Loading...
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/kmemcheck.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/tcp.h>
51#include <linux/udp.h>
52#include <linux/netdevice.h>
53#ifdef CONFIG_NET_CLS_ACT
54#include <net/pkt_sched.h>
55#endif
56#include <linux/string.h>
57#include <linux/skbuff.h>
58#include <linux/splice.h>
59#include <linux/cache.h>
60#include <linux/rtnetlink.h>
61#include <linux/init.h>
62#include <linux/scatterlist.h>
63#include <linux/errqueue.h>
64#include <linux/prefetch.h>
65#include <linux/if_vlan.h>
66
67#include <net/protocol.h>
68#include <net/dst.h>
69#include <net/sock.h>
70#include <net/checksum.h>
71#include <net/ip6_checksum.h>
72#include <net/xfrm.h>
73
74#include <asm/uaccess.h>
75#include <trace/events/skb.h>
76#include <linux/highmem.h>
77#include <linux/capability.h>
78#include <linux/user_namespace.h>
79
80struct kmem_cache *skbuff_head_cache __read_mostly;
81static struct kmem_cache *skbuff_fclone_cache __read_mostly;
82int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83EXPORT_SYMBOL(sysctl_max_skb_frags);
84
85/**
86 * skb_panic - private function for out-of-line support
87 * @skb: buffer
88 * @sz: size
89 * @addr: address
90 * @msg: skb_over_panic or skb_under_panic
91 *
92 * Out-of-line support for skb_put() and skb_push().
93 * Called via the wrapper skb_over_panic() or skb_under_panic().
94 * Keep out of line to prevent kernel bloat.
95 * __builtin_return_address is not used because it is not always reliable.
96 */
97static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
98 const char msg[])
99{
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
104 BUG();
105}
106
107static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
111
112static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117/*
118 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
119 * the caller if emergency pfmemalloc reserves are being used. If it is and
120 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
121 * may be used. Otherwise, the packet data may be discarded until enough
122 * memory is free
123 */
124#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
126
127static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
129{
130 void *obj;
131 bool ret_pfmemalloc = false;
132
133 /*
134 * Try a regular allocation, when that fails and we're not entitled
135 * to the reserves, fail.
136 */
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
139 node);
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
141 goto out;
142
143 /* Try again but now we are using pfmemalloc reserves */
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
146
147out:
148 if (pfmemalloc)
149 *pfmemalloc = ret_pfmemalloc;
150
151 return obj;
152}
153
154/* Allocate a new skbuff. We do this ourselves so we can fill in a few
155 * 'private' fields and also do memory statistics to find all the
156 * [BEEP] leaks.
157 *
158 */
159
160struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
161{
162 struct sk_buff *skb;
163
164 /* Get the HEAD */
165 skb = kmem_cache_alloc_node(skbuff_head_cache,
166 gfp_mask & ~__GFP_DMA, node);
167 if (!skb)
168 goto out;
169
170 /*
171 * Only clear those fields we need to clear, not those that we will
172 * actually initialise below. Hence, don't put any more fields after
173 * the tail pointer in struct sk_buff!
174 */
175 memset(skb, 0, offsetof(struct sk_buff, tail));
176 skb->head = NULL;
177 skb->truesize = sizeof(struct sk_buff);
178 atomic_set(&skb->users, 1);
179
180 skb->mac_header = (typeof(skb->mac_header))~0U;
181out:
182 return skb;
183}
184
185/**
186 * __alloc_skb - allocate a network buffer
187 * @size: size to allocate
188 * @gfp_mask: allocation mask
189 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
190 * instead of head cache and allocate a cloned (child) skb.
191 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
192 * allocations in case the data is required for writeback
193 * @node: numa node to allocate memory on
194 *
195 * Allocate a new &sk_buff. The returned buffer has no headroom and a
196 * tail room of at least size bytes. The object has a reference count
197 * of one. The return is the buffer. On a failure the return is %NULL.
198 *
199 * Buffers may only be allocated from interrupts using a @gfp_mask of
200 * %GFP_ATOMIC.
201 */
202struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
203 int flags, int node)
204{
205 struct kmem_cache *cache;
206 struct skb_shared_info *shinfo;
207 struct sk_buff *skb;
208 u8 *data;
209 bool pfmemalloc;
210
211 cache = (flags & SKB_ALLOC_FCLONE)
212 ? skbuff_fclone_cache : skbuff_head_cache;
213
214 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
215 gfp_mask |= __GFP_MEMALLOC;
216
217 /* Get the HEAD */
218 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
219 if (!skb)
220 goto out;
221 prefetchw(skb);
222
223 /* We do our best to align skb_shared_info on a separate cache
224 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
225 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
226 * Both skb->head and skb_shared_info are cache line aligned.
227 */
228 size = SKB_DATA_ALIGN(size);
229 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
230 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
231 if (!data)
232 goto nodata;
233 /* kmalloc(size) might give us more room than requested.
234 * Put skb_shared_info exactly at the end of allocated zone,
235 * to allow max possible filling before reallocation.
236 */
237 size = SKB_WITH_OVERHEAD(ksize(data));
238 prefetchw(data + size);
239
240 /*
241 * Only clear those fields we need to clear, not those that we will
242 * actually initialise below. Hence, don't put any more fields after
243 * the tail pointer in struct sk_buff!
244 */
245 memset(skb, 0, offsetof(struct sk_buff, tail));
246 /* Account for allocated memory : skb + skb->head */
247 skb->truesize = SKB_TRUESIZE(size);
248 skb->pfmemalloc = pfmemalloc;
249 atomic_set(&skb->users, 1);
250 skb->head = data;
251 skb->data = data;
252 skb_reset_tail_pointer(skb);
253 skb->end = skb->tail + size;
254 skb->mac_header = (typeof(skb->mac_header))~0U;
255 skb->transport_header = (typeof(skb->transport_header))~0U;
256
257 /* make sure we initialize shinfo sequentially */
258 shinfo = skb_shinfo(skb);
259 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
260 atomic_set(&shinfo->dataref, 1);
261 kmemcheck_annotate_variable(shinfo->destructor_arg);
262
263 if (flags & SKB_ALLOC_FCLONE) {
264 struct sk_buff_fclones *fclones;
265
266 fclones = container_of(skb, struct sk_buff_fclones, skb1);
267
268 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
269 skb->fclone = SKB_FCLONE_ORIG;
270 atomic_set(&fclones->fclone_ref, 1);
271
272 fclones->skb2.fclone = SKB_FCLONE_CLONE;
273 fclones->skb2.pfmemalloc = pfmemalloc;
274 }
275out:
276 return skb;
277nodata:
278 kmem_cache_free(cache, skb);
279 skb = NULL;
280 goto out;
281}
282EXPORT_SYMBOL(__alloc_skb);
283
284/**
285 * __build_skb - build a network buffer
286 * @data: data buffer provided by caller
287 * @frag_size: size of data, or 0 if head was kmalloced
288 *
289 * Allocate a new &sk_buff. Caller provides space holding head and
290 * skb_shared_info. @data must have been allocated by kmalloc() only if
291 * @frag_size is 0, otherwise data should come from the page allocator
292 * or vmalloc()
293 * The return is the new skb buffer.
294 * On a failure the return is %NULL, and @data is not freed.
295 * Notes :
296 * Before IO, driver allocates only data buffer where NIC put incoming frame
297 * Driver should add room at head (NET_SKB_PAD) and
298 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
299 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
300 * before giving packet to stack.
301 * RX rings only contains data buffers, not full skbs.
302 */
303struct sk_buff *__build_skb(void *data, unsigned int frag_size)
304{
305 struct skb_shared_info *shinfo;
306 struct sk_buff *skb;
307 unsigned int size = frag_size ? : ksize(data);
308
309 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
310 if (!skb)
311 return NULL;
312
313 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
314
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316 skb->truesize = SKB_TRUESIZE(size);
317 atomic_set(&skb->users, 1);
318 skb->head = data;
319 skb->data = data;
320 skb_reset_tail_pointer(skb);
321 skb->end = skb->tail + size;
322 skb->mac_header = (typeof(skb->mac_header))~0U;
323 skb->transport_header = (typeof(skb->transport_header))~0U;
324
325 /* make sure we initialize shinfo sequentially */
326 shinfo = skb_shinfo(skb);
327 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
328 atomic_set(&shinfo->dataref, 1);
329 kmemcheck_annotate_variable(shinfo->destructor_arg);
330
331 return skb;
332}
333
334/* build_skb() is wrapper over __build_skb(), that specifically
335 * takes care of skb->head and skb->pfmemalloc
336 * This means that if @frag_size is not zero, then @data must be backed
337 * by a page fragment, not kmalloc() or vmalloc()
338 */
339struct sk_buff *build_skb(void *data, unsigned int frag_size)
340{
341 struct sk_buff *skb = __build_skb(data, frag_size);
342
343 if (skb && frag_size) {
344 skb->head_frag = 1;
345 if (page_is_pfmemalloc(virt_to_head_page(data)))
346 skb->pfmemalloc = 1;
347 }
348 return skb;
349}
350EXPORT_SYMBOL(build_skb);
351
352#define NAPI_SKB_CACHE_SIZE 64
353
354struct napi_alloc_cache {
355 struct page_frag_cache page;
356 size_t skb_count;
357 void *skb_cache[NAPI_SKB_CACHE_SIZE];
358};
359
360static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
361static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
362
363static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
364{
365 struct page_frag_cache *nc;
366 unsigned long flags;
367 void *data;
368
369 local_irq_save(flags);
370 nc = this_cpu_ptr(&netdev_alloc_cache);
371 data = __alloc_page_frag(nc, fragsz, gfp_mask);
372 local_irq_restore(flags);
373 return data;
374}
375
376/**
377 * netdev_alloc_frag - allocate a page fragment
378 * @fragsz: fragment size
379 *
380 * Allocates a frag from a page for receive buffer.
381 * Uses GFP_ATOMIC allocations.
382 */
383void *netdev_alloc_frag(unsigned int fragsz)
384{
385 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
386}
387EXPORT_SYMBOL(netdev_alloc_frag);
388
389static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
390{
391 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
392
393 return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
394}
395
396void *napi_alloc_frag(unsigned int fragsz)
397{
398 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
399}
400EXPORT_SYMBOL(napi_alloc_frag);
401
402/**
403 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
404 * @dev: network device to receive on
405 * @len: length to allocate
406 * @gfp_mask: get_free_pages mask, passed to alloc_skb
407 *
408 * Allocate a new &sk_buff and assign it a usage count of one. The
409 * buffer has NET_SKB_PAD headroom built in. Users should allocate
410 * the headroom they think they need without accounting for the
411 * built in space. The built in space is used for optimisations.
412 *
413 * %NULL is returned if there is no free memory.
414 */
415struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
416 gfp_t gfp_mask)
417{
418 struct page_frag_cache *nc;
419 unsigned long flags;
420 struct sk_buff *skb;
421 bool pfmemalloc;
422 void *data;
423
424 len += NET_SKB_PAD;
425
426 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
427 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
428 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
429 if (!skb)
430 goto skb_fail;
431 goto skb_success;
432 }
433
434 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
435 len = SKB_DATA_ALIGN(len);
436
437 if (sk_memalloc_socks())
438 gfp_mask |= __GFP_MEMALLOC;
439
440 local_irq_save(flags);
441
442 nc = this_cpu_ptr(&netdev_alloc_cache);
443 data = __alloc_page_frag(nc, len, gfp_mask);
444 pfmemalloc = nc->pfmemalloc;
445
446 local_irq_restore(flags);
447
448 if (unlikely(!data))
449 return NULL;
450
451 skb = __build_skb(data, len);
452 if (unlikely(!skb)) {
453 skb_free_frag(data);
454 return NULL;
455 }
456
457 /* use OR instead of assignment to avoid clearing of bits in mask */
458 if (pfmemalloc)
459 skb->pfmemalloc = 1;
460 skb->head_frag = 1;
461
462skb_success:
463 skb_reserve(skb, NET_SKB_PAD);
464 skb->dev = dev;
465
466skb_fail:
467 return skb;
468}
469EXPORT_SYMBOL(__netdev_alloc_skb);
470
471/**
472 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
473 * @napi: napi instance this buffer was allocated for
474 * @len: length to allocate
475 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
476 *
477 * Allocate a new sk_buff for use in NAPI receive. This buffer will
478 * attempt to allocate the head from a special reserved region used
479 * only for NAPI Rx allocation. By doing this we can save several
480 * CPU cycles by avoiding having to disable and re-enable IRQs.
481 *
482 * %NULL is returned if there is no free memory.
483 */
484struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
485 gfp_t gfp_mask)
486{
487 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
488 struct sk_buff *skb;
489 void *data;
490
491 len += NET_SKB_PAD + NET_IP_ALIGN;
492
493 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
494 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
495 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
496 if (!skb)
497 goto skb_fail;
498 goto skb_success;
499 }
500
501 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
502 len = SKB_DATA_ALIGN(len);
503
504 if (sk_memalloc_socks())
505 gfp_mask |= __GFP_MEMALLOC;
506
507 data = __alloc_page_frag(&nc->page, len, gfp_mask);
508 if (unlikely(!data))
509 return NULL;
510
511 skb = __build_skb(data, len);
512 if (unlikely(!skb)) {
513 skb_free_frag(data);
514 return NULL;
515 }
516
517 /* use OR instead of assignment to avoid clearing of bits in mask */
518 if (nc->page.pfmemalloc)
519 skb->pfmemalloc = 1;
520 skb->head_frag = 1;
521
522skb_success:
523 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
524 skb->dev = napi->dev;
525
526skb_fail:
527 return skb;
528}
529EXPORT_SYMBOL(__napi_alloc_skb);
530
531void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
532 int size, unsigned int truesize)
533{
534 skb_fill_page_desc(skb, i, page, off, size);
535 skb->len += size;
536 skb->data_len += size;
537 skb->truesize += truesize;
538}
539EXPORT_SYMBOL(skb_add_rx_frag);
540
541void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
542 unsigned int truesize)
543{
544 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
545
546 skb_frag_size_add(frag, size);
547 skb->len += size;
548 skb->data_len += size;
549 skb->truesize += truesize;
550}
551EXPORT_SYMBOL(skb_coalesce_rx_frag);
552
553static void skb_drop_list(struct sk_buff **listp)
554{
555 kfree_skb_list(*listp);
556 *listp = NULL;
557}
558
559static inline void skb_drop_fraglist(struct sk_buff *skb)
560{
561 skb_drop_list(&skb_shinfo(skb)->frag_list);
562}
563
564static void skb_clone_fraglist(struct sk_buff *skb)
565{
566 struct sk_buff *list;
567
568 skb_walk_frags(skb, list)
569 skb_get(list);
570}
571
572static void skb_free_head(struct sk_buff *skb)
573{
574 unsigned char *head = skb->head;
575
576 if (skb->head_frag)
577 skb_free_frag(head);
578 else
579 kfree(head);
580}
581
582static void skb_release_data(struct sk_buff *skb)
583{
584 struct skb_shared_info *shinfo = skb_shinfo(skb);
585 int i;
586
587 if (skb->cloned &&
588 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
589 &shinfo->dataref))
590 return;
591
592 for (i = 0; i < shinfo->nr_frags; i++)
593 __skb_frag_unref(&shinfo->frags[i]);
594
595 /*
596 * If skb buf is from userspace, we need to notify the caller
597 * the lower device DMA has done;
598 */
599 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
600 struct ubuf_info *uarg;
601
602 uarg = shinfo->destructor_arg;
603 if (uarg->callback)
604 uarg->callback(uarg, true);
605 }
606
607 if (shinfo->frag_list)
608 kfree_skb_list(shinfo->frag_list);
609
610 skb_free_head(skb);
611}
612
613/*
614 * Free an skbuff by memory without cleaning the state.
615 */
616static void kfree_skbmem(struct sk_buff *skb)
617{
618 struct sk_buff_fclones *fclones;
619
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
623 return;
624
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
627
628 /* We usually free the clone (TX completion) before original skb
629 * This test would have no chance to be true for the clone,
630 * while here, branch prediction will be good.
631 */
632 if (atomic_read(&fclones->fclone_ref) == 1)
633 goto fastpath;
634 break;
635
636 default: /* SKB_FCLONE_CLONE */
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
638 break;
639 }
640 if (!atomic_dec_and_test(&fclones->fclone_ref))
641 return;
642fastpath:
643 kmem_cache_free(skbuff_fclone_cache, fclones);
644}
645
646static void skb_release_head_state(struct sk_buff *skb)
647{
648 skb_dst_drop(skb);
649#ifdef CONFIG_XFRM
650 secpath_put(skb->sp);
651#endif
652 if (skb->destructor) {
653 WARN_ON(in_irq());
654 skb->destructor(skb);
655 }
656#if IS_ENABLED(CONFIG_NF_CONNTRACK)
657 nf_conntrack_put(skb->nfct);
658#endif
659#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
660 nf_bridge_put(skb->nf_bridge);
661#endif
662}
663
664/* Free everything but the sk_buff shell. */
665static void skb_release_all(struct sk_buff *skb)
666{
667 skb_release_head_state(skb);
668 if (likely(skb->head))
669 skb_release_data(skb);
670}
671
672/**
673 * __kfree_skb - private function
674 * @skb: buffer
675 *
676 * Free an sk_buff. Release anything attached to the buffer.
677 * Clean the state. This is an internal helper function. Users should
678 * always call kfree_skb
679 */
680
681void __kfree_skb(struct sk_buff *skb)
682{
683 skb_release_all(skb);
684 kfree_skbmem(skb);
685}
686EXPORT_SYMBOL(__kfree_skb);
687
688/**
689 * kfree_skb - free an sk_buff
690 * @skb: buffer to free
691 *
692 * Drop a reference to the buffer and free it if the usage count has
693 * hit zero.
694 */
695void kfree_skb(struct sk_buff *skb)
696{
697 if (unlikely(!skb))
698 return;
699 if (likely(atomic_read(&skb->users) == 1))
700 smp_rmb();
701 else if (likely(!atomic_dec_and_test(&skb->users)))
702 return;
703 trace_kfree_skb(skb, __builtin_return_address(0));
704 __kfree_skb(skb);
705}
706EXPORT_SYMBOL(kfree_skb);
707
708void kfree_skb_list(struct sk_buff *segs)
709{
710 while (segs) {
711 struct sk_buff *next = segs->next;
712
713 kfree_skb(segs);
714 segs = next;
715 }
716}
717EXPORT_SYMBOL(kfree_skb_list);
718
719/**
720 * skb_tx_error - report an sk_buff xmit error
721 * @skb: buffer that triggered an error
722 *
723 * Report xmit error if a device callback is tracking this skb.
724 * skb must be freed afterwards.
725 */
726void skb_tx_error(struct sk_buff *skb)
727{
728 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
729 struct ubuf_info *uarg;
730
731 uarg = skb_shinfo(skb)->destructor_arg;
732 if (uarg->callback)
733 uarg->callback(uarg, false);
734 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
735 }
736}
737EXPORT_SYMBOL(skb_tx_error);
738
739/**
740 * consume_skb - free an skbuff
741 * @skb: buffer to free
742 *
743 * Drop a ref to the buffer and free it if the usage count has hit zero
744 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
745 * is being dropped after a failure and notes that
746 */
747void consume_skb(struct sk_buff *skb)
748{
749 if (unlikely(!skb))
750 return;
751 if (likely(atomic_read(&skb->users) == 1))
752 smp_rmb();
753 else if (likely(!atomic_dec_and_test(&skb->users)))
754 return;
755 trace_consume_skb(skb);
756 __kfree_skb(skb);
757}
758EXPORT_SYMBOL(consume_skb);
759
760void __kfree_skb_flush(void)
761{
762 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
763
764 /* flush skb_cache if containing objects */
765 if (nc->skb_count) {
766 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
767 nc->skb_cache);
768 nc->skb_count = 0;
769 }
770}
771
772static inline void _kfree_skb_defer(struct sk_buff *skb)
773{
774 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
775
776 /* drop skb->head and call any destructors for packet */
777 skb_release_all(skb);
778
779 /* record skb to CPU local list */
780 nc->skb_cache[nc->skb_count++] = skb;
781
782#ifdef CONFIG_SLUB
783 /* SLUB writes into objects when freeing */
784 prefetchw(skb);
785#endif
786
787 /* flush skb_cache if it is filled */
788 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
789 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
790 nc->skb_cache);
791 nc->skb_count = 0;
792 }
793}
794void __kfree_skb_defer(struct sk_buff *skb)
795{
796 _kfree_skb_defer(skb);
797}
798
799void napi_consume_skb(struct sk_buff *skb, int budget)
800{
801 if (unlikely(!skb))
802 return;
803
804 /* Zero budget indicate non-NAPI context called us, like netpoll */
805 if (unlikely(!budget)) {
806 dev_consume_skb_any(skb);
807 return;
808 }
809
810 if (likely(atomic_read(&skb->users) == 1))
811 smp_rmb();
812 else if (likely(!atomic_dec_and_test(&skb->users)))
813 return;
814 /* if reaching here SKB is ready to free */
815 trace_consume_skb(skb);
816
817 /* if SKB is a clone, don't handle this case */
818 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
819 __kfree_skb(skb);
820 return;
821 }
822
823 _kfree_skb_defer(skb);
824}
825EXPORT_SYMBOL(napi_consume_skb);
826
827/* Make sure a field is enclosed inside headers_start/headers_end section */
828#define CHECK_SKB_FIELD(field) \
829 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
830 offsetof(struct sk_buff, headers_start)); \
831 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
832 offsetof(struct sk_buff, headers_end)); \
833
834static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
835{
836 new->tstamp = old->tstamp;
837 /* We do not copy old->sk */
838 new->dev = old->dev;
839 memcpy(new->cb, old->cb, sizeof(old->cb));
840 skb_dst_copy(new, old);
841#ifdef CONFIG_XFRM
842 new->sp = secpath_get(old->sp);
843#endif
844 __nf_copy(new, old, false);
845
846 /* Note : this field could be in headers_start/headers_end section
847 * It is not yet because we do not want to have a 16 bit hole
848 */
849 new->queue_mapping = old->queue_mapping;
850
851 memcpy(&new->headers_start, &old->headers_start,
852 offsetof(struct sk_buff, headers_end) -
853 offsetof(struct sk_buff, headers_start));
854 CHECK_SKB_FIELD(protocol);
855 CHECK_SKB_FIELD(csum);
856 CHECK_SKB_FIELD(hash);
857 CHECK_SKB_FIELD(priority);
858 CHECK_SKB_FIELD(skb_iif);
859 CHECK_SKB_FIELD(vlan_proto);
860 CHECK_SKB_FIELD(vlan_tci);
861 CHECK_SKB_FIELD(transport_header);
862 CHECK_SKB_FIELD(network_header);
863 CHECK_SKB_FIELD(mac_header);
864 CHECK_SKB_FIELD(inner_protocol);
865 CHECK_SKB_FIELD(inner_transport_header);
866 CHECK_SKB_FIELD(inner_network_header);
867 CHECK_SKB_FIELD(inner_mac_header);
868 CHECK_SKB_FIELD(mark);
869#ifdef CONFIG_NETWORK_SECMARK
870 CHECK_SKB_FIELD(secmark);
871#endif
872#ifdef CONFIG_NET_RX_BUSY_POLL
873 CHECK_SKB_FIELD(napi_id);
874#endif
875#ifdef CONFIG_XPS
876 CHECK_SKB_FIELD(sender_cpu);
877#endif
878#ifdef CONFIG_NET_SCHED
879 CHECK_SKB_FIELD(tc_index);
880#ifdef CONFIG_NET_CLS_ACT
881 CHECK_SKB_FIELD(tc_verd);
882#endif
883#endif
884
885}
886
887/*
888 * You should not add any new code to this function. Add it to
889 * __copy_skb_header above instead.
890 */
891static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
892{
893#define C(x) n->x = skb->x
894
895 n->next = n->prev = NULL;
896 n->sk = NULL;
897 __copy_skb_header(n, skb);
898
899 C(len);
900 C(data_len);
901 C(mac_len);
902 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
903 n->cloned = 1;
904 n->nohdr = 0;
905 n->destructor = NULL;
906 C(tail);
907 C(end);
908 C(head);
909 C(head_frag);
910 C(data);
911 C(truesize);
912 atomic_set(&n->users, 1);
913
914 atomic_inc(&(skb_shinfo(skb)->dataref));
915 skb->cloned = 1;
916
917 return n;
918#undef C
919}
920
921/**
922 * skb_morph - morph one skb into another
923 * @dst: the skb to receive the contents
924 * @src: the skb to supply the contents
925 *
926 * This is identical to skb_clone except that the target skb is
927 * supplied by the user.
928 *
929 * The target skb is returned upon exit.
930 */
931struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
932{
933 skb_release_all(dst);
934 return __skb_clone(dst, src);
935}
936EXPORT_SYMBOL_GPL(skb_morph);
937
938/**
939 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
940 * @skb: the skb to modify
941 * @gfp_mask: allocation priority
942 *
943 * This must be called on SKBTX_DEV_ZEROCOPY skb.
944 * It will copy all frags into kernel and drop the reference
945 * to userspace pages.
946 *
947 * If this function is called from an interrupt gfp_mask() must be
948 * %GFP_ATOMIC.
949 *
950 * Returns 0 on success or a negative error code on failure
951 * to allocate kernel memory to copy to.
952 */
953int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
954{
955 int i;
956 int num_frags = skb_shinfo(skb)->nr_frags;
957 struct page *page, *head = NULL;
958 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
959
960 for (i = 0; i < num_frags; i++) {
961 u8 *vaddr;
962 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
963
964 page = alloc_page(gfp_mask);
965 if (!page) {
966 while (head) {
967 struct page *next = (struct page *)page_private(head);
968 put_page(head);
969 head = next;
970 }
971 return -ENOMEM;
972 }
973 vaddr = kmap_atomic(skb_frag_page(f));
974 memcpy(page_address(page),
975 vaddr + f->page_offset, skb_frag_size(f));
976 kunmap_atomic(vaddr);
977 set_page_private(page, (unsigned long)head);
978 head = page;
979 }
980
981 /* skb frags release userspace buffers */
982 for (i = 0; i < num_frags; i++)
983 skb_frag_unref(skb, i);
984
985 uarg->callback(uarg, false);
986
987 /* skb frags point to kernel buffers */
988 for (i = num_frags - 1; i >= 0; i--) {
989 __skb_fill_page_desc(skb, i, head, 0,
990 skb_shinfo(skb)->frags[i].size);
991 head = (struct page *)page_private(head);
992 }
993
994 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
995 return 0;
996}
997EXPORT_SYMBOL_GPL(skb_copy_ubufs);
998
999/**
1000 * skb_clone - duplicate an sk_buff
1001 * @skb: buffer to clone
1002 * @gfp_mask: allocation priority
1003 *
1004 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1005 * copies share the same packet data but not structure. The new
1006 * buffer has a reference count of 1. If the allocation fails the
1007 * function returns %NULL otherwise the new buffer is returned.
1008 *
1009 * If this function is called from an interrupt gfp_mask() must be
1010 * %GFP_ATOMIC.
1011 */
1012
1013struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1014{
1015 struct sk_buff_fclones *fclones = container_of(skb,
1016 struct sk_buff_fclones,
1017 skb1);
1018 struct sk_buff *n;
1019
1020 if (skb_orphan_frags(skb, gfp_mask))
1021 return NULL;
1022
1023 if (skb->fclone == SKB_FCLONE_ORIG &&
1024 atomic_read(&fclones->fclone_ref) == 1) {
1025 n = &fclones->skb2;
1026 atomic_set(&fclones->fclone_ref, 2);
1027 } else {
1028 if (skb_pfmemalloc(skb))
1029 gfp_mask |= __GFP_MEMALLOC;
1030
1031 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1032 if (!n)
1033 return NULL;
1034
1035 kmemcheck_annotate_bitfield(n, flags1);
1036 n->fclone = SKB_FCLONE_UNAVAILABLE;
1037 }
1038
1039 return __skb_clone(n, skb);
1040}
1041EXPORT_SYMBOL(skb_clone);
1042
1043static void skb_headers_offset_update(struct sk_buff *skb, int off)
1044{
1045 /* Only adjust this if it actually is csum_start rather than csum */
1046 if (skb->ip_summed == CHECKSUM_PARTIAL)
1047 skb->csum_start += off;
1048 /* {transport,network,mac}_header and tail are relative to skb->head */
1049 skb->transport_header += off;
1050 skb->network_header += off;
1051 if (skb_mac_header_was_set(skb))
1052 skb->mac_header += off;
1053 skb->inner_transport_header += off;
1054 skb->inner_network_header += off;
1055 skb->inner_mac_header += off;
1056}
1057
1058static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1059{
1060 __copy_skb_header(new, old);
1061
1062 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1063 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1064 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1065}
1066
1067static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1068{
1069 if (skb_pfmemalloc(skb))
1070 return SKB_ALLOC_RX;
1071 return 0;
1072}
1073
1074/**
1075 * skb_copy - create private copy of an sk_buff
1076 * @skb: buffer to copy
1077 * @gfp_mask: allocation priority
1078 *
1079 * Make a copy of both an &sk_buff and its data. This is used when the
1080 * caller wishes to modify the data and needs a private copy of the
1081 * data to alter. Returns %NULL on failure or the pointer to the buffer
1082 * on success. The returned buffer has a reference count of 1.
1083 *
1084 * As by-product this function converts non-linear &sk_buff to linear
1085 * one, so that &sk_buff becomes completely private and caller is allowed
1086 * to modify all the data of returned buffer. This means that this
1087 * function is not recommended for use in circumstances when only
1088 * header is going to be modified. Use pskb_copy() instead.
1089 */
1090
1091struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1092{
1093 int headerlen = skb_headroom(skb);
1094 unsigned int size = skb_end_offset(skb) + skb->data_len;
1095 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1096 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1097
1098 if (!n)
1099 return NULL;
1100
1101 /* Set the data pointer */
1102 skb_reserve(n, headerlen);
1103 /* Set the tail pointer and length */
1104 skb_put(n, skb->len);
1105
1106 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1107 BUG();
1108
1109 copy_skb_header(n, skb);
1110 return n;
1111}
1112EXPORT_SYMBOL(skb_copy);
1113
1114/**
1115 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1116 * @skb: buffer to copy
1117 * @headroom: headroom of new skb
1118 * @gfp_mask: allocation priority
1119 * @fclone: if true allocate the copy of the skb from the fclone
1120 * cache instead of the head cache; it is recommended to set this
1121 * to true for the cases where the copy will likely be cloned
1122 *
1123 * Make a copy of both an &sk_buff and part of its data, located
1124 * in header. Fragmented data remain shared. This is used when
1125 * the caller wishes to modify only header of &sk_buff and needs
1126 * private copy of the header to alter. Returns %NULL on failure
1127 * or the pointer to the buffer on success.
1128 * The returned buffer has a reference count of 1.
1129 */
1130
1131struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1132 gfp_t gfp_mask, bool fclone)
1133{
1134 unsigned int size = skb_headlen(skb) + headroom;
1135 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1136 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1137
1138 if (!n)
1139 goto out;
1140
1141 /* Set the data pointer */
1142 skb_reserve(n, headroom);
1143 /* Set the tail pointer and length */
1144 skb_put(n, skb_headlen(skb));
1145 /* Copy the bytes */
1146 skb_copy_from_linear_data(skb, n->data, n->len);
1147
1148 n->truesize += skb->data_len;
1149 n->data_len = skb->data_len;
1150 n->len = skb->len;
1151
1152 if (skb_shinfo(skb)->nr_frags) {
1153 int i;
1154
1155 if (skb_orphan_frags(skb, gfp_mask)) {
1156 kfree_skb(n);
1157 n = NULL;
1158 goto out;
1159 }
1160 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1162 skb_frag_ref(skb, i);
1163 }
1164 skb_shinfo(n)->nr_frags = i;
1165 }
1166
1167 if (skb_has_frag_list(skb)) {
1168 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1169 skb_clone_fraglist(n);
1170 }
1171
1172 copy_skb_header(n, skb);
1173out:
1174 return n;
1175}
1176EXPORT_SYMBOL(__pskb_copy_fclone);
1177
1178/**
1179 * pskb_expand_head - reallocate header of &sk_buff
1180 * @skb: buffer to reallocate
1181 * @nhead: room to add at head
1182 * @ntail: room to add at tail
1183 * @gfp_mask: allocation priority
1184 *
1185 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1186 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1187 * reference count of 1. Returns zero in the case of success or error,
1188 * if expansion failed. In the last case, &sk_buff is not changed.
1189 *
1190 * All the pointers pointing into skb header may change and must be
1191 * reloaded after call to this function.
1192 */
1193
1194int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1195 gfp_t gfp_mask)
1196{
1197 int i;
1198 u8 *data;
1199 int size = nhead + skb_end_offset(skb) + ntail;
1200 long off;
1201
1202 BUG_ON(nhead < 0);
1203
1204 if (skb_shared(skb))
1205 BUG();
1206
1207 size = SKB_DATA_ALIGN(size);
1208
1209 if (skb_pfmemalloc(skb))
1210 gfp_mask |= __GFP_MEMALLOC;
1211 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1212 gfp_mask, NUMA_NO_NODE, NULL);
1213 if (!data)
1214 goto nodata;
1215 size = SKB_WITH_OVERHEAD(ksize(data));
1216
1217 /* Copy only real data... and, alas, header. This should be
1218 * optimized for the cases when header is void.
1219 */
1220 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1221
1222 memcpy((struct skb_shared_info *)(data + size),
1223 skb_shinfo(skb),
1224 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1225
1226 /*
1227 * if shinfo is shared we must drop the old head gracefully, but if it
1228 * is not we can just drop the old head and let the existing refcount
1229 * be since all we did is relocate the values
1230 */
1231 if (skb_cloned(skb)) {
1232 /* copy this zero copy skb frags */
1233 if (skb_orphan_frags(skb, gfp_mask))
1234 goto nofrags;
1235 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1236 skb_frag_ref(skb, i);
1237
1238 if (skb_has_frag_list(skb))
1239 skb_clone_fraglist(skb);
1240
1241 skb_release_data(skb);
1242 } else {
1243 skb_free_head(skb);
1244 }
1245 off = (data + nhead) - skb->head;
1246
1247 skb->head = data;
1248 skb->head_frag = 0;
1249 skb->data += off;
1250#ifdef NET_SKBUFF_DATA_USES_OFFSET
1251 skb->end = size;
1252 off = nhead;
1253#else
1254 skb->end = skb->head + size;
1255#endif
1256 skb->tail += off;
1257 skb_headers_offset_update(skb, nhead);
1258 skb->cloned = 0;
1259 skb->hdr_len = 0;
1260 skb->nohdr = 0;
1261 atomic_set(&skb_shinfo(skb)->dataref, 1);
1262 return 0;
1263
1264nofrags:
1265 kfree(data);
1266nodata:
1267 return -ENOMEM;
1268}
1269EXPORT_SYMBOL(pskb_expand_head);
1270
1271/* Make private copy of skb with writable head and some headroom */
1272
1273struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1274{
1275 struct sk_buff *skb2;
1276 int delta = headroom - skb_headroom(skb);
1277
1278 if (delta <= 0)
1279 skb2 = pskb_copy(skb, GFP_ATOMIC);
1280 else {
1281 skb2 = skb_clone(skb, GFP_ATOMIC);
1282 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1283 GFP_ATOMIC)) {
1284 kfree_skb(skb2);
1285 skb2 = NULL;
1286 }
1287 }
1288 return skb2;
1289}
1290EXPORT_SYMBOL(skb_realloc_headroom);
1291
1292/**
1293 * skb_copy_expand - copy and expand sk_buff
1294 * @skb: buffer to copy
1295 * @newheadroom: new free bytes at head
1296 * @newtailroom: new free bytes at tail
1297 * @gfp_mask: allocation priority
1298 *
1299 * Make a copy of both an &sk_buff and its data and while doing so
1300 * allocate additional space.
1301 *
1302 * This is used when the caller wishes to modify the data and needs a
1303 * private copy of the data to alter as well as more space for new fields.
1304 * Returns %NULL on failure or the pointer to the buffer
1305 * on success. The returned buffer has a reference count of 1.
1306 *
1307 * You must pass %GFP_ATOMIC as the allocation priority if this function
1308 * is called from an interrupt.
1309 */
1310struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1311 int newheadroom, int newtailroom,
1312 gfp_t gfp_mask)
1313{
1314 /*
1315 * Allocate the copy buffer
1316 */
1317 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1318 gfp_mask, skb_alloc_rx_flag(skb),
1319 NUMA_NO_NODE);
1320 int oldheadroom = skb_headroom(skb);
1321 int head_copy_len, head_copy_off;
1322
1323 if (!n)
1324 return NULL;
1325
1326 skb_reserve(n, newheadroom);
1327
1328 /* Set the tail pointer and length */
1329 skb_put(n, skb->len);
1330
1331 head_copy_len = oldheadroom;
1332 head_copy_off = 0;
1333 if (newheadroom <= head_copy_len)
1334 head_copy_len = newheadroom;
1335 else
1336 head_copy_off = newheadroom - head_copy_len;
1337
1338 /* Copy the linear header and data. */
1339 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1340 skb->len + head_copy_len))
1341 BUG();
1342
1343 copy_skb_header(n, skb);
1344
1345 skb_headers_offset_update(n, newheadroom - oldheadroom);
1346
1347 return n;
1348}
1349EXPORT_SYMBOL(skb_copy_expand);
1350
1351/**
1352 * skb_pad - zero pad the tail of an skb
1353 * @skb: buffer to pad
1354 * @pad: space to pad
1355 *
1356 * Ensure that a buffer is followed by a padding area that is zero
1357 * filled. Used by network drivers which may DMA or transfer data
1358 * beyond the buffer end onto the wire.
1359 *
1360 * May return error in out of memory cases. The skb is freed on error.
1361 */
1362
1363int skb_pad(struct sk_buff *skb, int pad)
1364{
1365 int err;
1366 int ntail;
1367
1368 /* If the skbuff is non linear tailroom is always zero.. */
1369 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1370 memset(skb->data+skb->len, 0, pad);
1371 return 0;
1372 }
1373
1374 ntail = skb->data_len + pad - (skb->end - skb->tail);
1375 if (likely(skb_cloned(skb) || ntail > 0)) {
1376 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1377 if (unlikely(err))
1378 goto free_skb;
1379 }
1380
1381 /* FIXME: The use of this function with non-linear skb's really needs
1382 * to be audited.
1383 */
1384 err = skb_linearize(skb);
1385 if (unlikely(err))
1386 goto free_skb;
1387
1388 memset(skb->data + skb->len, 0, pad);
1389 return 0;
1390
1391free_skb:
1392 kfree_skb(skb);
1393 return err;
1394}
1395EXPORT_SYMBOL(skb_pad);
1396
1397/**
1398 * pskb_put - add data to the tail of a potentially fragmented buffer
1399 * @skb: start of the buffer to use
1400 * @tail: tail fragment of the buffer to use
1401 * @len: amount of data to add
1402 *
1403 * This function extends the used data area of the potentially
1404 * fragmented buffer. @tail must be the last fragment of @skb -- or
1405 * @skb itself. If this would exceed the total buffer size the kernel
1406 * will panic. A pointer to the first byte of the extra data is
1407 * returned.
1408 */
1409
1410unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1411{
1412 if (tail != skb) {
1413 skb->data_len += len;
1414 skb->len += len;
1415 }
1416 return skb_put(tail, len);
1417}
1418EXPORT_SYMBOL_GPL(pskb_put);
1419
1420/**
1421 * skb_put - add data to a buffer
1422 * @skb: buffer to use
1423 * @len: amount of data to add
1424 *
1425 * This function extends the used data area of the buffer. If this would
1426 * exceed the total buffer size the kernel will panic. A pointer to the
1427 * first byte of the extra data is returned.
1428 */
1429unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1430{
1431 unsigned char *tmp = skb_tail_pointer(skb);
1432 SKB_LINEAR_ASSERT(skb);
1433 skb->tail += len;
1434 skb->len += len;
1435 if (unlikely(skb->tail > skb->end))
1436 skb_over_panic(skb, len, __builtin_return_address(0));
1437 return tmp;
1438}
1439EXPORT_SYMBOL(skb_put);
1440
1441/**
1442 * skb_push - add data to the start of a buffer
1443 * @skb: buffer to use
1444 * @len: amount of data to add
1445 *
1446 * This function extends the used data area of the buffer at the buffer
1447 * start. If this would exceed the total buffer headroom the kernel will
1448 * panic. A pointer to the first byte of the extra data is returned.
1449 */
1450unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1451{
1452 skb->data -= len;
1453 skb->len += len;
1454 if (unlikely(skb->data<skb->head))
1455 skb_under_panic(skb, len, __builtin_return_address(0));
1456 return skb->data;
1457}
1458EXPORT_SYMBOL(skb_push);
1459
1460/**
1461 * skb_pull - remove data from the start of a buffer
1462 * @skb: buffer to use
1463 * @len: amount of data to remove
1464 *
1465 * This function removes data from the start of a buffer, returning
1466 * the memory to the headroom. A pointer to the next data in the buffer
1467 * is returned. Once the data has been pulled future pushes will overwrite
1468 * the old data.
1469 */
1470unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1471{
1472 return skb_pull_inline(skb, len);
1473}
1474EXPORT_SYMBOL(skb_pull);
1475
1476/**
1477 * skb_trim - remove end from a buffer
1478 * @skb: buffer to alter
1479 * @len: new length
1480 *
1481 * Cut the length of a buffer down by removing data from the tail. If
1482 * the buffer is already under the length specified it is not modified.
1483 * The skb must be linear.
1484 */
1485void skb_trim(struct sk_buff *skb, unsigned int len)
1486{
1487 if (skb->len > len)
1488 __skb_trim(skb, len);
1489}
1490EXPORT_SYMBOL(skb_trim);
1491
1492/* Trims skb to length len. It can change skb pointers.
1493 */
1494
1495int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1496{
1497 struct sk_buff **fragp;
1498 struct sk_buff *frag;
1499 int offset = skb_headlen(skb);
1500 int nfrags = skb_shinfo(skb)->nr_frags;
1501 int i;
1502 int err;
1503
1504 if (skb_cloned(skb) &&
1505 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1506 return err;
1507
1508 i = 0;
1509 if (offset >= len)
1510 goto drop_pages;
1511
1512 for (; i < nfrags; i++) {
1513 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1514
1515 if (end < len) {
1516 offset = end;
1517 continue;
1518 }
1519
1520 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1521
1522drop_pages:
1523 skb_shinfo(skb)->nr_frags = i;
1524
1525 for (; i < nfrags; i++)
1526 skb_frag_unref(skb, i);
1527
1528 if (skb_has_frag_list(skb))
1529 skb_drop_fraglist(skb);
1530 goto done;
1531 }
1532
1533 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1534 fragp = &frag->next) {
1535 int end = offset + frag->len;
1536
1537 if (skb_shared(frag)) {
1538 struct sk_buff *nfrag;
1539
1540 nfrag = skb_clone(frag, GFP_ATOMIC);
1541 if (unlikely(!nfrag))
1542 return -ENOMEM;
1543
1544 nfrag->next = frag->next;
1545 consume_skb(frag);
1546 frag = nfrag;
1547 *fragp = frag;
1548 }
1549
1550 if (end < len) {
1551 offset = end;
1552 continue;
1553 }
1554
1555 if (end > len &&
1556 unlikely((err = pskb_trim(frag, len - offset))))
1557 return err;
1558
1559 if (frag->next)
1560 skb_drop_list(&frag->next);
1561 break;
1562 }
1563
1564done:
1565 if (len > skb_headlen(skb)) {
1566 skb->data_len -= skb->len - len;
1567 skb->len = len;
1568 } else {
1569 skb->len = len;
1570 skb->data_len = 0;
1571 skb_set_tail_pointer(skb, len);
1572 }
1573
1574 return 0;
1575}
1576EXPORT_SYMBOL(___pskb_trim);
1577
1578/**
1579 * __pskb_pull_tail - advance tail of skb header
1580 * @skb: buffer to reallocate
1581 * @delta: number of bytes to advance tail
1582 *
1583 * The function makes a sense only on a fragmented &sk_buff,
1584 * it expands header moving its tail forward and copying necessary
1585 * data from fragmented part.
1586 *
1587 * &sk_buff MUST have reference count of 1.
1588 *
1589 * Returns %NULL (and &sk_buff does not change) if pull failed
1590 * or value of new tail of skb in the case of success.
1591 *
1592 * All the pointers pointing into skb header may change and must be
1593 * reloaded after call to this function.
1594 */
1595
1596/* Moves tail of skb head forward, copying data from fragmented part,
1597 * when it is necessary.
1598 * 1. It may fail due to malloc failure.
1599 * 2. It may change skb pointers.
1600 *
1601 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1602 */
1603unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1604{
1605 /* If skb has not enough free space at tail, get new one
1606 * plus 128 bytes for future expansions. If we have enough
1607 * room at tail, reallocate without expansion only if skb is cloned.
1608 */
1609 int i, k, eat = (skb->tail + delta) - skb->end;
1610
1611 if (eat > 0 || skb_cloned(skb)) {
1612 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1613 GFP_ATOMIC))
1614 return NULL;
1615 }
1616
1617 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1618 BUG();
1619
1620 /* Optimization: no fragments, no reasons to preestimate
1621 * size of pulled pages. Superb.
1622 */
1623 if (!skb_has_frag_list(skb))
1624 goto pull_pages;
1625
1626 /* Estimate size of pulled pages. */
1627 eat = delta;
1628 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1629 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1630
1631 if (size >= eat)
1632 goto pull_pages;
1633 eat -= size;
1634 }
1635
1636 /* If we need update frag list, we are in troubles.
1637 * Certainly, it possible to add an offset to skb data,
1638 * but taking into account that pulling is expected to
1639 * be very rare operation, it is worth to fight against
1640 * further bloating skb head and crucify ourselves here instead.
1641 * Pure masohism, indeed. 8)8)
1642 */
1643 if (eat) {
1644 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1645 struct sk_buff *clone = NULL;
1646 struct sk_buff *insp = NULL;
1647
1648 do {
1649 BUG_ON(!list);
1650
1651 if (list->len <= eat) {
1652 /* Eaten as whole. */
1653 eat -= list->len;
1654 list = list->next;
1655 insp = list;
1656 } else {
1657 /* Eaten partially. */
1658
1659 if (skb_shared(list)) {
1660 /* Sucks! We need to fork list. :-( */
1661 clone = skb_clone(list, GFP_ATOMIC);
1662 if (!clone)
1663 return NULL;
1664 insp = list->next;
1665 list = clone;
1666 } else {
1667 /* This may be pulled without
1668 * problems. */
1669 insp = list;
1670 }
1671 if (!pskb_pull(list, eat)) {
1672 kfree_skb(clone);
1673 return NULL;
1674 }
1675 break;
1676 }
1677 } while (eat);
1678
1679 /* Free pulled out fragments. */
1680 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1681 skb_shinfo(skb)->frag_list = list->next;
1682 kfree_skb(list);
1683 }
1684 /* And insert new clone at head. */
1685 if (clone) {
1686 clone->next = list;
1687 skb_shinfo(skb)->frag_list = clone;
1688 }
1689 }
1690 /* Success! Now we may commit changes to skb data. */
1691
1692pull_pages:
1693 eat = delta;
1694 k = 0;
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1696 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1697
1698 if (size <= eat) {
1699 skb_frag_unref(skb, i);
1700 eat -= size;
1701 } else {
1702 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1703 if (eat) {
1704 skb_shinfo(skb)->frags[k].page_offset += eat;
1705 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1706 eat = 0;
1707 }
1708 k++;
1709 }
1710 }
1711 skb_shinfo(skb)->nr_frags = k;
1712
1713 skb->tail += delta;
1714 skb->data_len -= delta;
1715
1716 return skb_tail_pointer(skb);
1717}
1718EXPORT_SYMBOL(__pskb_pull_tail);
1719
1720/**
1721 * skb_copy_bits - copy bits from skb to kernel buffer
1722 * @skb: source skb
1723 * @offset: offset in source
1724 * @to: destination buffer
1725 * @len: number of bytes to copy
1726 *
1727 * Copy the specified number of bytes from the source skb to the
1728 * destination buffer.
1729 *
1730 * CAUTION ! :
1731 * If its prototype is ever changed,
1732 * check arch/{*}/net/{*}.S files,
1733 * since it is called from BPF assembly code.
1734 */
1735int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1736{
1737 int start = skb_headlen(skb);
1738 struct sk_buff *frag_iter;
1739 int i, copy;
1740
1741 if (offset > (int)skb->len - len)
1742 goto fault;
1743
1744 /* Copy header. */
1745 if ((copy = start - offset) > 0) {
1746 if (copy > len)
1747 copy = len;
1748 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1749 if ((len -= copy) == 0)
1750 return 0;
1751 offset += copy;
1752 to += copy;
1753 }
1754
1755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1756 int end;
1757 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1758
1759 WARN_ON(start > offset + len);
1760
1761 end = start + skb_frag_size(f);
1762 if ((copy = end - offset) > 0) {
1763 u8 *vaddr;
1764
1765 if (copy > len)
1766 copy = len;
1767
1768 vaddr = kmap_atomic(skb_frag_page(f));
1769 memcpy(to,
1770 vaddr + f->page_offset + offset - start,
1771 copy);
1772 kunmap_atomic(vaddr);
1773
1774 if ((len -= copy) == 0)
1775 return 0;
1776 offset += copy;
1777 to += copy;
1778 }
1779 start = end;
1780 }
1781
1782 skb_walk_frags(skb, frag_iter) {
1783 int end;
1784
1785 WARN_ON(start > offset + len);
1786
1787 end = start + frag_iter->len;
1788 if ((copy = end - offset) > 0) {
1789 if (copy > len)
1790 copy = len;
1791 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1792 goto fault;
1793 if ((len -= copy) == 0)
1794 return 0;
1795 offset += copy;
1796 to += copy;
1797 }
1798 start = end;
1799 }
1800
1801 if (!len)
1802 return 0;
1803
1804fault:
1805 return -EFAULT;
1806}
1807EXPORT_SYMBOL(skb_copy_bits);
1808
1809/*
1810 * Callback from splice_to_pipe(), if we need to release some pages
1811 * at the end of the spd in case we error'ed out in filling the pipe.
1812 */
1813static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1814{
1815 put_page(spd->pages[i]);
1816}
1817
1818static struct page *linear_to_page(struct page *page, unsigned int *len,
1819 unsigned int *offset,
1820 struct sock *sk)
1821{
1822 struct page_frag *pfrag = sk_page_frag(sk);
1823
1824 if (!sk_page_frag_refill(sk, pfrag))
1825 return NULL;
1826
1827 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1828
1829 memcpy(page_address(pfrag->page) + pfrag->offset,
1830 page_address(page) + *offset, *len);
1831 *offset = pfrag->offset;
1832 pfrag->offset += *len;
1833
1834 return pfrag->page;
1835}
1836
1837static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1838 struct page *page,
1839 unsigned int offset)
1840{
1841 return spd->nr_pages &&
1842 spd->pages[spd->nr_pages - 1] == page &&
1843 (spd->partial[spd->nr_pages - 1].offset +
1844 spd->partial[spd->nr_pages - 1].len == offset);
1845}
1846
1847/*
1848 * Fill page/offset/length into spd, if it can hold more pages.
1849 */
1850static bool spd_fill_page(struct splice_pipe_desc *spd,
1851 struct pipe_inode_info *pipe, struct page *page,
1852 unsigned int *len, unsigned int offset,
1853 bool linear,
1854 struct sock *sk)
1855{
1856 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1857 return true;
1858
1859 if (linear) {
1860 page = linear_to_page(page, len, &offset, sk);
1861 if (!page)
1862 return true;
1863 }
1864 if (spd_can_coalesce(spd, page, offset)) {
1865 spd->partial[spd->nr_pages - 1].len += *len;
1866 return false;
1867 }
1868 get_page(page);
1869 spd->pages[spd->nr_pages] = page;
1870 spd->partial[spd->nr_pages].len = *len;
1871 spd->partial[spd->nr_pages].offset = offset;
1872 spd->nr_pages++;
1873
1874 return false;
1875}
1876
1877static bool __splice_segment(struct page *page, unsigned int poff,
1878 unsigned int plen, unsigned int *off,
1879 unsigned int *len,
1880 struct splice_pipe_desc *spd, bool linear,
1881 struct sock *sk,
1882 struct pipe_inode_info *pipe)
1883{
1884 if (!*len)
1885 return true;
1886
1887 /* skip this segment if already processed */
1888 if (*off >= plen) {
1889 *off -= plen;
1890 return false;
1891 }
1892
1893 /* ignore any bits we already processed */
1894 poff += *off;
1895 plen -= *off;
1896 *off = 0;
1897
1898 do {
1899 unsigned int flen = min(*len, plen);
1900
1901 if (spd_fill_page(spd, pipe, page, &flen, poff,
1902 linear, sk))
1903 return true;
1904 poff += flen;
1905 plen -= flen;
1906 *len -= flen;
1907 } while (*len && plen);
1908
1909 return false;
1910}
1911
1912/*
1913 * Map linear and fragment data from the skb to spd. It reports true if the
1914 * pipe is full or if we already spliced the requested length.
1915 */
1916static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1917 unsigned int *offset, unsigned int *len,
1918 struct splice_pipe_desc *spd, struct sock *sk)
1919{
1920 int seg;
1921 struct sk_buff *iter;
1922
1923 /* map the linear part :
1924 * If skb->head_frag is set, this 'linear' part is backed by a
1925 * fragment, and if the head is not shared with any clones then
1926 * we can avoid a copy since we own the head portion of this page.
1927 */
1928 if (__splice_segment(virt_to_page(skb->data),
1929 (unsigned long) skb->data & (PAGE_SIZE - 1),
1930 skb_headlen(skb),
1931 offset, len, spd,
1932 skb_head_is_locked(skb),
1933 sk, pipe))
1934 return true;
1935
1936 /*
1937 * then map the fragments
1938 */
1939 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1940 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1941
1942 if (__splice_segment(skb_frag_page(f),
1943 f->page_offset, skb_frag_size(f),
1944 offset, len, spd, false, sk, pipe))
1945 return true;
1946 }
1947
1948 skb_walk_frags(skb, iter) {
1949 if (*offset >= iter->len) {
1950 *offset -= iter->len;
1951 continue;
1952 }
1953 /* __skb_splice_bits() only fails if the output has no room
1954 * left, so no point in going over the frag_list for the error
1955 * case.
1956 */
1957 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1958 return true;
1959 }
1960
1961 return false;
1962}
1963
1964ssize_t skb_socket_splice(struct sock *sk,
1965 struct pipe_inode_info *pipe,
1966 struct splice_pipe_desc *spd)
1967{
1968 int ret;
1969
1970 /* Drop the socket lock, otherwise we have reverse
1971 * locking dependencies between sk_lock and i_mutex
1972 * here as compared to sendfile(). We enter here
1973 * with the socket lock held, and splice_to_pipe() will
1974 * grab the pipe inode lock. For sendfile() emulation,
1975 * we call into ->sendpage() with the i_mutex lock held
1976 * and networking will grab the socket lock.
1977 */
1978 release_sock(sk);
1979 ret = splice_to_pipe(pipe, spd);
1980 lock_sock(sk);
1981
1982 return ret;
1983}
1984
1985/*
1986 * Map data from the skb to a pipe. Should handle both the linear part,
1987 * the fragments, and the frag list.
1988 */
1989int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1990 struct pipe_inode_info *pipe, unsigned int tlen,
1991 unsigned int flags,
1992 ssize_t (*splice_cb)(struct sock *,
1993 struct pipe_inode_info *,
1994 struct splice_pipe_desc *))
1995{
1996 struct partial_page partial[MAX_SKB_FRAGS];
1997 struct page *pages[MAX_SKB_FRAGS];
1998 struct splice_pipe_desc spd = {
1999 .pages = pages,
2000 .partial = partial,
2001 .nr_pages_max = MAX_SKB_FRAGS,
2002 .flags = flags,
2003 .ops = &nosteal_pipe_buf_ops,
2004 .spd_release = sock_spd_release,
2005 };
2006 int ret = 0;
2007
2008 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2009
2010 if (spd.nr_pages)
2011 ret = splice_cb(sk, pipe, &spd);
2012
2013 return ret;
2014}
2015EXPORT_SYMBOL_GPL(skb_splice_bits);
2016
2017/**
2018 * skb_store_bits - store bits from kernel buffer to skb
2019 * @skb: destination buffer
2020 * @offset: offset in destination
2021 * @from: source buffer
2022 * @len: number of bytes to copy
2023 *
2024 * Copy the specified number of bytes from the source buffer to the
2025 * destination skb. This function handles all the messy bits of
2026 * traversing fragment lists and such.
2027 */
2028
2029int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2030{
2031 int start = skb_headlen(skb);
2032 struct sk_buff *frag_iter;
2033 int i, copy;
2034
2035 if (offset > (int)skb->len - len)
2036 goto fault;
2037
2038 if ((copy = start - offset) > 0) {
2039 if (copy > len)
2040 copy = len;
2041 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2042 if ((len -= copy) == 0)
2043 return 0;
2044 offset += copy;
2045 from += copy;
2046 }
2047
2048 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2049 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2050 int end;
2051
2052 WARN_ON(start > offset + len);
2053
2054 end = start + skb_frag_size(frag);
2055 if ((copy = end - offset) > 0) {
2056 u8 *vaddr;
2057
2058 if (copy > len)
2059 copy = len;
2060
2061 vaddr = kmap_atomic(skb_frag_page(frag));
2062 memcpy(vaddr + frag->page_offset + offset - start,
2063 from, copy);
2064 kunmap_atomic(vaddr);
2065
2066 if ((len -= copy) == 0)
2067 return 0;
2068 offset += copy;
2069 from += copy;
2070 }
2071 start = end;
2072 }
2073
2074 skb_walk_frags(skb, frag_iter) {
2075 int end;
2076
2077 WARN_ON(start > offset + len);
2078
2079 end = start + frag_iter->len;
2080 if ((copy = end - offset) > 0) {
2081 if (copy > len)
2082 copy = len;
2083 if (skb_store_bits(frag_iter, offset - start,
2084 from, copy))
2085 goto fault;
2086 if ((len -= copy) == 0)
2087 return 0;
2088 offset += copy;
2089 from += copy;
2090 }
2091 start = end;
2092 }
2093 if (!len)
2094 return 0;
2095
2096fault:
2097 return -EFAULT;
2098}
2099EXPORT_SYMBOL(skb_store_bits);
2100
2101/* Checksum skb data. */
2102__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2103 __wsum csum, const struct skb_checksum_ops *ops)
2104{
2105 int start = skb_headlen(skb);
2106 int i, copy = start - offset;
2107 struct sk_buff *frag_iter;
2108 int pos = 0;
2109
2110 /* Checksum header. */
2111 if (copy > 0) {
2112 if (copy > len)
2113 copy = len;
2114 csum = ops->update(skb->data + offset, copy, csum);
2115 if ((len -= copy) == 0)
2116 return csum;
2117 offset += copy;
2118 pos = copy;
2119 }
2120
2121 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2122 int end;
2123 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2124
2125 WARN_ON(start > offset + len);
2126
2127 end = start + skb_frag_size(frag);
2128 if ((copy = end - offset) > 0) {
2129 __wsum csum2;
2130 u8 *vaddr;
2131
2132 if (copy > len)
2133 copy = len;
2134 vaddr = kmap_atomic(skb_frag_page(frag));
2135 csum2 = ops->update(vaddr + frag->page_offset +
2136 offset - start, copy, 0);
2137 kunmap_atomic(vaddr);
2138 csum = ops->combine(csum, csum2, pos, copy);
2139 if (!(len -= copy))
2140 return csum;
2141 offset += copy;
2142 pos += copy;
2143 }
2144 start = end;
2145 }
2146
2147 skb_walk_frags(skb, frag_iter) {
2148 int end;
2149
2150 WARN_ON(start > offset + len);
2151
2152 end = start + frag_iter->len;
2153 if ((copy = end - offset) > 0) {
2154 __wsum csum2;
2155 if (copy > len)
2156 copy = len;
2157 csum2 = __skb_checksum(frag_iter, offset - start,
2158 copy, 0, ops);
2159 csum = ops->combine(csum, csum2, pos, copy);
2160 if ((len -= copy) == 0)
2161 return csum;
2162 offset += copy;
2163 pos += copy;
2164 }
2165 start = end;
2166 }
2167 BUG_ON(len);
2168
2169 return csum;
2170}
2171EXPORT_SYMBOL(__skb_checksum);
2172
2173__wsum skb_checksum(const struct sk_buff *skb, int offset,
2174 int len, __wsum csum)
2175{
2176 const struct skb_checksum_ops ops = {
2177 .update = csum_partial_ext,
2178 .combine = csum_block_add_ext,
2179 };
2180
2181 return __skb_checksum(skb, offset, len, csum, &ops);
2182}
2183EXPORT_SYMBOL(skb_checksum);
2184
2185/* Both of above in one bottle. */
2186
2187__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2188 u8 *to, int len, __wsum csum)
2189{
2190 int start = skb_headlen(skb);
2191 int i, copy = start - offset;
2192 struct sk_buff *frag_iter;
2193 int pos = 0;
2194
2195 /* Copy header. */
2196 if (copy > 0) {
2197 if (copy > len)
2198 copy = len;
2199 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2200 copy, csum);
2201 if ((len -= copy) == 0)
2202 return csum;
2203 offset += copy;
2204 to += copy;
2205 pos = copy;
2206 }
2207
2208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209 int end;
2210
2211 WARN_ON(start > offset + len);
2212
2213 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2214 if ((copy = end - offset) > 0) {
2215 __wsum csum2;
2216 u8 *vaddr;
2217 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2218
2219 if (copy > len)
2220 copy = len;
2221 vaddr = kmap_atomic(skb_frag_page(frag));
2222 csum2 = csum_partial_copy_nocheck(vaddr +
2223 frag->page_offset +
2224 offset - start, to,
2225 copy, 0);
2226 kunmap_atomic(vaddr);
2227 csum = csum_block_add(csum, csum2, pos);
2228 if (!(len -= copy))
2229 return csum;
2230 offset += copy;
2231 to += copy;
2232 pos += copy;
2233 }
2234 start = end;
2235 }
2236
2237 skb_walk_frags(skb, frag_iter) {
2238 __wsum csum2;
2239 int end;
2240
2241 WARN_ON(start > offset + len);
2242
2243 end = start + frag_iter->len;
2244 if ((copy = end - offset) > 0) {
2245 if (copy > len)
2246 copy = len;
2247 csum2 = skb_copy_and_csum_bits(frag_iter,
2248 offset - start,
2249 to, copy, 0);
2250 csum = csum_block_add(csum, csum2, pos);
2251 if ((len -= copy) == 0)
2252 return csum;
2253 offset += copy;
2254 to += copy;
2255 pos += copy;
2256 }
2257 start = end;
2258 }
2259 BUG_ON(len);
2260 return csum;
2261}
2262EXPORT_SYMBOL(skb_copy_and_csum_bits);
2263
2264 /**
2265 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2266 * @from: source buffer
2267 *
2268 * Calculates the amount of linear headroom needed in the 'to' skb passed
2269 * into skb_zerocopy().
2270 */
2271unsigned int
2272skb_zerocopy_headlen(const struct sk_buff *from)
2273{
2274 unsigned int hlen = 0;
2275
2276 if (!from->head_frag ||
2277 skb_headlen(from) < L1_CACHE_BYTES ||
2278 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2279 hlen = skb_headlen(from);
2280
2281 if (skb_has_frag_list(from))
2282 hlen = from->len;
2283
2284 return hlen;
2285}
2286EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2287
2288/**
2289 * skb_zerocopy - Zero copy skb to skb
2290 * @to: destination buffer
2291 * @from: source buffer
2292 * @len: number of bytes to copy from source buffer
2293 * @hlen: size of linear headroom in destination buffer
2294 *
2295 * Copies up to `len` bytes from `from` to `to` by creating references
2296 * to the frags in the source buffer.
2297 *
2298 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2299 * headroom in the `to` buffer.
2300 *
2301 * Return value:
2302 * 0: everything is OK
2303 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2304 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2305 */
2306int
2307skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2308{
2309 int i, j = 0;
2310 int plen = 0; /* length of skb->head fragment */
2311 int ret;
2312 struct page *page;
2313 unsigned int offset;
2314
2315 BUG_ON(!from->head_frag && !hlen);
2316
2317 /* dont bother with small payloads */
2318 if (len <= skb_tailroom(to))
2319 return skb_copy_bits(from, 0, skb_put(to, len), len);
2320
2321 if (hlen) {
2322 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2323 if (unlikely(ret))
2324 return ret;
2325 len -= hlen;
2326 } else {
2327 plen = min_t(int, skb_headlen(from), len);
2328 if (plen) {
2329 page = virt_to_head_page(from->head);
2330 offset = from->data - (unsigned char *)page_address(page);
2331 __skb_fill_page_desc(to, 0, page, offset, plen);
2332 get_page(page);
2333 j = 1;
2334 len -= plen;
2335 }
2336 }
2337
2338 to->truesize += len + plen;
2339 to->len += len + plen;
2340 to->data_len += len + plen;
2341
2342 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2343 skb_tx_error(from);
2344 return -ENOMEM;
2345 }
2346
2347 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2348 if (!len)
2349 break;
2350 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2351 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2352 len -= skb_shinfo(to)->frags[j].size;
2353 skb_frag_ref(to, j);
2354 j++;
2355 }
2356 skb_shinfo(to)->nr_frags = j;
2357
2358 return 0;
2359}
2360EXPORT_SYMBOL_GPL(skb_zerocopy);
2361
2362void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2363{
2364 __wsum csum;
2365 long csstart;
2366
2367 if (skb->ip_summed == CHECKSUM_PARTIAL)
2368 csstart = skb_checksum_start_offset(skb);
2369 else
2370 csstart = skb_headlen(skb);
2371
2372 BUG_ON(csstart > skb_headlen(skb));
2373
2374 skb_copy_from_linear_data(skb, to, csstart);
2375
2376 csum = 0;
2377 if (csstart != skb->len)
2378 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2379 skb->len - csstart, 0);
2380
2381 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2382 long csstuff = csstart + skb->csum_offset;
2383
2384 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2385 }
2386}
2387EXPORT_SYMBOL(skb_copy_and_csum_dev);
2388
2389/**
2390 * skb_dequeue - remove from the head of the queue
2391 * @list: list to dequeue from
2392 *
2393 * Remove the head of the list. The list lock is taken so the function
2394 * may be used safely with other locking list functions. The head item is
2395 * returned or %NULL if the list is empty.
2396 */
2397
2398struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2399{
2400 unsigned long flags;
2401 struct sk_buff *result;
2402
2403 spin_lock_irqsave(&list->lock, flags);
2404 result = __skb_dequeue(list);
2405 spin_unlock_irqrestore(&list->lock, flags);
2406 return result;
2407}
2408EXPORT_SYMBOL(skb_dequeue);
2409
2410/**
2411 * skb_dequeue_tail - remove from the tail of the queue
2412 * @list: list to dequeue from
2413 *
2414 * Remove the tail of the list. The list lock is taken so the function
2415 * may be used safely with other locking list functions. The tail item is
2416 * returned or %NULL if the list is empty.
2417 */
2418struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2419{
2420 unsigned long flags;
2421 struct sk_buff *result;
2422
2423 spin_lock_irqsave(&list->lock, flags);
2424 result = __skb_dequeue_tail(list);
2425 spin_unlock_irqrestore(&list->lock, flags);
2426 return result;
2427}
2428EXPORT_SYMBOL(skb_dequeue_tail);
2429
2430/**
2431 * skb_queue_purge - empty a list
2432 * @list: list to empty
2433 *
2434 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2435 * the list and one reference dropped. This function takes the list
2436 * lock and is atomic with respect to other list locking functions.
2437 */
2438void skb_queue_purge(struct sk_buff_head *list)
2439{
2440 struct sk_buff *skb;
2441 while ((skb = skb_dequeue(list)) != NULL)
2442 kfree_skb(skb);
2443}
2444EXPORT_SYMBOL(skb_queue_purge);
2445
2446/**
2447 * skb_queue_head - queue a buffer at the list head
2448 * @list: list to use
2449 * @newsk: buffer to queue
2450 *
2451 * Queue a buffer at the start of the list. This function takes the
2452 * list lock and can be used safely with other locking &sk_buff functions
2453 * safely.
2454 *
2455 * A buffer cannot be placed on two lists at the same time.
2456 */
2457void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2458{
2459 unsigned long flags;
2460
2461 spin_lock_irqsave(&list->lock, flags);
2462 __skb_queue_head(list, newsk);
2463 spin_unlock_irqrestore(&list->lock, flags);
2464}
2465EXPORT_SYMBOL(skb_queue_head);
2466
2467/**
2468 * skb_queue_tail - queue a buffer at the list tail
2469 * @list: list to use
2470 * @newsk: buffer to queue
2471 *
2472 * Queue a buffer at the tail of the list. This function takes the
2473 * list lock and can be used safely with other locking &sk_buff functions
2474 * safely.
2475 *
2476 * A buffer cannot be placed on two lists at the same time.
2477 */
2478void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2479{
2480 unsigned long flags;
2481
2482 spin_lock_irqsave(&list->lock, flags);
2483 __skb_queue_tail(list, newsk);
2484 spin_unlock_irqrestore(&list->lock, flags);
2485}
2486EXPORT_SYMBOL(skb_queue_tail);
2487
2488/**
2489 * skb_unlink - remove a buffer from a list
2490 * @skb: buffer to remove
2491 * @list: list to use
2492 *
2493 * Remove a packet from a list. The list locks are taken and this
2494 * function is atomic with respect to other list locked calls
2495 *
2496 * You must know what list the SKB is on.
2497 */
2498void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2499{
2500 unsigned long flags;
2501
2502 spin_lock_irqsave(&list->lock, flags);
2503 __skb_unlink(skb, list);
2504 spin_unlock_irqrestore(&list->lock, flags);
2505}
2506EXPORT_SYMBOL(skb_unlink);
2507
2508/**
2509 * skb_append - append a buffer
2510 * @old: buffer to insert after
2511 * @newsk: buffer to insert
2512 * @list: list to use
2513 *
2514 * Place a packet after a given packet in a list. The list locks are taken
2515 * and this function is atomic with respect to other list locked calls.
2516 * A buffer cannot be placed on two lists at the same time.
2517 */
2518void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2519{
2520 unsigned long flags;
2521
2522 spin_lock_irqsave(&list->lock, flags);
2523 __skb_queue_after(list, old, newsk);
2524 spin_unlock_irqrestore(&list->lock, flags);
2525}
2526EXPORT_SYMBOL(skb_append);
2527
2528/**
2529 * skb_insert - insert a buffer
2530 * @old: buffer to insert before
2531 * @newsk: buffer to insert
2532 * @list: list to use
2533 *
2534 * Place a packet before a given packet in a list. The list locks are
2535 * taken and this function is atomic with respect to other list locked
2536 * calls.
2537 *
2538 * A buffer cannot be placed on two lists at the same time.
2539 */
2540void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2541{
2542 unsigned long flags;
2543
2544 spin_lock_irqsave(&list->lock, flags);
2545 __skb_insert(newsk, old->prev, old, list);
2546 spin_unlock_irqrestore(&list->lock, flags);
2547}
2548EXPORT_SYMBOL(skb_insert);
2549
2550static inline void skb_split_inside_header(struct sk_buff *skb,
2551 struct sk_buff* skb1,
2552 const u32 len, const int pos)
2553{
2554 int i;
2555
2556 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2557 pos - len);
2558 /* And move data appendix as is. */
2559 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2560 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2561
2562 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2563 skb_shinfo(skb)->nr_frags = 0;
2564 skb1->data_len = skb->data_len;
2565 skb1->len += skb1->data_len;
2566 skb->data_len = 0;
2567 skb->len = len;
2568 skb_set_tail_pointer(skb, len);
2569}
2570
2571static inline void skb_split_no_header(struct sk_buff *skb,
2572 struct sk_buff* skb1,
2573 const u32 len, int pos)
2574{
2575 int i, k = 0;
2576 const int nfrags = skb_shinfo(skb)->nr_frags;
2577
2578 skb_shinfo(skb)->nr_frags = 0;
2579 skb1->len = skb1->data_len = skb->len - len;
2580 skb->len = len;
2581 skb->data_len = len - pos;
2582
2583 for (i = 0; i < nfrags; i++) {
2584 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2585
2586 if (pos + size > len) {
2587 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2588
2589 if (pos < len) {
2590 /* Split frag.
2591 * We have two variants in this case:
2592 * 1. Move all the frag to the second
2593 * part, if it is possible. F.e.
2594 * this approach is mandatory for TUX,
2595 * where splitting is expensive.
2596 * 2. Split is accurately. We make this.
2597 */
2598 skb_frag_ref(skb, i);
2599 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2600 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2601 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2602 skb_shinfo(skb)->nr_frags++;
2603 }
2604 k++;
2605 } else
2606 skb_shinfo(skb)->nr_frags++;
2607 pos += size;
2608 }
2609 skb_shinfo(skb1)->nr_frags = k;
2610}
2611
2612/**
2613 * skb_split - Split fragmented skb to two parts at length len.
2614 * @skb: the buffer to split
2615 * @skb1: the buffer to receive the second part
2616 * @len: new length for skb
2617 */
2618void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2619{
2620 int pos = skb_headlen(skb);
2621
2622 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2623 if (len < pos) /* Split line is inside header. */
2624 skb_split_inside_header(skb, skb1, len, pos);
2625 else /* Second chunk has no header, nothing to copy. */
2626 skb_split_no_header(skb, skb1, len, pos);
2627}
2628EXPORT_SYMBOL(skb_split);
2629
2630/* Shifting from/to a cloned skb is a no-go.
2631 *
2632 * Caller cannot keep skb_shinfo related pointers past calling here!
2633 */
2634static int skb_prepare_for_shift(struct sk_buff *skb)
2635{
2636 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2637}
2638
2639/**
2640 * skb_shift - Shifts paged data partially from skb to another
2641 * @tgt: buffer into which tail data gets added
2642 * @skb: buffer from which the paged data comes from
2643 * @shiftlen: shift up to this many bytes
2644 *
2645 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2646 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2647 * It's up to caller to free skb if everything was shifted.
2648 *
2649 * If @tgt runs out of frags, the whole operation is aborted.
2650 *
2651 * Skb cannot include anything else but paged data while tgt is allowed
2652 * to have non-paged data as well.
2653 *
2654 * TODO: full sized shift could be optimized but that would need
2655 * specialized skb free'er to handle frags without up-to-date nr_frags.
2656 */
2657int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2658{
2659 int from, to, merge, todo;
2660 struct skb_frag_struct *fragfrom, *fragto;
2661
2662 BUG_ON(shiftlen > skb->len);
2663 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2664
2665 todo = shiftlen;
2666 from = 0;
2667 to = skb_shinfo(tgt)->nr_frags;
2668 fragfrom = &skb_shinfo(skb)->frags[from];
2669
2670 /* Actual merge is delayed until the point when we know we can
2671 * commit all, so that we don't have to undo partial changes
2672 */
2673 if (!to ||
2674 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675 fragfrom->page_offset)) {
2676 merge = -1;
2677 } else {
2678 merge = to - 1;
2679
2680 todo -= skb_frag_size(fragfrom);
2681 if (todo < 0) {
2682 if (skb_prepare_for_shift(skb) ||
2683 skb_prepare_for_shift(tgt))
2684 return 0;
2685
2686 /* All previous frag pointers might be stale! */
2687 fragfrom = &skb_shinfo(skb)->frags[from];
2688 fragto = &skb_shinfo(tgt)->frags[merge];
2689
2690 skb_frag_size_add(fragto, shiftlen);
2691 skb_frag_size_sub(fragfrom, shiftlen);
2692 fragfrom->page_offset += shiftlen;
2693
2694 goto onlymerged;
2695 }
2696
2697 from++;
2698 }
2699
2700 /* Skip full, not-fitting skb to avoid expensive operations */
2701 if ((shiftlen == skb->len) &&
2702 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2703 return 0;
2704
2705 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2706 return 0;
2707
2708 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709 if (to == MAX_SKB_FRAGS)
2710 return 0;
2711
2712 fragfrom = &skb_shinfo(skb)->frags[from];
2713 fragto = &skb_shinfo(tgt)->frags[to];
2714
2715 if (todo >= skb_frag_size(fragfrom)) {
2716 *fragto = *fragfrom;
2717 todo -= skb_frag_size(fragfrom);
2718 from++;
2719 to++;
2720
2721 } else {
2722 __skb_frag_ref(fragfrom);
2723 fragto->page = fragfrom->page;
2724 fragto->page_offset = fragfrom->page_offset;
2725 skb_frag_size_set(fragto, todo);
2726
2727 fragfrom->page_offset += todo;
2728 skb_frag_size_sub(fragfrom, todo);
2729 todo = 0;
2730
2731 to++;
2732 break;
2733 }
2734 }
2735
2736 /* Ready to "commit" this state change to tgt */
2737 skb_shinfo(tgt)->nr_frags = to;
2738
2739 if (merge >= 0) {
2740 fragfrom = &skb_shinfo(skb)->frags[0];
2741 fragto = &skb_shinfo(tgt)->frags[merge];
2742
2743 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744 __skb_frag_unref(fragfrom);
2745 }
2746
2747 /* Reposition in the original skb */
2748 to = 0;
2749 while (from < skb_shinfo(skb)->nr_frags)
2750 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751 skb_shinfo(skb)->nr_frags = to;
2752
2753 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2754
2755onlymerged:
2756 /* Most likely the tgt won't ever need its checksum anymore, skb on
2757 * the other hand might need it if it needs to be resent
2758 */
2759 tgt->ip_summed = CHECKSUM_PARTIAL;
2760 skb->ip_summed = CHECKSUM_PARTIAL;
2761
2762 /* Yak, is it really working this way? Some helper please? */
2763 skb->len -= shiftlen;
2764 skb->data_len -= shiftlen;
2765 skb->truesize -= shiftlen;
2766 tgt->len += shiftlen;
2767 tgt->data_len += shiftlen;
2768 tgt->truesize += shiftlen;
2769
2770 return shiftlen;
2771}
2772
2773/**
2774 * skb_prepare_seq_read - Prepare a sequential read of skb data
2775 * @skb: the buffer to read
2776 * @from: lower offset of data to be read
2777 * @to: upper offset of data to be read
2778 * @st: state variable
2779 *
2780 * Initializes the specified state variable. Must be called before
2781 * invoking skb_seq_read() for the first time.
2782 */
2783void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784 unsigned int to, struct skb_seq_state *st)
2785{
2786 st->lower_offset = from;
2787 st->upper_offset = to;
2788 st->root_skb = st->cur_skb = skb;
2789 st->frag_idx = st->stepped_offset = 0;
2790 st->frag_data = NULL;
2791}
2792EXPORT_SYMBOL(skb_prepare_seq_read);
2793
2794/**
2795 * skb_seq_read - Sequentially read skb data
2796 * @consumed: number of bytes consumed by the caller so far
2797 * @data: destination pointer for data to be returned
2798 * @st: state variable
2799 *
2800 * Reads a block of skb data at @consumed relative to the
2801 * lower offset specified to skb_prepare_seq_read(). Assigns
2802 * the head of the data block to @data and returns the length
2803 * of the block or 0 if the end of the skb data or the upper
2804 * offset has been reached.
2805 *
2806 * The caller is not required to consume all of the data
2807 * returned, i.e. @consumed is typically set to the number
2808 * of bytes already consumed and the next call to
2809 * skb_seq_read() will return the remaining part of the block.
2810 *
2811 * Note 1: The size of each block of data returned can be arbitrary,
2812 * this limitation is the cost for zerocopy sequential
2813 * reads of potentially non linear data.
2814 *
2815 * Note 2: Fragment lists within fragments are not implemented
2816 * at the moment, state->root_skb could be replaced with
2817 * a stack for this purpose.
2818 */
2819unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820 struct skb_seq_state *st)
2821{
2822 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2823 skb_frag_t *frag;
2824
2825 if (unlikely(abs_offset >= st->upper_offset)) {
2826 if (st->frag_data) {
2827 kunmap_atomic(st->frag_data);
2828 st->frag_data = NULL;
2829 }
2830 return 0;
2831 }
2832
2833next_skb:
2834 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2835
2836 if (abs_offset < block_limit && !st->frag_data) {
2837 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838 return block_limit - abs_offset;
2839 }
2840
2841 if (st->frag_idx == 0 && !st->frag_data)
2842 st->stepped_offset += skb_headlen(st->cur_skb);
2843
2844 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846 block_limit = skb_frag_size(frag) + st->stepped_offset;
2847
2848 if (abs_offset < block_limit) {
2849 if (!st->frag_data)
2850 st->frag_data = kmap_atomic(skb_frag_page(frag));
2851
2852 *data = (u8 *) st->frag_data + frag->page_offset +
2853 (abs_offset - st->stepped_offset);
2854
2855 return block_limit - abs_offset;
2856 }
2857
2858 if (st->frag_data) {
2859 kunmap_atomic(st->frag_data);
2860 st->frag_data = NULL;
2861 }
2862
2863 st->frag_idx++;
2864 st->stepped_offset += skb_frag_size(frag);
2865 }
2866
2867 if (st->frag_data) {
2868 kunmap_atomic(st->frag_data);
2869 st->frag_data = NULL;
2870 }
2871
2872 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2874 st->frag_idx = 0;
2875 goto next_skb;
2876 } else if (st->cur_skb->next) {
2877 st->cur_skb = st->cur_skb->next;
2878 st->frag_idx = 0;
2879 goto next_skb;
2880 }
2881
2882 return 0;
2883}
2884EXPORT_SYMBOL(skb_seq_read);
2885
2886/**
2887 * skb_abort_seq_read - Abort a sequential read of skb data
2888 * @st: state variable
2889 *
2890 * Must be called if skb_seq_read() was not called until it
2891 * returned 0.
2892 */
2893void skb_abort_seq_read(struct skb_seq_state *st)
2894{
2895 if (st->frag_data)
2896 kunmap_atomic(st->frag_data);
2897}
2898EXPORT_SYMBOL(skb_abort_seq_read);
2899
2900#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2901
2902static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903 struct ts_config *conf,
2904 struct ts_state *state)
2905{
2906 return skb_seq_read(offset, text, TS_SKB_CB(state));
2907}
2908
2909static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2910{
2911 skb_abort_seq_read(TS_SKB_CB(state));
2912}
2913
2914/**
2915 * skb_find_text - Find a text pattern in skb data
2916 * @skb: the buffer to look in
2917 * @from: search offset
2918 * @to: search limit
2919 * @config: textsearch configuration
2920 *
2921 * Finds a pattern in the skb data according to the specified
2922 * textsearch configuration. Use textsearch_next() to retrieve
2923 * subsequent occurrences of the pattern. Returns the offset
2924 * to the first occurrence or UINT_MAX if no match was found.
2925 */
2926unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927 unsigned int to, struct ts_config *config)
2928{
2929 struct ts_state state;
2930 unsigned int ret;
2931
2932 config->get_next_block = skb_ts_get_next_block;
2933 config->finish = skb_ts_finish;
2934
2935 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2936
2937 ret = textsearch_find(config, &state);
2938 return (ret <= to - from ? ret : UINT_MAX);
2939}
2940EXPORT_SYMBOL(skb_find_text);
2941
2942/**
2943 * skb_append_datato_frags - append the user data to a skb
2944 * @sk: sock structure
2945 * @skb: skb structure to be appended with user data.
2946 * @getfrag: call back function to be used for getting the user data
2947 * @from: pointer to user message iov
2948 * @length: length of the iov message
2949 *
2950 * Description: This procedure append the user data in the fragment part
2951 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2952 */
2953int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954 int (*getfrag)(void *from, char *to, int offset,
2955 int len, int odd, struct sk_buff *skb),
2956 void *from, int length)
2957{
2958 int frg_cnt = skb_shinfo(skb)->nr_frags;
2959 int copy;
2960 int offset = 0;
2961 int ret;
2962 struct page_frag *pfrag = ¤t->task_frag;
2963
2964 do {
2965 /* Return error if we don't have space for new frag */
2966 if (frg_cnt >= MAX_SKB_FRAGS)
2967 return -EMSGSIZE;
2968
2969 if (!sk_page_frag_refill(sk, pfrag))
2970 return -ENOMEM;
2971
2972 /* copy the user data to page */
2973 copy = min_t(int, length, pfrag->size - pfrag->offset);
2974
2975 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976 offset, copy, 0, skb);
2977 if (ret < 0)
2978 return -EFAULT;
2979
2980 /* copy was successful so update the size parameters */
2981 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2982 copy);
2983 frg_cnt++;
2984 pfrag->offset += copy;
2985 get_page(pfrag->page);
2986
2987 skb->truesize += copy;
2988 atomic_add(copy, &sk->sk_wmem_alloc);
2989 skb->len += copy;
2990 skb->data_len += copy;
2991 offset += copy;
2992 length -= copy;
2993
2994 } while (length > 0);
2995
2996 return 0;
2997}
2998EXPORT_SYMBOL(skb_append_datato_frags);
2999
3000int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001 int offset, size_t size)
3002{
3003 int i = skb_shinfo(skb)->nr_frags;
3004
3005 if (skb_can_coalesce(skb, i, page, offset)) {
3006 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007 } else if (i < MAX_SKB_FRAGS) {
3008 get_page(page);
3009 skb_fill_page_desc(skb, i, page, offset, size);
3010 } else {
3011 return -EMSGSIZE;
3012 }
3013
3014 return 0;
3015}
3016EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3017
3018/**
3019 * skb_push_rcsum - push skb and update receive checksum
3020 * @skb: buffer to update
3021 * @len: length of data pulled
3022 *
3023 * This function performs an skb_push on the packet and updates
3024 * the CHECKSUM_COMPLETE checksum. It should be used on
3025 * receive path processing instead of skb_push unless you know
3026 * that the checksum difference is zero (e.g., a valid IP header)
3027 * or you are setting ip_summed to CHECKSUM_NONE.
3028 */
3029static unsigned char *skb_push_rcsum(struct sk_buff *skb, unsigned len)
3030{
3031 skb_push(skb, len);
3032 skb_postpush_rcsum(skb, skb->data, len);
3033 return skb->data;
3034}
3035
3036/**
3037 * skb_pull_rcsum - pull skb and update receive checksum
3038 * @skb: buffer to update
3039 * @len: length of data pulled
3040 *
3041 * This function performs an skb_pull on the packet and updates
3042 * the CHECKSUM_COMPLETE checksum. It should be used on
3043 * receive path processing instead of skb_pull unless you know
3044 * that the checksum difference is zero (e.g., a valid IP header)
3045 * or you are setting ip_summed to CHECKSUM_NONE.
3046 */
3047unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3048{
3049 unsigned char *data = skb->data;
3050
3051 BUG_ON(len > skb->len);
3052 __skb_pull(skb, len);
3053 skb_postpull_rcsum(skb, data, len);
3054 return skb->data;
3055}
3056EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3057
3058/**
3059 * skb_segment - Perform protocol segmentation on skb.
3060 * @head_skb: buffer to segment
3061 * @features: features for the output path (see dev->features)
3062 *
3063 * This function performs segmentation on the given skb. It returns
3064 * a pointer to the first in a list of new skbs for the segments.
3065 * In case of error it returns ERR_PTR(err).
3066 */
3067struct sk_buff *skb_segment(struct sk_buff *head_skb,
3068 netdev_features_t features)
3069{
3070 struct sk_buff *segs = NULL;
3071 struct sk_buff *tail = NULL;
3072 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3073 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3074 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3075 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3076 struct sk_buff *frag_skb = head_skb;
3077 unsigned int offset = doffset;
3078 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3079 unsigned int headroom;
3080 unsigned int len;
3081 __be16 proto;
3082 bool csum;
3083 int sg = !!(features & NETIF_F_SG);
3084 int nfrags = skb_shinfo(head_skb)->nr_frags;
3085 int err = -ENOMEM;
3086 int i = 0;
3087 int pos;
3088 int dummy;
3089
3090 __skb_push(head_skb, doffset);
3091 proto = skb_network_protocol(head_skb, &dummy);
3092 if (unlikely(!proto))
3093 return ERR_PTR(-EINVAL);
3094
3095 csum = !!can_checksum_protocol(features, proto);
3096
3097 headroom = skb_headroom(head_skb);
3098 pos = skb_headlen(head_skb);
3099
3100 do {
3101 struct sk_buff *nskb;
3102 skb_frag_t *nskb_frag;
3103 int hsize;
3104 int size;
3105
3106 len = head_skb->len - offset;
3107 if (len > mss)
3108 len = mss;
3109
3110 hsize = skb_headlen(head_skb) - offset;
3111 if (hsize < 0)
3112 hsize = 0;
3113 if (hsize > len || !sg)
3114 hsize = len;
3115
3116 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3117 (skb_headlen(list_skb) == len || sg)) {
3118 BUG_ON(skb_headlen(list_skb) > len);
3119
3120 i = 0;
3121 nfrags = skb_shinfo(list_skb)->nr_frags;
3122 frag = skb_shinfo(list_skb)->frags;
3123 frag_skb = list_skb;
3124 pos += skb_headlen(list_skb);
3125
3126 while (pos < offset + len) {
3127 BUG_ON(i >= nfrags);
3128
3129 size = skb_frag_size(frag);
3130 if (pos + size > offset + len)
3131 break;
3132
3133 i++;
3134 pos += size;
3135 frag++;
3136 }
3137
3138 nskb = skb_clone(list_skb, GFP_ATOMIC);
3139 list_skb = list_skb->next;
3140
3141 if (unlikely(!nskb))
3142 goto err;
3143
3144 if (unlikely(pskb_trim(nskb, len))) {
3145 kfree_skb(nskb);
3146 goto err;
3147 }
3148
3149 hsize = skb_end_offset(nskb);
3150 if (skb_cow_head(nskb, doffset + headroom)) {
3151 kfree_skb(nskb);
3152 goto err;
3153 }
3154
3155 nskb->truesize += skb_end_offset(nskb) - hsize;
3156 skb_release_head_state(nskb);
3157 __skb_push(nskb, doffset);
3158 } else {
3159 nskb = __alloc_skb(hsize + doffset + headroom,
3160 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3161 NUMA_NO_NODE);
3162
3163 if (unlikely(!nskb))
3164 goto err;
3165
3166 skb_reserve(nskb, headroom);
3167 __skb_put(nskb, doffset);
3168 }
3169
3170 if (segs)
3171 tail->next = nskb;
3172 else
3173 segs = nskb;
3174 tail = nskb;
3175
3176 __copy_skb_header(nskb, head_skb);
3177
3178 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3179 skb_reset_mac_len(nskb);
3180
3181 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3182 nskb->data - tnl_hlen,
3183 doffset + tnl_hlen);
3184
3185 if (nskb->len == len + doffset)
3186 goto perform_csum_check;
3187
3188 if (!sg) {
3189 if (!nskb->remcsum_offload)
3190 nskb->ip_summed = CHECKSUM_NONE;
3191 SKB_GSO_CB(nskb)->csum =
3192 skb_copy_and_csum_bits(head_skb, offset,
3193 skb_put(nskb, len),
3194 len, 0);
3195 SKB_GSO_CB(nskb)->csum_start =
3196 skb_headroom(nskb) + doffset;
3197 continue;
3198 }
3199
3200 nskb_frag = skb_shinfo(nskb)->frags;
3201
3202 skb_copy_from_linear_data_offset(head_skb, offset,
3203 skb_put(nskb, hsize), hsize);
3204
3205 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3206 SKBTX_SHARED_FRAG;
3207
3208 while (pos < offset + len) {
3209 if (i >= nfrags) {
3210 BUG_ON(skb_headlen(list_skb));
3211
3212 i = 0;
3213 nfrags = skb_shinfo(list_skb)->nr_frags;
3214 frag = skb_shinfo(list_skb)->frags;
3215 frag_skb = list_skb;
3216
3217 BUG_ON(!nfrags);
3218
3219 list_skb = list_skb->next;
3220 }
3221
3222 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3223 MAX_SKB_FRAGS)) {
3224 net_warn_ratelimited(
3225 "skb_segment: too many frags: %u %u\n",
3226 pos, mss);
3227 goto err;
3228 }
3229
3230 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3231 goto err;
3232
3233 *nskb_frag = *frag;
3234 __skb_frag_ref(nskb_frag);
3235 size = skb_frag_size(nskb_frag);
3236
3237 if (pos < offset) {
3238 nskb_frag->page_offset += offset - pos;
3239 skb_frag_size_sub(nskb_frag, offset - pos);
3240 }
3241
3242 skb_shinfo(nskb)->nr_frags++;
3243
3244 if (pos + size <= offset + len) {
3245 i++;
3246 frag++;
3247 pos += size;
3248 } else {
3249 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3250 goto skip_fraglist;
3251 }
3252
3253 nskb_frag++;
3254 }
3255
3256skip_fraglist:
3257 nskb->data_len = len - hsize;
3258 nskb->len += nskb->data_len;
3259 nskb->truesize += nskb->data_len;
3260
3261perform_csum_check:
3262 if (!csum) {
3263 if (skb_has_shared_frag(nskb)) {
3264 err = __skb_linearize(nskb);
3265 if (err)
3266 goto err;
3267 }
3268 if (!nskb->remcsum_offload)
3269 nskb->ip_summed = CHECKSUM_NONE;
3270 SKB_GSO_CB(nskb)->csum =
3271 skb_checksum(nskb, doffset,
3272 nskb->len - doffset, 0);
3273 SKB_GSO_CB(nskb)->csum_start =
3274 skb_headroom(nskb) + doffset;
3275 }
3276 } while ((offset += len) < head_skb->len);
3277
3278 /* Some callers want to get the end of the list.
3279 * Put it in segs->prev to avoid walking the list.
3280 * (see validate_xmit_skb_list() for example)
3281 */
3282 segs->prev = tail;
3283
3284 /* Following permits correct backpressure, for protocols
3285 * using skb_set_owner_w().
3286 * Idea is to tranfert ownership from head_skb to last segment.
3287 */
3288 if (head_skb->destructor == sock_wfree) {
3289 swap(tail->truesize, head_skb->truesize);
3290 swap(tail->destructor, head_skb->destructor);
3291 swap(tail->sk, head_skb->sk);
3292 }
3293 return segs;
3294
3295err:
3296 kfree_skb_list(segs);
3297 return ERR_PTR(err);
3298}
3299EXPORT_SYMBOL_GPL(skb_segment);
3300
3301int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3302{
3303 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3304 unsigned int offset = skb_gro_offset(skb);
3305 unsigned int headlen = skb_headlen(skb);
3306 unsigned int len = skb_gro_len(skb);
3307 struct sk_buff *lp, *p = *head;
3308 unsigned int delta_truesize;
3309
3310 if (unlikely(p->len + len >= 65536))
3311 return -E2BIG;
3312
3313 lp = NAPI_GRO_CB(p)->last;
3314 pinfo = skb_shinfo(lp);
3315
3316 if (headlen <= offset) {
3317 skb_frag_t *frag;
3318 skb_frag_t *frag2;
3319 int i = skbinfo->nr_frags;
3320 int nr_frags = pinfo->nr_frags + i;
3321
3322 if (nr_frags > MAX_SKB_FRAGS)
3323 goto merge;
3324
3325 offset -= headlen;
3326 pinfo->nr_frags = nr_frags;
3327 skbinfo->nr_frags = 0;
3328
3329 frag = pinfo->frags + nr_frags;
3330 frag2 = skbinfo->frags + i;
3331 do {
3332 *--frag = *--frag2;
3333 } while (--i);
3334
3335 frag->page_offset += offset;
3336 skb_frag_size_sub(frag, offset);
3337
3338 /* all fragments truesize : remove (head size + sk_buff) */
3339 delta_truesize = skb->truesize -
3340 SKB_TRUESIZE(skb_end_offset(skb));
3341
3342 skb->truesize -= skb->data_len;
3343 skb->len -= skb->data_len;
3344 skb->data_len = 0;
3345
3346 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3347 goto done;
3348 } else if (skb->head_frag) {
3349 int nr_frags = pinfo->nr_frags;
3350 skb_frag_t *frag = pinfo->frags + nr_frags;
3351 struct page *page = virt_to_head_page(skb->head);
3352 unsigned int first_size = headlen - offset;
3353 unsigned int first_offset;
3354
3355 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3356 goto merge;
3357
3358 first_offset = skb->data -
3359 (unsigned char *)page_address(page) +
3360 offset;
3361
3362 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3363
3364 frag->page.p = page;
3365 frag->page_offset = first_offset;
3366 skb_frag_size_set(frag, first_size);
3367
3368 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3369 /* We dont need to clear skbinfo->nr_frags here */
3370
3371 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3372 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3373 goto done;
3374 }
3375
3376merge:
3377 delta_truesize = skb->truesize;
3378 if (offset > headlen) {
3379 unsigned int eat = offset - headlen;
3380
3381 skbinfo->frags[0].page_offset += eat;
3382 skb_frag_size_sub(&skbinfo->frags[0], eat);
3383 skb->data_len -= eat;
3384 skb->len -= eat;
3385 offset = headlen;
3386 }
3387
3388 __skb_pull(skb, offset);
3389
3390 if (NAPI_GRO_CB(p)->last == p)
3391 skb_shinfo(p)->frag_list = skb;
3392 else
3393 NAPI_GRO_CB(p)->last->next = skb;
3394 NAPI_GRO_CB(p)->last = skb;
3395 __skb_header_release(skb);
3396 lp = p;
3397
3398done:
3399 NAPI_GRO_CB(p)->count++;
3400 p->data_len += len;
3401 p->truesize += delta_truesize;
3402 p->len += len;
3403 if (lp != p) {
3404 lp->data_len += len;
3405 lp->truesize += delta_truesize;
3406 lp->len += len;
3407 }
3408 NAPI_GRO_CB(skb)->same_flow = 1;
3409 return 0;
3410}
3411
3412void __init skb_init(void)
3413{
3414 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3415 sizeof(struct sk_buff),
3416 0,
3417 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3418 NULL);
3419 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3420 sizeof(struct sk_buff_fclones),
3421 0,
3422 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3423 NULL);
3424}
3425
3426/**
3427 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3428 * @skb: Socket buffer containing the buffers to be mapped
3429 * @sg: The scatter-gather list to map into
3430 * @offset: The offset into the buffer's contents to start mapping
3431 * @len: Length of buffer space to be mapped
3432 *
3433 * Fill the specified scatter-gather list with mappings/pointers into a
3434 * region of the buffer space attached to a socket buffer.
3435 */
3436static int
3437__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3438{
3439 int start = skb_headlen(skb);
3440 int i, copy = start - offset;
3441 struct sk_buff *frag_iter;
3442 int elt = 0;
3443
3444 if (copy > 0) {
3445 if (copy > len)
3446 copy = len;
3447 sg_set_buf(sg, skb->data + offset, copy);
3448 elt++;
3449 if ((len -= copy) == 0)
3450 return elt;
3451 offset += copy;
3452 }
3453
3454 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3455 int end;
3456
3457 WARN_ON(start > offset + len);
3458
3459 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3460 if ((copy = end - offset) > 0) {
3461 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3462
3463 if (copy > len)
3464 copy = len;
3465 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3466 frag->page_offset+offset-start);
3467 elt++;
3468 if (!(len -= copy))
3469 return elt;
3470 offset += copy;
3471 }
3472 start = end;
3473 }
3474
3475 skb_walk_frags(skb, frag_iter) {
3476 int end;
3477
3478 WARN_ON(start > offset + len);
3479
3480 end = start + frag_iter->len;
3481 if ((copy = end - offset) > 0) {
3482 if (copy > len)
3483 copy = len;
3484 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3485 copy);
3486 if ((len -= copy) == 0)
3487 return elt;
3488 offset += copy;
3489 }
3490 start = end;
3491 }
3492 BUG_ON(len);
3493 return elt;
3494}
3495
3496/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3497 * sglist without mark the sg which contain last skb data as the end.
3498 * So the caller can mannipulate sg list as will when padding new data after
3499 * the first call without calling sg_unmark_end to expend sg list.
3500 *
3501 * Scenario to use skb_to_sgvec_nomark:
3502 * 1. sg_init_table
3503 * 2. skb_to_sgvec_nomark(payload1)
3504 * 3. skb_to_sgvec_nomark(payload2)
3505 *
3506 * This is equivalent to:
3507 * 1. sg_init_table
3508 * 2. skb_to_sgvec(payload1)
3509 * 3. sg_unmark_end
3510 * 4. skb_to_sgvec(payload2)
3511 *
3512 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3513 * is more preferable.
3514 */
3515int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3516 int offset, int len)
3517{
3518 return __skb_to_sgvec(skb, sg, offset, len);
3519}
3520EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3521
3522int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3523{
3524 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3525
3526 sg_mark_end(&sg[nsg - 1]);
3527
3528 return nsg;
3529}
3530EXPORT_SYMBOL_GPL(skb_to_sgvec);
3531
3532/**
3533 * skb_cow_data - Check that a socket buffer's data buffers are writable
3534 * @skb: The socket buffer to check.
3535 * @tailbits: Amount of trailing space to be added
3536 * @trailer: Returned pointer to the skb where the @tailbits space begins
3537 *
3538 * Make sure that the data buffers attached to a socket buffer are
3539 * writable. If they are not, private copies are made of the data buffers
3540 * and the socket buffer is set to use these instead.
3541 *
3542 * If @tailbits is given, make sure that there is space to write @tailbits
3543 * bytes of data beyond current end of socket buffer. @trailer will be
3544 * set to point to the skb in which this space begins.
3545 *
3546 * The number of scatterlist elements required to completely map the
3547 * COW'd and extended socket buffer will be returned.
3548 */
3549int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3550{
3551 int copyflag;
3552 int elt;
3553 struct sk_buff *skb1, **skb_p;
3554
3555 /* If skb is cloned or its head is paged, reallocate
3556 * head pulling out all the pages (pages are considered not writable
3557 * at the moment even if they are anonymous).
3558 */
3559 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3560 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3561 return -ENOMEM;
3562
3563 /* Easy case. Most of packets will go this way. */
3564 if (!skb_has_frag_list(skb)) {
3565 /* A little of trouble, not enough of space for trailer.
3566 * This should not happen, when stack is tuned to generate
3567 * good frames. OK, on miss we reallocate and reserve even more
3568 * space, 128 bytes is fair. */
3569
3570 if (skb_tailroom(skb) < tailbits &&
3571 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3572 return -ENOMEM;
3573
3574 /* Voila! */
3575 *trailer = skb;
3576 return 1;
3577 }
3578
3579 /* Misery. We are in troubles, going to mincer fragments... */
3580
3581 elt = 1;
3582 skb_p = &skb_shinfo(skb)->frag_list;
3583 copyflag = 0;
3584
3585 while ((skb1 = *skb_p) != NULL) {
3586 int ntail = 0;
3587
3588 /* The fragment is partially pulled by someone,
3589 * this can happen on input. Copy it and everything
3590 * after it. */
3591
3592 if (skb_shared(skb1))
3593 copyflag = 1;
3594
3595 /* If the skb is the last, worry about trailer. */
3596
3597 if (skb1->next == NULL && tailbits) {
3598 if (skb_shinfo(skb1)->nr_frags ||
3599 skb_has_frag_list(skb1) ||
3600 skb_tailroom(skb1) < tailbits)
3601 ntail = tailbits + 128;
3602 }
3603
3604 if (copyflag ||
3605 skb_cloned(skb1) ||
3606 ntail ||
3607 skb_shinfo(skb1)->nr_frags ||
3608 skb_has_frag_list(skb1)) {
3609 struct sk_buff *skb2;
3610
3611 /* Fuck, we are miserable poor guys... */
3612 if (ntail == 0)
3613 skb2 = skb_copy(skb1, GFP_ATOMIC);
3614 else
3615 skb2 = skb_copy_expand(skb1,
3616 skb_headroom(skb1),
3617 ntail,
3618 GFP_ATOMIC);
3619 if (unlikely(skb2 == NULL))
3620 return -ENOMEM;
3621
3622 if (skb1->sk)
3623 skb_set_owner_w(skb2, skb1->sk);
3624
3625 /* Looking around. Are we still alive?
3626 * OK, link new skb, drop old one */
3627
3628 skb2->next = skb1->next;
3629 *skb_p = skb2;
3630 kfree_skb(skb1);
3631 skb1 = skb2;
3632 }
3633 elt++;
3634 *trailer = skb1;
3635 skb_p = &skb1->next;
3636 }
3637
3638 return elt;
3639}
3640EXPORT_SYMBOL_GPL(skb_cow_data);
3641
3642static void sock_rmem_free(struct sk_buff *skb)
3643{
3644 struct sock *sk = skb->sk;
3645
3646 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3647}
3648
3649/*
3650 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3651 */
3652int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3653{
3654 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3655 (unsigned int)sk->sk_rcvbuf)
3656 return -ENOMEM;
3657
3658 skb_orphan(skb);
3659 skb->sk = sk;
3660 skb->destructor = sock_rmem_free;
3661 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3662
3663 /* before exiting rcu section, make sure dst is refcounted */
3664 skb_dst_force(skb);
3665
3666 skb_queue_tail(&sk->sk_error_queue, skb);
3667 if (!sock_flag(sk, SOCK_DEAD))
3668 sk->sk_data_ready(sk);
3669 return 0;
3670}
3671EXPORT_SYMBOL(sock_queue_err_skb);
3672
3673struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3674{
3675 struct sk_buff_head *q = &sk->sk_error_queue;
3676 struct sk_buff *skb, *skb_next;
3677 unsigned long flags;
3678 int err = 0;
3679
3680 spin_lock_irqsave(&q->lock, flags);
3681 skb = __skb_dequeue(q);
3682 if (skb && (skb_next = skb_peek(q)))
3683 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3684 spin_unlock_irqrestore(&q->lock, flags);
3685
3686 sk->sk_err = err;
3687 if (err)
3688 sk->sk_error_report(sk);
3689
3690 return skb;
3691}
3692EXPORT_SYMBOL(sock_dequeue_err_skb);
3693
3694/**
3695 * skb_clone_sk - create clone of skb, and take reference to socket
3696 * @skb: the skb to clone
3697 *
3698 * This function creates a clone of a buffer that holds a reference on
3699 * sk_refcnt. Buffers created via this function are meant to be
3700 * returned using sock_queue_err_skb, or free via kfree_skb.
3701 *
3702 * When passing buffers allocated with this function to sock_queue_err_skb
3703 * it is necessary to wrap the call with sock_hold/sock_put in order to
3704 * prevent the socket from being released prior to being enqueued on
3705 * the sk_error_queue.
3706 */
3707struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3708{
3709 struct sock *sk = skb->sk;
3710 struct sk_buff *clone;
3711
3712 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3713 return NULL;
3714
3715 clone = skb_clone(skb, GFP_ATOMIC);
3716 if (!clone) {
3717 sock_put(sk);
3718 return NULL;
3719 }
3720
3721 clone->sk = sk;
3722 clone->destructor = sock_efree;
3723
3724 return clone;
3725}
3726EXPORT_SYMBOL(skb_clone_sk);
3727
3728static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3729 struct sock *sk,
3730 int tstype)
3731{
3732 struct sock_exterr_skb *serr;
3733 int err;
3734
3735 serr = SKB_EXT_ERR(skb);
3736 memset(serr, 0, sizeof(*serr));
3737 serr->ee.ee_errno = ENOMSG;
3738 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3739 serr->ee.ee_info = tstype;
3740 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3741 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3742 if (sk->sk_protocol == IPPROTO_TCP &&
3743 sk->sk_type == SOCK_STREAM)
3744 serr->ee.ee_data -= sk->sk_tskey;
3745 }
3746
3747 err = sock_queue_err_skb(sk, skb);
3748
3749 if (err)
3750 kfree_skb(skb);
3751}
3752
3753static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3754{
3755 bool ret;
3756
3757 if (likely(sysctl_tstamp_allow_data || tsonly))
3758 return true;
3759
3760 read_lock_bh(&sk->sk_callback_lock);
3761 ret = sk->sk_socket && sk->sk_socket->file &&
3762 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3763 read_unlock_bh(&sk->sk_callback_lock);
3764 return ret;
3765}
3766
3767void skb_complete_tx_timestamp(struct sk_buff *skb,
3768 struct skb_shared_hwtstamps *hwtstamps)
3769{
3770 struct sock *sk = skb->sk;
3771
3772 if (!skb_may_tx_timestamp(sk, false))
3773 return;
3774
3775 /* take a reference to prevent skb_orphan() from freeing the socket */
3776 sock_hold(sk);
3777
3778 *skb_hwtstamps(skb) = *hwtstamps;
3779 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3780
3781 sock_put(sk);
3782}
3783EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3784
3785void __skb_tstamp_tx(struct sk_buff *orig_skb,
3786 struct skb_shared_hwtstamps *hwtstamps,
3787 struct sock *sk, int tstype)
3788{
3789 struct sk_buff *skb;
3790 bool tsonly;
3791
3792 if (!sk)
3793 return;
3794
3795 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3796 if (!skb_may_tx_timestamp(sk, tsonly))
3797 return;
3798
3799 if (tsonly)
3800 skb = alloc_skb(0, GFP_ATOMIC);
3801 else
3802 skb = skb_clone(orig_skb, GFP_ATOMIC);
3803 if (!skb)
3804 return;
3805
3806 if (tsonly) {
3807 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3808 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3809 }
3810
3811 if (hwtstamps)
3812 *skb_hwtstamps(skb) = *hwtstamps;
3813 else
3814 skb->tstamp = ktime_get_real();
3815
3816 __skb_complete_tx_timestamp(skb, sk, tstype);
3817}
3818EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3819
3820void skb_tstamp_tx(struct sk_buff *orig_skb,
3821 struct skb_shared_hwtstamps *hwtstamps)
3822{
3823 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3824 SCM_TSTAMP_SND);
3825}
3826EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3827
3828void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3829{
3830 struct sock *sk = skb->sk;
3831 struct sock_exterr_skb *serr;
3832 int err;
3833
3834 skb->wifi_acked_valid = 1;
3835 skb->wifi_acked = acked;
3836
3837 serr = SKB_EXT_ERR(skb);
3838 memset(serr, 0, sizeof(*serr));
3839 serr->ee.ee_errno = ENOMSG;
3840 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3841
3842 /* take a reference to prevent skb_orphan() from freeing the socket */
3843 sock_hold(sk);
3844
3845 err = sock_queue_err_skb(sk, skb);
3846 if (err)
3847 kfree_skb(skb);
3848
3849 sock_put(sk);
3850}
3851EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3852
3853/**
3854 * skb_partial_csum_set - set up and verify partial csum values for packet
3855 * @skb: the skb to set
3856 * @start: the number of bytes after skb->data to start checksumming.
3857 * @off: the offset from start to place the checksum.
3858 *
3859 * For untrusted partially-checksummed packets, we need to make sure the values
3860 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3861 *
3862 * This function checks and sets those values and skb->ip_summed: if this
3863 * returns false you should drop the packet.
3864 */
3865bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3866{
3867 if (unlikely(start > skb_headlen(skb)) ||
3868 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3869 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3870 start, off, skb_headlen(skb));
3871 return false;
3872 }
3873 skb->ip_summed = CHECKSUM_PARTIAL;
3874 skb->csum_start = skb_headroom(skb) + start;
3875 skb->csum_offset = off;
3876 skb_set_transport_header(skb, start);
3877 return true;
3878}
3879EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3880
3881static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3882 unsigned int max)
3883{
3884 if (skb_headlen(skb) >= len)
3885 return 0;
3886
3887 /* If we need to pullup then pullup to the max, so we
3888 * won't need to do it again.
3889 */
3890 if (max > skb->len)
3891 max = skb->len;
3892
3893 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3894 return -ENOMEM;
3895
3896 if (skb_headlen(skb) < len)
3897 return -EPROTO;
3898
3899 return 0;
3900}
3901
3902#define MAX_TCP_HDR_LEN (15 * 4)
3903
3904static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3905 typeof(IPPROTO_IP) proto,
3906 unsigned int off)
3907{
3908 switch (proto) {
3909 int err;
3910
3911 case IPPROTO_TCP:
3912 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3913 off + MAX_TCP_HDR_LEN);
3914 if (!err && !skb_partial_csum_set(skb, off,
3915 offsetof(struct tcphdr,
3916 check)))
3917 err = -EPROTO;
3918 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3919
3920 case IPPROTO_UDP:
3921 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3922 off + sizeof(struct udphdr));
3923 if (!err && !skb_partial_csum_set(skb, off,
3924 offsetof(struct udphdr,
3925 check)))
3926 err = -EPROTO;
3927 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3928 }
3929
3930 return ERR_PTR(-EPROTO);
3931}
3932
3933/* This value should be large enough to cover a tagged ethernet header plus
3934 * maximally sized IP and TCP or UDP headers.
3935 */
3936#define MAX_IP_HDR_LEN 128
3937
3938static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3939{
3940 unsigned int off;
3941 bool fragment;
3942 __sum16 *csum;
3943 int err;
3944
3945 fragment = false;
3946
3947 err = skb_maybe_pull_tail(skb,
3948 sizeof(struct iphdr),
3949 MAX_IP_HDR_LEN);
3950 if (err < 0)
3951 goto out;
3952
3953 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3954 fragment = true;
3955
3956 off = ip_hdrlen(skb);
3957
3958 err = -EPROTO;
3959
3960 if (fragment)
3961 goto out;
3962
3963 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3964 if (IS_ERR(csum))
3965 return PTR_ERR(csum);
3966
3967 if (recalculate)
3968 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3969 ip_hdr(skb)->daddr,
3970 skb->len - off,
3971 ip_hdr(skb)->protocol, 0);
3972 err = 0;
3973
3974out:
3975 return err;
3976}
3977
3978/* This value should be large enough to cover a tagged ethernet header plus
3979 * an IPv6 header, all options, and a maximal TCP or UDP header.
3980 */
3981#define MAX_IPV6_HDR_LEN 256
3982
3983#define OPT_HDR(type, skb, off) \
3984 (type *)(skb_network_header(skb) + (off))
3985
3986static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3987{
3988 int err;
3989 u8 nexthdr;
3990 unsigned int off;
3991 unsigned int len;
3992 bool fragment;
3993 bool done;
3994 __sum16 *csum;
3995
3996 fragment = false;
3997 done = false;
3998
3999 off = sizeof(struct ipv6hdr);
4000
4001 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4002 if (err < 0)
4003 goto out;
4004
4005 nexthdr = ipv6_hdr(skb)->nexthdr;
4006
4007 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4008 while (off <= len && !done) {
4009 switch (nexthdr) {
4010 case IPPROTO_DSTOPTS:
4011 case IPPROTO_HOPOPTS:
4012 case IPPROTO_ROUTING: {
4013 struct ipv6_opt_hdr *hp;
4014
4015 err = skb_maybe_pull_tail(skb,
4016 off +
4017 sizeof(struct ipv6_opt_hdr),
4018 MAX_IPV6_HDR_LEN);
4019 if (err < 0)
4020 goto out;
4021
4022 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4023 nexthdr = hp->nexthdr;
4024 off += ipv6_optlen(hp);
4025 break;
4026 }
4027 case IPPROTO_AH: {
4028 struct ip_auth_hdr *hp;
4029
4030 err = skb_maybe_pull_tail(skb,
4031 off +
4032 sizeof(struct ip_auth_hdr),
4033 MAX_IPV6_HDR_LEN);
4034 if (err < 0)
4035 goto out;
4036
4037 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4038 nexthdr = hp->nexthdr;
4039 off += ipv6_authlen(hp);
4040 break;
4041 }
4042 case IPPROTO_FRAGMENT: {
4043 struct frag_hdr *hp;
4044
4045 err = skb_maybe_pull_tail(skb,
4046 off +
4047 sizeof(struct frag_hdr),
4048 MAX_IPV6_HDR_LEN);
4049 if (err < 0)
4050 goto out;
4051
4052 hp = OPT_HDR(struct frag_hdr, skb, off);
4053
4054 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4055 fragment = true;
4056
4057 nexthdr = hp->nexthdr;
4058 off += sizeof(struct frag_hdr);
4059 break;
4060 }
4061 default:
4062 done = true;
4063 break;
4064 }
4065 }
4066
4067 err = -EPROTO;
4068
4069 if (!done || fragment)
4070 goto out;
4071
4072 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4073 if (IS_ERR(csum))
4074 return PTR_ERR(csum);
4075
4076 if (recalculate)
4077 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4078 &ipv6_hdr(skb)->daddr,
4079 skb->len - off, nexthdr, 0);
4080 err = 0;
4081
4082out:
4083 return err;
4084}
4085
4086/**
4087 * skb_checksum_setup - set up partial checksum offset
4088 * @skb: the skb to set up
4089 * @recalculate: if true the pseudo-header checksum will be recalculated
4090 */
4091int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4092{
4093 int err;
4094
4095 switch (skb->protocol) {
4096 case htons(ETH_P_IP):
4097 err = skb_checksum_setup_ipv4(skb, recalculate);
4098 break;
4099
4100 case htons(ETH_P_IPV6):
4101 err = skb_checksum_setup_ipv6(skb, recalculate);
4102 break;
4103
4104 default:
4105 err = -EPROTO;
4106 break;
4107 }
4108
4109 return err;
4110}
4111EXPORT_SYMBOL(skb_checksum_setup);
4112
4113/**
4114 * skb_checksum_maybe_trim - maybe trims the given skb
4115 * @skb: the skb to check
4116 * @transport_len: the data length beyond the network header
4117 *
4118 * Checks whether the given skb has data beyond the given transport length.
4119 * If so, returns a cloned skb trimmed to this transport length.
4120 * Otherwise returns the provided skb. Returns NULL in error cases
4121 * (e.g. transport_len exceeds skb length or out-of-memory).
4122 *
4123 * Caller needs to set the skb transport header and free any returned skb if it
4124 * differs from the provided skb.
4125 */
4126static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4127 unsigned int transport_len)
4128{
4129 struct sk_buff *skb_chk;
4130 unsigned int len = skb_transport_offset(skb) + transport_len;
4131 int ret;
4132
4133 if (skb->len < len)
4134 return NULL;
4135 else if (skb->len == len)
4136 return skb;
4137
4138 skb_chk = skb_clone(skb, GFP_ATOMIC);
4139 if (!skb_chk)
4140 return NULL;
4141
4142 ret = pskb_trim_rcsum(skb_chk, len);
4143 if (ret) {
4144 kfree_skb(skb_chk);
4145 return NULL;
4146 }
4147
4148 return skb_chk;
4149}
4150
4151/**
4152 * skb_checksum_trimmed - validate checksum of an skb
4153 * @skb: the skb to check
4154 * @transport_len: the data length beyond the network header
4155 * @skb_chkf: checksum function to use
4156 *
4157 * Applies the given checksum function skb_chkf to the provided skb.
4158 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4159 *
4160 * If the skb has data beyond the given transport length, then a
4161 * trimmed & cloned skb is checked and returned.
4162 *
4163 * Caller needs to set the skb transport header and free any returned skb if it
4164 * differs from the provided skb.
4165 */
4166struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4167 unsigned int transport_len,
4168 __sum16(*skb_chkf)(struct sk_buff *skb))
4169{
4170 struct sk_buff *skb_chk;
4171 unsigned int offset = skb_transport_offset(skb);
4172 __sum16 ret;
4173
4174 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4175 if (!skb_chk)
4176 goto err;
4177
4178 if (!pskb_may_pull(skb_chk, offset))
4179 goto err;
4180
4181 skb_pull_rcsum(skb_chk, offset);
4182 ret = skb_chkf(skb_chk);
4183 skb_push_rcsum(skb_chk, offset);
4184
4185 if (ret)
4186 goto err;
4187
4188 return skb_chk;
4189
4190err:
4191 if (skb_chk && skb_chk != skb)
4192 kfree_skb(skb_chk);
4193
4194 return NULL;
4195
4196}
4197EXPORT_SYMBOL(skb_checksum_trimmed);
4198
4199void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4200{
4201 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4202 skb->dev->name);
4203}
4204EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4205
4206void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4207{
4208 if (head_stolen) {
4209 skb_release_head_state(skb);
4210 kmem_cache_free(skbuff_head_cache, skb);
4211 } else {
4212 __kfree_skb(skb);
4213 }
4214}
4215EXPORT_SYMBOL(kfree_skb_partial);
4216
4217/**
4218 * skb_try_coalesce - try to merge skb to prior one
4219 * @to: prior buffer
4220 * @from: buffer to add
4221 * @fragstolen: pointer to boolean
4222 * @delta_truesize: how much more was allocated than was requested
4223 */
4224bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4225 bool *fragstolen, int *delta_truesize)
4226{
4227 int i, delta, len = from->len;
4228
4229 *fragstolen = false;
4230
4231 if (skb_cloned(to))
4232 return false;
4233
4234 if (len <= skb_tailroom(to)) {
4235 if (len)
4236 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4237 *delta_truesize = 0;
4238 return true;
4239 }
4240
4241 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4242 return false;
4243
4244 if (skb_headlen(from) != 0) {
4245 struct page *page;
4246 unsigned int offset;
4247
4248 if (skb_shinfo(to)->nr_frags +
4249 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4250 return false;
4251
4252 if (skb_head_is_locked(from))
4253 return false;
4254
4255 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4256
4257 page = virt_to_head_page(from->head);
4258 offset = from->data - (unsigned char *)page_address(page);
4259
4260 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4261 page, offset, skb_headlen(from));
4262 *fragstolen = true;
4263 } else {
4264 if (skb_shinfo(to)->nr_frags +
4265 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4266 return false;
4267
4268 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4269 }
4270
4271 WARN_ON_ONCE(delta < len);
4272
4273 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4274 skb_shinfo(from)->frags,
4275 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4276 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4277
4278 if (!skb_cloned(from))
4279 skb_shinfo(from)->nr_frags = 0;
4280
4281 /* if the skb is not cloned this does nothing
4282 * since we set nr_frags to 0.
4283 */
4284 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4285 skb_frag_ref(from, i);
4286
4287 to->truesize += delta;
4288 to->len += len;
4289 to->data_len += len;
4290
4291 *delta_truesize = delta;
4292 return true;
4293}
4294EXPORT_SYMBOL(skb_try_coalesce);
4295
4296/**
4297 * skb_scrub_packet - scrub an skb
4298 *
4299 * @skb: buffer to clean
4300 * @xnet: packet is crossing netns
4301 *
4302 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4303 * into/from a tunnel. Some information have to be cleared during these
4304 * operations.
4305 * skb_scrub_packet can also be used to clean a skb before injecting it in
4306 * another namespace (@xnet == true). We have to clear all information in the
4307 * skb that could impact namespace isolation.
4308 */
4309void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4310{
4311 skb->tstamp.tv64 = 0;
4312 skb->pkt_type = PACKET_HOST;
4313 skb->skb_iif = 0;
4314 skb->ignore_df = 0;
4315 skb_dst_drop(skb);
4316 secpath_reset(skb);
4317 nf_reset(skb);
4318 nf_reset_trace(skb);
4319
4320 if (!xnet)
4321 return;
4322
4323 skb_orphan(skb);
4324 skb->mark = 0;
4325}
4326EXPORT_SYMBOL_GPL(skb_scrub_packet);
4327
4328/**
4329 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4330 *
4331 * @skb: GSO skb
4332 *
4333 * skb_gso_transport_seglen is used to determine the real size of the
4334 * individual segments, including Layer4 headers (TCP/UDP).
4335 *
4336 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4337 */
4338unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4339{
4340 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4341 unsigned int thlen = 0;
4342
4343 if (skb->encapsulation) {
4344 thlen = skb_inner_transport_header(skb) -
4345 skb_transport_header(skb);
4346
4347 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4348 thlen += inner_tcp_hdrlen(skb);
4349 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4350 thlen = tcp_hdrlen(skb);
4351 }
4352 /* UFO sets gso_size to the size of the fragmentation
4353 * payload, i.e. the size of the L4 (UDP) header is already
4354 * accounted for.
4355 */
4356 return thlen + shinfo->gso_size;
4357}
4358EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4359
4360static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4361{
4362 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4363 kfree_skb(skb);
4364 return NULL;
4365 }
4366
4367 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4368 2 * ETH_ALEN);
4369 skb->mac_header += VLAN_HLEN;
4370 return skb;
4371}
4372
4373struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4374{
4375 struct vlan_hdr *vhdr;
4376 u16 vlan_tci;
4377
4378 if (unlikely(skb_vlan_tag_present(skb))) {
4379 /* vlan_tci is already set-up so leave this for another time */
4380 return skb;
4381 }
4382
4383 skb = skb_share_check(skb, GFP_ATOMIC);
4384 if (unlikely(!skb))
4385 goto err_free;
4386
4387 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4388 goto err_free;
4389
4390 vhdr = (struct vlan_hdr *)skb->data;
4391 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4392 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4393
4394 skb_pull_rcsum(skb, VLAN_HLEN);
4395 vlan_set_encap_proto(skb, vhdr);
4396
4397 skb = skb_reorder_vlan_header(skb);
4398 if (unlikely(!skb))
4399 goto err_free;
4400
4401 skb_reset_network_header(skb);
4402 skb_reset_transport_header(skb);
4403 skb_reset_mac_len(skb);
4404
4405 return skb;
4406
4407err_free:
4408 kfree_skb(skb);
4409 return NULL;
4410}
4411EXPORT_SYMBOL(skb_vlan_untag);
4412
4413int skb_ensure_writable(struct sk_buff *skb, int write_len)
4414{
4415 if (!pskb_may_pull(skb, write_len))
4416 return -ENOMEM;
4417
4418 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4419 return 0;
4420
4421 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4422}
4423EXPORT_SYMBOL(skb_ensure_writable);
4424
4425/* remove VLAN header from packet and update csum accordingly. */
4426static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4427{
4428 struct vlan_hdr *vhdr;
4429 unsigned int offset = skb->data - skb_mac_header(skb);
4430 int err;
4431
4432 __skb_push(skb, offset);
4433 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4434 if (unlikely(err))
4435 goto pull;
4436
4437 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4438
4439 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4440 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4441
4442 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4443 __skb_pull(skb, VLAN_HLEN);
4444
4445 vlan_set_encap_proto(skb, vhdr);
4446 skb->mac_header += VLAN_HLEN;
4447
4448 if (skb_network_offset(skb) < ETH_HLEN)
4449 skb_set_network_header(skb, ETH_HLEN);
4450
4451 skb_reset_mac_len(skb);
4452pull:
4453 __skb_pull(skb, offset);
4454
4455 return err;
4456}
4457
4458int skb_vlan_pop(struct sk_buff *skb)
4459{
4460 u16 vlan_tci;
4461 __be16 vlan_proto;
4462 int err;
4463
4464 if (likely(skb_vlan_tag_present(skb))) {
4465 skb->vlan_tci = 0;
4466 } else {
4467 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4468 skb->protocol != htons(ETH_P_8021AD)) ||
4469 skb->len < VLAN_ETH_HLEN))
4470 return 0;
4471
4472 err = __skb_vlan_pop(skb, &vlan_tci);
4473 if (err)
4474 return err;
4475 }
4476 /* move next vlan tag to hw accel tag */
4477 if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4478 skb->protocol != htons(ETH_P_8021AD)) ||
4479 skb->len < VLAN_ETH_HLEN))
4480 return 0;
4481
4482 vlan_proto = skb->protocol;
4483 err = __skb_vlan_pop(skb, &vlan_tci);
4484 if (unlikely(err))
4485 return err;
4486
4487 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4488 return 0;
4489}
4490EXPORT_SYMBOL(skb_vlan_pop);
4491
4492int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4493{
4494 if (skb_vlan_tag_present(skb)) {
4495 unsigned int offset = skb->data - skb_mac_header(skb);
4496 int err;
4497
4498 /* __vlan_insert_tag expect skb->data pointing to mac header.
4499 * So change skb->data before calling it and change back to
4500 * original position later
4501 */
4502 __skb_push(skb, offset);
4503 err = __vlan_insert_tag(skb, skb->vlan_proto,
4504 skb_vlan_tag_get(skb));
4505 if (err) {
4506 __skb_pull(skb, offset);
4507 return err;
4508 }
4509
4510 skb->protocol = skb->vlan_proto;
4511 skb->mac_len += VLAN_HLEN;
4512
4513 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4514 __skb_pull(skb, offset);
4515 }
4516 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4517 return 0;
4518}
4519EXPORT_SYMBOL(skb_vlan_push);
4520
4521/**
4522 * alloc_skb_with_frags - allocate skb with page frags
4523 *
4524 * @header_len: size of linear part
4525 * @data_len: needed length in frags
4526 * @max_page_order: max page order desired.
4527 * @errcode: pointer to error code if any
4528 * @gfp_mask: allocation mask
4529 *
4530 * This can be used to allocate a paged skb, given a maximal order for frags.
4531 */
4532struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4533 unsigned long data_len,
4534 int max_page_order,
4535 int *errcode,
4536 gfp_t gfp_mask)
4537{
4538 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4539 unsigned long chunk;
4540 struct sk_buff *skb;
4541 struct page *page;
4542 gfp_t gfp_head;
4543 int i;
4544
4545 *errcode = -EMSGSIZE;
4546 /* Note this test could be relaxed, if we succeed to allocate
4547 * high order pages...
4548 */
4549 if (npages > MAX_SKB_FRAGS)
4550 return NULL;
4551
4552 gfp_head = gfp_mask;
4553 if (gfp_head & __GFP_DIRECT_RECLAIM)
4554 gfp_head |= __GFP_REPEAT;
4555
4556 *errcode = -ENOBUFS;
4557 skb = alloc_skb(header_len, gfp_head);
4558 if (!skb)
4559 return NULL;
4560
4561 skb->truesize += npages << PAGE_SHIFT;
4562
4563 for (i = 0; npages > 0; i++) {
4564 int order = max_page_order;
4565
4566 while (order) {
4567 if (npages >= 1 << order) {
4568 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4569 __GFP_COMP |
4570 __GFP_NOWARN |
4571 __GFP_NORETRY,
4572 order);
4573 if (page)
4574 goto fill_page;
4575 /* Do not retry other high order allocations */
4576 order = 1;
4577 max_page_order = 0;
4578 }
4579 order--;
4580 }
4581 page = alloc_page(gfp_mask);
4582 if (!page)
4583 goto failure;
4584fill_page:
4585 chunk = min_t(unsigned long, data_len,
4586 PAGE_SIZE << order);
4587 skb_fill_page_desc(skb, i, page, 0, chunk);
4588 data_len -= chunk;
4589 npages -= 1 << order;
4590 }
4591 return skb;
4592
4593failure:
4594 kfree_skb(skb);
4595 return NULL;
4596}
4597EXPORT_SYMBOL(alloc_skb_with_frags);
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/kmemcheck.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/tcp.h>
51#include <linux/udp.h>
52#include <linux/sctp.h>
53#include <linux/netdevice.h>
54#ifdef CONFIG_NET_CLS_ACT
55#include <net/pkt_sched.h>
56#endif
57#include <linux/string.h>
58#include <linux/skbuff.h>
59#include <linux/splice.h>
60#include <linux/cache.h>
61#include <linux/rtnetlink.h>
62#include <linux/init.h>
63#include <linux/scatterlist.h>
64#include <linux/errqueue.h>
65#include <linux/prefetch.h>
66#include <linux/if_vlan.h>
67
68#include <net/protocol.h>
69#include <net/dst.h>
70#include <net/sock.h>
71#include <net/checksum.h>
72#include <net/ip6_checksum.h>
73#include <net/xfrm.h>
74
75#include <linux/uaccess.h>
76#include <trace/events/skb.h>
77#include <linux/highmem.h>
78#include <linux/capability.h>
79#include <linux/user_namespace.h>
80
81struct kmem_cache *skbuff_head_cache __read_mostly;
82static struct kmem_cache *skbuff_fclone_cache __read_mostly;
83int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
84EXPORT_SYMBOL(sysctl_max_skb_frags);
85
86/**
87 * skb_panic - private function for out-of-line support
88 * @skb: buffer
89 * @sz: size
90 * @addr: address
91 * @msg: skb_over_panic or skb_under_panic
92 *
93 * Out-of-line support for skb_put() and skb_push().
94 * Called via the wrapper skb_over_panic() or skb_under_panic().
95 * Keep out of line to prevent kernel bloat.
96 * __builtin_return_address is not used because it is not always reliable.
97 */
98static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
99 const char msg[])
100{
101 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
102 msg, addr, skb->len, sz, skb->head, skb->data,
103 (unsigned long)skb->tail, (unsigned long)skb->end,
104 skb->dev ? skb->dev->name : "<NULL>");
105 BUG();
106}
107
108static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
109{
110 skb_panic(skb, sz, addr, __func__);
111}
112
113static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
114{
115 skb_panic(skb, sz, addr, __func__);
116}
117
118/*
119 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
120 * the caller if emergency pfmemalloc reserves are being used. If it is and
121 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
122 * may be used. Otherwise, the packet data may be discarded until enough
123 * memory is free
124 */
125#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
126 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
127
128static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
129 unsigned long ip, bool *pfmemalloc)
130{
131 void *obj;
132 bool ret_pfmemalloc = false;
133
134 /*
135 * Try a regular allocation, when that fails and we're not entitled
136 * to the reserves, fail.
137 */
138 obj = kmalloc_node_track_caller(size,
139 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
140 node);
141 if (obj || !(gfp_pfmemalloc_allowed(flags)))
142 goto out;
143
144 /* Try again but now we are using pfmemalloc reserves */
145 ret_pfmemalloc = true;
146 obj = kmalloc_node_track_caller(size, flags, node);
147
148out:
149 if (pfmemalloc)
150 *pfmemalloc = ret_pfmemalloc;
151
152 return obj;
153}
154
155/* Allocate a new skbuff. We do this ourselves so we can fill in a few
156 * 'private' fields and also do memory statistics to find all the
157 * [BEEP] leaks.
158 *
159 */
160
161struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
162{
163 struct sk_buff *skb;
164
165 /* Get the HEAD */
166 skb = kmem_cache_alloc_node(skbuff_head_cache,
167 gfp_mask & ~__GFP_DMA, node);
168 if (!skb)
169 goto out;
170
171 /*
172 * Only clear those fields we need to clear, not those that we will
173 * actually initialise below. Hence, don't put any more fields after
174 * the tail pointer in struct sk_buff!
175 */
176 memset(skb, 0, offsetof(struct sk_buff, tail));
177 skb->head = NULL;
178 skb->truesize = sizeof(struct sk_buff);
179 atomic_set(&skb->users, 1);
180
181 skb->mac_header = (typeof(skb->mac_header))~0U;
182out:
183 return skb;
184}
185
186/**
187 * __alloc_skb - allocate a network buffer
188 * @size: size to allocate
189 * @gfp_mask: allocation mask
190 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
191 * instead of head cache and allocate a cloned (child) skb.
192 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
193 * allocations in case the data is required for writeback
194 * @node: numa node to allocate memory on
195 *
196 * Allocate a new &sk_buff. The returned buffer has no headroom and a
197 * tail room of at least size bytes. The object has a reference count
198 * of one. The return is the buffer. On a failure the return is %NULL.
199 *
200 * Buffers may only be allocated from interrupts using a @gfp_mask of
201 * %GFP_ATOMIC.
202 */
203struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
204 int flags, int node)
205{
206 struct kmem_cache *cache;
207 struct skb_shared_info *shinfo;
208 struct sk_buff *skb;
209 u8 *data;
210 bool pfmemalloc;
211
212 cache = (flags & SKB_ALLOC_FCLONE)
213 ? skbuff_fclone_cache : skbuff_head_cache;
214
215 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
216 gfp_mask |= __GFP_MEMALLOC;
217
218 /* Get the HEAD */
219 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
220 if (!skb)
221 goto out;
222 prefetchw(skb);
223
224 /* We do our best to align skb_shared_info on a separate cache
225 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
226 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
227 * Both skb->head and skb_shared_info are cache line aligned.
228 */
229 size = SKB_DATA_ALIGN(size);
230 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
231 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
232 if (!data)
233 goto nodata;
234 /* kmalloc(size) might give us more room than requested.
235 * Put skb_shared_info exactly at the end of allocated zone,
236 * to allow max possible filling before reallocation.
237 */
238 size = SKB_WITH_OVERHEAD(ksize(data));
239 prefetchw(data + size);
240
241 /*
242 * Only clear those fields we need to clear, not those that we will
243 * actually initialise below. Hence, don't put any more fields after
244 * the tail pointer in struct sk_buff!
245 */
246 memset(skb, 0, offsetof(struct sk_buff, tail));
247 /* Account for allocated memory : skb + skb->head */
248 skb->truesize = SKB_TRUESIZE(size);
249 skb->pfmemalloc = pfmemalloc;
250 atomic_set(&skb->users, 1);
251 skb->head = data;
252 skb->data = data;
253 skb_reset_tail_pointer(skb);
254 skb->end = skb->tail + size;
255 skb->mac_header = (typeof(skb->mac_header))~0U;
256 skb->transport_header = (typeof(skb->transport_header))~0U;
257
258 /* make sure we initialize shinfo sequentially */
259 shinfo = skb_shinfo(skb);
260 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
261 atomic_set(&shinfo->dataref, 1);
262 kmemcheck_annotate_variable(shinfo->destructor_arg);
263
264 if (flags & SKB_ALLOC_FCLONE) {
265 struct sk_buff_fclones *fclones;
266
267 fclones = container_of(skb, struct sk_buff_fclones, skb1);
268
269 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
270 skb->fclone = SKB_FCLONE_ORIG;
271 atomic_set(&fclones->fclone_ref, 1);
272
273 fclones->skb2.fclone = SKB_FCLONE_CLONE;
274 fclones->skb2.pfmemalloc = pfmemalloc;
275 }
276out:
277 return skb;
278nodata:
279 kmem_cache_free(cache, skb);
280 skb = NULL;
281 goto out;
282}
283EXPORT_SYMBOL(__alloc_skb);
284
285/**
286 * __build_skb - build a network buffer
287 * @data: data buffer provided by caller
288 * @frag_size: size of data, or 0 if head was kmalloced
289 *
290 * Allocate a new &sk_buff. Caller provides space holding head and
291 * skb_shared_info. @data must have been allocated by kmalloc() only if
292 * @frag_size is 0, otherwise data should come from the page allocator
293 * or vmalloc()
294 * The return is the new skb buffer.
295 * On a failure the return is %NULL, and @data is not freed.
296 * Notes :
297 * Before IO, driver allocates only data buffer where NIC put incoming frame
298 * Driver should add room at head (NET_SKB_PAD) and
299 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
300 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
301 * before giving packet to stack.
302 * RX rings only contains data buffers, not full skbs.
303 */
304struct sk_buff *__build_skb(void *data, unsigned int frag_size)
305{
306 struct skb_shared_info *shinfo;
307 struct sk_buff *skb;
308 unsigned int size = frag_size ? : ksize(data);
309
310 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
311 if (!skb)
312 return NULL;
313
314 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
315
316 memset(skb, 0, offsetof(struct sk_buff, tail));
317 skb->truesize = SKB_TRUESIZE(size);
318 atomic_set(&skb->users, 1);
319 skb->head = data;
320 skb->data = data;
321 skb_reset_tail_pointer(skb);
322 skb->end = skb->tail + size;
323 skb->mac_header = (typeof(skb->mac_header))~0U;
324 skb->transport_header = (typeof(skb->transport_header))~0U;
325
326 /* make sure we initialize shinfo sequentially */
327 shinfo = skb_shinfo(skb);
328 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
329 atomic_set(&shinfo->dataref, 1);
330 kmemcheck_annotate_variable(shinfo->destructor_arg);
331
332 return skb;
333}
334
335/* build_skb() is wrapper over __build_skb(), that specifically
336 * takes care of skb->head and skb->pfmemalloc
337 * This means that if @frag_size is not zero, then @data must be backed
338 * by a page fragment, not kmalloc() or vmalloc()
339 */
340struct sk_buff *build_skb(void *data, unsigned int frag_size)
341{
342 struct sk_buff *skb = __build_skb(data, frag_size);
343
344 if (skb && frag_size) {
345 skb->head_frag = 1;
346 if (page_is_pfmemalloc(virt_to_head_page(data)))
347 skb->pfmemalloc = 1;
348 }
349 return skb;
350}
351EXPORT_SYMBOL(build_skb);
352
353#define NAPI_SKB_CACHE_SIZE 64
354
355struct napi_alloc_cache {
356 struct page_frag_cache page;
357 unsigned int skb_count;
358 void *skb_cache[NAPI_SKB_CACHE_SIZE];
359};
360
361static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
362static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
363
364static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
365{
366 struct page_frag_cache *nc;
367 unsigned long flags;
368 void *data;
369
370 local_irq_save(flags);
371 nc = this_cpu_ptr(&netdev_alloc_cache);
372 data = page_frag_alloc(nc, fragsz, gfp_mask);
373 local_irq_restore(flags);
374 return data;
375}
376
377/**
378 * netdev_alloc_frag - allocate a page fragment
379 * @fragsz: fragment size
380 *
381 * Allocates a frag from a page for receive buffer.
382 * Uses GFP_ATOMIC allocations.
383 */
384void *netdev_alloc_frag(unsigned int fragsz)
385{
386 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
387}
388EXPORT_SYMBOL(netdev_alloc_frag);
389
390static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
391{
392 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
393
394 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
395}
396
397void *napi_alloc_frag(unsigned int fragsz)
398{
399 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
400}
401EXPORT_SYMBOL(napi_alloc_frag);
402
403/**
404 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
405 * @dev: network device to receive on
406 * @len: length to allocate
407 * @gfp_mask: get_free_pages mask, passed to alloc_skb
408 *
409 * Allocate a new &sk_buff and assign it a usage count of one. The
410 * buffer has NET_SKB_PAD headroom built in. Users should allocate
411 * the headroom they think they need without accounting for the
412 * built in space. The built in space is used for optimisations.
413 *
414 * %NULL is returned if there is no free memory.
415 */
416struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
417 gfp_t gfp_mask)
418{
419 struct page_frag_cache *nc;
420 unsigned long flags;
421 struct sk_buff *skb;
422 bool pfmemalloc;
423 void *data;
424
425 len += NET_SKB_PAD;
426
427 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
428 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
429 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
430 if (!skb)
431 goto skb_fail;
432 goto skb_success;
433 }
434
435 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
436 len = SKB_DATA_ALIGN(len);
437
438 if (sk_memalloc_socks())
439 gfp_mask |= __GFP_MEMALLOC;
440
441 local_irq_save(flags);
442
443 nc = this_cpu_ptr(&netdev_alloc_cache);
444 data = page_frag_alloc(nc, len, gfp_mask);
445 pfmemalloc = nc->pfmemalloc;
446
447 local_irq_restore(flags);
448
449 if (unlikely(!data))
450 return NULL;
451
452 skb = __build_skb(data, len);
453 if (unlikely(!skb)) {
454 skb_free_frag(data);
455 return NULL;
456 }
457
458 /* use OR instead of assignment to avoid clearing of bits in mask */
459 if (pfmemalloc)
460 skb->pfmemalloc = 1;
461 skb->head_frag = 1;
462
463skb_success:
464 skb_reserve(skb, NET_SKB_PAD);
465 skb->dev = dev;
466
467skb_fail:
468 return skb;
469}
470EXPORT_SYMBOL(__netdev_alloc_skb);
471
472/**
473 * __napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
474 * @napi: napi instance this buffer was allocated for
475 * @len: length to allocate
476 * @gfp_mask: get_free_pages mask, passed to alloc_skb and alloc_pages
477 *
478 * Allocate a new sk_buff for use in NAPI receive. This buffer will
479 * attempt to allocate the head from a special reserved region used
480 * only for NAPI Rx allocation. By doing this we can save several
481 * CPU cycles by avoiding having to disable and re-enable IRQs.
482 *
483 * %NULL is returned if there is no free memory.
484 */
485struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
486 gfp_t gfp_mask)
487{
488 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
489 struct sk_buff *skb;
490 void *data;
491
492 len += NET_SKB_PAD + NET_IP_ALIGN;
493
494 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
495 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
496 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
497 if (!skb)
498 goto skb_fail;
499 goto skb_success;
500 }
501
502 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
503 len = SKB_DATA_ALIGN(len);
504
505 if (sk_memalloc_socks())
506 gfp_mask |= __GFP_MEMALLOC;
507
508 data = page_frag_alloc(&nc->page, len, gfp_mask);
509 if (unlikely(!data))
510 return NULL;
511
512 skb = __build_skb(data, len);
513 if (unlikely(!skb)) {
514 skb_free_frag(data);
515 return NULL;
516 }
517
518 /* use OR instead of assignment to avoid clearing of bits in mask */
519 if (nc->page.pfmemalloc)
520 skb->pfmemalloc = 1;
521 skb->head_frag = 1;
522
523skb_success:
524 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
525 skb->dev = napi->dev;
526
527skb_fail:
528 return skb;
529}
530EXPORT_SYMBOL(__napi_alloc_skb);
531
532void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
533 int size, unsigned int truesize)
534{
535 skb_fill_page_desc(skb, i, page, off, size);
536 skb->len += size;
537 skb->data_len += size;
538 skb->truesize += truesize;
539}
540EXPORT_SYMBOL(skb_add_rx_frag);
541
542void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
543 unsigned int truesize)
544{
545 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
546
547 skb_frag_size_add(frag, size);
548 skb->len += size;
549 skb->data_len += size;
550 skb->truesize += truesize;
551}
552EXPORT_SYMBOL(skb_coalesce_rx_frag);
553
554static void skb_drop_list(struct sk_buff **listp)
555{
556 kfree_skb_list(*listp);
557 *listp = NULL;
558}
559
560static inline void skb_drop_fraglist(struct sk_buff *skb)
561{
562 skb_drop_list(&skb_shinfo(skb)->frag_list);
563}
564
565static void skb_clone_fraglist(struct sk_buff *skb)
566{
567 struct sk_buff *list;
568
569 skb_walk_frags(skb, list)
570 skb_get(list);
571}
572
573static void skb_free_head(struct sk_buff *skb)
574{
575 unsigned char *head = skb->head;
576
577 if (skb->head_frag)
578 skb_free_frag(head);
579 else
580 kfree(head);
581}
582
583static void skb_release_data(struct sk_buff *skb)
584{
585 struct skb_shared_info *shinfo = skb_shinfo(skb);
586 int i;
587
588 if (skb->cloned &&
589 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
590 &shinfo->dataref))
591 return;
592
593 for (i = 0; i < shinfo->nr_frags; i++)
594 __skb_frag_unref(&shinfo->frags[i]);
595
596 /*
597 * If skb buf is from userspace, we need to notify the caller
598 * the lower device DMA has done;
599 */
600 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
601 struct ubuf_info *uarg;
602
603 uarg = shinfo->destructor_arg;
604 if (uarg->callback)
605 uarg->callback(uarg, true);
606 }
607
608 if (shinfo->frag_list)
609 kfree_skb_list(shinfo->frag_list);
610
611 skb_free_head(skb);
612}
613
614/*
615 * Free an skbuff by memory without cleaning the state.
616 */
617static void kfree_skbmem(struct sk_buff *skb)
618{
619 struct sk_buff_fclones *fclones;
620
621 switch (skb->fclone) {
622 case SKB_FCLONE_UNAVAILABLE:
623 kmem_cache_free(skbuff_head_cache, skb);
624 return;
625
626 case SKB_FCLONE_ORIG:
627 fclones = container_of(skb, struct sk_buff_fclones, skb1);
628
629 /* We usually free the clone (TX completion) before original skb
630 * This test would have no chance to be true for the clone,
631 * while here, branch prediction will be good.
632 */
633 if (atomic_read(&fclones->fclone_ref) == 1)
634 goto fastpath;
635 break;
636
637 default: /* SKB_FCLONE_CLONE */
638 fclones = container_of(skb, struct sk_buff_fclones, skb2);
639 break;
640 }
641 if (!atomic_dec_and_test(&fclones->fclone_ref))
642 return;
643fastpath:
644 kmem_cache_free(skbuff_fclone_cache, fclones);
645}
646
647static void skb_release_head_state(struct sk_buff *skb)
648{
649 skb_dst_drop(skb);
650#ifdef CONFIG_XFRM
651 secpath_put(skb->sp);
652#endif
653 if (skb->destructor) {
654 WARN_ON(in_irq());
655 skb->destructor(skb);
656 }
657#if IS_ENABLED(CONFIG_NF_CONNTRACK)
658 nf_conntrack_put(skb->nfct);
659#endif
660#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
661 nf_bridge_put(skb->nf_bridge);
662#endif
663}
664
665/* Free everything but the sk_buff shell. */
666static void skb_release_all(struct sk_buff *skb)
667{
668 skb_release_head_state(skb);
669 if (likely(skb->head))
670 skb_release_data(skb);
671}
672
673/**
674 * __kfree_skb - private function
675 * @skb: buffer
676 *
677 * Free an sk_buff. Release anything attached to the buffer.
678 * Clean the state. This is an internal helper function. Users should
679 * always call kfree_skb
680 */
681
682void __kfree_skb(struct sk_buff *skb)
683{
684 skb_release_all(skb);
685 kfree_skbmem(skb);
686}
687EXPORT_SYMBOL(__kfree_skb);
688
689/**
690 * kfree_skb - free an sk_buff
691 * @skb: buffer to free
692 *
693 * Drop a reference to the buffer and free it if the usage count has
694 * hit zero.
695 */
696void kfree_skb(struct sk_buff *skb)
697{
698 if (unlikely(!skb))
699 return;
700 if (likely(atomic_read(&skb->users) == 1))
701 smp_rmb();
702 else if (likely(!atomic_dec_and_test(&skb->users)))
703 return;
704 trace_kfree_skb(skb, __builtin_return_address(0));
705 __kfree_skb(skb);
706}
707EXPORT_SYMBOL(kfree_skb);
708
709void kfree_skb_list(struct sk_buff *segs)
710{
711 while (segs) {
712 struct sk_buff *next = segs->next;
713
714 kfree_skb(segs);
715 segs = next;
716 }
717}
718EXPORT_SYMBOL(kfree_skb_list);
719
720/**
721 * skb_tx_error - report an sk_buff xmit error
722 * @skb: buffer that triggered an error
723 *
724 * Report xmit error if a device callback is tracking this skb.
725 * skb must be freed afterwards.
726 */
727void skb_tx_error(struct sk_buff *skb)
728{
729 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
730 struct ubuf_info *uarg;
731
732 uarg = skb_shinfo(skb)->destructor_arg;
733 if (uarg->callback)
734 uarg->callback(uarg, false);
735 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
736 }
737}
738EXPORT_SYMBOL(skb_tx_error);
739
740/**
741 * consume_skb - free an skbuff
742 * @skb: buffer to free
743 *
744 * Drop a ref to the buffer and free it if the usage count has hit zero
745 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
746 * is being dropped after a failure and notes that
747 */
748void consume_skb(struct sk_buff *skb)
749{
750 if (unlikely(!skb))
751 return;
752 if (likely(atomic_read(&skb->users) == 1))
753 smp_rmb();
754 else if (likely(!atomic_dec_and_test(&skb->users)))
755 return;
756 trace_consume_skb(skb);
757 __kfree_skb(skb);
758}
759EXPORT_SYMBOL(consume_skb);
760
761void __kfree_skb_flush(void)
762{
763 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
764
765 /* flush skb_cache if containing objects */
766 if (nc->skb_count) {
767 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
768 nc->skb_cache);
769 nc->skb_count = 0;
770 }
771}
772
773static inline void _kfree_skb_defer(struct sk_buff *skb)
774{
775 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
776
777 /* drop skb->head and call any destructors for packet */
778 skb_release_all(skb);
779
780 /* record skb to CPU local list */
781 nc->skb_cache[nc->skb_count++] = skb;
782
783#ifdef CONFIG_SLUB
784 /* SLUB writes into objects when freeing */
785 prefetchw(skb);
786#endif
787
788 /* flush skb_cache if it is filled */
789 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
790 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
791 nc->skb_cache);
792 nc->skb_count = 0;
793 }
794}
795void __kfree_skb_defer(struct sk_buff *skb)
796{
797 _kfree_skb_defer(skb);
798}
799
800void napi_consume_skb(struct sk_buff *skb, int budget)
801{
802 if (unlikely(!skb))
803 return;
804
805 /* Zero budget indicate non-NAPI context called us, like netpoll */
806 if (unlikely(!budget)) {
807 dev_consume_skb_any(skb);
808 return;
809 }
810
811 if (likely(atomic_read(&skb->users) == 1))
812 smp_rmb();
813 else if (likely(!atomic_dec_and_test(&skb->users)))
814 return;
815 /* if reaching here SKB is ready to free */
816 trace_consume_skb(skb);
817
818 /* if SKB is a clone, don't handle this case */
819 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
820 __kfree_skb(skb);
821 return;
822 }
823
824 _kfree_skb_defer(skb);
825}
826EXPORT_SYMBOL(napi_consume_skb);
827
828/* Make sure a field is enclosed inside headers_start/headers_end section */
829#define CHECK_SKB_FIELD(field) \
830 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
831 offsetof(struct sk_buff, headers_start)); \
832 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
833 offsetof(struct sk_buff, headers_end)); \
834
835static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
836{
837 new->tstamp = old->tstamp;
838 /* We do not copy old->sk */
839 new->dev = old->dev;
840 memcpy(new->cb, old->cb, sizeof(old->cb));
841 skb_dst_copy(new, old);
842#ifdef CONFIG_XFRM
843 new->sp = secpath_get(old->sp);
844#endif
845 __nf_copy(new, old, false);
846
847 /* Note : this field could be in headers_start/headers_end section
848 * It is not yet because we do not want to have a 16 bit hole
849 */
850 new->queue_mapping = old->queue_mapping;
851
852 memcpy(&new->headers_start, &old->headers_start,
853 offsetof(struct sk_buff, headers_end) -
854 offsetof(struct sk_buff, headers_start));
855 CHECK_SKB_FIELD(protocol);
856 CHECK_SKB_FIELD(csum);
857 CHECK_SKB_FIELD(hash);
858 CHECK_SKB_FIELD(priority);
859 CHECK_SKB_FIELD(skb_iif);
860 CHECK_SKB_FIELD(vlan_proto);
861 CHECK_SKB_FIELD(vlan_tci);
862 CHECK_SKB_FIELD(transport_header);
863 CHECK_SKB_FIELD(network_header);
864 CHECK_SKB_FIELD(mac_header);
865 CHECK_SKB_FIELD(inner_protocol);
866 CHECK_SKB_FIELD(inner_transport_header);
867 CHECK_SKB_FIELD(inner_network_header);
868 CHECK_SKB_FIELD(inner_mac_header);
869 CHECK_SKB_FIELD(mark);
870#ifdef CONFIG_NETWORK_SECMARK
871 CHECK_SKB_FIELD(secmark);
872#endif
873#ifdef CONFIG_NET_RX_BUSY_POLL
874 CHECK_SKB_FIELD(napi_id);
875#endif
876#ifdef CONFIG_XPS
877 CHECK_SKB_FIELD(sender_cpu);
878#endif
879#ifdef CONFIG_NET_SCHED
880 CHECK_SKB_FIELD(tc_index);
881#ifdef CONFIG_NET_CLS_ACT
882 CHECK_SKB_FIELD(tc_verd);
883#endif
884#endif
885
886}
887
888/*
889 * You should not add any new code to this function. Add it to
890 * __copy_skb_header above instead.
891 */
892static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
893{
894#define C(x) n->x = skb->x
895
896 n->next = n->prev = NULL;
897 n->sk = NULL;
898 __copy_skb_header(n, skb);
899
900 C(len);
901 C(data_len);
902 C(mac_len);
903 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
904 n->cloned = 1;
905 n->nohdr = 0;
906 n->destructor = NULL;
907 C(tail);
908 C(end);
909 C(head);
910 C(head_frag);
911 C(data);
912 C(truesize);
913 atomic_set(&n->users, 1);
914
915 atomic_inc(&(skb_shinfo(skb)->dataref));
916 skb->cloned = 1;
917
918 return n;
919#undef C
920}
921
922/**
923 * skb_morph - morph one skb into another
924 * @dst: the skb to receive the contents
925 * @src: the skb to supply the contents
926 *
927 * This is identical to skb_clone except that the target skb is
928 * supplied by the user.
929 *
930 * The target skb is returned upon exit.
931 */
932struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
933{
934 skb_release_all(dst);
935 return __skb_clone(dst, src);
936}
937EXPORT_SYMBOL_GPL(skb_morph);
938
939/**
940 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
941 * @skb: the skb to modify
942 * @gfp_mask: allocation priority
943 *
944 * This must be called on SKBTX_DEV_ZEROCOPY skb.
945 * It will copy all frags into kernel and drop the reference
946 * to userspace pages.
947 *
948 * If this function is called from an interrupt gfp_mask() must be
949 * %GFP_ATOMIC.
950 *
951 * Returns 0 on success or a negative error code on failure
952 * to allocate kernel memory to copy to.
953 */
954int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
955{
956 int i;
957 int num_frags = skb_shinfo(skb)->nr_frags;
958 struct page *page, *head = NULL;
959 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
960
961 for (i = 0; i < num_frags; i++) {
962 u8 *vaddr;
963 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
964
965 page = alloc_page(gfp_mask);
966 if (!page) {
967 while (head) {
968 struct page *next = (struct page *)page_private(head);
969 put_page(head);
970 head = next;
971 }
972 return -ENOMEM;
973 }
974 vaddr = kmap_atomic(skb_frag_page(f));
975 memcpy(page_address(page),
976 vaddr + f->page_offset, skb_frag_size(f));
977 kunmap_atomic(vaddr);
978 set_page_private(page, (unsigned long)head);
979 head = page;
980 }
981
982 /* skb frags release userspace buffers */
983 for (i = 0; i < num_frags; i++)
984 skb_frag_unref(skb, i);
985
986 uarg->callback(uarg, false);
987
988 /* skb frags point to kernel buffers */
989 for (i = num_frags - 1; i >= 0; i--) {
990 __skb_fill_page_desc(skb, i, head, 0,
991 skb_shinfo(skb)->frags[i].size);
992 head = (struct page *)page_private(head);
993 }
994
995 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
996 return 0;
997}
998EXPORT_SYMBOL_GPL(skb_copy_ubufs);
999
1000/**
1001 * skb_clone - duplicate an sk_buff
1002 * @skb: buffer to clone
1003 * @gfp_mask: allocation priority
1004 *
1005 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
1006 * copies share the same packet data but not structure. The new
1007 * buffer has a reference count of 1. If the allocation fails the
1008 * function returns %NULL otherwise the new buffer is returned.
1009 *
1010 * If this function is called from an interrupt gfp_mask() must be
1011 * %GFP_ATOMIC.
1012 */
1013
1014struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1015{
1016 struct sk_buff_fclones *fclones = container_of(skb,
1017 struct sk_buff_fclones,
1018 skb1);
1019 struct sk_buff *n;
1020
1021 if (skb_orphan_frags(skb, gfp_mask))
1022 return NULL;
1023
1024 if (skb->fclone == SKB_FCLONE_ORIG &&
1025 atomic_read(&fclones->fclone_ref) == 1) {
1026 n = &fclones->skb2;
1027 atomic_set(&fclones->fclone_ref, 2);
1028 } else {
1029 if (skb_pfmemalloc(skb))
1030 gfp_mask |= __GFP_MEMALLOC;
1031
1032 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1033 if (!n)
1034 return NULL;
1035
1036 kmemcheck_annotate_bitfield(n, flags1);
1037 n->fclone = SKB_FCLONE_UNAVAILABLE;
1038 }
1039
1040 return __skb_clone(n, skb);
1041}
1042EXPORT_SYMBOL(skb_clone);
1043
1044static void skb_headers_offset_update(struct sk_buff *skb, int off)
1045{
1046 /* Only adjust this if it actually is csum_start rather than csum */
1047 if (skb->ip_summed == CHECKSUM_PARTIAL)
1048 skb->csum_start += off;
1049 /* {transport,network,mac}_header and tail are relative to skb->head */
1050 skb->transport_header += off;
1051 skb->network_header += off;
1052 if (skb_mac_header_was_set(skb))
1053 skb->mac_header += off;
1054 skb->inner_transport_header += off;
1055 skb->inner_network_header += off;
1056 skb->inner_mac_header += off;
1057}
1058
1059static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1060{
1061 __copy_skb_header(new, old);
1062
1063 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1064 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1065 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1066}
1067
1068static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1069{
1070 if (skb_pfmemalloc(skb))
1071 return SKB_ALLOC_RX;
1072 return 0;
1073}
1074
1075/**
1076 * skb_copy - create private copy of an sk_buff
1077 * @skb: buffer to copy
1078 * @gfp_mask: allocation priority
1079 *
1080 * Make a copy of both an &sk_buff and its data. This is used when the
1081 * caller wishes to modify the data and needs a private copy of the
1082 * data to alter. Returns %NULL on failure or the pointer to the buffer
1083 * on success. The returned buffer has a reference count of 1.
1084 *
1085 * As by-product this function converts non-linear &sk_buff to linear
1086 * one, so that &sk_buff becomes completely private and caller is allowed
1087 * to modify all the data of returned buffer. This means that this
1088 * function is not recommended for use in circumstances when only
1089 * header is going to be modified. Use pskb_copy() instead.
1090 */
1091
1092struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1093{
1094 int headerlen = skb_headroom(skb);
1095 unsigned int size = skb_end_offset(skb) + skb->data_len;
1096 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1097 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1098
1099 if (!n)
1100 return NULL;
1101
1102 /* Set the data pointer */
1103 skb_reserve(n, headerlen);
1104 /* Set the tail pointer and length */
1105 skb_put(n, skb->len);
1106
1107 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1108 BUG();
1109
1110 copy_skb_header(n, skb);
1111 return n;
1112}
1113EXPORT_SYMBOL(skb_copy);
1114
1115/**
1116 * __pskb_copy_fclone - create copy of an sk_buff with private head.
1117 * @skb: buffer to copy
1118 * @headroom: headroom of new skb
1119 * @gfp_mask: allocation priority
1120 * @fclone: if true allocate the copy of the skb from the fclone
1121 * cache instead of the head cache; it is recommended to set this
1122 * to true for the cases where the copy will likely be cloned
1123 *
1124 * Make a copy of both an &sk_buff and part of its data, located
1125 * in header. Fragmented data remain shared. This is used when
1126 * the caller wishes to modify only header of &sk_buff and needs
1127 * private copy of the header to alter. Returns %NULL on failure
1128 * or the pointer to the buffer on success.
1129 * The returned buffer has a reference count of 1.
1130 */
1131
1132struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1133 gfp_t gfp_mask, bool fclone)
1134{
1135 unsigned int size = skb_headlen(skb) + headroom;
1136 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1137 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1138
1139 if (!n)
1140 goto out;
1141
1142 /* Set the data pointer */
1143 skb_reserve(n, headroom);
1144 /* Set the tail pointer and length */
1145 skb_put(n, skb_headlen(skb));
1146 /* Copy the bytes */
1147 skb_copy_from_linear_data(skb, n->data, n->len);
1148
1149 n->truesize += skb->data_len;
1150 n->data_len = skb->data_len;
1151 n->len = skb->len;
1152
1153 if (skb_shinfo(skb)->nr_frags) {
1154 int i;
1155
1156 if (skb_orphan_frags(skb, gfp_mask)) {
1157 kfree_skb(n);
1158 n = NULL;
1159 goto out;
1160 }
1161 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1162 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1163 skb_frag_ref(skb, i);
1164 }
1165 skb_shinfo(n)->nr_frags = i;
1166 }
1167
1168 if (skb_has_frag_list(skb)) {
1169 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1170 skb_clone_fraglist(n);
1171 }
1172
1173 copy_skb_header(n, skb);
1174out:
1175 return n;
1176}
1177EXPORT_SYMBOL(__pskb_copy_fclone);
1178
1179/**
1180 * pskb_expand_head - reallocate header of &sk_buff
1181 * @skb: buffer to reallocate
1182 * @nhead: room to add at head
1183 * @ntail: room to add at tail
1184 * @gfp_mask: allocation priority
1185 *
1186 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1187 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1188 * reference count of 1. Returns zero in the case of success or error,
1189 * if expansion failed. In the last case, &sk_buff is not changed.
1190 *
1191 * All the pointers pointing into skb header may change and must be
1192 * reloaded after call to this function.
1193 */
1194
1195int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1196 gfp_t gfp_mask)
1197{
1198 int i;
1199 u8 *data;
1200 int size = nhead + skb_end_offset(skb) + ntail;
1201 long off;
1202
1203 BUG_ON(nhead < 0);
1204
1205 if (skb_shared(skb))
1206 BUG();
1207
1208 size = SKB_DATA_ALIGN(size);
1209
1210 if (skb_pfmemalloc(skb))
1211 gfp_mask |= __GFP_MEMALLOC;
1212 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1213 gfp_mask, NUMA_NO_NODE, NULL);
1214 if (!data)
1215 goto nodata;
1216 size = SKB_WITH_OVERHEAD(ksize(data));
1217
1218 /* Copy only real data... and, alas, header. This should be
1219 * optimized for the cases when header is void.
1220 */
1221 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1222
1223 memcpy((struct skb_shared_info *)(data + size),
1224 skb_shinfo(skb),
1225 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1226
1227 /*
1228 * if shinfo is shared we must drop the old head gracefully, but if it
1229 * is not we can just drop the old head and let the existing refcount
1230 * be since all we did is relocate the values
1231 */
1232 if (skb_cloned(skb)) {
1233 /* copy this zero copy skb frags */
1234 if (skb_orphan_frags(skb, gfp_mask))
1235 goto nofrags;
1236 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1237 skb_frag_ref(skb, i);
1238
1239 if (skb_has_frag_list(skb))
1240 skb_clone_fraglist(skb);
1241
1242 skb_release_data(skb);
1243 } else {
1244 skb_free_head(skb);
1245 }
1246 off = (data + nhead) - skb->head;
1247
1248 skb->head = data;
1249 skb->head_frag = 0;
1250 skb->data += off;
1251#ifdef NET_SKBUFF_DATA_USES_OFFSET
1252 skb->end = size;
1253 off = nhead;
1254#else
1255 skb->end = skb->head + size;
1256#endif
1257 skb->tail += off;
1258 skb_headers_offset_update(skb, nhead);
1259 skb->cloned = 0;
1260 skb->hdr_len = 0;
1261 skb->nohdr = 0;
1262 atomic_set(&skb_shinfo(skb)->dataref, 1);
1263 return 0;
1264
1265nofrags:
1266 kfree(data);
1267nodata:
1268 return -ENOMEM;
1269}
1270EXPORT_SYMBOL(pskb_expand_head);
1271
1272/* Make private copy of skb with writable head and some headroom */
1273
1274struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1275{
1276 struct sk_buff *skb2;
1277 int delta = headroom - skb_headroom(skb);
1278
1279 if (delta <= 0)
1280 skb2 = pskb_copy(skb, GFP_ATOMIC);
1281 else {
1282 skb2 = skb_clone(skb, GFP_ATOMIC);
1283 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1284 GFP_ATOMIC)) {
1285 kfree_skb(skb2);
1286 skb2 = NULL;
1287 }
1288 }
1289 return skb2;
1290}
1291EXPORT_SYMBOL(skb_realloc_headroom);
1292
1293/**
1294 * skb_copy_expand - copy and expand sk_buff
1295 * @skb: buffer to copy
1296 * @newheadroom: new free bytes at head
1297 * @newtailroom: new free bytes at tail
1298 * @gfp_mask: allocation priority
1299 *
1300 * Make a copy of both an &sk_buff and its data and while doing so
1301 * allocate additional space.
1302 *
1303 * This is used when the caller wishes to modify the data and needs a
1304 * private copy of the data to alter as well as more space for new fields.
1305 * Returns %NULL on failure or the pointer to the buffer
1306 * on success. The returned buffer has a reference count of 1.
1307 *
1308 * You must pass %GFP_ATOMIC as the allocation priority if this function
1309 * is called from an interrupt.
1310 */
1311struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1312 int newheadroom, int newtailroom,
1313 gfp_t gfp_mask)
1314{
1315 /*
1316 * Allocate the copy buffer
1317 */
1318 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1319 gfp_mask, skb_alloc_rx_flag(skb),
1320 NUMA_NO_NODE);
1321 int oldheadroom = skb_headroom(skb);
1322 int head_copy_len, head_copy_off;
1323
1324 if (!n)
1325 return NULL;
1326
1327 skb_reserve(n, newheadroom);
1328
1329 /* Set the tail pointer and length */
1330 skb_put(n, skb->len);
1331
1332 head_copy_len = oldheadroom;
1333 head_copy_off = 0;
1334 if (newheadroom <= head_copy_len)
1335 head_copy_len = newheadroom;
1336 else
1337 head_copy_off = newheadroom - head_copy_len;
1338
1339 /* Copy the linear header and data. */
1340 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1341 skb->len + head_copy_len))
1342 BUG();
1343
1344 copy_skb_header(n, skb);
1345
1346 skb_headers_offset_update(n, newheadroom - oldheadroom);
1347
1348 return n;
1349}
1350EXPORT_SYMBOL(skb_copy_expand);
1351
1352/**
1353 * skb_pad - zero pad the tail of an skb
1354 * @skb: buffer to pad
1355 * @pad: space to pad
1356 *
1357 * Ensure that a buffer is followed by a padding area that is zero
1358 * filled. Used by network drivers which may DMA or transfer data
1359 * beyond the buffer end onto the wire.
1360 *
1361 * May return error in out of memory cases. The skb is freed on error.
1362 */
1363
1364int skb_pad(struct sk_buff *skb, int pad)
1365{
1366 int err;
1367 int ntail;
1368
1369 /* If the skbuff is non linear tailroom is always zero.. */
1370 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1371 memset(skb->data+skb->len, 0, pad);
1372 return 0;
1373 }
1374
1375 ntail = skb->data_len + pad - (skb->end - skb->tail);
1376 if (likely(skb_cloned(skb) || ntail > 0)) {
1377 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1378 if (unlikely(err))
1379 goto free_skb;
1380 }
1381
1382 /* FIXME: The use of this function with non-linear skb's really needs
1383 * to be audited.
1384 */
1385 err = skb_linearize(skb);
1386 if (unlikely(err))
1387 goto free_skb;
1388
1389 memset(skb->data + skb->len, 0, pad);
1390 return 0;
1391
1392free_skb:
1393 kfree_skb(skb);
1394 return err;
1395}
1396EXPORT_SYMBOL(skb_pad);
1397
1398/**
1399 * pskb_put - add data to the tail of a potentially fragmented buffer
1400 * @skb: start of the buffer to use
1401 * @tail: tail fragment of the buffer to use
1402 * @len: amount of data to add
1403 *
1404 * This function extends the used data area of the potentially
1405 * fragmented buffer. @tail must be the last fragment of @skb -- or
1406 * @skb itself. If this would exceed the total buffer size the kernel
1407 * will panic. A pointer to the first byte of the extra data is
1408 * returned.
1409 */
1410
1411unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1412{
1413 if (tail != skb) {
1414 skb->data_len += len;
1415 skb->len += len;
1416 }
1417 return skb_put(tail, len);
1418}
1419EXPORT_SYMBOL_GPL(pskb_put);
1420
1421/**
1422 * skb_put - add data to a buffer
1423 * @skb: buffer to use
1424 * @len: amount of data to add
1425 *
1426 * This function extends the used data area of the buffer. If this would
1427 * exceed the total buffer size the kernel will panic. A pointer to the
1428 * first byte of the extra data is returned.
1429 */
1430unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1431{
1432 unsigned char *tmp = skb_tail_pointer(skb);
1433 SKB_LINEAR_ASSERT(skb);
1434 skb->tail += len;
1435 skb->len += len;
1436 if (unlikely(skb->tail > skb->end))
1437 skb_over_panic(skb, len, __builtin_return_address(0));
1438 return tmp;
1439}
1440EXPORT_SYMBOL(skb_put);
1441
1442/**
1443 * skb_push - add data to the start of a buffer
1444 * @skb: buffer to use
1445 * @len: amount of data to add
1446 *
1447 * This function extends the used data area of the buffer at the buffer
1448 * start. If this would exceed the total buffer headroom the kernel will
1449 * panic. A pointer to the first byte of the extra data is returned.
1450 */
1451unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1452{
1453 skb->data -= len;
1454 skb->len += len;
1455 if (unlikely(skb->data<skb->head))
1456 skb_under_panic(skb, len, __builtin_return_address(0));
1457 return skb->data;
1458}
1459EXPORT_SYMBOL(skb_push);
1460
1461/**
1462 * skb_pull - remove data from the start of a buffer
1463 * @skb: buffer to use
1464 * @len: amount of data to remove
1465 *
1466 * This function removes data from the start of a buffer, returning
1467 * the memory to the headroom. A pointer to the next data in the buffer
1468 * is returned. Once the data has been pulled future pushes will overwrite
1469 * the old data.
1470 */
1471unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1472{
1473 return skb_pull_inline(skb, len);
1474}
1475EXPORT_SYMBOL(skb_pull);
1476
1477/**
1478 * skb_trim - remove end from a buffer
1479 * @skb: buffer to alter
1480 * @len: new length
1481 *
1482 * Cut the length of a buffer down by removing data from the tail. If
1483 * the buffer is already under the length specified it is not modified.
1484 * The skb must be linear.
1485 */
1486void skb_trim(struct sk_buff *skb, unsigned int len)
1487{
1488 if (skb->len > len)
1489 __skb_trim(skb, len);
1490}
1491EXPORT_SYMBOL(skb_trim);
1492
1493/* Trims skb to length len. It can change skb pointers.
1494 */
1495
1496int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1497{
1498 struct sk_buff **fragp;
1499 struct sk_buff *frag;
1500 int offset = skb_headlen(skb);
1501 int nfrags = skb_shinfo(skb)->nr_frags;
1502 int i;
1503 int err;
1504
1505 if (skb_cloned(skb) &&
1506 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1507 return err;
1508
1509 i = 0;
1510 if (offset >= len)
1511 goto drop_pages;
1512
1513 for (; i < nfrags; i++) {
1514 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1515
1516 if (end < len) {
1517 offset = end;
1518 continue;
1519 }
1520
1521 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1522
1523drop_pages:
1524 skb_shinfo(skb)->nr_frags = i;
1525
1526 for (; i < nfrags; i++)
1527 skb_frag_unref(skb, i);
1528
1529 if (skb_has_frag_list(skb))
1530 skb_drop_fraglist(skb);
1531 goto done;
1532 }
1533
1534 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1535 fragp = &frag->next) {
1536 int end = offset + frag->len;
1537
1538 if (skb_shared(frag)) {
1539 struct sk_buff *nfrag;
1540
1541 nfrag = skb_clone(frag, GFP_ATOMIC);
1542 if (unlikely(!nfrag))
1543 return -ENOMEM;
1544
1545 nfrag->next = frag->next;
1546 consume_skb(frag);
1547 frag = nfrag;
1548 *fragp = frag;
1549 }
1550
1551 if (end < len) {
1552 offset = end;
1553 continue;
1554 }
1555
1556 if (end > len &&
1557 unlikely((err = pskb_trim(frag, len - offset))))
1558 return err;
1559
1560 if (frag->next)
1561 skb_drop_list(&frag->next);
1562 break;
1563 }
1564
1565done:
1566 if (len > skb_headlen(skb)) {
1567 skb->data_len -= skb->len - len;
1568 skb->len = len;
1569 } else {
1570 skb->len = len;
1571 skb->data_len = 0;
1572 skb_set_tail_pointer(skb, len);
1573 }
1574
1575 return 0;
1576}
1577EXPORT_SYMBOL(___pskb_trim);
1578
1579/**
1580 * __pskb_pull_tail - advance tail of skb header
1581 * @skb: buffer to reallocate
1582 * @delta: number of bytes to advance tail
1583 *
1584 * The function makes a sense only on a fragmented &sk_buff,
1585 * it expands header moving its tail forward and copying necessary
1586 * data from fragmented part.
1587 *
1588 * &sk_buff MUST have reference count of 1.
1589 *
1590 * Returns %NULL (and &sk_buff does not change) if pull failed
1591 * or value of new tail of skb in the case of success.
1592 *
1593 * All the pointers pointing into skb header may change and must be
1594 * reloaded after call to this function.
1595 */
1596
1597/* Moves tail of skb head forward, copying data from fragmented part,
1598 * when it is necessary.
1599 * 1. It may fail due to malloc failure.
1600 * 2. It may change skb pointers.
1601 *
1602 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1603 */
1604unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1605{
1606 /* If skb has not enough free space at tail, get new one
1607 * plus 128 bytes for future expansions. If we have enough
1608 * room at tail, reallocate without expansion only if skb is cloned.
1609 */
1610 int i, k, eat = (skb->tail + delta) - skb->end;
1611
1612 if (eat > 0 || skb_cloned(skb)) {
1613 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1614 GFP_ATOMIC))
1615 return NULL;
1616 }
1617
1618 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1619 BUG();
1620
1621 /* Optimization: no fragments, no reasons to preestimate
1622 * size of pulled pages. Superb.
1623 */
1624 if (!skb_has_frag_list(skb))
1625 goto pull_pages;
1626
1627 /* Estimate size of pulled pages. */
1628 eat = delta;
1629 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1630 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1631
1632 if (size >= eat)
1633 goto pull_pages;
1634 eat -= size;
1635 }
1636
1637 /* If we need update frag list, we are in troubles.
1638 * Certainly, it possible to add an offset to skb data,
1639 * but taking into account that pulling is expected to
1640 * be very rare operation, it is worth to fight against
1641 * further bloating skb head and crucify ourselves here instead.
1642 * Pure masohism, indeed. 8)8)
1643 */
1644 if (eat) {
1645 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1646 struct sk_buff *clone = NULL;
1647 struct sk_buff *insp = NULL;
1648
1649 do {
1650 BUG_ON(!list);
1651
1652 if (list->len <= eat) {
1653 /* Eaten as whole. */
1654 eat -= list->len;
1655 list = list->next;
1656 insp = list;
1657 } else {
1658 /* Eaten partially. */
1659
1660 if (skb_shared(list)) {
1661 /* Sucks! We need to fork list. :-( */
1662 clone = skb_clone(list, GFP_ATOMIC);
1663 if (!clone)
1664 return NULL;
1665 insp = list->next;
1666 list = clone;
1667 } else {
1668 /* This may be pulled without
1669 * problems. */
1670 insp = list;
1671 }
1672 if (!pskb_pull(list, eat)) {
1673 kfree_skb(clone);
1674 return NULL;
1675 }
1676 break;
1677 }
1678 } while (eat);
1679
1680 /* Free pulled out fragments. */
1681 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1682 skb_shinfo(skb)->frag_list = list->next;
1683 kfree_skb(list);
1684 }
1685 /* And insert new clone at head. */
1686 if (clone) {
1687 clone->next = list;
1688 skb_shinfo(skb)->frag_list = clone;
1689 }
1690 }
1691 /* Success! Now we may commit changes to skb data. */
1692
1693pull_pages:
1694 eat = delta;
1695 k = 0;
1696 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1697 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1698
1699 if (size <= eat) {
1700 skb_frag_unref(skb, i);
1701 eat -= size;
1702 } else {
1703 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1704 if (eat) {
1705 skb_shinfo(skb)->frags[k].page_offset += eat;
1706 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1707 eat = 0;
1708 }
1709 k++;
1710 }
1711 }
1712 skb_shinfo(skb)->nr_frags = k;
1713
1714 skb->tail += delta;
1715 skb->data_len -= delta;
1716
1717 return skb_tail_pointer(skb);
1718}
1719EXPORT_SYMBOL(__pskb_pull_tail);
1720
1721/**
1722 * skb_copy_bits - copy bits from skb to kernel buffer
1723 * @skb: source skb
1724 * @offset: offset in source
1725 * @to: destination buffer
1726 * @len: number of bytes to copy
1727 *
1728 * Copy the specified number of bytes from the source skb to the
1729 * destination buffer.
1730 *
1731 * CAUTION ! :
1732 * If its prototype is ever changed,
1733 * check arch/{*}/net/{*}.S files,
1734 * since it is called from BPF assembly code.
1735 */
1736int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1737{
1738 int start = skb_headlen(skb);
1739 struct sk_buff *frag_iter;
1740 int i, copy;
1741
1742 if (offset > (int)skb->len - len)
1743 goto fault;
1744
1745 /* Copy header. */
1746 if ((copy = start - offset) > 0) {
1747 if (copy > len)
1748 copy = len;
1749 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1750 if ((len -= copy) == 0)
1751 return 0;
1752 offset += copy;
1753 to += copy;
1754 }
1755
1756 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1757 int end;
1758 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1759
1760 WARN_ON(start > offset + len);
1761
1762 end = start + skb_frag_size(f);
1763 if ((copy = end - offset) > 0) {
1764 u8 *vaddr;
1765
1766 if (copy > len)
1767 copy = len;
1768
1769 vaddr = kmap_atomic(skb_frag_page(f));
1770 memcpy(to,
1771 vaddr + f->page_offset + offset - start,
1772 copy);
1773 kunmap_atomic(vaddr);
1774
1775 if ((len -= copy) == 0)
1776 return 0;
1777 offset += copy;
1778 to += copy;
1779 }
1780 start = end;
1781 }
1782
1783 skb_walk_frags(skb, frag_iter) {
1784 int end;
1785
1786 WARN_ON(start > offset + len);
1787
1788 end = start + frag_iter->len;
1789 if ((copy = end - offset) > 0) {
1790 if (copy > len)
1791 copy = len;
1792 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1793 goto fault;
1794 if ((len -= copy) == 0)
1795 return 0;
1796 offset += copy;
1797 to += copy;
1798 }
1799 start = end;
1800 }
1801
1802 if (!len)
1803 return 0;
1804
1805fault:
1806 return -EFAULT;
1807}
1808EXPORT_SYMBOL(skb_copy_bits);
1809
1810/*
1811 * Callback from splice_to_pipe(), if we need to release some pages
1812 * at the end of the spd in case we error'ed out in filling the pipe.
1813 */
1814static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1815{
1816 put_page(spd->pages[i]);
1817}
1818
1819static struct page *linear_to_page(struct page *page, unsigned int *len,
1820 unsigned int *offset,
1821 struct sock *sk)
1822{
1823 struct page_frag *pfrag = sk_page_frag(sk);
1824
1825 if (!sk_page_frag_refill(sk, pfrag))
1826 return NULL;
1827
1828 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1829
1830 memcpy(page_address(pfrag->page) + pfrag->offset,
1831 page_address(page) + *offset, *len);
1832 *offset = pfrag->offset;
1833 pfrag->offset += *len;
1834
1835 return pfrag->page;
1836}
1837
1838static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1839 struct page *page,
1840 unsigned int offset)
1841{
1842 return spd->nr_pages &&
1843 spd->pages[spd->nr_pages - 1] == page &&
1844 (spd->partial[spd->nr_pages - 1].offset +
1845 spd->partial[spd->nr_pages - 1].len == offset);
1846}
1847
1848/*
1849 * Fill page/offset/length into spd, if it can hold more pages.
1850 */
1851static bool spd_fill_page(struct splice_pipe_desc *spd,
1852 struct pipe_inode_info *pipe, struct page *page,
1853 unsigned int *len, unsigned int offset,
1854 bool linear,
1855 struct sock *sk)
1856{
1857 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1858 return true;
1859
1860 if (linear) {
1861 page = linear_to_page(page, len, &offset, sk);
1862 if (!page)
1863 return true;
1864 }
1865 if (spd_can_coalesce(spd, page, offset)) {
1866 spd->partial[spd->nr_pages - 1].len += *len;
1867 return false;
1868 }
1869 get_page(page);
1870 spd->pages[spd->nr_pages] = page;
1871 spd->partial[spd->nr_pages].len = *len;
1872 spd->partial[spd->nr_pages].offset = offset;
1873 spd->nr_pages++;
1874
1875 return false;
1876}
1877
1878static bool __splice_segment(struct page *page, unsigned int poff,
1879 unsigned int plen, unsigned int *off,
1880 unsigned int *len,
1881 struct splice_pipe_desc *spd, bool linear,
1882 struct sock *sk,
1883 struct pipe_inode_info *pipe)
1884{
1885 if (!*len)
1886 return true;
1887
1888 /* skip this segment if already processed */
1889 if (*off >= plen) {
1890 *off -= plen;
1891 return false;
1892 }
1893
1894 /* ignore any bits we already processed */
1895 poff += *off;
1896 plen -= *off;
1897 *off = 0;
1898
1899 do {
1900 unsigned int flen = min(*len, plen);
1901
1902 if (spd_fill_page(spd, pipe, page, &flen, poff,
1903 linear, sk))
1904 return true;
1905 poff += flen;
1906 plen -= flen;
1907 *len -= flen;
1908 } while (*len && plen);
1909
1910 return false;
1911}
1912
1913/*
1914 * Map linear and fragment data from the skb to spd. It reports true if the
1915 * pipe is full or if we already spliced the requested length.
1916 */
1917static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1918 unsigned int *offset, unsigned int *len,
1919 struct splice_pipe_desc *spd, struct sock *sk)
1920{
1921 int seg;
1922 struct sk_buff *iter;
1923
1924 /* map the linear part :
1925 * If skb->head_frag is set, this 'linear' part is backed by a
1926 * fragment, and if the head is not shared with any clones then
1927 * we can avoid a copy since we own the head portion of this page.
1928 */
1929 if (__splice_segment(virt_to_page(skb->data),
1930 (unsigned long) skb->data & (PAGE_SIZE - 1),
1931 skb_headlen(skb),
1932 offset, len, spd,
1933 skb_head_is_locked(skb),
1934 sk, pipe))
1935 return true;
1936
1937 /*
1938 * then map the fragments
1939 */
1940 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1941 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1942
1943 if (__splice_segment(skb_frag_page(f),
1944 f->page_offset, skb_frag_size(f),
1945 offset, len, spd, false, sk, pipe))
1946 return true;
1947 }
1948
1949 skb_walk_frags(skb, iter) {
1950 if (*offset >= iter->len) {
1951 *offset -= iter->len;
1952 continue;
1953 }
1954 /* __skb_splice_bits() only fails if the output has no room
1955 * left, so no point in going over the frag_list for the error
1956 * case.
1957 */
1958 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1959 return true;
1960 }
1961
1962 return false;
1963}
1964
1965/*
1966 * Map data from the skb to a pipe. Should handle both the linear part,
1967 * the fragments, and the frag list.
1968 */
1969int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1970 struct pipe_inode_info *pipe, unsigned int tlen,
1971 unsigned int flags)
1972{
1973 struct partial_page partial[MAX_SKB_FRAGS];
1974 struct page *pages[MAX_SKB_FRAGS];
1975 struct splice_pipe_desc spd = {
1976 .pages = pages,
1977 .partial = partial,
1978 .nr_pages_max = MAX_SKB_FRAGS,
1979 .flags = flags,
1980 .ops = &nosteal_pipe_buf_ops,
1981 .spd_release = sock_spd_release,
1982 };
1983 int ret = 0;
1984
1985 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
1986
1987 if (spd.nr_pages)
1988 ret = splice_to_pipe(pipe, &spd);
1989
1990 return ret;
1991}
1992EXPORT_SYMBOL_GPL(skb_splice_bits);
1993
1994/**
1995 * skb_store_bits - store bits from kernel buffer to skb
1996 * @skb: destination buffer
1997 * @offset: offset in destination
1998 * @from: source buffer
1999 * @len: number of bytes to copy
2000 *
2001 * Copy the specified number of bytes from the source buffer to the
2002 * destination skb. This function handles all the messy bits of
2003 * traversing fragment lists and such.
2004 */
2005
2006int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2007{
2008 int start = skb_headlen(skb);
2009 struct sk_buff *frag_iter;
2010 int i, copy;
2011
2012 if (offset > (int)skb->len - len)
2013 goto fault;
2014
2015 if ((copy = start - offset) > 0) {
2016 if (copy > len)
2017 copy = len;
2018 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2019 if ((len -= copy) == 0)
2020 return 0;
2021 offset += copy;
2022 from += copy;
2023 }
2024
2025 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2026 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2027 int end;
2028
2029 WARN_ON(start > offset + len);
2030
2031 end = start + skb_frag_size(frag);
2032 if ((copy = end - offset) > 0) {
2033 u8 *vaddr;
2034
2035 if (copy > len)
2036 copy = len;
2037
2038 vaddr = kmap_atomic(skb_frag_page(frag));
2039 memcpy(vaddr + frag->page_offset + offset - start,
2040 from, copy);
2041 kunmap_atomic(vaddr);
2042
2043 if ((len -= copy) == 0)
2044 return 0;
2045 offset += copy;
2046 from += copy;
2047 }
2048 start = end;
2049 }
2050
2051 skb_walk_frags(skb, frag_iter) {
2052 int end;
2053
2054 WARN_ON(start > offset + len);
2055
2056 end = start + frag_iter->len;
2057 if ((copy = end - offset) > 0) {
2058 if (copy > len)
2059 copy = len;
2060 if (skb_store_bits(frag_iter, offset - start,
2061 from, copy))
2062 goto fault;
2063 if ((len -= copy) == 0)
2064 return 0;
2065 offset += copy;
2066 from += copy;
2067 }
2068 start = end;
2069 }
2070 if (!len)
2071 return 0;
2072
2073fault:
2074 return -EFAULT;
2075}
2076EXPORT_SYMBOL(skb_store_bits);
2077
2078/* Checksum skb data. */
2079__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2080 __wsum csum, const struct skb_checksum_ops *ops)
2081{
2082 int start = skb_headlen(skb);
2083 int i, copy = start - offset;
2084 struct sk_buff *frag_iter;
2085 int pos = 0;
2086
2087 /* Checksum header. */
2088 if (copy > 0) {
2089 if (copy > len)
2090 copy = len;
2091 csum = ops->update(skb->data + offset, copy, csum);
2092 if ((len -= copy) == 0)
2093 return csum;
2094 offset += copy;
2095 pos = copy;
2096 }
2097
2098 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2099 int end;
2100 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2101
2102 WARN_ON(start > offset + len);
2103
2104 end = start + skb_frag_size(frag);
2105 if ((copy = end - offset) > 0) {
2106 __wsum csum2;
2107 u8 *vaddr;
2108
2109 if (copy > len)
2110 copy = len;
2111 vaddr = kmap_atomic(skb_frag_page(frag));
2112 csum2 = ops->update(vaddr + frag->page_offset +
2113 offset - start, copy, 0);
2114 kunmap_atomic(vaddr);
2115 csum = ops->combine(csum, csum2, pos, copy);
2116 if (!(len -= copy))
2117 return csum;
2118 offset += copy;
2119 pos += copy;
2120 }
2121 start = end;
2122 }
2123
2124 skb_walk_frags(skb, frag_iter) {
2125 int end;
2126
2127 WARN_ON(start > offset + len);
2128
2129 end = start + frag_iter->len;
2130 if ((copy = end - offset) > 0) {
2131 __wsum csum2;
2132 if (copy > len)
2133 copy = len;
2134 csum2 = __skb_checksum(frag_iter, offset - start,
2135 copy, 0, ops);
2136 csum = ops->combine(csum, csum2, pos, copy);
2137 if ((len -= copy) == 0)
2138 return csum;
2139 offset += copy;
2140 pos += copy;
2141 }
2142 start = end;
2143 }
2144 BUG_ON(len);
2145
2146 return csum;
2147}
2148EXPORT_SYMBOL(__skb_checksum);
2149
2150__wsum skb_checksum(const struct sk_buff *skb, int offset,
2151 int len, __wsum csum)
2152{
2153 const struct skb_checksum_ops ops = {
2154 .update = csum_partial_ext,
2155 .combine = csum_block_add_ext,
2156 };
2157
2158 return __skb_checksum(skb, offset, len, csum, &ops);
2159}
2160EXPORT_SYMBOL(skb_checksum);
2161
2162/* Both of above in one bottle. */
2163
2164__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2165 u8 *to, int len, __wsum csum)
2166{
2167 int start = skb_headlen(skb);
2168 int i, copy = start - offset;
2169 struct sk_buff *frag_iter;
2170 int pos = 0;
2171
2172 /* Copy header. */
2173 if (copy > 0) {
2174 if (copy > len)
2175 copy = len;
2176 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2177 copy, csum);
2178 if ((len -= copy) == 0)
2179 return csum;
2180 offset += copy;
2181 to += copy;
2182 pos = copy;
2183 }
2184
2185 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2186 int end;
2187
2188 WARN_ON(start > offset + len);
2189
2190 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2191 if ((copy = end - offset) > 0) {
2192 __wsum csum2;
2193 u8 *vaddr;
2194 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2195
2196 if (copy > len)
2197 copy = len;
2198 vaddr = kmap_atomic(skb_frag_page(frag));
2199 csum2 = csum_partial_copy_nocheck(vaddr +
2200 frag->page_offset +
2201 offset - start, to,
2202 copy, 0);
2203 kunmap_atomic(vaddr);
2204 csum = csum_block_add(csum, csum2, pos);
2205 if (!(len -= copy))
2206 return csum;
2207 offset += copy;
2208 to += copy;
2209 pos += copy;
2210 }
2211 start = end;
2212 }
2213
2214 skb_walk_frags(skb, frag_iter) {
2215 __wsum csum2;
2216 int end;
2217
2218 WARN_ON(start > offset + len);
2219
2220 end = start + frag_iter->len;
2221 if ((copy = end - offset) > 0) {
2222 if (copy > len)
2223 copy = len;
2224 csum2 = skb_copy_and_csum_bits(frag_iter,
2225 offset - start,
2226 to, copy, 0);
2227 csum = csum_block_add(csum, csum2, pos);
2228 if ((len -= copy) == 0)
2229 return csum;
2230 offset += copy;
2231 to += copy;
2232 pos += copy;
2233 }
2234 start = end;
2235 }
2236 BUG_ON(len);
2237 return csum;
2238}
2239EXPORT_SYMBOL(skb_copy_and_csum_bits);
2240
2241 /**
2242 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2243 * @from: source buffer
2244 *
2245 * Calculates the amount of linear headroom needed in the 'to' skb passed
2246 * into skb_zerocopy().
2247 */
2248unsigned int
2249skb_zerocopy_headlen(const struct sk_buff *from)
2250{
2251 unsigned int hlen = 0;
2252
2253 if (!from->head_frag ||
2254 skb_headlen(from) < L1_CACHE_BYTES ||
2255 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2256 hlen = skb_headlen(from);
2257
2258 if (skb_has_frag_list(from))
2259 hlen = from->len;
2260
2261 return hlen;
2262}
2263EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2264
2265/**
2266 * skb_zerocopy - Zero copy skb to skb
2267 * @to: destination buffer
2268 * @from: source buffer
2269 * @len: number of bytes to copy from source buffer
2270 * @hlen: size of linear headroom in destination buffer
2271 *
2272 * Copies up to `len` bytes from `from` to `to` by creating references
2273 * to the frags in the source buffer.
2274 *
2275 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2276 * headroom in the `to` buffer.
2277 *
2278 * Return value:
2279 * 0: everything is OK
2280 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2281 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2282 */
2283int
2284skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2285{
2286 int i, j = 0;
2287 int plen = 0; /* length of skb->head fragment */
2288 int ret;
2289 struct page *page;
2290 unsigned int offset;
2291
2292 BUG_ON(!from->head_frag && !hlen);
2293
2294 /* dont bother with small payloads */
2295 if (len <= skb_tailroom(to))
2296 return skb_copy_bits(from, 0, skb_put(to, len), len);
2297
2298 if (hlen) {
2299 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2300 if (unlikely(ret))
2301 return ret;
2302 len -= hlen;
2303 } else {
2304 plen = min_t(int, skb_headlen(from), len);
2305 if (plen) {
2306 page = virt_to_head_page(from->head);
2307 offset = from->data - (unsigned char *)page_address(page);
2308 __skb_fill_page_desc(to, 0, page, offset, plen);
2309 get_page(page);
2310 j = 1;
2311 len -= plen;
2312 }
2313 }
2314
2315 to->truesize += len + plen;
2316 to->len += len + plen;
2317 to->data_len += len + plen;
2318
2319 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2320 skb_tx_error(from);
2321 return -ENOMEM;
2322 }
2323
2324 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2325 if (!len)
2326 break;
2327 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2328 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2329 len -= skb_shinfo(to)->frags[j].size;
2330 skb_frag_ref(to, j);
2331 j++;
2332 }
2333 skb_shinfo(to)->nr_frags = j;
2334
2335 return 0;
2336}
2337EXPORT_SYMBOL_GPL(skb_zerocopy);
2338
2339void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2340{
2341 __wsum csum;
2342 long csstart;
2343
2344 if (skb->ip_summed == CHECKSUM_PARTIAL)
2345 csstart = skb_checksum_start_offset(skb);
2346 else
2347 csstart = skb_headlen(skb);
2348
2349 BUG_ON(csstart > skb_headlen(skb));
2350
2351 skb_copy_from_linear_data(skb, to, csstart);
2352
2353 csum = 0;
2354 if (csstart != skb->len)
2355 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2356 skb->len - csstart, 0);
2357
2358 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2359 long csstuff = csstart + skb->csum_offset;
2360
2361 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2362 }
2363}
2364EXPORT_SYMBOL(skb_copy_and_csum_dev);
2365
2366/**
2367 * skb_dequeue - remove from the head of the queue
2368 * @list: list to dequeue from
2369 *
2370 * Remove the head of the list. The list lock is taken so the function
2371 * may be used safely with other locking list functions. The head item is
2372 * returned or %NULL if the list is empty.
2373 */
2374
2375struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2376{
2377 unsigned long flags;
2378 struct sk_buff *result;
2379
2380 spin_lock_irqsave(&list->lock, flags);
2381 result = __skb_dequeue(list);
2382 spin_unlock_irqrestore(&list->lock, flags);
2383 return result;
2384}
2385EXPORT_SYMBOL(skb_dequeue);
2386
2387/**
2388 * skb_dequeue_tail - remove from the tail of the queue
2389 * @list: list to dequeue from
2390 *
2391 * Remove the tail of the list. The list lock is taken so the function
2392 * may be used safely with other locking list functions. The tail item is
2393 * returned or %NULL if the list is empty.
2394 */
2395struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2396{
2397 unsigned long flags;
2398 struct sk_buff *result;
2399
2400 spin_lock_irqsave(&list->lock, flags);
2401 result = __skb_dequeue_tail(list);
2402 spin_unlock_irqrestore(&list->lock, flags);
2403 return result;
2404}
2405EXPORT_SYMBOL(skb_dequeue_tail);
2406
2407/**
2408 * skb_queue_purge - empty a list
2409 * @list: list to empty
2410 *
2411 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2412 * the list and one reference dropped. This function takes the list
2413 * lock and is atomic with respect to other list locking functions.
2414 */
2415void skb_queue_purge(struct sk_buff_head *list)
2416{
2417 struct sk_buff *skb;
2418 while ((skb = skb_dequeue(list)) != NULL)
2419 kfree_skb(skb);
2420}
2421EXPORT_SYMBOL(skb_queue_purge);
2422
2423/**
2424 * skb_rbtree_purge - empty a skb rbtree
2425 * @root: root of the rbtree to empty
2426 *
2427 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
2428 * the list and one reference dropped. This function does not take
2429 * any lock. Synchronization should be handled by the caller (e.g., TCP
2430 * out-of-order queue is protected by the socket lock).
2431 */
2432void skb_rbtree_purge(struct rb_root *root)
2433{
2434 struct sk_buff *skb, *next;
2435
2436 rbtree_postorder_for_each_entry_safe(skb, next, root, rbnode)
2437 kfree_skb(skb);
2438
2439 *root = RB_ROOT;
2440}
2441
2442/**
2443 * skb_queue_head - queue a buffer at the list head
2444 * @list: list to use
2445 * @newsk: buffer to queue
2446 *
2447 * Queue a buffer at the start of the list. This function takes the
2448 * list lock and can be used safely with other locking &sk_buff functions
2449 * safely.
2450 *
2451 * A buffer cannot be placed on two lists at the same time.
2452 */
2453void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2454{
2455 unsigned long flags;
2456
2457 spin_lock_irqsave(&list->lock, flags);
2458 __skb_queue_head(list, newsk);
2459 spin_unlock_irqrestore(&list->lock, flags);
2460}
2461EXPORT_SYMBOL(skb_queue_head);
2462
2463/**
2464 * skb_queue_tail - queue a buffer at the list tail
2465 * @list: list to use
2466 * @newsk: buffer to queue
2467 *
2468 * Queue a buffer at the tail of the list. This function takes the
2469 * list lock and can be used safely with other locking &sk_buff functions
2470 * safely.
2471 *
2472 * A buffer cannot be placed on two lists at the same time.
2473 */
2474void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2475{
2476 unsigned long flags;
2477
2478 spin_lock_irqsave(&list->lock, flags);
2479 __skb_queue_tail(list, newsk);
2480 spin_unlock_irqrestore(&list->lock, flags);
2481}
2482EXPORT_SYMBOL(skb_queue_tail);
2483
2484/**
2485 * skb_unlink - remove a buffer from a list
2486 * @skb: buffer to remove
2487 * @list: list to use
2488 *
2489 * Remove a packet from a list. The list locks are taken and this
2490 * function is atomic with respect to other list locked calls
2491 *
2492 * You must know what list the SKB is on.
2493 */
2494void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2495{
2496 unsigned long flags;
2497
2498 spin_lock_irqsave(&list->lock, flags);
2499 __skb_unlink(skb, list);
2500 spin_unlock_irqrestore(&list->lock, flags);
2501}
2502EXPORT_SYMBOL(skb_unlink);
2503
2504/**
2505 * skb_append - append a buffer
2506 * @old: buffer to insert after
2507 * @newsk: buffer to insert
2508 * @list: list to use
2509 *
2510 * Place a packet after a given packet in a list. The list locks are taken
2511 * and this function is atomic with respect to other list locked calls.
2512 * A buffer cannot be placed on two lists at the same time.
2513 */
2514void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2515{
2516 unsigned long flags;
2517
2518 spin_lock_irqsave(&list->lock, flags);
2519 __skb_queue_after(list, old, newsk);
2520 spin_unlock_irqrestore(&list->lock, flags);
2521}
2522EXPORT_SYMBOL(skb_append);
2523
2524/**
2525 * skb_insert - insert a buffer
2526 * @old: buffer to insert before
2527 * @newsk: buffer to insert
2528 * @list: list to use
2529 *
2530 * Place a packet before a given packet in a list. The list locks are
2531 * taken and this function is atomic with respect to other list locked
2532 * calls.
2533 *
2534 * A buffer cannot be placed on two lists at the same time.
2535 */
2536void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2537{
2538 unsigned long flags;
2539
2540 spin_lock_irqsave(&list->lock, flags);
2541 __skb_insert(newsk, old->prev, old, list);
2542 spin_unlock_irqrestore(&list->lock, flags);
2543}
2544EXPORT_SYMBOL(skb_insert);
2545
2546static inline void skb_split_inside_header(struct sk_buff *skb,
2547 struct sk_buff* skb1,
2548 const u32 len, const int pos)
2549{
2550 int i;
2551
2552 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2553 pos - len);
2554 /* And move data appendix as is. */
2555 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2556 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2557
2558 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2559 skb_shinfo(skb)->nr_frags = 0;
2560 skb1->data_len = skb->data_len;
2561 skb1->len += skb1->data_len;
2562 skb->data_len = 0;
2563 skb->len = len;
2564 skb_set_tail_pointer(skb, len);
2565}
2566
2567static inline void skb_split_no_header(struct sk_buff *skb,
2568 struct sk_buff* skb1,
2569 const u32 len, int pos)
2570{
2571 int i, k = 0;
2572 const int nfrags = skb_shinfo(skb)->nr_frags;
2573
2574 skb_shinfo(skb)->nr_frags = 0;
2575 skb1->len = skb1->data_len = skb->len - len;
2576 skb->len = len;
2577 skb->data_len = len - pos;
2578
2579 for (i = 0; i < nfrags; i++) {
2580 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2581
2582 if (pos + size > len) {
2583 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2584
2585 if (pos < len) {
2586 /* Split frag.
2587 * We have two variants in this case:
2588 * 1. Move all the frag to the second
2589 * part, if it is possible. F.e.
2590 * this approach is mandatory for TUX,
2591 * where splitting is expensive.
2592 * 2. Split is accurately. We make this.
2593 */
2594 skb_frag_ref(skb, i);
2595 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2596 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2597 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2598 skb_shinfo(skb)->nr_frags++;
2599 }
2600 k++;
2601 } else
2602 skb_shinfo(skb)->nr_frags++;
2603 pos += size;
2604 }
2605 skb_shinfo(skb1)->nr_frags = k;
2606}
2607
2608/**
2609 * skb_split - Split fragmented skb to two parts at length len.
2610 * @skb: the buffer to split
2611 * @skb1: the buffer to receive the second part
2612 * @len: new length for skb
2613 */
2614void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2615{
2616 int pos = skb_headlen(skb);
2617
2618 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2619 if (len < pos) /* Split line is inside header. */
2620 skb_split_inside_header(skb, skb1, len, pos);
2621 else /* Second chunk has no header, nothing to copy. */
2622 skb_split_no_header(skb, skb1, len, pos);
2623}
2624EXPORT_SYMBOL(skb_split);
2625
2626/* Shifting from/to a cloned skb is a no-go.
2627 *
2628 * Caller cannot keep skb_shinfo related pointers past calling here!
2629 */
2630static int skb_prepare_for_shift(struct sk_buff *skb)
2631{
2632 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2633}
2634
2635/**
2636 * skb_shift - Shifts paged data partially from skb to another
2637 * @tgt: buffer into which tail data gets added
2638 * @skb: buffer from which the paged data comes from
2639 * @shiftlen: shift up to this many bytes
2640 *
2641 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2642 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2643 * It's up to caller to free skb if everything was shifted.
2644 *
2645 * If @tgt runs out of frags, the whole operation is aborted.
2646 *
2647 * Skb cannot include anything else but paged data while tgt is allowed
2648 * to have non-paged data as well.
2649 *
2650 * TODO: full sized shift could be optimized but that would need
2651 * specialized skb free'er to handle frags without up-to-date nr_frags.
2652 */
2653int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2654{
2655 int from, to, merge, todo;
2656 struct skb_frag_struct *fragfrom, *fragto;
2657
2658 BUG_ON(shiftlen > skb->len);
2659
2660 if (skb_headlen(skb))
2661 return 0;
2662
2663 todo = shiftlen;
2664 from = 0;
2665 to = skb_shinfo(tgt)->nr_frags;
2666 fragfrom = &skb_shinfo(skb)->frags[from];
2667
2668 /* Actual merge is delayed until the point when we know we can
2669 * commit all, so that we don't have to undo partial changes
2670 */
2671 if (!to ||
2672 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2673 fragfrom->page_offset)) {
2674 merge = -1;
2675 } else {
2676 merge = to - 1;
2677
2678 todo -= skb_frag_size(fragfrom);
2679 if (todo < 0) {
2680 if (skb_prepare_for_shift(skb) ||
2681 skb_prepare_for_shift(tgt))
2682 return 0;
2683
2684 /* All previous frag pointers might be stale! */
2685 fragfrom = &skb_shinfo(skb)->frags[from];
2686 fragto = &skb_shinfo(tgt)->frags[merge];
2687
2688 skb_frag_size_add(fragto, shiftlen);
2689 skb_frag_size_sub(fragfrom, shiftlen);
2690 fragfrom->page_offset += shiftlen;
2691
2692 goto onlymerged;
2693 }
2694
2695 from++;
2696 }
2697
2698 /* Skip full, not-fitting skb to avoid expensive operations */
2699 if ((shiftlen == skb->len) &&
2700 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2701 return 0;
2702
2703 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2704 return 0;
2705
2706 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2707 if (to == MAX_SKB_FRAGS)
2708 return 0;
2709
2710 fragfrom = &skb_shinfo(skb)->frags[from];
2711 fragto = &skb_shinfo(tgt)->frags[to];
2712
2713 if (todo >= skb_frag_size(fragfrom)) {
2714 *fragto = *fragfrom;
2715 todo -= skb_frag_size(fragfrom);
2716 from++;
2717 to++;
2718
2719 } else {
2720 __skb_frag_ref(fragfrom);
2721 fragto->page = fragfrom->page;
2722 fragto->page_offset = fragfrom->page_offset;
2723 skb_frag_size_set(fragto, todo);
2724
2725 fragfrom->page_offset += todo;
2726 skb_frag_size_sub(fragfrom, todo);
2727 todo = 0;
2728
2729 to++;
2730 break;
2731 }
2732 }
2733
2734 /* Ready to "commit" this state change to tgt */
2735 skb_shinfo(tgt)->nr_frags = to;
2736
2737 if (merge >= 0) {
2738 fragfrom = &skb_shinfo(skb)->frags[0];
2739 fragto = &skb_shinfo(tgt)->frags[merge];
2740
2741 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2742 __skb_frag_unref(fragfrom);
2743 }
2744
2745 /* Reposition in the original skb */
2746 to = 0;
2747 while (from < skb_shinfo(skb)->nr_frags)
2748 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2749 skb_shinfo(skb)->nr_frags = to;
2750
2751 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2752
2753onlymerged:
2754 /* Most likely the tgt won't ever need its checksum anymore, skb on
2755 * the other hand might need it if it needs to be resent
2756 */
2757 tgt->ip_summed = CHECKSUM_PARTIAL;
2758 skb->ip_summed = CHECKSUM_PARTIAL;
2759
2760 /* Yak, is it really working this way? Some helper please? */
2761 skb->len -= shiftlen;
2762 skb->data_len -= shiftlen;
2763 skb->truesize -= shiftlen;
2764 tgt->len += shiftlen;
2765 tgt->data_len += shiftlen;
2766 tgt->truesize += shiftlen;
2767
2768 return shiftlen;
2769}
2770
2771/**
2772 * skb_prepare_seq_read - Prepare a sequential read of skb data
2773 * @skb: the buffer to read
2774 * @from: lower offset of data to be read
2775 * @to: upper offset of data to be read
2776 * @st: state variable
2777 *
2778 * Initializes the specified state variable. Must be called before
2779 * invoking skb_seq_read() for the first time.
2780 */
2781void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2782 unsigned int to, struct skb_seq_state *st)
2783{
2784 st->lower_offset = from;
2785 st->upper_offset = to;
2786 st->root_skb = st->cur_skb = skb;
2787 st->frag_idx = st->stepped_offset = 0;
2788 st->frag_data = NULL;
2789}
2790EXPORT_SYMBOL(skb_prepare_seq_read);
2791
2792/**
2793 * skb_seq_read - Sequentially read skb data
2794 * @consumed: number of bytes consumed by the caller so far
2795 * @data: destination pointer for data to be returned
2796 * @st: state variable
2797 *
2798 * Reads a block of skb data at @consumed relative to the
2799 * lower offset specified to skb_prepare_seq_read(). Assigns
2800 * the head of the data block to @data and returns the length
2801 * of the block or 0 if the end of the skb data or the upper
2802 * offset has been reached.
2803 *
2804 * The caller is not required to consume all of the data
2805 * returned, i.e. @consumed is typically set to the number
2806 * of bytes already consumed and the next call to
2807 * skb_seq_read() will return the remaining part of the block.
2808 *
2809 * Note 1: The size of each block of data returned can be arbitrary,
2810 * this limitation is the cost for zerocopy sequential
2811 * reads of potentially non linear data.
2812 *
2813 * Note 2: Fragment lists within fragments are not implemented
2814 * at the moment, state->root_skb could be replaced with
2815 * a stack for this purpose.
2816 */
2817unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2818 struct skb_seq_state *st)
2819{
2820 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2821 skb_frag_t *frag;
2822
2823 if (unlikely(abs_offset >= st->upper_offset)) {
2824 if (st->frag_data) {
2825 kunmap_atomic(st->frag_data);
2826 st->frag_data = NULL;
2827 }
2828 return 0;
2829 }
2830
2831next_skb:
2832 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2833
2834 if (abs_offset < block_limit && !st->frag_data) {
2835 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2836 return block_limit - abs_offset;
2837 }
2838
2839 if (st->frag_idx == 0 && !st->frag_data)
2840 st->stepped_offset += skb_headlen(st->cur_skb);
2841
2842 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2843 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2844 block_limit = skb_frag_size(frag) + st->stepped_offset;
2845
2846 if (abs_offset < block_limit) {
2847 if (!st->frag_data)
2848 st->frag_data = kmap_atomic(skb_frag_page(frag));
2849
2850 *data = (u8 *) st->frag_data + frag->page_offset +
2851 (abs_offset - st->stepped_offset);
2852
2853 return block_limit - abs_offset;
2854 }
2855
2856 if (st->frag_data) {
2857 kunmap_atomic(st->frag_data);
2858 st->frag_data = NULL;
2859 }
2860
2861 st->frag_idx++;
2862 st->stepped_offset += skb_frag_size(frag);
2863 }
2864
2865 if (st->frag_data) {
2866 kunmap_atomic(st->frag_data);
2867 st->frag_data = NULL;
2868 }
2869
2870 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2871 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2872 st->frag_idx = 0;
2873 goto next_skb;
2874 } else if (st->cur_skb->next) {
2875 st->cur_skb = st->cur_skb->next;
2876 st->frag_idx = 0;
2877 goto next_skb;
2878 }
2879
2880 return 0;
2881}
2882EXPORT_SYMBOL(skb_seq_read);
2883
2884/**
2885 * skb_abort_seq_read - Abort a sequential read of skb data
2886 * @st: state variable
2887 *
2888 * Must be called if skb_seq_read() was not called until it
2889 * returned 0.
2890 */
2891void skb_abort_seq_read(struct skb_seq_state *st)
2892{
2893 if (st->frag_data)
2894 kunmap_atomic(st->frag_data);
2895}
2896EXPORT_SYMBOL(skb_abort_seq_read);
2897
2898#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2899
2900static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2901 struct ts_config *conf,
2902 struct ts_state *state)
2903{
2904 return skb_seq_read(offset, text, TS_SKB_CB(state));
2905}
2906
2907static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2908{
2909 skb_abort_seq_read(TS_SKB_CB(state));
2910}
2911
2912/**
2913 * skb_find_text - Find a text pattern in skb data
2914 * @skb: the buffer to look in
2915 * @from: search offset
2916 * @to: search limit
2917 * @config: textsearch configuration
2918 *
2919 * Finds a pattern in the skb data according to the specified
2920 * textsearch configuration. Use textsearch_next() to retrieve
2921 * subsequent occurrences of the pattern. Returns the offset
2922 * to the first occurrence or UINT_MAX if no match was found.
2923 */
2924unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2925 unsigned int to, struct ts_config *config)
2926{
2927 struct ts_state state;
2928 unsigned int ret;
2929
2930 config->get_next_block = skb_ts_get_next_block;
2931 config->finish = skb_ts_finish;
2932
2933 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2934
2935 ret = textsearch_find(config, &state);
2936 return (ret <= to - from ? ret : UINT_MAX);
2937}
2938EXPORT_SYMBOL(skb_find_text);
2939
2940/**
2941 * skb_append_datato_frags - append the user data to a skb
2942 * @sk: sock structure
2943 * @skb: skb structure to be appended with user data.
2944 * @getfrag: call back function to be used for getting the user data
2945 * @from: pointer to user message iov
2946 * @length: length of the iov message
2947 *
2948 * Description: This procedure append the user data in the fragment part
2949 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2950 */
2951int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2952 int (*getfrag)(void *from, char *to, int offset,
2953 int len, int odd, struct sk_buff *skb),
2954 void *from, int length)
2955{
2956 int frg_cnt = skb_shinfo(skb)->nr_frags;
2957 int copy;
2958 int offset = 0;
2959 int ret;
2960 struct page_frag *pfrag = ¤t->task_frag;
2961
2962 do {
2963 /* Return error if we don't have space for new frag */
2964 if (frg_cnt >= MAX_SKB_FRAGS)
2965 return -EMSGSIZE;
2966
2967 if (!sk_page_frag_refill(sk, pfrag))
2968 return -ENOMEM;
2969
2970 /* copy the user data to page */
2971 copy = min_t(int, length, pfrag->size - pfrag->offset);
2972
2973 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2974 offset, copy, 0, skb);
2975 if (ret < 0)
2976 return -EFAULT;
2977
2978 /* copy was successful so update the size parameters */
2979 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2980 copy);
2981 frg_cnt++;
2982 pfrag->offset += copy;
2983 get_page(pfrag->page);
2984
2985 skb->truesize += copy;
2986 atomic_add(copy, &sk->sk_wmem_alloc);
2987 skb->len += copy;
2988 skb->data_len += copy;
2989 offset += copy;
2990 length -= copy;
2991
2992 } while (length > 0);
2993
2994 return 0;
2995}
2996EXPORT_SYMBOL(skb_append_datato_frags);
2997
2998int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
2999 int offset, size_t size)
3000{
3001 int i = skb_shinfo(skb)->nr_frags;
3002
3003 if (skb_can_coalesce(skb, i, page, offset)) {
3004 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3005 } else if (i < MAX_SKB_FRAGS) {
3006 get_page(page);
3007 skb_fill_page_desc(skb, i, page, offset, size);
3008 } else {
3009 return -EMSGSIZE;
3010 }
3011
3012 return 0;
3013}
3014EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3015
3016/**
3017 * skb_pull_rcsum - pull skb and update receive checksum
3018 * @skb: buffer to update
3019 * @len: length of data pulled
3020 *
3021 * This function performs an skb_pull on the packet and updates
3022 * the CHECKSUM_COMPLETE checksum. It should be used on
3023 * receive path processing instead of skb_pull unless you know
3024 * that the checksum difference is zero (e.g., a valid IP header)
3025 * or you are setting ip_summed to CHECKSUM_NONE.
3026 */
3027unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3028{
3029 unsigned char *data = skb->data;
3030
3031 BUG_ON(len > skb->len);
3032 __skb_pull(skb, len);
3033 skb_postpull_rcsum(skb, data, len);
3034 return skb->data;
3035}
3036EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3037
3038/**
3039 * skb_segment - Perform protocol segmentation on skb.
3040 * @head_skb: buffer to segment
3041 * @features: features for the output path (see dev->features)
3042 *
3043 * This function performs segmentation on the given skb. It returns
3044 * a pointer to the first in a list of new skbs for the segments.
3045 * In case of error it returns ERR_PTR(err).
3046 */
3047struct sk_buff *skb_segment(struct sk_buff *head_skb,
3048 netdev_features_t features)
3049{
3050 struct sk_buff *segs = NULL;
3051 struct sk_buff *tail = NULL;
3052 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3053 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3054 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3055 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3056 struct sk_buff *frag_skb = head_skb;
3057 unsigned int offset = doffset;
3058 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3059 unsigned int partial_segs = 0;
3060 unsigned int headroom;
3061 unsigned int len = head_skb->len;
3062 __be16 proto;
3063 bool csum, sg;
3064 int nfrags = skb_shinfo(head_skb)->nr_frags;
3065 int err = -ENOMEM;
3066 int i = 0;
3067 int pos;
3068 int dummy;
3069
3070 __skb_push(head_skb, doffset);
3071 proto = skb_network_protocol(head_skb, &dummy);
3072 if (unlikely(!proto))
3073 return ERR_PTR(-EINVAL);
3074
3075 sg = !!(features & NETIF_F_SG);
3076 csum = !!can_checksum_protocol(features, proto);
3077
3078 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3079 if (!(features & NETIF_F_GSO_PARTIAL)) {
3080 struct sk_buff *iter;
3081
3082 if (!list_skb ||
3083 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3084 goto normal;
3085
3086 /* Split the buffer at the frag_list pointer.
3087 * This is based on the assumption that all
3088 * buffers in the chain excluding the last
3089 * containing the same amount of data.
3090 */
3091 skb_walk_frags(head_skb, iter) {
3092 if (skb_headlen(iter))
3093 goto normal;
3094
3095 len -= iter->len;
3096 }
3097 }
3098
3099 /* GSO partial only requires that we trim off any excess that
3100 * doesn't fit into an MSS sized block, so take care of that
3101 * now.
3102 */
3103 partial_segs = len / mss;
3104 if (partial_segs > 1)
3105 mss *= partial_segs;
3106 else
3107 partial_segs = 0;
3108 }
3109
3110normal:
3111 headroom = skb_headroom(head_skb);
3112 pos = skb_headlen(head_skb);
3113
3114 do {
3115 struct sk_buff *nskb;
3116 skb_frag_t *nskb_frag;
3117 int hsize;
3118 int size;
3119
3120 if (unlikely(mss == GSO_BY_FRAGS)) {
3121 len = list_skb->len;
3122 } else {
3123 len = head_skb->len - offset;
3124 if (len > mss)
3125 len = mss;
3126 }
3127
3128 hsize = skb_headlen(head_skb) - offset;
3129 if (hsize < 0)
3130 hsize = 0;
3131 if (hsize > len || !sg)
3132 hsize = len;
3133
3134 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3135 (skb_headlen(list_skb) == len || sg)) {
3136 BUG_ON(skb_headlen(list_skb) > len);
3137
3138 i = 0;
3139 nfrags = skb_shinfo(list_skb)->nr_frags;
3140 frag = skb_shinfo(list_skb)->frags;
3141 frag_skb = list_skb;
3142 pos += skb_headlen(list_skb);
3143
3144 while (pos < offset + len) {
3145 BUG_ON(i >= nfrags);
3146
3147 size = skb_frag_size(frag);
3148 if (pos + size > offset + len)
3149 break;
3150
3151 i++;
3152 pos += size;
3153 frag++;
3154 }
3155
3156 nskb = skb_clone(list_skb, GFP_ATOMIC);
3157 list_skb = list_skb->next;
3158
3159 if (unlikely(!nskb))
3160 goto err;
3161
3162 if (unlikely(pskb_trim(nskb, len))) {
3163 kfree_skb(nskb);
3164 goto err;
3165 }
3166
3167 hsize = skb_end_offset(nskb);
3168 if (skb_cow_head(nskb, doffset + headroom)) {
3169 kfree_skb(nskb);
3170 goto err;
3171 }
3172
3173 nskb->truesize += skb_end_offset(nskb) - hsize;
3174 skb_release_head_state(nskb);
3175 __skb_push(nskb, doffset);
3176 } else {
3177 nskb = __alloc_skb(hsize + doffset + headroom,
3178 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3179 NUMA_NO_NODE);
3180
3181 if (unlikely(!nskb))
3182 goto err;
3183
3184 skb_reserve(nskb, headroom);
3185 __skb_put(nskb, doffset);
3186 }
3187
3188 if (segs)
3189 tail->next = nskb;
3190 else
3191 segs = nskb;
3192 tail = nskb;
3193
3194 __copy_skb_header(nskb, head_skb);
3195
3196 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3197 skb_reset_mac_len(nskb);
3198
3199 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3200 nskb->data - tnl_hlen,
3201 doffset + tnl_hlen);
3202
3203 if (nskb->len == len + doffset)
3204 goto perform_csum_check;
3205
3206 if (!sg) {
3207 if (!nskb->remcsum_offload)
3208 nskb->ip_summed = CHECKSUM_NONE;
3209 SKB_GSO_CB(nskb)->csum =
3210 skb_copy_and_csum_bits(head_skb, offset,
3211 skb_put(nskb, len),
3212 len, 0);
3213 SKB_GSO_CB(nskb)->csum_start =
3214 skb_headroom(nskb) + doffset;
3215 continue;
3216 }
3217
3218 nskb_frag = skb_shinfo(nskb)->frags;
3219
3220 skb_copy_from_linear_data_offset(head_skb, offset,
3221 skb_put(nskb, hsize), hsize);
3222
3223 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3224 SKBTX_SHARED_FRAG;
3225
3226 while (pos < offset + len) {
3227 if (i >= nfrags) {
3228 BUG_ON(skb_headlen(list_skb));
3229
3230 i = 0;
3231 nfrags = skb_shinfo(list_skb)->nr_frags;
3232 frag = skb_shinfo(list_skb)->frags;
3233 frag_skb = list_skb;
3234
3235 BUG_ON(!nfrags);
3236
3237 list_skb = list_skb->next;
3238 }
3239
3240 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3241 MAX_SKB_FRAGS)) {
3242 net_warn_ratelimited(
3243 "skb_segment: too many frags: %u %u\n",
3244 pos, mss);
3245 goto err;
3246 }
3247
3248 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3249 goto err;
3250
3251 *nskb_frag = *frag;
3252 __skb_frag_ref(nskb_frag);
3253 size = skb_frag_size(nskb_frag);
3254
3255 if (pos < offset) {
3256 nskb_frag->page_offset += offset - pos;
3257 skb_frag_size_sub(nskb_frag, offset - pos);
3258 }
3259
3260 skb_shinfo(nskb)->nr_frags++;
3261
3262 if (pos + size <= offset + len) {
3263 i++;
3264 frag++;
3265 pos += size;
3266 } else {
3267 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3268 goto skip_fraglist;
3269 }
3270
3271 nskb_frag++;
3272 }
3273
3274skip_fraglist:
3275 nskb->data_len = len - hsize;
3276 nskb->len += nskb->data_len;
3277 nskb->truesize += nskb->data_len;
3278
3279perform_csum_check:
3280 if (!csum) {
3281 if (skb_has_shared_frag(nskb)) {
3282 err = __skb_linearize(nskb);
3283 if (err)
3284 goto err;
3285 }
3286 if (!nskb->remcsum_offload)
3287 nskb->ip_summed = CHECKSUM_NONE;
3288 SKB_GSO_CB(nskb)->csum =
3289 skb_checksum(nskb, doffset,
3290 nskb->len - doffset, 0);
3291 SKB_GSO_CB(nskb)->csum_start =
3292 skb_headroom(nskb) + doffset;
3293 }
3294 } while ((offset += len) < head_skb->len);
3295
3296 /* Some callers want to get the end of the list.
3297 * Put it in segs->prev to avoid walking the list.
3298 * (see validate_xmit_skb_list() for example)
3299 */
3300 segs->prev = tail;
3301
3302 if (partial_segs) {
3303 struct sk_buff *iter;
3304 int type = skb_shinfo(head_skb)->gso_type;
3305 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
3306
3307 /* Update type to add partial and then remove dodgy if set */
3308 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
3309 type &= ~SKB_GSO_DODGY;
3310
3311 /* Update GSO info and prepare to start updating headers on
3312 * our way back down the stack of protocols.
3313 */
3314 for (iter = segs; iter; iter = iter->next) {
3315 skb_shinfo(iter)->gso_size = gso_size;
3316 skb_shinfo(iter)->gso_segs = partial_segs;
3317 skb_shinfo(iter)->gso_type = type;
3318 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
3319 }
3320
3321 if (tail->len - doffset <= gso_size)
3322 skb_shinfo(tail)->gso_size = 0;
3323 else if (tail != segs)
3324 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
3325 }
3326
3327 /* Following permits correct backpressure, for protocols
3328 * using skb_set_owner_w().
3329 * Idea is to tranfert ownership from head_skb to last segment.
3330 */
3331 if (head_skb->destructor == sock_wfree) {
3332 swap(tail->truesize, head_skb->truesize);
3333 swap(tail->destructor, head_skb->destructor);
3334 swap(tail->sk, head_skb->sk);
3335 }
3336 return segs;
3337
3338err:
3339 kfree_skb_list(segs);
3340 return ERR_PTR(err);
3341}
3342EXPORT_SYMBOL_GPL(skb_segment);
3343
3344int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3345{
3346 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3347 unsigned int offset = skb_gro_offset(skb);
3348 unsigned int headlen = skb_headlen(skb);
3349 unsigned int len = skb_gro_len(skb);
3350 struct sk_buff *lp, *p = *head;
3351 unsigned int delta_truesize;
3352
3353 if (unlikely(p->len + len >= 65536))
3354 return -E2BIG;
3355
3356 lp = NAPI_GRO_CB(p)->last;
3357 pinfo = skb_shinfo(lp);
3358
3359 if (headlen <= offset) {
3360 skb_frag_t *frag;
3361 skb_frag_t *frag2;
3362 int i = skbinfo->nr_frags;
3363 int nr_frags = pinfo->nr_frags + i;
3364
3365 if (nr_frags > MAX_SKB_FRAGS)
3366 goto merge;
3367
3368 offset -= headlen;
3369 pinfo->nr_frags = nr_frags;
3370 skbinfo->nr_frags = 0;
3371
3372 frag = pinfo->frags + nr_frags;
3373 frag2 = skbinfo->frags + i;
3374 do {
3375 *--frag = *--frag2;
3376 } while (--i);
3377
3378 frag->page_offset += offset;
3379 skb_frag_size_sub(frag, offset);
3380
3381 /* all fragments truesize : remove (head size + sk_buff) */
3382 delta_truesize = skb->truesize -
3383 SKB_TRUESIZE(skb_end_offset(skb));
3384
3385 skb->truesize -= skb->data_len;
3386 skb->len -= skb->data_len;
3387 skb->data_len = 0;
3388
3389 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3390 goto done;
3391 } else if (skb->head_frag) {
3392 int nr_frags = pinfo->nr_frags;
3393 skb_frag_t *frag = pinfo->frags + nr_frags;
3394 struct page *page = virt_to_head_page(skb->head);
3395 unsigned int first_size = headlen - offset;
3396 unsigned int first_offset;
3397
3398 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3399 goto merge;
3400
3401 first_offset = skb->data -
3402 (unsigned char *)page_address(page) +
3403 offset;
3404
3405 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3406
3407 frag->page.p = page;
3408 frag->page_offset = first_offset;
3409 skb_frag_size_set(frag, first_size);
3410
3411 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3412 /* We dont need to clear skbinfo->nr_frags here */
3413
3414 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3415 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3416 goto done;
3417 }
3418
3419merge:
3420 delta_truesize = skb->truesize;
3421 if (offset > headlen) {
3422 unsigned int eat = offset - headlen;
3423
3424 skbinfo->frags[0].page_offset += eat;
3425 skb_frag_size_sub(&skbinfo->frags[0], eat);
3426 skb->data_len -= eat;
3427 skb->len -= eat;
3428 offset = headlen;
3429 }
3430
3431 __skb_pull(skb, offset);
3432
3433 if (NAPI_GRO_CB(p)->last == p)
3434 skb_shinfo(p)->frag_list = skb;
3435 else
3436 NAPI_GRO_CB(p)->last->next = skb;
3437 NAPI_GRO_CB(p)->last = skb;
3438 __skb_header_release(skb);
3439 lp = p;
3440
3441done:
3442 NAPI_GRO_CB(p)->count++;
3443 p->data_len += len;
3444 p->truesize += delta_truesize;
3445 p->len += len;
3446 if (lp != p) {
3447 lp->data_len += len;
3448 lp->truesize += delta_truesize;
3449 lp->len += len;
3450 }
3451 NAPI_GRO_CB(skb)->same_flow = 1;
3452 return 0;
3453}
3454EXPORT_SYMBOL_GPL(skb_gro_receive);
3455
3456void __init skb_init(void)
3457{
3458 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3459 sizeof(struct sk_buff),
3460 0,
3461 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3462 NULL);
3463 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3464 sizeof(struct sk_buff_fclones),
3465 0,
3466 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3467 NULL);
3468}
3469
3470/**
3471 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3472 * @skb: Socket buffer containing the buffers to be mapped
3473 * @sg: The scatter-gather list to map into
3474 * @offset: The offset into the buffer's contents to start mapping
3475 * @len: Length of buffer space to be mapped
3476 *
3477 * Fill the specified scatter-gather list with mappings/pointers into a
3478 * region of the buffer space attached to a socket buffer.
3479 */
3480static int
3481__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3482{
3483 int start = skb_headlen(skb);
3484 int i, copy = start - offset;
3485 struct sk_buff *frag_iter;
3486 int elt = 0;
3487
3488 if (copy > 0) {
3489 if (copy > len)
3490 copy = len;
3491 sg_set_buf(sg, skb->data + offset, copy);
3492 elt++;
3493 if ((len -= copy) == 0)
3494 return elt;
3495 offset += copy;
3496 }
3497
3498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3499 int end;
3500
3501 WARN_ON(start > offset + len);
3502
3503 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3504 if ((copy = end - offset) > 0) {
3505 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3506
3507 if (copy > len)
3508 copy = len;
3509 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3510 frag->page_offset+offset-start);
3511 elt++;
3512 if (!(len -= copy))
3513 return elt;
3514 offset += copy;
3515 }
3516 start = end;
3517 }
3518
3519 skb_walk_frags(skb, frag_iter) {
3520 int end;
3521
3522 WARN_ON(start > offset + len);
3523
3524 end = start + frag_iter->len;
3525 if ((copy = end - offset) > 0) {
3526 if (copy > len)
3527 copy = len;
3528 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3529 copy);
3530 if ((len -= copy) == 0)
3531 return elt;
3532 offset += copy;
3533 }
3534 start = end;
3535 }
3536 BUG_ON(len);
3537 return elt;
3538}
3539
3540/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3541 * sglist without mark the sg which contain last skb data as the end.
3542 * So the caller can mannipulate sg list as will when padding new data after
3543 * the first call without calling sg_unmark_end to expend sg list.
3544 *
3545 * Scenario to use skb_to_sgvec_nomark:
3546 * 1. sg_init_table
3547 * 2. skb_to_sgvec_nomark(payload1)
3548 * 3. skb_to_sgvec_nomark(payload2)
3549 *
3550 * This is equivalent to:
3551 * 1. sg_init_table
3552 * 2. skb_to_sgvec(payload1)
3553 * 3. sg_unmark_end
3554 * 4. skb_to_sgvec(payload2)
3555 *
3556 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3557 * is more preferable.
3558 */
3559int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3560 int offset, int len)
3561{
3562 return __skb_to_sgvec(skb, sg, offset, len);
3563}
3564EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3565
3566int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3567{
3568 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3569
3570 sg_mark_end(&sg[nsg - 1]);
3571
3572 return nsg;
3573}
3574EXPORT_SYMBOL_GPL(skb_to_sgvec);
3575
3576/**
3577 * skb_cow_data - Check that a socket buffer's data buffers are writable
3578 * @skb: The socket buffer to check.
3579 * @tailbits: Amount of trailing space to be added
3580 * @trailer: Returned pointer to the skb where the @tailbits space begins
3581 *
3582 * Make sure that the data buffers attached to a socket buffer are
3583 * writable. If they are not, private copies are made of the data buffers
3584 * and the socket buffer is set to use these instead.
3585 *
3586 * If @tailbits is given, make sure that there is space to write @tailbits
3587 * bytes of data beyond current end of socket buffer. @trailer will be
3588 * set to point to the skb in which this space begins.
3589 *
3590 * The number of scatterlist elements required to completely map the
3591 * COW'd and extended socket buffer will be returned.
3592 */
3593int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3594{
3595 int copyflag;
3596 int elt;
3597 struct sk_buff *skb1, **skb_p;
3598
3599 /* If skb is cloned or its head is paged, reallocate
3600 * head pulling out all the pages (pages are considered not writable
3601 * at the moment even if they are anonymous).
3602 */
3603 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3604 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3605 return -ENOMEM;
3606
3607 /* Easy case. Most of packets will go this way. */
3608 if (!skb_has_frag_list(skb)) {
3609 /* A little of trouble, not enough of space for trailer.
3610 * This should not happen, when stack is tuned to generate
3611 * good frames. OK, on miss we reallocate and reserve even more
3612 * space, 128 bytes is fair. */
3613
3614 if (skb_tailroom(skb) < tailbits &&
3615 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3616 return -ENOMEM;
3617
3618 /* Voila! */
3619 *trailer = skb;
3620 return 1;
3621 }
3622
3623 /* Misery. We are in troubles, going to mincer fragments... */
3624
3625 elt = 1;
3626 skb_p = &skb_shinfo(skb)->frag_list;
3627 copyflag = 0;
3628
3629 while ((skb1 = *skb_p) != NULL) {
3630 int ntail = 0;
3631
3632 /* The fragment is partially pulled by someone,
3633 * this can happen on input. Copy it and everything
3634 * after it. */
3635
3636 if (skb_shared(skb1))
3637 copyflag = 1;
3638
3639 /* If the skb is the last, worry about trailer. */
3640
3641 if (skb1->next == NULL && tailbits) {
3642 if (skb_shinfo(skb1)->nr_frags ||
3643 skb_has_frag_list(skb1) ||
3644 skb_tailroom(skb1) < tailbits)
3645 ntail = tailbits + 128;
3646 }
3647
3648 if (copyflag ||
3649 skb_cloned(skb1) ||
3650 ntail ||
3651 skb_shinfo(skb1)->nr_frags ||
3652 skb_has_frag_list(skb1)) {
3653 struct sk_buff *skb2;
3654
3655 /* Fuck, we are miserable poor guys... */
3656 if (ntail == 0)
3657 skb2 = skb_copy(skb1, GFP_ATOMIC);
3658 else
3659 skb2 = skb_copy_expand(skb1,
3660 skb_headroom(skb1),
3661 ntail,
3662 GFP_ATOMIC);
3663 if (unlikely(skb2 == NULL))
3664 return -ENOMEM;
3665
3666 if (skb1->sk)
3667 skb_set_owner_w(skb2, skb1->sk);
3668
3669 /* Looking around. Are we still alive?
3670 * OK, link new skb, drop old one */
3671
3672 skb2->next = skb1->next;
3673 *skb_p = skb2;
3674 kfree_skb(skb1);
3675 skb1 = skb2;
3676 }
3677 elt++;
3678 *trailer = skb1;
3679 skb_p = &skb1->next;
3680 }
3681
3682 return elt;
3683}
3684EXPORT_SYMBOL_GPL(skb_cow_data);
3685
3686static void sock_rmem_free(struct sk_buff *skb)
3687{
3688 struct sock *sk = skb->sk;
3689
3690 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3691}
3692
3693/*
3694 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3695 */
3696int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3697{
3698 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3699 (unsigned int)sk->sk_rcvbuf)
3700 return -ENOMEM;
3701
3702 skb_orphan(skb);
3703 skb->sk = sk;
3704 skb->destructor = sock_rmem_free;
3705 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3706
3707 /* before exiting rcu section, make sure dst is refcounted */
3708 skb_dst_force(skb);
3709
3710 skb_queue_tail(&sk->sk_error_queue, skb);
3711 if (!sock_flag(sk, SOCK_DEAD))
3712 sk->sk_data_ready(sk);
3713 return 0;
3714}
3715EXPORT_SYMBOL(sock_queue_err_skb);
3716
3717static bool is_icmp_err_skb(const struct sk_buff *skb)
3718{
3719 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
3720 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
3721}
3722
3723struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3724{
3725 struct sk_buff_head *q = &sk->sk_error_queue;
3726 struct sk_buff *skb, *skb_next = NULL;
3727 bool icmp_next = false;
3728 unsigned long flags;
3729
3730 spin_lock_irqsave(&q->lock, flags);
3731 skb = __skb_dequeue(q);
3732 if (skb && (skb_next = skb_peek(q)))
3733 icmp_next = is_icmp_err_skb(skb_next);
3734 spin_unlock_irqrestore(&q->lock, flags);
3735
3736 if (is_icmp_err_skb(skb) && !icmp_next)
3737 sk->sk_err = 0;
3738
3739 if (skb_next)
3740 sk->sk_error_report(sk);
3741
3742 return skb;
3743}
3744EXPORT_SYMBOL(sock_dequeue_err_skb);
3745
3746/**
3747 * skb_clone_sk - create clone of skb, and take reference to socket
3748 * @skb: the skb to clone
3749 *
3750 * This function creates a clone of a buffer that holds a reference on
3751 * sk_refcnt. Buffers created via this function are meant to be
3752 * returned using sock_queue_err_skb, or free via kfree_skb.
3753 *
3754 * When passing buffers allocated with this function to sock_queue_err_skb
3755 * it is necessary to wrap the call with sock_hold/sock_put in order to
3756 * prevent the socket from being released prior to being enqueued on
3757 * the sk_error_queue.
3758 */
3759struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3760{
3761 struct sock *sk = skb->sk;
3762 struct sk_buff *clone;
3763
3764 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3765 return NULL;
3766
3767 clone = skb_clone(skb, GFP_ATOMIC);
3768 if (!clone) {
3769 sock_put(sk);
3770 return NULL;
3771 }
3772
3773 clone->sk = sk;
3774 clone->destructor = sock_efree;
3775
3776 return clone;
3777}
3778EXPORT_SYMBOL(skb_clone_sk);
3779
3780static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3781 struct sock *sk,
3782 int tstype)
3783{
3784 struct sock_exterr_skb *serr;
3785 int err;
3786
3787 serr = SKB_EXT_ERR(skb);
3788 memset(serr, 0, sizeof(*serr));
3789 serr->ee.ee_errno = ENOMSG;
3790 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3791 serr->ee.ee_info = tstype;
3792 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3793 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3794 if (sk->sk_protocol == IPPROTO_TCP &&
3795 sk->sk_type == SOCK_STREAM)
3796 serr->ee.ee_data -= sk->sk_tskey;
3797 }
3798
3799 err = sock_queue_err_skb(sk, skb);
3800
3801 if (err)
3802 kfree_skb(skb);
3803}
3804
3805static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3806{
3807 bool ret;
3808
3809 if (likely(sysctl_tstamp_allow_data || tsonly))
3810 return true;
3811
3812 read_lock_bh(&sk->sk_callback_lock);
3813 ret = sk->sk_socket && sk->sk_socket->file &&
3814 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3815 read_unlock_bh(&sk->sk_callback_lock);
3816 return ret;
3817}
3818
3819void skb_complete_tx_timestamp(struct sk_buff *skb,
3820 struct skb_shared_hwtstamps *hwtstamps)
3821{
3822 struct sock *sk = skb->sk;
3823
3824 if (!skb_may_tx_timestamp(sk, false))
3825 return;
3826
3827 /* Take a reference to prevent skb_orphan() from freeing the socket,
3828 * but only if the socket refcount is not zero.
3829 */
3830 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3831 *skb_hwtstamps(skb) = *hwtstamps;
3832 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3833 sock_put(sk);
3834 }
3835}
3836EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3837
3838void __skb_tstamp_tx(struct sk_buff *orig_skb,
3839 struct skb_shared_hwtstamps *hwtstamps,
3840 struct sock *sk, int tstype)
3841{
3842 struct sk_buff *skb;
3843 bool tsonly;
3844
3845 if (!sk)
3846 return;
3847
3848 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3849 if (!skb_may_tx_timestamp(sk, tsonly))
3850 return;
3851
3852 if (tsonly) {
3853#ifdef CONFIG_INET
3854 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
3855 sk->sk_protocol == IPPROTO_TCP &&
3856 sk->sk_type == SOCK_STREAM)
3857 skb = tcp_get_timestamping_opt_stats(sk);
3858 else
3859#endif
3860 skb = alloc_skb(0, GFP_ATOMIC);
3861 } else {
3862 skb = skb_clone(orig_skb, GFP_ATOMIC);
3863 }
3864 if (!skb)
3865 return;
3866
3867 if (tsonly) {
3868 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3869 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3870 }
3871
3872 if (hwtstamps)
3873 *skb_hwtstamps(skb) = *hwtstamps;
3874 else
3875 skb->tstamp = ktime_get_real();
3876
3877 __skb_complete_tx_timestamp(skb, sk, tstype);
3878}
3879EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3880
3881void skb_tstamp_tx(struct sk_buff *orig_skb,
3882 struct skb_shared_hwtstamps *hwtstamps)
3883{
3884 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3885 SCM_TSTAMP_SND);
3886}
3887EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3888
3889void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3890{
3891 struct sock *sk = skb->sk;
3892 struct sock_exterr_skb *serr;
3893 int err = 1;
3894
3895 skb->wifi_acked_valid = 1;
3896 skb->wifi_acked = acked;
3897
3898 serr = SKB_EXT_ERR(skb);
3899 memset(serr, 0, sizeof(*serr));
3900 serr->ee.ee_errno = ENOMSG;
3901 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3902
3903 /* Take a reference to prevent skb_orphan() from freeing the socket,
3904 * but only if the socket refcount is not zero.
3905 */
3906 if (likely(atomic_inc_not_zero(&sk->sk_refcnt))) {
3907 err = sock_queue_err_skb(sk, skb);
3908 sock_put(sk);
3909 }
3910 if (err)
3911 kfree_skb(skb);
3912}
3913EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3914
3915/**
3916 * skb_partial_csum_set - set up and verify partial csum values for packet
3917 * @skb: the skb to set
3918 * @start: the number of bytes after skb->data to start checksumming.
3919 * @off: the offset from start to place the checksum.
3920 *
3921 * For untrusted partially-checksummed packets, we need to make sure the values
3922 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3923 *
3924 * This function checks and sets those values and skb->ip_summed: if this
3925 * returns false you should drop the packet.
3926 */
3927bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3928{
3929 if (unlikely(start > skb_headlen(skb)) ||
3930 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3931 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3932 start, off, skb_headlen(skb));
3933 return false;
3934 }
3935 skb->ip_summed = CHECKSUM_PARTIAL;
3936 skb->csum_start = skb_headroom(skb) + start;
3937 skb->csum_offset = off;
3938 skb_set_transport_header(skb, start);
3939 return true;
3940}
3941EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3942
3943static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3944 unsigned int max)
3945{
3946 if (skb_headlen(skb) >= len)
3947 return 0;
3948
3949 /* If we need to pullup then pullup to the max, so we
3950 * won't need to do it again.
3951 */
3952 if (max > skb->len)
3953 max = skb->len;
3954
3955 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3956 return -ENOMEM;
3957
3958 if (skb_headlen(skb) < len)
3959 return -EPROTO;
3960
3961 return 0;
3962}
3963
3964#define MAX_TCP_HDR_LEN (15 * 4)
3965
3966static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3967 typeof(IPPROTO_IP) proto,
3968 unsigned int off)
3969{
3970 switch (proto) {
3971 int err;
3972
3973 case IPPROTO_TCP:
3974 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3975 off + MAX_TCP_HDR_LEN);
3976 if (!err && !skb_partial_csum_set(skb, off,
3977 offsetof(struct tcphdr,
3978 check)))
3979 err = -EPROTO;
3980 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3981
3982 case IPPROTO_UDP:
3983 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3984 off + sizeof(struct udphdr));
3985 if (!err && !skb_partial_csum_set(skb, off,
3986 offsetof(struct udphdr,
3987 check)))
3988 err = -EPROTO;
3989 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3990 }
3991
3992 return ERR_PTR(-EPROTO);
3993}
3994
3995/* This value should be large enough to cover a tagged ethernet header plus
3996 * maximally sized IP and TCP or UDP headers.
3997 */
3998#define MAX_IP_HDR_LEN 128
3999
4000static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4001{
4002 unsigned int off;
4003 bool fragment;
4004 __sum16 *csum;
4005 int err;
4006
4007 fragment = false;
4008
4009 err = skb_maybe_pull_tail(skb,
4010 sizeof(struct iphdr),
4011 MAX_IP_HDR_LEN);
4012 if (err < 0)
4013 goto out;
4014
4015 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4016 fragment = true;
4017
4018 off = ip_hdrlen(skb);
4019
4020 err = -EPROTO;
4021
4022 if (fragment)
4023 goto out;
4024
4025 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4026 if (IS_ERR(csum))
4027 return PTR_ERR(csum);
4028
4029 if (recalculate)
4030 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4031 ip_hdr(skb)->daddr,
4032 skb->len - off,
4033 ip_hdr(skb)->protocol, 0);
4034 err = 0;
4035
4036out:
4037 return err;
4038}
4039
4040/* This value should be large enough to cover a tagged ethernet header plus
4041 * an IPv6 header, all options, and a maximal TCP or UDP header.
4042 */
4043#define MAX_IPV6_HDR_LEN 256
4044
4045#define OPT_HDR(type, skb, off) \
4046 (type *)(skb_network_header(skb) + (off))
4047
4048static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4049{
4050 int err;
4051 u8 nexthdr;
4052 unsigned int off;
4053 unsigned int len;
4054 bool fragment;
4055 bool done;
4056 __sum16 *csum;
4057
4058 fragment = false;
4059 done = false;
4060
4061 off = sizeof(struct ipv6hdr);
4062
4063 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4064 if (err < 0)
4065 goto out;
4066
4067 nexthdr = ipv6_hdr(skb)->nexthdr;
4068
4069 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4070 while (off <= len && !done) {
4071 switch (nexthdr) {
4072 case IPPROTO_DSTOPTS:
4073 case IPPROTO_HOPOPTS:
4074 case IPPROTO_ROUTING: {
4075 struct ipv6_opt_hdr *hp;
4076
4077 err = skb_maybe_pull_tail(skb,
4078 off +
4079 sizeof(struct ipv6_opt_hdr),
4080 MAX_IPV6_HDR_LEN);
4081 if (err < 0)
4082 goto out;
4083
4084 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4085 nexthdr = hp->nexthdr;
4086 off += ipv6_optlen(hp);
4087 break;
4088 }
4089 case IPPROTO_AH: {
4090 struct ip_auth_hdr *hp;
4091
4092 err = skb_maybe_pull_tail(skb,
4093 off +
4094 sizeof(struct ip_auth_hdr),
4095 MAX_IPV6_HDR_LEN);
4096 if (err < 0)
4097 goto out;
4098
4099 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4100 nexthdr = hp->nexthdr;
4101 off += ipv6_authlen(hp);
4102 break;
4103 }
4104 case IPPROTO_FRAGMENT: {
4105 struct frag_hdr *hp;
4106
4107 err = skb_maybe_pull_tail(skb,
4108 off +
4109 sizeof(struct frag_hdr),
4110 MAX_IPV6_HDR_LEN);
4111 if (err < 0)
4112 goto out;
4113
4114 hp = OPT_HDR(struct frag_hdr, skb, off);
4115
4116 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4117 fragment = true;
4118
4119 nexthdr = hp->nexthdr;
4120 off += sizeof(struct frag_hdr);
4121 break;
4122 }
4123 default:
4124 done = true;
4125 break;
4126 }
4127 }
4128
4129 err = -EPROTO;
4130
4131 if (!done || fragment)
4132 goto out;
4133
4134 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4135 if (IS_ERR(csum))
4136 return PTR_ERR(csum);
4137
4138 if (recalculate)
4139 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4140 &ipv6_hdr(skb)->daddr,
4141 skb->len - off, nexthdr, 0);
4142 err = 0;
4143
4144out:
4145 return err;
4146}
4147
4148/**
4149 * skb_checksum_setup - set up partial checksum offset
4150 * @skb: the skb to set up
4151 * @recalculate: if true the pseudo-header checksum will be recalculated
4152 */
4153int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4154{
4155 int err;
4156
4157 switch (skb->protocol) {
4158 case htons(ETH_P_IP):
4159 err = skb_checksum_setup_ipv4(skb, recalculate);
4160 break;
4161
4162 case htons(ETH_P_IPV6):
4163 err = skb_checksum_setup_ipv6(skb, recalculate);
4164 break;
4165
4166 default:
4167 err = -EPROTO;
4168 break;
4169 }
4170
4171 return err;
4172}
4173EXPORT_SYMBOL(skb_checksum_setup);
4174
4175/**
4176 * skb_checksum_maybe_trim - maybe trims the given skb
4177 * @skb: the skb to check
4178 * @transport_len: the data length beyond the network header
4179 *
4180 * Checks whether the given skb has data beyond the given transport length.
4181 * If so, returns a cloned skb trimmed to this transport length.
4182 * Otherwise returns the provided skb. Returns NULL in error cases
4183 * (e.g. transport_len exceeds skb length or out-of-memory).
4184 *
4185 * Caller needs to set the skb transport header and free any returned skb if it
4186 * differs from the provided skb.
4187 */
4188static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4189 unsigned int transport_len)
4190{
4191 struct sk_buff *skb_chk;
4192 unsigned int len = skb_transport_offset(skb) + transport_len;
4193 int ret;
4194
4195 if (skb->len < len)
4196 return NULL;
4197 else if (skb->len == len)
4198 return skb;
4199
4200 skb_chk = skb_clone(skb, GFP_ATOMIC);
4201 if (!skb_chk)
4202 return NULL;
4203
4204 ret = pskb_trim_rcsum(skb_chk, len);
4205 if (ret) {
4206 kfree_skb(skb_chk);
4207 return NULL;
4208 }
4209
4210 return skb_chk;
4211}
4212
4213/**
4214 * skb_checksum_trimmed - validate checksum of an skb
4215 * @skb: the skb to check
4216 * @transport_len: the data length beyond the network header
4217 * @skb_chkf: checksum function to use
4218 *
4219 * Applies the given checksum function skb_chkf to the provided skb.
4220 * Returns a checked and maybe trimmed skb. Returns NULL on error.
4221 *
4222 * If the skb has data beyond the given transport length, then a
4223 * trimmed & cloned skb is checked and returned.
4224 *
4225 * Caller needs to set the skb transport header and free any returned skb if it
4226 * differs from the provided skb.
4227 */
4228struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4229 unsigned int transport_len,
4230 __sum16(*skb_chkf)(struct sk_buff *skb))
4231{
4232 struct sk_buff *skb_chk;
4233 unsigned int offset = skb_transport_offset(skb);
4234 __sum16 ret;
4235
4236 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4237 if (!skb_chk)
4238 goto err;
4239
4240 if (!pskb_may_pull(skb_chk, offset))
4241 goto err;
4242
4243 skb_pull_rcsum(skb_chk, offset);
4244 ret = skb_chkf(skb_chk);
4245 skb_push_rcsum(skb_chk, offset);
4246
4247 if (ret)
4248 goto err;
4249
4250 return skb_chk;
4251
4252err:
4253 if (skb_chk && skb_chk != skb)
4254 kfree_skb(skb_chk);
4255
4256 return NULL;
4257
4258}
4259EXPORT_SYMBOL(skb_checksum_trimmed);
4260
4261void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4262{
4263 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4264 skb->dev->name);
4265}
4266EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4267
4268void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4269{
4270 if (head_stolen) {
4271 skb_release_head_state(skb);
4272 kmem_cache_free(skbuff_head_cache, skb);
4273 } else {
4274 __kfree_skb(skb);
4275 }
4276}
4277EXPORT_SYMBOL(kfree_skb_partial);
4278
4279/**
4280 * skb_try_coalesce - try to merge skb to prior one
4281 * @to: prior buffer
4282 * @from: buffer to add
4283 * @fragstolen: pointer to boolean
4284 * @delta_truesize: how much more was allocated than was requested
4285 */
4286bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4287 bool *fragstolen, int *delta_truesize)
4288{
4289 int i, delta, len = from->len;
4290
4291 *fragstolen = false;
4292
4293 if (skb_cloned(to))
4294 return false;
4295
4296 if (len <= skb_tailroom(to)) {
4297 if (len)
4298 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4299 *delta_truesize = 0;
4300 return true;
4301 }
4302
4303 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4304 return false;
4305
4306 if (skb_headlen(from) != 0) {
4307 struct page *page;
4308 unsigned int offset;
4309
4310 if (skb_shinfo(to)->nr_frags +
4311 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4312 return false;
4313
4314 if (skb_head_is_locked(from))
4315 return false;
4316
4317 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4318
4319 page = virt_to_head_page(from->head);
4320 offset = from->data - (unsigned char *)page_address(page);
4321
4322 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4323 page, offset, skb_headlen(from));
4324 *fragstolen = true;
4325 } else {
4326 if (skb_shinfo(to)->nr_frags +
4327 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4328 return false;
4329
4330 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4331 }
4332
4333 WARN_ON_ONCE(delta < len);
4334
4335 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4336 skb_shinfo(from)->frags,
4337 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4338 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4339
4340 if (!skb_cloned(from))
4341 skb_shinfo(from)->nr_frags = 0;
4342
4343 /* if the skb is not cloned this does nothing
4344 * since we set nr_frags to 0.
4345 */
4346 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4347 skb_frag_ref(from, i);
4348
4349 to->truesize += delta;
4350 to->len += len;
4351 to->data_len += len;
4352
4353 *delta_truesize = delta;
4354 return true;
4355}
4356EXPORT_SYMBOL(skb_try_coalesce);
4357
4358/**
4359 * skb_scrub_packet - scrub an skb
4360 *
4361 * @skb: buffer to clean
4362 * @xnet: packet is crossing netns
4363 *
4364 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
4365 * into/from a tunnel. Some information have to be cleared during these
4366 * operations.
4367 * skb_scrub_packet can also be used to clean a skb before injecting it in
4368 * another namespace (@xnet == true). We have to clear all information in the
4369 * skb that could impact namespace isolation.
4370 */
4371void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4372{
4373 skb->tstamp = 0;
4374 skb->pkt_type = PACKET_HOST;
4375 skb->skb_iif = 0;
4376 skb->ignore_df = 0;
4377 skb_dst_drop(skb);
4378 secpath_reset(skb);
4379 nf_reset(skb);
4380 nf_reset_trace(skb);
4381
4382 if (!xnet)
4383 return;
4384
4385 skb_orphan(skb);
4386 skb->mark = 0;
4387}
4388EXPORT_SYMBOL_GPL(skb_scrub_packet);
4389
4390/**
4391 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
4392 *
4393 * @skb: GSO skb
4394 *
4395 * skb_gso_transport_seglen is used to determine the real size of the
4396 * individual segments, including Layer4 headers (TCP/UDP).
4397 *
4398 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
4399 */
4400unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4401{
4402 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4403 unsigned int thlen = 0;
4404
4405 if (skb->encapsulation) {
4406 thlen = skb_inner_transport_header(skb) -
4407 skb_transport_header(skb);
4408
4409 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4410 thlen += inner_tcp_hdrlen(skb);
4411 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4412 thlen = tcp_hdrlen(skb);
4413 } else if (unlikely(shinfo->gso_type & SKB_GSO_SCTP)) {
4414 thlen = sizeof(struct sctphdr);
4415 }
4416 /* UFO sets gso_size to the size of the fragmentation
4417 * payload, i.e. the size of the L4 (UDP) header is already
4418 * accounted for.
4419 */
4420 return thlen + shinfo->gso_size;
4421}
4422EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4423
4424/**
4425 * skb_gso_validate_mtu - Return in case such skb fits a given MTU
4426 *
4427 * @skb: GSO skb
4428 * @mtu: MTU to validate against
4429 *
4430 * skb_gso_validate_mtu validates if a given skb will fit a wanted MTU
4431 * once split.
4432 */
4433bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu)
4434{
4435 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4436 const struct sk_buff *iter;
4437 unsigned int hlen;
4438
4439 hlen = skb_gso_network_seglen(skb);
4440
4441 if (shinfo->gso_size != GSO_BY_FRAGS)
4442 return hlen <= mtu;
4443
4444 /* Undo this so we can re-use header sizes */
4445 hlen -= GSO_BY_FRAGS;
4446
4447 skb_walk_frags(skb, iter) {
4448 if (hlen + skb_headlen(iter) > mtu)
4449 return false;
4450 }
4451
4452 return true;
4453}
4454EXPORT_SYMBOL_GPL(skb_gso_validate_mtu);
4455
4456static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4457{
4458 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4459 kfree_skb(skb);
4460 return NULL;
4461 }
4462
4463 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4464 2 * ETH_ALEN);
4465 skb->mac_header += VLAN_HLEN;
4466 return skb;
4467}
4468
4469struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4470{
4471 struct vlan_hdr *vhdr;
4472 u16 vlan_tci;
4473
4474 if (unlikely(skb_vlan_tag_present(skb))) {
4475 /* vlan_tci is already set-up so leave this for another time */
4476 return skb;
4477 }
4478
4479 skb = skb_share_check(skb, GFP_ATOMIC);
4480 if (unlikely(!skb))
4481 goto err_free;
4482
4483 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4484 goto err_free;
4485
4486 vhdr = (struct vlan_hdr *)skb->data;
4487 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4488 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4489
4490 skb_pull_rcsum(skb, VLAN_HLEN);
4491 vlan_set_encap_proto(skb, vhdr);
4492
4493 skb = skb_reorder_vlan_header(skb);
4494 if (unlikely(!skb))
4495 goto err_free;
4496
4497 skb_reset_network_header(skb);
4498 skb_reset_transport_header(skb);
4499 skb_reset_mac_len(skb);
4500
4501 return skb;
4502
4503err_free:
4504 kfree_skb(skb);
4505 return NULL;
4506}
4507EXPORT_SYMBOL(skb_vlan_untag);
4508
4509int skb_ensure_writable(struct sk_buff *skb, int write_len)
4510{
4511 if (!pskb_may_pull(skb, write_len))
4512 return -ENOMEM;
4513
4514 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4515 return 0;
4516
4517 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4518}
4519EXPORT_SYMBOL(skb_ensure_writable);
4520
4521/* remove VLAN header from packet and update csum accordingly.
4522 * expects a non skb_vlan_tag_present skb with a vlan tag payload
4523 */
4524int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4525{
4526 struct vlan_hdr *vhdr;
4527 int offset = skb->data - skb_mac_header(skb);
4528 int err;
4529
4530 if (WARN_ONCE(offset,
4531 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
4532 offset)) {
4533 return -EINVAL;
4534 }
4535
4536 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4537 if (unlikely(err))
4538 return err;
4539
4540 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4541
4542 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4543 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4544
4545 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4546 __skb_pull(skb, VLAN_HLEN);
4547
4548 vlan_set_encap_proto(skb, vhdr);
4549 skb->mac_header += VLAN_HLEN;
4550
4551 if (skb_network_offset(skb) < ETH_HLEN)
4552 skb_set_network_header(skb, ETH_HLEN);
4553
4554 skb_reset_mac_len(skb);
4555
4556 return err;
4557}
4558EXPORT_SYMBOL(__skb_vlan_pop);
4559
4560/* Pop a vlan tag either from hwaccel or from payload.
4561 * Expects skb->data at mac header.
4562 */
4563int skb_vlan_pop(struct sk_buff *skb)
4564{
4565 u16 vlan_tci;
4566 __be16 vlan_proto;
4567 int err;
4568
4569 if (likely(skb_vlan_tag_present(skb))) {
4570 skb->vlan_tci = 0;
4571 } else {
4572 if (unlikely(!eth_type_vlan(skb->protocol)))
4573 return 0;
4574
4575 err = __skb_vlan_pop(skb, &vlan_tci);
4576 if (err)
4577 return err;
4578 }
4579 /* move next vlan tag to hw accel tag */
4580 if (likely(!eth_type_vlan(skb->protocol)))
4581 return 0;
4582
4583 vlan_proto = skb->protocol;
4584 err = __skb_vlan_pop(skb, &vlan_tci);
4585 if (unlikely(err))
4586 return err;
4587
4588 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4589 return 0;
4590}
4591EXPORT_SYMBOL(skb_vlan_pop);
4592
4593/* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
4594 * Expects skb->data at mac header.
4595 */
4596int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4597{
4598 if (skb_vlan_tag_present(skb)) {
4599 int offset = skb->data - skb_mac_header(skb);
4600 int err;
4601
4602 if (WARN_ONCE(offset,
4603 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
4604 offset)) {
4605 return -EINVAL;
4606 }
4607
4608 err = __vlan_insert_tag(skb, skb->vlan_proto,
4609 skb_vlan_tag_get(skb));
4610 if (err)
4611 return err;
4612
4613 skb->protocol = skb->vlan_proto;
4614 skb->mac_len += VLAN_HLEN;
4615
4616 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4617 }
4618 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4619 return 0;
4620}
4621EXPORT_SYMBOL(skb_vlan_push);
4622
4623/**
4624 * alloc_skb_with_frags - allocate skb with page frags
4625 *
4626 * @header_len: size of linear part
4627 * @data_len: needed length in frags
4628 * @max_page_order: max page order desired.
4629 * @errcode: pointer to error code if any
4630 * @gfp_mask: allocation mask
4631 *
4632 * This can be used to allocate a paged skb, given a maximal order for frags.
4633 */
4634struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4635 unsigned long data_len,
4636 int max_page_order,
4637 int *errcode,
4638 gfp_t gfp_mask)
4639{
4640 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4641 unsigned long chunk;
4642 struct sk_buff *skb;
4643 struct page *page;
4644 gfp_t gfp_head;
4645 int i;
4646
4647 *errcode = -EMSGSIZE;
4648 /* Note this test could be relaxed, if we succeed to allocate
4649 * high order pages...
4650 */
4651 if (npages > MAX_SKB_FRAGS)
4652 return NULL;
4653
4654 gfp_head = gfp_mask;
4655 if (gfp_head & __GFP_DIRECT_RECLAIM)
4656 gfp_head |= __GFP_REPEAT;
4657
4658 *errcode = -ENOBUFS;
4659 skb = alloc_skb(header_len, gfp_head);
4660 if (!skb)
4661 return NULL;
4662
4663 skb->truesize += npages << PAGE_SHIFT;
4664
4665 for (i = 0; npages > 0; i++) {
4666 int order = max_page_order;
4667
4668 while (order) {
4669 if (npages >= 1 << order) {
4670 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4671 __GFP_COMP |
4672 __GFP_NOWARN |
4673 __GFP_NORETRY,
4674 order);
4675 if (page)
4676 goto fill_page;
4677 /* Do not retry other high order allocations */
4678 order = 1;
4679 max_page_order = 0;
4680 }
4681 order--;
4682 }
4683 page = alloc_page(gfp_mask);
4684 if (!page)
4685 goto failure;
4686fill_page:
4687 chunk = min_t(unsigned long, data_len,
4688 PAGE_SIZE << order);
4689 skb_fill_page_desc(skb, i, page, 0, chunk);
4690 data_len -= chunk;
4691 npages -= 1 << order;
4692 }
4693 return skb;
4694
4695failure:
4696 kfree_skb(skb);
4697 return NULL;
4698}
4699EXPORT_SYMBOL(alloc_skb_with_frags);
4700
4701/* carve out the first off bytes from skb when off < headlen */
4702static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4703 const int headlen, gfp_t gfp_mask)
4704{
4705 int i;
4706 int size = skb_end_offset(skb);
4707 int new_hlen = headlen - off;
4708 u8 *data;
4709
4710 size = SKB_DATA_ALIGN(size);
4711
4712 if (skb_pfmemalloc(skb))
4713 gfp_mask |= __GFP_MEMALLOC;
4714 data = kmalloc_reserve(size +
4715 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4716 gfp_mask, NUMA_NO_NODE, NULL);
4717 if (!data)
4718 return -ENOMEM;
4719
4720 size = SKB_WITH_OVERHEAD(ksize(data));
4721
4722 /* Copy real data, and all frags */
4723 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4724 skb->len -= off;
4725
4726 memcpy((struct skb_shared_info *)(data + size),
4727 skb_shinfo(skb),
4728 offsetof(struct skb_shared_info,
4729 frags[skb_shinfo(skb)->nr_frags]));
4730 if (skb_cloned(skb)) {
4731 /* drop the old head gracefully */
4732 if (skb_orphan_frags(skb, gfp_mask)) {
4733 kfree(data);
4734 return -ENOMEM;
4735 }
4736 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4737 skb_frag_ref(skb, i);
4738 if (skb_has_frag_list(skb))
4739 skb_clone_fraglist(skb);
4740 skb_release_data(skb);
4741 } else {
4742 /* we can reuse existing recount- all we did was
4743 * relocate values
4744 */
4745 skb_free_head(skb);
4746 }
4747
4748 skb->head = data;
4749 skb->data = data;
4750 skb->head_frag = 0;
4751#ifdef NET_SKBUFF_DATA_USES_OFFSET
4752 skb->end = size;
4753#else
4754 skb->end = skb->head + size;
4755#endif
4756 skb_set_tail_pointer(skb, skb_headlen(skb));
4757 skb_headers_offset_update(skb, 0);
4758 skb->cloned = 0;
4759 skb->hdr_len = 0;
4760 skb->nohdr = 0;
4761 atomic_set(&skb_shinfo(skb)->dataref, 1);
4762
4763 return 0;
4764}
4765
4766static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4767
4768/* carve out the first eat bytes from skb's frag_list. May recurse into
4769 * pskb_carve()
4770 */
4771static int pskb_carve_frag_list(struct sk_buff *skb,
4772 struct skb_shared_info *shinfo, int eat,
4773 gfp_t gfp_mask)
4774{
4775 struct sk_buff *list = shinfo->frag_list;
4776 struct sk_buff *clone = NULL;
4777 struct sk_buff *insp = NULL;
4778
4779 do {
4780 if (!list) {
4781 pr_err("Not enough bytes to eat. Want %d\n", eat);
4782 return -EFAULT;
4783 }
4784 if (list->len <= eat) {
4785 /* Eaten as whole. */
4786 eat -= list->len;
4787 list = list->next;
4788 insp = list;
4789 } else {
4790 /* Eaten partially. */
4791 if (skb_shared(list)) {
4792 clone = skb_clone(list, gfp_mask);
4793 if (!clone)
4794 return -ENOMEM;
4795 insp = list->next;
4796 list = clone;
4797 } else {
4798 /* This may be pulled without problems. */
4799 insp = list;
4800 }
4801 if (pskb_carve(list, eat, gfp_mask) < 0) {
4802 kfree_skb(clone);
4803 return -ENOMEM;
4804 }
4805 break;
4806 }
4807 } while (eat);
4808
4809 /* Free pulled out fragments. */
4810 while ((list = shinfo->frag_list) != insp) {
4811 shinfo->frag_list = list->next;
4812 kfree_skb(list);
4813 }
4814 /* And insert new clone at head. */
4815 if (clone) {
4816 clone->next = list;
4817 shinfo->frag_list = clone;
4818 }
4819 return 0;
4820}
4821
4822/* carve off first len bytes from skb. Split line (off) is in the
4823 * non-linear part of skb
4824 */
4825static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4826 int pos, gfp_t gfp_mask)
4827{
4828 int i, k = 0;
4829 int size = skb_end_offset(skb);
4830 u8 *data;
4831 const int nfrags = skb_shinfo(skb)->nr_frags;
4832 struct skb_shared_info *shinfo;
4833
4834 size = SKB_DATA_ALIGN(size);
4835
4836 if (skb_pfmemalloc(skb))
4837 gfp_mask |= __GFP_MEMALLOC;
4838 data = kmalloc_reserve(size +
4839 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4840 gfp_mask, NUMA_NO_NODE, NULL);
4841 if (!data)
4842 return -ENOMEM;
4843
4844 size = SKB_WITH_OVERHEAD(ksize(data));
4845
4846 memcpy((struct skb_shared_info *)(data + size),
4847 skb_shinfo(skb), offsetof(struct skb_shared_info,
4848 frags[skb_shinfo(skb)->nr_frags]));
4849 if (skb_orphan_frags(skb, gfp_mask)) {
4850 kfree(data);
4851 return -ENOMEM;
4852 }
4853 shinfo = (struct skb_shared_info *)(data + size);
4854 for (i = 0; i < nfrags; i++) {
4855 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4856
4857 if (pos + fsize > off) {
4858 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4859
4860 if (pos < off) {
4861 /* Split frag.
4862 * We have two variants in this case:
4863 * 1. Move all the frag to the second
4864 * part, if it is possible. F.e.
4865 * this approach is mandatory for TUX,
4866 * where splitting is expensive.
4867 * 2. Split is accurately. We make this.
4868 */
4869 shinfo->frags[0].page_offset += off - pos;
4870 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4871 }
4872 skb_frag_ref(skb, i);
4873 k++;
4874 }
4875 pos += fsize;
4876 }
4877 shinfo->nr_frags = k;
4878 if (skb_has_frag_list(skb))
4879 skb_clone_fraglist(skb);
4880
4881 if (k == 0) {
4882 /* split line is in frag list */
4883 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4884 }
4885 skb_release_data(skb);
4886
4887 skb->head = data;
4888 skb->head_frag = 0;
4889 skb->data = data;
4890#ifdef NET_SKBUFF_DATA_USES_OFFSET
4891 skb->end = size;
4892#else
4893 skb->end = skb->head + size;
4894#endif
4895 skb_reset_tail_pointer(skb);
4896 skb_headers_offset_update(skb, 0);
4897 skb->cloned = 0;
4898 skb->hdr_len = 0;
4899 skb->nohdr = 0;
4900 skb->len -= off;
4901 skb->data_len = skb->len;
4902 atomic_set(&skb_shinfo(skb)->dataref, 1);
4903 return 0;
4904}
4905
4906/* remove len bytes from the beginning of the skb */
4907static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4908{
4909 int headlen = skb_headlen(skb);
4910
4911 if (len < headlen)
4912 return pskb_carve_inside_header(skb, len, headlen, gfp);
4913 else
4914 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4915}
4916
4917/* Extract to_copy bytes starting at off from skb, and return this in
4918 * a new skb
4919 */
4920struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4921 int to_copy, gfp_t gfp)
4922{
4923 struct sk_buff *clone = skb_clone(skb, gfp);
4924
4925 if (!clone)
4926 return NULL;
4927
4928 if (pskb_carve(clone, off, gfp) < 0 ||
4929 pskb_trim(clone, to_copy)) {
4930 kfree_skb(clone);
4931 return NULL;
4932 }
4933 return clone;
4934}
4935EXPORT_SYMBOL(pskb_extract);
4936
4937/**
4938 * skb_condense - try to get rid of fragments/frag_list if possible
4939 * @skb: buffer
4940 *
4941 * Can be used to save memory before skb is added to a busy queue.
4942 * If packet has bytes in frags and enough tail room in skb->head,
4943 * pull all of them, so that we can free the frags right now and adjust
4944 * truesize.
4945 * Notes:
4946 * We do not reallocate skb->head thus can not fail.
4947 * Caller must re-evaluate skb->truesize if needed.
4948 */
4949void skb_condense(struct sk_buff *skb)
4950{
4951 if (skb->data_len) {
4952 if (skb->data_len > skb->end - skb->tail ||
4953 skb_cloned(skb))
4954 return;
4955
4956 /* Nice, we can free page frag(s) right now */
4957 __pskb_pull_tail(skb, skb->data_len);
4958 }
4959 /* At this point, skb->truesize might be over estimated,
4960 * because skb had a fragment, and fragments do not tell
4961 * their truesize.
4962 * When we pulled its content into skb->head, fragment
4963 * was freed, but __pskb_pull_tail() could not possibly
4964 * adjust skb->truesize, not knowing the frag truesize.
4965 */
4966 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
4967}