but paged data while tgt is allowed
2652 * to have non-paged data as well.
2653 *
2654 * TODO: full sized shift could be optimized but that would need
2655 * specialized skb free'er to handle frags without up-to-date nr_frags.
2656 */
2657int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2658{
2659 int from, to, merge, todo;
2660 struct skb_frag_struct *fragfrom, *fragto;
2661
2662 BUG_ON(shiftlen > skb->len);
2663 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2664
2665 todo = shiftlen;
2666 from = 0;
2667 to = skb_shinfo(tgt)->nr_frags;
2668 fragfrom = &skb_shinfo(skb)->frags[from];
2669
2670 /* Actual merge is delayed until the point when we know we can
2671 * commit all, so that we don't have to undo partial changes
2672 */
2673 if (!to ||
2674 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675 fragfrom->page_offset)) {
2676 merge = -1;
2677 } else {
2678 merge = to - 1;
2679
2680 todo -= skb_frag_size(fragfrom);
2681 if (todo < 0) {
2682 if (skb_prepare_for_shift(skb) ||
2683 skb_prepare_for_shift(tgt))
2684 return 0;
2685
2686 /* All previous frag pointers might be stale! */
2687 fragfrom = &skb_shinfo(skb)->frags[from];
2688 fragto = &skb_shinfo(tgt)->frags[merge];
2689
2690 skb_frag_size_add(fragto, shiftlen);
2691 skb_frag_size_sub(fragfrom, shiftlen);
2692 fragfrom->page_offset += shiftlen;
2693
2694 goto onlymerged;
2695 }
2696
2697 from++;
2698 }
2699
2700 /* Skip full, not-fitting skb to avoid expensive operations */
2701 if ((shiftlen == skb->len) &&
2702 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2703 return 0;
2704
2705 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2706 return 0;
2707
2708 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709 if (to == MAX_SKB_FRAGS)
2710 return 0;
2711
2712 fragfrom = &skb_shinfo(skb)->frags[from];
2713 fragto = &skb_shinfo(tgt)->frags[to];
2714
2715 if (todo >= skb_frag_size(fragfrom)) {
2716 *fragto = *fragfrom;
2717 todo -= skb_frag_size(fragfrom);
2718 from++;
2719 to++;
2720
2721 } else {
2722 __skb_frag_ref(fragfrom);
2723 fragto->page = fragfrom->page;
2724 fragto->page_offset = fragfrom->page_offset;
2725 skb_frag_size_set(fragto, todo);
2726
2727 fragfrom->page_offset += todo;
2728 skb_frag_size_sub(fragfrom, todo);
2729 todo = 0;
2730
2731 to++;
2732 break;
2733 }
2734 }
2735
2736 /* Ready to "commit" this state change to tgt */
2737 skb_shinfo(tgt)->nr_frags = to;
2738
2739 if (merge >= 0) {
2740 fragfrom = &skb_shinfo(skb)->frags[0];
2741 fragto = &skb_shinfo(tgt)->frags[merge];
2742
2743 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744 __skb_frag_unref(fragfrom);
2745 }
2746
2747 /* Reposition in the original skb */
2748 to = 0;
2749 while (from < skb_shinfo(skb)->nr_frags)
2750 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751 skb_shinfo(skb)->nr_frags = to;
2752
2753 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2754
2755onlymerged:
2756 /* Most likely the tgt won't ever need its checksum anymore, skb on
2757 * the other hand might need it if it needs to be resent
2758 */
2759 tgt->ip_summed = CHECKSUM_PARTIAL;
2760 skb->ip_summed = CHECKSUM_PARTIAL;
2761
2762 /* Yak, is it really working this way? Some helper please? */
2763 skb->len -= shiftlen;
2764 skb->data_len -= shiftlen;
2765 skb->truesize -= shiftlen;
2766 tgt->len += shiftlen;
2767 tgt->data_len += shiftlen;
2768 tgt->truesize += shiftlen;
2769
2770 return shiftlen;
2771}
2772
2773/**
2774 * skb_prepare_seq_read - Prepare a sequential read of skb data
2775 * @skb: the buffer to read
2776 * @from: lower offset of data to be read
2777 * @to: upper offset of data to be read
2778 * @st: state variable
2779 *
2780 * Initializes the specified state variable. Must be called before
2781 * invoking skb_seq_read() for the first time.
2782 */
2783void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784 unsigned int to, struct skb_seq_state *st)
2785{
2786 st->lower_offset = from;
2787 st->upper_offset = to;
2788 st->root_skb = st->cur_skb = skb;
2789 st->frag_idx = st->stepped_offset = 0;
2790 st->frag_data = NULL;
2791}
2792EXPORT_SYMBOL(skb_prepare_seq_read);
2793
2794/**
2795 * skb_seq_read - Sequentially read skb data
2796 * @consumed: number of bytes consumed by the caller so far
2797 * @data: destination pointer for data to be returned
2798 * @st: state variable
2799 *
2800 * Reads a block of skb data at @consumed relative to the
2801 * lower offset specified to skb_prepare_seq_read(). Assigns
2802 * the head of the data block to @data and returns the length
2803 * of the block or 0 if the end of the skb data or the upper
2804 * offset has been reached.
2805 *
2806 * The caller is not required to consume all of the data
2807 * returned, i.e. @consumed is typically set to the number
2808 * of bytes already consumed and the next call to
2809 * skb_seq_read() will return the remaining part of the block.
2810 *
2811 * Note 1: The size of each block of data returned can be arbitrary,
2812 * this limitation is the cost for zerocopy sequential
2813 * reads of potentially non linear data.
2814 *
2815 * Note 2: Fragment lists within fragments are not implemented
2816 * at the moment, state->root_skb could be replaced with
2817 * a stack for this purpose.
2818 */
2819unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820 struct skb_seq_state *st)
2821{
2822 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2823 skb_frag_t *frag;
2824
2825 if (unlikely(abs_offset >= st->upper_offset)) {
2826 if (st->frag_data) {
2827 kunmap_atomic(st->frag_data);
2828 st->frag_data = NULL;
2829 }
2830 return 0;
2831 }
2832
2833next_skb:
2834 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2835
2836 if (abs_offset < block_limit && !st->frag_data) {
2837 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838 return block_limit - abs_offset;
2839 }
2840
2841 if (st->frag_idx == 0 && !st->frag_data)
2842 st->stepped_offset += skb_headlen(st->cur_skb);
2843
2844 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846 block_limit = skb_frag_size(frag) + st->stepped_offset;
2847
2848 if (abs_offset < block_limit) {
2849 if (!st->frag_data)
2850 st->frag_data = kmap_atomic(skb_frag_page(frag));
2851
2852 *data = (u8 *) st->frag_data + frag->page_offset +
2853 (abs_offset - st->stepped_offset);
2854
2855 return block_limit - abs_offset;
2856 }
2857
2858 if (st->frag_data) {
2859 kunmap_atomic(st->frag_data);
2860 st->frag_data = NULL;
2861 }
2862
2863 st->frag_idx++;
2864 st->stepped_offset += skb_frag_size(frag);
2865 }
2866
2867 if (st->frag_data) {
2868 kunmap_atomic(st->frag_data);
2869 st->frag_data = NULL;
2870 }
2871
2872 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2874 st->frag_idx = 0;
2875 goto next_skb;
2876 } else if (st->cur_skb->next) {
2877 st->cur_skb = st->cur_skb->next;
2878 st->frag_idx = 0;
2879 goto next_skb;
2880 }
2881
2882 return 0;
2883}
2884EXPORT_SYMBOL(skb_seq_read);
2885
2886/**
2887 * skb_abort_seq_read - Abort a sequential read of skb data
2888 * @st: state variable
2889 *
2890 * Must be called if skb_seq_read() was not called until it
2891 * returned 0.
2892 */
2893void skb_abort_seq_read(struct skb_seq_state *st)
2894{
2895 if (st->frag_data)
2896 kunmap_atomic(st->frag_data);
2897}
2898EXPORT_SYMBOL(skb_abort_seq_read);
2899
2900#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2901
2902static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903 struct ts_config *conf,
2904 struct ts_state *state)
2905{
2906 return skb_seq_read(offset, text, TS_SKB_CB(state));
2907}
2908
2909static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2910{
2911 skb_abort_seq_read(TS_SKB_CB(state));
2912}
2913
2914/**
2915 * skb_find_text - Find a text pattern in skb data
2916 * @skb: the buffer to look in
2917 * @from: search offset
2918 * @to: search limit
2919 * @config: textsearch configuration
2920 *
2921 * Finds a pattern in the skb data according to the specified
2922 * textsearch configuration. Use textsearch_next() to retrieve
2923 * subsequent occurrences of the pattern. Returns the offset
2924 * to the first occurrence or UINT_MAX if no match was found.
2925 */
2926unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927 unsigned int to, struct ts_config *config)
2928{
2929 struct ts_state state;
2930 unsigned int ret;
2931
2932 config->get_next_block = skb_ts_get_next_block;
2933 config->finish = skb_ts_finish;
2934
2935 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2936
2937 ret = textsearch_find(config, &state);
2938 return (ret <= to - from ? ret : UINT_MAX);
2939}
2940EXPORT_SYMBOL(skb_find_text);
2941
2942/**
2943 * skb_append_datato_frags - append the user data to a skb
2944 * @sk: sock structure
2945 * @skb: skb structure to be appended with user data.
2946 * @getfrag: call back function to be used for getting the user data
2947 * @from: pointer to user message iov
2948 * @length: length of the iov message
2949 *
2950 * Description: This procedure append the user data in the fragment part
2951 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2952 */
2953int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954 int (*getfrag)(void *from, char *to, int offset,
2955 int len, int odd, struct sk_buff *skb),
2956 void *from, int length)
2957{
2958 int frg_cnt = skb_shinfo(skb)->nr_frags;
2959 int copy;
2960 int offset = 0;
2961 int ret;
2962 struct page_frag *pfrag = ¤t->task_frag;
2963
2964 do {
2965 /* Return error if we don't have space for new frag */
2966 if (frg_cnt >= MAX_SKB_FRAGS)
2967 return -EMSGSIZE;
2968
2969 if (!sk_page_frag_refill(sk, pfrag))
2970 return -ENOMEM;
2971
2972 /* copy the user data to page */
2973 copy = min_t(int, length, pfrag->size - pfrag->offset);
2974
2975 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976 offset, copy, 0, skb);
2977 if (ret < 0)
2978 return -EFAULT;
2979
2980 /* copy was successful so update the size parameters */
2981 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2982 copy);
2983 frg_cnt++;
2984 pfrag->offset += copy;
2985 get_page(pfrag->page);
2986
2987 skb->truesize += copy;
2988 atomic_add(copy, &sk->sk_wmem_alloc);
2989 skb->len += copy;
2990 skb->data_len += copy;
2991 offset += copy;
2992 length -= copy;
2993
2994 } while (length > 0);
2995
2996 return 0;
2997}
2998EXPORT_SYMBOL(skb_append_datato_frags);
2999
3000int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001 int offset, size_t size)
3002{
3003 int i = skb_shinfo(skb)->nr_frags;
3004
3005 if (skb_can_coalesce(skb, i, page, offset)) {
3006 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007 } else if (i < MAX_SKB_FRAGS) {
3008 get_page(page);
3009 skb_fill_page_desc(skb, i, page, offset, size);
3010 } else {
3011 return -EMSGSIZE;
3012 }
3013
3014 return 0;
3015}
3016EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3017
3018/**
3019 * skb_push_rcsum - push skb and update receive checksum
3020 * @skb: buffer to update
3021 * @len: length of data pulled
3022 *
3023 * This function performs an skb_push on the packet and updates
3024 * the CHECKSUM_COMPLETE checksum. It should be used on
3025 * receive path processing instead of skb_push unless you know
3026 * that the checksum difference is zero (e.g., a valid IP header)
3027 * or you are setting ip_summed to CHECKSUM_NONE.
3028 */
3029static unsigned char *skb_push_rcsum(struct sk_buff *skb, unsigned len)
3030{
3031 skb_push(skb, len);
3032 skb_postpush_rcsum(skb, skb->data, len);
3033 return skb->data;
3034}
3035
3036/**
3037 * skb_pull_rcsum - pull skb and update receive checksum
3038 * @skb: buffer to update
3039 * @len: length of data pulled
3040 *
3041 * This function performs an skb_pull on the packet and updates
3042 * the CHECKSUM_COMPLETE checksum. It should be used on
3043 * receive path processing instead of skb_pull unless you know
3044 * that the checksum difference is zero (e.g., a valid IP header)
3045 * or you are setting ip_summed to CHECKSUM_NONE.
3046 */
3047unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3048{
3049 unsigned char *data = skb->data;
3050
3051 BUG_ON(len > skb->len);
3052 __skb_pull(skb, len);
3053 skb_postpull_rcsum(skb, data, len);
3054 return skb->data;
3055}
3056EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3057
3058/**
3059 * skb_segment - Perform protocol segmentation on skb.
3060 * @head_skb: buffer to segment
3061 * @features: features for the output path (see dev->features)
3062 *
3063 * This function performs segmentation on the given skb. It returns
3064 * a pointer to the first in a list of new skbs for the segments.
3065 * In case of error it returns ERR_PTR(err).
3066 */
3067struct sk_buff *skb_segment(struct sk_buff *head_skb,
3068 netdev_features_t features)
3069{
3070 struct sk_buff *segs = NULL;
3071 struct sk_buff *tail = NULL;
3072 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3073 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3074 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3075 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3076 struct sk_buff *frag_skb = head_skb;
3077 unsigned int offset = doffset;
3078 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3079 unsigned int headroom;
3080 unsigned int len;
3081 __be16 proto;
3082 bool csum;
3083 int sg = !!(features & NETIF_F_SG);
3084 int nfrags = skb_shinfo(head_skb)->nr_frags;
3085 int err = -ENOMEM;
3086 int i = 0;
3087 int pos;
3088 int dummy;
3089
3090 __skb_push(head_skb, doffset);
3091 proto = skb_network_protocol(head_skb, &dummy);
3092 if (unlikely(!proto))
3093 return ERR_PTR(-EINVAL);
3094
3095 csum = !!can_checksum_protocol(features, proto);
3096
3097 headroom = skb_headroom(head_skb);
3098 pos = skb_headlen(head_skb);
3099
3100 do {
3101 struct sk_buff *nskb;
3102 skb_frag_t *nskb_frag;
3103 int hsize;
3104 int size;
3105
3106 len = head_skb->len - offset;
3107 if (len > mss)
3108 len = mss;
3109
3110 hsize = skb_headlen(head_skb) - offset;
3111 if (hsize < 0)
3112 hsize = 0;
3113 if (hsize > len || !sg)
3114 hsize = len;
3115
3116 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3117 (skb_headlen(list_skb) == len || sg)) {
3118 BUG_ON(skb_headlen(list_skb) > len);
3119
3120 i = 0;
3121 nfrags = skb_shinfo(list_skb)->nr_frags;
3122 frag = skb_shinfo(list_skb)->frags;
3123 frag_skb = list_skb;
3124 pos += skb_headlen(list_skb);
3125
3126 while (pos < offset + len) {
3127 BUG_ON(i >= nfrags);
3128
3129 size = skb_frag_size(frag);
3130 if (pos + size > offset + len)
3131 break;
3132
3133 i++;
3134 pos += size;
3135 frag++;
3136 }
3137
3138 nskb = skb_clone(list_skb, GFP_ATOMIC);
3139 list_skb = list_skb->next;
3140
3141 if (unlikely(!nskb))
3142 goto err;
3143
3144 if (unlikely(pskb_trim(nskb, len))) {
3145 kfree_skb(nskb);
3146 goto err;
3147 }
3148
3149 hsize = skb_end_offset(nskb);
3150 if (skb_cow_head(nskb, doffset + headroom)) {
3151 kfree_skb(nskb);
3152 goto err;
3153 }
3154
3155 nskb->truesize += skb_end_offset(nskb) - hsize;
3156 skb_release_head_state(nskb);
3157 __skb_push(nskb, doffset);
3158 } else {
3159 nskb = __alloc_skb(hsize + doffset + headroom,
3160 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3161 NUMA_NO_NODE);
3162
3163 if (unlikely(!nskb))
3164 goto err;
3165
3166 skb_reserve(nskb, headroom);
3167 __skb_put(nskb, doffset);
3168 }
3169
3170 if (segs)
3171 tail->next = nskb;
3172 else
3173 segs = nskb;
3174 tail = nskb;
3175
3176 __copy_skb_header(nskb, head_skb);
3177
3178 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3179 skb_reset_mac_len(nskb);
3180
3181 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3182 nskb->data - tnl_hlen,
3183 doffset + tnl_hlen);
3184
3185 if (nskb->len == len + doffset)
3186 goto perform_csum_check;
3187
3188 if (!sg) {
3189 if (!nskb->remcsum_offload)
3190 nskb->ip_summed = CHECKSUM_NONE;
3191 SKB_GSO_CB(nskb)->csum =
3192 skb_copy_and_csum_bits(head_skb, offset,
3193 skb_put(nskb, len),
3194 len, 0);
3195 SKB_GSO_CB(nskb)->csum_start =
3196 skb_headroom(nskb) + doffset;
3197 continue;
3198 }
3199
3200 nskb_frag = skb_shinfo(nskb)->frags;
3201
3202 skb_copy_from_linear_data_offset(head_skb, offset,
3203 skb_put(nskb, hsize), hsize);
3204
3205 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3206 SKBTX_SHARED_FRAG;
3207
3208 while (pos < offset + len) {
3209 if (i >= nfrags) {
3210 BUG_ON(skb_headlen(list_skb));
3211
3212 i = 0;
3213 nfrags = skb_shinfo(list_skb)->nr_frags;
3214 frag = skb_shinfo(list_skb)->frags;
3215 frag_skb = list_skb;
3216
3217 BUG_ON(!nfrags);
3218
3219 list_skb = list_skb->next;
3220 }
3221
3222 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3223 MAX_SKB_FRAGS)) {
3224 net_warn_ratelimited(
3225 "skb_segment: too many frags: %u %u\n",
3226 pos, mss);
3227 goto err;
3228 }
3229
3230 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3231 goto err;
3232
3233 *nskb_frag = *frag;
3234 __skb_frag_ref(nskb_frag);
3235 size = skb_frag_size(nskb_frag);
3236
3237 if (pos < offset) {
3238 nskb_frag->page_offset += offset - pos;
3239 skb_frag_size_sub(nskb_frag, offset - pos);
3240 }
3241
3242 skb_shinfo(nskb)->nr_frags++;
3243
3244 if (pos + size <= offset + len) {
3245 i++;
3246 frag++;
3247 pos += size;
3248 } else {
3249 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3250 goto skip_fraglist;
3251 }
3252
3253 nskb_frag++;
3254 }
3255
3256skip_fraglist:
3257 nskb->data_len = len - hsize;
3258 nskb->len += nskb->data_len;
3259 nskb->truesize += nskb->data_len;
3260
3261perform_csum_check:
3262 if (!csum) {
3263 if (skb_has_shared_frag(nskb)) {
3264 err = __skb_linearize(nskb);
3265 if (err)
3266 goto err;
3267 }
3268 if (!nskb->remcsum_offload)
3269 nskb->ip_summed = CHECKSUM_NONE;
3270 SKB_GSO_CB(nskb)->csum =
3271 skb_checksum(nskb, doffset,
3272 nskb->len - doffset, 0);
3273 SKB_GSO_CB(nskb)->csum_start =
3274 skb_headroom(nskb) + doffset;
3275 }
3276 } while ((offset += len) < head_skb->len);
3277
3278 /* Some callers want to get the end of the list.
3279 * Put it in segs->prev to avoid walking the list.
3280 * (see validate_xmit_skb_list() for example)
3281 */
3282 segs->prev = tail;
3283
3284 /* Following permits correct backpressure, for protocols
3285 * using skb_set_owner_w().
3286 * Idea is to tranfert ownership from head_skb to last segment.
3287 */
3288 if (head_skb->destructor == sock_wfree) {
3289 swap(tail->truesize, head_skb->truesize);
3290 swap(tail->destructor, head_skb->destructor);
3291 swap(tail->sk, head_skb->sk);
3292 }
3293 return segs;
3294
3295err:
3296 kfree_skb_list(segs);
3297 return ERR_PTR(err);
3298}
3299EXPORT_SYMBOL_GPL(skb_segment);
3300
3301int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3302{
3303 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3304 unsigned int offset = skb_gro_offset(skb);
3305 unsigned int headlen = skb_headlen(skb);
3306 unsigned int len = skb_gro_len(skb);
3307 struct sk_buff *lp, *p = *head;
3308 unsigned int delta_truesize;
3309
3310 if (unlikely(p->len + len >= 65536))
3311 return -E2BIG;
3312
3313 lp = NAPI_GRO_CB(p)->last;
3314 pinfo = skb_shinfo(lp);
3315
3316 if (headlen <= offset) {
3317 skb_frag_t *frag;
3318 skb_frag_t *frag2;
3319 int i = skbinfo->nr_frags;
3320 int nr_frags = pinfo->nr_frags + i;
3321
3322 if (nr_frags > MAX_SKB_FRAGS)
3323 goto merge;
3324
3325 offset -= headlen;
3326 pinfo->nr_frags = nr_frags;
3327 skbinfo->nr_frags = 0;
3328
3329 frag = pinfo->frags + nr_frags;
3330 frag2 = skbinfo->frags + i;
3331 do {
3332 *--frag = *--frag2;
3333 } while (--i);
3334
3335 frag->page_offset += offset;
3336 skb_frag_size_sub(frag, offset);
3337
3338 /* all fragments truesize : remove (head size + sk_buff) */
3339 delta_truesize = skb->truesize -
3340 SKB_TRUESIZE(skb_end_offset(skb));
3341
3342 skb->truesize -= skb->data_len;
3343 skb->len -= skb->data_len;
3344 skb->data_len = 0;
3345
3346 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3347 goto done;
3348 } else if (skb->head_frag) {
3349 int nr_frags = pinfo->nr_frags;
3350 skb_frag_t *frag = pinfo->frags + nr_frags;
3351 struct page *page = virt_to_head_page(skb->head);
3352 unsigned int first_size = headlen - offset;
3353 unsigned int first_offset;
3354
3355 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3356 goto merge;
3357
3358 first_offset = skb->data -
3359 (unsigned char *)page_address(page) +
3360 offset;
3361
3362 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3363
3364 frag->page.p = page;
3365 frag->page_offset = first_offset;
3366 skb_frag_size_set(frag, first_size);
3367
3368 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3369 /* We dont need to clear skbinfo->nr_frags here */
3370
3371 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3372 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3373 goto done;
3374 }
3375
3376merge:
3377 delta_truesize = skb->truesize;
3378 if (offset > headlen) {
3379 unsigned int eat = offset - headlen;
3380
3381 skbinfo->frags[0].page_offset += eat;
3382 skb_frag_size_sub(&skbinfo->frags[0], eat);
3383 skb->data_len -= eat;
3384 skb->len -= eat;
3385 offset = headlen;
3386 }
3387
3388 __skb_pull(skb, offset);
3389
3390 if (NAPI_GRO_CB(p)->last == p)
3391 skb_shinfo(p)->frag_list = skb;
3392 else
3393 NAPI_GRO_CB(p)->last->next = skb;
3394 NAPI_GRO_CB(p)->last = skb;
3395 __skb_header_release(skb);
3396 lp = p;
3397
3398done:
3399 NAPI_GRO_CB(p)->count++;
3400 p->data_len += len;
3401 p->truesize += delta_truesize;
3402 p->len += len;
3403 if (lp != p) {
3404 lp->data_len += len;
3405 lp->truesize += delta_truesize;
3406 lp->len += len;
3407 }
3408 NAPI_GRO_CB(skb)->same_flow = 1;
3409 return 0;
3410}
3411
3412void __init skb_init(void)
3413{
3414 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3415 sizeof(struct sk_buff),
3416 0,
3417 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3418 NULL);
3419 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3420 sizeof(struct sk_buff_fclones),
3421 0,
3422 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3423 NULL);
3424}
3425
3426/**
3427 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3428 * @skb: Socket buffer containing the buffers to be mapped
3429 * @sg: The scatter-gather list to map into
3430 * @offset: The offset into the buffer's contents to start mapping
3431 * @len: Length of buffer space to be mapped
3432 *
3433 * Fill the specified scatter-gather list with mappings/pointers into a
3434 * region of the buffer space attached to a socket buffer.
3435 */
3436static int
3437__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3438{
3439 int start = skb_headlen(skb);
3440 int i, copy = start - offset;
3441 struct sk_buff *frag_iter;
3442 int elt = 0;
3443
3444 if (copy > 0) {
3445 if (copy > len)
3446 copy = len;
3447 sg_set_buf(sg, skb->data + offset, copy);
3448 elt++;
3449 if ((len -= copy) == 0)
3450 return elt;
3451 offset += copy;
3452 }
3453
3454 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3455 int end;
3456
3457 WARN_ON(start > offset + len);
3458
3459 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3460 if ((copy = end - offset) > 0) {
3461 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3462
3463 if (copy > len)
3464 copy = len;
3465 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3466 frag->page_offset+offset-start);
3467 elt++;
3468 if (!(len -= copy))
3469 return elt;
3470 offset += copy;
3471 }
3472 start = end;
3473 }
3474
3475 skb_walk_frags(skb, frag_iter) {
3476 int end;
3477
3478 WARN_ON(start > offset + len);
3479
3480 end = start + frag_iter->len;
3481 if ((copy = end - offset) > 0) {
3482 if (copy > len)
3483 copy = len;
3484 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3485 copy);
3486 if ((len -= copy) == 0)
3487 return elt;
3488 offset += copy;
3489 }
3490 start = end;
3491 }
3492 BUG_ON(len);
3493 return elt;
3494}
3495
3496/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3497 * sglist without mark the sg which contain last skb data as the end.
3498 * So the caller can mannipulate sg list as will when padding new data after
3499 * the first call without calling sg_unmark_end to expend sg list.
3500 *
3501 * Scenario to use skb_to_sgvec_nomark:
3502 * 1. sg_init_table
3503 * 2. skb_to_sgvec_nomark(payload1)
3504 * 3. skb_to_sgvec_nomark(payload2)
3505 *
3506 * This is equivalent to:
3507 * 1. sg_init_table
3508 * 2. skb_to_sgvec(payload1)
3509 * 3. sg_unmark_end
3510 * 4. skb_to_sgvec(payload2)
3511 *
3512 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3513 * is more preferable.
3514 */
3515int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3516 int offset, int len)
3517{
3518 return __skb_to_sgvec(skb, sg, offset, len);
3519}
3520EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3521
3522int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3523{
3524 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3525
3526 sg_mark_end(&sg[nsg - 1]);
3527
3528 return nsg;
3529}
3530EXPORT_SYMBOL_GPL(skb_to_sgvec);
3531
3532/**
3533 * skb_cow_data - Check that a socket buffer's data buffers are writable
3534 * @skb: The socket buffer to check.
3535 * @tailbits: Amount of trailing space to be added
3536 * @trailer: Returned pointer to the skb where the @tailbits space begins
3537 *
3538 * Make sure that the data buffers attached to a socket buffer are
3539 * writable. If they are not, private copies are made of the data buffers
3540 * and the socket buffer is set to use these instead.
3541 *
3542 * If @tailbits is given, make sure that there is space to write @tailbits
3543 * bytes of data beyond current end of socket buffer. @trailer will be
3544 * set to point to the skb in which this space begins.
3545 *
3546 * The number of scatterlist elements required to completely map the
3547 * COW'd and extended socket buffer will be returned.
3548 */
3549int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3550{
3551 int copyflag;
3552 int elt;
3553 struct sk_buff *skb1, **skb_p;
3554
3555 /* If skb is cloned or its head is paged, reallocate
3556 * head pulling out all the pages (pages are considered not writable
3557 * at the moment even if they are anonymous).
3558 */
3559 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3560 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3561 return -ENOMEM;
3562
3563 /* Easy case. Most of packets will go this way. */
3564 if (!skb_has_frag_list(skb)) {
3565 /* A little of trouble, not enough of space for trailer.
3566 * This should not happen, when stack is tuned to generate
3567 * good frames. OK, on miss we reallocate and reserve even more
3568 * space, 128 bytes is fair. */
3569
3570 if (skb_tailroom(skb) < tailbits &&
3571 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3572 return -ENOMEM;
3573
3574 /* Voila! */
3575 *trailer = skb;
3576 return 1;
3577 }
3578
3579 /* Misery. We are in troubles, going to mincer fragments... */
3580
3581 elt = 1;
3582 skb_p = &skb_shinfo(skb)->frag_list;
3583 copyflag = 0;
3584
3585 while ((skb1 = *skb_p) != NULL) {
3586 int ntail = 0;
3587
3588 /* The fragment is partially pulled by someone,
3589 * this can happen on input. Copy it and everything
3590 * after it. */
3591
3592 if (skb_shared(skb1))
3593 copyflag = 1;
3594
3595 /* If the skb is the last, worry about trailer. */
3596
3597 if (skb1->next == NULL && tailbits) {
3598 if (skb_shinfo(skb1)->nr_frags ||
3599 skb_has_frag_list(skb1) ||
3600 skb_tailroom(skb1) < tailbits)
3601 ntail = tailbits + 128;
3602 }
3603
3604 if (copyflag ||
3605 skb_cloned(skb1) ||
3606 ntail ||
3607 skb_shinfo(skb1)->nr_frags ||
3608 skb_has_frag_list(skb1)) {
3609 struct sk_buff *skb2;
3610
3611 /* Fuck, we are miserable poor guys... */
3612 if (ntail == 0)
3613 skb2 = skb_copy(skb1, GFP_ATOMIC);
3614 else
3615 skb2 = skb_copy_expand(skb1,
3616 skb_headroom(skb1),
3617 ntail,
3618 GFP_ATOMIC);
3619 if (unlikely(skb2 == NULL))
3620 return -ENOMEM;
3621
3622 if (skb1->sk)
3623 skb_set_owner_w(skb2, skb1->sk);
3624
3625 /* Looking around. Are we still alive?
3626 * OK, link new skb, drop old one */
3627
3628 skb2->next = skb1->next;
3629 *skb_p = skb2;
3630 kfree_skb(skb1);
3631 skb1 = skb2;
3632 }
3633 elt++;
3634 *trailer = skb1;
3635 skb_p = &skb1->next;
3636 }
3637
3638 return elt;
3639}
3640EXPORT_SYMBOL_GPL(skb_cow_data);
3641
3642static void sock_rmem_free(struct sk_buff *skb)
3643{
3644 struct sock *sk = skb->sk;
3645
3646 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3647}
3648
3649/*
3650 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3651 */
3652int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3653{
3654 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3655 (unsigned int)sk->sk_rcvbuf)
3656 return -ENOMEM;
3657
3658 skb_orphan(skb);
3659 skb->sk = sk;
3660 skb->destructor = sock_rmem_free;
3661 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3662
3663 /* before exiting rcu section, make sure dst is refcounted */
3664 skb_dst_force(skb);
3665
3666 skb_queue_tail(&sk->sk_error_queue, skb);
3667 if (!sock_flag(sk, SOCK_DEAD))
3668 sk->sk_data_ready(sk);
3669 return 0;
3670}
3671EXPORT_SYMBOL(sock_queue_err_skb);
3672
3673struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3674{
3675 struct sk_buff_head *q = &sk->sk_error_queue;
3676 struct sk_buff *skb, *skb_next;
3677 unsigned long flags;
3678 int err = 0;
3679
3680 spin_lock_irqsave(&q->lock, flags);
3681 skb = __skb_dequeue(q);
3682 if (skb && (skb_next = skb_peek(q)))
3683 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3684 spin_unlock_irqrestore(&q->lock, flags);
3685
3686 sk->sk_err = err;
3687 if (err)
3688 sk->sk_error_report(sk);
3689
3690 return skb;
3691}
3692EXPORT_SYMBOL(sock_dequeue_err_skb);
3693
3694/**
3695 * skb_clone_sk - create clone of skb, and take reference to socket
3696 * @skb: the skb to clone
3697 *
3698 * This function creates a clone of a buffer that holds a reference on
3699 * sk_refcnt. Buffers created via this function are meant to be
3700 * returned using sock_queue_err_skb, or free via kfree_skb.
3701 *
3702 * When passing buffers allocated with this function to sock_queue_err_skb
3703 * it is necessary to wrap the call with sock_hold/sock_put in order to
3704 * prevent the socket from being released prior to being enqueued on
3705 * the sk_error_queue.
3706 */
3707struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3708{
3709 struct sock *sk = skb->sk;
3710 struct sk_buff *clone;
3711
3712 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3713 return NULL;
3714
3715 clone = skb_clone(skb, GFP_ATOMIC);
3716 if (!clone) {
3717 sock_put(sk);
3718 return NULL;
3719 }
3720
3721 clone->sk = sk;
3722 clone->destructor = sock_efree;
3723
3724 return clone;
3725}
3726EXPORT_SYMBOL(skb_clone_sk);
3727
3728static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3729 struct sock *sk,
3730 int tstype)
3731{
3732 struct sock_exterr_skb *serr;
3733 int err;
3734
3735 serr = SKB_EXT_ERR(skb);
3736 memset(serr, 0, sizeof(*serr));
3737 serr->ee.ee_errno = ENOMSG;
3738 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3739 serr->ee.ee_info = tstype;
3740 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3741 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3742 if (sk->sk_protocol == IPPROTO_TCP &&
3743 sk->sk_type == SOCK_STREAM)
3744 serr->ee.ee_data -= sk->sk_tskey;
3745 }
3746
3747 err = sock_queue_err_skb(sk, skb);
3748
3749 if (err)
3750 kfree_skb(skb);
3751}
3752
3753static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3754{
3755 bool ret;
3756
3757 if (likely(sysctl_tstamp_allow_data || tsonly))
3758 return true;
3759
3760 read_lock_bh(&sk->sk_callback_lock);
3761 ret = sk->sk_socket && sk->sk_socket->file &&
3762 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3763 read_unlock_bh(&sk->sk_callback_lock);
3764 return ret;
3765}
3766
3767void skb_complete_tx_timestamp(struct sk_buff *skb,
3768 struct skb_shared_hwtstamps *hwtstamps)
3769{
3770 struct sock *sk = skb->sk;
3771
3772 if (!skb_may_tx_timestamp(sk, false))
3773 return;
3774
3775 /* take a reference to prevent skb_orphan() from freeing the socket */
3776 sock_hold(sk);
3777
3778 *skb_hwtstamps(skb) = *hwtstamps;
3779 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3780
3781 sock_put(sk);
3782}
3783EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3784
3785void __skb_tstamp_tx(struct sk_buff *orig_skb,
3786 struct skb_shared_hwtstamps *hwtstamps,
3787 struct sock *sk, int tstype)
3788{
3789 struct sk_buff *skb;
3790 bool tsonly;
3791
3792 if (!sk)
3793 return;
3794
3795 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3796 if (!skb_may_tx_timestamp(sk, tsonly))
3797 return;
3798
3799 if (tsonly)
3800 skb = alloc_skb(0, GFP_ATOMIC);
3801 else
3802 skb = skb_clone(orig_skb, GFP_ATOMIC);
3803 if (!skb)
3804 return;
3805
3806 if (tsonly) {
3807 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3808 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3809 }
3810
3811 if (hwtstamps)
3812 *skb_hwtstamps(skb) = *hwtstamps;
3813 else
3814 skb->tstamp = ktime_get_real();
3815
3816 __skb_complete_tx_timestamp(skb, sk, tstype);
3817}
3818EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3819
3820void skb_tstamp_tx(struct sk_buff *orig_skb,
3821 struct skb_shared_hwtstamps *hwtstamps)
3822{
3823 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3824 SCM_TSTAMP_SND);
3825}
3826EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3827
3828void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3829{
3830 struct sock *sk = skb->sk;
3831 struct sock_exterr_skb *serr;
3832 int err;
3833
3834 skb->wifi_acked_valid = 1;
3835 skb->wifi_acked = acked;
3836
3837 serr = SKB_EXT_ERR(skb);
3838 memset(serr, 0, sizeof(*serr));
3839 serr->ee.ee_errno = ENOMSG;
3840 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3841
3842 /* take a reference to prevent skb_orphan() from freeing the socket */
3843 sock_hold(sk);
3844
3845 err = sock_queue_err_skb(sk, skb);
3846 if (err)
3847 kfree_skb(skb);
3848
3849 sock_put(sk);
3850}
3851EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3852
3853/**
3854 * skb_partial_csum_set - set up and verify partial csum values for packet
3855 * @skb: the skb to set
3856 * @start: the number of bytes after skb->data to start checksumming.
3857 * @off: the offset from start to place the checksum.
3858 *
3859 * For untrusted partially-checksummed packets, we need to make sure the values
3860 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3861 *
3862 * This function checks and sets those values and skb->ip_summed: if this
3863 * returns false you should drop the packet.
3864 */
3865bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3866{
3867 if (unlikely(start > skb_headlen(skb)) ||
3868 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3869 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3870 start, off, skb_headlen(skb));
3871 return false;
3872 }
3873 skb->ip_summed = CHECKSUM_PARTIAL;
3874 skb->csum_start = skb_headroom(skb) + start;
3875 skb->csum_offset = off;
3876 skb_set_transport_header(skb, start);
3877 return true;
3878}
3879EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3880
3881static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3882 unsigned int max)
3883{
3884 if (skb_headlen(skb) >= len)
3885 return 0;
3886
3887 /* If we need to pullup then pullup to the max, so we
3888 * won't need to do it again.
3889 */
3890 if (max > skb->len)
3891 max = skb->len;
3892
3893 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3894 return -ENOMEM;
3895
3896 if (skb_headlen(skb) < len)
3897 return -EPROTO;
3898
3899 return 0;
3900}
3901
3902#define MAX_TCP_HDR_LEN (15 * 4)
3903
3904static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3905 typeof(IPPROTO_IP) proto,
3906 unsigned int off)
3907{
3908 switch (proto) {
3909 int err;
3910
3911 case IPPROTO_TCP:
3912 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3913 off + MAX_TCP_HDR_LEN);
3914 if (!err && !skb_partial_csum_set(skb, off,
3915 offsetof(struct tcphdr,
3916 check)))
3917 err = -EPROTO;
3918 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3919
3920 case IPPROTO_UDP:
3921 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3922 off + sizeof(struct udphdr));
3923 if (!err && !skb_partial_csum_set(skb, off,
3924 offsetof(struct udphdr,
3925 check)))
3926 err = -EPROTO;
3927 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3928 }
3929
3930 return ERR_PTR(-EPROTO);
3931}
3932
3933/* This value should be large enough to cover a tagged ethernet header plus
3934 * maximally sized IP and TCP or UDP headers.
3935 */
3936#define MAX_IP_HDR_LEN 128
3937
3938static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3939{
3940 unsigned int off;
3941 bool fragment;
3942 __sum16 *csum;
3943 int err;
3944
3945 fragment = false;
3946
3947 err = skb_maybe_pull_tail(skb,
3948 sizeof(struct iphdr),
3949 MAX_IP_HDR_LEN);
3950 if (err < 0)
3951 goto out;
3952
3953 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3954 fragment = true;
3955
3956 off = ip_hdrlen(skb);
3957
3958 err = -EPROTO;
3959
3960 if (fragment)
3961 goto out;
3962
3963 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3964 if (IS_ERR(csum))
3965 return PTR_ERR(csum);
3966
3967 if (recalculate)
3968 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3969 ip_hdr(skb)->daddr,
3970 skb->len - off,
3971 ip_hdr(skb)->protocol, 0);
3972 err = 0;
3973
3974out:
3975 return err;
3976}
3977
3978/* This value should be large enough to cover a tagged ethernet header plus
3979 * an IPv6 header, all options, and a maximal TCP or UDP header.
3980 */
3981#define MAX_IPV6_HDR_LEN 256
3982
3983#define OPT_HDR(type, skb, off) \
3984 (type *)(skb_network_header(skb) + (off))
3985
3986static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3987{
3988 int err;
3989 u8 nexthdr;
3990 unsigned int off;
3991 unsigned int len;
3992 bool fragment;
3993 bool done;
3994 __sum16 *csum;
3995
3996 fragment = false;
3997 done = false;
3998
3999 off = sizeof(struct ipv6hdr);
4000
4001 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4002 if (err < 0)
4003 goto out;
4004
4005 nexthdr = ipv6_hdr(skb)->nexthdr;
4006
4007 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4008 while (off <= len && !done) {
4009 switch (nexthdr) {
4010 case IPPROTO_DSTOPTS:
4011 case IPPROTO_HOPOPTS:
4012 case IPPROTO_ROUTING: {
4013 struct ipv6_opt_hdr *hp;
4014
4015 err = skb_maybe_pull_tail(skb,
4016 off +
4017 sizeof(struct ipv6_opt_hdr),
4018 MAX_IPV6_HDR_LEN);
4019 if (err < 0)
4020 goto out;
4021
4022 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4023 nexthdr = hp->nexthdr;
4024 off += ipv6_optlen(hp);
4025 break;
4026 }
4027 case IPPROTO_AH: {
4028 struct ip_auth_hdr *hp;
4029
4030 err = skb_maybe_pull_tail(skb,
4031 off +
4032 sizeof(struct ip_auth_hdr),
4033 MAX_IPV6_HDR_LEN);
4034 if (err < 0)
4035 goto out;
4036
4037 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4038 nexthdr = hp->nexthdr;
4039 off += ipv6_authlen(hp);
4040 break;
4041 }
4042 case IPPROTO_FRAGMENT: {
4043 struct frag_hdr *hp;
4044
4045 err = skb_maybe_pull_tail(skb,
4046 off +
4047 sizeof(struct frag_hdr),
4048 MAX_IPV6_HDR_LEN);
4049 if (err < 0)
4050 goto out;
4051
4052 hp = OPT_HDR(struct frag_hdr, skb, off);
4053
4054 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4055 fragment = true;
4056
4057 nexthdr = hp->nexthdr;
4058 off += sizeof(struct frag_hdr);
4059 break;
4060 }
4061 default:
4062 done = true;
4063 break;
4064 }
4065 }
4066
4067 err = -EPROTO;
4068
4069 if (!done || fragment)
4070 goto out;
4071
4072 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4073 if (IS_ERR(csum))
4074 return PTR_ERR(csum);
4075
4076 if (recalculate)
4077 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4078 &ipv6_hdr(skb)->daddr,
4079 skb->len - off, nexthdr, 0);
4080 err = 0;
4081
4082out:
4083 return err;
4084}
4085
4086/**
4087 * skb_checksum_setup - set up partial checksum offset
4088 * @skb: the skb to set up
4089 * @recalculate: if true the pseudo-header checksum will be recalculated
4090 */
4091int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4092{
4093 int err;
4094
4095 switch (skb->protocol) {
4096 case htons(ETH_P_IP):
4097 err = skb_checksum_setup_ipv4(skb, recalculate);
4098 break;
4099
4100 case htons(ETH_P_IPV6):
4101 err = skb_checksum_setup_ipv6(skb, recalculate);
4102 break;
4103
4104 default:
4105 err = -EPROTO;
4106 break;
4107 }
4108
4109 return err;
4110}
4111EXPORT_SYMBOL(skb_checksum_setup);
4112
4113/**
4114 * skb_checksum_maybe_trim - maybe trims the given skb
4115 * @skb: the skb to check
4116 * @transport_len: the data length beyond the network header
4117 *
4118 * Checks whether the given skb has data beyond the given transport length.
4119 * If so, returns a cloned skb trimmed to this transport length.
4120 * Otherwise returns the provided skb. Returns NULL in error cases
4121 * (e.g. transport_len exceeds skb length or out-of-memory).
4122 *
4123 * Caller needs to set the skb transport header and free any returned skb if it
4124 * differs from the provided skb.
4125 */
4126static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4127 unsigned int transport_len)
4128{
4129 struct sk_buff *skb_chk;
4130 unsigned int len = skb_transport_offset(skb) + transport_len;
4131 int ret;
4132
4133 if (skb->len < len)
4134 return NULL;
4135 else if (skb->len == len)
4136 return skb;
4137
4138 skb_chk = skb_clone(skb, GFP_ATOMIC);
4139 if (!skb_chk)
4140 return NULL;
4141
4142 ret = pskb_trim_rcsum(skb_chk, len);
4143 if (ret) {
4144 kfree_skb(skb_chk);
4145 return NULL;
4146 }
4147
4148 return skb_chk;
4149}
4150
4151/**
4152 * skb_checksum_trimmed - validate checksum of an skb
4153 * @skb: the skb to check
4154 * @transport_len: the data length beyond the network header
4155 * @skb_chkf: checksum function to use
4156 *
4157 * Applies the given checksum function skb_chkf to the provided skb.
4158 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4159 *
4160 * If the skb has data beyond the given transport length, then a
4161 * trimmed & cloned skb is checked and returned.
4162 *
4163 * Caller needs to set the skb transport header and free any returned skb if it
4164 * differs from the provided skb.
4165 */
4166struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4167 unsigned int transport_len,
4168 __sum16(*skb_chkf)(struct sk_buff *skb))
4169{
4170 struct sk_buff *skb_chk;
4171 unsigned int offset = skb_transport_offset(skb);
4172 __sum16 ret;
4173
4174 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4175 if (!skb_chk)
4176 goto err;
4177
4178 if (!pskb_may_pull(skb_chk, offset))
4179 goto err;
4180
4181 skb_pull_rcsum(skb_chk, offset);
4182 ret = skb_chkf(skb_chk);
4183 skb_push_rcsum(skb_chk, offset);
4184
4185 if (ret)
4186 goto err;
4187
4188 return skb_chk;
4189
4190err:
4191 if (skb_chk && skb_chk != skb)
4192 kfree_skb(skb_chk);
4193
4194 return NULL;
4195
4196}
4197EXPORT_SYMBOL(skb_checksum_trimmed);
4198
4199void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4200{
4201 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4202 skb->dev->name);
4203}
4204EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4205
4206void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4207{
4208 if (head_stolen) {
4209 skb_release_head_state(skb);
4210 kmem_cache_free(skbuff_head_cache, skb);
4211 } else {
4212 __kfree_skb(skb);
4213 }
4214}
4215EXPORT_SYMBOL(kfree_skb_partial);
4216
4217/**
4218 * skb_try_coalesce - try to merge skb to prior one
4219 * @to: prior buffer
4220 * @from: buffer to add
4221 * @fragstolen: pointer to boolean
4222 * @delta_truesize: how much more was allocated than was requested
4223 */
4224bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4225 bool *fragstolen, int *delta_truesize)
4226{
4227 int i, delta, len = from->len;
4228
4229 *fragstolen = false;
4230
4231 if (skb_cloned(to))
4232 return false;
4233
4234 if (len <= skb_tailroom(to)) {
4235 if (len)
4236 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4237 *delta_truesize = 0;
4238 return true;
4239 }
4240
4241 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4242 return false;
4243
4244 if (skb_headlen(from) != 0) {
4245 struct page *page;
4246 unsigned int offset;
4247
4248 if (skb_shinfo(to)->nr_frags +
4249 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4250 return false;
4251
4252 if (skb_head_is_locked(from))
4253 return false;
4254
4255 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4256
4257 page = virt_to_head_page(from->head);
4258 offset = from->data - (unsigned char *)page_address(page);
4259
4260 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4261 page, offset, skb_headlen(from));
4262 *fragstolen = true;
4263 } else {
4264 if (skb_shinfo(to)->nr_frags +
4265 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4266 return false;
4267
4268 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4269 }
4270
4271 WARN_ON_ONCE(delta < len);
4272
4273 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4274 skb_shinfo(from)->frags,
4275 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4276 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4277
4278 if (!skb_cloned(from))
4279 skb_shinfo(from)->nr_frags = 0;
4280
4281 /* if the skb is not cloned this does nothing
4282 * since we set nr_frags to 0.
4283 */
4284 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4285 skb_frag_ref(from, i);
4286
4287 to->truesize += delta;
4288 to->len += len;
4289 to->data_len += len;
4290
4291 *delta_truesize = delta;
4292 return true;
4293}
4294EXPORT_SYMBOL(skb_try_coalesce);
4295
4296/**
4297 * skb_scrub_packet - scrub an skb
4298 *
4299 * @skb: buffer to clean
4300 * @xnet: packet is crossing netns
4301 *
4302 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4303 * into/from a tunnel. Some information have to be cleared during these
4304 * operations.
4305 * skb_scrub_packet can also be used to clean a skb before injecting it in
4306 * another namespace (@xnet == true). We have to clear all information in the
4307 * skb that could impact namespace isolation.
4308 */
4309void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4310{
4311 skb->tstamp.tv64 = 0;
4312 skb->pkt_type = PACKET_HOST;
4313 skb->skb_iif = 0;
4314 skb->ignore_df = 0;
4315 skb_dst_drop(skb);
4316 secpath_reset(skb);
4317 nf_reset(skb);
4318 nf_reset_trace(skb);
4319
4320 if (!xnet)
4321 return;
4322
4323 skb_orphan(skb);
4324 skb->mark = 0;
4325}
4326EXPORT_SYMBOL_GPL(skb_scrub_packet);
4327
4328/**
4329 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4330 *
4331 * @skb: GSO skb
4332 *
4333 * skb_gso_transport_seglen is used to determine the real size of the
4334 * individual segments, including Layer4 headers (TCP/UDP).
4335 *
4336 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4337 */
4338unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4339{
4340 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4341 unsigned int thlen = 0;
4342
4343 if (skb->encapsulation) {
4344 thlen = skb_inner_transport_header(skb) -
4345 skb_transport_header(skb);
4346
4347 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4348 thlen += inner_tcp_hdrlen(skb);
4349 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4350 thlen = tcp_hdrlen(skb);
4351 }
4352 /* UFO sets gso_size to the size of the fragmentation
4353 * payload, i.e. the size of the L4 (UDP) header is already
4354 * accounted for.
4355 */
4356 return thlen + shinfo->gso_size;
4357}
4358EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4359
4360static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4361{
4362 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4363 kfree_skb(skb);
4364 return NULL;
4365 }
4366
4367 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4368 2 * ETH_ALEN);
4369 skb->mac_header += VLAN_HLEN;
4370 return skb;
4371}
4372
4373struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4374{
4375 struct vlan_hdr *vhdr;
4376 u16 vlan_tci;
4377
4378 if (unlikely(skb_vlan_tag_present(skb))) {
4379 /* vlan_tci is already set-up so leave this for another time */
4380 return skb;
4381 }
4382
4383 skb = skb_share_check(skb, GFP_ATOMIC);
4384 if (unlikely(!skb))
4385 goto err_free;
4386
4387 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4388 goto err_free;
4389
4390 vhdr = (struct vlan_hdr *)skb->data;
4391 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4392 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4393
4394 skb_pull_rcsum(skb, VLAN_HLEN);
4395 vlan_set_encap_proto(skb, vhdr);
4396
4397 skb = skb_reorder_vlan_header(skb);
4398 if (unlikely(!skb))
4399 goto err_free;
4400
4401 skb_reset_network_header(skb);
4402 skb_reset_transport_header(skb);
4403 skb_reset_mac_len(skb);
4404
4405 return skb;
4406
4407err_free:
4408 kfree_skb(skb);
4409 return NULL;
4410}
4411EXPORT_SYMBOL(skb_vlan_untag);
4412
4413int skb_ensure_writable(struct sk_buff *skb, int write_len)
4414{
4415 if (!pskb_may_pull(skb, write_len))
4416 return -ENOMEM;
4417
4418 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4419 return 0;
4420
4421 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4422}
4423EXPORT_SYMBOL(skb_ensure_writable);
4424
4425/* remove VLAN header from packet and update csum accordingly. */
4426static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4427{
4428 struct vlan_hdr *vhdr;
4429 unsigned int offset = skb->data - skb_mac_header(skb);
4430 int err;
4431
4432 __skb_push(skb, offset);
4433 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4434 if (unlikely(err))
4435 goto pull;
4436
4437 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4438
4439 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4440 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4441
4442 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4443 __skb_pull(skb, VLAN_HLEN);
4444
4445 vlan_set_encap_proto(skb, vhdr);
4446 skb->mac_header += VLAN_HLEN;
4447
4448 if (skb_network_offset(skb) < ETH_HLEN)
4449 skb_set_network_header(skb, ETH_HLEN);
4450
4451 skb_reset_mac_len(skb);
4452pull:
4453 __skb_pull(skb, offset);
4454
4455 return err;
4456}
4457
4458int skb_vlan_pop(struct sk_buff *skb)
4459{
4460 u16 vlan_tci;
4461 __be16 vlan_proto;
4462 int err;
4463
4464 if (likely(skb_vlan_tag_present(skb))) {
4465 skb->vlan_tci = 0;
4466 } else {
4467 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4468 skb->protocol != htons(ETH_P_8021AD)) ||
4469 skb->len < VLAN_ETH_HLEN))
4470 return 0;
4471
4472 err = __skb_vlan_pop(skb, &vlan_tci);
4473 if (err)
4474 return err;
4475 }
4476 /* move next vlan tag to hw accel tag */
4477 if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4478 skb->protocol != htons(ETH_P_8021AD)) ||
4479 skb->len < VLAN_ETH_HLEN))
4480 return 0;
4481
4482 vlan_proto = skb->protocol;
4483 err = __skb_vlan_pop(skb, &vlan_tci);
4484 if (unlikely(err))
4485 return err;
4486
4487 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4488 return 0;
4489}
4490EXPORT_SYMBOL(skb_vlan_pop);
4491
4492int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4493{
4494 if (skb_vlan_tag_present(skb)) {
4495 unsigned int offset = skb->data - skb_mac_header(skb);
4496 int err;
4497
4498 /* __vlan_insert_tag expect skb->data pointing to mac header.
4499 * So change skb->data before calling it and change back to
4500 * original position later
4501 */
4502 __skb_push(skb, offset);
4503 err = __vlan_insert_tag(skb, skb->vlan_proto,
4504 skb_vlan_tag_get(skb));
4505 if (err) {
4506 __skb_pull(skb, offset);
4507 return err;
4508 }
4509
4510 skb->protocol = skb->vlan_proto;
4511 skb->mac_len += VLAN_HLEN;
4512
4513 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4514 __skb_pull(skb, offset);
4515 }
4516 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4517 return 0;
4518}
4519EXPORT_SYMBOL(skb_vlan_push);
4520
4521/**
4522 * alloc_skb_with_frags - allocate skb with page frags
4523 *
4524 * @header_len: size of linear part
4525 * @data_len: needed length in frags
4526 * @max_page_order: max page order desired.
4527 * @errcode: pointer to error code if any
4528 * @gfp_mask: allocation mask
4529 *
4530 * This can be used to allocate a paged skb, given a maximal order for frags.
4531 */
4532struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4533 unsigned long data_len,
4534 int max_page_order,
4535 int *errcode,
4536 gfp_t gfp_mask)
4537{
4538 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4539 unsigned long chunk;
4540 struct sk_buff *skb;
4541 struct page *page;
4542 gfp_t gfp_head;
4543 int i;
4544
4545 *errcode = -EMSGSIZE;
4546 /* Note this test could be relaxed, if we succeed to allocate
4547 * high order pages...
4548 */
4549 if (npages > MAX_SKB_FRAGS)
4550 return NULL;
4551
4552 gfp_head = gfp_mask;
4553 if (gfp_head & __GFP_DIRECT_RECLAIM)
4554 gfp_head |= __GFP_REPEAT;
4555
4556 *errcode = -ENOBUFS;
4557 skb = alloc_skb(header_len, gfp_head);
4558 if (!skb)
4559 return NULL;
4560
4561 skb->truesize += npages << PAGE_SHIFT;
4562
4563 for (i = 0; npages > 0; i++) {
4564 int order = max_page_order;
4565
4566 while (order) {
4567 if (npages >= 1 << order) {
4568 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4569 __GFP_COMP |
4570 __GFP_NOWARN |
4571 __GFP_NORETRY,
4572 order);
4573 if (page)
4574 goto fill_page;
4575 /* Do not retry other high order allocations */
4576 order = 1;
4577 max_page_order = 0;
4578 }
4579 order--;
4580 }
4581 page = alloc_page(gfp_mask);
4582 if (!page)
4583 goto failure;
4584fill_page:
4585 chunk = min_t(unsigned long, data_len,
4586 PAGE_SIZE << order);
4587 skb_fill_page_desc(skb, i, page, 0, chunk);
4588 data_len -= chunk;
4589 npages -= 1 << order;
4590 }
4591 return skb;
4592
4593failure:
4594 kfree_skb(skb);
4595 return NULL;
4596}
4597EXPORT_SYMBOL(alloc_skb_with_frags);
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/kmemcheck.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/tcp.h>
51#include <linux/udp.h>
52#include <linux/netdevice.h>
53#ifdef CONFIG_NET_CLS_ACT
54#include <net/pkt_sched.h>
55#endif
56#include <linux/string.h>
57#include <linux/skbuff.h>
58#include <linux/splice.h>
59#include <linux/cache.h>
60#include <linux/rtnetlink.h>
61#include <linux/init.h>
62#include <linux/scatterlist.h>
63#include <linux/errqueue.h>
64#include <linux/prefetch.h>
65
66#include <net/protocol.h>
67#include <net/dst.h>
68#include <net/sock.h>
69#include <net/checksum.h>
70#include <net/ip6_checksum.h>
71#include <net/xfrm.h>
72
73#include <asm/uaccess.h>
74#include <trace/events/skb.h>
75#include <linux/highmem.h>
76
77struct kmem_cache *skbuff_head_cache __read_mostly;
78static struct kmem_cache *skbuff_fclone_cache __read_mostly;
79
80/**
81 * skb_panic - private function for out-of-line support
82 * @skb: buffer
83 * @sz: size
84 * @addr: address
85 * @msg: skb_over_panic or skb_under_panic
86 *
87 * Out-of-line support for skb_put() and skb_push().
88 * Called via the wrapper skb_over_panic() or skb_under_panic().
89 * Keep out of line to prevent kernel bloat.
90 * __builtin_return_address is not used because it is not always reliable.
91 */
92static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
93 const char msg[])
94{
95 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
96 msg, addr, skb->len, sz, skb->head, skb->data,
97 (unsigned long)skb->tail, (unsigned long)skb->end,
98 skb->dev ? skb->dev->name : "<NULL>");
99 BUG();
100}
101
102static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
103{
104 skb_panic(skb, sz, addr, __func__);
105}
106
107static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
111
112/*
113 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
114 * the caller if emergency pfmemalloc reserves are being used. If it is and
115 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
116 * may be used. Otherwise, the packet data may be discarded until enough
117 * memory is free
118 */
119#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
120 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
121
122static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
123 unsigned long ip, bool *pfmemalloc)
124{
125 void *obj;
126 bool ret_pfmemalloc = false;
127
128 /*
129 * Try a regular allocation, when that fails and we're not entitled
130 * to the reserves, fail.
131 */
132 obj = kmalloc_node_track_caller(size,
133 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
134 node);
135 if (obj || !(gfp_pfmemalloc_allowed(flags)))
136 goto out;
137
138 /* Try again but now we are using pfmemalloc reserves */
139 ret_pfmemalloc = true;
140 obj = kmalloc_node_track_caller(size, flags, node);
141
142out:
143 if (pfmemalloc)
144 *pfmemalloc = ret_pfmemalloc;
145
146 return obj;
147}
148
149/* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
151 * [BEEP] leaks.
152 *
153 */
154
155struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
156{
157 struct sk_buff *skb;
158
159 /* Get the HEAD */
160 skb = kmem_cache_alloc_node(skbuff_head_cache,
161 gfp_mask & ~__GFP_DMA, node);
162 if (!skb)
163 goto out;
164
165 /*
166 * Only clear those fields we need to clear, not those that we will
167 * actually initialise below. Hence, don't put any more fields after
168 * the tail pointer in struct sk_buff!
169 */
170 memset(skb, 0, offsetof(struct sk_buff, tail));
171 skb->head = NULL;
172 skb->truesize = sizeof(struct sk_buff);
173 atomic_set(&skb->users, 1);
174
175 skb->mac_header = (typeof(skb->mac_header))~0U;
176out:
177 return skb;
178}
179
180/**
181 * __alloc_skb - allocate a network buffer
182 * @size: size to allocate
183 * @gfp_mask: allocation mask
184 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
185 * instead of head cache and allocate a cloned (child) skb.
186 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
187 * allocations in case the data is required for writeback
188 * @node: numa node to allocate memory on
189 *
190 * Allocate a new &sk_buff. The returned buffer has no headroom and a
191 * tail room of at least size bytes. The object has a reference count
192 * of one. The return is the buffer. On a failure the return is %NULL.
193 *
194 * Buffers may only be allocated from interrupts using a @gfp_mask of
195 * %GFP_ATOMIC.
196 */
197struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
198 int flags, int node)
199{
200 struct kmem_cache *cache;
201 struct skb_shared_info *shinfo;
202 struct sk_buff *skb;
203 u8 *data;
204 bool pfmemalloc;
205
206 cache = (flags & SKB_ALLOC_FCLONE)
207 ? skbuff_fclone_cache : skbuff_head_cache;
208
209 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
210 gfp_mask |= __GFP_MEMALLOC;
211
212 /* Get the HEAD */
213 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
214 if (!skb)
215 goto out;
216 prefetchw(skb);
217
218 /* We do our best to align skb_shared_info on a separate cache
219 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
220 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
221 * Both skb->head and skb_shared_info are cache line aligned.
222 */
223 size = SKB_DATA_ALIGN(size);
224 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
225 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
226 if (!data)
227 goto nodata;
228 /* kmalloc(size) might give us more room than requested.
229 * Put skb_shared_info exactly at the end of allocated zone,
230 * to allow max possible filling before reallocation.
231 */
232 size = SKB_WITH_OVERHEAD(ksize(data));
233 prefetchw(data + size);
234
235 /*
236 * Only clear those fields we need to clear, not those that we will
237 * actually initialise below. Hence, don't put any more fields after
238 * the tail pointer in struct sk_buff!
239 */
240 memset(skb, 0, offsetof(struct sk_buff, tail));
241 /* Account for allocated memory : skb + skb->head */
242 skb->truesize = SKB_TRUESIZE(size);
243 skb->pfmemalloc = pfmemalloc;
244 atomic_set(&skb->users, 1);
245 skb->head = data;
246 skb->data = data;
247 skb_reset_tail_pointer(skb);
248 skb->end = skb->tail + size;
249 skb->mac_header = (typeof(skb->mac_header))~0U;
250 skb->transport_header = (typeof(skb->transport_header))~0U;
251
252 /* make sure we initialize shinfo sequentially */
253 shinfo = skb_shinfo(skb);
254 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
255 atomic_set(&shinfo->dataref, 1);
256 kmemcheck_annotate_variable(shinfo->destructor_arg);
257
258 if (flags & SKB_ALLOC_FCLONE) {
259 struct sk_buff *child = skb + 1;
260 atomic_t *fclone_ref = (atomic_t *) (child + 1);
261
262 kmemcheck_annotate_bitfield(child, flags1);
263 kmemcheck_annotate_bitfield(child, flags2);
264 skb->fclone = SKB_FCLONE_ORIG;
265 atomic_set(fclone_ref, 1);
266
267 child->fclone = SKB_FCLONE_UNAVAILABLE;
268 child->pfmemalloc = pfmemalloc;
269 }
270out:
271 return skb;
272nodata:
273 kmem_cache_free(cache, skb);
274 skb = NULL;
275 goto out;
276}
277EXPORT_SYMBOL(__alloc_skb);
278
279/**
280 * build_skb - build a network buffer
281 * @data: data buffer provided by caller
282 * @frag_size: size of fragment, or 0 if head was kmalloced
283 *
284 * Allocate a new &sk_buff. Caller provides space holding head and
285 * skb_shared_info. @data must have been allocated by kmalloc() only if
286 * @frag_size is 0, otherwise data should come from the page allocator.
287 * The return is the new skb buffer.
288 * On a failure the return is %NULL, and @data is not freed.
289 * Notes :
290 * Before IO, driver allocates only data buffer where NIC put incoming frame
291 * Driver should add room at head (NET_SKB_PAD) and
292 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
293 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
294 * before giving packet to stack.
295 * RX rings only contains data buffers, not full skbs.
296 */
297struct sk_buff *build_skb(void *data, unsigned int frag_size)
298{
299 struct skb_shared_info *shinfo;
300 struct sk_buff *skb;
301 unsigned int size = frag_size ? : ksize(data);
302
303 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
304 if (!skb)
305 return NULL;
306
307 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
308
309 memset(skb, 0, offsetof(struct sk_buff, tail));
310 skb->truesize = SKB_TRUESIZE(size);
311 skb->head_frag = frag_size != 0;
312 atomic_set(&skb->users, 1);
313 skb->head = data;
314 skb->data = data;
315 skb_reset_tail_pointer(skb);
316 skb->end = skb->tail + size;
317 skb->mac_header = (typeof(skb->mac_header))~0U;
318 skb->transport_header = (typeof(skb->transport_header))~0U;
319
320 /* make sure we initialize shinfo sequentially */
321 shinfo = skb_shinfo(skb);
322 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
323 atomic_set(&shinfo->dataref, 1);
324 kmemcheck_annotate_variable(shinfo->destructor_arg);
325
326 return skb;
327}
328EXPORT_SYMBOL(build_skb);
329
330struct netdev_alloc_cache {
331 struct page_frag frag;
332 /* we maintain a pagecount bias, so that we dont dirty cache line
333 * containing page->_count every time we allocate a fragment.
334 */
335 unsigned int pagecnt_bias;
336};
337static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
338
339static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
340{
341 struct netdev_alloc_cache *nc;
342 void *data = NULL;
343 int order;
344 unsigned long flags;
345
346 local_irq_save(flags);
347 nc = &__get_cpu_var(netdev_alloc_cache);
348 if (unlikely(!nc->frag.page)) {
349refill:
350 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
351 gfp_t gfp = gfp_mask;
352
353 if (order)
354 gfp |= __GFP_COMP | __GFP_NOWARN;
355 nc->frag.page = alloc_pages(gfp, order);
356 if (likely(nc->frag.page))
357 break;
358 if (--order < 0)
359 goto end;
360 }
361 nc->frag.size = PAGE_SIZE << order;
362recycle:
363 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
364 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
365 nc->frag.offset = 0;
366 }
367
368 if (nc->frag.offset + fragsz > nc->frag.size) {
369 /* avoid unnecessary locked operations if possible */
370 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
371 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
372 goto recycle;
373 goto refill;
374 }
375
376 data = page_address(nc->frag.page) + nc->frag.offset;
377 nc->frag.offset += fragsz;
378 nc->pagecnt_bias--;
379end:
380 local_irq_restore(flags);
381 return data;
382}
383
384/**
385 * netdev_alloc_frag - allocate a page fragment
386 * @fragsz: fragment size
387 *
388 * Allocates a frag from a page for receive buffer.
389 * Uses GFP_ATOMIC allocations.
390 */
391void *netdev_alloc_frag(unsigned int fragsz)
392{
393 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
394}
395EXPORT_SYMBOL(netdev_alloc_frag);
396
397/**
398 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
399 * @dev: network device to receive on
400 * @length: length to allocate
401 * @gfp_mask: get_free_pages mask, passed to alloc_skb
402 *
403 * Allocate a new &sk_buff and assign it a usage count of one. The
404 * buffer has unspecified headroom built in. Users should allocate
405 * the headroom they think they need without accounting for the
406 * built in space. The built in space is used for optimisations.
407 *
408 * %NULL is returned if there is no free memory.
409 */
410struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
411 unsigned int length, gfp_t gfp_mask)
412{
413 struct sk_buff *skb = NULL;
414 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
415 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
416
417 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
418 void *data;
419
420 if (sk_memalloc_socks())
421 gfp_mask |= __GFP_MEMALLOC;
422
423 data = __netdev_alloc_frag(fragsz, gfp_mask);
424
425 if (likely(data)) {
426 skb = build_skb(data, fragsz);
427 if (unlikely(!skb))
428 put_page(virt_to_head_page(data));
429 }
430 } else {
431 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
432 SKB_ALLOC_RX, NUMA_NO_NODE);
433 }
434 if (likely(skb)) {
435 skb_reserve(skb, NET_SKB_PAD);
436 skb->dev = dev;
437 }
438 return skb;
439}
440EXPORT_SYMBOL(__netdev_alloc_skb);
441
442void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
443 int size, unsigned int truesize)
444{
445 skb_fill_page_desc(skb, i, page, off, size);
446 skb->len += size;
447 skb->data_len += size;
448 skb->truesize += truesize;
449}
450EXPORT_SYMBOL(skb_add_rx_frag);
451
452void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
453 unsigned int truesize)
454{
455 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
456
457 skb_frag_size_add(frag, size);
458 skb->len += size;
459 skb->data_len += size;
460 skb->truesize += truesize;
461}
462EXPORT_SYMBOL(skb_coalesce_rx_frag);
463
464static void skb_drop_list(struct sk_buff **listp)
465{
466 kfree_skb_list(*listp);
467 *listp = NULL;
468}
469
470static inline void skb_drop_fraglist(struct sk_buff *skb)
471{
472 skb_drop_list(&skb_shinfo(skb)->frag_list);
473}
474
475static void skb_clone_fraglist(struct sk_buff *skb)
476{
477 struct sk_buff *list;
478
479 skb_walk_frags(skb, list)
480 skb_get(list);
481}
482
483static void skb_free_head(struct sk_buff *skb)
484{
485 if (skb->head_frag)
486 put_page(virt_to_head_page(skb->head));
487 else
488 kfree(skb->head);
489}
490
491static void skb_release_data(struct sk_buff *skb)
492{
493 if (!skb->cloned ||
494 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
495 &skb_shinfo(skb)->dataref)) {
496 if (skb_shinfo(skb)->nr_frags) {
497 int i;
498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
499 skb_frag_unref(skb, i);
500 }
501
502 /*
503 * If skb buf is from userspace, we need to notify the caller
504 * the lower device DMA has done;
505 */
506 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
507 struct ubuf_info *uarg;
508
509 uarg = skb_shinfo(skb)->destructor_arg;
510 if (uarg->callback)
511 uarg->callback(uarg, true);
512 }
513
514 if (skb_has_frag_list(skb))
515 skb_drop_fraglist(skb);
516
517 skb_free_head(skb);
518 }
519}
520
521/*
522 * Free an skbuff by memory without cleaning the state.
523 */
524static void kfree_skbmem(struct sk_buff *skb)
525{
526 struct sk_buff *other;
527 atomic_t *fclone_ref;
528
529 switch (skb->fclone) {
530 case SKB_FCLONE_UNAVAILABLE:
531 kmem_cache_free(skbuff_head_cache, skb);
532 break;
533
534 case SKB_FCLONE_ORIG:
535 fclone_ref = (atomic_t *) (skb + 2);
536 if (atomic_dec_and_test(fclone_ref))
537 kmem_cache_free(skbuff_fclone_cache, skb);
538 break;
539
540 case SKB_FCLONE_CLONE:
541 fclone_ref = (atomic_t *) (skb + 1);
542 other = skb - 1;
543
544 /* The clone portion is available for
545 * fast-cloning again.
546 */
547 skb->fclone = SKB_FCLONE_UNAVAILABLE;
548
549 if (atomic_dec_and_test(fclone_ref))
550 kmem_cache_free(skbuff_fclone_cache, other);
551 break;
552 }
553}
554
555static void skb_release_head_state(struct sk_buff *skb)
556{
557 skb_dst_drop(skb);
558#ifdef CONFIG_XFRM
559 secpath_put(skb->sp);
560#endif
561 if (skb->destructor) {
562 WARN_ON(in_irq());
563 skb->destructor(skb);
564 }
565#if IS_ENABLED(CONFIG_NF_CONNTRACK)
566 nf_conntrack_put(skb->nfct);
567#endif
568#ifdef CONFIG_BRIDGE_NETFILTER
569 nf_bridge_put(skb->nf_bridge);
570#endif
571/* XXX: IS this still necessary? - JHS */
572#ifdef CONFIG_NET_SCHED
573 skb->tc_index = 0;
574#ifdef CONFIG_NET_CLS_ACT
575 skb->tc_verd = 0;
576#endif
577#endif
578}
579
580/* Free everything but the sk_buff shell. */
581static void skb_release_all(struct sk_buff *skb)
582{
583 skb_release_head_state(skb);
584 if (likely(skb->head))
585 skb_release_data(skb);
586}
587
588/**
589 * __kfree_skb - private function
590 * @skb: buffer
591 *
592 * Free an sk_buff. Release anything attached to the buffer.
593 * Clean the state. This is an internal helper function. Users should
594 * always call kfree_skb
595 */
596
597void __kfree_skb(struct sk_buff *skb)
598{
599 skb_release_all(skb);
600 kfree_skbmem(skb);
601}
602EXPORT_SYMBOL(__kfree_skb);
603
604/**
605 * kfree_skb - free an sk_buff
606 * @skb: buffer to free
607 *
608 * Drop a reference to the buffer and free it if the usage count has
609 * hit zero.
610 */
611void kfree_skb(struct sk_buff *skb)
612{
613 if (unlikely(!skb))
614 return;
615 if (likely(atomic_read(&skb->users) == 1))
616 smp_rmb();
617 else if (likely(!atomic_dec_and_test(&skb->users)))
618 return;
619 trace_kfree_skb(skb, __builtin_return_address(0));
620 __kfree_skb(skb);
621}
622EXPORT_SYMBOL(kfree_skb);
623
624void kfree_skb_list(struct sk_buff *segs)
625{
626 while (segs) {
627 struct sk_buff *next = segs->next;
628
629 kfree_skb(segs);
630 segs = next;
631 }
632}
633EXPORT_SYMBOL(kfree_skb_list);
634
635/**
636 * skb_tx_error - report an sk_buff xmit error
637 * @skb: buffer that triggered an error
638 *
639 * Report xmit error if a device callback is tracking this skb.
640 * skb must be freed afterwards.
641 */
642void skb_tx_error(struct sk_buff *skb)
643{
644 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
645 struct ubuf_info *uarg;
646
647 uarg = skb_shinfo(skb)->destructor_arg;
648 if (uarg->callback)
649 uarg->callback(uarg, false);
650 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
651 }
652}
653EXPORT_SYMBOL(skb_tx_error);
654
655/**
656 * consume_skb - free an skbuff
657 * @skb: buffer to free
658 *
659 * Drop a ref to the buffer and free it if the usage count has hit zero
660 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
661 * is being dropped after a failure and notes that
662 */
663void consume_skb(struct sk_buff *skb)
664{
665 if (unlikely(!skb))
666 return;
667 if (likely(atomic_read(&skb->users) == 1))
668 smp_rmb();
669 else if (likely(!atomic_dec_and_test(&skb->users)))
670 return;
671 trace_consume_skb(skb);
672 __kfree_skb(skb);
673}
674EXPORT_SYMBOL(consume_skb);
675
676static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
677{
678 new->tstamp = old->tstamp;
679 new->dev = old->dev;
680 new->transport_header = old->transport_header;
681 new->network_header = old->network_header;
682 new->mac_header = old->mac_header;
683 new->inner_protocol = old->inner_protocol;
684 new->inner_transport_header = old->inner_transport_header;
685 new->inner_network_header = old->inner_network_header;
686 new->inner_mac_header = old->inner_mac_header;
687 skb_dst_copy(new, old);
688 skb_copy_hash(new, old);
689 new->ooo_okay = old->ooo_okay;
690 new->no_fcs = old->no_fcs;
691 new->encapsulation = old->encapsulation;
692#ifdef CONFIG_XFRM
693 new->sp = secpath_get(old->sp);
694#endif
695 memcpy(new->cb, old->cb, sizeof(old->cb));
696 new->csum = old->csum;
697 new->local_df = old->local_df;
698 new->pkt_type = old->pkt_type;
699 new->ip_summed = old->ip_summed;
700 skb_copy_queue_mapping(new, old);
701 new->priority = old->priority;
702#if IS_ENABLED(CONFIG_IP_VS)
703 new->ipvs_property = old->ipvs_property;
704#endif
705 new->pfmemalloc = old->pfmemalloc;
706 new->protocol = old->protocol;
707 new->mark = old->mark;
708 new->skb_iif = old->skb_iif;
709 __nf_copy(new, old);
710#ifdef CONFIG_NET_SCHED
711 new->tc_index = old->tc_index;
712#ifdef CONFIG_NET_CLS_ACT
713 new->tc_verd = old->tc_verd;
714#endif
715#endif
716 new->vlan_proto = old->vlan_proto;
717 new->vlan_tci = old->vlan_tci;
718
719 skb_copy_secmark(new, old);
720
721#ifdef CONFIG_NET_RX_BUSY_POLL
722 new->napi_id = old->napi_id;
723#endif
724}
725
726/*
727 * You should not add any new code to this function. Add it to
728 * __copy_skb_header above instead.
729 */
730static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
731{
732#define C(x) n->x = skb->x
733
734 n->next = n->prev = NULL;
735 n->sk = NULL;
736 __copy_skb_header(n, skb);
737
738 C(len);
739 C(data_len);
740 C(mac_len);
741 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
742 n->cloned = 1;
743 n->nohdr = 0;
744 n->destructor = NULL;
745 C(tail);
746 C(end);
747 C(head);
748 C(head_frag);
749 C(data);
750 C(truesize);
751 atomic_set(&n->users, 1);
752
753 atomic_inc(&(skb_shinfo(skb)->dataref));
754 skb->cloned = 1;
755
756 return n;
757#undef C
758}
759
760/**
761 * skb_morph - morph one skb into another
762 * @dst: the skb to receive the contents
763 * @src: the skb to supply the contents
764 *
765 * This is identical to skb_clone except that the target skb is
766 * supplied by the user.
767 *
768 * The target skb is returned upon exit.
769 */
770struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
771{
772 skb_release_all(dst);
773 return __skb_clone(dst, src);
774}
775EXPORT_SYMBOL_GPL(skb_morph);
776
777/**
778 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
779 * @skb: the skb to modify
780 * @gfp_mask: allocation priority
781 *
782 * This must be called on SKBTX_DEV_ZEROCOPY skb.
783 * It will copy all frags into kernel and drop the reference
784 * to userspace pages.
785 *
786 * If this function is called from an interrupt gfp_mask() must be
787 * %GFP_ATOMIC.
788 *
789 * Returns 0 on success or a negative error code on failure
790 * to allocate kernel memory to copy to.
791 */
792int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
793{
794 int i;
795 int num_frags = skb_shinfo(skb)->nr_frags;
796 struct page *page, *head = NULL;
797 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
798
799 for (i = 0; i < num_frags; i++) {
800 u8 *vaddr;
801 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
802
803 page = alloc_page(gfp_mask);
804 if (!page) {
805 while (head) {
806 struct page *next = (struct page *)page_private(head);
807 put_page(head);
808 head = next;
809 }
810 return -ENOMEM;
811 }
812 vaddr = kmap_atomic(skb_frag_page(f));
813 memcpy(page_address(page),
814 vaddr + f->page_offset, skb_frag_size(f));
815 kunmap_atomic(vaddr);
816 set_page_private(page, (unsigned long)head);
817 head = page;
818 }
819
820 /* skb frags release userspace buffers */
821 for (i = 0; i < num_frags; i++)
822 skb_frag_unref(skb, i);
823
824 uarg->callback(uarg, false);
825
826 /* skb frags point to kernel buffers */
827 for (i = num_frags - 1; i >= 0; i--) {
828 __skb_fill_page_desc(skb, i, head, 0,
829 skb_shinfo(skb)->frags[i].size);
830 head = (struct page *)page_private(head);
831 }
832
833 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
834 return 0;
835}
836EXPORT_SYMBOL_GPL(skb_copy_ubufs);
837
838/**
839 * skb_clone - duplicate an sk_buff
840 * @skb: buffer to clone
841 * @gfp_mask: allocation priority
842 *
843 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
844 * copies share the same packet data but not structure. The new
845 * buffer has a reference count of 1. If the allocation fails the
846 * function returns %NULL otherwise the new buffer is returned.
847 *
848 * If this function is called from an interrupt gfp_mask() must be
849 * %GFP_ATOMIC.
850 */
851
852struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
853{
854 struct sk_buff *n;
855
856 if (skb_orphan_frags(skb, gfp_mask))
857 return NULL;
858
859 n = skb + 1;
860 if (skb->fclone == SKB_FCLONE_ORIG &&
861 n->fclone == SKB_FCLONE_UNAVAILABLE) {
862 atomic_t *fclone_ref = (atomic_t *) (n + 1);
863 n->fclone = SKB_FCLONE_CLONE;
864 atomic_inc(fclone_ref);
865 } else {
866 if (skb_pfmemalloc(skb))
867 gfp_mask |= __GFP_MEMALLOC;
868
869 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
870 if (!n)
871 return NULL;
872
873 kmemcheck_annotate_bitfield(n, flags1);
874 kmemcheck_annotate_bitfield(n, flags2);
875 n->fclone = SKB_FCLONE_UNAVAILABLE;
876 }
877
878 return __skb_clone(n, skb);
879}
880EXPORT_SYMBOL(skb_clone);
881
882static void skb_headers_offset_update(struct sk_buff *skb, int off)
883{
884 /* Only adjust this if it actually is csum_start rather than csum */
885 if (skb->ip_summed == CHECKSUM_PARTIAL)
886 skb->csum_start += off;
887 /* {transport,network,mac}_header and tail are relative to skb->head */
888 skb->transport_header += off;
889 skb->network_header += off;
890 if (skb_mac_header_was_set(skb))
891 skb->mac_header += off;
892 skb->inner_transport_header += off;
893 skb->inner_network_header += off;
894 skb->inner_mac_header += off;
895}
896
897static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
898{
899 __copy_skb_header(new, old);
900
901 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
902 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
903 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
904}
905
906static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
907{
908 if (skb_pfmemalloc(skb))
909 return SKB_ALLOC_RX;
910 return 0;
911}
912
913/**
914 * skb_copy - create private copy of an sk_buff
915 * @skb: buffer to copy
916 * @gfp_mask: allocation priority
917 *
918 * Make a copy of both an &sk_buff and its data. This is used when the
919 * caller wishes to modify the data and needs a private copy of the
920 * data to alter. Returns %NULL on failure or the pointer to the buffer
921 * on success. The returned buffer has a reference count of 1.
922 *
923 * As by-product this function converts non-linear &sk_buff to linear
924 * one, so that &sk_buff becomes completely private and caller is allowed
925 * to modify all the data of returned buffer. This means that this
926 * function is not recommended for use in circumstances when only
927 * header is going to be modified. Use pskb_copy() instead.
928 */
929
930struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
931{
932 int headerlen = skb_headroom(skb);
933 unsigned int size = skb_end_offset(skb) + skb->data_len;
934 struct sk_buff *n = __alloc_skb(size, gfp_mask,
935 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
936
937 if (!n)
938 return NULL;
939
940 /* Set the data pointer */
941 skb_reserve(n, headerlen);
942 /* Set the tail pointer and length */
943 skb_put(n, skb->len);
944
945 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
946 BUG();
947
948 copy_skb_header(n, skb);
949 return n;
950}
951EXPORT_SYMBOL(skb_copy);
952
953/**
954 * __pskb_copy - create copy of an sk_buff with private head.
955 * @skb: buffer to copy
956 * @headroom: headroom of new skb
957 * @gfp_mask: allocation priority
958 *
959 * Make a copy of both an &sk_buff and part of its data, located
960 * in header. Fragmented data remain shared. This is used when
961 * the caller wishes to modify only header of &sk_buff and needs
962 * private copy of the header to alter. Returns %NULL on failure
963 * or the pointer to the buffer on success.
964 * The returned buffer has a reference count of 1.
965 */
966
967struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
968{
969 unsigned int size = skb_headlen(skb) + headroom;
970 struct sk_buff *n = __alloc_skb(size, gfp_mask,
971 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
972
973 if (!n)
974 goto out;
975
976 /* Set the data pointer */
977 skb_reserve(n, headroom);
978 /* Set the tail pointer and length */
979 skb_put(n, skb_headlen(skb));
980 /* Copy the bytes */
981 skb_copy_from_linear_data(skb, n->data, n->len);
982
983 n->truesize += skb->data_len;
984 n->data_len = skb->data_len;
985 n->len = skb->len;
986
987 if (skb_shinfo(skb)->nr_frags) {
988 int i;
989
990 if (skb_orphan_frags(skb, gfp_mask)) {
991 kfree_skb(n);
992 n = NULL;
993 goto out;
994 }
995 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
996 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
997 skb_frag_ref(skb, i);
998 }
999 skb_shinfo(n)->nr_frags = i;
1000 }
1001
1002 if (skb_has_frag_list(skb)) {
1003 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1004 skb_clone_fraglist(n);
1005 }
1006
1007 copy_skb_header(n, skb);
1008out:
1009 return n;
1010}
1011EXPORT_SYMBOL(__pskb_copy);
1012
1013/**
1014 * pskb_expand_head - reallocate header of &sk_buff
1015 * @skb: buffer to reallocate
1016 * @nhead: room to add at head
1017 * @ntail: room to add at tail
1018 * @gfp_mask: allocation priority
1019 *
1020 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1021 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1022 * reference count of 1. Returns zero in the case of success or error,
1023 * if expansion failed. In the last case, &sk_buff is not changed.
1024 *
1025 * All the pointers pointing into skb header may change and must be
1026 * reloaded after call to this function.
1027 */
1028
1029int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1030 gfp_t gfp_mask)
1031{
1032 int i;
1033 u8 *data;
1034 int size = nhead + skb_end_offset(skb) + ntail;
1035 long off;
1036
1037 BUG_ON(nhead < 0);
1038
1039 if (skb_shared(skb))
1040 BUG();
1041
1042 size = SKB_DATA_ALIGN(size);
1043
1044 if (skb_pfmemalloc(skb))
1045 gfp_mask |= __GFP_MEMALLOC;
1046 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1047 gfp_mask, NUMA_NO_NODE, NULL);
1048 if (!data)
1049 goto nodata;
1050 size = SKB_WITH_OVERHEAD(ksize(data));
1051
1052 /* Copy only real data... and, alas, header. This should be
1053 * optimized for the cases when header is void.
1054 */
1055 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1056
1057 memcpy((struct skb_shared_info *)(data + size),
1058 skb_shinfo(skb),
1059 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1060
1061 /*
1062 * if shinfo is shared we must drop the old head gracefully, but if it
1063 * is not we can just drop the old head and let the existing refcount
1064 * be since all we did is relocate the values
1065 */
1066 if (skb_cloned(skb)) {
1067 /* copy this zero copy skb frags */
1068 if (skb_orphan_frags(skb, gfp_mask))
1069 goto nofrags;
1070 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1071 skb_frag_ref(skb, i);
1072
1073 if (skb_has_frag_list(skb))
1074 skb_clone_fraglist(skb);
1075
1076 skb_release_data(skb);
1077 } else {
1078 skb_free_head(skb);
1079 }
1080 off = (data + nhead) - skb->head;
1081
1082 skb->head = data;
1083 skb->head_frag = 0;
1084 skb->data += off;
1085#ifdef NET_SKBUFF_DATA_USES_OFFSET
1086 skb->end = size;
1087 off = nhead;
1088#else
1089 skb->end = skb->head + size;
1090#endif
1091 skb->tail += off;
1092 skb_headers_offset_update(skb, nhead);
1093 skb->cloned = 0;
1094 skb->hdr_len = 0;
1095 skb->nohdr = 0;
1096 atomic_set(&skb_shinfo(skb)->dataref, 1);
1097 return 0;
1098
1099nofrags:
1100 kfree(data);
1101nodata:
1102 return -ENOMEM;
1103}
1104EXPORT_SYMBOL(pskb_expand_head);
1105
1106/* Make private copy of skb with writable head and some headroom */
1107
1108struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1109{
1110 struct sk_buff *skb2;
1111 int delta = headroom - skb_headroom(skb);
1112
1113 if (delta <= 0)
1114 skb2 = pskb_copy(skb, GFP_ATOMIC);
1115 else {
1116 skb2 = skb_clone(skb, GFP_ATOMIC);
1117 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1118 GFP_ATOMIC)) {
1119 kfree_skb(skb2);
1120 skb2 = NULL;
1121 }
1122 }
1123 return skb2;
1124}
1125EXPORT_SYMBOL(skb_realloc_headroom);
1126
1127/**
1128 * skb_copy_expand - copy and expand sk_buff
1129 * @skb: buffer to copy
1130 * @newheadroom: new free bytes at head
1131 * @newtailroom: new free bytes at tail
1132 * @gfp_mask: allocation priority
1133 *
1134 * Make a copy of both an &sk_buff and its data and while doing so
1135 * allocate additional space.
1136 *
1137 * This is used when the caller wishes to modify the data and needs a
1138 * private copy of the data to alter as well as more space for new fields.
1139 * Returns %NULL on failure or the pointer to the buffer
1140 * on success. The returned buffer has a reference count of 1.
1141 *
1142 * You must pass %GFP_ATOMIC as the allocation priority if this function
1143 * is called from an interrupt.
1144 */
1145struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1146 int newheadroom, int newtailroom,
1147 gfp_t gfp_mask)
1148{
1149 /*
1150 * Allocate the copy buffer
1151 */
1152 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1153 gfp_mask, skb_alloc_rx_flag(skb),
1154 NUMA_NO_NODE);
1155 int oldheadroom = skb_headroom(skb);
1156 int head_copy_len, head_copy_off;
1157
1158 if (!n)
1159 return NULL;
1160
1161 skb_reserve(n, newheadroom);
1162
1163 /* Set the tail pointer and length */
1164 skb_put(n, skb->len);
1165
1166 head_copy_len = oldheadroom;
1167 head_copy_off = 0;
1168 if (newheadroom <= head_copy_len)
1169 head_copy_len = newheadroom;
1170 else
1171 head_copy_off = newheadroom - head_copy_len;
1172
1173 /* Copy the linear header and data. */
1174 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1175 skb->len + head_copy_len))
1176 BUG();
1177
1178 copy_skb_header(n, skb);
1179
1180 skb_headers_offset_update(n, newheadroom - oldheadroom);
1181
1182 return n;
1183}
1184EXPORT_SYMBOL(skb_copy_expand);
1185
1186/**
1187 * skb_pad - zero pad the tail of an skb
1188 * @skb: buffer to pad
1189 * @pad: space to pad
1190 *
1191 * Ensure that a buffer is followed by a padding area that is zero
1192 * filled. Used by network drivers which may DMA or transfer data
1193 * beyond the buffer end onto the wire.
1194 *
1195 * May return error in out of memory cases. The skb is freed on error.
1196 */
1197
1198int skb_pad(struct sk_buff *skb, int pad)
1199{
1200 int err;
1201 int ntail;
1202
1203 /* If the skbuff is non linear tailroom is always zero.. */
1204 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1205 memset(skb->data+skb->len, 0, pad);
1206 return 0;
1207 }
1208
1209 ntail = skb->data_len + pad - (skb->end - skb->tail);
1210 if (likely(skb_cloned(skb) || ntail > 0)) {
1211 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1212 if (unlikely(err))
1213 goto free_skb;
1214 }
1215
1216 /* FIXME: The use of this function with non-linear skb's really needs
1217 * to be audited.
1218 */
1219 err = skb_linearize(skb);
1220 if (unlikely(err))
1221 goto free_skb;
1222
1223 memset(skb->data + skb->len, 0, pad);
1224 return 0;
1225
1226free_skb:
1227 kfree_skb(skb);
1228 return err;
1229}
1230EXPORT_SYMBOL(skb_pad);
1231
1232/**
1233 * pskb_put - add data to the tail of a potentially fragmented buffer
1234 * @skb: start of the buffer to use
1235 * @tail: tail fragment of the buffer to use
1236 * @len: amount of data to add
1237 *
1238 * This function extends the used data area of the potentially
1239 * fragmented buffer. @tail must be the last fragment of @skb -- or
1240 * @skb itself. If this would exceed the total buffer size the kernel
1241 * will panic. A pointer to the first byte of the extra data is
1242 * returned.
1243 */
1244
1245unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1246{
1247 if (tail != skb) {
1248 skb->data_len += len;
1249 skb->len += len;
1250 }
1251 return skb_put(tail, len);
1252}
1253EXPORT_SYMBOL_GPL(pskb_put);
1254
1255/**
1256 * skb_put - add data to a buffer
1257 * @skb: buffer to use
1258 * @len: amount of data to add
1259 *
1260 * This function extends the used data area of the buffer. If this would
1261 * exceed the total buffer size the kernel will panic. A pointer to the
1262 * first byte of the extra data is returned.
1263 */
1264unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1265{
1266 unsigned char *tmp = skb_tail_pointer(skb);
1267 SKB_LINEAR_ASSERT(skb);
1268 skb->tail += len;
1269 skb->len += len;
1270 if (unlikely(skb->tail > skb->end))
1271 skb_over_panic(skb, len, __builtin_return_address(0));
1272 return tmp;
1273}
1274EXPORT_SYMBOL(skb_put);
1275
1276/**
1277 * skb_push - add data to the start of a buffer
1278 * @skb: buffer to use
1279 * @len: amount of data to add
1280 *
1281 * This function extends the used data area of the buffer at the buffer
1282 * start. If this would exceed the total buffer headroom the kernel will
1283 * panic. A pointer to the first byte of the extra data is returned.
1284 */
1285unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1286{
1287 skb->data -= len;
1288 skb->len += len;
1289 if (unlikely(skb->data<skb->head))
1290 skb_under_panic(skb, len, __builtin_return_address(0));
1291 return skb->data;
1292}
1293EXPORT_SYMBOL(skb_push);
1294
1295/**
1296 * skb_pull - remove data from the start of a buffer
1297 * @skb: buffer to use
1298 * @len: amount of data to remove
1299 *
1300 * This function removes data from the start of a buffer, returning
1301 * the memory to the headroom. A pointer to the next data in the buffer
1302 * is returned. Once the data has been pulled future pushes will overwrite
1303 * the old data.
1304 */
1305unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1306{
1307 return skb_pull_inline(skb, len);
1308}
1309EXPORT_SYMBOL(skb_pull);
1310
1311/**
1312 * skb_trim - remove end from a buffer
1313 * @skb: buffer to alter
1314 * @len: new length
1315 *
1316 * Cut the length of a buffer down by removing data from the tail. If
1317 * the buffer is already under the length specified it is not modified.
1318 * The skb must be linear.
1319 */
1320void skb_trim(struct sk_buff *skb, unsigned int len)
1321{
1322 if (skb->len > len)
1323 __skb_trim(skb, len);
1324}
1325EXPORT_SYMBOL(skb_trim);
1326
1327/* Trims skb to length len. It can change skb pointers.
1328 */
1329
1330int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1331{
1332 struct sk_buff **fragp;
1333 struct sk_buff *frag;
1334 int offset = skb_headlen(skb);
1335 int nfrags = skb_shinfo(skb)->nr_frags;
1336 int i;
1337 int err;
1338
1339 if (skb_cloned(skb) &&
1340 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1341 return err;
1342
1343 i = 0;
1344 if (offset >= len)
1345 goto drop_pages;
1346
1347 for (; i < nfrags; i++) {
1348 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1349
1350 if (end < len) {
1351 offset = end;
1352 continue;
1353 }
1354
1355 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1356
1357drop_pages:
1358 skb_shinfo(skb)->nr_frags = i;
1359
1360 for (; i < nfrags; i++)
1361 skb_frag_unref(skb, i);
1362
1363 if (skb_has_frag_list(skb))
1364 skb_drop_fraglist(skb);
1365 goto done;
1366 }
1367
1368 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1369 fragp = &frag->next) {
1370 int end = offset + frag->len;
1371
1372 if (skb_shared(frag)) {
1373 struct sk_buff *nfrag;
1374
1375 nfrag = skb_clone(frag, GFP_ATOMIC);
1376 if (unlikely(!nfrag))
1377 return -ENOMEM;
1378
1379 nfrag->next = frag->next;
1380 consume_skb(frag);
1381 frag = nfrag;
1382 *fragp = frag;
1383 }
1384
1385 if (end < len) {
1386 offset = end;
1387 continue;
1388 }
1389
1390 if (end > len &&
1391 unlikely((err = pskb_trim(frag, len - offset))))
1392 return err;
1393
1394 if (frag->next)
1395 skb_drop_list(&frag->next);
1396 break;
1397 }
1398
1399done:
1400 if (len > skb_headlen(skb)) {
1401 skb->data_len -= skb->len - len;
1402 skb->len = len;
1403 } else {
1404 skb->len = len;
1405 skb->data_len = 0;
1406 skb_set_tail_pointer(skb, len);
1407 }
1408
1409 return 0;
1410}
1411EXPORT_SYMBOL(___pskb_trim);
1412
1413/**
1414 * __pskb_pull_tail - advance tail of skb header
1415 * @skb: buffer to reallocate
1416 * @delta: number of bytes to advance tail
1417 *
1418 * The function makes a sense only on a fragmented &sk_buff,
1419 * it expands header moving its tail forward and copying necessary
1420 * data from fragmented part.
1421 *
1422 * &sk_buff MUST have reference count of 1.
1423 *
1424 * Returns %NULL (and &sk_buff does not change) if pull failed
1425 * or value of new tail of skb in the case of success.
1426 *
1427 * All the pointers pointing into skb header may change and must be
1428 * reloaded after call to this function.
1429 */
1430
1431/* Moves tail of skb head forward, copying data from fragmented part,
1432 * when it is necessary.
1433 * 1. It may fail due to malloc failure.
1434 * 2. It may change skb pointers.
1435 *
1436 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1437 */
1438unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1439{
1440 /* If skb has not enough free space at tail, get new one
1441 * plus 128 bytes for future expansions. If we have enough
1442 * room at tail, reallocate without expansion only if skb is cloned.
1443 */
1444 int i, k, eat = (skb->tail + delta) - skb->end;
1445
1446 if (eat > 0 || skb_cloned(skb)) {
1447 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1448 GFP_ATOMIC))
1449 return NULL;
1450 }
1451
1452 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1453 BUG();
1454
1455 /* Optimization: no fragments, no reasons to preestimate
1456 * size of pulled pages. Superb.
1457 */
1458 if (!skb_has_frag_list(skb))
1459 goto pull_pages;
1460
1461 /* Estimate size of pulled pages. */
1462 eat = delta;
1463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1464 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1465
1466 if (size >= eat)
1467 goto pull_pages;
1468 eat -= size;
1469 }
1470
1471 /* If we need update frag list, we are in troubles.
1472 * Certainly, it possible to add an offset to skb data,
1473 * but taking into account that pulling is expected to
1474 * be very rare operation, it is worth to fight against
1475 * further bloating skb head and crucify ourselves here instead.
1476 * Pure masohism, indeed. 8)8)
1477 */
1478 if (eat) {
1479 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1480 struct sk_buff *clone = NULL;
1481 struct sk_buff *insp = NULL;
1482
1483 do {
1484 BUG_ON(!list);
1485
1486 if (list->len <= eat) {
1487 /* Eaten as whole. */
1488 eat -= list->len;
1489 list = list->next;
1490 insp = list;
1491 } else {
1492 /* Eaten partially. */
1493
1494 if (skb_shared(list)) {
1495 /* Sucks! We need to fork list. :-( */
1496 clone = skb_clone(list, GFP_ATOMIC);
1497 if (!clone)
1498 return NULL;
1499 insp = list->next;
1500 list = clone;
1501 } else {
1502 /* This may be pulled without
1503 * problems. */
1504 insp = list;
1505 }
1506 if (!pskb_pull(list, eat)) {
1507 kfree_skb(clone);
1508 return NULL;
1509 }
1510 break;
1511 }
1512 } while (eat);
1513
1514 /* Free pulled out fragments. */
1515 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1516 skb_shinfo(skb)->frag_list = list->next;
1517 kfree_skb(list);
1518 }
1519 /* And insert new clone at head. */
1520 if (clone) {
1521 clone->next = list;
1522 skb_shinfo(skb)->frag_list = clone;
1523 }
1524 }
1525 /* Success! Now we may commit changes to skb data. */
1526
1527pull_pages:
1528 eat = delta;
1529 k = 0;
1530 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1531 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1532
1533 if (size <= eat) {
1534 skb_frag_unref(skb, i);
1535 eat -= size;
1536 } else {
1537 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1538 if (eat) {
1539 skb_shinfo(skb)->frags[k].page_offset += eat;
1540 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1541 eat = 0;
1542 }
1543 k++;
1544 }
1545 }
1546 skb_shinfo(skb)->nr_frags = k;
1547
1548 skb->tail += delta;
1549 skb->data_len -= delta;
1550
1551 return skb_tail_pointer(skb);
1552}
1553EXPORT_SYMBOL(__pskb_pull_tail);
1554
1555/**
1556 * skb_copy_bits - copy bits from skb to kernel buffer
1557 * @skb: source skb
1558 * @offset: offset in source
1559 * @to: destination buffer
1560 * @len: number of bytes to copy
1561 *
1562 * Copy the specified number of bytes from the source skb to the
1563 * destination buffer.
1564 *
1565 * CAUTION ! :
1566 * If its prototype is ever changed,
1567 * check arch/{*}/net/{*}.S files,
1568 * since it is called from BPF assembly code.
1569 */
1570int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1571{
1572 int start = skb_headlen(skb);
1573 struct sk_buff *frag_iter;
1574 int i, copy;
1575
1576 if (offset > (int)skb->len - len)
1577 goto fault;
1578
1579 /* Copy header. */
1580 if ((copy = start - offset) > 0) {
1581 if (copy > len)
1582 copy = len;
1583 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1584 if ((len -= copy) == 0)
1585 return 0;
1586 offset += copy;
1587 to += copy;
1588 }
1589
1590 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1591 int end;
1592 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1593
1594 WARN_ON(start > offset + len);
1595
1596 end = start + skb_frag_size(f);
1597 if ((copy = end - offset) > 0) {
1598 u8 *vaddr;
1599
1600 if (copy > len)
1601 copy = len;
1602
1603 vaddr = kmap_atomic(skb_frag_page(f));
1604 memcpy(to,
1605 vaddr + f->page_offset + offset - start,
1606 copy);
1607 kunmap_atomic(vaddr);
1608
1609 if ((len -= copy) == 0)
1610 return 0;
1611 offset += copy;
1612 to += copy;
1613 }
1614 start = end;
1615 }
1616
1617 skb_walk_frags(skb, frag_iter) {
1618 int end;
1619
1620 WARN_ON(start > offset + len);
1621
1622 end = start + frag_iter->len;
1623 if ((copy = end - offset) > 0) {
1624 if (copy > len)
1625 copy = len;
1626 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1627 goto fault;
1628 if ((len -= copy) == 0)
1629 return 0;
1630 offset += copy;
1631 to += copy;
1632 }
1633 start = end;
1634 }
1635
1636 if (!len)
1637 return 0;
1638
1639fault:
1640 return -EFAULT;
1641}
1642EXPORT_SYMBOL(skb_copy_bits);
1643
1644/*
1645 * Callback from splice_to_pipe(), if we need to release some pages
1646 * at the end of the spd in case we error'ed out in filling the pipe.
1647 */
1648static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1649{
1650 put_page(spd->pages[i]);
1651}
1652
1653static struct page *linear_to_page(struct page *page, unsigned int *len,
1654 unsigned int *offset,
1655 struct sock *sk)
1656{
1657 struct page_frag *pfrag = sk_page_frag(sk);
1658
1659 if (!sk_page_frag_refill(sk, pfrag))
1660 return NULL;
1661
1662 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1663
1664 memcpy(page_address(pfrag->page) + pfrag->offset,
1665 page_address(page) + *offset, *len);
1666 *offset = pfrag->offset;
1667 pfrag->offset += *len;
1668
1669 return pfrag->page;
1670}
1671
1672static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1673 struct page *page,
1674 unsigned int offset)
1675{
1676 return spd->nr_pages &&
1677 spd->pages[spd->nr_pages - 1] == page &&
1678 (spd->partial[spd->nr_pages - 1].offset +
1679 spd->partial[spd->nr_pages - 1].len == offset);
1680}
1681
1682/*
1683 * Fill page/offset/length into spd, if it can hold more pages.
1684 */
1685static bool spd_fill_page(struct splice_pipe_desc *spd,
1686 struct pipe_inode_info *pipe, struct page *page,
1687 unsigned int *len, unsigned int offset,
1688 bool linear,
1689 struct sock *sk)
1690{
1691 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1692 return true;
1693
1694 if (linear) {
1695 page = linear_to_page(page, len, &offset, sk);
1696 if (!page)
1697 return true;
1698 }
1699 if (spd_can_coalesce(spd, page, offset)) {
1700 spd->partial[spd->nr_pages - 1].len += *len;
1701 return false;
1702 }
1703 get_page(page);
1704 spd->pages[spd->nr_pages] = page;
1705 spd->partial[spd->nr_pages].len = *len;
1706 spd->partial[spd->nr_pages].offset = offset;
1707 spd->nr_pages++;
1708
1709 return false;
1710}
1711
1712static bool __splice_segment(struct page *page, unsigned int poff,
1713 unsigned int plen, unsigned int *off,
1714 unsigned int *len,
1715 struct splice_pipe_desc *spd, bool linear,
1716 struct sock *sk,
1717 struct pipe_inode_info *pipe)
1718{
1719 if (!*len)
1720 return true;
1721
1722 /* skip this segment if already processed */
1723 if (*off >= plen) {
1724 *off -= plen;
1725 return false;
1726 }
1727
1728 /* ignore any bits we already processed */
1729 poff += *off;
1730 plen -= *off;
1731 *off = 0;
1732
1733 do {
1734 unsigned int flen = min(*len, plen);
1735
1736 if (spd_fill_page(spd, pipe, page, &flen, poff,
1737 linear, sk))
1738 return true;
1739 poff += flen;
1740 plen -= flen;
1741 *len -= flen;
1742 } while (*len && plen);
1743
1744 return false;
1745}
1746
1747/*
1748 * Map linear and fragment data from the skb to spd. It reports true if the
1749 * pipe is full or if we already spliced the requested length.
1750 */
1751static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1752 unsigned int *offset, unsigned int *len,
1753 struct splice_pipe_desc *spd, struct sock *sk)
1754{
1755 int seg;
1756
1757 /* map the linear part :
1758 * If skb->head_frag is set, this 'linear' part is backed by a
1759 * fragment, and if the head is not shared with any clones then
1760 * we can avoid a copy since we own the head portion of this page.
1761 */
1762 if (__splice_segment(virt_to_page(skb->data),
1763 (unsigned long) skb->data & (PAGE_SIZE - 1),
1764 skb_headlen(skb),
1765 offset, len, spd,
1766 skb_head_is_locked(skb),
1767 sk, pipe))
1768 return true;
1769
1770 /*
1771 * then map the fragments
1772 */
1773 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1774 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1775
1776 if (__splice_segment(skb_frag_page(f),
1777 f->page_offset, skb_frag_size(f),
1778 offset, len, spd, false, sk, pipe))
1779 return true;
1780 }
1781
1782 return false;
1783}
1784
1785/*
1786 * Map data from the skb to a pipe. Should handle both the linear part,
1787 * the fragments, and the frag list. It does NOT handle frag lists within
1788 * the frag list, if such a thing exists. We'd probably need to recurse to
1789 * handle that cleanly.
1790 */
1791int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1792 struct pipe_inode_info *pipe, unsigned int tlen,
1793 unsigned int flags)
1794{
1795 struct partial_page partial[MAX_SKB_FRAGS];
1796 struct page *pages[MAX_SKB_FRAGS];
1797 struct splice_pipe_desc spd = {
1798 .pages = pages,
1799 .partial = partial,
1800 .nr_pages_max = MAX_SKB_FRAGS,
1801 .flags = flags,
1802 .ops = &nosteal_pipe_buf_ops,
1803 .spd_release = sock_spd_release,
1804 };
1805 struct sk_buff *frag_iter;
1806 struct sock *sk = skb->sk;
1807 int ret = 0;
1808
1809 /*
1810 * __skb_splice_bits() only fails if the output has no room left,
1811 * so no point in going over the frag_list for the error case.
1812 */
1813 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1814 goto done;
1815 else if (!tlen)
1816 goto done;
1817
1818 /*
1819 * now see if we have a frag_list to map
1820 */
1821 skb_walk_frags(skb, frag_iter) {
1822 if (!tlen)
1823 break;
1824 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1825 break;
1826 }
1827
1828done:
1829 if (spd.nr_pages) {
1830 /*
1831 * Drop the socket lock, otherwise we have reverse
1832 * locking dependencies between sk_lock and i_mutex
1833 * here as compared to sendfile(). We enter here
1834 * with the socket lock held, and splice_to_pipe() will
1835 * grab the pipe inode lock. For sendfile() emulation,
1836 * we call into ->sendpage() with the i_mutex lock held
1837 * and networking will grab the socket lock.
1838 */
1839 release_sock(sk);
1840 ret = splice_to_pipe(pipe, &spd);
1841 lock_sock(sk);
1842 }
1843
1844 return ret;
1845}
1846
1847/**
1848 * skb_store_bits - store bits from kernel buffer to skb
1849 * @skb: destination buffer
1850 * @offset: offset in destination
1851 * @from: source buffer
1852 * @len: number of bytes to copy
1853 *
1854 * Copy the specified number of bytes from the source buffer to the
1855 * destination skb. This function handles all the messy bits of
1856 * traversing fragment lists and such.
1857 */
1858
1859int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1860{
1861 int start = skb_headlen(skb);
1862 struct sk_buff *frag_iter;
1863 int i, copy;
1864
1865 if (offset > (int)skb->len - len)
1866 goto fault;
1867
1868 if ((copy = start - offset) > 0) {
1869 if (copy > len)
1870 copy = len;
1871 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1872 if ((len -= copy) == 0)
1873 return 0;
1874 offset += copy;
1875 from += copy;
1876 }
1877
1878 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1879 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1880 int end;
1881
1882 WARN_ON(start > offset + len);
1883
1884 end = start + skb_frag_size(frag);
1885 if ((copy = end - offset) > 0) {
1886 u8 *vaddr;
1887
1888 if (copy > len)
1889 copy = len;
1890
1891 vaddr = kmap_atomic(skb_frag_page(frag));
1892 memcpy(vaddr + frag->page_offset + offset - start,
1893 from, copy);
1894 kunmap_atomic(vaddr);
1895
1896 if ((len -= copy) == 0)
1897 return 0;
1898 offset += copy;
1899 from += copy;
1900 }
1901 start = end;
1902 }
1903
1904 skb_walk_frags(skb, frag_iter) {
1905 int end;
1906
1907 WARN_ON(start > offset + len);
1908
1909 end = start + frag_iter->len;
1910 if ((copy = end - offset) > 0) {
1911 if (copy > len)
1912 copy = len;
1913 if (skb_store_bits(frag_iter, offset - start,
1914 from, copy))
1915 goto fault;
1916 if ((len -= copy) == 0)
1917 return 0;
1918 offset += copy;
1919 from += copy;
1920 }
1921 start = end;
1922 }
1923 if (!len)
1924 return 0;
1925
1926fault:
1927 return -EFAULT;
1928}
1929EXPORT_SYMBOL(skb_store_bits);
1930
1931/* Checksum skb data. */
1932__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1933 __wsum csum, const struct skb_checksum_ops *ops)
1934{
1935 int start = skb_headlen(skb);
1936 int i, copy = start - offset;
1937 struct sk_buff *frag_iter;
1938 int pos = 0;
1939
1940 /* Checksum header. */
1941 if (copy > 0) {
1942 if (copy > len)
1943 copy = len;
1944 csum = ops->update(skb->data + offset, copy, csum);
1945 if ((len -= copy) == 0)
1946 return csum;
1947 offset += copy;
1948 pos = copy;
1949 }
1950
1951 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1952 int end;
1953 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1954
1955 WARN_ON(start > offset + len);
1956
1957 end = start + skb_frag_size(frag);
1958 if ((copy = end - offset) > 0) {
1959 __wsum csum2;
1960 u8 *vaddr;
1961
1962 if (copy > len)
1963 copy = len;
1964 vaddr = kmap_atomic(skb_frag_page(frag));
1965 csum2 = ops->update(vaddr + frag->page_offset +
1966 offset - start, copy, 0);
1967 kunmap_atomic(vaddr);
1968 csum = ops->combine(csum, csum2, pos, copy);
1969 if (!(len -= copy))
1970 return csum;
1971 offset += copy;
1972 pos += copy;
1973 }
1974 start = end;
1975 }
1976
1977 skb_walk_frags(skb, frag_iter) {
1978 int end;
1979
1980 WARN_ON(start > offset + len);
1981
1982 end = start + frag_iter->len;
1983 if ((copy = end - offset) > 0) {
1984 __wsum csum2;
1985 if (copy > len)
1986 copy = len;
1987 csum2 = __skb_checksum(frag_iter, offset - start,
1988 copy, 0, ops);
1989 csum = ops->combine(csum, csum2, pos, copy);
1990 if ((len -= copy) == 0)
1991 return csum;
1992 offset += copy;
1993 pos += copy;
1994 }
1995 start = end;
1996 }
1997 BUG_ON(len);
1998
1999 return csum;
2000}
2001EXPORT_SYMBOL(__skb_checksum);
2002
2003__wsum skb_checksum(const struct sk_buff *skb, int offset,
2004 int len, __wsum csum)
2005{
2006 const struct skb_checksum_ops ops = {
2007 .update = csum_partial_ext,
2008 .combine = csum_block_add_ext,
2009 };
2010
2011 return __skb_checksum(skb, offset, len, csum, &ops);
2012}
2013EXPORT_SYMBOL(skb_checksum);
2014
2015/* Both of above in one bottle. */
2016
2017__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2018 u8 *to, int len, __wsum csum)
2019{
2020 int start = skb_headlen(skb);
2021 int i, copy = start - offset;
2022 struct sk_buff *frag_iter;
2023 int pos = 0;
2024
2025 /* Copy header. */
2026 if (copy > 0) {
2027 if (copy > len)
2028 copy = len;
2029 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2030 copy, csum);
2031 if ((len -= copy) == 0)
2032 return csum;
2033 offset += copy;
2034 to += copy;
2035 pos = copy;
2036 }
2037
2038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2039 int end;
2040
2041 WARN_ON(start > offset + len);
2042
2043 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2044 if ((copy = end - offset) > 0) {
2045 __wsum csum2;
2046 u8 *vaddr;
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048
2049 if (copy > len)
2050 copy = len;
2051 vaddr = kmap_atomic(skb_frag_page(frag));
2052 csum2 = csum_partial_copy_nocheck(vaddr +
2053 frag->page_offset +
2054 offset - start, to,
2055 copy, 0);
2056 kunmap_atomic(vaddr);
2057 csum = csum_block_add(csum, csum2, pos);
2058 if (!(len -= copy))
2059 return csum;
2060 offset += copy;
2061 to += copy;
2062 pos += copy;
2063 }
2064 start = end;
2065 }
2066
2067 skb_walk_frags(skb, frag_iter) {
2068 __wsum csum2;
2069 int end;
2070
2071 WARN_ON(start > offset + len);
2072
2073 end = start + frag_iter->len;
2074 if ((copy = end - offset) > 0) {
2075 if (copy > len)
2076 copy = len;
2077 csum2 = skb_copy_and_csum_bits(frag_iter,
2078 offset - start,
2079 to, copy, 0);
2080 csum = csum_block_add(csum, csum2, pos);
2081 if ((len -= copy) == 0)
2082 return csum;
2083 offset += copy;
2084 to += copy;
2085 pos += copy;
2086 }
2087 start = end;
2088 }
2089 BUG_ON(len);
2090 return csum;
2091}
2092EXPORT_SYMBOL(skb_copy_and_csum_bits);
2093
2094 /**
2095 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2096 * @from: source buffer
2097 *
2098 * Calculates the amount of linear headroom needed in the 'to' skb passed
2099 * into skb_zerocopy().
2100 */
2101unsigned int
2102skb_zerocopy_headlen(const struct sk_buff *from)
2103{
2104 unsigned int hlen = 0;
2105
2106 if (!from->head_frag ||
2107 skb_headlen(from) < L1_CACHE_BYTES ||
2108 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2109 hlen = skb_headlen(from);
2110
2111 if (skb_has_frag_list(from))
2112 hlen = from->len;
2113
2114 return hlen;
2115}
2116EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2117
2118/**
2119 * skb_zerocopy - Zero copy skb to skb
2120 * @to: destination buffer
2121 * @from: source buffer
2122 * @len: number of bytes to copy from source buffer
2123 * @hlen: size of linear headroom in destination buffer
2124 *
2125 * Copies up to `len` bytes from `from` to `to` by creating references
2126 * to the frags in the source buffer.
2127 *
2128 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2129 * headroom in the `to` buffer.
2130 *
2131 * Return value:
2132 * 0: everything is OK
2133 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2134 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2135 */
2136int
2137skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2138{
2139 int i, j = 0;
2140 int plen = 0; /* length of skb->head fragment */
2141 int ret;
2142 struct page *page;
2143 unsigned int offset;
2144
2145 BUG_ON(!from->head_frag && !hlen);
2146
2147 /* dont bother with small payloads */
2148 if (len <= skb_tailroom(to))
2149 return skb_copy_bits(from, 0, skb_put(to, len), len);
2150
2151 if (hlen) {
2152 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2153 if (unlikely(ret))
2154 return ret;
2155 len -= hlen;
2156 } else {
2157 plen = min_t(int, skb_headlen(from), len);
2158 if (plen) {
2159 page = virt_to_head_page(from->head);
2160 offset = from->data - (unsigned char *)page_address(page);
2161 __skb_fill_page_desc(to, 0, page, offset, plen);
2162 get_page(page);
2163 j = 1;
2164 len -= plen;
2165 }
2166 }
2167
2168 to->truesize += len + plen;
2169 to->len += len + plen;
2170 to->data_len += len + plen;
2171
2172 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2173 skb_tx_error(from);
2174 return -ENOMEM;
2175 }
2176
2177 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2178 if (!len)
2179 break;
2180 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2181 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2182 len -= skb_shinfo(to)->frags[j].size;
2183 skb_frag_ref(to, j);
2184 j++;
2185 }
2186 skb_shinfo(to)->nr_frags = j;
2187
2188 return 0;
2189}
2190EXPORT_SYMBOL_GPL(skb_zerocopy);
2191
2192void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2193{
2194 __wsum csum;
2195 long csstart;
2196
2197 if (skb->ip_summed == CHECKSUM_PARTIAL)
2198 csstart = skb_checksum_start_offset(skb);
2199 else
2200 csstart = skb_headlen(skb);
2201
2202 BUG_ON(csstart > skb_headlen(skb));
2203
2204 skb_copy_from_linear_data(skb, to, csstart);
2205
2206 csum = 0;
2207 if (csstart != skb->len)
2208 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2209 skb->len - csstart, 0);
2210
2211 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2212 long csstuff = csstart + skb->csum_offset;
2213
2214 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2215 }
2216}
2217EXPORT_SYMBOL(skb_copy_and_csum_dev);
2218
2219/**
2220 * skb_dequeue - remove from the head of the queue
2221 * @list: list to dequeue from
2222 *
2223 * Remove the head of the list. The list lock is taken so the function
2224 * may be used safely with other locking list functions. The head item is
2225 * returned or %NULL if the list is empty.
2226 */
2227
2228struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2229{
2230 unsigned long flags;
2231 struct sk_buff *result;
2232
2233 spin_lock_irqsave(&list->lock, flags);
2234 result = __skb_dequeue(list);
2235 spin_unlock_irqrestore(&list->lock, flags);
2236 return result;
2237}
2238EXPORT_SYMBOL(skb_dequeue);
2239
2240/**
2241 * skb_dequeue_tail - remove from the tail of the queue
2242 * @list: list to dequeue from
2243 *
2244 * Remove the tail of the list. The list lock is taken so the function
2245 * may be used safely with other locking list functions. The tail item is
2246 * returned or %NULL if the list is empty.
2247 */
2248struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2249{
2250 unsigned long flags;
2251 struct sk_buff *result;
2252
2253 spin_lock_irqsave(&list->lock, flags);
2254 result = __skb_dequeue_tail(list);
2255 spin_unlock_irqrestore(&list->lock, flags);
2256 return result;
2257}
2258EXPORT_SYMBOL(skb_dequeue_tail);
2259
2260/**
2261 * skb_queue_purge - empty a list
2262 * @list: list to empty
2263 *
2264 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2265 * the list and one reference dropped. This function takes the list
2266 * lock and is atomic with respect to other list locking functions.
2267 */
2268void skb_queue_purge(struct sk_buff_head *list)
2269{
2270 struct sk_buff *skb;
2271 while ((skb = skb_dequeue(list)) != NULL)
2272 kfree_skb(skb);
2273}
2274EXPORT_SYMBOL(skb_queue_purge);
2275
2276/**
2277 * skb_queue_head - queue a buffer at the list head
2278 * @list: list to use
2279 * @newsk: buffer to queue
2280 *
2281 * Queue a buffer at the start of the list. This function takes the
2282 * list lock and can be used safely with other locking &sk_buff functions
2283 * safely.
2284 *
2285 * A buffer cannot be placed on two lists at the same time.
2286 */
2287void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2288{
2289 unsigned long flags;
2290
2291 spin_lock_irqsave(&list->lock, flags);
2292 __skb_queue_head(list, newsk);
2293 spin_unlock_irqrestore(&list->lock, flags);
2294}
2295EXPORT_SYMBOL(skb_queue_head);
2296
2297/**
2298 * skb_queue_tail - queue a buffer at the list tail
2299 * @list: list to use
2300 * @newsk: buffer to queue
2301 *
2302 * Queue a buffer at the tail of the list. This function takes the
2303 * list lock and can be used safely with other locking &sk_buff functions
2304 * safely.
2305 *
2306 * A buffer cannot be placed on two lists at the same time.
2307 */
2308void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2309{
2310 unsigned long flags;
2311
2312 spin_lock_irqsave(&list->lock, flags);
2313 __skb_queue_tail(list, newsk);
2314 spin_unlock_irqrestore(&list->lock, flags);
2315}
2316EXPORT_SYMBOL(skb_queue_tail);
2317
2318/**
2319 * skb_unlink - remove a buffer from a list
2320 * @skb: buffer to remove
2321 * @list: list to use
2322 *
2323 * Remove a packet from a list. The list locks are taken and this
2324 * function is atomic with respect to other list locked calls
2325 *
2326 * You must know what list the SKB is on.
2327 */
2328void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2329{
2330 unsigned long flags;
2331
2332 spin_lock_irqsave(&list->lock, flags);
2333 __skb_unlink(skb, list);
2334 spin_unlock_irqrestore(&list->lock, flags);
2335}
2336EXPORT_SYMBOL(skb_unlink);
2337
2338/**
2339 * skb_append - append a buffer
2340 * @old: buffer to insert after
2341 * @newsk: buffer to insert
2342 * @list: list to use
2343 *
2344 * Place a packet after a given packet in a list. The list locks are taken
2345 * and this function is atomic with respect to other list locked calls.
2346 * A buffer cannot be placed on two lists at the same time.
2347 */
2348void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2349{
2350 unsigned long flags;
2351
2352 spin_lock_irqsave(&list->lock, flags);
2353 __skb_queue_after(list, old, newsk);
2354 spin_unlock_irqrestore(&list->lock, flags);
2355}
2356EXPORT_SYMBOL(skb_append);
2357
2358/**
2359 * skb_insert - insert a buffer
2360 * @old: buffer to insert before
2361 * @newsk: buffer to insert
2362 * @list: list to use
2363 *
2364 * Place a packet before a given packet in a list. The list locks are
2365 * taken and this function is atomic with respect to other list locked
2366 * calls.
2367 *
2368 * A buffer cannot be placed on two lists at the same time.
2369 */
2370void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2371{
2372 unsigned long flags;
2373
2374 spin_lock_irqsave(&list->lock, flags);
2375 __skb_insert(newsk, old->prev, old, list);
2376 spin_unlock_irqrestore(&list->lock, flags);
2377}
2378EXPORT_SYMBOL(skb_insert);
2379
2380static inline void skb_split_inside_header(struct sk_buff *skb,
2381 struct sk_buff* skb1,
2382 const u32 len, const int pos)
2383{
2384 int i;
2385
2386 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2387 pos - len);
2388 /* And move data appendix as is. */
2389 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2390 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2391
2392 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2393 skb_shinfo(skb)->nr_frags = 0;
2394 skb1->data_len = skb->data_len;
2395 skb1->len += skb1->data_len;
2396 skb->data_len = 0;
2397 skb->len = len;
2398 skb_set_tail_pointer(skb, len);
2399}
2400
2401static inline void skb_split_no_header(struct sk_buff *skb,
2402 struct sk_buff* skb1,
2403 const u32 len, int pos)
2404{
2405 int i, k = 0;
2406 const int nfrags = skb_shinfo(skb)->nr_frags;
2407
2408 skb_shinfo(skb)->nr_frags = 0;
2409 skb1->len = skb1->data_len = skb->len - len;
2410 skb->len = len;
2411 skb->data_len = len - pos;
2412
2413 for (i = 0; i < nfrags; i++) {
2414 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2415
2416 if (pos + size > len) {
2417 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2418
2419 if (pos < len) {
2420 /* Split frag.
2421 * We have two variants in this case:
2422 * 1. Move all the frag to the second
2423 * part, if it is possible. F.e.
2424 * this approach is mandatory for TUX,
2425 * where splitting is expensive.
2426 * 2. Split is accurately. We make this.
2427 */
2428 skb_frag_ref(skb, i);
2429 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2430 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2431 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2432 skb_shinfo(skb)->nr_frags++;
2433 }
2434 k++;
2435 } else
2436 skb_shinfo(skb)->nr_frags++;
2437 pos += size;
2438 }
2439 skb_shinfo(skb1)->nr_frags = k;
2440}
2441
2442/**
2443 * skb_split - Split fragmented skb to two parts at length len.
2444 * @skb: the buffer to split
2445 * @skb1: the buffer to receive the second part
2446 * @len: new length for skb
2447 */
2448void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2449{
2450 int pos = skb_headlen(skb);
2451
2452 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2453 if (len < pos) /* Split line is inside header. */
2454 skb_split_inside_header(skb, skb1, len, pos);
2455 else /* Second chunk has no header, nothing to copy. */
2456 skb_split_no_header(skb, skb1, len, pos);
2457}
2458EXPORT_SYMBOL(skb_split);
2459
2460/* Shifting from/to a cloned skb is a no-go.
2461 *
2462 * Caller cannot keep skb_shinfo related pointers past calling here!
2463 */
2464static int skb_prepare_for_shift(struct sk_buff *skb)
2465{
2466 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2467}
2468
2469/**
2470 * skb_shift - Shifts paged data partially from skb to another
2471 * @tgt: buffer into which tail data gets added
2472 * @skb: buffer from which the paged data comes from
2473 * @shiftlen: shift up to this many bytes
2474 *
2475 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2476 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2477 * It's up to caller to free skb if everything was shifted.
2478 *
2479 * If @tgt runs out of frags, the whole operation is aborted.
2480 *
2481 * Skb cannot include anything else but paged data while tgt is allowed
2482 * to have non-paged data as well.
2483 *
2484 * TODO: full sized shift could be optimized but that would need
2485 * specialized skb free'er to handle frags without up-to-date nr_frags.
2486 */
2487int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2488{
2489 int from, to, merge, todo;
2490 struct skb_frag_struct *fragfrom, *fragto;
2491
2492 BUG_ON(shiftlen > skb->len);
2493 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2494
2495 todo = shiftlen;
2496 from = 0;
2497 to = skb_shinfo(tgt)->nr_frags;
2498 fragfrom = &skb_shinfo(skb)->frags[from];
2499
2500 /* Actual merge is delayed until the point when we know we can
2501 * commit all, so that we don't have to undo partial changes
2502 */
2503 if (!to ||
2504 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2505 fragfrom->page_offset)) {
2506 merge = -1;
2507 } else {
2508 merge = to - 1;
2509
2510 todo -= skb_frag_size(fragfrom);
2511 if (todo < 0) {
2512 if (skb_prepare_for_shift(skb) ||
2513 skb_prepare_for_shift(tgt))
2514 return 0;
2515
2516 /* All previous frag pointers might be stale! */
2517 fragfrom = &skb_shinfo(skb)->frags[from];
2518 fragto = &skb_shinfo(tgt)->frags[merge];
2519
2520 skb_frag_size_add(fragto, shiftlen);
2521 skb_frag_size_sub(fragfrom, shiftlen);
2522 fragfrom->page_offset += shiftlen;
2523
2524 goto onlymerged;
2525 }
2526
2527 from++;
2528 }
2529
2530 /* Skip full, not-fitting skb to avoid expensive operations */
2531 if ((shiftlen == skb->len) &&
2532 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2533 return 0;
2534
2535 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2536 return 0;
2537
2538 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2539 if (to == MAX_SKB_FRAGS)
2540 return 0;
2541
2542 fragfrom = &skb_shinfo(skb)->frags[from];
2543 fragto = &skb_shinfo(tgt)->frags[to];
2544
2545 if (todo >= skb_frag_size(fragfrom)) {
2546 *fragto = *fragfrom;
2547 todo -= skb_frag_size(fragfrom);
2548 from++;
2549 to++;
2550
2551 } else {
2552 __skb_frag_ref(fragfrom);
2553 fragto->page = fragfrom->page;
2554 fragto->page_offset = fragfrom->page_offset;
2555 skb_frag_size_set(fragto, todo);
2556
2557 fragfrom->page_offset += todo;
2558 skb_frag_size_sub(fragfrom, todo);
2559 todo = 0;
2560
2561 to++;
2562 break;
2563 }
2564 }
2565
2566 /* Ready to "commit" this state change to tgt */
2567 skb_shinfo(tgt)->nr_frags = to;
2568
2569 if (merge >= 0) {
2570 fragfrom = &skb_shinfo(skb)->frags[0];
2571 fragto = &skb_shinfo(tgt)->frags[merge];
2572
2573 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2574 __skb_frag_unref(fragfrom);
2575 }
2576
2577 /* Reposition in the original skb */
2578 to = 0;
2579 while (from < skb_shinfo(skb)->nr_frags)
2580 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2581 skb_shinfo(skb)->nr_frags = to;
2582
2583 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2584
2585onlymerged:
2586 /* Most likely the tgt won't ever need its checksum anymore, skb on
2587 * the other hand might need it if it needs to be resent
2588 */
2589 tgt->ip_summed = CHECKSUM_PARTIAL;
2590 skb->ip_summed = CHECKSUM_PARTIAL;
2591
2592 /* Yak, is it really working this way? Some helper please? */
2593 skb->len -= shiftlen;
2594 skb->data_len -= shiftlen;
2595 skb->truesize -= shiftlen;
2596 tgt->len += shiftlen;
2597 tgt->data_len += shiftlen;
2598 tgt->truesize += shiftlen;
2599
2600 return shiftlen;
2601}
2602
2603/**
2604 * skb_prepare_seq_read - Prepare a sequential read of skb data
2605 * @skb: the buffer to read
2606 * @from: lower offset of data to be read
2607 * @to: upper offset of data to be read
2608 * @st: state variable
2609 *
2610 * Initializes the specified state variable. Must be called before
2611 * invoking skb_seq_read() for the first time.
2612 */
2613void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2614 unsigned int to, struct skb_seq_state *st)
2615{
2616 st->lower_offset = from;
2617 st->upper_offset = to;
2618 st->root_skb = st->cur_skb = skb;
2619 st->frag_idx = st->stepped_offset = 0;
2620 st->frag_data = NULL;
2621}
2622EXPORT_SYMBOL(skb_prepare_seq_read);
2623
2624/**
2625 * skb_seq_read - Sequentially read skb data
2626 * @consumed: number of bytes consumed by the caller so far
2627 * @data: destination pointer for data to be returned
2628 * @st: state variable
2629 *
2630 * Reads a block of skb data at @consumed relative to the
2631 * lower offset specified to skb_prepare_seq_read(). Assigns
2632 * the head of the data block to @data and returns the length
2633 * of the block or 0 if the end of the skb data or the upper
2634 * offset has been reached.
2635 *
2636 * The caller is not required to consume all of the data
2637 * returned, i.e. @consumed is typically set to the number
2638 * of bytes already consumed and the next call to
2639 * skb_seq_read() will return the remaining part of the block.
2640 *
2641 * Note 1: The size of each block of data returned can be arbitrary,
2642 * this limitation is the cost for zerocopy seqeuental
2643 * reads of potentially non linear data.
2644 *
2645 * Note 2: Fragment lists within fragments are not implemented
2646 * at the moment, state->root_skb could be replaced with
2647 * a stack for this purpose.
2648 */
2649unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2650 struct skb_seq_state *st)
2651{
2652 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2653 skb_frag_t *frag;
2654
2655 if (unlikely(abs_offset >= st->upper_offset)) {
2656 if (st->frag_data) {
2657 kunmap_atomic(st->frag_data);
2658 st->frag_data = NULL;
2659 }
2660 return 0;
2661 }
2662
2663next_skb:
2664 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2665
2666 if (abs_offset < block_limit && !st->frag_data) {
2667 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2668 return block_limit - abs_offset;
2669 }
2670
2671 if (st->frag_idx == 0 && !st->frag_data)
2672 st->stepped_offset += skb_headlen(st->cur_skb);
2673
2674 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2675 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2676 block_limit = skb_frag_size(frag) + st->stepped_offset;
2677
2678 if (abs_offset < block_limit) {
2679 if (!st->frag_data)
2680 st->frag_data = kmap_atomic(skb_frag_page(frag));
2681
2682 *data = (u8 *) st->frag_data + frag->page_offset +
2683 (abs_offset - st->stepped_offset);
2684
2685 return block_limit - abs_offset;
2686 }
2687
2688 if (st->frag_data) {
2689 kunmap_atomic(st->frag_data);
2690 st->frag_data = NULL;
2691 }
2692
2693 st->frag_idx++;
2694 st->stepped_offset += skb_frag_size(frag);
2695 }
2696
2697 if (st->frag_data) {
2698 kunmap_atomic(st->frag_data);
2699 st->frag_data = NULL;
2700 }
2701
2702 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2703 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2704 st->frag_idx = 0;
2705 goto next_skb;
2706 } else if (st->cur_skb->next) {
2707 st->cur_skb = st->cur_skb->next;
2708 st->frag_idx = 0;
2709 goto next_skb;
2710 }
2711
2712 return 0;
2713}
2714EXPORT_SYMBOL(skb_seq_read);
2715
2716/**
2717 * skb_abort_seq_read - Abort a sequential read of skb data
2718 * @st: state variable
2719 *
2720 * Must be called if skb_seq_read() was not called until it
2721 * returned 0.
2722 */
2723void skb_abort_seq_read(struct skb_seq_state *st)
2724{
2725 if (st->frag_data)
2726 kunmap_atomic(st->frag_data);
2727}
2728EXPORT_SYMBOL(skb_abort_seq_read);
2729
2730#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2731
2732static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2733 struct ts_config *conf,
2734 struct ts_state *state)
2735{
2736 return skb_seq_read(offset, text, TS_SKB_CB(state));
2737}
2738
2739static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2740{
2741 skb_abort_seq_read(TS_SKB_CB(state));
2742}
2743
2744/**
2745 * skb_find_text - Find a text pattern in skb data
2746 * @skb: the buffer to look in
2747 * @from: search offset
2748 * @to: search limit
2749 * @config: textsearch configuration
2750 * @state: uninitialized textsearch state variable
2751 *
2752 * Finds a pattern in the skb data according to the specified
2753 * textsearch configuration. Use textsearch_next() to retrieve
2754 * subsequent occurrences of the pattern. Returns the offset
2755 * to the first occurrence or UINT_MAX if no match was found.
2756 */
2757unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2758 unsigned int to, struct ts_config *config,
2759 struct ts_state *state)
2760{
2761 unsigned int ret;
2762
2763 config->get_next_block = skb_ts_get_next_block;
2764 config->finish = skb_ts_finish;
2765
2766 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2767
2768 ret = textsearch_find(config, state);
2769 return (ret <= to - from ? ret : UINT_MAX);
2770}
2771EXPORT_SYMBOL(skb_find_text);
2772
2773/**
2774 * skb_append_datato_frags - append the user data to a skb
2775 * @sk: sock structure
2776 * @skb: skb structure to be appened with user data.
2777 * @getfrag: call back function to be used for getting the user data
2778 * @from: pointer to user message iov
2779 * @length: length of the iov message
2780 *
2781 * Description: This procedure append the user data in the fragment part
2782 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2783 */
2784int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2785 int (*getfrag)(void *from, char *to, int offset,
2786 int len, int odd, struct sk_buff *skb),
2787 void *from, int length)
2788{
2789 int frg_cnt = skb_shinfo(skb)->nr_frags;
2790 int copy;
2791 int offset = 0;
2792 int ret;
2793 struct page_frag *pfrag = ¤t->task_frag;
2794
2795 do {
2796 /* Return error if we don't have space for new frag */
2797 if (frg_cnt >= MAX_SKB_FRAGS)
2798 return -EMSGSIZE;
2799
2800 if (!sk_page_frag_refill(sk, pfrag))
2801 return -ENOMEM;
2802
2803 /* copy the user data to page */
2804 copy = min_t(int, length, pfrag->size - pfrag->offset);
2805
2806 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2807 offset, copy, 0, skb);
2808 if (ret < 0)
2809 return -EFAULT;
2810
2811 /* copy was successful so update the size parameters */
2812 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2813 copy);
2814 frg_cnt++;
2815 pfrag->offset += copy;
2816 get_page(pfrag->page);
2817
2818 skb->truesize += copy;
2819 atomic_add(copy, &sk->sk_wmem_alloc);
2820 skb->len += copy;
2821 skb->data_len += copy;
2822 offset += copy;
2823 length -= copy;
2824
2825 } while (length > 0);
2826
2827 return 0;
2828}
2829EXPORT_SYMBOL(skb_append_datato_frags);
2830
2831/**
2832 * skb_pull_rcsum - pull skb and update receive checksum
2833 * @skb: buffer to update
2834 * @len: length of data pulled
2835 *
2836 * This function performs an skb_pull on the packet and updates
2837 * the CHECKSUM_COMPLETE checksum. It should be used on
2838 * receive path processing instead of skb_pull unless you know
2839 * that the checksum difference is zero (e.g., a valid IP header)
2840 * or you are setting ip_summed to CHECKSUM_NONE.
2841 */
2842unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2843{
2844 BUG_ON(len > skb->len);
2845 skb->len -= len;
2846 BUG_ON(skb->len < skb->data_len);
2847 skb_postpull_rcsum(skb, skb->data, len);
2848 return skb->data += len;
2849}
2850EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2851
2852/**
2853 * skb_segment - Perform protocol segmentation on skb.
2854 * @head_skb: buffer to segment
2855 * @features: features for the output path (see dev->features)
2856 *
2857 * This function performs segmentation on the given skb. It returns
2858 * a pointer to the first in a list of new skbs for the segments.
2859 * In case of error it returns ERR_PTR(err).
2860 */
2861struct sk_buff *skb_segment(struct sk_buff *head_skb,
2862 netdev_features_t features)
2863{
2864 struct sk_buff *segs = NULL;
2865 struct sk_buff *tail = NULL;
2866 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
2867 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2868 unsigned int mss = skb_shinfo(head_skb)->gso_size;
2869 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
2870 struct sk_buff *frag_skb = head_skb;
2871 unsigned int offset = doffset;
2872 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
2873 unsigned int headroom;
2874 unsigned int len;
2875 __be16 proto;
2876 bool csum;
2877 int sg = !!(features & NETIF_F_SG);
2878 int nfrags = skb_shinfo(head_skb)->nr_frags;
2879 int err = -ENOMEM;
2880 int i = 0;
2881 int pos;
2882 int dummy;
2883
2884 proto = skb_network_protocol(head_skb, &dummy);
2885 if (unlikely(!proto))
2886 return ERR_PTR(-EINVAL);
2887
2888 csum = !!can_checksum_protocol(features, proto);
2889 __skb_push(head_skb, doffset);
2890 headroom = skb_headroom(head_skb);
2891 pos = skb_headlen(head_skb);
2892
2893 do {
2894 struct sk_buff *nskb;
2895 skb_frag_t *nskb_frag;
2896 int hsize;
2897 int size;
2898
2899 len = head_skb->len - offset;
2900 if (len > mss)
2901 len = mss;
2902
2903 hsize = skb_headlen(head_skb) - offset;
2904 if (hsize < 0)
2905 hsize = 0;
2906 if (hsize > len || !sg)
2907 hsize = len;
2908
2909 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
2910 (skb_headlen(list_skb) == len || sg)) {
2911 BUG_ON(skb_headlen(list_skb) > len);
2912
2913 i = 0;
2914 nfrags = skb_shinfo(list_skb)->nr_frags;
2915 frag = skb_shinfo(list_skb)->frags;
2916 frag_skb = list_skb;
2917 pos += skb_headlen(list_skb);
2918
2919 while (pos < offset + len) {
2920 BUG_ON(i >= nfrags);
2921
2922 size = skb_frag_size(frag);
2923 if (pos + size > offset + len)
2924 break;
2925
2926 i++;
2927 pos += size;
2928 frag++;
2929 }
2930
2931 nskb = skb_clone(list_skb, GFP_ATOMIC);
2932 list_skb = list_skb->next;
2933
2934 if (unlikely(!nskb))
2935 goto err;
2936
2937 if (unlikely(pskb_trim(nskb, len))) {
2938 kfree_skb(nskb);
2939 goto err;
2940 }
2941
2942 hsize = skb_end_offset(nskb);
2943 if (skb_cow_head(nskb, doffset + headroom)) {
2944 kfree_skb(nskb);
2945 goto err;
2946 }
2947
2948 nskb->truesize += skb_end_offset(nskb) - hsize;
2949 skb_release_head_state(nskb);
2950 __skb_push(nskb, doffset);
2951 } else {
2952 nskb = __alloc_skb(hsize + doffset + headroom,
2953 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
2954 NUMA_NO_NODE);
2955
2956 if (unlikely(!nskb))
2957 goto err;
2958
2959 skb_reserve(nskb, headroom);
2960 __skb_put(nskb, doffset);
2961 }
2962
2963 if (segs)
2964 tail->next = nskb;
2965 else
2966 segs = nskb;
2967 tail = nskb;
2968
2969 __copy_skb_header(nskb, head_skb);
2970 nskb->mac_len = head_skb->mac_len;
2971
2972 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2973
2974 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
2975 nskb->data - tnl_hlen,
2976 doffset + tnl_hlen);
2977
2978 if (nskb->len == len + doffset)
2979 goto perform_csum_check;
2980
2981 if (!sg) {
2982 nskb->ip_summed = CHECKSUM_NONE;
2983 nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
2984 skb_put(nskb, len),
2985 len, 0);
2986 continue;
2987 }
2988
2989 nskb_frag = skb_shinfo(nskb)->frags;
2990
2991 skb_copy_from_linear_data_offset(head_skb, offset,
2992 skb_put(nskb, hsize), hsize);
2993
2994 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
2995 SKBTX_SHARED_FRAG;
2996
2997 while (pos < offset + len) {
2998 if (i >= nfrags) {
2999 BUG_ON(skb_headlen(list_skb));
3000
3001 i = 0;
3002 nfrags = skb_shinfo(list_skb)->nr_frags;
3003 frag = skb_shinfo(list_skb)->frags;
3004 frag_skb = list_skb;
3005
3006 BUG_ON(!nfrags);
3007
3008 list_skb = list_skb->next;
3009 }
3010
3011 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3012 MAX_SKB_FRAGS)) {
3013 net_warn_ratelimited(
3014 "skb_segment: too many frags: %u %u\n",
3015 pos, mss);
3016 goto err;
3017 }
3018
3019 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3020 goto err;
3021
3022 *nskb_frag = *frag;
3023 __skb_frag_ref(nskb_frag);
3024 size = skb_frag_size(nskb_frag);
3025
3026 if (pos < offset) {
3027 nskb_frag->page_offset += offset - pos;
3028 skb_frag_size_sub(nskb_frag, offset - pos);
3029 }
3030
3031 skb_shinfo(nskb)->nr_frags++;
3032
3033 if (pos + size <= offset + len) {
3034 i++;
3035 frag++;
3036 pos += size;
3037 } else {
3038 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3039 goto skip_fraglist;
3040 }
3041
3042 nskb_frag++;
3043 }
3044
3045skip_fraglist:
3046 nskb->data_len = len - hsize;
3047 nskb->len += nskb->data_len;
3048 nskb->truesize += nskb->data_len;
3049
3050perform_csum_check:
3051 if (!csum) {
3052 nskb->csum = skb_checksum(nskb, doffset,
3053 nskb->len - doffset, 0);
3054 nskb->ip_summed = CHECKSUM_NONE;
3055 }
3056 } while ((offset += len) < head_skb->len);
3057
3058 return segs;
3059
3060err:
3061 kfree_skb_list(segs);
3062 return ERR_PTR(err);
3063}
3064EXPORT_SYMBOL_GPL(skb_segment);
3065
3066int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3067{
3068 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3069 unsigned int offset = skb_gro_offset(skb);
3070 unsigned int headlen = skb_headlen(skb);
3071 struct sk_buff *nskb, *lp, *p = *head;
3072 unsigned int len = skb_gro_len(skb);
3073 unsigned int delta_truesize;
3074 unsigned int headroom;
3075
3076 if (unlikely(p->len + len >= 65536))
3077 return -E2BIG;
3078
3079 lp = NAPI_GRO_CB(p)->last;
3080 pinfo = skb_shinfo(lp);
3081
3082 if (headlen <= offset) {
3083 skb_frag_t *frag;
3084 skb_frag_t *frag2;
3085 int i = skbinfo->nr_frags;
3086 int nr_frags = pinfo->nr_frags + i;
3087
3088 if (nr_frags > MAX_SKB_FRAGS)
3089 goto merge;
3090
3091 offset -= headlen;
3092 pinfo->nr_frags = nr_frags;
3093 skbinfo->nr_frags = 0;
3094
3095 frag = pinfo->frags + nr_frags;
3096 frag2 = skbinfo->frags + i;
3097 do {
3098 *--frag = *--frag2;
3099 } while (--i);
3100
3101 frag->page_offset += offset;
3102 skb_frag_size_sub(frag, offset);
3103
3104 /* all fragments truesize : remove (head size + sk_buff) */
3105 delta_truesize = skb->truesize -
3106 SKB_TRUESIZE(skb_end_offset(skb));
3107
3108 skb->truesize -= skb->data_len;
3109 skb->len -= skb->data_len;
3110 skb->data_len = 0;
3111
3112 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3113 goto done;
3114 } else if (skb->head_frag) {
3115 int nr_frags = pinfo->nr_frags;
3116 skb_frag_t *frag = pinfo->frags + nr_frags;
3117 struct page *page = virt_to_head_page(skb->head);
3118 unsigned int first_size = headlen - offset;
3119 unsigned int first_offset;
3120
3121 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3122 goto merge;
3123
3124 first_offset = skb->data -
3125 (unsigned char *)page_address(page) +
3126 offset;
3127
3128 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3129
3130 frag->page.p = page;
3131 frag->page_offset = first_offset;
3132 skb_frag_size_set(frag, first_size);
3133
3134 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3135 /* We dont need to clear skbinfo->nr_frags here */
3136
3137 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3138 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3139 goto done;
3140 }
3141 if (pinfo->frag_list)
3142 goto merge;
3143 if (skb_gro_len(p) != pinfo->gso_size)
3144 return -E2BIG;
3145
3146 headroom = skb_headroom(p);
3147 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3148 if (unlikely(!nskb))
3149 return -ENOMEM;
3150
3151 __copy_skb_header(nskb, p);
3152 nskb->mac_len = p->mac_len;
3153
3154 skb_reserve(nskb, headroom);
3155 __skb_put(nskb, skb_gro_offset(p));
3156
3157 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3158 skb_set_network_header(nskb, skb_network_offset(p));
3159 skb_set_transport_header(nskb, skb_transport_offset(p));
3160
3161 __skb_pull(p, skb_gro_offset(p));
3162 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3163 p->data - skb_mac_header(p));
3164
3165 skb_shinfo(nskb)->frag_list = p;
3166 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3167 pinfo->gso_size = 0;
3168 skb_header_release(p);
3169 NAPI_GRO_CB(nskb)->last = p;
3170
3171 nskb->data_len += p->len;
3172 nskb->truesize += p->truesize;
3173 nskb->len += p->len;
3174
3175 *head = nskb;
3176 nskb->next = p->next;
3177 p->next = NULL;
3178
3179 p = nskb;
3180
3181merge:
3182 delta_truesize = skb->truesize;
3183 if (offset > headlen) {
3184 unsigned int eat = offset - headlen;
3185
3186 skbinfo->frags[0].page_offset += eat;
3187 skb_frag_size_sub(&skbinfo->frags[0], eat);
3188 skb->data_len -= eat;
3189 skb->len -= eat;
3190 offset = headlen;
3191 }
3192
3193 __skb_pull(skb, offset);
3194
3195 if (NAPI_GRO_CB(p)->last == p)
3196 skb_shinfo(p)->frag_list = skb;
3197 else
3198 NAPI_GRO_CB(p)->last->next = skb;
3199 NAPI_GRO_CB(p)->last = skb;
3200 skb_header_release(skb);
3201 lp = p;
3202
3203done:
3204 NAPI_GRO_CB(p)->count++;
3205 p->data_len += len;
3206 p->truesize += delta_truesize;
3207 p->len += len;
3208 if (lp != p) {
3209 lp->data_len += len;
3210 lp->truesize += delta_truesize;
3211 lp->len += len;
3212 }
3213 NAPI_GRO_CB(skb)->same_flow = 1;
3214 return 0;
3215}
3216EXPORT_SYMBOL_GPL(skb_gro_receive);
3217
3218void __init skb_init(void)
3219{
3220 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3221 sizeof(struct sk_buff),
3222 0,
3223 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3224 NULL);
3225 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3226 (2*sizeof(struct sk_buff)) +
3227 sizeof(atomic_t),
3228 0,
3229 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3230 NULL);
3231}
3232
3233/**
3234 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3235 * @skb: Socket buffer containing the buffers to be mapped
3236 * @sg: The scatter-gather list to map into
3237 * @offset: The offset into the buffer's contents to start mapping
3238 * @len: Length of buffer space to be mapped
3239 *
3240 * Fill the specified scatter-gather list with mappings/pointers into a
3241 * region of the buffer space attached to a socket buffer.
3242 */
3243static int
3244__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3245{
3246 int start = skb_headlen(skb);
3247 int i, copy = start - offset;
3248 struct sk_buff *frag_iter;
3249 int elt = 0;
3250
3251 if (copy > 0) {
3252 if (copy > len)
3253 copy = len;
3254 sg_set_buf(sg, skb->data + offset, copy);
3255 elt++;
3256 if ((len -= copy) == 0)
3257 return elt;
3258 offset += copy;
3259 }
3260
3261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3262 int end;
3263
3264 WARN_ON(start > offset + len);
3265
3266 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3267 if ((copy = end - offset) > 0) {
3268 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3269
3270 if (copy > len)
3271 copy = len;
3272 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3273 frag->page_offset+offset-start);
3274 elt++;
3275 if (!(len -= copy))
3276 return elt;
3277 offset += copy;
3278 }
3279 start = end;
3280 }
3281
3282 skb_walk_frags(skb, frag_iter) {
3283 int end;
3284
3285 WARN_ON(start > offset + len);
3286
3287 end = start + frag_iter->len;
3288 if ((copy = end - offset) > 0) {
3289 if (copy > len)
3290 copy = len;
3291 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3292 copy);
3293 if ((len -= copy) == 0)
3294 return elt;
3295 offset += copy;
3296 }
3297 start = end;
3298 }
3299 BUG_ON(len);
3300 return elt;
3301}
3302
3303/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3304 * sglist without mark the sg which contain last skb data as the end.
3305 * So the caller can mannipulate sg list as will when padding new data after
3306 * the first call without calling sg_unmark_end to expend sg list.
3307 *
3308 * Scenario to use skb_to_sgvec_nomark:
3309 * 1. sg_init_table
3310 * 2. skb_to_sgvec_nomark(payload1)
3311 * 3. skb_to_sgvec_nomark(payload2)
3312 *
3313 * This is equivalent to:
3314 * 1. sg_init_table
3315 * 2. skb_to_sgvec(payload1)
3316 * 3. sg_unmark_end
3317 * 4. skb_to_sgvec(payload2)
3318 *
3319 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3320 * is more preferable.
3321 */
3322int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3323 int offset, int len)
3324{
3325 return __skb_to_sgvec(skb, sg, offset, len);
3326}
3327EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3328
3329int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3330{
3331 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3332
3333 sg_mark_end(&sg[nsg - 1]);
3334
3335 return nsg;
3336}
3337EXPORT_SYMBOL_GPL(skb_to_sgvec);
3338
3339/**
3340 * skb_cow_data - Check that a socket buffer's data buffers are writable
3341 * @skb: The socket buffer to check.
3342 * @tailbits: Amount of trailing space to be added
3343 * @trailer: Returned pointer to the skb where the @tailbits space begins
3344 *
3345 * Make sure that the data buffers attached to a socket buffer are
3346 * writable. If they are not, private copies are made of the data buffers
3347 * and the socket buffer is set to use these instead.
3348 *
3349 * If @tailbits is given, make sure that there is space to write @tailbits
3350 * bytes of data beyond current end of socket buffer. @trailer will be
3351 * set to point to the skb in which this space begins.
3352 *
3353 * The number of scatterlist elements required to completely map the
3354 * COW'd and extended socket buffer will be returned.
3355 */
3356int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3357{
3358 int copyflag;
3359 int elt;
3360 struct sk_buff *skb1, **skb_p;
3361
3362 /* If skb is cloned or its head is paged, reallocate
3363 * head pulling out all the pages (pages are considered not writable
3364 * at the moment even if they are anonymous).
3365 */
3366 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3367 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3368 return -ENOMEM;
3369
3370 /* Easy case. Most of packets will go this way. */
3371 if (!skb_has_frag_list(skb)) {
3372 /* A little of trouble, not enough of space for trailer.
3373 * This should not happen, when stack is tuned to generate
3374 * good frames. OK, on miss we reallocate and reserve even more
3375 * space, 128 bytes is fair. */
3376
3377 if (skb_tailroom(skb) < tailbits &&
3378 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3379 return -ENOMEM;
3380
3381 /* Voila! */
3382 *trailer = skb;
3383 return 1;
3384 }
3385
3386 /* Misery. We are in troubles, going to mincer fragments... */
3387
3388 elt = 1;
3389 skb_p = &skb_shinfo(skb)->frag_list;
3390 copyflag = 0;
3391
3392 while ((skb1 = *skb_p) != NULL) {
3393 int ntail = 0;
3394
3395 /* The fragment is partially pulled by someone,
3396 * this can happen on input. Copy it and everything
3397 * after it. */
3398
3399 if (skb_shared(skb1))
3400 copyflag = 1;
3401
3402 /* If the skb is the last, worry about trailer. */
3403
3404 if (skb1->next == NULL && tailbits) {
3405 if (skb_shinfo(skb1)->nr_frags ||
3406 skb_has_frag_list(skb1) ||
3407 skb_tailroom(skb1) < tailbits)
3408 ntail = tailbits + 128;
3409 }
3410
3411 if (copyflag ||
3412 skb_cloned(skb1) ||
3413 ntail ||
3414 skb_shinfo(skb1)->nr_frags ||
3415 skb_has_frag_list(skb1)) {
3416 struct sk_buff *skb2;
3417
3418 /* Fuck, we are miserable poor guys... */
3419 if (ntail == 0)
3420 skb2 = skb_copy(skb1, GFP_ATOMIC);
3421 else
3422 skb2 = skb_copy_expand(skb1,
3423 skb_headroom(skb1),
3424 ntail,
3425 GFP_ATOMIC);
3426 if (unlikely(skb2 == NULL))
3427 return -ENOMEM;
3428
3429 if (skb1->sk)
3430 skb_set_owner_w(skb2, skb1->sk);
3431
3432 /* Looking around. Are we still alive?
3433 * OK, link new skb, drop old one */
3434
3435 skb2->next = skb1->next;
3436 *skb_p = skb2;
3437 kfree_skb(skb1);
3438 skb1 = skb2;
3439 }
3440 elt++;
3441 *trailer = skb1;
3442 skb_p = &skb1->next;
3443 }
3444
3445 return elt;
3446}
3447EXPORT_SYMBOL_GPL(skb_cow_data);
3448
3449static void sock_rmem_free(struct sk_buff *skb)
3450{
3451 struct sock *sk = skb->sk;
3452
3453 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3454}
3455
3456/*
3457 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3458 */
3459int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3460{
3461 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3462 (unsigned int)sk->sk_rcvbuf)
3463 return -ENOMEM;
3464
3465 skb_orphan(skb);
3466 skb->sk = sk;
3467 skb->destructor = sock_rmem_free;
3468 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3469
3470 /* before exiting rcu section, make sure dst is refcounted */
3471 skb_dst_force(skb);
3472
3473 skb_queue_tail(&sk->sk_error_queue, skb);
3474 if (!sock_flag(sk, SOCK_DEAD))
3475 sk->sk_data_ready(sk);
3476 return 0;
3477}
3478EXPORT_SYMBOL(sock_queue_err_skb);
3479
3480void skb_tstamp_tx(struct sk_buff *orig_skb,
3481 struct skb_shared_hwtstamps *hwtstamps)
3482{
3483 struct sock *sk = orig_skb->sk;
3484 struct sock_exterr_skb *serr;
3485 struct sk_buff *skb;
3486 int err;
3487
3488 if (!sk)
3489 return;
3490
3491 if (hwtstamps) {
3492 *skb_hwtstamps(orig_skb) =
3493 *hwtstamps;
3494 } else {
3495 /*
3496 * no hardware time stamps available,
3497 * so keep the shared tx_flags and only
3498 * store software time stamp
3499 */
3500 orig_skb->tstamp = ktime_get_real();
3501 }
3502
3503 skb = skb_clone(orig_skb, GFP_ATOMIC);
3504 if (!skb)
3505 return;
3506
3507 serr = SKB_EXT_ERR(skb);
3508 memset(serr, 0, sizeof(*serr));
3509 serr->ee.ee_errno = ENOMSG;
3510 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3511
3512 err = sock_queue_err_skb(sk, skb);
3513
3514 if (err)
3515 kfree_skb(skb);
3516}
3517EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3518
3519void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3520{
3521 struct sock *sk = skb->sk;
3522 struct sock_exterr_skb *serr;
3523 int err;
3524
3525 skb->wifi_acked_valid = 1;
3526 skb->wifi_acked = acked;
3527
3528 serr = SKB_EXT_ERR(skb);
3529 memset(serr, 0, sizeof(*serr));
3530 serr->ee.ee_errno = ENOMSG;
3531 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3532
3533 err = sock_queue_err_skb(sk, skb);
3534 if (err)
3535 kfree_skb(skb);
3536}
3537EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3538
3539
3540/**
3541 * skb_partial_csum_set - set up and verify partial csum values for packet
3542 * @skb: the skb to set
3543 * @start: the number of bytes after skb->data to start checksumming.
3544 * @off: the offset from start to place the checksum.
3545 *
3546 * For untrusted partially-checksummed packets, we need to make sure the values
3547 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3548 *
3549 * This function checks and sets those values and skb->ip_summed: if this
3550 * returns false you should drop the packet.
3551 */
3552bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3553{
3554 if (unlikely(start > skb_headlen(skb)) ||
3555 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3556 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3557 start, off, skb_headlen(skb));
3558 return false;
3559 }
3560 skb->ip_summed = CHECKSUM_PARTIAL;
3561 skb->csum_start = skb_headroom(skb) + start;
3562 skb->csum_offset = off;
3563 skb_set_transport_header(skb, start);
3564 return true;
3565}
3566EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3567
3568static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3569 unsigned int max)
3570{
3571 if (skb_headlen(skb) >= len)
3572 return 0;
3573
3574 /* If we need to pullup then pullup to the max, so we
3575 * won't need to do it again.
3576 */
3577 if (max > skb->len)
3578 max = skb->len;
3579
3580 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3581 return -ENOMEM;
3582
3583 if (skb_headlen(skb) < len)
3584 return -EPROTO;
3585
3586 return 0;
3587}
3588
3589#define MAX_TCP_HDR_LEN (15 * 4)
3590
3591static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3592 typeof(IPPROTO_IP) proto,
3593 unsigned int off)
3594{
3595 switch (proto) {
3596 int err;
3597
3598 case IPPROTO_TCP:
3599 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3600 off + MAX_TCP_HDR_LEN);
3601 if (!err && !skb_partial_csum_set(skb, off,
3602 offsetof(struct tcphdr,
3603 check)))
3604 err = -EPROTO;
3605 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3606
3607 case IPPROTO_UDP:
3608 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3609 off + sizeof(struct udphdr));
3610 if (!err && !skb_partial_csum_set(skb, off,
3611 offsetof(struct udphdr,
3612 check)))
3613 err = -EPROTO;
3614 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3615 }
3616
3617 return ERR_PTR(-EPROTO);
3618}
3619
3620/* This value should be large enough to cover a tagged ethernet header plus
3621 * maximally sized IP and TCP or UDP headers.
3622 */
3623#define MAX_IP_HDR_LEN 128
3624
3625static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3626{
3627 unsigned int off;
3628 bool fragment;
3629 __sum16 *csum;
3630 int err;
3631
3632 fragment = false;
3633
3634 err = skb_maybe_pull_tail(skb,
3635 sizeof(struct iphdr),
3636 MAX_IP_HDR_LEN);
3637 if (err < 0)
3638 goto out;
3639
3640 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3641 fragment = true;
3642
3643 off = ip_hdrlen(skb);
3644
3645 err = -EPROTO;
3646
3647 if (fragment)
3648 goto out;
3649
3650 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3651 if (IS_ERR(csum))
3652 return PTR_ERR(csum);
3653
3654 if (recalculate)
3655 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3656 ip_hdr(skb)->daddr,
3657 skb->len - off,
3658 ip_hdr(skb)->protocol, 0);
3659 err = 0;
3660
3661out:
3662 return err;
3663}
3664
3665/* This value should be large enough to cover a tagged ethernet header plus
3666 * an IPv6 header, all options, and a maximal TCP or UDP header.
3667 */
3668#define MAX_IPV6_HDR_LEN 256
3669
3670#define OPT_HDR(type, skb, off) \
3671 (type *)(skb_network_header(skb) + (off))
3672
3673static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3674{
3675 int err;
3676 u8 nexthdr;
3677 unsigned int off;
3678 unsigned int len;
3679 bool fragment;
3680 bool done;
3681 __sum16 *csum;
3682
3683 fragment = false;
3684 done = false;
3685
3686 off = sizeof(struct ipv6hdr);
3687
3688 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3689 if (err < 0)
3690 goto out;
3691
3692 nexthdr = ipv6_hdr(skb)->nexthdr;
3693
3694 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3695 while (off <= len && !done) {
3696 switch (nexthdr) {
3697 case IPPROTO_DSTOPTS:
3698 case IPPROTO_HOPOPTS:
3699 case IPPROTO_ROUTING: {
3700 struct ipv6_opt_hdr *hp;
3701
3702 err = skb_maybe_pull_tail(skb,
3703 off +
3704 sizeof(struct ipv6_opt_hdr),
3705 MAX_IPV6_HDR_LEN);
3706 if (err < 0)
3707 goto out;
3708
3709 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3710 nexthdr = hp->nexthdr;
3711 off += ipv6_optlen(hp);
3712 break;
3713 }
3714 case IPPROTO_AH: {
3715 struct ip_auth_hdr *hp;
3716
3717 err = skb_maybe_pull_tail(skb,
3718 off +
3719 sizeof(struct ip_auth_hdr),
3720 MAX_IPV6_HDR_LEN);
3721 if (err < 0)
3722 goto out;
3723
3724 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3725 nexthdr = hp->nexthdr;
3726 off += ipv6_authlen(hp);
3727 break;
3728 }
3729 case IPPROTO_FRAGMENT: {
3730 struct frag_hdr *hp;
3731
3732 err = skb_maybe_pull_tail(skb,
3733 off +
3734 sizeof(struct frag_hdr),
3735 MAX_IPV6_HDR_LEN);
3736 if (err < 0)
3737 goto out;
3738
3739 hp = OPT_HDR(struct frag_hdr, skb, off);
3740
3741 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3742 fragment = true;
3743
3744 nexthdr = hp->nexthdr;
3745 off += sizeof(struct frag_hdr);
3746 break;
3747 }
3748 default:
3749 done = true;
3750 break;
3751 }
3752 }
3753
3754 err = -EPROTO;
3755
3756 if (!done || fragment)
3757 goto out;
3758
3759 csum = skb_checksum_setup_ip(skb, nexthdr, off);
3760 if (IS_ERR(csum))
3761 return PTR_ERR(csum);
3762
3763 if (recalculate)
3764 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3765 &ipv6_hdr(skb)->daddr,
3766 skb->len - off, nexthdr, 0);
3767 err = 0;
3768
3769out:
3770 return err;
3771}
3772
3773/**
3774 * skb_checksum_setup - set up partial checksum offset
3775 * @skb: the skb to set up
3776 * @recalculate: if true the pseudo-header checksum will be recalculated
3777 */
3778int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3779{
3780 int err;
3781
3782 switch (skb->protocol) {
3783 case htons(ETH_P_IP):
3784 err = skb_checksum_setup_ipv4(skb, recalculate);
3785 break;
3786
3787 case htons(ETH_P_IPV6):
3788 err = skb_checksum_setup_ipv6(skb, recalculate);
3789 break;
3790
3791 default:
3792 err = -EPROTO;
3793 break;
3794 }
3795
3796 return err;
3797}
3798EXPORT_SYMBOL(skb_checksum_setup);
3799
3800void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3801{
3802 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3803 skb->dev->name);
3804}
3805EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3806
3807void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3808{
3809 if (head_stolen) {
3810 skb_release_head_state(skb);
3811 kmem_cache_free(skbuff_head_cache, skb);
3812 } else {
3813 __kfree_skb(skb);
3814 }
3815}
3816EXPORT_SYMBOL(kfree_skb_partial);
3817
3818/**
3819 * skb_try_coalesce - try to merge skb to prior one
3820 * @to: prior buffer
3821 * @from: buffer to add
3822 * @fragstolen: pointer to boolean
3823 * @delta_truesize: how much more was allocated than was requested
3824 */
3825bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3826 bool *fragstolen, int *delta_truesize)
3827{
3828 int i, delta, len = from->len;
3829
3830 *fragstolen = false;
3831
3832 if (skb_cloned(to))
3833 return false;
3834
3835 if (len <= skb_tailroom(to)) {
3836 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3837 *delta_truesize = 0;
3838 return true;
3839 }
3840
3841 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3842 return false;
3843
3844 if (skb_headlen(from) != 0) {
3845 struct page *page;
3846 unsigned int offset;
3847
3848 if (skb_shinfo(to)->nr_frags +
3849 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3850 return false;
3851
3852 if (skb_head_is_locked(from))
3853 return false;
3854
3855 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3856
3857 page = virt_to_head_page(from->head);
3858 offset = from->data - (unsigned char *)page_address(page);
3859
3860 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3861 page, offset, skb_headlen(from));
3862 *fragstolen = true;
3863 } else {
3864 if (skb_shinfo(to)->nr_frags +
3865 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3866 return false;
3867
3868 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3869 }
3870
3871 WARN_ON_ONCE(delta < len);
3872
3873 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3874 skb_shinfo(from)->frags,
3875 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3876 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3877
3878 if (!skb_cloned(from))
3879 skb_shinfo(from)->nr_frags = 0;
3880
3881 /* if the skb is not cloned this does nothing
3882 * since we set nr_frags to 0.
3883 */
3884 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3885 skb_frag_ref(from, i);
3886
3887 to->truesize += delta;
3888 to->len += len;
3889 to->data_len += len;
3890
3891 *delta_truesize = delta;
3892 return true;
3893}
3894EXPORT_SYMBOL(skb_try_coalesce);
3895
3896/**
3897 * skb_scrub_packet - scrub an skb
3898 *
3899 * @skb: buffer to clean
3900 * @xnet: packet is crossing netns
3901 *
3902 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3903 * into/from a tunnel. Some information have to be cleared during these
3904 * operations.
3905 * skb_scrub_packet can also be used to clean a skb before injecting it in
3906 * another namespace (@xnet == true). We have to clear all information in the
3907 * skb that could impact namespace isolation.
3908 */
3909void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3910{
3911 if (xnet)
3912 skb_orphan(skb);
3913 skb->tstamp.tv64 = 0;
3914 skb->pkt_type = PACKET_HOST;
3915 skb->skb_iif = 0;
3916 skb->local_df = 0;
3917 skb_dst_drop(skb);
3918 skb->mark = 0;
3919 secpath_reset(skb);
3920 nf_reset(skb);
3921 nf_reset_trace(skb);
3922}
3923EXPORT_SYMBOL_GPL(skb_scrub_packet);
3924
3925/**
3926 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3927 *
3928 * @skb: GSO skb
3929 *
3930 * skb_gso_transport_seglen is used to determine the real size of the
3931 * individual segments, including Layer4 headers (TCP/UDP).
3932 *
3933 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3934 */
3935unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
3936{
3937 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3938
3939 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3940 return tcp_hdrlen(skb) + shinfo->gso_size;
3941
3942 /* UFO sets gso_size to the size of the fragmentation
3943 * payload, i.e. the size of the L4 (UDP) header is already
3944 * accounted for.
3945 */
3946 return shinfo->gso_size;
3947}
3948EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);