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/netdevice.h>
51#ifdef CONFIG_NET_CLS_ACT
52#include <net/pkt_sched.h>
53#endif
54#include <linux/string.h>
55#include <linux/skbuff.h>
56#include <linux/splice.h>
57#include <linux/cache.h>
58#include <linux/rtnetlink.h>
59#include <linux/init.h>
60#include <linux/scatterlist.h>
61#include <linux/errqueue.h>
62#include <linux/prefetch.h>
63
64#include <net/protocol.h>
65#include <net/dst.h>
66#include <net/sock.h>
67#include <net/checksum.h>
68#include <net/xfrm.h>
69
70#include <asm/uaccess.h>
71#include <trace/events/skb.h>
72#include <linux/highmem.h>
73
74struct kmem_cache *skbuff_head_cache __read_mostly;
75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
76
77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
79{
80 put_page(buf->page);
81}
82
83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
85{
86 get_page(buf->page);
87}
88
89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
91{
92 return 1;
93}
94
95
96/* Pipe buffer operations for a socket. */
97static const struct pipe_buf_operations sock_pipe_buf_ops = {
98 .can_merge = 0,
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
105};
106
107/*
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
110 * reliable.
111 */
112
113/**
114 * skb_over_panic - private function
115 * @skb: buffer
116 * @sz: size
117 * @here: address
118 *
119 * Out of line support code for skb_put(). Not user callable.
120 */
121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
122{
123 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
124 __func__, here, skb->len, sz, skb->head, skb->data,
125 (unsigned long)skb->tail, (unsigned long)skb->end,
126 skb->dev ? skb->dev->name : "<NULL>");
127 BUG();
128}
129
130/**
131 * skb_under_panic - private function
132 * @skb: buffer
133 * @sz: size
134 * @here: address
135 *
136 * Out of line support code for skb_push(). Not user callable.
137 */
138
139static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
140{
141 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
142 __func__, here, skb->len, sz, skb->head, skb->data,
143 (unsigned long)skb->tail, (unsigned long)skb->end,
144 skb->dev ? skb->dev->name : "<NULL>");
145 BUG();
146}
147
148/* Allocate a new skbuff. We do this ourselves so we can fill in a few
149 * 'private' fields and also do memory statistics to find all the
150 * [BEEP] leaks.
151 *
152 */
153
154/**
155 * __alloc_skb - allocate a network buffer
156 * @size: size to allocate
157 * @gfp_mask: allocation mask
158 * @fclone: allocate from fclone cache instead of head cache
159 * and allocate a cloned (child) skb
160 * @node: numa node to allocate memory on
161 *
162 * Allocate a new &sk_buff. The returned buffer has no headroom and a
163 * tail room of size bytes. The object has a reference count of one.
164 * The return is the buffer. On a failure the return is %NULL.
165 *
166 * Buffers may only be allocated from interrupts using a @gfp_mask of
167 * %GFP_ATOMIC.
168 */
169struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
170 int fclone, int node)
171{
172 struct kmem_cache *cache;
173 struct skb_shared_info *shinfo;
174 struct sk_buff *skb;
175 u8 *data;
176
177 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
178
179 /* Get the HEAD */
180 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
181 if (!skb)
182 goto out;
183 prefetchw(skb);
184
185 /* We do our best to align skb_shared_info on a separate cache
186 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
187 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
188 * Both skb->head and skb_shared_info are cache line aligned.
189 */
190 size = SKB_DATA_ALIGN(size);
191 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
192 data = kmalloc_node_track_caller(size, gfp_mask, node);
193 if (!data)
194 goto nodata;
195 /* kmalloc(size) might give us more room than requested.
196 * Put skb_shared_info exactly at the end of allocated zone,
197 * to allow max possible filling before reallocation.
198 */
199 size = SKB_WITH_OVERHEAD(ksize(data));
200 prefetchw(data + size);
201
202 /*
203 * Only clear those fields we need to clear, not those that we will
204 * actually initialise below. Hence, don't put any more fields after
205 * the tail pointer in struct sk_buff!
206 */
207 memset(skb, 0, offsetof(struct sk_buff, tail));
208 /* Account for allocated memory : skb + skb->head */
209 skb->truesize = SKB_TRUESIZE(size);
210 atomic_set(&skb->users, 1);
211 skb->head = data;
212 skb->data = data;
213 skb_reset_tail_pointer(skb);
214 skb->end = skb->tail + size;
215#ifdef NET_SKBUFF_DATA_USES_OFFSET
216 skb->mac_header = ~0U;
217#endif
218
219 /* make sure we initialize shinfo sequentially */
220 shinfo = skb_shinfo(skb);
221 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
222 atomic_set(&shinfo->dataref, 1);
223 kmemcheck_annotate_variable(shinfo->destructor_arg);
224
225 if (fclone) {
226 struct sk_buff *child = skb + 1;
227 atomic_t *fclone_ref = (atomic_t *) (child + 1);
228
229 kmemcheck_annotate_bitfield(child, flags1);
230 kmemcheck_annotate_bitfield(child, flags2);
231 skb->fclone = SKB_FCLONE_ORIG;
232 atomic_set(fclone_ref, 1);
233
234 child->fclone = SKB_FCLONE_UNAVAILABLE;
235 }
236out:
237 return skb;
238nodata:
239 kmem_cache_free(cache, skb);
240 skb = NULL;
241 goto out;
242}
243EXPORT_SYMBOL(__alloc_skb);
244
245/**
246 * build_skb - build a network buffer
247 * @data: data buffer provided by caller
248 * @frag_size: size of fragment, or 0 if head was kmalloced
249 *
250 * Allocate a new &sk_buff. Caller provides space holding head and
251 * skb_shared_info. @data must have been allocated by kmalloc()
252 * The return is the new skb buffer.
253 * On a failure the return is %NULL, and @data is not freed.
254 * Notes :
255 * Before IO, driver allocates only data buffer where NIC put incoming frame
256 * Driver should add room at head (NET_SKB_PAD) and
257 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
258 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
259 * before giving packet to stack.
260 * RX rings only contains data buffers, not full skbs.
261 */
262struct sk_buff *build_skb(void *data, unsigned int frag_size)
263{
264 struct skb_shared_info *shinfo;
265 struct sk_buff *skb;
266 unsigned int size = frag_size ? : ksize(data);
267
268 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
269 if (!skb)
270 return NULL;
271
272 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
273
274 memset(skb, 0, offsetof(struct sk_buff, tail));
275 skb->truesize = SKB_TRUESIZE(size);
276 skb->head_frag = frag_size != 0;
277 atomic_set(&skb->users, 1);
278 skb->head = data;
279 skb->data = data;
280 skb_reset_tail_pointer(skb);
281 skb->end = skb->tail + size;
282#ifdef NET_SKBUFF_DATA_USES_OFFSET
283 skb->mac_header = ~0U;
284#endif
285
286 /* make sure we initialize shinfo sequentially */
287 shinfo = skb_shinfo(skb);
288 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
289 atomic_set(&shinfo->dataref, 1);
290 kmemcheck_annotate_variable(shinfo->destructor_arg);
291
292 return skb;
293}
294EXPORT_SYMBOL(build_skb);
295
296struct netdev_alloc_cache {
297 struct page *page;
298 unsigned int offset;
299};
300static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
301
302/**
303 * netdev_alloc_frag - allocate a page fragment
304 * @fragsz: fragment size
305 *
306 * Allocates a frag from a page for receive buffer.
307 * Uses GFP_ATOMIC allocations.
308 */
309void *netdev_alloc_frag(unsigned int fragsz)
310{
311 struct netdev_alloc_cache *nc;
312 void *data = NULL;
313 unsigned long flags;
314
315 local_irq_save(flags);
316 nc = &__get_cpu_var(netdev_alloc_cache);
317 if (unlikely(!nc->page)) {
318refill:
319 nc->page = alloc_page(GFP_ATOMIC | __GFP_COLD);
320 nc->offset = 0;
321 }
322 if (likely(nc->page)) {
323 if (nc->offset + fragsz > PAGE_SIZE) {
324 put_page(nc->page);
325 goto refill;
326 }
327 data = page_address(nc->page) + nc->offset;
328 nc->offset += fragsz;
329 get_page(nc->page);
330 }
331 local_irq_restore(flags);
332 return data;
333}
334EXPORT_SYMBOL(netdev_alloc_frag);
335
336/**
337 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
338 * @dev: network device to receive on
339 * @length: length to allocate
340 * @gfp_mask: get_free_pages mask, passed to alloc_skb
341 *
342 * Allocate a new &sk_buff and assign it a usage count of one. The
343 * buffer has unspecified headroom built in. Users should allocate
344 * the headroom they think they need without accounting for the
345 * built in space. The built in space is used for optimisations.
346 *
347 * %NULL is returned if there is no free memory.
348 */
349struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
350 unsigned int length, gfp_t gfp_mask)
351{
352 struct sk_buff *skb = NULL;
353 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
354 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
355
356 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
357 void *data = netdev_alloc_frag(fragsz);
358
359 if (likely(data)) {
360 skb = build_skb(data, fragsz);
361 if (unlikely(!skb))
362 put_page(virt_to_head_page(data));
363 }
364 } else {
365 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
366 }
367 if (likely(skb)) {
368 skb_reserve(skb, NET_SKB_PAD);
369 skb->dev = dev;
370 }
371 return skb;
372}
373EXPORT_SYMBOL(__netdev_alloc_skb);
374
375void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
376 int size, unsigned int truesize)
377{
378 skb_fill_page_desc(skb, i, page, off, size);
379 skb->len += size;
380 skb->data_len += size;
381 skb->truesize += truesize;
382}
383EXPORT_SYMBOL(skb_add_rx_frag);
384
385static void skb_drop_list(struct sk_buff **listp)
386{
387 struct sk_buff *list = *listp;
388
389 *listp = NULL;
390
391 do {
392 struct sk_buff *this = list;
393 list = list->next;
394 kfree_skb(this);
395 } while (list);
396}
397
398static inline void skb_drop_fraglist(struct sk_buff *skb)
399{
400 skb_drop_list(&skb_shinfo(skb)->frag_list);
401}
402
403static void skb_clone_fraglist(struct sk_buff *skb)
404{
405 struct sk_buff *list;
406
407 skb_walk_frags(skb, list)
408 skb_get(list);
409}
410
411static void skb_free_head(struct sk_buff *skb)
412{
413 if (skb->head_frag)
414 put_page(virt_to_head_page(skb->head));
415 else
416 kfree(skb->head);
417}
418
419static void skb_release_data(struct sk_buff *skb)
420{
421 if (!skb->cloned ||
422 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
423 &skb_shinfo(skb)->dataref)) {
424 if (skb_shinfo(skb)->nr_frags) {
425 int i;
426 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
427 skb_frag_unref(skb, i);
428 }
429
430 /*
431 * If skb buf is from userspace, we need to notify the caller
432 * the lower device DMA has done;
433 */
434 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
435 struct ubuf_info *uarg;
436
437 uarg = skb_shinfo(skb)->destructor_arg;
438 if (uarg->callback)
439 uarg->callback(uarg);
440 }
441
442 if (skb_has_frag_list(skb))
443 skb_drop_fraglist(skb);
444
445 skb_free_head(skb);
446 }
447}
448
449/*
450 * Free an skbuff by memory without cleaning the state.
451 */
452static void kfree_skbmem(struct sk_buff *skb)
453{
454 struct sk_buff *other;
455 atomic_t *fclone_ref;
456
457 switch (skb->fclone) {
458 case SKB_FCLONE_UNAVAILABLE:
459 kmem_cache_free(skbuff_head_cache, skb);
460 break;
461
462 case SKB_FCLONE_ORIG:
463 fclone_ref = (atomic_t *) (skb + 2);
464 if (atomic_dec_and_test(fclone_ref))
465 kmem_cache_free(skbuff_fclone_cache, skb);
466 break;
467
468 case SKB_FCLONE_CLONE:
469 fclone_ref = (atomic_t *) (skb + 1);
470 other = skb - 1;
471
472 /* The clone portion is available for
473 * fast-cloning again.
474 */
475 skb->fclone = SKB_FCLONE_UNAVAILABLE;
476
477 if (atomic_dec_and_test(fclone_ref))
478 kmem_cache_free(skbuff_fclone_cache, other);
479 break;
480 }
481}
482
483static void skb_release_head_state(struct sk_buff *skb)
484{
485 skb_dst_drop(skb);
486#ifdef CONFIG_XFRM
487 secpath_put(skb->sp);
488#endif
489 if (skb->destructor) {
490 WARN_ON(in_irq());
491 skb->destructor(skb);
492 }
493#if IS_ENABLED(CONFIG_NF_CONNTRACK)
494 nf_conntrack_put(skb->nfct);
495#endif
496#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
497 nf_conntrack_put_reasm(skb->nfct_reasm);
498#endif
499#ifdef CONFIG_BRIDGE_NETFILTER
500 nf_bridge_put(skb->nf_bridge);
501#endif
502/* XXX: IS this still necessary? - JHS */
503#ifdef CONFIG_NET_SCHED
504 skb->tc_index = 0;
505#ifdef CONFIG_NET_CLS_ACT
506 skb->tc_verd = 0;
507#endif
508#endif
509}
510
511/* Free everything but the sk_buff shell. */
512static void skb_release_all(struct sk_buff *skb)
513{
514 skb_release_head_state(skb);
515 skb_release_data(skb);
516}
517
518/**
519 * __kfree_skb - private function
520 * @skb: buffer
521 *
522 * Free an sk_buff. Release anything attached to the buffer.
523 * Clean the state. This is an internal helper function. Users should
524 * always call kfree_skb
525 */
526
527void __kfree_skb(struct sk_buff *skb)
528{
529 skb_release_all(skb);
530 kfree_skbmem(skb);
531}
532EXPORT_SYMBOL(__kfree_skb);
533
534/**
535 * kfree_skb - free an sk_buff
536 * @skb: buffer to free
537 *
538 * Drop a reference to the buffer and free it if the usage count has
539 * hit zero.
540 */
541void kfree_skb(struct sk_buff *skb)
542{
543 if (unlikely(!skb))
544 return;
545 if (likely(atomic_read(&skb->users) == 1))
546 smp_rmb();
547 else if (likely(!atomic_dec_and_test(&skb->users)))
548 return;
549 trace_kfree_skb(skb, __builtin_return_address(0));
550 __kfree_skb(skb);
551}
552EXPORT_SYMBOL(kfree_skb);
553
554/**
555 * consume_skb - free an skbuff
556 * @skb: buffer to free
557 *
558 * Drop a ref to the buffer and free it if the usage count has hit zero
559 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
560 * is being dropped after a failure and notes that
561 */
562void consume_skb(struct sk_buff *skb)
563{
564 if (unlikely(!skb))
565 return;
566 if (likely(atomic_read(&skb->users) == 1))
567 smp_rmb();
568 else if (likely(!atomic_dec_and_test(&skb->users)))
569 return;
570 trace_consume_skb(skb);
571 __kfree_skb(skb);
572}
573EXPORT_SYMBOL(consume_skb);
574
575/**
576 * skb_recycle - clean up an skb for reuse
577 * @skb: buffer
578 *
579 * Recycles the skb to be reused as a receive buffer. This
580 * function does any necessary reference count dropping, and
581 * cleans up the skbuff as if it just came from __alloc_skb().
582 */
583void skb_recycle(struct sk_buff *skb)
584{
585 struct skb_shared_info *shinfo;
586
587 skb_release_head_state(skb);
588
589 shinfo = skb_shinfo(skb);
590 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
591 atomic_set(&shinfo->dataref, 1);
592
593 memset(skb, 0, offsetof(struct sk_buff, tail));
594 skb->data = skb->head + NET_SKB_PAD;
595 skb_reset_tail_pointer(skb);
596}
597EXPORT_SYMBOL(skb_recycle);
598
599/**
600 * skb_recycle_check - check if skb can be reused for receive
601 * @skb: buffer
602 * @skb_size: minimum receive buffer size
603 *
604 * Checks that the skb passed in is not shared or cloned, and
605 * that it is linear and its head portion at least as large as
606 * skb_size so that it can be recycled as a receive buffer.
607 * If these conditions are met, this function does any necessary
608 * reference count dropping and cleans up the skbuff as if it
609 * just came from __alloc_skb().
610 */
611bool skb_recycle_check(struct sk_buff *skb, int skb_size)
612{
613 if (!skb_is_recycleable(skb, skb_size))
614 return false;
615
616 skb_recycle(skb);
617
618 return true;
619}
620EXPORT_SYMBOL(skb_recycle_check);
621
622static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
623{
624 new->tstamp = old->tstamp;
625 new->dev = old->dev;
626 new->transport_header = old->transport_header;
627 new->network_header = old->network_header;
628 new->mac_header = old->mac_header;
629 skb_dst_copy(new, old);
630 new->rxhash = old->rxhash;
631 new->ooo_okay = old->ooo_okay;
632 new->l4_rxhash = old->l4_rxhash;
633 new->no_fcs = old->no_fcs;
634#ifdef CONFIG_XFRM
635 new->sp = secpath_get(old->sp);
636#endif
637 memcpy(new->cb, old->cb, sizeof(old->cb));
638 new->csum = old->csum;
639 new->local_df = old->local_df;
640 new->pkt_type = old->pkt_type;
641 new->ip_summed = old->ip_summed;
642 skb_copy_queue_mapping(new, old);
643 new->priority = old->priority;
644#if IS_ENABLED(CONFIG_IP_VS)
645 new->ipvs_property = old->ipvs_property;
646#endif
647 new->protocol = old->protocol;
648 new->mark = old->mark;
649 new->skb_iif = old->skb_iif;
650 __nf_copy(new, old);
651#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
652 new->nf_trace = old->nf_trace;
653#endif
654#ifdef CONFIG_NET_SCHED
655 new->tc_index = old->tc_index;
656#ifdef CONFIG_NET_CLS_ACT
657 new->tc_verd = old->tc_verd;
658#endif
659#endif
660 new->vlan_tci = old->vlan_tci;
661
662 skb_copy_secmark(new, old);
663}
664
665/*
666 * You should not add any new code to this function. Add it to
667 * __copy_skb_header above instead.
668 */
669static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
670{
671#define C(x) n->x = skb->x
672
673 n->next = n->prev = NULL;
674 n->sk = NULL;
675 __copy_skb_header(n, skb);
676
677 C(len);
678 C(data_len);
679 C(mac_len);
680 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
681 n->cloned = 1;
682 n->nohdr = 0;
683 n->destructor = NULL;
684 C(tail);
685 C(end);
686 C(head);
687 C(head_frag);
688 C(data);
689 C(truesize);
690 atomic_set(&n->users, 1);
691
692 atomic_inc(&(skb_shinfo(skb)->dataref));
693 skb->cloned = 1;
694
695 return n;
696#undef C
697}
698
699/**
700 * skb_morph - morph one skb into another
701 * @dst: the skb to receive the contents
702 * @src: the skb to supply the contents
703 *
704 * This is identical to skb_clone except that the target skb is
705 * supplied by the user.
706 *
707 * The target skb is returned upon exit.
708 */
709struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
710{
711 skb_release_all(dst);
712 return __skb_clone(dst, src);
713}
714EXPORT_SYMBOL_GPL(skb_morph);
715
716/* skb_copy_ubufs - copy userspace skb frags buffers to kernel
717 * @skb: the skb to modify
718 * @gfp_mask: allocation priority
719 *
720 * This must be called on SKBTX_DEV_ZEROCOPY skb.
721 * It will copy all frags into kernel and drop the reference
722 * to userspace pages.
723 *
724 * If this function is called from an interrupt gfp_mask() must be
725 * %GFP_ATOMIC.
726 *
727 * Returns 0 on success or a negative error code on failure
728 * to allocate kernel memory to copy to.
729 */
730int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
731{
732 int i;
733 int num_frags = skb_shinfo(skb)->nr_frags;
734 struct page *page, *head = NULL;
735 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
736
737 for (i = 0; i < num_frags; i++) {
738 u8 *vaddr;
739 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
740
741 page = alloc_page(GFP_ATOMIC);
742 if (!page) {
743 while (head) {
744 struct page *next = (struct page *)head->private;
745 put_page(head);
746 head = next;
747 }
748 return -ENOMEM;
749 }
750 vaddr = kmap_atomic(skb_frag_page(f));
751 memcpy(page_address(page),
752 vaddr + f->page_offset, skb_frag_size(f));
753 kunmap_atomic(vaddr);
754 page->private = (unsigned long)head;
755 head = page;
756 }
757
758 /* skb frags release userspace buffers */
759 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
760 skb_frag_unref(skb, i);
761
762 uarg->callback(uarg);
763
764 /* skb frags point to kernel buffers */
765 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
766 __skb_fill_page_desc(skb, i-1, head, 0,
767 skb_shinfo(skb)->frags[i - 1].size);
768 head = (struct page *)head->private;
769 }
770
771 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
772 return 0;
773}
774
775
776/**
777 * skb_clone - duplicate an sk_buff
778 * @skb: buffer to clone
779 * @gfp_mask: allocation priority
780 *
781 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
782 * copies share the same packet data but not structure. The new
783 * buffer has a reference count of 1. If the allocation fails the
784 * function returns %NULL otherwise the new buffer is returned.
785 *
786 * If this function is called from an interrupt gfp_mask() must be
787 * %GFP_ATOMIC.
788 */
789
790struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
791{
792 struct sk_buff *n;
793
794 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
795 if (skb_copy_ubufs(skb, gfp_mask))
796 return NULL;
797 }
798
799 n = skb + 1;
800 if (skb->fclone == SKB_FCLONE_ORIG &&
801 n->fclone == SKB_FCLONE_UNAVAILABLE) {
802 atomic_t *fclone_ref = (atomic_t *) (n + 1);
803 n->fclone = SKB_FCLONE_CLONE;
804 atomic_inc(fclone_ref);
805 } else {
806 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
807 if (!n)
808 return NULL;
809
810 kmemcheck_annotate_bitfield(n, flags1);
811 kmemcheck_annotate_bitfield(n, flags2);
812 n->fclone = SKB_FCLONE_UNAVAILABLE;
813 }
814
815 return __skb_clone(n, skb);
816}
817EXPORT_SYMBOL(skb_clone);
818
819static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
820{
821#ifndef NET_SKBUFF_DATA_USES_OFFSET
822 /*
823 * Shift between the two data areas in bytes
824 */
825 unsigned long offset = new->data - old->data;
826#endif
827
828 __copy_skb_header(new, old);
829
830#ifndef NET_SKBUFF_DATA_USES_OFFSET
831 /* {transport,network,mac}_header are relative to skb->head */
832 new->transport_header += offset;
833 new->network_header += offset;
834 if (skb_mac_header_was_set(new))
835 new->mac_header += offset;
836#endif
837 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
838 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
839 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
840}
841
842/**
843 * skb_copy - create private copy of an sk_buff
844 * @skb: buffer to copy
845 * @gfp_mask: allocation priority
846 *
847 * Make a copy of both an &sk_buff and its data. This is used when the
848 * caller wishes to modify the data and needs a private copy of the
849 * data to alter. Returns %NULL on failure or the pointer to the buffer
850 * on success. The returned buffer has a reference count of 1.
851 *
852 * As by-product this function converts non-linear &sk_buff to linear
853 * one, so that &sk_buff becomes completely private and caller is allowed
854 * to modify all the data of returned buffer. This means that this
855 * function is not recommended for use in circumstances when only
856 * header is going to be modified. Use pskb_copy() instead.
857 */
858
859struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
860{
861 int headerlen = skb_headroom(skb);
862 unsigned int size = skb_end_offset(skb) + skb->data_len;
863 struct sk_buff *n = alloc_skb(size, gfp_mask);
864
865 if (!n)
866 return NULL;
867
868 /* Set the data pointer */
869 skb_reserve(n, headerlen);
870 /* Set the tail pointer and length */
871 skb_put(n, skb->len);
872
873 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
874 BUG();
875
876 copy_skb_header(n, skb);
877 return n;
878}
879EXPORT_SYMBOL(skb_copy);
880
881/**
882 * __pskb_copy - create copy of an sk_buff with private head.
883 * @skb: buffer to copy
884 * @headroom: headroom of new skb
885 * @gfp_mask: allocation priority
886 *
887 * Make a copy of both an &sk_buff and part of its data, located
888 * in header. Fragmented data remain shared. This is used when
889 * the caller wishes to modify only header of &sk_buff and needs
890 * private copy of the header to alter. Returns %NULL on failure
891 * or the pointer to the buffer on success.
892 * The returned buffer has a reference count of 1.
893 */
894
895struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
896{
897 unsigned int size = skb_headlen(skb) + headroom;
898 struct sk_buff *n = alloc_skb(size, gfp_mask);
899
900 if (!n)
901 goto out;
902
903 /* Set the data pointer */
904 skb_reserve(n, headroom);
905 /* Set the tail pointer and length */
906 skb_put(n, skb_headlen(skb));
907 /* Copy the bytes */
908 skb_copy_from_linear_data(skb, n->data, n->len);
909
910 n->truesize += skb->data_len;
911 n->data_len = skb->data_len;
912 n->len = skb->len;
913
914 if (skb_shinfo(skb)->nr_frags) {
915 int i;
916
917 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
918 if (skb_copy_ubufs(skb, gfp_mask)) {
919 kfree_skb(n);
920 n = NULL;
921 goto out;
922 }
923 }
924 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
925 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
926 skb_frag_ref(skb, i);
927 }
928 skb_shinfo(n)->nr_frags = i;
929 }
930
931 if (skb_has_frag_list(skb)) {
932 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
933 skb_clone_fraglist(n);
934 }
935
936 copy_skb_header(n, skb);
937out:
938 return n;
939}
940EXPORT_SYMBOL(__pskb_copy);
941
942/**
943 * pskb_expand_head - reallocate header of &sk_buff
944 * @skb: buffer to reallocate
945 * @nhead: room to add at head
946 * @ntail: room to add at tail
947 * @gfp_mask: allocation priority
948 *
949 * Expands (or creates identical copy, if &nhead and &ntail are zero)
950 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
951 * reference count of 1. Returns zero in the case of success or error,
952 * if expansion failed. In the last case, &sk_buff is not changed.
953 *
954 * All the pointers pointing into skb header may change and must be
955 * reloaded after call to this function.
956 */
957
958int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
959 gfp_t gfp_mask)
960{
961 int i;
962 u8 *data;
963 int size = nhead + skb_end_offset(skb) + ntail;
964 long off;
965
966 BUG_ON(nhead < 0);
967
968 if (skb_shared(skb))
969 BUG();
970
971 size = SKB_DATA_ALIGN(size);
972
973 data = kmalloc(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
974 gfp_mask);
975 if (!data)
976 goto nodata;
977 size = SKB_WITH_OVERHEAD(ksize(data));
978
979 /* Copy only real data... and, alas, header. This should be
980 * optimized for the cases when header is void.
981 */
982 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
983
984 memcpy((struct skb_shared_info *)(data + size),
985 skb_shinfo(skb),
986 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
987
988 /*
989 * if shinfo is shared we must drop the old head gracefully, but if it
990 * is not we can just drop the old head and let the existing refcount
991 * be since all we did is relocate the values
992 */
993 if (skb_cloned(skb)) {
994 /* copy this zero copy skb frags */
995 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
996 if (skb_copy_ubufs(skb, gfp_mask))
997 goto nofrags;
998 }
999 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1000 skb_frag_ref(skb, i);
1001
1002 if (skb_has_frag_list(skb))
1003 skb_clone_fraglist(skb);
1004
1005 skb_release_data(skb);
1006 } else {
1007 skb_free_head(skb);
1008 }
1009 off = (data + nhead) - skb->head;
1010
1011 skb->head = data;
1012 skb->head_frag = 0;
1013 skb->data += off;
1014#ifdef NET_SKBUFF_DATA_USES_OFFSET
1015 skb->end = size;
1016 off = nhead;
1017#else
1018 skb->end = skb->head + size;
1019#endif
1020 /* {transport,network,mac}_header and tail are relative to skb->head */
1021 skb->tail += off;
1022 skb->transport_header += off;
1023 skb->network_header += off;
1024 if (skb_mac_header_was_set(skb))
1025 skb->mac_header += off;
1026 /* Only adjust this if it actually is csum_start rather than csum */
1027 if (skb->ip_summed == CHECKSUM_PARTIAL)
1028 skb->csum_start += nhead;
1029 skb->cloned = 0;
1030 skb->hdr_len = 0;
1031 skb->nohdr = 0;
1032 atomic_set(&skb_shinfo(skb)->dataref, 1);
1033 return 0;
1034
1035nofrags:
1036 kfree(data);
1037nodata:
1038 return -ENOMEM;
1039}
1040EXPORT_SYMBOL(pskb_expand_head);
1041
1042/* Make private copy of skb with writable head and some headroom */
1043
1044struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1045{
1046 struct sk_buff *skb2;
1047 int delta = headroom - skb_headroom(skb);
1048
1049 if (delta <= 0)
1050 skb2 = pskb_copy(skb, GFP_ATOMIC);
1051 else {
1052 skb2 = skb_clone(skb, GFP_ATOMIC);
1053 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1054 GFP_ATOMIC)) {
1055 kfree_skb(skb2);
1056 skb2 = NULL;
1057 }
1058 }
1059 return skb2;
1060}
1061EXPORT_SYMBOL(skb_realloc_headroom);
1062
1063/**
1064 * skb_copy_expand - copy and expand sk_buff
1065 * @skb: buffer to copy
1066 * @newheadroom: new free bytes at head
1067 * @newtailroom: new free bytes at tail
1068 * @gfp_mask: allocation priority
1069 *
1070 * Make a copy of both an &sk_buff and its data and while doing so
1071 * allocate additional space.
1072 *
1073 * This is used when the caller wishes to modify the data and needs a
1074 * private copy of the data to alter as well as more space for new fields.
1075 * Returns %NULL on failure or the pointer to the buffer
1076 * on success. The returned buffer has a reference count of 1.
1077 *
1078 * You must pass %GFP_ATOMIC as the allocation priority if this function
1079 * is called from an interrupt.
1080 */
1081struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1082 int newheadroom, int newtailroom,
1083 gfp_t gfp_mask)
1084{
1085 /*
1086 * Allocate the copy buffer
1087 */
1088 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1089 gfp_mask);
1090 int oldheadroom = skb_headroom(skb);
1091 int head_copy_len, head_copy_off;
1092 int off;
1093
1094 if (!n)
1095 return NULL;
1096
1097 skb_reserve(n, newheadroom);
1098
1099 /* Set the tail pointer and length */
1100 skb_put(n, skb->len);
1101
1102 head_copy_len = oldheadroom;
1103 head_copy_off = 0;
1104 if (newheadroom <= head_copy_len)
1105 head_copy_len = newheadroom;
1106 else
1107 head_copy_off = newheadroom - head_copy_len;
1108
1109 /* Copy the linear header and data. */
1110 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1111 skb->len + head_copy_len))
1112 BUG();
1113
1114 copy_skb_header(n, skb);
1115
1116 off = newheadroom - oldheadroom;
1117 if (n->ip_summed == CHECKSUM_PARTIAL)
1118 n->csum_start += off;
1119#ifdef NET_SKBUFF_DATA_USES_OFFSET
1120 n->transport_header += off;
1121 n->network_header += off;
1122 if (skb_mac_header_was_set(skb))
1123 n->mac_header += off;
1124#endif
1125
1126 return n;
1127}
1128EXPORT_SYMBOL(skb_copy_expand);
1129
1130/**
1131 * skb_pad - zero pad the tail of an skb
1132 * @skb: buffer to pad
1133 * @pad: space to pad
1134 *
1135 * Ensure that a buffer is followed by a padding area that is zero
1136 * filled. Used by network drivers which may DMA or transfer data
1137 * beyond the buffer end onto the wire.
1138 *
1139 * May return error in out of memory cases. The skb is freed on error.
1140 */
1141
1142int skb_pad(struct sk_buff *skb, int pad)
1143{
1144 int err;
1145 int ntail;
1146
1147 /* If the skbuff is non linear tailroom is always zero.. */
1148 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1149 memset(skb->data+skb->len, 0, pad);
1150 return 0;
1151 }
1152
1153 ntail = skb->data_len + pad - (skb->end - skb->tail);
1154 if (likely(skb_cloned(skb) || ntail > 0)) {
1155 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1156 if (unlikely(err))
1157 goto free_skb;
1158 }
1159
1160 /* FIXME: The use of this function with non-linear skb's really needs
1161 * to be audited.
1162 */
1163 err = skb_linearize(skb);
1164 if (unlikely(err))
1165 goto free_skb;
1166
1167 memset(skb->data + skb->len, 0, pad);
1168 return 0;
1169
1170free_skb:
1171 kfree_skb(skb);
1172 return err;
1173}
1174EXPORT_SYMBOL(skb_pad);
1175
1176/**
1177 * skb_put - add data to a buffer
1178 * @skb: buffer to use
1179 * @len: amount of data to add
1180 *
1181 * This function extends the used data area of the buffer. If this would
1182 * exceed the total buffer size the kernel will panic. A pointer to the
1183 * first byte of the extra data is returned.
1184 */
1185unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1186{
1187 unsigned char *tmp = skb_tail_pointer(skb);
1188 SKB_LINEAR_ASSERT(skb);
1189 skb->tail += len;
1190 skb->len += len;
1191 if (unlikely(skb->tail > skb->end))
1192 skb_over_panic(skb, len, __builtin_return_address(0));
1193 return tmp;
1194}
1195EXPORT_SYMBOL(skb_put);
1196
1197/**
1198 * skb_push - add data to the start of a buffer
1199 * @skb: buffer to use
1200 * @len: amount of data to add
1201 *
1202 * This function extends the used data area of the buffer at the buffer
1203 * start. If this would exceed the total buffer headroom the kernel will
1204 * panic. A pointer to the first byte of the extra data is returned.
1205 */
1206unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1207{
1208 skb->data -= len;
1209 skb->len += len;
1210 if (unlikely(skb->data<skb->head))
1211 skb_under_panic(skb, len, __builtin_return_address(0));
1212 return skb->data;
1213}
1214EXPORT_SYMBOL(skb_push);
1215
1216/**
1217 * skb_pull - remove data from the start of a buffer
1218 * @skb: buffer to use
1219 * @len: amount of data to remove
1220 *
1221 * This function removes data from the start of a buffer, returning
1222 * the memory to the headroom. A pointer to the next data in the buffer
1223 * is returned. Once the data has been pulled future pushes will overwrite
1224 * the old data.
1225 */
1226unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1227{
1228 return skb_pull_inline(skb, len);
1229}
1230EXPORT_SYMBOL(skb_pull);
1231
1232/**
1233 * skb_trim - remove end from a buffer
1234 * @skb: buffer to alter
1235 * @len: new length
1236 *
1237 * Cut the length of a buffer down by removing data from the tail. If
1238 * the buffer is already under the length specified it is not modified.
1239 * The skb must be linear.
1240 */
1241void skb_trim(struct sk_buff *skb, unsigned int len)
1242{
1243 if (skb->len > len)
1244 __skb_trim(skb, len);
1245}
1246EXPORT_SYMBOL(skb_trim);
1247
1248/* Trims skb to length len. It can change skb pointers.
1249 */
1250
1251int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1252{
1253 struct sk_buff **fragp;
1254 struct sk_buff *frag;
1255 int offset = skb_headlen(skb);
1256 int nfrags = skb_shinfo(skb)->nr_frags;
1257 int i;
1258 int err;
1259
1260 if (skb_cloned(skb) &&
1261 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1262 return err;
1263
1264 i = 0;
1265 if (offset >= len)
1266 goto drop_pages;
1267
1268 for (; i < nfrags; i++) {
1269 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1270
1271 if (end < len) {
1272 offset = end;
1273 continue;
1274 }
1275
1276 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1277
1278drop_pages:
1279 skb_shinfo(skb)->nr_frags = i;
1280
1281 for (; i < nfrags; i++)
1282 skb_frag_unref(skb, i);
1283
1284 if (skb_has_frag_list(skb))
1285 skb_drop_fraglist(skb);
1286 goto done;
1287 }
1288
1289 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1290 fragp = &frag->next) {
1291 int end = offset + frag->len;
1292
1293 if (skb_shared(frag)) {
1294 struct sk_buff *nfrag;
1295
1296 nfrag = skb_clone(frag, GFP_ATOMIC);
1297 if (unlikely(!nfrag))
1298 return -ENOMEM;
1299
1300 nfrag->next = frag->next;
1301 consume_skb(frag);
1302 frag = nfrag;
1303 *fragp = frag;
1304 }
1305
1306 if (end < len) {
1307 offset = end;
1308 continue;
1309 }
1310
1311 if (end > len &&
1312 unlikely((err = pskb_trim(frag, len - offset))))
1313 return err;
1314
1315 if (frag->next)
1316 skb_drop_list(&frag->next);
1317 break;
1318 }
1319
1320done:
1321 if (len > skb_headlen(skb)) {
1322 skb->data_len -= skb->len - len;
1323 skb->len = len;
1324 } else {
1325 skb->len = len;
1326 skb->data_len = 0;
1327 skb_set_tail_pointer(skb, len);
1328 }
1329
1330 return 0;
1331}
1332EXPORT_SYMBOL(___pskb_trim);
1333
1334/**
1335 * __pskb_pull_tail - advance tail of skb header
1336 * @skb: buffer to reallocate
1337 * @delta: number of bytes to advance tail
1338 *
1339 * The function makes a sense only on a fragmented &sk_buff,
1340 * it expands header moving its tail forward and copying necessary
1341 * data from fragmented part.
1342 *
1343 * &sk_buff MUST have reference count of 1.
1344 *
1345 * Returns %NULL (and &sk_buff does not change) if pull failed
1346 * or value of new tail of skb in the case of success.
1347 *
1348 * All the pointers pointing into skb header may change and must be
1349 * reloaded after call to this function.
1350 */
1351
1352/* Moves tail of skb head forward, copying data from fragmented part,
1353 * when it is necessary.
1354 * 1. It may fail due to malloc failure.
1355 * 2. It may change skb pointers.
1356 *
1357 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1358 */
1359unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1360{
1361 /* If skb has not enough free space at tail, get new one
1362 * plus 128 bytes for future expansions. If we have enough
1363 * room at tail, reallocate without expansion only if skb is cloned.
1364 */
1365 int i, k, eat = (skb->tail + delta) - skb->end;
1366
1367 if (eat > 0 || skb_cloned(skb)) {
1368 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1369 GFP_ATOMIC))
1370 return NULL;
1371 }
1372
1373 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1374 BUG();
1375
1376 /* Optimization: no fragments, no reasons to preestimate
1377 * size of pulled pages. Superb.
1378 */
1379 if (!skb_has_frag_list(skb))
1380 goto pull_pages;
1381
1382 /* Estimate size of pulled pages. */
1383 eat = delta;
1384 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1385 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1386
1387 if (size >= eat)
1388 goto pull_pages;
1389 eat -= size;
1390 }
1391
1392 /* If we need update frag list, we are in troubles.
1393 * Certainly, it possible to add an offset to skb data,
1394 * but taking into account that pulling is expected to
1395 * be very rare operation, it is worth to fight against
1396 * further bloating skb head and crucify ourselves here instead.
1397 * Pure masohism, indeed. 8)8)
1398 */
1399 if (eat) {
1400 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1401 struct sk_buff *clone = NULL;
1402 struct sk_buff *insp = NULL;
1403
1404 do {
1405 BUG_ON(!list);
1406
1407 if (list->len <= eat) {
1408 /* Eaten as whole. */
1409 eat -= list->len;
1410 list = list->next;
1411 insp = list;
1412 } else {
1413 /* Eaten partially. */
1414
1415 if (skb_shared(list)) {
1416 /* Sucks! We need to fork list. :-( */
1417 clone = skb_clone(list, GFP_ATOMIC);
1418 if (!clone)
1419 return NULL;
1420 insp = list->next;
1421 list = clone;
1422 } else {
1423 /* This may be pulled without
1424 * problems. */
1425 insp = list;
1426 }
1427 if (!pskb_pull(list, eat)) {
1428 kfree_skb(clone);
1429 return NULL;
1430 }
1431 break;
1432 }
1433 } while (eat);
1434
1435 /* Free pulled out fragments. */
1436 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1437 skb_shinfo(skb)->frag_list = list->next;
1438 kfree_skb(list);
1439 }
1440 /* And insert new clone at head. */
1441 if (clone) {
1442 clone->next = list;
1443 skb_shinfo(skb)->frag_list = clone;
1444 }
1445 }
1446 /* Success! Now we may commit changes to skb data. */
1447
1448pull_pages:
1449 eat = delta;
1450 k = 0;
1451 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1452 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1453
1454 if (size <= eat) {
1455 skb_frag_unref(skb, i);
1456 eat -= size;
1457 } else {
1458 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1459 if (eat) {
1460 skb_shinfo(skb)->frags[k].page_offset += eat;
1461 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1462 eat = 0;
1463 }
1464 k++;
1465 }
1466 }
1467 skb_shinfo(skb)->nr_frags = k;
1468
1469 skb->tail += delta;
1470 skb->data_len -= delta;
1471
1472 return skb_tail_pointer(skb);
1473}
1474EXPORT_SYMBOL(__pskb_pull_tail);
1475
1476/**
1477 * skb_copy_bits - copy bits from skb to kernel buffer
1478 * @skb: source skb
1479 * @offset: offset in source
1480 * @to: destination buffer
1481 * @len: number of bytes to copy
1482 *
1483 * Copy the specified number of bytes from the source skb to the
1484 * destination buffer.
1485 *
1486 * CAUTION ! :
1487 * If its prototype is ever changed,
1488 * check arch/{*}/net/{*}.S files,
1489 * since it is called from BPF assembly code.
1490 */
1491int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1492{
1493 int start = skb_headlen(skb);
1494 struct sk_buff *frag_iter;
1495 int i, copy;
1496
1497 if (offset > (int)skb->len - len)
1498 goto fault;
1499
1500 /* Copy header. */
1501 if ((copy = start - offset) > 0) {
1502 if (copy > len)
1503 copy = len;
1504 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1505 if ((len -= copy) == 0)
1506 return 0;
1507 offset += copy;
1508 to += copy;
1509 }
1510
1511 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1512 int end;
1513 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1514
1515 WARN_ON(start > offset + len);
1516
1517 end = start + skb_frag_size(f);
1518 if ((copy = end - offset) > 0) {
1519 u8 *vaddr;
1520
1521 if (copy > len)
1522 copy = len;
1523
1524 vaddr = kmap_atomic(skb_frag_page(f));
1525 memcpy(to,
1526 vaddr + f->page_offset + offset - start,
1527 copy);
1528 kunmap_atomic(vaddr);
1529
1530 if ((len -= copy) == 0)
1531 return 0;
1532 offset += copy;
1533 to += copy;
1534 }
1535 start = end;
1536 }
1537
1538 skb_walk_frags(skb, frag_iter) {
1539 int end;
1540
1541 WARN_ON(start > offset + len);
1542
1543 end = start + frag_iter->len;
1544 if ((copy = end - offset) > 0) {
1545 if (copy > len)
1546 copy = len;
1547 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1548 goto fault;
1549 if ((len -= copy) == 0)
1550 return 0;
1551 offset += copy;
1552 to += copy;
1553 }
1554 start = end;
1555 }
1556
1557 if (!len)
1558 return 0;
1559
1560fault:
1561 return -EFAULT;
1562}
1563EXPORT_SYMBOL(skb_copy_bits);
1564
1565/*
1566 * Callback from splice_to_pipe(), if we need to release some pages
1567 * at the end of the spd in case we error'ed out in filling the pipe.
1568 */
1569static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1570{
1571 put_page(spd->pages[i]);
1572}
1573
1574static struct page *linear_to_page(struct page *page, unsigned int *len,
1575 unsigned int *offset,
1576 struct sk_buff *skb, struct sock *sk)
1577{
1578 struct page *p = sk->sk_sndmsg_page;
1579 unsigned int off;
1580
1581 if (!p) {
1582new_page:
1583 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1584 if (!p)
1585 return NULL;
1586
1587 off = sk->sk_sndmsg_off = 0;
1588 /* hold one ref to this page until it's full */
1589 } else {
1590 unsigned int mlen;
1591
1592 /* If we are the only user of the page, we can reset offset */
1593 if (page_count(p) == 1)
1594 sk->sk_sndmsg_off = 0;
1595 off = sk->sk_sndmsg_off;
1596 mlen = PAGE_SIZE - off;
1597 if (mlen < 64 && mlen < *len) {
1598 put_page(p);
1599 goto new_page;
1600 }
1601
1602 *len = min_t(unsigned int, *len, mlen);
1603 }
1604
1605 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1606 sk->sk_sndmsg_off += *len;
1607 *offset = off;
1608
1609 return p;
1610}
1611
1612static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1613 struct page *page,
1614 unsigned int offset)
1615{
1616 return spd->nr_pages &&
1617 spd->pages[spd->nr_pages - 1] == page &&
1618 (spd->partial[spd->nr_pages - 1].offset +
1619 spd->partial[spd->nr_pages - 1].len == offset);
1620}
1621
1622/*
1623 * Fill page/offset/length into spd, if it can hold more pages.
1624 */
1625static bool spd_fill_page(struct splice_pipe_desc *spd,
1626 struct pipe_inode_info *pipe, struct page *page,
1627 unsigned int *len, unsigned int offset,
1628 struct sk_buff *skb, bool linear,
1629 struct sock *sk)
1630{
1631 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1632 return true;
1633
1634 if (linear) {
1635 page = linear_to_page(page, len, &offset, skb, sk);
1636 if (!page)
1637 return true;
1638 }
1639 if (spd_can_coalesce(spd, page, offset)) {
1640 spd->partial[spd->nr_pages - 1].len += *len;
1641 return false;
1642 }
1643 get_page(page);
1644 spd->pages[spd->nr_pages] = page;
1645 spd->partial[spd->nr_pages].len = *len;
1646 spd->partial[spd->nr_pages].offset = offset;
1647 spd->nr_pages++;
1648
1649 return false;
1650}
1651
1652static inline void __segment_seek(struct page **page, unsigned int *poff,
1653 unsigned int *plen, unsigned int off)
1654{
1655 unsigned long n;
1656
1657 *poff += off;
1658 n = *poff / PAGE_SIZE;
1659 if (n)
1660 *page = nth_page(*page, n);
1661
1662 *poff = *poff % PAGE_SIZE;
1663 *plen -= off;
1664}
1665
1666static bool __splice_segment(struct page *page, unsigned int poff,
1667 unsigned int plen, unsigned int *off,
1668 unsigned int *len, struct sk_buff *skb,
1669 struct splice_pipe_desc *spd, bool linear,
1670 struct sock *sk,
1671 struct pipe_inode_info *pipe)
1672{
1673 if (!*len)
1674 return true;
1675
1676 /* skip this segment if already processed */
1677 if (*off >= plen) {
1678 *off -= plen;
1679 return false;
1680 }
1681
1682 /* ignore any bits we already processed */
1683 if (*off) {
1684 __segment_seek(&page, &poff, &plen, *off);
1685 *off = 0;
1686 }
1687
1688 do {
1689 unsigned int flen = min(*len, plen);
1690
1691 /* the linear region may spread across several pages */
1692 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1693
1694 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1695 return true;
1696
1697 __segment_seek(&page, &poff, &plen, flen);
1698 *len -= flen;
1699
1700 } while (*len && plen);
1701
1702 return false;
1703}
1704
1705/*
1706 * Map linear and fragment data from the skb to spd. It reports true if the
1707 * pipe is full or if we already spliced the requested length.
1708 */
1709static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1710 unsigned int *offset, unsigned int *len,
1711 struct splice_pipe_desc *spd, struct sock *sk)
1712{
1713 int seg;
1714
1715 /* map the linear part :
1716 * If skb->head_frag is set, this 'linear' part is backed by a
1717 * fragment, and if the head is not shared with any clones then
1718 * we can avoid a copy since we own the head portion of this page.
1719 */
1720 if (__splice_segment(virt_to_page(skb->data),
1721 (unsigned long) skb->data & (PAGE_SIZE - 1),
1722 skb_headlen(skb),
1723 offset, len, skb, spd,
1724 skb_head_is_locked(skb),
1725 sk, pipe))
1726 return true;
1727
1728 /*
1729 * then map the fragments
1730 */
1731 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1732 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1733
1734 if (__splice_segment(skb_frag_page(f),
1735 f->page_offset, skb_frag_size(f),
1736 offset, len, skb, spd, false, sk, pipe))
1737 return true;
1738 }
1739
1740 return false;
1741}
1742
1743/*
1744 * Map data from the skb to a pipe. Should handle both the linear part,
1745 * the fragments, and the frag list. It does NOT handle frag lists within
1746 * the frag list, if such a thing exists. We'd probably need to recurse to
1747 * handle that cleanly.
1748 */
1749int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1750 struct pipe_inode_info *pipe, unsigned int tlen,
1751 unsigned int flags)
1752{
1753 struct partial_page partial[MAX_SKB_FRAGS];
1754 struct page *pages[MAX_SKB_FRAGS];
1755 struct splice_pipe_desc spd = {
1756 .pages = pages,
1757 .partial = partial,
1758 .nr_pages_max = MAX_SKB_FRAGS,
1759 .flags = flags,
1760 .ops = &sock_pipe_buf_ops,
1761 .spd_release = sock_spd_release,
1762 };
1763 struct sk_buff *frag_iter;
1764 struct sock *sk = skb->sk;
1765 int ret = 0;
1766
1767 /*
1768 * __skb_splice_bits() only fails if the output has no room left,
1769 * so no point in going over the frag_list for the error case.
1770 */
1771 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1772 goto done;
1773 else if (!tlen)
1774 goto done;
1775
1776 /*
1777 * now see if we have a frag_list to map
1778 */
1779 skb_walk_frags(skb, frag_iter) {
1780 if (!tlen)
1781 break;
1782 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1783 break;
1784 }
1785
1786done:
1787 if (spd.nr_pages) {
1788 /*
1789 * Drop the socket lock, otherwise we have reverse
1790 * locking dependencies between sk_lock and i_mutex
1791 * here as compared to sendfile(). We enter here
1792 * with the socket lock held, and splice_to_pipe() will
1793 * grab the pipe inode lock. For sendfile() emulation,
1794 * we call into ->sendpage() with the i_mutex lock held
1795 * and networking will grab the socket lock.
1796 */
1797 release_sock(sk);
1798 ret = splice_to_pipe(pipe, &spd);
1799 lock_sock(sk);
1800 }
1801
1802 return ret;
1803}
1804
1805/**
1806 * skb_store_bits - store bits from kernel buffer to skb
1807 * @skb: destination buffer
1808 * @offset: offset in destination
1809 * @from: source buffer
1810 * @len: number of bytes to copy
1811 *
1812 * Copy the specified number of bytes from the source buffer to the
1813 * destination skb. This function handles all the messy bits of
1814 * traversing fragment lists and such.
1815 */
1816
1817int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1818{
1819 int start = skb_headlen(skb);
1820 struct sk_buff *frag_iter;
1821 int i, copy;
1822
1823 if (offset > (int)skb->len - len)
1824 goto fault;
1825
1826 if ((copy = start - offset) > 0) {
1827 if (copy > len)
1828 copy = len;
1829 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1830 if ((len -= copy) == 0)
1831 return 0;
1832 offset += copy;
1833 from += copy;
1834 }
1835
1836 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1837 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1838 int end;
1839
1840 WARN_ON(start > offset + len);
1841
1842 end = start + skb_frag_size(frag);
1843 if ((copy = end - offset) > 0) {
1844 u8 *vaddr;
1845
1846 if (copy > len)
1847 copy = len;
1848
1849 vaddr = kmap_atomic(skb_frag_page(frag));
1850 memcpy(vaddr + frag->page_offset + offset - start,
1851 from, copy);
1852 kunmap_atomic(vaddr);
1853
1854 if ((len -= copy) == 0)
1855 return 0;
1856 offset += copy;
1857 from += copy;
1858 }
1859 start = end;
1860 }
1861
1862 skb_walk_frags(skb, frag_iter) {
1863 int end;
1864
1865 WARN_ON(start > offset + len);
1866
1867 end = start + frag_iter->len;
1868 if ((copy = end - offset) > 0) {
1869 if (copy > len)
1870 copy = len;
1871 if (skb_store_bits(frag_iter, offset - start,
1872 from, copy))
1873 goto fault;
1874 if ((len -= copy) == 0)
1875 return 0;
1876 offset += copy;
1877 from += copy;
1878 }
1879 start = end;
1880 }
1881 if (!len)
1882 return 0;
1883
1884fault:
1885 return -EFAULT;
1886}
1887EXPORT_SYMBOL(skb_store_bits);
1888
1889/* Checksum skb data. */
1890
1891__wsum skb_checksum(const struct sk_buff *skb, int offset,
1892 int len, __wsum csum)
1893{
1894 int start = skb_headlen(skb);
1895 int i, copy = start - offset;
1896 struct sk_buff *frag_iter;
1897 int pos = 0;
1898
1899 /* Checksum header. */
1900 if (copy > 0) {
1901 if (copy > len)
1902 copy = len;
1903 csum = csum_partial(skb->data + offset, copy, csum);
1904 if ((len -= copy) == 0)
1905 return csum;
1906 offset += copy;
1907 pos = copy;
1908 }
1909
1910 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1911 int end;
1912 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1913
1914 WARN_ON(start > offset + len);
1915
1916 end = start + skb_frag_size(frag);
1917 if ((copy = end - offset) > 0) {
1918 __wsum csum2;
1919 u8 *vaddr;
1920
1921 if (copy > len)
1922 copy = len;
1923 vaddr = kmap_atomic(skb_frag_page(frag));
1924 csum2 = csum_partial(vaddr + frag->page_offset +
1925 offset - start, copy, 0);
1926 kunmap_atomic(vaddr);
1927 csum = csum_block_add(csum, csum2, pos);
1928 if (!(len -= copy))
1929 return csum;
1930 offset += copy;
1931 pos += copy;
1932 }
1933 start = end;
1934 }
1935
1936 skb_walk_frags(skb, frag_iter) {
1937 int end;
1938
1939 WARN_ON(start > offset + len);
1940
1941 end = start + frag_iter->len;
1942 if ((copy = end - offset) > 0) {
1943 __wsum csum2;
1944 if (copy > len)
1945 copy = len;
1946 csum2 = skb_checksum(frag_iter, offset - start,
1947 copy, 0);
1948 csum = csum_block_add(csum, csum2, pos);
1949 if ((len -= copy) == 0)
1950 return csum;
1951 offset += copy;
1952 pos += copy;
1953 }
1954 start = end;
1955 }
1956 BUG_ON(len);
1957
1958 return csum;
1959}
1960EXPORT_SYMBOL(skb_checksum);
1961
1962/* Both of above in one bottle. */
1963
1964__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1965 u8 *to, int len, __wsum csum)
1966{
1967 int start = skb_headlen(skb);
1968 int i, copy = start - offset;
1969 struct sk_buff *frag_iter;
1970 int pos = 0;
1971
1972 /* Copy header. */
1973 if (copy > 0) {
1974 if (copy > len)
1975 copy = len;
1976 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1977 copy, csum);
1978 if ((len -= copy) == 0)
1979 return csum;
1980 offset += copy;
1981 to += copy;
1982 pos = copy;
1983 }
1984
1985 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1986 int end;
1987
1988 WARN_ON(start > offset + len);
1989
1990 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1991 if ((copy = end - offset) > 0) {
1992 __wsum csum2;
1993 u8 *vaddr;
1994 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1995
1996 if (copy > len)
1997 copy = len;
1998 vaddr = kmap_atomic(skb_frag_page(frag));
1999 csum2 = csum_partial_copy_nocheck(vaddr +
2000 frag->page_offset +
2001 offset - start, to,
2002 copy, 0);
2003 kunmap_atomic(vaddr);
2004 csum = csum_block_add(csum, csum2, pos);
2005 if (!(len -= copy))
2006 return csum;
2007 offset += copy;
2008 to += copy;
2009 pos += copy;
2010 }
2011 start = end;
2012 }
2013
2014 skb_walk_frags(skb, frag_iter) {
2015 __wsum csum2;
2016 int end;
2017
2018 WARN_ON(start > offset + len);
2019
2020 end = start + frag_iter->len;
2021 if ((copy = end - offset) > 0) {
2022 if (copy > len)
2023 copy = len;
2024 csum2 = skb_copy_and_csum_bits(frag_iter,
2025 offset - start,
2026 to, copy, 0);
2027 csum = csum_block_add(csum, csum2, pos);
2028 if ((len -= copy) == 0)
2029 return csum;
2030 offset += copy;
2031 to += copy;
2032 pos += copy;
2033 }
2034 start = end;
2035 }
2036 BUG_ON(len);
2037 return csum;
2038}
2039EXPORT_SYMBOL(skb_copy_and_csum_bits);
2040
2041void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2042{
2043 __wsum csum;
2044 long csstart;
2045
2046 if (skb->ip_summed == CHECKSUM_PARTIAL)
2047 csstart = skb_checksum_start_offset(skb);
2048 else
2049 csstart = skb_headlen(skb);
2050
2051 BUG_ON(csstart > skb_headlen(skb));
2052
2053 skb_copy_from_linear_data(skb, to, csstart);
2054
2055 csum = 0;
2056 if (csstart != skb->len)
2057 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2058 skb->len - csstart, 0);
2059
2060 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2061 long csstuff = csstart + skb->csum_offset;
2062
2063 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2064 }
2065}
2066EXPORT_SYMBOL(skb_copy_and_csum_dev);
2067
2068/**
2069 * skb_dequeue - remove from the head of the queue
2070 * @list: list to dequeue from
2071 *
2072 * Remove the head of the list. The list lock is taken so the function
2073 * may be used safely with other locking list functions. The head item is
2074 * returned or %NULL if the list is empty.
2075 */
2076
2077struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2078{
2079 unsigned long flags;
2080 struct sk_buff *result;
2081
2082 spin_lock_irqsave(&list->lock, flags);
2083 result = __skb_dequeue(list);
2084 spin_unlock_irqrestore(&list->lock, flags);
2085 return result;
2086}
2087EXPORT_SYMBOL(skb_dequeue);
2088
2089/**
2090 * skb_dequeue_tail - remove from the tail of the queue
2091 * @list: list to dequeue from
2092 *
2093 * Remove the tail of the list. The list lock is taken so the function
2094 * may be used safely with other locking list functions. The tail item is
2095 * returned or %NULL if the list is empty.
2096 */
2097struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2098{
2099 unsigned long flags;
2100 struct sk_buff *result;
2101
2102 spin_lock_irqsave(&list->lock, flags);
2103 result = __skb_dequeue_tail(list);
2104 spin_unlock_irqrestore(&list->lock, flags);
2105 return result;
2106}
2107EXPORT_SYMBOL(skb_dequeue_tail);
2108
2109/**
2110 * skb_queue_purge - empty a list
2111 * @list: list to empty
2112 *
2113 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2114 * the list and one reference dropped. This function takes the list
2115 * lock and is atomic with respect to other list locking functions.
2116 */
2117void skb_queue_purge(struct sk_buff_head *list)
2118{
2119 struct sk_buff *skb;
2120 while ((skb = skb_dequeue(list)) != NULL)
2121 kfree_skb(skb);
2122}
2123EXPORT_SYMBOL(skb_queue_purge);
2124
2125/**
2126 * skb_queue_head - queue a buffer at the list head
2127 * @list: list to use
2128 * @newsk: buffer to queue
2129 *
2130 * Queue a buffer at the start of the list. This function takes the
2131 * list lock and can be used safely with other locking &sk_buff functions
2132 * safely.
2133 *
2134 * A buffer cannot be placed on two lists at the same time.
2135 */
2136void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2137{
2138 unsigned long flags;
2139
2140 spin_lock_irqsave(&list->lock, flags);
2141 __skb_queue_head(list, newsk);
2142 spin_unlock_irqrestore(&list->lock, flags);
2143}
2144EXPORT_SYMBOL(skb_queue_head);
2145
2146/**
2147 * skb_queue_tail - queue a buffer at the list tail
2148 * @list: list to use
2149 * @newsk: buffer to queue
2150 *
2151 * Queue a buffer at the tail of the list. This function takes the
2152 * list lock and can be used safely with other locking &sk_buff functions
2153 * safely.
2154 *
2155 * A buffer cannot be placed on two lists at the same time.
2156 */
2157void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2158{
2159 unsigned long flags;
2160
2161 spin_lock_irqsave(&list->lock, flags);
2162 __skb_queue_tail(list, newsk);
2163 spin_unlock_irqrestore(&list->lock, flags);
2164}
2165EXPORT_SYMBOL(skb_queue_tail);
2166
2167/**
2168 * skb_unlink - remove a buffer from a list
2169 * @skb: buffer to remove
2170 * @list: list to use
2171 *
2172 * Remove a packet from a list. The list locks are taken and this
2173 * function is atomic with respect to other list locked calls
2174 *
2175 * You must know what list the SKB is on.
2176 */
2177void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2178{
2179 unsigned long flags;
2180
2181 spin_lock_irqsave(&list->lock, flags);
2182 __skb_unlink(skb, list);
2183 spin_unlock_irqrestore(&list->lock, flags);
2184}
2185EXPORT_SYMBOL(skb_unlink);
2186
2187/**
2188 * skb_append - append a buffer
2189 * @old: buffer to insert after
2190 * @newsk: buffer to insert
2191 * @list: list to use
2192 *
2193 * Place a packet after a given packet in a list. The list locks are taken
2194 * and this function is atomic with respect to other list locked calls.
2195 * A buffer cannot be placed on two lists at the same time.
2196 */
2197void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2198{
2199 unsigned long flags;
2200
2201 spin_lock_irqsave(&list->lock, flags);
2202 __skb_queue_after(list, old, newsk);
2203 spin_unlock_irqrestore(&list->lock, flags);
2204}
2205EXPORT_SYMBOL(skb_append);
2206
2207/**
2208 * skb_insert - insert a buffer
2209 * @old: buffer to insert before
2210 * @newsk: buffer to insert
2211 * @list: list to use
2212 *
2213 * Place a packet before a given packet in a list. The list locks are
2214 * taken and this function is atomic with respect to other list locked
2215 * calls.
2216 *
2217 * A buffer cannot be placed on two lists at the same time.
2218 */
2219void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2220{
2221 unsigned long flags;
2222
2223 spin_lock_irqsave(&list->lock, flags);
2224 __skb_insert(newsk, old->prev, old, list);
2225 spin_unlock_irqrestore(&list->lock, flags);
2226}
2227EXPORT_SYMBOL(skb_insert);
2228
2229static inline void skb_split_inside_header(struct sk_buff *skb,
2230 struct sk_buff* skb1,
2231 const u32 len, const int pos)
2232{
2233 int i;
2234
2235 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2236 pos - len);
2237 /* And move data appendix as is. */
2238 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2239 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2240
2241 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2242 skb_shinfo(skb)->nr_frags = 0;
2243 skb1->data_len = skb->data_len;
2244 skb1->len += skb1->data_len;
2245 skb->data_len = 0;
2246 skb->len = len;
2247 skb_set_tail_pointer(skb, len);
2248}
2249
2250static inline void skb_split_no_header(struct sk_buff *skb,
2251 struct sk_buff* skb1,
2252 const u32 len, int pos)
2253{
2254 int i, k = 0;
2255 const int nfrags = skb_shinfo(skb)->nr_frags;
2256
2257 skb_shinfo(skb)->nr_frags = 0;
2258 skb1->len = skb1->data_len = skb->len - len;
2259 skb->len = len;
2260 skb->data_len = len - pos;
2261
2262 for (i = 0; i < nfrags; i++) {
2263 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2264
2265 if (pos + size > len) {
2266 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2267
2268 if (pos < len) {
2269 /* Split frag.
2270 * We have two variants in this case:
2271 * 1. Move all the frag to the second
2272 * part, if it is possible. F.e.
2273 * this approach is mandatory for TUX,
2274 * where splitting is expensive.
2275 * 2. Split is accurately. We make this.
2276 */
2277 skb_frag_ref(skb, i);
2278 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2279 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2280 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2281 skb_shinfo(skb)->nr_frags++;
2282 }
2283 k++;
2284 } else
2285 skb_shinfo(skb)->nr_frags++;
2286 pos += size;
2287 }
2288 skb_shinfo(skb1)->nr_frags = k;
2289}
2290
2291/**
2292 * skb_split - Split fragmented skb to two parts at length len.
2293 * @skb: the buffer to split
2294 * @skb1: the buffer to receive the second part
2295 * @len: new length for skb
2296 */
2297void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2298{
2299 int pos = skb_headlen(skb);
2300
2301 if (len < pos) /* Split line is inside header. */
2302 skb_split_inside_header(skb, skb1, len, pos);
2303 else /* Second chunk has no header, nothing to copy. */
2304 skb_split_no_header(skb, skb1, len, pos);
2305}
2306EXPORT_SYMBOL(skb_split);
2307
2308/* Shifting from/to a cloned skb is a no-go.
2309 *
2310 * Caller cannot keep skb_shinfo related pointers past calling here!
2311 */
2312static int skb_prepare_for_shift(struct sk_buff *skb)
2313{
2314 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2315}
2316
2317/**
2318 * skb_shift - Shifts paged data partially from skb to another
2319 * @tgt: buffer into which tail data gets added
2320 * @skb: buffer from which the paged data comes from
2321 * @shiftlen: shift up to this many bytes
2322 *
2323 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2324 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2325 * It's up to caller to free skb if everything was shifted.
2326 *
2327 * If @tgt runs out of frags, the whole operation is aborted.
2328 *
2329 * Skb cannot include anything else but paged data while tgt is allowed
2330 * to have non-paged data as well.
2331 *
2332 * TODO: full sized shift could be optimized but that would need
2333 * specialized skb free'er to handle frags without up-to-date nr_frags.
2334 */
2335int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2336{
2337 int from, to, merge, todo;
2338 struct skb_frag_struct *fragfrom, *fragto;
2339
2340 BUG_ON(shiftlen > skb->len);
2341 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2342
2343 todo = shiftlen;
2344 from = 0;
2345 to = skb_shinfo(tgt)->nr_frags;
2346 fragfrom = &skb_shinfo(skb)->frags[from];
2347
2348 /* Actual merge is delayed until the point when we know we can
2349 * commit all, so that we don't have to undo partial changes
2350 */
2351 if (!to ||
2352 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2353 fragfrom->page_offset)) {
2354 merge = -1;
2355 } else {
2356 merge = to - 1;
2357
2358 todo -= skb_frag_size(fragfrom);
2359 if (todo < 0) {
2360 if (skb_prepare_for_shift(skb) ||
2361 skb_prepare_for_shift(tgt))
2362 return 0;
2363
2364 /* All previous frag pointers might be stale! */
2365 fragfrom = &skb_shinfo(skb)->frags[from];
2366 fragto = &skb_shinfo(tgt)->frags[merge];
2367
2368 skb_frag_size_add(fragto, shiftlen);
2369 skb_frag_size_sub(fragfrom, shiftlen);
2370 fragfrom->page_offset += shiftlen;
2371
2372 goto onlymerged;
2373 }
2374
2375 from++;
2376 }
2377
2378 /* Skip full, not-fitting skb to avoid expensive operations */
2379 if ((shiftlen == skb->len) &&
2380 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2381 return 0;
2382
2383 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2384 return 0;
2385
2386 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2387 if (to == MAX_SKB_FRAGS)
2388 return 0;
2389
2390 fragfrom = &skb_shinfo(skb)->frags[from];
2391 fragto = &skb_shinfo(tgt)->frags[to];
2392
2393 if (todo >= skb_frag_size(fragfrom)) {
2394 *fragto = *fragfrom;
2395 todo -= skb_frag_size(fragfrom);
2396 from++;
2397 to++;
2398
2399 } else {
2400 __skb_frag_ref(fragfrom);
2401 fragto->page = fragfrom->page;
2402 fragto->page_offset = fragfrom->page_offset;
2403 skb_frag_size_set(fragto, todo);
2404
2405 fragfrom->page_offset += todo;
2406 skb_frag_size_sub(fragfrom, todo);
2407 todo = 0;
2408
2409 to++;
2410 break;
2411 }
2412 }
2413
2414 /* Ready to "commit" this state change to tgt */
2415 skb_shinfo(tgt)->nr_frags = to;
2416
2417 if (merge >= 0) {
2418 fragfrom = &skb_shinfo(skb)->frags[0];
2419 fragto = &skb_shinfo(tgt)->frags[merge];
2420
2421 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2422 __skb_frag_unref(fragfrom);
2423 }
2424
2425 /* Reposition in the original skb */
2426 to = 0;
2427 while (from < skb_shinfo(skb)->nr_frags)
2428 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2429 skb_shinfo(skb)->nr_frags = to;
2430
2431 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2432
2433onlymerged:
2434 /* Most likely the tgt won't ever need its checksum anymore, skb on
2435 * the other hand might need it if it needs to be resent
2436 */
2437 tgt->ip_summed = CHECKSUM_PARTIAL;
2438 skb->ip_summed = CHECKSUM_PARTIAL;
2439
2440 /* Yak, is it really working this way? Some helper please? */
2441 skb->len -= shiftlen;
2442 skb->data_len -= shiftlen;
2443 skb->truesize -= shiftlen;
2444 tgt->len += shiftlen;
2445 tgt->data_len += shiftlen;
2446 tgt->truesize += shiftlen;
2447
2448 return shiftlen;
2449}
2450
2451/**
2452 * skb_prepare_seq_read - Prepare a sequential read of skb data
2453 * @skb: the buffer to read
2454 * @from: lower offset of data to be read
2455 * @to: upper offset of data to be read
2456 * @st: state variable
2457 *
2458 * Initializes the specified state variable. Must be called before
2459 * invoking skb_seq_read() for the first time.
2460 */
2461void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2462 unsigned int to, struct skb_seq_state *st)
2463{
2464 st->lower_offset = from;
2465 st->upper_offset = to;
2466 st->root_skb = st->cur_skb = skb;
2467 st->frag_idx = st->stepped_offset = 0;
2468 st->frag_data = NULL;
2469}
2470EXPORT_SYMBOL(skb_prepare_seq_read);
2471
2472/**
2473 * skb_seq_read - Sequentially read skb data
2474 * @consumed: number of bytes consumed by the caller so far
2475 * @data: destination pointer for data to be returned
2476 * @st: state variable
2477 *
2478 * Reads a block of skb data at &consumed relative to the
2479 * lower offset specified to skb_prepare_seq_read(). Assigns
2480 * the head of the data block to &data and returns the length
2481 * of the block or 0 if the end of the skb data or the upper
2482 * offset has been reached.
2483 *
2484 * The caller is not required to consume all of the data
2485 * returned, i.e. &consumed is typically set to the number
2486 * of bytes already consumed and the next call to
2487 * skb_seq_read() will return the remaining part of the block.
2488 *
2489 * Note 1: The size of each block of data returned can be arbitrary,
2490 * this limitation is the cost for zerocopy seqeuental
2491 * reads of potentially non linear data.
2492 *
2493 * Note 2: Fragment lists within fragments are not implemented
2494 * at the moment, state->root_skb could be replaced with
2495 * a stack for this purpose.
2496 */
2497unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2498 struct skb_seq_state *st)
2499{
2500 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2501 skb_frag_t *frag;
2502
2503 if (unlikely(abs_offset >= st->upper_offset))
2504 return 0;
2505
2506next_skb:
2507 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2508
2509 if (abs_offset < block_limit && !st->frag_data) {
2510 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2511 return block_limit - abs_offset;
2512 }
2513
2514 if (st->frag_idx == 0 && !st->frag_data)
2515 st->stepped_offset += skb_headlen(st->cur_skb);
2516
2517 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2518 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2519 block_limit = skb_frag_size(frag) + st->stepped_offset;
2520
2521 if (abs_offset < block_limit) {
2522 if (!st->frag_data)
2523 st->frag_data = kmap_atomic(skb_frag_page(frag));
2524
2525 *data = (u8 *) st->frag_data + frag->page_offset +
2526 (abs_offset - st->stepped_offset);
2527
2528 return block_limit - abs_offset;
2529 }
2530
2531 if (st->frag_data) {
2532 kunmap_atomic(st->frag_data);
2533 st->frag_data = NULL;
2534 }
2535
2536 st->frag_idx++;
2537 st->stepped_offset += skb_frag_size(frag);
2538 }
2539
2540 if (st->frag_data) {
2541 kunmap_atomic(st->frag_data);
2542 st->frag_data = NULL;
2543 }
2544
2545 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2546 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2547 st->frag_idx = 0;
2548 goto next_skb;
2549 } else if (st->cur_skb->next) {
2550 st->cur_skb = st->cur_skb->next;
2551 st->frag_idx = 0;
2552 goto next_skb;
2553 }
2554
2555 return 0;
2556}
2557EXPORT_SYMBOL(skb_seq_read);
2558
2559/**
2560 * skb_abort_seq_read - Abort a sequential read of skb data
2561 * @st: state variable
2562 *
2563 * Must be called if skb_seq_read() was not called until it
2564 * returned 0.
2565 */
2566void skb_abort_seq_read(struct skb_seq_state *st)
2567{
2568 if (st->frag_data)
2569 kunmap_atomic(st->frag_data);
2570}
2571EXPORT_SYMBOL(skb_abort_seq_read);
2572
2573#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2574
2575static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2576 struct ts_config *conf,
2577 struct ts_state *state)
2578{
2579 return skb_seq_read(offset, text, TS_SKB_CB(state));
2580}
2581
2582static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2583{
2584 skb_abort_seq_read(TS_SKB_CB(state));
2585}
2586
2587/**
2588 * skb_find_text - Find a text pattern in skb data
2589 * @skb: the buffer to look in
2590 * @from: search offset
2591 * @to: search limit
2592 * @config: textsearch configuration
2593 * @state: uninitialized textsearch state variable
2594 *
2595 * Finds a pattern in the skb data according to the specified
2596 * textsearch configuration. Use textsearch_next() to retrieve
2597 * subsequent occurrences of the pattern. Returns the offset
2598 * to the first occurrence or UINT_MAX if no match was found.
2599 */
2600unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2601 unsigned int to, struct ts_config *config,
2602 struct ts_state *state)
2603{
2604 unsigned int ret;
2605
2606 config->get_next_block = skb_ts_get_next_block;
2607 config->finish = skb_ts_finish;
2608
2609 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2610
2611 ret = textsearch_find(config, state);
2612 return (ret <= to - from ? ret : UINT_MAX);
2613}
2614EXPORT_SYMBOL(skb_find_text);
2615
2616/**
2617 * skb_append_datato_frags: - append the user data to a skb
2618 * @sk: sock structure
2619 * @skb: skb structure to be appened with user data.
2620 * @getfrag: call back function to be used for getting the user data
2621 * @from: pointer to user message iov
2622 * @length: length of the iov message
2623 *
2624 * Description: This procedure append the user data in the fragment part
2625 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2626 */
2627int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2628 int (*getfrag)(void *from, char *to, int offset,
2629 int len, int odd, struct sk_buff *skb),
2630 void *from, int length)
2631{
2632 int frg_cnt = 0;
2633 skb_frag_t *frag = NULL;
2634 struct page *page = NULL;
2635 int copy, left;
2636 int offset = 0;
2637 int ret;
2638
2639 do {
2640 /* Return error if we don't have space for new frag */
2641 frg_cnt = skb_shinfo(skb)->nr_frags;
2642 if (frg_cnt >= MAX_SKB_FRAGS)
2643 return -EFAULT;
2644
2645 /* allocate a new page for next frag */
2646 page = alloc_pages(sk->sk_allocation, 0);
2647
2648 /* If alloc_page fails just return failure and caller will
2649 * free previous allocated pages by doing kfree_skb()
2650 */
2651 if (page == NULL)
2652 return -ENOMEM;
2653
2654 /* initialize the next frag */
2655 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2656 skb->truesize += PAGE_SIZE;
2657 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2658
2659 /* get the new initialized frag */
2660 frg_cnt = skb_shinfo(skb)->nr_frags;
2661 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2662
2663 /* copy the user data to page */
2664 left = PAGE_SIZE - frag->page_offset;
2665 copy = (length > left)? left : length;
2666
2667 ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2668 offset, copy, 0, skb);
2669 if (ret < 0)
2670 return -EFAULT;
2671
2672 /* copy was successful so update the size parameters */
2673 skb_frag_size_add(frag, copy);
2674 skb->len += copy;
2675 skb->data_len += copy;
2676 offset += copy;
2677 length -= copy;
2678
2679 } while (length > 0);
2680
2681 return 0;
2682}
2683EXPORT_SYMBOL(skb_append_datato_frags);
2684
2685/**
2686 * skb_pull_rcsum - pull skb and update receive checksum
2687 * @skb: buffer to update
2688 * @len: length of data pulled
2689 *
2690 * This function performs an skb_pull on the packet and updates
2691 * the CHECKSUM_COMPLETE checksum. It should be used on
2692 * receive path processing instead of skb_pull unless you know
2693 * that the checksum difference is zero (e.g., a valid IP header)
2694 * or you are setting ip_summed to CHECKSUM_NONE.
2695 */
2696unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2697{
2698 BUG_ON(len > skb->len);
2699 skb->len -= len;
2700 BUG_ON(skb->len < skb->data_len);
2701 skb_postpull_rcsum(skb, skb->data, len);
2702 return skb->data += len;
2703}
2704EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2705
2706/**
2707 * skb_segment - Perform protocol segmentation on skb.
2708 * @skb: buffer to segment
2709 * @features: features for the output path (see dev->features)
2710 *
2711 * This function performs segmentation on the given skb. It returns
2712 * a pointer to the first in a list of new skbs for the segments.
2713 * In case of error it returns ERR_PTR(err).
2714 */
2715struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2716{
2717 struct sk_buff *segs = NULL;
2718 struct sk_buff *tail = NULL;
2719 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2720 unsigned int mss = skb_shinfo(skb)->gso_size;
2721 unsigned int doffset = skb->data - skb_mac_header(skb);
2722 unsigned int offset = doffset;
2723 unsigned int headroom;
2724 unsigned int len;
2725 int sg = !!(features & NETIF_F_SG);
2726 int nfrags = skb_shinfo(skb)->nr_frags;
2727 int err = -ENOMEM;
2728 int i = 0;
2729 int pos;
2730
2731 __skb_push(skb, doffset);
2732 headroom = skb_headroom(skb);
2733 pos = skb_headlen(skb);
2734
2735 do {
2736 struct sk_buff *nskb;
2737 skb_frag_t *frag;
2738 int hsize;
2739 int size;
2740
2741 len = skb->len - offset;
2742 if (len > mss)
2743 len = mss;
2744
2745 hsize = skb_headlen(skb) - offset;
2746 if (hsize < 0)
2747 hsize = 0;
2748 if (hsize > len || !sg)
2749 hsize = len;
2750
2751 if (!hsize && i >= nfrags) {
2752 BUG_ON(fskb->len != len);
2753
2754 pos += len;
2755 nskb = skb_clone(fskb, GFP_ATOMIC);
2756 fskb = fskb->next;
2757
2758 if (unlikely(!nskb))
2759 goto err;
2760
2761 hsize = skb_end_offset(nskb);
2762 if (skb_cow_head(nskb, doffset + headroom)) {
2763 kfree_skb(nskb);
2764 goto err;
2765 }
2766
2767 nskb->truesize += skb_end_offset(nskb) - hsize;
2768 skb_release_head_state(nskb);
2769 __skb_push(nskb, doffset);
2770 } else {
2771 nskb = alloc_skb(hsize + doffset + headroom,
2772 GFP_ATOMIC);
2773
2774 if (unlikely(!nskb))
2775 goto err;
2776
2777 skb_reserve(nskb, headroom);
2778 __skb_put(nskb, doffset);
2779 }
2780
2781 if (segs)
2782 tail->next = nskb;
2783 else
2784 segs = nskb;
2785 tail = nskb;
2786
2787 __copy_skb_header(nskb, skb);
2788 nskb->mac_len = skb->mac_len;
2789
2790 /* nskb and skb might have different headroom */
2791 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2792 nskb->csum_start += skb_headroom(nskb) - headroom;
2793
2794 skb_reset_mac_header(nskb);
2795 skb_set_network_header(nskb, skb->mac_len);
2796 nskb->transport_header = (nskb->network_header +
2797 skb_network_header_len(skb));
2798 skb_copy_from_linear_data(skb, nskb->data, doffset);
2799
2800 if (fskb != skb_shinfo(skb)->frag_list)
2801 continue;
2802
2803 if (!sg) {
2804 nskb->ip_summed = CHECKSUM_NONE;
2805 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2806 skb_put(nskb, len),
2807 len, 0);
2808 continue;
2809 }
2810
2811 frag = skb_shinfo(nskb)->frags;
2812
2813 skb_copy_from_linear_data_offset(skb, offset,
2814 skb_put(nskb, hsize), hsize);
2815
2816 while (pos < offset + len && i < nfrags) {
2817 *frag = skb_shinfo(skb)->frags[i];
2818 __skb_frag_ref(frag);
2819 size = skb_frag_size(frag);
2820
2821 if (pos < offset) {
2822 frag->page_offset += offset - pos;
2823 skb_frag_size_sub(frag, offset - pos);
2824 }
2825
2826 skb_shinfo(nskb)->nr_frags++;
2827
2828 if (pos + size <= offset + len) {
2829 i++;
2830 pos += size;
2831 } else {
2832 skb_frag_size_sub(frag, pos + size - (offset + len));
2833 goto skip_fraglist;
2834 }
2835
2836 frag++;
2837 }
2838
2839 if (pos < offset + len) {
2840 struct sk_buff *fskb2 = fskb;
2841
2842 BUG_ON(pos + fskb->len != offset + len);
2843
2844 pos += fskb->len;
2845 fskb = fskb->next;
2846
2847 if (fskb2->next) {
2848 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2849 if (!fskb2)
2850 goto err;
2851 } else
2852 skb_get(fskb2);
2853
2854 SKB_FRAG_ASSERT(nskb);
2855 skb_shinfo(nskb)->frag_list = fskb2;
2856 }
2857
2858skip_fraglist:
2859 nskb->data_len = len - hsize;
2860 nskb->len += nskb->data_len;
2861 nskb->truesize += nskb->data_len;
2862 } while ((offset += len) < skb->len);
2863
2864 return segs;
2865
2866err:
2867 while ((skb = segs)) {
2868 segs = skb->next;
2869 kfree_skb(skb);
2870 }
2871 return ERR_PTR(err);
2872}
2873EXPORT_SYMBOL_GPL(skb_segment);
2874
2875int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2876{
2877 struct sk_buff *p = *head;
2878 struct sk_buff *nskb;
2879 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2880 struct skb_shared_info *pinfo = skb_shinfo(p);
2881 unsigned int headroom;
2882 unsigned int len = skb_gro_len(skb);
2883 unsigned int offset = skb_gro_offset(skb);
2884 unsigned int headlen = skb_headlen(skb);
2885 unsigned int delta_truesize;
2886
2887 if (p->len + len >= 65536)
2888 return -E2BIG;
2889
2890 if (pinfo->frag_list)
2891 goto merge;
2892 else if (headlen <= offset) {
2893 skb_frag_t *frag;
2894 skb_frag_t *frag2;
2895 int i = skbinfo->nr_frags;
2896 int nr_frags = pinfo->nr_frags + i;
2897
2898 offset -= headlen;
2899
2900 if (nr_frags > MAX_SKB_FRAGS)
2901 return -E2BIG;
2902
2903 pinfo->nr_frags = nr_frags;
2904 skbinfo->nr_frags = 0;
2905
2906 frag = pinfo->frags + nr_frags;
2907 frag2 = skbinfo->frags + i;
2908 do {
2909 *--frag = *--frag2;
2910 } while (--i);
2911
2912 frag->page_offset += offset;
2913 skb_frag_size_sub(frag, offset);
2914
2915 /* all fragments truesize : remove (head size + sk_buff) */
2916 delta_truesize = skb->truesize -
2917 SKB_TRUESIZE(skb_end_offset(skb));
2918
2919 skb->truesize -= skb->data_len;
2920 skb->len -= skb->data_len;
2921 skb->data_len = 0;
2922
2923 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
2924 goto done;
2925 } else if (skb->head_frag) {
2926 int nr_frags = pinfo->nr_frags;
2927 skb_frag_t *frag = pinfo->frags + nr_frags;
2928 struct page *page = virt_to_head_page(skb->head);
2929 unsigned int first_size = headlen - offset;
2930 unsigned int first_offset;
2931
2932 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
2933 return -E2BIG;
2934
2935 first_offset = skb->data -
2936 (unsigned char *)page_address(page) +
2937 offset;
2938
2939 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
2940
2941 frag->page.p = page;
2942 frag->page_offset = first_offset;
2943 skb_frag_size_set(frag, first_size);
2944
2945 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
2946 /* We dont need to clear skbinfo->nr_frags here */
2947
2948 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
2949 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
2950 goto done;
2951 } else if (skb_gro_len(p) != pinfo->gso_size)
2952 return -E2BIG;
2953
2954 headroom = skb_headroom(p);
2955 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2956 if (unlikely(!nskb))
2957 return -ENOMEM;
2958
2959 __copy_skb_header(nskb, p);
2960 nskb->mac_len = p->mac_len;
2961
2962 skb_reserve(nskb, headroom);
2963 __skb_put(nskb, skb_gro_offset(p));
2964
2965 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2966 skb_set_network_header(nskb, skb_network_offset(p));
2967 skb_set_transport_header(nskb, skb_transport_offset(p));
2968
2969 __skb_pull(p, skb_gro_offset(p));
2970 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2971 p->data - skb_mac_header(p));
2972
2973 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2974 skb_shinfo(nskb)->frag_list = p;
2975 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2976 pinfo->gso_size = 0;
2977 skb_header_release(p);
2978 nskb->prev = p;
2979
2980 nskb->data_len += p->len;
2981 nskb->truesize += p->truesize;
2982 nskb->len += p->len;
2983
2984 *head = nskb;
2985 nskb->next = p->next;
2986 p->next = NULL;
2987
2988 p = nskb;
2989
2990merge:
2991 delta_truesize = skb->truesize;
2992 if (offset > headlen) {
2993 unsigned int eat = offset - headlen;
2994
2995 skbinfo->frags[0].page_offset += eat;
2996 skb_frag_size_sub(&skbinfo->frags[0], eat);
2997 skb->data_len -= eat;
2998 skb->len -= eat;
2999 offset = headlen;
3000 }
3001
3002 __skb_pull(skb, offset);
3003
3004 p->prev->next = skb;
3005 p->prev = skb;
3006 skb_header_release(skb);
3007
3008done:
3009 NAPI_GRO_CB(p)->count++;
3010 p->data_len += len;
3011 p->truesize += delta_truesize;
3012 p->len += len;
3013
3014 NAPI_GRO_CB(skb)->same_flow = 1;
3015 return 0;
3016}
3017EXPORT_SYMBOL_GPL(skb_gro_receive);
3018
3019void __init skb_init(void)
3020{
3021 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3022 sizeof(struct sk_buff),
3023 0,
3024 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3025 NULL);
3026 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3027 (2*sizeof(struct sk_buff)) +
3028 sizeof(atomic_t),
3029 0,
3030 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3031 NULL);
3032}
3033
3034/**
3035 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3036 * @skb: Socket buffer containing the buffers to be mapped
3037 * @sg: The scatter-gather list to map into
3038 * @offset: The offset into the buffer's contents to start mapping
3039 * @len: Length of buffer space to be mapped
3040 *
3041 * Fill the specified scatter-gather list with mappings/pointers into a
3042 * region of the buffer space attached to a socket buffer.
3043 */
3044static int
3045__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3046{
3047 int start = skb_headlen(skb);
3048 int i, copy = start - offset;
3049 struct sk_buff *frag_iter;
3050 int elt = 0;
3051
3052 if (copy > 0) {
3053 if (copy > len)
3054 copy = len;
3055 sg_set_buf(sg, skb->data + offset, copy);
3056 elt++;
3057 if ((len -= copy) == 0)
3058 return elt;
3059 offset += copy;
3060 }
3061
3062 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3063 int end;
3064
3065 WARN_ON(start > offset + len);
3066
3067 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3068 if ((copy = end - offset) > 0) {
3069 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3070
3071 if (copy > len)
3072 copy = len;
3073 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3074 frag->page_offset+offset-start);
3075 elt++;
3076 if (!(len -= copy))
3077 return elt;
3078 offset += copy;
3079 }
3080 start = end;
3081 }
3082
3083 skb_walk_frags(skb, frag_iter) {
3084 int end;
3085
3086 WARN_ON(start > offset + len);
3087
3088 end = start + frag_iter->len;
3089 if ((copy = end - offset) > 0) {
3090 if (copy > len)
3091 copy = len;
3092 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3093 copy);
3094 if ((len -= copy) == 0)
3095 return elt;
3096 offset += copy;
3097 }
3098 start = end;
3099 }
3100 BUG_ON(len);
3101 return elt;
3102}
3103
3104int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3105{
3106 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3107
3108 sg_mark_end(&sg[nsg - 1]);
3109
3110 return nsg;
3111}
3112EXPORT_SYMBOL_GPL(skb_to_sgvec);
3113
3114/**
3115 * skb_cow_data - Check that a socket buffer's data buffers are writable
3116 * @skb: The socket buffer to check.
3117 * @tailbits: Amount of trailing space to be added
3118 * @trailer: Returned pointer to the skb where the @tailbits space begins
3119 *
3120 * Make sure that the data buffers attached to a socket buffer are
3121 * writable. If they are not, private copies are made of the data buffers
3122 * and the socket buffer is set to use these instead.
3123 *
3124 * If @tailbits is given, make sure that there is space to write @tailbits
3125 * bytes of data beyond current end of socket buffer. @trailer will be
3126 * set to point to the skb in which this space begins.
3127 *
3128 * The number of scatterlist elements required to completely map the
3129 * COW'd and extended socket buffer will be returned.
3130 */
3131int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3132{
3133 int copyflag;
3134 int elt;
3135 struct sk_buff *skb1, **skb_p;
3136
3137 /* If skb is cloned or its head is paged, reallocate
3138 * head pulling out all the pages (pages are considered not writable
3139 * at the moment even if they are anonymous).
3140 */
3141 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3142 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3143 return -ENOMEM;
3144
3145 /* Easy case. Most of packets will go this way. */
3146 if (!skb_has_frag_list(skb)) {
3147 /* A little of trouble, not enough of space for trailer.
3148 * This should not happen, when stack is tuned to generate
3149 * good frames. OK, on miss we reallocate and reserve even more
3150 * space, 128 bytes is fair. */
3151
3152 if (skb_tailroom(skb) < tailbits &&
3153 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3154 return -ENOMEM;
3155
3156 /* Voila! */
3157 *trailer = skb;
3158 return 1;
3159 }
3160
3161 /* Misery. We are in troubles, going to mincer fragments... */
3162
3163 elt = 1;
3164 skb_p = &skb_shinfo(skb)->frag_list;
3165 copyflag = 0;
3166
3167 while ((skb1 = *skb_p) != NULL) {
3168 int ntail = 0;
3169
3170 /* The fragment is partially pulled by someone,
3171 * this can happen on input. Copy it and everything
3172 * after it. */
3173
3174 if (skb_shared(skb1))
3175 copyflag = 1;
3176
3177 /* If the skb is the last, worry about trailer. */
3178
3179 if (skb1->next == NULL && tailbits) {
3180 if (skb_shinfo(skb1)->nr_frags ||
3181 skb_has_frag_list(skb1) ||
3182 skb_tailroom(skb1) < tailbits)
3183 ntail = tailbits + 128;
3184 }
3185
3186 if (copyflag ||
3187 skb_cloned(skb1) ||
3188 ntail ||
3189 skb_shinfo(skb1)->nr_frags ||
3190 skb_has_frag_list(skb1)) {
3191 struct sk_buff *skb2;
3192
3193 /* Fuck, we are miserable poor guys... */
3194 if (ntail == 0)
3195 skb2 = skb_copy(skb1, GFP_ATOMIC);
3196 else
3197 skb2 = skb_copy_expand(skb1,
3198 skb_headroom(skb1),
3199 ntail,
3200 GFP_ATOMIC);
3201 if (unlikely(skb2 == NULL))
3202 return -ENOMEM;
3203
3204 if (skb1->sk)
3205 skb_set_owner_w(skb2, skb1->sk);
3206
3207 /* Looking around. Are we still alive?
3208 * OK, link new skb, drop old one */
3209
3210 skb2->next = skb1->next;
3211 *skb_p = skb2;
3212 kfree_skb(skb1);
3213 skb1 = skb2;
3214 }
3215 elt++;
3216 *trailer = skb1;
3217 skb_p = &skb1->next;
3218 }
3219
3220 return elt;
3221}
3222EXPORT_SYMBOL_GPL(skb_cow_data);
3223
3224static void sock_rmem_free(struct sk_buff *skb)
3225{
3226 struct sock *sk = skb->sk;
3227
3228 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3229}
3230
3231/*
3232 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3233 */
3234int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3235{
3236 int len = skb->len;
3237
3238 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3239 (unsigned int)sk->sk_rcvbuf)
3240 return -ENOMEM;
3241
3242 skb_orphan(skb);
3243 skb->sk = sk;
3244 skb->destructor = sock_rmem_free;
3245 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3246
3247 /* before exiting rcu section, make sure dst is refcounted */
3248 skb_dst_force(skb);
3249
3250 skb_queue_tail(&sk->sk_error_queue, skb);
3251 if (!sock_flag(sk, SOCK_DEAD))
3252 sk->sk_data_ready(sk, len);
3253 return 0;
3254}
3255EXPORT_SYMBOL(sock_queue_err_skb);
3256
3257void skb_tstamp_tx(struct sk_buff *orig_skb,
3258 struct skb_shared_hwtstamps *hwtstamps)
3259{
3260 struct sock *sk = orig_skb->sk;
3261 struct sock_exterr_skb *serr;
3262 struct sk_buff *skb;
3263 int err;
3264
3265 if (!sk)
3266 return;
3267
3268 skb = skb_clone(orig_skb, GFP_ATOMIC);
3269 if (!skb)
3270 return;
3271
3272 if (hwtstamps) {
3273 *skb_hwtstamps(skb) =
3274 *hwtstamps;
3275 } else {
3276 /*
3277 * no hardware time stamps available,
3278 * so keep the shared tx_flags and only
3279 * store software time stamp
3280 */
3281 skb->tstamp = ktime_get_real();
3282 }
3283
3284 serr = SKB_EXT_ERR(skb);
3285 memset(serr, 0, sizeof(*serr));
3286 serr->ee.ee_errno = ENOMSG;
3287 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3288
3289 err = sock_queue_err_skb(sk, skb);
3290
3291 if (err)
3292 kfree_skb(skb);
3293}
3294EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3295
3296void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3297{
3298 struct sock *sk = skb->sk;
3299 struct sock_exterr_skb *serr;
3300 int err;
3301
3302 skb->wifi_acked_valid = 1;
3303 skb->wifi_acked = acked;
3304
3305 serr = SKB_EXT_ERR(skb);
3306 memset(serr, 0, sizeof(*serr));
3307 serr->ee.ee_errno = ENOMSG;
3308 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3309
3310 err = sock_queue_err_skb(sk, skb);
3311 if (err)
3312 kfree_skb(skb);
3313}
3314EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3315
3316
3317/**
3318 * skb_partial_csum_set - set up and verify partial csum values for packet
3319 * @skb: the skb to set
3320 * @start: the number of bytes after skb->data to start checksumming.
3321 * @off: the offset from start to place the checksum.
3322 *
3323 * For untrusted partially-checksummed packets, we need to make sure the values
3324 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3325 *
3326 * This function checks and sets those values and skb->ip_summed: if this
3327 * returns false you should drop the packet.
3328 */
3329bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3330{
3331 if (unlikely(start > skb_headlen(skb)) ||
3332 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3333 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3334 start, off, skb_headlen(skb));
3335 return false;
3336 }
3337 skb->ip_summed = CHECKSUM_PARTIAL;
3338 skb->csum_start = skb_headroom(skb) + start;
3339 skb->csum_offset = off;
3340 return true;
3341}
3342EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3343
3344void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3345{
3346 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3347 skb->dev->name);
3348}
3349EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3350
3351void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3352{
3353 if (head_stolen)
3354 kmem_cache_free(skbuff_head_cache, skb);
3355 else
3356 __kfree_skb(skb);
3357}
3358EXPORT_SYMBOL(kfree_skb_partial);
3359
3360/**
3361 * skb_try_coalesce - try to merge skb to prior one
3362 * @to: prior buffer
3363 * @from: buffer to add
3364 * @fragstolen: pointer to boolean
3365 * @delta_truesize: how much more was allocated than was requested
3366 */
3367bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3368 bool *fragstolen, int *delta_truesize)
3369{
3370 int i, delta, len = from->len;
3371
3372 *fragstolen = false;
3373
3374 if (skb_cloned(to))
3375 return false;
3376
3377 if (len <= skb_tailroom(to)) {
3378 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3379 *delta_truesize = 0;
3380 return true;
3381 }
3382
3383 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3384 return false;
3385
3386 if (skb_headlen(from) != 0) {
3387 struct page *page;
3388 unsigned int offset;
3389
3390 if (skb_shinfo(to)->nr_frags +
3391 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3392 return false;
3393
3394 if (skb_head_is_locked(from))
3395 return false;
3396
3397 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3398
3399 page = virt_to_head_page(from->head);
3400 offset = from->data - (unsigned char *)page_address(page);
3401
3402 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3403 page, offset, skb_headlen(from));
3404 *fragstolen = true;
3405 } else {
3406 if (skb_shinfo(to)->nr_frags +
3407 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3408 return false;
3409
3410 delta = from->truesize -
3411 SKB_TRUESIZE(skb_end_pointer(from) - from->head);
3412 }
3413
3414 WARN_ON_ONCE(delta < len);
3415
3416 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3417 skb_shinfo(from)->frags,
3418 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3419 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3420
3421 if (!skb_cloned(from))
3422 skb_shinfo(from)->nr_frags = 0;
3423
3424 /* if the skb is cloned this does nothing since we set nr_frags to 0 */
3425 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3426 skb_frag_ref(from, i);
3427
3428 to->truesize += delta;
3429 to->len += len;
3430 to->data_len += len;
3431
3432 *delta_truesize = delta;
3433 return true;
3434}
3435EXPORT_SYMBOL(skb_try_coalesce);