xref: /linux/net/core/skbuff.c (revision 8ab79ed50cf10f338465c296012500de1081646f)
1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3  *	Routines having to do with the 'struct sk_buff' memory handlers.
4  *
5  *	Authors:	Alan Cox <alan@lxorguk.ukuu.org.uk>
6  *			Florian La Roche <rzsfl@rz.uni-sb.de>
7  *
8  *	Fixes:
9  *		Alan Cox	:	Fixed the worst of the load
10  *					balancer bugs.
11  *		Dave Platt	:	Interrupt stacking fix.
12  *	Richard Kooijman	:	Timestamp fixes.
13  *		Alan Cox	:	Changed buffer format.
14  *		Alan Cox	:	destructor hook for AF_UNIX etc.
15  *		Linus Torvalds	:	Better skb_clone.
16  *		Alan Cox	:	Added skb_copy.
17  *		Alan Cox	:	Added all the changed routines Linus
18  *					only put in the headers
19  *		Ray VanTassle	:	Fixed --skb->lock in free
20  *		Alan Cox	:	skb_copy copy arp field
21  *		Andi Kleen	:	slabified it.
22  *		Robert Olsson	:	Removed skb_head_pool
23  *
24  *	NOTE:
25  *		The __skb_ routines should be called with interrupts
26  *	disabled, or you better be *real* sure that the operation is atomic
27  *	with respect to whatever list is being frobbed (e.g. via lock_sock()
28  *	or via disabling bottom half handlers, etc).
29  */
30 
31 /*
32  *	The functions in this file will not compile correctly with gcc 2.4.x
33  */
34 
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36 
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/in.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
51 #endif
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/skbuff_ref.h>
55 #include <linux/splice.h>
56 #include <linux/cache.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/init.h>
59 #include <linux/scatterlist.h>
60 #include <linux/errqueue.h>
61 #include <linux/prefetch.h>
62 #include <linux/bitfield.h>
63 #include <linux/if_vlan.h>
64 #include <linux/mpls.h>
65 #include <linux/kcov.h>
66 #include <linux/iov_iter.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/gso.h>
73 #include <net/hotdata.h>
74 #include <net/ip6_checksum.h>
75 #include <net/xfrm.h>
76 #include <net/mpls.h>
77 #include <net/mptcp.h>
78 #include <net/mctp.h>
79 #include <net/page_pool/helpers.h>
80 #include <net/dropreason.h>
81 
82 #include <linux/uaccess.h>
83 #include <trace/events/skb.h>
84 #include <linux/highmem.h>
85 #include <linux/capability.h>
86 #include <linux/user_namespace.h>
87 #include <linux/indirect_call_wrapper.h>
88 #include <linux/textsearch.h>
89 
90 #include "dev.h"
91 #include "netmem_priv.h"
92 #include "sock_destructor.h"
93 
94 #ifdef CONFIG_SKB_EXTENSIONS
95 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
96 #endif
97 
98 #define SKB_SMALL_HEAD_SIZE SKB_HEAD_ALIGN(MAX_TCP_HEADER)
99 
100 /* We want SKB_SMALL_HEAD_CACHE_SIZE to not be a power of two.
101  * This should ensure that SKB_SMALL_HEAD_HEADROOM is a unique
102  * size, and we can differentiate heads from skb_small_head_cache
103  * vs system slabs by looking at their size (skb_end_offset()).
104  */
105 #define SKB_SMALL_HEAD_CACHE_SIZE					\
106 	(is_power_of_2(SKB_SMALL_HEAD_SIZE) ?			\
107 		(SKB_SMALL_HEAD_SIZE + L1_CACHE_BYTES) :	\
108 		SKB_SMALL_HEAD_SIZE)
109 
110 #define SKB_SMALL_HEAD_HEADROOM						\
111 	SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE)
112 
113 /* kcm_write_msgs() relies on casting paged frags to bio_vec to use
114  * iov_iter_bvec(). These static asserts ensure the cast is valid is long as the
115  * netmem is a page.
116  */
117 static_assert(offsetof(struct bio_vec, bv_page) ==
118 	      offsetof(skb_frag_t, netmem));
119 static_assert(sizeof_field(struct bio_vec, bv_page) ==
120 	      sizeof_field(skb_frag_t, netmem));
121 
122 static_assert(offsetof(struct bio_vec, bv_len) == offsetof(skb_frag_t, len));
123 static_assert(sizeof_field(struct bio_vec, bv_len) ==
124 	      sizeof_field(skb_frag_t, len));
125 
126 static_assert(offsetof(struct bio_vec, bv_offset) ==
127 	      offsetof(skb_frag_t, offset));
128 static_assert(sizeof_field(struct bio_vec, bv_offset) ==
129 	      sizeof_field(skb_frag_t, offset));
130 
131 #undef FN
132 #define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
133 static const char * const drop_reasons[] = {
134 	[SKB_CONSUMED] = "CONSUMED",
135 	DEFINE_DROP_REASON(FN, FN)
136 };
137 
138 static const struct drop_reason_list drop_reasons_core = {
139 	.reasons = drop_reasons,
140 	.n_reasons = ARRAY_SIZE(drop_reasons),
141 };
142 
143 const struct drop_reason_list __rcu *
144 drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_NUM] = {
145 	[SKB_DROP_REASON_SUBSYS_CORE] = RCU_INITIALIZER(&drop_reasons_core),
146 };
147 EXPORT_SYMBOL(drop_reasons_by_subsys);
148 
149 /**
150  * drop_reasons_register_subsys - register another drop reason subsystem
151  * @subsys: the subsystem to register, must not be the core
152  * @list: the list of drop reasons within the subsystem, must point to
153  *	a statically initialized list
154  */
155 void drop_reasons_register_subsys(enum skb_drop_reason_subsys subsys,
156 				  const struct drop_reason_list *list)
157 {
158 	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
159 		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
160 		 "invalid subsystem %d\n", subsys))
161 		return;
162 
163 	/* must point to statically allocated memory, so INIT is OK */
164 	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], list);
165 }
166 EXPORT_SYMBOL_GPL(drop_reasons_register_subsys);
167 
168 /**
169  * drop_reasons_unregister_subsys - unregister a drop reason subsystem
170  * @subsys: the subsystem to remove, must not be the core
171  *
172  * Note: This will synchronize_rcu() to ensure no users when it returns.
173  */
174 void drop_reasons_unregister_subsys(enum skb_drop_reason_subsys subsys)
175 {
176 	if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
177 		 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
178 		 "invalid subsystem %d\n", subsys))
179 		return;
180 
181 	RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], NULL);
182 
183 	synchronize_rcu();
184 }
185 EXPORT_SYMBOL_GPL(drop_reasons_unregister_subsys);
186 
187 /**
188  *	skb_panic - private function for out-of-line support
189  *	@skb:	buffer
190  *	@sz:	size
191  *	@addr:	address
192  *	@msg:	skb_over_panic or skb_under_panic
193  *
194  *	Out-of-line support for skb_put() and skb_push().
195  *	Called via the wrapper skb_over_panic() or skb_under_panic().
196  *	Keep out of line to prevent kernel bloat.
197  *	__builtin_return_address is not used because it is not always reliable.
198  */
199 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
200 		      const char msg[])
201 {
202 	pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
203 		 msg, addr, skb->len, sz, skb->head, skb->data,
204 		 (unsigned long)skb->tail, (unsigned long)skb->end,
205 		 skb->dev ? skb->dev->name : "<NULL>");
206 	BUG();
207 }
208 
209 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
210 {
211 	skb_panic(skb, sz, addr, __func__);
212 }
213 
214 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
215 {
216 	skb_panic(skb, sz, addr, __func__);
217 }
218 
219 #define NAPI_SKB_CACHE_SIZE	64
220 #define NAPI_SKB_CACHE_BULK	16
221 #define NAPI_SKB_CACHE_HALF	(NAPI_SKB_CACHE_SIZE / 2)
222 
223 #if PAGE_SIZE == SZ_4K
224 
225 #define NAPI_HAS_SMALL_PAGE_FRAG	1
226 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	((nc).pfmemalloc)
227 
228 /* specialized page frag allocator using a single order 0 page
229  * and slicing it into 1K sized fragment. Constrained to systems
230  * with a very limited amount of 1K fragments fitting a single
231  * page - to avoid excessive truesize underestimation
232  */
233 
234 struct page_frag_1k {
235 	void *va;
236 	u16 offset;
237 	bool pfmemalloc;
238 };
239 
240 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
241 {
242 	struct page *page;
243 	int offset;
244 
245 	offset = nc->offset - SZ_1K;
246 	if (likely(offset >= 0))
247 		goto use_frag;
248 
249 	page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
250 	if (!page)
251 		return NULL;
252 
253 	nc->va = page_address(page);
254 	nc->pfmemalloc = page_is_pfmemalloc(page);
255 	offset = PAGE_SIZE - SZ_1K;
256 	page_ref_add(page, offset / SZ_1K);
257 
258 use_frag:
259 	nc->offset = offset;
260 	return nc->va + offset;
261 }
262 #else
263 
264 /* the small page is actually unused in this build; add dummy helpers
265  * to please the compiler and avoid later preprocessor's conditionals
266  */
267 #define NAPI_HAS_SMALL_PAGE_FRAG	0
268 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc)	false
269 
270 struct page_frag_1k {
271 };
272 
273 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
274 {
275 	return NULL;
276 }
277 
278 #endif
279 
280 struct napi_alloc_cache {
281 	local_lock_t bh_lock;
282 	struct page_frag_cache page;
283 	struct page_frag_1k page_small;
284 	unsigned int skb_count;
285 	void *skb_cache[NAPI_SKB_CACHE_SIZE];
286 };
287 
288 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
289 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = {
290 	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
291 };
292 
293 /* Double check that napi_get_frags() allocates skbs with
294  * skb->head being backed by slab, not a page fragment.
295  * This is to make sure bug fixed in 3226b158e67c
296  * ("net: avoid 32 x truesize under-estimation for tiny skbs")
297  * does not accidentally come back.
298  */
299 void napi_get_frags_check(struct napi_struct *napi)
300 {
301 	struct sk_buff *skb;
302 
303 	local_bh_disable();
304 	skb = napi_get_frags(napi);
305 	WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
306 	napi_free_frags(napi);
307 	local_bh_enable();
308 }
309 
310 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
311 {
312 	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
313 	void *data;
314 
315 	fragsz = SKB_DATA_ALIGN(fragsz);
316 
317 	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
318 	data = __page_frag_alloc_align(&nc->page, fragsz,
319 				       GFP_ATOMIC | __GFP_NOWARN, align_mask);
320 	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
321 	return data;
322 
323 }
324 EXPORT_SYMBOL(__napi_alloc_frag_align);
325 
326 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
327 {
328 	void *data;
329 
330 	if (in_hardirq() || irqs_disabled()) {
331 		struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
332 
333 		fragsz = SKB_DATA_ALIGN(fragsz);
334 		data = __page_frag_alloc_align(nc, fragsz,
335 					       GFP_ATOMIC | __GFP_NOWARN,
336 					       align_mask);
337 	} else {
338 		local_bh_disable();
339 		data = __napi_alloc_frag_align(fragsz, align_mask);
340 		local_bh_enable();
341 	}
342 	return data;
343 }
344 EXPORT_SYMBOL(__netdev_alloc_frag_align);
345 
346 static struct sk_buff *napi_skb_cache_get(void)
347 {
348 	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
349 	struct sk_buff *skb;
350 
351 	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
352 	if (unlikely(!nc->skb_count)) {
353 		nc->skb_count = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
354 						      GFP_ATOMIC | __GFP_NOWARN,
355 						      NAPI_SKB_CACHE_BULK,
356 						      nc->skb_cache);
357 		if (unlikely(!nc->skb_count)) {
358 			local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
359 			return NULL;
360 		}
361 	}
362 
363 	skb = nc->skb_cache[--nc->skb_count];
364 	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
365 	kasan_mempool_unpoison_object(skb, kmem_cache_size(net_hotdata.skbuff_cache));
366 
367 	return skb;
368 }
369 
370 static inline void __finalize_skb_around(struct sk_buff *skb, void *data,
371 					 unsigned int size)
372 {
373 	struct skb_shared_info *shinfo;
374 
375 	size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
376 
377 	/* Assumes caller memset cleared SKB */
378 	skb->truesize = SKB_TRUESIZE(size);
379 	refcount_set(&skb->users, 1);
380 	skb->head = data;
381 	skb->data = data;
382 	skb_reset_tail_pointer(skb);
383 	skb_set_end_offset(skb, size);
384 	skb->mac_header = (typeof(skb->mac_header))~0U;
385 	skb->transport_header = (typeof(skb->transport_header))~0U;
386 	skb->alloc_cpu = raw_smp_processor_id();
387 	/* make sure we initialize shinfo sequentially */
388 	shinfo = skb_shinfo(skb);
389 	memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
390 	atomic_set(&shinfo->dataref, 1);
391 
392 	skb_set_kcov_handle(skb, kcov_common_handle());
393 }
394 
395 static inline void *__slab_build_skb(struct sk_buff *skb, void *data,
396 				     unsigned int *size)
397 {
398 	void *resized;
399 
400 	/* Must find the allocation size (and grow it to match). */
401 	*size = ksize(data);
402 	/* krealloc() will immediately return "data" when
403 	 * "ksize(data)" is requested: it is the existing upper
404 	 * bounds. As a result, GFP_ATOMIC will be ignored. Note
405 	 * that this "new" pointer needs to be passed back to the
406 	 * caller for use so the __alloc_size hinting will be
407 	 * tracked correctly.
408 	 */
409 	resized = krealloc(data, *size, GFP_ATOMIC);
410 	WARN_ON_ONCE(resized != data);
411 	return resized;
412 }
413 
414 /* build_skb() variant which can operate on slab buffers.
415  * Note that this should be used sparingly as slab buffers
416  * cannot be combined efficiently by GRO!
417  */
418 struct sk_buff *slab_build_skb(void *data)
419 {
420 	struct sk_buff *skb;
421 	unsigned int size;
422 
423 	skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
424 			       GFP_ATOMIC | __GFP_NOWARN);
425 	if (unlikely(!skb))
426 		return NULL;
427 
428 	memset(skb, 0, offsetof(struct sk_buff, tail));
429 	data = __slab_build_skb(skb, data, &size);
430 	__finalize_skb_around(skb, data, size);
431 
432 	return skb;
433 }
434 EXPORT_SYMBOL(slab_build_skb);
435 
436 /* Caller must provide SKB that is memset cleared */
437 static void __build_skb_around(struct sk_buff *skb, void *data,
438 			       unsigned int frag_size)
439 {
440 	unsigned int size = frag_size;
441 
442 	/* frag_size == 0 is considered deprecated now. Callers
443 	 * using slab buffer should use slab_build_skb() instead.
444 	 */
445 	if (WARN_ONCE(size == 0, "Use slab_build_skb() instead"))
446 		data = __slab_build_skb(skb, data, &size);
447 
448 	__finalize_skb_around(skb, data, size);
449 }
450 
451 /**
452  * __build_skb - build a network buffer
453  * @data: data buffer provided by caller
454  * @frag_size: size of data (must not be 0)
455  *
456  * Allocate a new &sk_buff. Caller provides space holding head and
457  * skb_shared_info. @data must have been allocated from the page
458  * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc()
459  * allocation is deprecated, and callers should use slab_build_skb()
460  * instead.)
461  * The return is the new skb buffer.
462  * On a failure the return is %NULL, and @data is not freed.
463  * Notes :
464  *  Before IO, driver allocates only data buffer where NIC put incoming frame
465  *  Driver should add room at head (NET_SKB_PAD) and
466  *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
467  *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
468  *  before giving packet to stack.
469  *  RX rings only contains data buffers, not full skbs.
470  */
471 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
472 {
473 	struct sk_buff *skb;
474 
475 	skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
476 			       GFP_ATOMIC | __GFP_NOWARN);
477 	if (unlikely(!skb))
478 		return NULL;
479 
480 	memset(skb, 0, offsetof(struct sk_buff, tail));
481 	__build_skb_around(skb, data, frag_size);
482 
483 	return skb;
484 }
485 
486 /* build_skb() is wrapper over __build_skb(), that specifically
487  * takes care of skb->head and skb->pfmemalloc
488  */
489 struct sk_buff *build_skb(void *data, unsigned int frag_size)
490 {
491 	struct sk_buff *skb = __build_skb(data, frag_size);
492 
493 	if (likely(skb && frag_size)) {
494 		skb->head_frag = 1;
495 		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
496 	}
497 	return skb;
498 }
499 EXPORT_SYMBOL(build_skb);
500 
501 /**
502  * build_skb_around - build a network buffer around provided skb
503  * @skb: sk_buff provide by caller, must be memset cleared
504  * @data: data buffer provided by caller
505  * @frag_size: size of data
506  */
507 struct sk_buff *build_skb_around(struct sk_buff *skb,
508 				 void *data, unsigned int frag_size)
509 {
510 	if (unlikely(!skb))
511 		return NULL;
512 
513 	__build_skb_around(skb, data, frag_size);
514 
515 	if (frag_size) {
516 		skb->head_frag = 1;
517 		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
518 	}
519 	return skb;
520 }
521 EXPORT_SYMBOL(build_skb_around);
522 
523 /**
524  * __napi_build_skb - build a network buffer
525  * @data: data buffer provided by caller
526  * @frag_size: size of data
527  *
528  * Version of __build_skb() that uses NAPI percpu caches to obtain
529  * skbuff_head instead of inplace allocation.
530  *
531  * Returns a new &sk_buff on success, %NULL on allocation failure.
532  */
533 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
534 {
535 	struct sk_buff *skb;
536 
537 	skb = napi_skb_cache_get();
538 	if (unlikely(!skb))
539 		return NULL;
540 
541 	memset(skb, 0, offsetof(struct sk_buff, tail));
542 	__build_skb_around(skb, data, frag_size);
543 
544 	return skb;
545 }
546 
547 /**
548  * napi_build_skb - build a network buffer
549  * @data: data buffer provided by caller
550  * @frag_size: size of data
551  *
552  * Version of __napi_build_skb() that takes care of skb->head_frag
553  * and skb->pfmemalloc when the data is a page or page fragment.
554  *
555  * Returns a new &sk_buff on success, %NULL on allocation failure.
556  */
557 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
558 {
559 	struct sk_buff *skb = __napi_build_skb(data, frag_size);
560 
561 	if (likely(skb) && frag_size) {
562 		skb->head_frag = 1;
563 		skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
564 	}
565 
566 	return skb;
567 }
568 EXPORT_SYMBOL(napi_build_skb);
569 
570 /*
571  * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
572  * the caller if emergency pfmemalloc reserves are being used. If it is and
573  * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
574  * may be used. Otherwise, the packet data may be discarded until enough
575  * memory is free
576  */
577 static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
578 			     bool *pfmemalloc)
579 {
580 	bool ret_pfmemalloc = false;
581 	size_t obj_size;
582 	void *obj;
583 
584 	obj_size = SKB_HEAD_ALIGN(*size);
585 	if (obj_size <= SKB_SMALL_HEAD_CACHE_SIZE &&
586 	    !(flags & KMALLOC_NOT_NORMAL_BITS)) {
587 		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache,
588 				flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
589 				node);
590 		*size = SKB_SMALL_HEAD_CACHE_SIZE;
591 		if (obj || !(gfp_pfmemalloc_allowed(flags)))
592 			goto out;
593 		/* Try again but now we are using pfmemalloc reserves */
594 		ret_pfmemalloc = true;
595 		obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache, flags, node);
596 		goto out;
597 	}
598 
599 	obj_size = kmalloc_size_roundup(obj_size);
600 	/* The following cast might truncate high-order bits of obj_size, this
601 	 * is harmless because kmalloc(obj_size >= 2^32) will fail anyway.
602 	 */
603 	*size = (unsigned int)obj_size;
604 
605 	/*
606 	 * Try a regular allocation, when that fails and we're not entitled
607 	 * to the reserves, fail.
608 	 */
609 	obj = kmalloc_node_track_caller(obj_size,
610 					flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
611 					node);
612 	if (obj || !(gfp_pfmemalloc_allowed(flags)))
613 		goto out;
614 
615 	/* Try again but now we are using pfmemalloc reserves */
616 	ret_pfmemalloc = true;
617 	obj = kmalloc_node_track_caller(obj_size, flags, node);
618 
619 out:
620 	if (pfmemalloc)
621 		*pfmemalloc = ret_pfmemalloc;
622 
623 	return obj;
624 }
625 
626 /* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
627  *	'private' fields and also do memory statistics to find all the
628  *	[BEEP] leaks.
629  *
630  */
631 
632 /**
633  *	__alloc_skb	-	allocate a network buffer
634  *	@size: size to allocate
635  *	@gfp_mask: allocation mask
636  *	@flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
637  *		instead of head cache and allocate a cloned (child) skb.
638  *		If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
639  *		allocations in case the data is required for writeback
640  *	@node: numa node to allocate memory on
641  *
642  *	Allocate a new &sk_buff. The returned buffer has no headroom and a
643  *	tail room of at least size bytes. The object has a reference count
644  *	of one. The return is the buffer. On a failure the return is %NULL.
645  *
646  *	Buffers may only be allocated from interrupts using a @gfp_mask of
647  *	%GFP_ATOMIC.
648  */
649 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
650 			    int flags, int node)
651 {
652 	struct kmem_cache *cache;
653 	struct sk_buff *skb;
654 	bool pfmemalloc;
655 	u8 *data;
656 
657 	cache = (flags & SKB_ALLOC_FCLONE)
658 		? net_hotdata.skbuff_fclone_cache : net_hotdata.skbuff_cache;
659 
660 	if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
661 		gfp_mask |= __GFP_MEMALLOC;
662 
663 	/* Get the HEAD */
664 	if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
665 	    likely(node == NUMA_NO_NODE || node == numa_mem_id()))
666 		skb = napi_skb_cache_get();
667 	else
668 		skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
669 	if (unlikely(!skb))
670 		return NULL;
671 	prefetchw(skb);
672 
673 	/* We do our best to align skb_shared_info on a separate cache
674 	 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
675 	 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
676 	 * Both skb->head and skb_shared_info are cache line aligned.
677 	 */
678 	data = kmalloc_reserve(&size, gfp_mask, node, &pfmemalloc);
679 	if (unlikely(!data))
680 		goto nodata;
681 	/* kmalloc_size_roundup() might give us more room than requested.
682 	 * Put skb_shared_info exactly at the end of allocated zone,
683 	 * to allow max possible filling before reallocation.
684 	 */
685 	prefetchw(data + SKB_WITH_OVERHEAD(size));
686 
687 	/*
688 	 * Only clear those fields we need to clear, not those that we will
689 	 * actually initialise below. Hence, don't put any more fields after
690 	 * the tail pointer in struct sk_buff!
691 	 */
692 	memset(skb, 0, offsetof(struct sk_buff, tail));
693 	__build_skb_around(skb, data, size);
694 	skb->pfmemalloc = pfmemalloc;
695 
696 	if (flags & SKB_ALLOC_FCLONE) {
697 		struct sk_buff_fclones *fclones;
698 
699 		fclones = container_of(skb, struct sk_buff_fclones, skb1);
700 
701 		skb->fclone = SKB_FCLONE_ORIG;
702 		refcount_set(&fclones->fclone_ref, 1);
703 	}
704 
705 	return skb;
706 
707 nodata:
708 	kmem_cache_free(cache, skb);
709 	return NULL;
710 }
711 EXPORT_SYMBOL(__alloc_skb);
712 
713 /**
714  *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
715  *	@dev: network device to receive on
716  *	@len: length to allocate
717  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
718  *
719  *	Allocate a new &sk_buff and assign it a usage count of one. The
720  *	buffer has NET_SKB_PAD headroom built in. Users should allocate
721  *	the headroom they think they need without accounting for the
722  *	built in space. The built in space is used for optimisations.
723  *
724  *	%NULL is returned if there is no free memory.
725  */
726 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
727 				   gfp_t gfp_mask)
728 {
729 	struct page_frag_cache *nc;
730 	struct sk_buff *skb;
731 	bool pfmemalloc;
732 	void *data;
733 
734 	len += NET_SKB_PAD;
735 
736 	/* If requested length is either too small or too big,
737 	 * we use kmalloc() for skb->head allocation.
738 	 */
739 	if (len <= SKB_WITH_OVERHEAD(1024) ||
740 	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
741 	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
742 		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
743 		if (!skb)
744 			goto skb_fail;
745 		goto skb_success;
746 	}
747 
748 	len = SKB_HEAD_ALIGN(len);
749 
750 	if (sk_memalloc_socks())
751 		gfp_mask |= __GFP_MEMALLOC;
752 
753 	if (in_hardirq() || irqs_disabled()) {
754 		nc = this_cpu_ptr(&netdev_alloc_cache);
755 		data = page_frag_alloc(nc, len, gfp_mask);
756 		pfmemalloc = nc->pfmemalloc;
757 	} else {
758 		local_bh_disable();
759 		local_lock_nested_bh(&napi_alloc_cache.bh_lock);
760 
761 		nc = this_cpu_ptr(&napi_alloc_cache.page);
762 		data = page_frag_alloc(nc, len, gfp_mask);
763 		pfmemalloc = nc->pfmemalloc;
764 
765 		local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
766 		local_bh_enable();
767 	}
768 
769 	if (unlikely(!data))
770 		return NULL;
771 
772 	skb = __build_skb(data, len);
773 	if (unlikely(!skb)) {
774 		skb_free_frag(data);
775 		return NULL;
776 	}
777 
778 	if (pfmemalloc)
779 		skb->pfmemalloc = 1;
780 	skb->head_frag = 1;
781 
782 skb_success:
783 	skb_reserve(skb, NET_SKB_PAD);
784 	skb->dev = dev;
785 
786 skb_fail:
787 	return skb;
788 }
789 EXPORT_SYMBOL(__netdev_alloc_skb);
790 
791 /**
792  *	napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
793  *	@napi: napi instance this buffer was allocated for
794  *	@len: length to allocate
795  *
796  *	Allocate a new sk_buff for use in NAPI receive.  This buffer will
797  *	attempt to allocate the head from a special reserved region used
798  *	only for NAPI Rx allocation.  By doing this we can save several
799  *	CPU cycles by avoiding having to disable and re-enable IRQs.
800  *
801  *	%NULL is returned if there is no free memory.
802  */
803 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len)
804 {
805 	gfp_t gfp_mask = GFP_ATOMIC | __GFP_NOWARN;
806 	struct napi_alloc_cache *nc;
807 	struct sk_buff *skb;
808 	bool pfmemalloc;
809 	void *data;
810 
811 	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
812 	len += NET_SKB_PAD + NET_IP_ALIGN;
813 
814 	/* If requested length is either too small or too big,
815 	 * we use kmalloc() for skb->head allocation.
816 	 * When the small frag allocator is available, prefer it over kmalloc
817 	 * for small fragments
818 	 */
819 	if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
820 	    len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
821 	    (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
822 		skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
823 				  NUMA_NO_NODE);
824 		if (!skb)
825 			goto skb_fail;
826 		goto skb_success;
827 	}
828 
829 	if (sk_memalloc_socks())
830 		gfp_mask |= __GFP_MEMALLOC;
831 
832 	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
833 	nc = this_cpu_ptr(&napi_alloc_cache);
834 	if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
835 		/* we are artificially inflating the allocation size, but
836 		 * that is not as bad as it may look like, as:
837 		 * - 'len' less than GRO_MAX_HEAD makes little sense
838 		 * - On most systems, larger 'len' values lead to fragment
839 		 *   size above 512 bytes
840 		 * - kmalloc would use the kmalloc-1k slab for such values
841 		 * - Builds with smaller GRO_MAX_HEAD will very likely do
842 		 *   little networking, as that implies no WiFi and no
843 		 *   tunnels support, and 32 bits arches.
844 		 */
845 		len = SZ_1K;
846 
847 		data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
848 		pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
849 	} else {
850 		len = SKB_HEAD_ALIGN(len);
851 
852 		data = page_frag_alloc(&nc->page, len, gfp_mask);
853 		pfmemalloc = nc->page.pfmemalloc;
854 	}
855 	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
856 
857 	if (unlikely(!data))
858 		return NULL;
859 
860 	skb = __napi_build_skb(data, len);
861 	if (unlikely(!skb)) {
862 		skb_free_frag(data);
863 		return NULL;
864 	}
865 
866 	if (pfmemalloc)
867 		skb->pfmemalloc = 1;
868 	skb->head_frag = 1;
869 
870 skb_success:
871 	skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
872 	skb->dev = napi->dev;
873 
874 skb_fail:
875 	return skb;
876 }
877 EXPORT_SYMBOL(napi_alloc_skb);
878 
879 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
880 			    int off, int size, unsigned int truesize)
881 {
882 	DEBUG_NET_WARN_ON_ONCE(size > truesize);
883 
884 	skb_fill_netmem_desc(skb, i, netmem, off, size);
885 	skb->len += size;
886 	skb->data_len += size;
887 	skb->truesize += truesize;
888 }
889 EXPORT_SYMBOL(skb_add_rx_frag_netmem);
890 
891 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
892 			  unsigned int truesize)
893 {
894 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
895 
896 	DEBUG_NET_WARN_ON_ONCE(size > truesize);
897 
898 	skb_frag_size_add(frag, size);
899 	skb->len += size;
900 	skb->data_len += size;
901 	skb->truesize += truesize;
902 }
903 EXPORT_SYMBOL(skb_coalesce_rx_frag);
904 
905 static void skb_drop_list(struct sk_buff **listp)
906 {
907 	kfree_skb_list(*listp);
908 	*listp = NULL;
909 }
910 
911 static inline void skb_drop_fraglist(struct sk_buff *skb)
912 {
913 	skb_drop_list(&skb_shinfo(skb)->frag_list);
914 }
915 
916 static void skb_clone_fraglist(struct sk_buff *skb)
917 {
918 	struct sk_buff *list;
919 
920 	skb_walk_frags(skb, list)
921 		skb_get(list);
922 }
923 
924 static bool is_pp_netmem(netmem_ref netmem)
925 {
926 	return (netmem_get_pp_magic(netmem) & ~0x3UL) == PP_SIGNATURE;
927 }
928 
929 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
930 		    unsigned int headroom)
931 {
932 #if IS_ENABLED(CONFIG_PAGE_POOL)
933 	u32 size, truesize, len, max_head_size, off;
934 	struct sk_buff *skb = *pskb, *nskb;
935 	int err, i, head_off;
936 	void *data;
937 
938 	/* XDP does not support fraglist so we need to linearize
939 	 * the skb.
940 	 */
941 	if (skb_has_frag_list(skb))
942 		return -EOPNOTSUPP;
943 
944 	max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE - headroom);
945 	if (skb->len > max_head_size + MAX_SKB_FRAGS * PAGE_SIZE)
946 		return -ENOMEM;
947 
948 	size = min_t(u32, skb->len, max_head_size);
949 	truesize = SKB_HEAD_ALIGN(size) + headroom;
950 	data = page_pool_dev_alloc_va(pool, &truesize);
951 	if (!data)
952 		return -ENOMEM;
953 
954 	nskb = napi_build_skb(data, truesize);
955 	if (!nskb) {
956 		page_pool_free_va(pool, data, true);
957 		return -ENOMEM;
958 	}
959 
960 	skb_reserve(nskb, headroom);
961 	skb_copy_header(nskb, skb);
962 	skb_mark_for_recycle(nskb);
963 
964 	err = skb_copy_bits(skb, 0, nskb->data, size);
965 	if (err) {
966 		consume_skb(nskb);
967 		return err;
968 	}
969 	skb_put(nskb, size);
970 
971 	head_off = skb_headroom(nskb) - skb_headroom(skb);
972 	skb_headers_offset_update(nskb, head_off);
973 
974 	off = size;
975 	len = skb->len - off;
976 	for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
977 		struct page *page;
978 		u32 page_off;
979 
980 		size = min_t(u32, len, PAGE_SIZE);
981 		truesize = size;
982 
983 		page = page_pool_dev_alloc(pool, &page_off, &truesize);
984 		if (!page) {
985 			consume_skb(nskb);
986 			return -ENOMEM;
987 		}
988 
989 		skb_add_rx_frag(nskb, i, page, page_off, size, truesize);
990 		err = skb_copy_bits(skb, off, page_address(page) + page_off,
991 				    size);
992 		if (err) {
993 			consume_skb(nskb);
994 			return err;
995 		}
996 
997 		len -= size;
998 		off += size;
999 	}
1000 
1001 	consume_skb(skb);
1002 	*pskb = nskb;
1003 
1004 	return 0;
1005 #else
1006 	return -EOPNOTSUPP;
1007 #endif
1008 }
1009 EXPORT_SYMBOL(skb_pp_cow_data);
1010 
1011 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
1012 			 struct bpf_prog *prog)
1013 {
1014 	if (!prog->aux->xdp_has_frags)
1015 		return -EINVAL;
1016 
1017 	return skb_pp_cow_data(pool, pskb, XDP_PACKET_HEADROOM);
1018 }
1019 EXPORT_SYMBOL(skb_cow_data_for_xdp);
1020 
1021 #if IS_ENABLED(CONFIG_PAGE_POOL)
1022 bool napi_pp_put_page(netmem_ref netmem)
1023 {
1024 	netmem = netmem_compound_head(netmem);
1025 
1026 	/* page->pp_magic is OR'ed with PP_SIGNATURE after the allocation
1027 	 * in order to preserve any existing bits, such as bit 0 for the
1028 	 * head page of compound page and bit 1 for pfmemalloc page, so
1029 	 * mask those bits for freeing side when doing below checking,
1030 	 * and page_is_pfmemalloc() is checked in __page_pool_put_page()
1031 	 * to avoid recycling the pfmemalloc page.
1032 	 */
1033 	if (unlikely(!is_pp_netmem(netmem)))
1034 		return false;
1035 
1036 	page_pool_put_full_netmem(netmem_get_pp(netmem), netmem, false);
1037 
1038 	return true;
1039 }
1040 EXPORT_SYMBOL(napi_pp_put_page);
1041 #endif
1042 
1043 static bool skb_pp_recycle(struct sk_buff *skb, void *data)
1044 {
1045 	if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
1046 		return false;
1047 	return napi_pp_put_page(page_to_netmem(virt_to_page(data)));
1048 }
1049 
1050 /**
1051  * skb_pp_frag_ref() - Increase fragment references of a page pool aware skb
1052  * @skb:	page pool aware skb
1053  *
1054  * Increase the fragment reference count (pp_ref_count) of a skb. This is
1055  * intended to gain fragment references only for page pool aware skbs,
1056  * i.e. when skb->pp_recycle is true, and not for fragments in a
1057  * non-pp-recycling skb. It has a fallback to increase references on normal
1058  * pages, as page pool aware skbs may also have normal page fragments.
1059  */
1060 static int skb_pp_frag_ref(struct sk_buff *skb)
1061 {
1062 	struct skb_shared_info *shinfo;
1063 	netmem_ref head_netmem;
1064 	int i;
1065 
1066 	if (!skb->pp_recycle)
1067 		return -EINVAL;
1068 
1069 	shinfo = skb_shinfo(skb);
1070 
1071 	for (i = 0; i < shinfo->nr_frags; i++) {
1072 		head_netmem = netmem_compound_head(shinfo->frags[i].netmem);
1073 		if (likely(is_pp_netmem(head_netmem)))
1074 			page_pool_ref_netmem(head_netmem);
1075 		else
1076 			page_ref_inc(netmem_to_page(head_netmem));
1077 	}
1078 	return 0;
1079 }
1080 
1081 static void skb_kfree_head(void *head, unsigned int end_offset)
1082 {
1083 	if (end_offset == SKB_SMALL_HEAD_HEADROOM)
1084 		kmem_cache_free(net_hotdata.skb_small_head_cache, head);
1085 	else
1086 		kfree(head);
1087 }
1088 
1089 static void skb_free_head(struct sk_buff *skb)
1090 {
1091 	unsigned char *head = skb->head;
1092 
1093 	if (skb->head_frag) {
1094 		if (skb_pp_recycle(skb, head))
1095 			return;
1096 		skb_free_frag(head);
1097 	} else {
1098 		skb_kfree_head(head, skb_end_offset(skb));
1099 	}
1100 }
1101 
1102 static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
1103 {
1104 	struct skb_shared_info *shinfo = skb_shinfo(skb);
1105 	int i;
1106 
1107 	if (!skb_data_unref(skb, shinfo))
1108 		goto exit;
1109 
1110 	if (skb_zcopy(skb)) {
1111 		bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
1112 
1113 		skb_zcopy_clear(skb, true);
1114 		if (skip_unref)
1115 			goto free_head;
1116 	}
1117 
1118 	for (i = 0; i < shinfo->nr_frags; i++)
1119 		__skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
1120 
1121 free_head:
1122 	if (shinfo->frag_list)
1123 		kfree_skb_list_reason(shinfo->frag_list, reason);
1124 
1125 	skb_free_head(skb);
1126 exit:
1127 	/* When we clone an SKB we copy the reycling bit. The pp_recycle
1128 	 * bit is only set on the head though, so in order to avoid races
1129 	 * while trying to recycle fragments on __skb_frag_unref() we need
1130 	 * to make one SKB responsible for triggering the recycle path.
1131 	 * So disable the recycling bit if an SKB is cloned and we have
1132 	 * additional references to the fragmented part of the SKB.
1133 	 * Eventually the last SKB will have the recycling bit set and it's
1134 	 * dataref set to 0, which will trigger the recycling
1135 	 */
1136 	skb->pp_recycle = 0;
1137 }
1138 
1139 /*
1140  *	Free an skbuff by memory without cleaning the state.
1141  */
1142 static void kfree_skbmem(struct sk_buff *skb)
1143 {
1144 	struct sk_buff_fclones *fclones;
1145 
1146 	switch (skb->fclone) {
1147 	case SKB_FCLONE_UNAVAILABLE:
1148 		kmem_cache_free(net_hotdata.skbuff_cache, skb);
1149 		return;
1150 
1151 	case SKB_FCLONE_ORIG:
1152 		fclones = container_of(skb, struct sk_buff_fclones, skb1);
1153 
1154 		/* We usually free the clone (TX completion) before original skb
1155 		 * This test would have no chance to be true for the clone,
1156 		 * while here, branch prediction will be good.
1157 		 */
1158 		if (refcount_read(&fclones->fclone_ref) == 1)
1159 			goto fastpath;
1160 		break;
1161 
1162 	default: /* SKB_FCLONE_CLONE */
1163 		fclones = container_of(skb, struct sk_buff_fclones, skb2);
1164 		break;
1165 	}
1166 	if (!refcount_dec_and_test(&fclones->fclone_ref))
1167 		return;
1168 fastpath:
1169 	kmem_cache_free(net_hotdata.skbuff_fclone_cache, fclones);
1170 }
1171 
1172 void skb_release_head_state(struct sk_buff *skb)
1173 {
1174 	skb_dst_drop(skb);
1175 	if (skb->destructor) {
1176 		DEBUG_NET_WARN_ON_ONCE(in_hardirq());
1177 		skb->destructor(skb);
1178 	}
1179 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1180 	nf_conntrack_put(skb_nfct(skb));
1181 #endif
1182 	skb_ext_put(skb);
1183 }
1184 
1185 /* Free everything but the sk_buff shell. */
1186 static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
1187 {
1188 	skb_release_head_state(skb);
1189 	if (likely(skb->head))
1190 		skb_release_data(skb, reason);
1191 }
1192 
1193 /**
1194  *	__kfree_skb - private function
1195  *	@skb: buffer
1196  *
1197  *	Free an sk_buff. Release anything attached to the buffer.
1198  *	Clean the state. This is an internal helper function. Users should
1199  *	always call kfree_skb
1200  */
1201 
1202 void __kfree_skb(struct sk_buff *skb)
1203 {
1204 	skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1205 	kfree_skbmem(skb);
1206 }
1207 EXPORT_SYMBOL(__kfree_skb);
1208 
1209 static __always_inline
1210 bool __sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1211 			  enum skb_drop_reason reason)
1212 {
1213 	if (unlikely(!skb_unref(skb)))
1214 		return false;
1215 
1216 	DEBUG_NET_WARN_ON_ONCE(reason == SKB_NOT_DROPPED_YET ||
1217 			       u32_get_bits(reason,
1218 					    SKB_DROP_REASON_SUBSYS_MASK) >=
1219 				SKB_DROP_REASON_SUBSYS_NUM);
1220 
1221 	if (reason == SKB_CONSUMED)
1222 		trace_consume_skb(skb, __builtin_return_address(0));
1223 	else
1224 		trace_kfree_skb(skb, __builtin_return_address(0), reason, sk);
1225 	return true;
1226 }
1227 
1228 /**
1229  *	sk_skb_reason_drop - free an sk_buff with special reason
1230  *	@sk: the socket to receive @skb, or NULL if not applicable
1231  *	@skb: buffer to free
1232  *	@reason: reason why this skb is dropped
1233  *
1234  *	Drop a reference to the buffer and free it if the usage count has hit
1235  *	zero. Meanwhile, pass the receiving socket and drop reason to
1236  *	'kfree_skb' tracepoint.
1237  */
1238 void __fix_address
1239 sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, enum skb_drop_reason reason)
1240 {
1241 	if (__sk_skb_reason_drop(sk, skb, reason))
1242 		__kfree_skb(skb);
1243 }
1244 EXPORT_SYMBOL(sk_skb_reason_drop);
1245 
1246 #define KFREE_SKB_BULK_SIZE	16
1247 
1248 struct skb_free_array {
1249 	unsigned int skb_count;
1250 	void *skb_array[KFREE_SKB_BULK_SIZE];
1251 };
1252 
1253 static void kfree_skb_add_bulk(struct sk_buff *skb,
1254 			       struct skb_free_array *sa,
1255 			       enum skb_drop_reason reason)
1256 {
1257 	/* if SKB is a clone, don't handle this case */
1258 	if (unlikely(skb->fclone != SKB_FCLONE_UNAVAILABLE)) {
1259 		__kfree_skb(skb);
1260 		return;
1261 	}
1262 
1263 	skb_release_all(skb, reason);
1264 	sa->skb_array[sa->skb_count++] = skb;
1265 
1266 	if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
1267 		kmem_cache_free_bulk(net_hotdata.skbuff_cache, KFREE_SKB_BULK_SIZE,
1268 				     sa->skb_array);
1269 		sa->skb_count = 0;
1270 	}
1271 }
1272 
1273 void __fix_address
1274 kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
1275 {
1276 	struct skb_free_array sa;
1277 
1278 	sa.skb_count = 0;
1279 
1280 	while (segs) {
1281 		struct sk_buff *next = segs->next;
1282 
1283 		if (__sk_skb_reason_drop(NULL, segs, reason)) {
1284 			skb_poison_list(segs);
1285 			kfree_skb_add_bulk(segs, &sa, reason);
1286 		}
1287 
1288 		segs = next;
1289 	}
1290 
1291 	if (sa.skb_count)
1292 		kmem_cache_free_bulk(net_hotdata.skbuff_cache, sa.skb_count, sa.skb_array);
1293 }
1294 EXPORT_SYMBOL(kfree_skb_list_reason);
1295 
1296 /* Dump skb information and contents.
1297  *
1298  * Must only be called from net_ratelimit()-ed paths.
1299  *
1300  * Dumps whole packets if full_pkt, only headers otherwise.
1301  */
1302 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
1303 {
1304 	struct skb_shared_info *sh = skb_shinfo(skb);
1305 	struct net_device *dev = skb->dev;
1306 	struct sock *sk = skb->sk;
1307 	struct sk_buff *list_skb;
1308 	bool has_mac, has_trans;
1309 	int headroom, tailroom;
1310 	int i, len, seg_len;
1311 
1312 	if (full_pkt)
1313 		len = skb->len;
1314 	else
1315 		len = min_t(int, skb->len, MAX_HEADER + 128);
1316 
1317 	headroom = skb_headroom(skb);
1318 	tailroom = skb_tailroom(skb);
1319 
1320 	has_mac = skb_mac_header_was_set(skb);
1321 	has_trans = skb_transport_header_was_set(skb);
1322 
1323 	printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1324 	       "mac=(%d,%d) mac_len=%u net=(%d,%d) trans=%d\n"
1325 	       "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1326 	       "csum(0x%x start=%u offset=%u ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1327 	       "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n"
1328 	       "priority=0x%x mark=0x%x alloc_cpu=%u vlan_all=0x%x\n"
1329 	       "encapsulation=%d inner(proto=0x%04x, mac=%u, net=%u, trans=%u)\n",
1330 	       level, skb->len, headroom, skb_headlen(skb), tailroom,
1331 	       has_mac ? skb->mac_header : -1,
1332 	       has_mac ? skb_mac_header_len(skb) : -1,
1333 	       skb->mac_len,
1334 	       skb->network_header,
1335 	       has_trans ? skb_network_header_len(skb) : -1,
1336 	       has_trans ? skb->transport_header : -1,
1337 	       sh->tx_flags, sh->nr_frags,
1338 	       sh->gso_size, sh->gso_type, sh->gso_segs,
1339 	       skb->csum, skb->csum_start, skb->csum_offset, skb->ip_summed,
1340 	       skb->csum_complete_sw, skb->csum_valid, skb->csum_level,
1341 	       skb->hash, skb->sw_hash, skb->l4_hash,
1342 	       ntohs(skb->protocol), skb->pkt_type, skb->skb_iif,
1343 	       skb->priority, skb->mark, skb->alloc_cpu, skb->vlan_all,
1344 	       skb->encapsulation, skb->inner_protocol, skb->inner_mac_header,
1345 	       skb->inner_network_header, skb->inner_transport_header);
1346 
1347 	if (dev)
1348 		printk("%sdev name=%s feat=%pNF\n",
1349 		       level, dev->name, &dev->features);
1350 	if (sk)
1351 		printk("%ssk family=%hu type=%u proto=%u\n",
1352 		       level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1353 
1354 	if (full_pkt && headroom)
1355 		print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1356 			       16, 1, skb->head, headroom, false);
1357 
1358 	seg_len = min_t(int, skb_headlen(skb), len);
1359 	if (seg_len)
1360 		print_hex_dump(level, "skb linear:   ", DUMP_PREFIX_OFFSET,
1361 			       16, 1, skb->data, seg_len, false);
1362 	len -= seg_len;
1363 
1364 	if (full_pkt && tailroom)
1365 		print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1366 			       16, 1, skb_tail_pointer(skb), tailroom, false);
1367 
1368 	for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1369 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1370 		u32 p_off, p_len, copied;
1371 		struct page *p;
1372 		u8 *vaddr;
1373 
1374 		skb_frag_foreach_page(frag, skb_frag_off(frag),
1375 				      skb_frag_size(frag), p, p_off, p_len,
1376 				      copied) {
1377 			seg_len = min_t(int, p_len, len);
1378 			vaddr = kmap_atomic(p);
1379 			print_hex_dump(level, "skb frag:     ",
1380 				       DUMP_PREFIX_OFFSET,
1381 				       16, 1, vaddr + p_off, seg_len, false);
1382 			kunmap_atomic(vaddr);
1383 			len -= seg_len;
1384 			if (!len)
1385 				break;
1386 		}
1387 	}
1388 
1389 	if (full_pkt && skb_has_frag_list(skb)) {
1390 		printk("skb fraglist:\n");
1391 		skb_walk_frags(skb, list_skb)
1392 			skb_dump(level, list_skb, true);
1393 	}
1394 }
1395 EXPORT_SYMBOL(skb_dump);
1396 
1397 /**
1398  *	skb_tx_error - report an sk_buff xmit error
1399  *	@skb: buffer that triggered an error
1400  *
1401  *	Report xmit error if a device callback is tracking this skb.
1402  *	skb must be freed afterwards.
1403  */
1404 void skb_tx_error(struct sk_buff *skb)
1405 {
1406 	if (skb) {
1407 		skb_zcopy_downgrade_managed(skb);
1408 		skb_zcopy_clear(skb, true);
1409 	}
1410 }
1411 EXPORT_SYMBOL(skb_tx_error);
1412 
1413 #ifdef CONFIG_TRACEPOINTS
1414 /**
1415  *	consume_skb - free an skbuff
1416  *	@skb: buffer to free
1417  *
1418  *	Drop a ref to the buffer and free it if the usage count has hit zero
1419  *	Functions identically to kfree_skb, but kfree_skb assumes that the frame
1420  *	is being dropped after a failure and notes that
1421  */
1422 void consume_skb(struct sk_buff *skb)
1423 {
1424 	if (!skb_unref(skb))
1425 		return;
1426 
1427 	trace_consume_skb(skb, __builtin_return_address(0));
1428 	__kfree_skb(skb);
1429 }
1430 EXPORT_SYMBOL(consume_skb);
1431 #endif
1432 
1433 /**
1434  *	__consume_stateless_skb - free an skbuff, assuming it is stateless
1435  *	@skb: buffer to free
1436  *
1437  *	Alike consume_skb(), but this variant assumes that this is the last
1438  *	skb reference and all the head states have been already dropped
1439  */
1440 void __consume_stateless_skb(struct sk_buff *skb)
1441 {
1442 	trace_consume_skb(skb, __builtin_return_address(0));
1443 	skb_release_data(skb, SKB_CONSUMED);
1444 	kfree_skbmem(skb);
1445 }
1446 
1447 static void napi_skb_cache_put(struct sk_buff *skb)
1448 {
1449 	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1450 	u32 i;
1451 
1452 	if (!kasan_mempool_poison_object(skb))
1453 		return;
1454 
1455 	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
1456 	nc->skb_cache[nc->skb_count++] = skb;
1457 
1458 	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1459 		for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1460 			kasan_mempool_unpoison_object(nc->skb_cache[i],
1461 						kmem_cache_size(net_hotdata.skbuff_cache));
1462 
1463 		kmem_cache_free_bulk(net_hotdata.skbuff_cache, NAPI_SKB_CACHE_HALF,
1464 				     nc->skb_cache + NAPI_SKB_CACHE_HALF);
1465 		nc->skb_count = NAPI_SKB_CACHE_HALF;
1466 	}
1467 	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
1468 }
1469 
1470 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
1471 {
1472 	skb_release_all(skb, reason);
1473 	napi_skb_cache_put(skb);
1474 }
1475 
1476 void napi_skb_free_stolen_head(struct sk_buff *skb)
1477 {
1478 	if (unlikely(skb->slow_gro)) {
1479 		nf_reset_ct(skb);
1480 		skb_dst_drop(skb);
1481 		skb_ext_put(skb);
1482 		skb_orphan(skb);
1483 		skb->slow_gro = 0;
1484 	}
1485 	napi_skb_cache_put(skb);
1486 }
1487 
1488 void napi_consume_skb(struct sk_buff *skb, int budget)
1489 {
1490 	/* Zero budget indicate non-NAPI context called us, like netpoll */
1491 	if (unlikely(!budget)) {
1492 		dev_consume_skb_any(skb);
1493 		return;
1494 	}
1495 
1496 	DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1497 
1498 	if (!skb_unref(skb))
1499 		return;
1500 
1501 	/* if reaching here SKB is ready to free */
1502 	trace_consume_skb(skb, __builtin_return_address(0));
1503 
1504 	/* if SKB is a clone, don't handle this case */
1505 	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1506 		__kfree_skb(skb);
1507 		return;
1508 	}
1509 
1510 	skb_release_all(skb, SKB_CONSUMED);
1511 	napi_skb_cache_put(skb);
1512 }
1513 EXPORT_SYMBOL(napi_consume_skb);
1514 
1515 /* Make sure a field is contained by headers group */
1516 #define CHECK_SKB_FIELD(field) \
1517 	BUILD_BUG_ON(offsetof(struct sk_buff, field) !=		\
1518 		     offsetof(struct sk_buff, headers.field));	\
1519 
1520 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1521 {
1522 	new->tstamp		= old->tstamp;
1523 	/* We do not copy old->sk */
1524 	new->dev		= old->dev;
1525 	memcpy(new->cb, old->cb, sizeof(old->cb));
1526 	skb_dst_copy(new, old);
1527 	__skb_ext_copy(new, old);
1528 	__nf_copy(new, old, false);
1529 
1530 	/* Note : this field could be in the headers group.
1531 	 * It is not yet because we do not want to have a 16 bit hole
1532 	 */
1533 	new->queue_mapping = old->queue_mapping;
1534 
1535 	memcpy(&new->headers, &old->headers, sizeof(new->headers));
1536 	CHECK_SKB_FIELD(protocol);
1537 	CHECK_SKB_FIELD(csum);
1538 	CHECK_SKB_FIELD(hash);
1539 	CHECK_SKB_FIELD(priority);
1540 	CHECK_SKB_FIELD(skb_iif);
1541 	CHECK_SKB_FIELD(vlan_proto);
1542 	CHECK_SKB_FIELD(vlan_tci);
1543 	CHECK_SKB_FIELD(transport_header);
1544 	CHECK_SKB_FIELD(network_header);
1545 	CHECK_SKB_FIELD(mac_header);
1546 	CHECK_SKB_FIELD(inner_protocol);
1547 	CHECK_SKB_FIELD(inner_transport_header);
1548 	CHECK_SKB_FIELD(inner_network_header);
1549 	CHECK_SKB_FIELD(inner_mac_header);
1550 	CHECK_SKB_FIELD(mark);
1551 #ifdef CONFIG_NETWORK_SECMARK
1552 	CHECK_SKB_FIELD(secmark);
1553 #endif
1554 #ifdef CONFIG_NET_RX_BUSY_POLL
1555 	CHECK_SKB_FIELD(napi_id);
1556 #endif
1557 	CHECK_SKB_FIELD(alloc_cpu);
1558 #ifdef CONFIG_XPS
1559 	CHECK_SKB_FIELD(sender_cpu);
1560 #endif
1561 #ifdef CONFIG_NET_SCHED
1562 	CHECK_SKB_FIELD(tc_index);
1563 #endif
1564 
1565 }
1566 
1567 /*
1568  * You should not add any new code to this function.  Add it to
1569  * __copy_skb_header above instead.
1570  */
1571 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1572 {
1573 #define C(x) n->x = skb->x
1574 
1575 	n->next = n->prev = NULL;
1576 	n->sk = NULL;
1577 	__copy_skb_header(n, skb);
1578 
1579 	C(len);
1580 	C(data_len);
1581 	C(mac_len);
1582 	n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1583 	n->cloned = 1;
1584 	n->nohdr = 0;
1585 	n->peeked = 0;
1586 	C(pfmemalloc);
1587 	C(pp_recycle);
1588 	n->destructor = NULL;
1589 	C(tail);
1590 	C(end);
1591 	C(head);
1592 	C(head_frag);
1593 	C(data);
1594 	C(truesize);
1595 	refcount_set(&n->users, 1);
1596 
1597 	atomic_inc(&(skb_shinfo(skb)->dataref));
1598 	skb->cloned = 1;
1599 
1600 	return n;
1601 #undef C
1602 }
1603 
1604 /**
1605  * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1606  * @first: first sk_buff of the msg
1607  */
1608 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1609 {
1610 	struct sk_buff *n;
1611 
1612 	n = alloc_skb(0, GFP_ATOMIC);
1613 	if (!n)
1614 		return NULL;
1615 
1616 	n->len = first->len;
1617 	n->data_len = first->len;
1618 	n->truesize = first->truesize;
1619 
1620 	skb_shinfo(n)->frag_list = first;
1621 
1622 	__copy_skb_header(n, first);
1623 	n->destructor = NULL;
1624 
1625 	return n;
1626 }
1627 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1628 
1629 /**
1630  *	skb_morph	-	morph one skb into another
1631  *	@dst: the skb to receive the contents
1632  *	@src: the skb to supply the contents
1633  *
1634  *	This is identical to skb_clone except that the target skb is
1635  *	supplied by the user.
1636  *
1637  *	The target skb is returned upon exit.
1638  */
1639 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1640 {
1641 	skb_release_all(dst, SKB_CONSUMED);
1642 	return __skb_clone(dst, src);
1643 }
1644 EXPORT_SYMBOL_GPL(skb_morph);
1645 
1646 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1647 {
1648 	unsigned long max_pg, num_pg, new_pg, old_pg, rlim;
1649 	struct user_struct *user;
1650 
1651 	if (capable(CAP_IPC_LOCK) || !size)
1652 		return 0;
1653 
1654 	rlim = rlimit(RLIMIT_MEMLOCK);
1655 	if (rlim == RLIM_INFINITY)
1656 		return 0;
1657 
1658 	num_pg = (size >> PAGE_SHIFT) + 2;	/* worst case */
1659 	max_pg = rlim >> PAGE_SHIFT;
1660 	user = mmp->user ? : current_user();
1661 
1662 	old_pg = atomic_long_read(&user->locked_vm);
1663 	do {
1664 		new_pg = old_pg + num_pg;
1665 		if (new_pg > max_pg)
1666 			return -ENOBUFS;
1667 	} while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1668 
1669 	if (!mmp->user) {
1670 		mmp->user = get_uid(user);
1671 		mmp->num_pg = num_pg;
1672 	} else {
1673 		mmp->num_pg += num_pg;
1674 	}
1675 
1676 	return 0;
1677 }
1678 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1679 
1680 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1681 {
1682 	if (mmp->user) {
1683 		atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1684 		free_uid(mmp->user);
1685 	}
1686 }
1687 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1688 
1689 static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1690 {
1691 	struct ubuf_info_msgzc *uarg;
1692 	struct sk_buff *skb;
1693 
1694 	WARN_ON_ONCE(!in_task());
1695 
1696 	skb = sock_omalloc(sk, 0, GFP_KERNEL);
1697 	if (!skb)
1698 		return NULL;
1699 
1700 	BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1701 	uarg = (void *)skb->cb;
1702 	uarg->mmp.user = NULL;
1703 
1704 	if (mm_account_pinned_pages(&uarg->mmp, size)) {
1705 		kfree_skb(skb);
1706 		return NULL;
1707 	}
1708 
1709 	uarg->ubuf.ops = &msg_zerocopy_ubuf_ops;
1710 	uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1711 	uarg->len = 1;
1712 	uarg->bytelen = size;
1713 	uarg->zerocopy = 1;
1714 	uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1715 	refcount_set(&uarg->ubuf.refcnt, 1);
1716 	sock_hold(sk);
1717 
1718 	return &uarg->ubuf;
1719 }
1720 
1721 static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1722 {
1723 	return container_of((void *)uarg, struct sk_buff, cb);
1724 }
1725 
1726 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1727 				       struct ubuf_info *uarg)
1728 {
1729 	if (uarg) {
1730 		struct ubuf_info_msgzc *uarg_zc;
1731 		const u32 byte_limit = 1 << 19;		/* limit to a few TSO */
1732 		u32 bytelen, next;
1733 
1734 		/* there might be non MSG_ZEROCOPY users */
1735 		if (uarg->ops != &msg_zerocopy_ubuf_ops)
1736 			return NULL;
1737 
1738 		/* realloc only when socket is locked (TCP, UDP cork),
1739 		 * so uarg->len and sk_zckey access is serialized
1740 		 */
1741 		if (!sock_owned_by_user(sk)) {
1742 			WARN_ON_ONCE(1);
1743 			return NULL;
1744 		}
1745 
1746 		uarg_zc = uarg_to_msgzc(uarg);
1747 		bytelen = uarg_zc->bytelen + size;
1748 		if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1749 			/* TCP can create new skb to attach new uarg */
1750 			if (sk->sk_type == SOCK_STREAM)
1751 				goto new_alloc;
1752 			return NULL;
1753 		}
1754 
1755 		next = (u32)atomic_read(&sk->sk_zckey);
1756 		if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1757 			if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1758 				return NULL;
1759 			uarg_zc->len++;
1760 			uarg_zc->bytelen = bytelen;
1761 			atomic_set(&sk->sk_zckey, ++next);
1762 
1763 			/* no extra ref when appending to datagram (MSG_MORE) */
1764 			if (sk->sk_type == SOCK_STREAM)
1765 				net_zcopy_get(uarg);
1766 
1767 			return uarg;
1768 		}
1769 	}
1770 
1771 new_alloc:
1772 	return msg_zerocopy_alloc(sk, size);
1773 }
1774 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1775 
1776 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1777 {
1778 	struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1779 	u32 old_lo, old_hi;
1780 	u64 sum_len;
1781 
1782 	old_lo = serr->ee.ee_info;
1783 	old_hi = serr->ee.ee_data;
1784 	sum_len = old_hi - old_lo + 1ULL + len;
1785 
1786 	if (sum_len >= (1ULL << 32))
1787 		return false;
1788 
1789 	if (lo != old_hi + 1)
1790 		return false;
1791 
1792 	serr->ee.ee_data += len;
1793 	return true;
1794 }
1795 
1796 static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1797 {
1798 	struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1799 	struct sock_exterr_skb *serr;
1800 	struct sock *sk = skb->sk;
1801 	struct sk_buff_head *q;
1802 	unsigned long flags;
1803 	bool is_zerocopy;
1804 	u32 lo, hi;
1805 	u16 len;
1806 
1807 	mm_unaccount_pinned_pages(&uarg->mmp);
1808 
1809 	/* if !len, there was only 1 call, and it was aborted
1810 	 * so do not queue a completion notification
1811 	 */
1812 	if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1813 		goto release;
1814 
1815 	len = uarg->len;
1816 	lo = uarg->id;
1817 	hi = uarg->id + len - 1;
1818 	is_zerocopy = uarg->zerocopy;
1819 
1820 	serr = SKB_EXT_ERR(skb);
1821 	memset(serr, 0, sizeof(*serr));
1822 	serr->ee.ee_errno = 0;
1823 	serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1824 	serr->ee.ee_data = hi;
1825 	serr->ee.ee_info = lo;
1826 	if (!is_zerocopy)
1827 		serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1828 
1829 	q = &sk->sk_error_queue;
1830 	spin_lock_irqsave(&q->lock, flags);
1831 	tail = skb_peek_tail(q);
1832 	if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1833 	    !skb_zerocopy_notify_extend(tail, lo, len)) {
1834 		__skb_queue_tail(q, skb);
1835 		skb = NULL;
1836 	}
1837 	spin_unlock_irqrestore(&q->lock, flags);
1838 
1839 	sk_error_report(sk);
1840 
1841 release:
1842 	consume_skb(skb);
1843 	sock_put(sk);
1844 }
1845 
1846 static void msg_zerocopy_complete(struct sk_buff *skb, struct ubuf_info *uarg,
1847 				  bool success)
1848 {
1849 	struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1850 
1851 	uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1852 
1853 	if (refcount_dec_and_test(&uarg->refcnt))
1854 		__msg_zerocopy_callback(uarg_zc);
1855 }
1856 
1857 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1858 {
1859 	struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1860 
1861 	atomic_dec(&sk->sk_zckey);
1862 	uarg_to_msgzc(uarg)->len--;
1863 
1864 	if (have_uref)
1865 		msg_zerocopy_complete(NULL, uarg, true);
1866 }
1867 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1868 
1869 const struct ubuf_info_ops msg_zerocopy_ubuf_ops = {
1870 	.complete = msg_zerocopy_complete,
1871 };
1872 EXPORT_SYMBOL_GPL(msg_zerocopy_ubuf_ops);
1873 
1874 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1875 			     struct msghdr *msg, int len,
1876 			     struct ubuf_info *uarg)
1877 {
1878 	int err, orig_len = skb->len;
1879 
1880 	if (uarg->ops->link_skb) {
1881 		err = uarg->ops->link_skb(skb, uarg);
1882 		if (err)
1883 			return err;
1884 	} else {
1885 		struct ubuf_info *orig_uarg = skb_zcopy(skb);
1886 
1887 		/* An skb can only point to one uarg. This edge case happens
1888 		 * when TCP appends to an skb, but zerocopy_realloc triggered
1889 		 * a new alloc.
1890 		 */
1891 		if (orig_uarg && uarg != orig_uarg)
1892 			return -EEXIST;
1893 	}
1894 
1895 	err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1896 	if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1897 		struct sock *save_sk = skb->sk;
1898 
1899 		/* Streams do not free skb on error. Reset to prev state. */
1900 		iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1901 		skb->sk = sk;
1902 		___pskb_trim(skb, orig_len);
1903 		skb->sk = save_sk;
1904 		return err;
1905 	}
1906 
1907 	skb_zcopy_set(skb, uarg, NULL);
1908 	return skb->len - orig_len;
1909 }
1910 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1911 
1912 void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1913 {
1914 	int i;
1915 
1916 	skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1917 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1918 		skb_frag_ref(skb, i);
1919 }
1920 EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1921 
1922 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1923 			      gfp_t gfp_mask)
1924 {
1925 	if (skb_zcopy(orig)) {
1926 		if (skb_zcopy(nskb)) {
1927 			/* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1928 			if (!gfp_mask) {
1929 				WARN_ON_ONCE(1);
1930 				return -ENOMEM;
1931 			}
1932 			if (skb_uarg(nskb) == skb_uarg(orig))
1933 				return 0;
1934 			if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1935 				return -EIO;
1936 		}
1937 		skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1938 	}
1939 	return 0;
1940 }
1941 
1942 /**
1943  *	skb_copy_ubufs	-	copy userspace skb frags buffers to kernel
1944  *	@skb: the skb to modify
1945  *	@gfp_mask: allocation priority
1946  *
1947  *	This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1948  *	It will copy all frags into kernel and drop the reference
1949  *	to userspace pages.
1950  *
1951  *	If this function is called from an interrupt gfp_mask() must be
1952  *	%GFP_ATOMIC.
1953  *
1954  *	Returns 0 on success or a negative error code on failure
1955  *	to allocate kernel memory to copy to.
1956  */
1957 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1958 {
1959 	int num_frags = skb_shinfo(skb)->nr_frags;
1960 	struct page *page, *head = NULL;
1961 	int i, order, psize, new_frags;
1962 	u32 d_off;
1963 
1964 	if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1965 		return -EINVAL;
1966 
1967 	if (!num_frags)
1968 		goto release;
1969 
1970 	/* We might have to allocate high order pages, so compute what minimum
1971 	 * page order is needed.
1972 	 */
1973 	order = 0;
1974 	while ((PAGE_SIZE << order) * MAX_SKB_FRAGS < __skb_pagelen(skb))
1975 		order++;
1976 	psize = (PAGE_SIZE << order);
1977 
1978 	new_frags = (__skb_pagelen(skb) + psize - 1) >> (PAGE_SHIFT + order);
1979 	for (i = 0; i < new_frags; i++) {
1980 		page = alloc_pages(gfp_mask | __GFP_COMP, order);
1981 		if (!page) {
1982 			while (head) {
1983 				struct page *next = (struct page *)page_private(head);
1984 				put_page(head);
1985 				head = next;
1986 			}
1987 			return -ENOMEM;
1988 		}
1989 		set_page_private(page, (unsigned long)head);
1990 		head = page;
1991 	}
1992 
1993 	page = head;
1994 	d_off = 0;
1995 	for (i = 0; i < num_frags; i++) {
1996 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1997 		u32 p_off, p_len, copied;
1998 		struct page *p;
1999 		u8 *vaddr;
2000 
2001 		skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
2002 				      p, p_off, p_len, copied) {
2003 			u32 copy, done = 0;
2004 			vaddr = kmap_atomic(p);
2005 
2006 			while (done < p_len) {
2007 				if (d_off == psize) {
2008 					d_off = 0;
2009 					page = (struct page *)page_private(page);
2010 				}
2011 				copy = min_t(u32, psize - d_off, p_len - done);
2012 				memcpy(page_address(page) + d_off,
2013 				       vaddr + p_off + done, copy);
2014 				done += copy;
2015 				d_off += copy;
2016 			}
2017 			kunmap_atomic(vaddr);
2018 		}
2019 	}
2020 
2021 	/* skb frags release userspace buffers */
2022 	for (i = 0; i < num_frags; i++)
2023 		skb_frag_unref(skb, i);
2024 
2025 	/* skb frags point to kernel buffers */
2026 	for (i = 0; i < new_frags - 1; i++) {
2027 		__skb_fill_netmem_desc(skb, i, page_to_netmem(head), 0, psize);
2028 		head = (struct page *)page_private(head);
2029 	}
2030 	__skb_fill_netmem_desc(skb, new_frags - 1, page_to_netmem(head), 0,
2031 			       d_off);
2032 	skb_shinfo(skb)->nr_frags = new_frags;
2033 
2034 release:
2035 	skb_zcopy_clear(skb, false);
2036 	return 0;
2037 }
2038 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
2039 
2040 /**
2041  *	skb_clone	-	duplicate an sk_buff
2042  *	@skb: buffer to clone
2043  *	@gfp_mask: allocation priority
2044  *
2045  *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
2046  *	copies share the same packet data but not structure. The new
2047  *	buffer has a reference count of 1. If the allocation fails the
2048  *	function returns %NULL otherwise the new buffer is returned.
2049  *
2050  *	If this function is called from an interrupt gfp_mask() must be
2051  *	%GFP_ATOMIC.
2052  */
2053 
2054 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
2055 {
2056 	struct sk_buff_fclones *fclones = container_of(skb,
2057 						       struct sk_buff_fclones,
2058 						       skb1);
2059 	struct sk_buff *n;
2060 
2061 	if (skb_orphan_frags(skb, gfp_mask))
2062 		return NULL;
2063 
2064 	if (skb->fclone == SKB_FCLONE_ORIG &&
2065 	    refcount_read(&fclones->fclone_ref) == 1) {
2066 		n = &fclones->skb2;
2067 		refcount_set(&fclones->fclone_ref, 2);
2068 		n->fclone = SKB_FCLONE_CLONE;
2069 	} else {
2070 		if (skb_pfmemalloc(skb))
2071 			gfp_mask |= __GFP_MEMALLOC;
2072 
2073 		n = kmem_cache_alloc(net_hotdata.skbuff_cache, gfp_mask);
2074 		if (!n)
2075 			return NULL;
2076 
2077 		n->fclone = SKB_FCLONE_UNAVAILABLE;
2078 	}
2079 
2080 	return __skb_clone(n, skb);
2081 }
2082 EXPORT_SYMBOL(skb_clone);
2083 
2084 void skb_headers_offset_update(struct sk_buff *skb, int off)
2085 {
2086 	/* Only adjust this if it actually is csum_start rather than csum */
2087 	if (skb->ip_summed == CHECKSUM_PARTIAL)
2088 		skb->csum_start += off;
2089 	/* {transport,network,mac}_header and tail are relative to skb->head */
2090 	skb->transport_header += off;
2091 	skb->network_header   += off;
2092 	if (skb_mac_header_was_set(skb))
2093 		skb->mac_header += off;
2094 	skb->inner_transport_header += off;
2095 	skb->inner_network_header += off;
2096 	skb->inner_mac_header += off;
2097 }
2098 EXPORT_SYMBOL(skb_headers_offset_update);
2099 
2100 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
2101 {
2102 	__copy_skb_header(new, old);
2103 
2104 	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
2105 	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
2106 	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
2107 }
2108 EXPORT_SYMBOL(skb_copy_header);
2109 
2110 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
2111 {
2112 	if (skb_pfmemalloc(skb))
2113 		return SKB_ALLOC_RX;
2114 	return 0;
2115 }
2116 
2117 /**
2118  *	skb_copy	-	create private copy of an sk_buff
2119  *	@skb: buffer to copy
2120  *	@gfp_mask: allocation priority
2121  *
2122  *	Make a copy of both an &sk_buff and its data. This is used when the
2123  *	caller wishes to modify the data and needs a private copy of the
2124  *	data to alter. Returns %NULL on failure or the pointer to the buffer
2125  *	on success. The returned buffer has a reference count of 1.
2126  *
2127  *	As by-product this function converts non-linear &sk_buff to linear
2128  *	one, so that &sk_buff becomes completely private and caller is allowed
2129  *	to modify all the data of returned buffer. This means that this
2130  *	function is not recommended for use in circumstances when only
2131  *	header is going to be modified. Use pskb_copy() instead.
2132  */
2133 
2134 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
2135 {
2136 	struct sk_buff *n;
2137 	unsigned int size;
2138 	int headerlen;
2139 
2140 	if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2141 		return NULL;
2142 
2143 	headerlen = skb_headroom(skb);
2144 	size = skb_end_offset(skb) + skb->data_len;
2145 	n = __alloc_skb(size, gfp_mask,
2146 			skb_alloc_rx_flag(skb), NUMA_NO_NODE);
2147 	if (!n)
2148 		return NULL;
2149 
2150 	/* Set the data pointer */
2151 	skb_reserve(n, headerlen);
2152 	/* Set the tail pointer and length */
2153 	skb_put(n, skb->len);
2154 
2155 	BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
2156 
2157 	skb_copy_header(n, skb);
2158 	return n;
2159 }
2160 EXPORT_SYMBOL(skb_copy);
2161 
2162 /**
2163  *	__pskb_copy_fclone	-  create copy of an sk_buff with private head.
2164  *	@skb: buffer to copy
2165  *	@headroom: headroom of new skb
2166  *	@gfp_mask: allocation priority
2167  *	@fclone: if true allocate the copy of the skb from the fclone
2168  *	cache instead of the head cache; it is recommended to set this
2169  *	to true for the cases where the copy will likely be cloned
2170  *
2171  *	Make a copy of both an &sk_buff and part of its data, located
2172  *	in header. Fragmented data remain shared. This is used when
2173  *	the caller wishes to modify only header of &sk_buff and needs
2174  *	private copy of the header to alter. Returns %NULL on failure
2175  *	or the pointer to the buffer on success.
2176  *	The returned buffer has a reference count of 1.
2177  */
2178 
2179 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
2180 				   gfp_t gfp_mask, bool fclone)
2181 {
2182 	unsigned int size = skb_headlen(skb) + headroom;
2183 	int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
2184 	struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
2185 
2186 	if (!n)
2187 		goto out;
2188 
2189 	/* Set the data pointer */
2190 	skb_reserve(n, headroom);
2191 	/* Set the tail pointer and length */
2192 	skb_put(n, skb_headlen(skb));
2193 	/* Copy the bytes */
2194 	skb_copy_from_linear_data(skb, n->data, n->len);
2195 
2196 	n->truesize += skb->data_len;
2197 	n->data_len  = skb->data_len;
2198 	n->len	     = skb->len;
2199 
2200 	if (skb_shinfo(skb)->nr_frags) {
2201 		int i;
2202 
2203 		if (skb_orphan_frags(skb, gfp_mask) ||
2204 		    skb_zerocopy_clone(n, skb, gfp_mask)) {
2205 			kfree_skb(n);
2206 			n = NULL;
2207 			goto out;
2208 		}
2209 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2210 			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
2211 			skb_frag_ref(skb, i);
2212 		}
2213 		skb_shinfo(n)->nr_frags = i;
2214 	}
2215 
2216 	if (skb_has_frag_list(skb)) {
2217 		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
2218 		skb_clone_fraglist(n);
2219 	}
2220 
2221 	skb_copy_header(n, skb);
2222 out:
2223 	return n;
2224 }
2225 EXPORT_SYMBOL(__pskb_copy_fclone);
2226 
2227 /**
2228  *	pskb_expand_head - reallocate header of &sk_buff
2229  *	@skb: buffer to reallocate
2230  *	@nhead: room to add at head
2231  *	@ntail: room to add at tail
2232  *	@gfp_mask: allocation priority
2233  *
2234  *	Expands (or creates identical copy, if @nhead and @ntail are zero)
2235  *	header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
2236  *	reference count of 1. Returns zero in the case of success or error,
2237  *	if expansion failed. In the last case, &sk_buff is not changed.
2238  *
2239  *	All the pointers pointing into skb header may change and must be
2240  *	reloaded after call to this function.
2241  */
2242 
2243 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
2244 		     gfp_t gfp_mask)
2245 {
2246 	unsigned int osize = skb_end_offset(skb);
2247 	unsigned int size = osize + nhead + ntail;
2248 	long off;
2249 	u8 *data;
2250 	int i;
2251 
2252 	BUG_ON(nhead < 0);
2253 
2254 	BUG_ON(skb_shared(skb));
2255 
2256 	skb_zcopy_downgrade_managed(skb);
2257 
2258 	if (skb_pfmemalloc(skb))
2259 		gfp_mask |= __GFP_MEMALLOC;
2260 
2261 	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
2262 	if (!data)
2263 		goto nodata;
2264 	size = SKB_WITH_OVERHEAD(size);
2265 
2266 	/* Copy only real data... and, alas, header. This should be
2267 	 * optimized for the cases when header is void.
2268 	 */
2269 	memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
2270 
2271 	memcpy((struct skb_shared_info *)(data + size),
2272 	       skb_shinfo(skb),
2273 	       offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
2274 
2275 	/*
2276 	 * if shinfo is shared we must drop the old head gracefully, but if it
2277 	 * is not we can just drop the old head and let the existing refcount
2278 	 * be since all we did is relocate the values
2279 	 */
2280 	if (skb_cloned(skb)) {
2281 		if (skb_orphan_frags(skb, gfp_mask))
2282 			goto nofrags;
2283 		if (skb_zcopy(skb))
2284 			refcount_inc(&skb_uarg(skb)->refcnt);
2285 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2286 			skb_frag_ref(skb, i);
2287 
2288 		if (skb_has_frag_list(skb))
2289 			skb_clone_fraglist(skb);
2290 
2291 		skb_release_data(skb, SKB_CONSUMED);
2292 	} else {
2293 		skb_free_head(skb);
2294 	}
2295 	off = (data + nhead) - skb->head;
2296 
2297 	skb->head     = data;
2298 	skb->head_frag = 0;
2299 	skb->data    += off;
2300 
2301 	skb_set_end_offset(skb, size);
2302 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2303 	off           = nhead;
2304 #endif
2305 	skb->tail	      += off;
2306 	skb_headers_offset_update(skb, nhead);
2307 	skb->cloned   = 0;
2308 	skb->hdr_len  = 0;
2309 	skb->nohdr    = 0;
2310 	atomic_set(&skb_shinfo(skb)->dataref, 1);
2311 
2312 	skb_metadata_clear(skb);
2313 
2314 	/* It is not generally safe to change skb->truesize.
2315 	 * For the moment, we really care of rx path, or
2316 	 * when skb is orphaned (not attached to a socket).
2317 	 */
2318 	if (!skb->sk || skb->destructor == sock_edemux)
2319 		skb->truesize += size - osize;
2320 
2321 	return 0;
2322 
2323 nofrags:
2324 	skb_kfree_head(data, size);
2325 nodata:
2326 	return -ENOMEM;
2327 }
2328 EXPORT_SYMBOL(pskb_expand_head);
2329 
2330 /* Make private copy of skb with writable head and some headroom */
2331 
2332 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
2333 {
2334 	struct sk_buff *skb2;
2335 	int delta = headroom - skb_headroom(skb);
2336 
2337 	if (delta <= 0)
2338 		skb2 = pskb_copy(skb, GFP_ATOMIC);
2339 	else {
2340 		skb2 = skb_clone(skb, GFP_ATOMIC);
2341 		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
2342 					     GFP_ATOMIC)) {
2343 			kfree_skb(skb2);
2344 			skb2 = NULL;
2345 		}
2346 	}
2347 	return skb2;
2348 }
2349 EXPORT_SYMBOL(skb_realloc_headroom);
2350 
2351 /* Note: We plan to rework this in linux-6.4 */
2352 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2353 {
2354 	unsigned int saved_end_offset, saved_truesize;
2355 	struct skb_shared_info *shinfo;
2356 	int res;
2357 
2358 	saved_end_offset = skb_end_offset(skb);
2359 	saved_truesize = skb->truesize;
2360 
2361 	res = pskb_expand_head(skb, 0, 0, pri);
2362 	if (res)
2363 		return res;
2364 
2365 	skb->truesize = saved_truesize;
2366 
2367 	if (likely(skb_end_offset(skb) == saved_end_offset))
2368 		return 0;
2369 
2370 	/* We can not change skb->end if the original or new value
2371 	 * is SKB_SMALL_HEAD_HEADROOM, as it might break skb_kfree_head().
2372 	 */
2373 	if (saved_end_offset == SKB_SMALL_HEAD_HEADROOM ||
2374 	    skb_end_offset(skb) == SKB_SMALL_HEAD_HEADROOM) {
2375 		/* We think this path should not be taken.
2376 		 * Add a temporary trace to warn us just in case.
2377 		 */
2378 		pr_err_once("__skb_unclone_keeptruesize() skb_end_offset() %u -> %u\n",
2379 			    saved_end_offset, skb_end_offset(skb));
2380 		WARN_ON_ONCE(1);
2381 		return 0;
2382 	}
2383 
2384 	shinfo = skb_shinfo(skb);
2385 
2386 	/* We are about to change back skb->end,
2387 	 * we need to move skb_shinfo() to its new location.
2388 	 */
2389 	memmove(skb->head + saved_end_offset,
2390 		shinfo,
2391 		offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2392 
2393 	skb_set_end_offset(skb, saved_end_offset);
2394 
2395 	return 0;
2396 }
2397 
2398 /**
2399  *	skb_expand_head - reallocate header of &sk_buff
2400  *	@skb: buffer to reallocate
2401  *	@headroom: needed headroom
2402  *
2403  *	Unlike skb_realloc_headroom, this one does not allocate a new skb
2404  *	if possible; copies skb->sk to new skb as needed
2405  *	and frees original skb in case of failures.
2406  *
2407  *	It expect increased headroom and generates warning otherwise.
2408  */
2409 
2410 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2411 {
2412 	int delta = headroom - skb_headroom(skb);
2413 	int osize = skb_end_offset(skb);
2414 	struct sock *sk = skb->sk;
2415 
2416 	if (WARN_ONCE(delta <= 0,
2417 		      "%s is expecting an increase in the headroom", __func__))
2418 		return skb;
2419 
2420 	delta = SKB_DATA_ALIGN(delta);
2421 	/* pskb_expand_head() might crash, if skb is shared. */
2422 	if (skb_shared(skb) || !is_skb_wmem(skb)) {
2423 		struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2424 
2425 		if (unlikely(!nskb))
2426 			goto fail;
2427 
2428 		if (sk)
2429 			skb_set_owner_w(nskb, sk);
2430 		consume_skb(skb);
2431 		skb = nskb;
2432 	}
2433 	if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2434 		goto fail;
2435 
2436 	if (sk && is_skb_wmem(skb)) {
2437 		delta = skb_end_offset(skb) - osize;
2438 		refcount_add(delta, &sk->sk_wmem_alloc);
2439 		skb->truesize += delta;
2440 	}
2441 	return skb;
2442 
2443 fail:
2444 	kfree_skb(skb);
2445 	return NULL;
2446 }
2447 EXPORT_SYMBOL(skb_expand_head);
2448 
2449 /**
2450  *	skb_copy_expand	-	copy and expand sk_buff
2451  *	@skb: buffer to copy
2452  *	@newheadroom: new free bytes at head
2453  *	@newtailroom: new free bytes at tail
2454  *	@gfp_mask: allocation priority
2455  *
2456  *	Make a copy of both an &sk_buff and its data and while doing so
2457  *	allocate additional space.
2458  *
2459  *	This is used when the caller wishes to modify the data and needs a
2460  *	private copy of the data to alter as well as more space for new fields.
2461  *	Returns %NULL on failure or the pointer to the buffer
2462  *	on success. The returned buffer has a reference count of 1.
2463  *
2464  *	You must pass %GFP_ATOMIC as the allocation priority if this function
2465  *	is called from an interrupt.
2466  */
2467 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2468 				int newheadroom, int newtailroom,
2469 				gfp_t gfp_mask)
2470 {
2471 	/*
2472 	 *	Allocate the copy buffer
2473 	 */
2474 	int head_copy_len, head_copy_off;
2475 	struct sk_buff *n;
2476 	int oldheadroom;
2477 
2478 	if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2479 		return NULL;
2480 
2481 	oldheadroom = skb_headroom(skb);
2482 	n = __alloc_skb(newheadroom + skb->len + newtailroom,
2483 			gfp_mask, skb_alloc_rx_flag(skb),
2484 			NUMA_NO_NODE);
2485 	if (!n)
2486 		return NULL;
2487 
2488 	skb_reserve(n, newheadroom);
2489 
2490 	/* Set the tail pointer and length */
2491 	skb_put(n, skb->len);
2492 
2493 	head_copy_len = oldheadroom;
2494 	head_copy_off = 0;
2495 	if (newheadroom <= head_copy_len)
2496 		head_copy_len = newheadroom;
2497 	else
2498 		head_copy_off = newheadroom - head_copy_len;
2499 
2500 	/* Copy the linear header and data. */
2501 	BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2502 			     skb->len + head_copy_len));
2503 
2504 	skb_copy_header(n, skb);
2505 
2506 	skb_headers_offset_update(n, newheadroom - oldheadroom);
2507 
2508 	return n;
2509 }
2510 EXPORT_SYMBOL(skb_copy_expand);
2511 
2512 /**
2513  *	__skb_pad		-	zero pad the tail of an skb
2514  *	@skb: buffer to pad
2515  *	@pad: space to pad
2516  *	@free_on_error: free buffer on error
2517  *
2518  *	Ensure that a buffer is followed by a padding area that is zero
2519  *	filled. Used by network drivers which may DMA or transfer data
2520  *	beyond the buffer end onto the wire.
2521  *
2522  *	May return error in out of memory cases. The skb is freed on error
2523  *	if @free_on_error is true.
2524  */
2525 
2526 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2527 {
2528 	int err;
2529 	int ntail;
2530 
2531 	/* If the skbuff is non linear tailroom is always zero.. */
2532 	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2533 		memset(skb->data+skb->len, 0, pad);
2534 		return 0;
2535 	}
2536 
2537 	ntail = skb->data_len + pad - (skb->end - skb->tail);
2538 	if (likely(skb_cloned(skb) || ntail > 0)) {
2539 		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2540 		if (unlikely(err))
2541 			goto free_skb;
2542 	}
2543 
2544 	/* FIXME: The use of this function with non-linear skb's really needs
2545 	 * to be audited.
2546 	 */
2547 	err = skb_linearize(skb);
2548 	if (unlikely(err))
2549 		goto free_skb;
2550 
2551 	memset(skb->data + skb->len, 0, pad);
2552 	return 0;
2553 
2554 free_skb:
2555 	if (free_on_error)
2556 		kfree_skb(skb);
2557 	return err;
2558 }
2559 EXPORT_SYMBOL(__skb_pad);
2560 
2561 /**
2562  *	pskb_put - add data to the tail of a potentially fragmented buffer
2563  *	@skb: start of the buffer to use
2564  *	@tail: tail fragment of the buffer to use
2565  *	@len: amount of data to add
2566  *
2567  *	This function extends the used data area of the potentially
2568  *	fragmented buffer. @tail must be the last fragment of @skb -- or
2569  *	@skb itself. If this would exceed the total buffer size the kernel
2570  *	will panic. A pointer to the first byte of the extra data is
2571  *	returned.
2572  */
2573 
2574 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2575 {
2576 	if (tail != skb) {
2577 		skb->data_len += len;
2578 		skb->len += len;
2579 	}
2580 	return skb_put(tail, len);
2581 }
2582 EXPORT_SYMBOL_GPL(pskb_put);
2583 
2584 /**
2585  *	skb_put - add data to a buffer
2586  *	@skb: buffer to use
2587  *	@len: amount of data to add
2588  *
2589  *	This function extends the used data area of the buffer. If this would
2590  *	exceed the total buffer size the kernel will panic. A pointer to the
2591  *	first byte of the extra data is returned.
2592  */
2593 void *skb_put(struct sk_buff *skb, unsigned int len)
2594 {
2595 	void *tmp = skb_tail_pointer(skb);
2596 	SKB_LINEAR_ASSERT(skb);
2597 	skb->tail += len;
2598 	skb->len  += len;
2599 	if (unlikely(skb->tail > skb->end))
2600 		skb_over_panic(skb, len, __builtin_return_address(0));
2601 	return tmp;
2602 }
2603 EXPORT_SYMBOL(skb_put);
2604 
2605 /**
2606  *	skb_push - add data to the start of a buffer
2607  *	@skb: buffer to use
2608  *	@len: amount of data to add
2609  *
2610  *	This function extends the used data area of the buffer at the buffer
2611  *	start. If this would exceed the total buffer headroom the kernel will
2612  *	panic. A pointer to the first byte of the extra data is returned.
2613  */
2614 void *skb_push(struct sk_buff *skb, unsigned int len)
2615 {
2616 	skb->data -= len;
2617 	skb->len  += len;
2618 	if (unlikely(skb->data < skb->head))
2619 		skb_under_panic(skb, len, __builtin_return_address(0));
2620 	return skb->data;
2621 }
2622 EXPORT_SYMBOL(skb_push);
2623 
2624 /**
2625  *	skb_pull - remove data from the start of a buffer
2626  *	@skb: buffer to use
2627  *	@len: amount of data to remove
2628  *
2629  *	This function removes data from the start of a buffer, returning
2630  *	the memory to the headroom. A pointer to the next data in the buffer
2631  *	is returned. Once the data has been pulled future pushes will overwrite
2632  *	the old data.
2633  */
2634 void *skb_pull(struct sk_buff *skb, unsigned int len)
2635 {
2636 	return skb_pull_inline(skb, len);
2637 }
2638 EXPORT_SYMBOL(skb_pull);
2639 
2640 /**
2641  *	skb_pull_data - remove data from the start of a buffer returning its
2642  *	original position.
2643  *	@skb: buffer to use
2644  *	@len: amount of data to remove
2645  *
2646  *	This function removes data from the start of a buffer, returning
2647  *	the memory to the headroom. A pointer to the original data in the buffer
2648  *	is returned after checking if there is enough data to pull. Once the
2649  *	data has been pulled future pushes will overwrite the old data.
2650  */
2651 void *skb_pull_data(struct sk_buff *skb, size_t len)
2652 {
2653 	void *data = skb->data;
2654 
2655 	if (skb->len < len)
2656 		return NULL;
2657 
2658 	skb_pull(skb, len);
2659 
2660 	return data;
2661 }
2662 EXPORT_SYMBOL(skb_pull_data);
2663 
2664 /**
2665  *	skb_trim - remove end from a buffer
2666  *	@skb: buffer to alter
2667  *	@len: new length
2668  *
2669  *	Cut the length of a buffer down by removing data from the tail. If
2670  *	the buffer is already under the length specified it is not modified.
2671  *	The skb must be linear.
2672  */
2673 void skb_trim(struct sk_buff *skb, unsigned int len)
2674 {
2675 	if (skb->len > len)
2676 		__skb_trim(skb, len);
2677 }
2678 EXPORT_SYMBOL(skb_trim);
2679 
2680 /* Trims skb to length len. It can change skb pointers.
2681  */
2682 
2683 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2684 {
2685 	struct sk_buff **fragp;
2686 	struct sk_buff *frag;
2687 	int offset = skb_headlen(skb);
2688 	int nfrags = skb_shinfo(skb)->nr_frags;
2689 	int i;
2690 	int err;
2691 
2692 	if (skb_cloned(skb) &&
2693 	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2694 		return err;
2695 
2696 	i = 0;
2697 	if (offset >= len)
2698 		goto drop_pages;
2699 
2700 	for (; i < nfrags; i++) {
2701 		int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2702 
2703 		if (end < len) {
2704 			offset = end;
2705 			continue;
2706 		}
2707 
2708 		skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2709 
2710 drop_pages:
2711 		skb_shinfo(skb)->nr_frags = i;
2712 
2713 		for (; i < nfrags; i++)
2714 			skb_frag_unref(skb, i);
2715 
2716 		if (skb_has_frag_list(skb))
2717 			skb_drop_fraglist(skb);
2718 		goto done;
2719 	}
2720 
2721 	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2722 	     fragp = &frag->next) {
2723 		int end = offset + frag->len;
2724 
2725 		if (skb_shared(frag)) {
2726 			struct sk_buff *nfrag;
2727 
2728 			nfrag = skb_clone(frag, GFP_ATOMIC);
2729 			if (unlikely(!nfrag))
2730 				return -ENOMEM;
2731 
2732 			nfrag->next = frag->next;
2733 			consume_skb(frag);
2734 			frag = nfrag;
2735 			*fragp = frag;
2736 		}
2737 
2738 		if (end < len) {
2739 			offset = end;
2740 			continue;
2741 		}
2742 
2743 		if (end > len &&
2744 		    unlikely((err = pskb_trim(frag, len - offset))))
2745 			return err;
2746 
2747 		if (frag->next)
2748 			skb_drop_list(&frag->next);
2749 		break;
2750 	}
2751 
2752 done:
2753 	if (len > skb_headlen(skb)) {
2754 		skb->data_len -= skb->len - len;
2755 		skb->len       = len;
2756 	} else {
2757 		skb->len       = len;
2758 		skb->data_len  = 0;
2759 		skb_set_tail_pointer(skb, len);
2760 	}
2761 
2762 	if (!skb->sk || skb->destructor == sock_edemux)
2763 		skb_condense(skb);
2764 	return 0;
2765 }
2766 EXPORT_SYMBOL(___pskb_trim);
2767 
2768 /* Note : use pskb_trim_rcsum() instead of calling this directly
2769  */
2770 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2771 {
2772 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
2773 		int delta = skb->len - len;
2774 
2775 		skb->csum = csum_block_sub(skb->csum,
2776 					   skb_checksum(skb, len, delta, 0),
2777 					   len);
2778 	} else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2779 		int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2780 		int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2781 
2782 		if (offset + sizeof(__sum16) > hdlen)
2783 			return -EINVAL;
2784 	}
2785 	return __pskb_trim(skb, len);
2786 }
2787 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2788 
2789 /**
2790  *	__pskb_pull_tail - advance tail of skb header
2791  *	@skb: buffer to reallocate
2792  *	@delta: number of bytes to advance tail
2793  *
2794  *	The function makes a sense only on a fragmented &sk_buff,
2795  *	it expands header moving its tail forward and copying necessary
2796  *	data from fragmented part.
2797  *
2798  *	&sk_buff MUST have reference count of 1.
2799  *
2800  *	Returns %NULL (and &sk_buff does not change) if pull failed
2801  *	or value of new tail of skb in the case of success.
2802  *
2803  *	All the pointers pointing into skb header may change and must be
2804  *	reloaded after call to this function.
2805  */
2806 
2807 /* Moves tail of skb head forward, copying data from fragmented part,
2808  * when it is necessary.
2809  * 1. It may fail due to malloc failure.
2810  * 2. It may change skb pointers.
2811  *
2812  * It is pretty complicated. Luckily, it is called only in exceptional cases.
2813  */
2814 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2815 {
2816 	/* If skb has not enough free space at tail, get new one
2817 	 * plus 128 bytes for future expansions. If we have enough
2818 	 * room at tail, reallocate without expansion only if skb is cloned.
2819 	 */
2820 	int i, k, eat = (skb->tail + delta) - skb->end;
2821 
2822 	if (eat > 0 || skb_cloned(skb)) {
2823 		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2824 				     GFP_ATOMIC))
2825 			return NULL;
2826 	}
2827 
2828 	BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2829 			     skb_tail_pointer(skb), delta));
2830 
2831 	/* Optimization: no fragments, no reasons to preestimate
2832 	 * size of pulled pages. Superb.
2833 	 */
2834 	if (!skb_has_frag_list(skb))
2835 		goto pull_pages;
2836 
2837 	/* Estimate size of pulled pages. */
2838 	eat = delta;
2839 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2840 		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2841 
2842 		if (size >= eat)
2843 			goto pull_pages;
2844 		eat -= size;
2845 	}
2846 
2847 	/* If we need update frag list, we are in troubles.
2848 	 * Certainly, it is possible to add an offset to skb data,
2849 	 * but taking into account that pulling is expected to
2850 	 * be very rare operation, it is worth to fight against
2851 	 * further bloating skb head and crucify ourselves here instead.
2852 	 * Pure masohism, indeed. 8)8)
2853 	 */
2854 	if (eat) {
2855 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2856 		struct sk_buff *clone = NULL;
2857 		struct sk_buff *insp = NULL;
2858 
2859 		do {
2860 			if (list->len <= eat) {
2861 				/* Eaten as whole. */
2862 				eat -= list->len;
2863 				list = list->next;
2864 				insp = list;
2865 			} else {
2866 				/* Eaten partially. */
2867 				if (skb_is_gso(skb) && !list->head_frag &&
2868 				    skb_headlen(list))
2869 					skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2870 
2871 				if (skb_shared(list)) {
2872 					/* Sucks! We need to fork list. :-( */
2873 					clone = skb_clone(list, GFP_ATOMIC);
2874 					if (!clone)
2875 						return NULL;
2876 					insp = list->next;
2877 					list = clone;
2878 				} else {
2879 					/* This may be pulled without
2880 					 * problems. */
2881 					insp = list;
2882 				}
2883 				if (!pskb_pull(list, eat)) {
2884 					kfree_skb(clone);
2885 					return NULL;
2886 				}
2887 				break;
2888 			}
2889 		} while (eat);
2890 
2891 		/* Free pulled out fragments. */
2892 		while ((list = skb_shinfo(skb)->frag_list) != insp) {
2893 			skb_shinfo(skb)->frag_list = list->next;
2894 			consume_skb(list);
2895 		}
2896 		/* And insert new clone at head. */
2897 		if (clone) {
2898 			clone->next = list;
2899 			skb_shinfo(skb)->frag_list = clone;
2900 		}
2901 	}
2902 	/* Success! Now we may commit changes to skb data. */
2903 
2904 pull_pages:
2905 	eat = delta;
2906 	k = 0;
2907 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2908 		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2909 
2910 		if (size <= eat) {
2911 			skb_frag_unref(skb, i);
2912 			eat -= size;
2913 		} else {
2914 			skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2915 
2916 			*frag = skb_shinfo(skb)->frags[i];
2917 			if (eat) {
2918 				skb_frag_off_add(frag, eat);
2919 				skb_frag_size_sub(frag, eat);
2920 				if (!i)
2921 					goto end;
2922 				eat = 0;
2923 			}
2924 			k++;
2925 		}
2926 	}
2927 	skb_shinfo(skb)->nr_frags = k;
2928 
2929 end:
2930 	skb->tail     += delta;
2931 	skb->data_len -= delta;
2932 
2933 	if (!skb->data_len)
2934 		skb_zcopy_clear(skb, false);
2935 
2936 	return skb_tail_pointer(skb);
2937 }
2938 EXPORT_SYMBOL(__pskb_pull_tail);
2939 
2940 /**
2941  *	skb_copy_bits - copy bits from skb to kernel buffer
2942  *	@skb: source skb
2943  *	@offset: offset in source
2944  *	@to: destination buffer
2945  *	@len: number of bytes to copy
2946  *
2947  *	Copy the specified number of bytes from the source skb to the
2948  *	destination buffer.
2949  *
2950  *	CAUTION ! :
2951  *		If its prototype is ever changed,
2952  *		check arch/{*}/net/{*}.S files,
2953  *		since it is called from BPF assembly code.
2954  */
2955 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2956 {
2957 	int start = skb_headlen(skb);
2958 	struct sk_buff *frag_iter;
2959 	int i, copy;
2960 
2961 	if (offset > (int)skb->len - len)
2962 		goto fault;
2963 
2964 	/* Copy header. */
2965 	if ((copy = start - offset) > 0) {
2966 		if (copy > len)
2967 			copy = len;
2968 		skb_copy_from_linear_data_offset(skb, offset, to, copy);
2969 		if ((len -= copy) == 0)
2970 			return 0;
2971 		offset += copy;
2972 		to     += copy;
2973 	}
2974 
2975 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2976 		int end;
2977 		skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2978 
2979 		WARN_ON(start > offset + len);
2980 
2981 		end = start + skb_frag_size(f);
2982 		if ((copy = end - offset) > 0) {
2983 			u32 p_off, p_len, copied;
2984 			struct page *p;
2985 			u8 *vaddr;
2986 
2987 			if (copy > len)
2988 				copy = len;
2989 
2990 			skb_frag_foreach_page(f,
2991 					      skb_frag_off(f) + offset - start,
2992 					      copy, p, p_off, p_len, copied) {
2993 				vaddr = kmap_atomic(p);
2994 				memcpy(to + copied, vaddr + p_off, p_len);
2995 				kunmap_atomic(vaddr);
2996 			}
2997 
2998 			if ((len -= copy) == 0)
2999 				return 0;
3000 			offset += copy;
3001 			to     += copy;
3002 		}
3003 		start = end;
3004 	}
3005 
3006 	skb_walk_frags(skb, frag_iter) {
3007 		int end;
3008 
3009 		WARN_ON(start > offset + len);
3010 
3011 		end = start + frag_iter->len;
3012 		if ((copy = end - offset) > 0) {
3013 			if (copy > len)
3014 				copy = len;
3015 			if (skb_copy_bits(frag_iter, offset - start, to, copy))
3016 				goto fault;
3017 			if ((len -= copy) == 0)
3018 				return 0;
3019 			offset += copy;
3020 			to     += copy;
3021 		}
3022 		start = end;
3023 	}
3024 
3025 	if (!len)
3026 		return 0;
3027 
3028 fault:
3029 	return -EFAULT;
3030 }
3031 EXPORT_SYMBOL(skb_copy_bits);
3032 
3033 /*
3034  * Callback from splice_to_pipe(), if we need to release some pages
3035  * at the end of the spd in case we error'ed out in filling the pipe.
3036  */
3037 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3038 {
3039 	put_page(spd->pages[i]);
3040 }
3041 
3042 static struct page *linear_to_page(struct page *page, unsigned int *len,
3043 				   unsigned int *offset,
3044 				   struct sock *sk)
3045 {
3046 	struct page_frag *pfrag = sk_page_frag(sk);
3047 
3048 	if (!sk_page_frag_refill(sk, pfrag))
3049 		return NULL;
3050 
3051 	*len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
3052 
3053 	memcpy(page_address(pfrag->page) + pfrag->offset,
3054 	       page_address(page) + *offset, *len);
3055 	*offset = pfrag->offset;
3056 	pfrag->offset += *len;
3057 
3058 	return pfrag->page;
3059 }
3060 
3061 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
3062 			     struct page *page,
3063 			     unsigned int offset)
3064 {
3065 	return	spd->nr_pages &&
3066 		spd->pages[spd->nr_pages - 1] == page &&
3067 		(spd->partial[spd->nr_pages - 1].offset +
3068 		 spd->partial[spd->nr_pages - 1].len == offset);
3069 }
3070 
3071 /*
3072  * Fill page/offset/length into spd, if it can hold more pages.
3073  */
3074 static bool spd_fill_page(struct splice_pipe_desc *spd,
3075 			  struct pipe_inode_info *pipe, struct page *page,
3076 			  unsigned int *len, unsigned int offset,
3077 			  bool linear,
3078 			  struct sock *sk)
3079 {
3080 	if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
3081 		return true;
3082 
3083 	if (linear) {
3084 		page = linear_to_page(page, len, &offset, sk);
3085 		if (!page)
3086 			return true;
3087 	}
3088 	if (spd_can_coalesce(spd, page, offset)) {
3089 		spd->partial[spd->nr_pages - 1].len += *len;
3090 		return false;
3091 	}
3092 	get_page(page);
3093 	spd->pages[spd->nr_pages] = page;
3094 	spd->partial[spd->nr_pages].len = *len;
3095 	spd->partial[spd->nr_pages].offset = offset;
3096 	spd->nr_pages++;
3097 
3098 	return false;
3099 }
3100 
3101 static bool __splice_segment(struct page *page, unsigned int poff,
3102 			     unsigned int plen, unsigned int *off,
3103 			     unsigned int *len,
3104 			     struct splice_pipe_desc *spd, bool linear,
3105 			     struct sock *sk,
3106 			     struct pipe_inode_info *pipe)
3107 {
3108 	if (!*len)
3109 		return true;
3110 
3111 	/* skip this segment if already processed */
3112 	if (*off >= plen) {
3113 		*off -= plen;
3114 		return false;
3115 	}
3116 
3117 	/* ignore any bits we already processed */
3118 	poff += *off;
3119 	plen -= *off;
3120 	*off = 0;
3121 
3122 	do {
3123 		unsigned int flen = min(*len, plen);
3124 
3125 		if (spd_fill_page(spd, pipe, page, &flen, poff,
3126 				  linear, sk))
3127 			return true;
3128 		poff += flen;
3129 		plen -= flen;
3130 		*len -= flen;
3131 	} while (*len && plen);
3132 
3133 	return false;
3134 }
3135 
3136 /*
3137  * Map linear and fragment data from the skb to spd. It reports true if the
3138  * pipe is full or if we already spliced the requested length.
3139  */
3140 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
3141 			      unsigned int *offset, unsigned int *len,
3142 			      struct splice_pipe_desc *spd, struct sock *sk)
3143 {
3144 	int seg;
3145 	struct sk_buff *iter;
3146 
3147 	/* map the linear part :
3148 	 * If skb->head_frag is set, this 'linear' part is backed by a
3149 	 * fragment, and if the head is not shared with any clones then
3150 	 * we can avoid a copy since we own the head portion of this page.
3151 	 */
3152 	if (__splice_segment(virt_to_page(skb->data),
3153 			     (unsigned long) skb->data & (PAGE_SIZE - 1),
3154 			     skb_headlen(skb),
3155 			     offset, len, spd,
3156 			     skb_head_is_locked(skb),
3157 			     sk, pipe))
3158 		return true;
3159 
3160 	/*
3161 	 * then map the fragments
3162 	 */
3163 	for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
3164 		const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
3165 
3166 		if (__splice_segment(skb_frag_page(f),
3167 				     skb_frag_off(f), skb_frag_size(f),
3168 				     offset, len, spd, false, sk, pipe))
3169 			return true;
3170 	}
3171 
3172 	skb_walk_frags(skb, iter) {
3173 		if (*offset >= iter->len) {
3174 			*offset -= iter->len;
3175 			continue;
3176 		}
3177 		/* __skb_splice_bits() only fails if the output has no room
3178 		 * left, so no point in going over the frag_list for the error
3179 		 * case.
3180 		 */
3181 		if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
3182 			return true;
3183 	}
3184 
3185 	return false;
3186 }
3187 
3188 /*
3189  * Map data from the skb to a pipe. Should handle both the linear part,
3190  * the fragments, and the frag list.
3191  */
3192 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3193 		    struct pipe_inode_info *pipe, unsigned int tlen,
3194 		    unsigned int flags)
3195 {
3196 	struct partial_page partial[MAX_SKB_FRAGS];
3197 	struct page *pages[MAX_SKB_FRAGS];
3198 	struct splice_pipe_desc spd = {
3199 		.pages = pages,
3200 		.partial = partial,
3201 		.nr_pages_max = MAX_SKB_FRAGS,
3202 		.ops = &nosteal_pipe_buf_ops,
3203 		.spd_release = sock_spd_release,
3204 	};
3205 	int ret = 0;
3206 
3207 	__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
3208 
3209 	if (spd.nr_pages)
3210 		ret = splice_to_pipe(pipe, &spd);
3211 
3212 	return ret;
3213 }
3214 EXPORT_SYMBOL_GPL(skb_splice_bits);
3215 
3216 static int sendmsg_locked(struct sock *sk, struct msghdr *msg)
3217 {
3218 	struct socket *sock = sk->sk_socket;
3219 	size_t size = msg_data_left(msg);
3220 
3221 	if (!sock)
3222 		return -EINVAL;
3223 
3224 	if (!sock->ops->sendmsg_locked)
3225 		return sock_no_sendmsg_locked(sk, msg, size);
3226 
3227 	return sock->ops->sendmsg_locked(sk, msg, size);
3228 }
3229 
3230 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg)
3231 {
3232 	struct socket *sock = sk->sk_socket;
3233 
3234 	if (!sock)
3235 		return -EINVAL;
3236 	return sock_sendmsg(sock, msg);
3237 }
3238 
3239 typedef int (*sendmsg_func)(struct sock *sk, struct msghdr *msg);
3240 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
3241 			   int len, sendmsg_func sendmsg)
3242 {
3243 	unsigned int orig_len = len;
3244 	struct sk_buff *head = skb;
3245 	unsigned short fragidx;
3246 	int slen, ret;
3247 
3248 do_frag_list:
3249 
3250 	/* Deal with head data */
3251 	while (offset < skb_headlen(skb) && len) {
3252 		struct kvec kv;
3253 		struct msghdr msg;
3254 
3255 		slen = min_t(int, len, skb_headlen(skb) - offset);
3256 		kv.iov_base = skb->data + offset;
3257 		kv.iov_len = slen;
3258 		memset(&msg, 0, sizeof(msg));
3259 		msg.msg_flags = MSG_DONTWAIT;
3260 
3261 		iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &kv, 1, slen);
3262 		ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3263 				      sendmsg_unlocked, sk, &msg);
3264 		if (ret <= 0)
3265 			goto error;
3266 
3267 		offset += ret;
3268 		len -= ret;
3269 	}
3270 
3271 	/* All the data was skb head? */
3272 	if (!len)
3273 		goto out;
3274 
3275 	/* Make offset relative to start of frags */
3276 	offset -= skb_headlen(skb);
3277 
3278 	/* Find where we are in frag list */
3279 	for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3280 		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3281 
3282 		if (offset < skb_frag_size(frag))
3283 			break;
3284 
3285 		offset -= skb_frag_size(frag);
3286 	}
3287 
3288 	for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3289 		skb_frag_t *frag  = &skb_shinfo(skb)->frags[fragidx];
3290 
3291 		slen = min_t(size_t, len, skb_frag_size(frag) - offset);
3292 
3293 		while (slen) {
3294 			struct bio_vec bvec;
3295 			struct msghdr msg = {
3296 				.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT,
3297 			};
3298 
3299 			bvec_set_page(&bvec, skb_frag_page(frag), slen,
3300 				      skb_frag_off(frag) + offset);
3301 			iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1,
3302 				      slen);
3303 
3304 			ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3305 					      sendmsg_unlocked, sk, &msg);
3306 			if (ret <= 0)
3307 				goto error;
3308 
3309 			len -= ret;
3310 			offset += ret;
3311 			slen -= ret;
3312 		}
3313 
3314 		offset = 0;
3315 	}
3316 
3317 	if (len) {
3318 		/* Process any frag lists */
3319 
3320 		if (skb == head) {
3321 			if (skb_has_frag_list(skb)) {
3322 				skb = skb_shinfo(skb)->frag_list;
3323 				goto do_frag_list;
3324 			}
3325 		} else if (skb->next) {
3326 			skb = skb->next;
3327 			goto do_frag_list;
3328 		}
3329 	}
3330 
3331 out:
3332 	return orig_len - len;
3333 
3334 error:
3335 	return orig_len == len ? ret : orig_len - len;
3336 }
3337 
3338 /* Send skb data on a socket. Socket must be locked. */
3339 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3340 			 int len)
3341 {
3342 	return __skb_send_sock(sk, skb, offset, len, sendmsg_locked);
3343 }
3344 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
3345 
3346 /* Send skb data on a socket. Socket must be unlocked. */
3347 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
3348 {
3349 	return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked);
3350 }
3351 
3352 /**
3353  *	skb_store_bits - store bits from kernel buffer to skb
3354  *	@skb: destination buffer
3355  *	@offset: offset in destination
3356  *	@from: source buffer
3357  *	@len: number of bytes to copy
3358  *
3359  *	Copy the specified number of bytes from the source buffer to the
3360  *	destination skb.  This function handles all the messy bits of
3361  *	traversing fragment lists and such.
3362  */
3363 
3364 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
3365 {
3366 	int start = skb_headlen(skb);
3367 	struct sk_buff *frag_iter;
3368 	int i, copy;
3369 
3370 	if (offset > (int)skb->len - len)
3371 		goto fault;
3372 
3373 	if ((copy = start - offset) > 0) {
3374 		if (copy > len)
3375 			copy = len;
3376 		skb_copy_to_linear_data_offset(skb, offset, from, copy);
3377 		if ((len -= copy) == 0)
3378 			return 0;
3379 		offset += copy;
3380 		from += copy;
3381 	}
3382 
3383 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3384 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3385 		int end;
3386 
3387 		WARN_ON(start > offset + len);
3388 
3389 		end = start + skb_frag_size(frag);
3390 		if ((copy = end - offset) > 0) {
3391 			u32 p_off, p_len, copied;
3392 			struct page *p;
3393 			u8 *vaddr;
3394 
3395 			if (copy > len)
3396 				copy = len;
3397 
3398 			skb_frag_foreach_page(frag,
3399 					      skb_frag_off(frag) + offset - start,
3400 					      copy, p, p_off, p_len, copied) {
3401 				vaddr = kmap_atomic(p);
3402 				memcpy(vaddr + p_off, from + copied, p_len);
3403 				kunmap_atomic(vaddr);
3404 			}
3405 
3406 			if ((len -= copy) == 0)
3407 				return 0;
3408 			offset += copy;
3409 			from += copy;
3410 		}
3411 		start = end;
3412 	}
3413 
3414 	skb_walk_frags(skb, frag_iter) {
3415 		int end;
3416 
3417 		WARN_ON(start > offset + len);
3418 
3419 		end = start + frag_iter->len;
3420 		if ((copy = end - offset) > 0) {
3421 			if (copy > len)
3422 				copy = len;
3423 			if (skb_store_bits(frag_iter, offset - start,
3424 					   from, copy))
3425 				goto fault;
3426 			if ((len -= copy) == 0)
3427 				return 0;
3428 			offset += copy;
3429 			from += copy;
3430 		}
3431 		start = end;
3432 	}
3433 	if (!len)
3434 		return 0;
3435 
3436 fault:
3437 	return -EFAULT;
3438 }
3439 EXPORT_SYMBOL(skb_store_bits);
3440 
3441 /* Checksum skb data. */
3442 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3443 		      __wsum csum, const struct skb_checksum_ops *ops)
3444 {
3445 	int start = skb_headlen(skb);
3446 	int i, copy = start - offset;
3447 	struct sk_buff *frag_iter;
3448 	int pos = 0;
3449 
3450 	/* Checksum header. */
3451 	if (copy > 0) {
3452 		if (copy > len)
3453 			copy = len;
3454 		csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3455 				       skb->data + offset, copy, csum);
3456 		if ((len -= copy) == 0)
3457 			return csum;
3458 		offset += copy;
3459 		pos	= copy;
3460 	}
3461 
3462 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3463 		int end;
3464 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3465 
3466 		WARN_ON(start > offset + len);
3467 
3468 		end = start + skb_frag_size(frag);
3469 		if ((copy = end - offset) > 0) {
3470 			u32 p_off, p_len, copied;
3471 			struct page *p;
3472 			__wsum csum2;
3473 			u8 *vaddr;
3474 
3475 			if (copy > len)
3476 				copy = len;
3477 
3478 			skb_frag_foreach_page(frag,
3479 					      skb_frag_off(frag) + offset - start,
3480 					      copy, p, p_off, p_len, copied) {
3481 				vaddr = kmap_atomic(p);
3482 				csum2 = INDIRECT_CALL_1(ops->update,
3483 							csum_partial_ext,
3484 							vaddr + p_off, p_len, 0);
3485 				kunmap_atomic(vaddr);
3486 				csum = INDIRECT_CALL_1(ops->combine,
3487 						       csum_block_add_ext, csum,
3488 						       csum2, pos, p_len);
3489 				pos += p_len;
3490 			}
3491 
3492 			if (!(len -= copy))
3493 				return csum;
3494 			offset += copy;
3495 		}
3496 		start = end;
3497 	}
3498 
3499 	skb_walk_frags(skb, frag_iter) {
3500 		int end;
3501 
3502 		WARN_ON(start > offset + len);
3503 
3504 		end = start + frag_iter->len;
3505 		if ((copy = end - offset) > 0) {
3506 			__wsum csum2;
3507 			if (copy > len)
3508 				copy = len;
3509 			csum2 = __skb_checksum(frag_iter, offset - start,
3510 					       copy, 0, ops);
3511 			csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3512 					       csum, csum2, pos, copy);
3513 			if ((len -= copy) == 0)
3514 				return csum;
3515 			offset += copy;
3516 			pos    += copy;
3517 		}
3518 		start = end;
3519 	}
3520 	BUG_ON(len);
3521 
3522 	return csum;
3523 }
3524 EXPORT_SYMBOL(__skb_checksum);
3525 
3526 __wsum skb_checksum(const struct sk_buff *skb, int offset,
3527 		    int len, __wsum csum)
3528 {
3529 	const struct skb_checksum_ops ops = {
3530 		.update  = csum_partial_ext,
3531 		.combine = csum_block_add_ext,
3532 	};
3533 
3534 	return __skb_checksum(skb, offset, len, csum, &ops);
3535 }
3536 EXPORT_SYMBOL(skb_checksum);
3537 
3538 /* Both of above in one bottle. */
3539 
3540 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3541 				    u8 *to, int len)
3542 {
3543 	int start = skb_headlen(skb);
3544 	int i, copy = start - offset;
3545 	struct sk_buff *frag_iter;
3546 	int pos = 0;
3547 	__wsum csum = 0;
3548 
3549 	/* Copy header. */
3550 	if (copy > 0) {
3551 		if (copy > len)
3552 			copy = len;
3553 		csum = csum_partial_copy_nocheck(skb->data + offset, to,
3554 						 copy);
3555 		if ((len -= copy) == 0)
3556 			return csum;
3557 		offset += copy;
3558 		to     += copy;
3559 		pos	= copy;
3560 	}
3561 
3562 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3563 		int end;
3564 
3565 		WARN_ON(start > offset + len);
3566 
3567 		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3568 		if ((copy = end - offset) > 0) {
3569 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3570 			u32 p_off, p_len, copied;
3571 			struct page *p;
3572 			__wsum csum2;
3573 			u8 *vaddr;
3574 
3575 			if (copy > len)
3576 				copy = len;
3577 
3578 			skb_frag_foreach_page(frag,
3579 					      skb_frag_off(frag) + offset - start,
3580 					      copy, p, p_off, p_len, copied) {
3581 				vaddr = kmap_atomic(p);
3582 				csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3583 								  to + copied,
3584 								  p_len);
3585 				kunmap_atomic(vaddr);
3586 				csum = csum_block_add(csum, csum2, pos);
3587 				pos += p_len;
3588 			}
3589 
3590 			if (!(len -= copy))
3591 				return csum;
3592 			offset += copy;
3593 			to     += copy;
3594 		}
3595 		start = end;
3596 	}
3597 
3598 	skb_walk_frags(skb, frag_iter) {
3599 		__wsum csum2;
3600 		int end;
3601 
3602 		WARN_ON(start > offset + len);
3603 
3604 		end = start + frag_iter->len;
3605 		if ((copy = end - offset) > 0) {
3606 			if (copy > len)
3607 				copy = len;
3608 			csum2 = skb_copy_and_csum_bits(frag_iter,
3609 						       offset - start,
3610 						       to, copy);
3611 			csum = csum_block_add(csum, csum2, pos);
3612 			if ((len -= copy) == 0)
3613 				return csum;
3614 			offset += copy;
3615 			to     += copy;
3616 			pos    += copy;
3617 		}
3618 		start = end;
3619 	}
3620 	BUG_ON(len);
3621 	return csum;
3622 }
3623 EXPORT_SYMBOL(skb_copy_and_csum_bits);
3624 
3625 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3626 {
3627 	__sum16 sum;
3628 
3629 	sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3630 	/* See comments in __skb_checksum_complete(). */
3631 	if (likely(!sum)) {
3632 		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3633 		    !skb->csum_complete_sw)
3634 			netdev_rx_csum_fault(skb->dev, skb);
3635 	}
3636 	if (!skb_shared(skb))
3637 		skb->csum_valid = !sum;
3638 	return sum;
3639 }
3640 EXPORT_SYMBOL(__skb_checksum_complete_head);
3641 
3642 /* This function assumes skb->csum already holds pseudo header's checksum,
3643  * which has been changed from the hardware checksum, for example, by
3644  * __skb_checksum_validate_complete(). And, the original skb->csum must
3645  * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3646  *
3647  * It returns non-zero if the recomputed checksum is still invalid, otherwise
3648  * zero. The new checksum is stored back into skb->csum unless the skb is
3649  * shared.
3650  */
3651 __sum16 __skb_checksum_complete(struct sk_buff *skb)
3652 {
3653 	__wsum csum;
3654 	__sum16 sum;
3655 
3656 	csum = skb_checksum(skb, 0, skb->len, 0);
3657 
3658 	sum = csum_fold(csum_add(skb->csum, csum));
3659 	/* This check is inverted, because we already knew the hardware
3660 	 * checksum is invalid before calling this function. So, if the
3661 	 * re-computed checksum is valid instead, then we have a mismatch
3662 	 * between the original skb->csum and skb_checksum(). This means either
3663 	 * the original hardware checksum is incorrect or we screw up skb->csum
3664 	 * when moving skb->data around.
3665 	 */
3666 	if (likely(!sum)) {
3667 		if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3668 		    !skb->csum_complete_sw)
3669 			netdev_rx_csum_fault(skb->dev, skb);
3670 	}
3671 
3672 	if (!skb_shared(skb)) {
3673 		/* Save full packet checksum */
3674 		skb->csum = csum;
3675 		skb->ip_summed = CHECKSUM_COMPLETE;
3676 		skb->csum_complete_sw = 1;
3677 		skb->csum_valid = !sum;
3678 	}
3679 
3680 	return sum;
3681 }
3682 EXPORT_SYMBOL(__skb_checksum_complete);
3683 
3684 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3685 {
3686 	net_warn_ratelimited(
3687 		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3688 		__func__);
3689 	return 0;
3690 }
3691 
3692 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3693 				       int offset, int len)
3694 {
3695 	net_warn_ratelimited(
3696 		"%s: attempt to compute crc32c without libcrc32c.ko\n",
3697 		__func__);
3698 	return 0;
3699 }
3700 
3701 static const struct skb_checksum_ops default_crc32c_ops = {
3702 	.update  = warn_crc32c_csum_update,
3703 	.combine = warn_crc32c_csum_combine,
3704 };
3705 
3706 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3707 	&default_crc32c_ops;
3708 EXPORT_SYMBOL(crc32c_csum_stub);
3709 
3710  /**
3711  *	skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3712  *	@from: source buffer
3713  *
3714  *	Calculates the amount of linear headroom needed in the 'to' skb passed
3715  *	into skb_zerocopy().
3716  */
3717 unsigned int
3718 skb_zerocopy_headlen(const struct sk_buff *from)
3719 {
3720 	unsigned int hlen = 0;
3721 
3722 	if (!from->head_frag ||
3723 	    skb_headlen(from) < L1_CACHE_BYTES ||
3724 	    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3725 		hlen = skb_headlen(from);
3726 		if (!hlen)
3727 			hlen = from->len;
3728 	}
3729 
3730 	if (skb_has_frag_list(from))
3731 		hlen = from->len;
3732 
3733 	return hlen;
3734 }
3735 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3736 
3737 /**
3738  *	skb_zerocopy - Zero copy skb to skb
3739  *	@to: destination buffer
3740  *	@from: source buffer
3741  *	@len: number of bytes to copy from source buffer
3742  *	@hlen: size of linear headroom in destination buffer
3743  *
3744  *	Copies up to `len` bytes from `from` to `to` by creating references
3745  *	to the frags in the source buffer.
3746  *
3747  *	The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3748  *	headroom in the `to` buffer.
3749  *
3750  *	Return value:
3751  *	0: everything is OK
3752  *	-ENOMEM: couldn't orphan frags of @from due to lack of memory
3753  *	-EFAULT: skb_copy_bits() found some problem with skb geometry
3754  */
3755 int
3756 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3757 {
3758 	int i, j = 0;
3759 	int plen = 0; /* length of skb->head fragment */
3760 	int ret;
3761 	struct page *page;
3762 	unsigned int offset;
3763 
3764 	BUG_ON(!from->head_frag && !hlen);
3765 
3766 	/* dont bother with small payloads */
3767 	if (len <= skb_tailroom(to))
3768 		return skb_copy_bits(from, 0, skb_put(to, len), len);
3769 
3770 	if (hlen) {
3771 		ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3772 		if (unlikely(ret))
3773 			return ret;
3774 		len -= hlen;
3775 	} else {
3776 		plen = min_t(int, skb_headlen(from), len);
3777 		if (plen) {
3778 			page = virt_to_head_page(from->head);
3779 			offset = from->data - (unsigned char *)page_address(page);
3780 			__skb_fill_netmem_desc(to, 0, page_to_netmem(page),
3781 					       offset, plen);
3782 			get_page(page);
3783 			j = 1;
3784 			len -= plen;
3785 		}
3786 	}
3787 
3788 	skb_len_add(to, len + plen);
3789 
3790 	if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3791 		skb_tx_error(from);
3792 		return -ENOMEM;
3793 	}
3794 	skb_zerocopy_clone(to, from, GFP_ATOMIC);
3795 
3796 	for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3797 		int size;
3798 
3799 		if (!len)
3800 			break;
3801 		skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3802 		size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3803 					len);
3804 		skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3805 		len -= size;
3806 		skb_frag_ref(to, j);
3807 		j++;
3808 	}
3809 	skb_shinfo(to)->nr_frags = j;
3810 
3811 	return 0;
3812 }
3813 EXPORT_SYMBOL_GPL(skb_zerocopy);
3814 
3815 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3816 {
3817 	__wsum csum;
3818 	long csstart;
3819 
3820 	if (skb->ip_summed == CHECKSUM_PARTIAL)
3821 		csstart = skb_checksum_start_offset(skb);
3822 	else
3823 		csstart = skb_headlen(skb);
3824 
3825 	BUG_ON(csstart > skb_headlen(skb));
3826 
3827 	skb_copy_from_linear_data(skb, to, csstart);
3828 
3829 	csum = 0;
3830 	if (csstart != skb->len)
3831 		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3832 					      skb->len - csstart);
3833 
3834 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
3835 		long csstuff = csstart + skb->csum_offset;
3836 
3837 		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
3838 	}
3839 }
3840 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3841 
3842 /**
3843  *	skb_dequeue - remove from the head of the queue
3844  *	@list: list to dequeue from
3845  *
3846  *	Remove the head of the list. The list lock is taken so the function
3847  *	may be used safely with other locking list functions. The head item is
3848  *	returned or %NULL if the list is empty.
3849  */
3850 
3851 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3852 {
3853 	unsigned long flags;
3854 	struct sk_buff *result;
3855 
3856 	spin_lock_irqsave(&list->lock, flags);
3857 	result = __skb_dequeue(list);
3858 	spin_unlock_irqrestore(&list->lock, flags);
3859 	return result;
3860 }
3861 EXPORT_SYMBOL(skb_dequeue);
3862 
3863 /**
3864  *	skb_dequeue_tail - remove from the tail of the queue
3865  *	@list: list to dequeue from
3866  *
3867  *	Remove the tail of the list. The list lock is taken so the function
3868  *	may be used safely with other locking list functions. The tail item is
3869  *	returned or %NULL if the list is empty.
3870  */
3871 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3872 {
3873 	unsigned long flags;
3874 	struct sk_buff *result;
3875 
3876 	spin_lock_irqsave(&list->lock, flags);
3877 	result = __skb_dequeue_tail(list);
3878 	spin_unlock_irqrestore(&list->lock, flags);
3879 	return result;
3880 }
3881 EXPORT_SYMBOL(skb_dequeue_tail);
3882 
3883 /**
3884  *	skb_queue_purge_reason - empty a list
3885  *	@list: list to empty
3886  *	@reason: drop reason
3887  *
3888  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
3889  *	the list and one reference dropped. This function takes the list
3890  *	lock and is atomic with respect to other list locking functions.
3891  */
3892 void skb_queue_purge_reason(struct sk_buff_head *list,
3893 			    enum skb_drop_reason reason)
3894 {
3895 	struct sk_buff_head tmp;
3896 	unsigned long flags;
3897 
3898 	if (skb_queue_empty_lockless(list))
3899 		return;
3900 
3901 	__skb_queue_head_init(&tmp);
3902 
3903 	spin_lock_irqsave(&list->lock, flags);
3904 	skb_queue_splice_init(list, &tmp);
3905 	spin_unlock_irqrestore(&list->lock, flags);
3906 
3907 	__skb_queue_purge_reason(&tmp, reason);
3908 }
3909 EXPORT_SYMBOL(skb_queue_purge_reason);
3910 
3911 /**
3912  *	skb_rbtree_purge - empty a skb rbtree
3913  *	@root: root of the rbtree to empty
3914  *	Return value: the sum of truesizes of all purged skbs.
3915  *
3916  *	Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3917  *	the list and one reference dropped. This function does not take
3918  *	any lock. Synchronization should be handled by the caller (e.g., TCP
3919  *	out-of-order queue is protected by the socket lock).
3920  */
3921 unsigned int skb_rbtree_purge(struct rb_root *root)
3922 {
3923 	struct rb_node *p = rb_first(root);
3924 	unsigned int sum = 0;
3925 
3926 	while (p) {
3927 		struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3928 
3929 		p = rb_next(p);
3930 		rb_erase(&skb->rbnode, root);
3931 		sum += skb->truesize;
3932 		kfree_skb(skb);
3933 	}
3934 	return sum;
3935 }
3936 
3937 void skb_errqueue_purge(struct sk_buff_head *list)
3938 {
3939 	struct sk_buff *skb, *next;
3940 	struct sk_buff_head kill;
3941 	unsigned long flags;
3942 
3943 	__skb_queue_head_init(&kill);
3944 
3945 	spin_lock_irqsave(&list->lock, flags);
3946 	skb_queue_walk_safe(list, skb, next) {
3947 		if (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ZEROCOPY ||
3948 		    SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING)
3949 			continue;
3950 		__skb_unlink(skb, list);
3951 		__skb_queue_tail(&kill, skb);
3952 	}
3953 	spin_unlock_irqrestore(&list->lock, flags);
3954 	__skb_queue_purge(&kill);
3955 }
3956 EXPORT_SYMBOL(skb_errqueue_purge);
3957 
3958 /**
3959  *	skb_queue_head - queue a buffer at the list head
3960  *	@list: list to use
3961  *	@newsk: buffer to queue
3962  *
3963  *	Queue a buffer at the start of the list. This function takes the
3964  *	list lock and can be used safely with other locking &sk_buff functions
3965  *	safely.
3966  *
3967  *	A buffer cannot be placed on two lists at the same time.
3968  */
3969 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3970 {
3971 	unsigned long flags;
3972 
3973 	spin_lock_irqsave(&list->lock, flags);
3974 	__skb_queue_head(list, newsk);
3975 	spin_unlock_irqrestore(&list->lock, flags);
3976 }
3977 EXPORT_SYMBOL(skb_queue_head);
3978 
3979 /**
3980  *	skb_queue_tail - queue a buffer at the list tail
3981  *	@list: list to use
3982  *	@newsk: buffer to queue
3983  *
3984  *	Queue a buffer at the tail of the list. This function takes the
3985  *	list lock and can be used safely with other locking &sk_buff functions
3986  *	safely.
3987  *
3988  *	A buffer cannot be placed on two lists at the same time.
3989  */
3990 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3991 {
3992 	unsigned long flags;
3993 
3994 	spin_lock_irqsave(&list->lock, flags);
3995 	__skb_queue_tail(list, newsk);
3996 	spin_unlock_irqrestore(&list->lock, flags);
3997 }
3998 EXPORT_SYMBOL(skb_queue_tail);
3999 
4000 /**
4001  *	skb_unlink	-	remove a buffer from a list
4002  *	@skb: buffer to remove
4003  *	@list: list to use
4004  *
4005  *	Remove a packet from a list. The list locks are taken and this
4006  *	function is atomic with respect to other list locked calls
4007  *
4008  *	You must know what list the SKB is on.
4009  */
4010 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
4011 {
4012 	unsigned long flags;
4013 
4014 	spin_lock_irqsave(&list->lock, flags);
4015 	__skb_unlink(skb, list);
4016 	spin_unlock_irqrestore(&list->lock, flags);
4017 }
4018 EXPORT_SYMBOL(skb_unlink);
4019 
4020 /**
4021  *	skb_append	-	append a buffer
4022  *	@old: buffer to insert after
4023  *	@newsk: buffer to insert
4024  *	@list: list to use
4025  *
4026  *	Place a packet after a given packet in a list. The list locks are taken
4027  *	and this function is atomic with respect to other list locked calls.
4028  *	A buffer cannot be placed on two lists at the same time.
4029  */
4030 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
4031 {
4032 	unsigned long flags;
4033 
4034 	spin_lock_irqsave(&list->lock, flags);
4035 	__skb_queue_after(list, old, newsk);
4036 	spin_unlock_irqrestore(&list->lock, flags);
4037 }
4038 EXPORT_SYMBOL(skb_append);
4039 
4040 static inline void skb_split_inside_header(struct sk_buff *skb,
4041 					   struct sk_buff* skb1,
4042 					   const u32 len, const int pos)
4043 {
4044 	int i;
4045 
4046 	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
4047 					 pos - len);
4048 	/* And move data appendix as is. */
4049 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4050 		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
4051 
4052 	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
4053 	skb_shinfo(skb)->nr_frags  = 0;
4054 	skb1->data_len		   = skb->data_len;
4055 	skb1->len		   += skb1->data_len;
4056 	skb->data_len		   = 0;
4057 	skb->len		   = len;
4058 	skb_set_tail_pointer(skb, len);
4059 }
4060 
4061 static inline void skb_split_no_header(struct sk_buff *skb,
4062 				       struct sk_buff* skb1,
4063 				       const u32 len, int pos)
4064 {
4065 	int i, k = 0;
4066 	const int nfrags = skb_shinfo(skb)->nr_frags;
4067 
4068 	skb_shinfo(skb)->nr_frags = 0;
4069 	skb1->len		  = skb1->data_len = skb->len - len;
4070 	skb->len		  = len;
4071 	skb->data_len		  = len - pos;
4072 
4073 	for (i = 0; i < nfrags; i++) {
4074 		int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4075 
4076 		if (pos + size > len) {
4077 			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
4078 
4079 			if (pos < len) {
4080 				/* Split frag.
4081 				 * We have two variants in this case:
4082 				 * 1. Move all the frag to the second
4083 				 *    part, if it is possible. F.e.
4084 				 *    this approach is mandatory for TUX,
4085 				 *    where splitting is expensive.
4086 				 * 2. Split is accurately. We make this.
4087 				 */
4088 				skb_frag_ref(skb, i);
4089 				skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
4090 				skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
4091 				skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
4092 				skb_shinfo(skb)->nr_frags++;
4093 			}
4094 			k++;
4095 		} else
4096 			skb_shinfo(skb)->nr_frags++;
4097 		pos += size;
4098 	}
4099 	skb_shinfo(skb1)->nr_frags = k;
4100 }
4101 
4102 /**
4103  * skb_split - Split fragmented skb to two parts at length len.
4104  * @skb: the buffer to split
4105  * @skb1: the buffer to receive the second part
4106  * @len: new length for skb
4107  */
4108 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
4109 {
4110 	int pos = skb_headlen(skb);
4111 	const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
4112 
4113 	skb_zcopy_downgrade_managed(skb);
4114 
4115 	skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
4116 	skb_zerocopy_clone(skb1, skb, 0);
4117 	if (len < pos)	/* Split line is inside header. */
4118 		skb_split_inside_header(skb, skb1, len, pos);
4119 	else		/* Second chunk has no header, nothing to copy. */
4120 		skb_split_no_header(skb, skb1, len, pos);
4121 }
4122 EXPORT_SYMBOL(skb_split);
4123 
4124 /* Shifting from/to a cloned skb is a no-go.
4125  *
4126  * Caller cannot keep skb_shinfo related pointers past calling here!
4127  */
4128 static int skb_prepare_for_shift(struct sk_buff *skb)
4129 {
4130 	return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
4131 }
4132 
4133 /**
4134  * skb_shift - Shifts paged data partially from skb to another
4135  * @tgt: buffer into which tail data gets added
4136  * @skb: buffer from which the paged data comes from
4137  * @shiftlen: shift up to this many bytes
4138  *
4139  * Attempts to shift up to shiftlen worth of bytes, which may be less than
4140  * the length of the skb, from skb to tgt. Returns number bytes shifted.
4141  * It's up to caller to free skb if everything was shifted.
4142  *
4143  * If @tgt runs out of frags, the whole operation is aborted.
4144  *
4145  * Skb cannot include anything else but paged data while tgt is allowed
4146  * to have non-paged data as well.
4147  *
4148  * TODO: full sized shift could be optimized but that would need
4149  * specialized skb free'er to handle frags without up-to-date nr_frags.
4150  */
4151 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
4152 {
4153 	int from, to, merge, todo;
4154 	skb_frag_t *fragfrom, *fragto;
4155 
4156 	BUG_ON(shiftlen > skb->len);
4157 
4158 	if (skb_headlen(skb))
4159 		return 0;
4160 	if (skb_zcopy(tgt) || skb_zcopy(skb))
4161 		return 0;
4162 
4163 	DEBUG_NET_WARN_ON_ONCE(tgt->pp_recycle != skb->pp_recycle);
4164 	DEBUG_NET_WARN_ON_ONCE(skb_cmp_decrypted(tgt, skb));
4165 
4166 	todo = shiftlen;
4167 	from = 0;
4168 	to = skb_shinfo(tgt)->nr_frags;
4169 	fragfrom = &skb_shinfo(skb)->frags[from];
4170 
4171 	/* Actual merge is delayed until the point when we know we can
4172 	 * commit all, so that we don't have to undo partial changes
4173 	 */
4174 	if (!skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
4175 			      skb_frag_off(fragfrom))) {
4176 		merge = -1;
4177 	} else {
4178 		merge = to - 1;
4179 
4180 		todo -= skb_frag_size(fragfrom);
4181 		if (todo < 0) {
4182 			if (skb_prepare_for_shift(skb) ||
4183 			    skb_prepare_for_shift(tgt))
4184 				return 0;
4185 
4186 			/* All previous frag pointers might be stale! */
4187 			fragfrom = &skb_shinfo(skb)->frags[from];
4188 			fragto = &skb_shinfo(tgt)->frags[merge];
4189 
4190 			skb_frag_size_add(fragto, shiftlen);
4191 			skb_frag_size_sub(fragfrom, shiftlen);
4192 			skb_frag_off_add(fragfrom, shiftlen);
4193 
4194 			goto onlymerged;
4195 		}
4196 
4197 		from++;
4198 	}
4199 
4200 	/* Skip full, not-fitting skb to avoid expensive operations */
4201 	if ((shiftlen == skb->len) &&
4202 	    (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
4203 		return 0;
4204 
4205 	if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
4206 		return 0;
4207 
4208 	while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
4209 		if (to == MAX_SKB_FRAGS)
4210 			return 0;
4211 
4212 		fragfrom = &skb_shinfo(skb)->frags[from];
4213 		fragto = &skb_shinfo(tgt)->frags[to];
4214 
4215 		if (todo >= skb_frag_size(fragfrom)) {
4216 			*fragto = *fragfrom;
4217 			todo -= skb_frag_size(fragfrom);
4218 			from++;
4219 			to++;
4220 
4221 		} else {
4222 			__skb_frag_ref(fragfrom);
4223 			skb_frag_page_copy(fragto, fragfrom);
4224 			skb_frag_off_copy(fragto, fragfrom);
4225 			skb_frag_size_set(fragto, todo);
4226 
4227 			skb_frag_off_add(fragfrom, todo);
4228 			skb_frag_size_sub(fragfrom, todo);
4229 			todo = 0;
4230 
4231 			to++;
4232 			break;
4233 		}
4234 	}
4235 
4236 	/* Ready to "commit" this state change to tgt */
4237 	skb_shinfo(tgt)->nr_frags = to;
4238 
4239 	if (merge >= 0) {
4240 		fragfrom = &skb_shinfo(skb)->frags[0];
4241 		fragto = &skb_shinfo(tgt)->frags[merge];
4242 
4243 		skb_frag_size_add(fragto, skb_frag_size(fragfrom));
4244 		__skb_frag_unref(fragfrom, skb->pp_recycle);
4245 	}
4246 
4247 	/* Reposition in the original skb */
4248 	to = 0;
4249 	while (from < skb_shinfo(skb)->nr_frags)
4250 		skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
4251 	skb_shinfo(skb)->nr_frags = to;
4252 
4253 	BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
4254 
4255 onlymerged:
4256 	/* Most likely the tgt won't ever need its checksum anymore, skb on
4257 	 * the other hand might need it if it needs to be resent
4258 	 */
4259 	tgt->ip_summed = CHECKSUM_PARTIAL;
4260 	skb->ip_summed = CHECKSUM_PARTIAL;
4261 
4262 	skb_len_add(skb, -shiftlen);
4263 	skb_len_add(tgt, shiftlen);
4264 
4265 	return shiftlen;
4266 }
4267 
4268 /**
4269  * skb_prepare_seq_read - Prepare a sequential read of skb data
4270  * @skb: the buffer to read
4271  * @from: lower offset of data to be read
4272  * @to: upper offset of data to be read
4273  * @st: state variable
4274  *
4275  * Initializes the specified state variable. Must be called before
4276  * invoking skb_seq_read() for the first time.
4277  */
4278 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
4279 			  unsigned int to, struct skb_seq_state *st)
4280 {
4281 	st->lower_offset = from;
4282 	st->upper_offset = to;
4283 	st->root_skb = st->cur_skb = skb;
4284 	st->frag_idx = st->stepped_offset = 0;
4285 	st->frag_data = NULL;
4286 	st->frag_off = 0;
4287 }
4288 EXPORT_SYMBOL(skb_prepare_seq_read);
4289 
4290 /**
4291  * skb_seq_read - Sequentially read skb data
4292  * @consumed: number of bytes consumed by the caller so far
4293  * @data: destination pointer for data to be returned
4294  * @st: state variable
4295  *
4296  * Reads a block of skb data at @consumed relative to the
4297  * lower offset specified to skb_prepare_seq_read(). Assigns
4298  * the head of the data block to @data and returns the length
4299  * of the block or 0 if the end of the skb data or the upper
4300  * offset has been reached.
4301  *
4302  * The caller is not required to consume all of the data
4303  * returned, i.e. @consumed is typically set to the number
4304  * of bytes already consumed and the next call to
4305  * skb_seq_read() will return the remaining part of the block.
4306  *
4307  * Note 1: The size of each block of data returned can be arbitrary,
4308  *       this limitation is the cost for zerocopy sequential
4309  *       reads of potentially non linear data.
4310  *
4311  * Note 2: Fragment lists within fragments are not implemented
4312  *       at the moment, state->root_skb could be replaced with
4313  *       a stack for this purpose.
4314  */
4315 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
4316 			  struct skb_seq_state *st)
4317 {
4318 	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
4319 	skb_frag_t *frag;
4320 
4321 	if (unlikely(abs_offset >= st->upper_offset)) {
4322 		if (st->frag_data) {
4323 			kunmap_atomic(st->frag_data);
4324 			st->frag_data = NULL;
4325 		}
4326 		return 0;
4327 	}
4328 
4329 next_skb:
4330 	block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
4331 
4332 	if (abs_offset < block_limit && !st->frag_data) {
4333 		*data = st->cur_skb->data + (abs_offset - st->stepped_offset);
4334 		return block_limit - abs_offset;
4335 	}
4336 
4337 	if (st->frag_idx == 0 && !st->frag_data)
4338 		st->stepped_offset += skb_headlen(st->cur_skb);
4339 
4340 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
4341 		unsigned int pg_idx, pg_off, pg_sz;
4342 
4343 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
4344 
4345 		pg_idx = 0;
4346 		pg_off = skb_frag_off(frag);
4347 		pg_sz = skb_frag_size(frag);
4348 
4349 		if (skb_frag_must_loop(skb_frag_page(frag))) {
4350 			pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
4351 			pg_off = offset_in_page(pg_off + st->frag_off);
4352 			pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
4353 						    PAGE_SIZE - pg_off);
4354 		}
4355 
4356 		block_limit = pg_sz + st->stepped_offset;
4357 		if (abs_offset < block_limit) {
4358 			if (!st->frag_data)
4359 				st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
4360 
4361 			*data = (u8 *)st->frag_data + pg_off +
4362 				(abs_offset - st->stepped_offset);
4363 
4364 			return block_limit - abs_offset;
4365 		}
4366 
4367 		if (st->frag_data) {
4368 			kunmap_atomic(st->frag_data);
4369 			st->frag_data = NULL;
4370 		}
4371 
4372 		st->stepped_offset += pg_sz;
4373 		st->frag_off += pg_sz;
4374 		if (st->frag_off == skb_frag_size(frag)) {
4375 			st->frag_off = 0;
4376 			st->frag_idx++;
4377 		}
4378 	}
4379 
4380 	if (st->frag_data) {
4381 		kunmap_atomic(st->frag_data);
4382 		st->frag_data = NULL;
4383 	}
4384 
4385 	if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
4386 		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
4387 		st->frag_idx = 0;
4388 		goto next_skb;
4389 	} else if (st->cur_skb->next) {
4390 		st->cur_skb = st->cur_skb->next;
4391 		st->frag_idx = 0;
4392 		goto next_skb;
4393 	}
4394 
4395 	return 0;
4396 }
4397 EXPORT_SYMBOL(skb_seq_read);
4398 
4399 /**
4400  * skb_abort_seq_read - Abort a sequential read of skb data
4401  * @st: state variable
4402  *
4403  * Must be called if skb_seq_read() was not called until it
4404  * returned 0.
4405  */
4406 void skb_abort_seq_read(struct skb_seq_state *st)
4407 {
4408 	if (st->frag_data)
4409 		kunmap_atomic(st->frag_data);
4410 }
4411 EXPORT_SYMBOL(skb_abort_seq_read);
4412 
4413 /**
4414  * skb_copy_seq_read() - copy from a skb_seq_state to a buffer
4415  * @st: source skb_seq_state
4416  * @offset: offset in source
4417  * @to: destination buffer
4418  * @len: number of bytes to copy
4419  *
4420  * Copy @len bytes from @offset bytes into the source @st to the destination
4421  * buffer @to. `offset` should increase (or be unchanged) with each subsequent
4422  * call to this function. If offset needs to decrease from the previous use `st`
4423  * should be reset first.
4424  *
4425  * Return: 0 on success or -EINVAL if the copy ended early
4426  */
4427 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len)
4428 {
4429 	const u8 *data;
4430 	u32 sqlen;
4431 
4432 	for (;;) {
4433 		sqlen = skb_seq_read(offset, &data, st);
4434 		if (sqlen == 0)
4435 			return -EINVAL;
4436 		if (sqlen >= len) {
4437 			memcpy(to, data, len);
4438 			return 0;
4439 		}
4440 		memcpy(to, data, sqlen);
4441 		to += sqlen;
4442 		offset += sqlen;
4443 		len -= sqlen;
4444 	}
4445 }
4446 EXPORT_SYMBOL(skb_copy_seq_read);
4447 
4448 #define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
4449 
4450 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
4451 					  struct ts_config *conf,
4452 					  struct ts_state *state)
4453 {
4454 	return skb_seq_read(offset, text, TS_SKB_CB(state));
4455 }
4456 
4457 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
4458 {
4459 	skb_abort_seq_read(TS_SKB_CB(state));
4460 }
4461 
4462 /**
4463  * skb_find_text - Find a text pattern in skb data
4464  * @skb: the buffer to look in
4465  * @from: search offset
4466  * @to: search limit
4467  * @config: textsearch configuration
4468  *
4469  * Finds a pattern in the skb data according to the specified
4470  * textsearch configuration. Use textsearch_next() to retrieve
4471  * subsequent occurrences of the pattern. Returns the offset
4472  * to the first occurrence or UINT_MAX if no match was found.
4473  */
4474 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4475 			   unsigned int to, struct ts_config *config)
4476 {
4477 	unsigned int patlen = config->ops->get_pattern_len(config);
4478 	struct ts_state state;
4479 	unsigned int ret;
4480 
4481 	BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4482 
4483 	config->get_next_block = skb_ts_get_next_block;
4484 	config->finish = skb_ts_finish;
4485 
4486 	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4487 
4488 	ret = textsearch_find(config, &state);
4489 	return (ret + patlen <= to - from ? ret : UINT_MAX);
4490 }
4491 EXPORT_SYMBOL(skb_find_text);
4492 
4493 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4494 			 int offset, size_t size, size_t max_frags)
4495 {
4496 	int i = skb_shinfo(skb)->nr_frags;
4497 
4498 	if (skb_can_coalesce(skb, i, page, offset)) {
4499 		skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4500 	} else if (i < max_frags) {
4501 		skb_zcopy_downgrade_managed(skb);
4502 		get_page(page);
4503 		skb_fill_page_desc_noacc(skb, i, page, offset, size);
4504 	} else {
4505 		return -EMSGSIZE;
4506 	}
4507 
4508 	return 0;
4509 }
4510 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4511 
4512 /**
4513  *	skb_pull_rcsum - pull skb and update receive checksum
4514  *	@skb: buffer to update
4515  *	@len: length of data pulled
4516  *
4517  *	This function performs an skb_pull on the packet and updates
4518  *	the CHECKSUM_COMPLETE checksum.  It should be used on
4519  *	receive path processing instead of skb_pull unless you know
4520  *	that the checksum difference is zero (e.g., a valid IP header)
4521  *	or you are setting ip_summed to CHECKSUM_NONE.
4522  */
4523 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4524 {
4525 	unsigned char *data = skb->data;
4526 
4527 	BUG_ON(len > skb->len);
4528 	__skb_pull(skb, len);
4529 	skb_postpull_rcsum(skb, data, len);
4530 	return skb->data;
4531 }
4532 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4533 
4534 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4535 {
4536 	skb_frag_t head_frag;
4537 	struct page *page;
4538 
4539 	page = virt_to_head_page(frag_skb->head);
4540 	skb_frag_fill_page_desc(&head_frag, page, frag_skb->data -
4541 				(unsigned char *)page_address(page),
4542 				skb_headlen(frag_skb));
4543 	return head_frag;
4544 }
4545 
4546 struct sk_buff *skb_segment_list(struct sk_buff *skb,
4547 				 netdev_features_t features,
4548 				 unsigned int offset)
4549 {
4550 	struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4551 	unsigned int tnl_hlen = skb_tnl_header_len(skb);
4552 	unsigned int delta_truesize = 0;
4553 	unsigned int delta_len = 0;
4554 	struct sk_buff *tail = NULL;
4555 	struct sk_buff *nskb, *tmp;
4556 	int len_diff, err;
4557 
4558 	skb_push(skb, -skb_network_offset(skb) + offset);
4559 
4560 	/* Ensure the head is writeable before touching the shared info */
4561 	err = skb_unclone(skb, GFP_ATOMIC);
4562 	if (err)
4563 		goto err_linearize;
4564 
4565 	skb_shinfo(skb)->frag_list = NULL;
4566 
4567 	while (list_skb) {
4568 		nskb = list_skb;
4569 		list_skb = list_skb->next;
4570 
4571 		err = 0;
4572 		delta_truesize += nskb->truesize;
4573 		if (skb_shared(nskb)) {
4574 			tmp = skb_clone(nskb, GFP_ATOMIC);
4575 			if (tmp) {
4576 				consume_skb(nskb);
4577 				nskb = tmp;
4578 				err = skb_unclone(nskb, GFP_ATOMIC);
4579 			} else {
4580 				err = -ENOMEM;
4581 			}
4582 		}
4583 
4584 		if (!tail)
4585 			skb->next = nskb;
4586 		else
4587 			tail->next = nskb;
4588 
4589 		if (unlikely(err)) {
4590 			nskb->next = list_skb;
4591 			goto err_linearize;
4592 		}
4593 
4594 		tail = nskb;
4595 
4596 		delta_len += nskb->len;
4597 
4598 		skb_push(nskb, -skb_network_offset(nskb) + offset);
4599 
4600 		skb_release_head_state(nskb);
4601 		len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4602 		__copy_skb_header(nskb, skb);
4603 
4604 		skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4605 		nskb->transport_header += len_diff;
4606 		skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4607 						 nskb->data - tnl_hlen,
4608 						 offset + tnl_hlen);
4609 
4610 		if (skb_needs_linearize(nskb, features) &&
4611 		    __skb_linearize(nskb))
4612 			goto err_linearize;
4613 	}
4614 
4615 	skb->truesize = skb->truesize - delta_truesize;
4616 	skb->data_len = skb->data_len - delta_len;
4617 	skb->len = skb->len - delta_len;
4618 
4619 	skb_gso_reset(skb);
4620 
4621 	skb->prev = tail;
4622 
4623 	if (skb_needs_linearize(skb, features) &&
4624 	    __skb_linearize(skb))
4625 		goto err_linearize;
4626 
4627 	skb_get(skb);
4628 
4629 	return skb;
4630 
4631 err_linearize:
4632 	kfree_skb_list(skb->next);
4633 	skb->next = NULL;
4634 	return ERR_PTR(-ENOMEM);
4635 }
4636 EXPORT_SYMBOL_GPL(skb_segment_list);
4637 
4638 /**
4639  *	skb_segment - Perform protocol segmentation on skb.
4640  *	@head_skb: buffer to segment
4641  *	@features: features for the output path (see dev->features)
4642  *
4643  *	This function performs segmentation on the given skb.  It returns
4644  *	a pointer to the first in a list of new skbs for the segments.
4645  *	In case of error it returns ERR_PTR(err).
4646  */
4647 struct sk_buff *skb_segment(struct sk_buff *head_skb,
4648 			    netdev_features_t features)
4649 {
4650 	struct sk_buff *segs = NULL;
4651 	struct sk_buff *tail = NULL;
4652 	struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4653 	unsigned int mss = skb_shinfo(head_skb)->gso_size;
4654 	unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4655 	unsigned int offset = doffset;
4656 	unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4657 	unsigned int partial_segs = 0;
4658 	unsigned int headroom;
4659 	unsigned int len = head_skb->len;
4660 	struct sk_buff *frag_skb;
4661 	skb_frag_t *frag;
4662 	__be16 proto;
4663 	bool csum, sg;
4664 	int err = -ENOMEM;
4665 	int i = 0;
4666 	int nfrags, pos;
4667 
4668 	if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4669 	    mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4670 		struct sk_buff *check_skb;
4671 
4672 		for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4673 			if (skb_headlen(check_skb) && !check_skb->head_frag) {
4674 				/* gso_size is untrusted, and we have a frag_list with
4675 				 * a linear non head_frag item.
4676 				 *
4677 				 * If head_skb's headlen does not fit requested gso_size,
4678 				 * it means that the frag_list members do NOT terminate
4679 				 * on exact gso_size boundaries. Hence we cannot perform
4680 				 * skb_frag_t page sharing. Therefore we must fallback to
4681 				 * copying the frag_list skbs; we do so by disabling SG.
4682 				 */
4683 				features &= ~NETIF_F_SG;
4684 				break;
4685 			}
4686 		}
4687 	}
4688 
4689 	__skb_push(head_skb, doffset);
4690 	proto = skb_network_protocol(head_skb, NULL);
4691 	if (unlikely(!proto))
4692 		return ERR_PTR(-EINVAL);
4693 
4694 	sg = !!(features & NETIF_F_SG);
4695 	csum = !!can_checksum_protocol(features, proto);
4696 
4697 	if (sg && csum && (mss != GSO_BY_FRAGS))  {
4698 		if (!(features & NETIF_F_GSO_PARTIAL)) {
4699 			struct sk_buff *iter;
4700 			unsigned int frag_len;
4701 
4702 			if (!list_skb ||
4703 			    !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4704 				goto normal;
4705 
4706 			/* If we get here then all the required
4707 			 * GSO features except frag_list are supported.
4708 			 * Try to split the SKB to multiple GSO SKBs
4709 			 * with no frag_list.
4710 			 * Currently we can do that only when the buffers don't
4711 			 * have a linear part and all the buffers except
4712 			 * the last are of the same length.
4713 			 */
4714 			frag_len = list_skb->len;
4715 			skb_walk_frags(head_skb, iter) {
4716 				if (frag_len != iter->len && iter->next)
4717 					goto normal;
4718 				if (skb_headlen(iter) && !iter->head_frag)
4719 					goto normal;
4720 
4721 				len -= iter->len;
4722 			}
4723 
4724 			if (len != frag_len)
4725 				goto normal;
4726 		}
4727 
4728 		/* GSO partial only requires that we trim off any excess that
4729 		 * doesn't fit into an MSS sized block, so take care of that
4730 		 * now.
4731 		 * Cap len to not accidentally hit GSO_BY_FRAGS.
4732 		 */
4733 		partial_segs = min(len, GSO_BY_FRAGS - 1) / mss;
4734 		if (partial_segs > 1)
4735 			mss *= partial_segs;
4736 		else
4737 			partial_segs = 0;
4738 	}
4739 
4740 normal:
4741 	headroom = skb_headroom(head_skb);
4742 	pos = skb_headlen(head_skb);
4743 
4744 	if (skb_orphan_frags(head_skb, GFP_ATOMIC))
4745 		return ERR_PTR(-ENOMEM);
4746 
4747 	nfrags = skb_shinfo(head_skb)->nr_frags;
4748 	frag = skb_shinfo(head_skb)->frags;
4749 	frag_skb = head_skb;
4750 
4751 	do {
4752 		struct sk_buff *nskb;
4753 		skb_frag_t *nskb_frag;
4754 		int hsize;
4755 		int size;
4756 
4757 		if (unlikely(mss == GSO_BY_FRAGS)) {
4758 			len = list_skb->len;
4759 		} else {
4760 			len = head_skb->len - offset;
4761 			if (len > mss)
4762 				len = mss;
4763 		}
4764 
4765 		hsize = skb_headlen(head_skb) - offset;
4766 
4767 		if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4768 		    (skb_headlen(list_skb) == len || sg)) {
4769 			BUG_ON(skb_headlen(list_skb) > len);
4770 
4771 			nskb = skb_clone(list_skb, GFP_ATOMIC);
4772 			if (unlikely(!nskb))
4773 				goto err;
4774 
4775 			i = 0;
4776 			nfrags = skb_shinfo(list_skb)->nr_frags;
4777 			frag = skb_shinfo(list_skb)->frags;
4778 			frag_skb = list_skb;
4779 			pos += skb_headlen(list_skb);
4780 
4781 			while (pos < offset + len) {
4782 				BUG_ON(i >= nfrags);
4783 
4784 				size = skb_frag_size(frag);
4785 				if (pos + size > offset + len)
4786 					break;
4787 
4788 				i++;
4789 				pos += size;
4790 				frag++;
4791 			}
4792 
4793 			list_skb = list_skb->next;
4794 
4795 			if (unlikely(pskb_trim(nskb, len))) {
4796 				kfree_skb(nskb);
4797 				goto err;
4798 			}
4799 
4800 			hsize = skb_end_offset(nskb);
4801 			if (skb_cow_head(nskb, doffset + headroom)) {
4802 				kfree_skb(nskb);
4803 				goto err;
4804 			}
4805 
4806 			nskb->truesize += skb_end_offset(nskb) - hsize;
4807 			skb_release_head_state(nskb);
4808 			__skb_push(nskb, doffset);
4809 		} else {
4810 			if (hsize < 0)
4811 				hsize = 0;
4812 			if (hsize > len || !sg)
4813 				hsize = len;
4814 
4815 			nskb = __alloc_skb(hsize + doffset + headroom,
4816 					   GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4817 					   NUMA_NO_NODE);
4818 
4819 			if (unlikely(!nskb))
4820 				goto err;
4821 
4822 			skb_reserve(nskb, headroom);
4823 			__skb_put(nskb, doffset);
4824 		}
4825 
4826 		if (segs)
4827 			tail->next = nskb;
4828 		else
4829 			segs = nskb;
4830 		tail = nskb;
4831 
4832 		__copy_skb_header(nskb, head_skb);
4833 
4834 		skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4835 		skb_reset_mac_len(nskb);
4836 
4837 		skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4838 						 nskb->data - tnl_hlen,
4839 						 doffset + tnl_hlen);
4840 
4841 		if (nskb->len == len + doffset)
4842 			goto perform_csum_check;
4843 
4844 		if (!sg) {
4845 			if (!csum) {
4846 				if (!nskb->remcsum_offload)
4847 					nskb->ip_summed = CHECKSUM_NONE;
4848 				SKB_GSO_CB(nskb)->csum =
4849 					skb_copy_and_csum_bits(head_skb, offset,
4850 							       skb_put(nskb,
4851 								       len),
4852 							       len);
4853 				SKB_GSO_CB(nskb)->csum_start =
4854 					skb_headroom(nskb) + doffset;
4855 			} else {
4856 				if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4857 					goto err;
4858 			}
4859 			continue;
4860 		}
4861 
4862 		nskb_frag = skb_shinfo(nskb)->frags;
4863 
4864 		skb_copy_from_linear_data_offset(head_skb, offset,
4865 						 skb_put(nskb, hsize), hsize);
4866 
4867 		skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4868 					   SKBFL_SHARED_FRAG;
4869 
4870 		if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4871 			goto err;
4872 
4873 		while (pos < offset + len) {
4874 			if (i >= nfrags) {
4875 				if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
4876 				    skb_zerocopy_clone(nskb, list_skb,
4877 						       GFP_ATOMIC))
4878 					goto err;
4879 
4880 				i = 0;
4881 				nfrags = skb_shinfo(list_skb)->nr_frags;
4882 				frag = skb_shinfo(list_skb)->frags;
4883 				frag_skb = list_skb;
4884 				if (!skb_headlen(list_skb)) {
4885 					BUG_ON(!nfrags);
4886 				} else {
4887 					BUG_ON(!list_skb->head_frag);
4888 
4889 					/* to make room for head_frag. */
4890 					i--;
4891 					frag--;
4892 				}
4893 
4894 				list_skb = list_skb->next;
4895 			}
4896 
4897 			if (unlikely(skb_shinfo(nskb)->nr_frags >=
4898 				     MAX_SKB_FRAGS)) {
4899 				net_warn_ratelimited(
4900 					"skb_segment: too many frags: %u %u\n",
4901 					pos, mss);
4902 				err = -EINVAL;
4903 				goto err;
4904 			}
4905 
4906 			*nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4907 			__skb_frag_ref(nskb_frag);
4908 			size = skb_frag_size(nskb_frag);
4909 
4910 			if (pos < offset) {
4911 				skb_frag_off_add(nskb_frag, offset - pos);
4912 				skb_frag_size_sub(nskb_frag, offset - pos);
4913 			}
4914 
4915 			skb_shinfo(nskb)->nr_frags++;
4916 
4917 			if (pos + size <= offset + len) {
4918 				i++;
4919 				frag++;
4920 				pos += size;
4921 			} else {
4922 				skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4923 				goto skip_fraglist;
4924 			}
4925 
4926 			nskb_frag++;
4927 		}
4928 
4929 skip_fraglist:
4930 		nskb->data_len = len - hsize;
4931 		nskb->len += nskb->data_len;
4932 		nskb->truesize += nskb->data_len;
4933 
4934 perform_csum_check:
4935 		if (!csum) {
4936 			if (skb_has_shared_frag(nskb) &&
4937 			    __skb_linearize(nskb))
4938 				goto err;
4939 
4940 			if (!nskb->remcsum_offload)
4941 				nskb->ip_summed = CHECKSUM_NONE;
4942 			SKB_GSO_CB(nskb)->csum =
4943 				skb_checksum(nskb, doffset,
4944 					     nskb->len - doffset, 0);
4945 			SKB_GSO_CB(nskb)->csum_start =
4946 				skb_headroom(nskb) + doffset;
4947 		}
4948 	} while ((offset += len) < head_skb->len);
4949 
4950 	/* Some callers want to get the end of the list.
4951 	 * Put it in segs->prev to avoid walking the list.
4952 	 * (see validate_xmit_skb_list() for example)
4953 	 */
4954 	segs->prev = tail;
4955 
4956 	if (partial_segs) {
4957 		struct sk_buff *iter;
4958 		int type = skb_shinfo(head_skb)->gso_type;
4959 		unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4960 
4961 		/* Update type to add partial and then remove dodgy if set */
4962 		type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4963 		type &= ~SKB_GSO_DODGY;
4964 
4965 		/* Update GSO info and prepare to start updating headers on
4966 		 * our way back down the stack of protocols.
4967 		 */
4968 		for (iter = segs; iter; iter = iter->next) {
4969 			skb_shinfo(iter)->gso_size = gso_size;
4970 			skb_shinfo(iter)->gso_segs = partial_segs;
4971 			skb_shinfo(iter)->gso_type = type;
4972 			SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4973 		}
4974 
4975 		if (tail->len - doffset <= gso_size)
4976 			skb_shinfo(tail)->gso_size = 0;
4977 		else if (tail != segs)
4978 			skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4979 	}
4980 
4981 	/* Following permits correct backpressure, for protocols
4982 	 * using skb_set_owner_w().
4983 	 * Idea is to tranfert ownership from head_skb to last segment.
4984 	 */
4985 	if (head_skb->destructor == sock_wfree) {
4986 		swap(tail->truesize, head_skb->truesize);
4987 		swap(tail->destructor, head_skb->destructor);
4988 		swap(tail->sk, head_skb->sk);
4989 	}
4990 	return segs;
4991 
4992 err:
4993 	kfree_skb_list(segs);
4994 	return ERR_PTR(err);
4995 }
4996 EXPORT_SYMBOL_GPL(skb_segment);
4997 
4998 #ifdef CONFIG_SKB_EXTENSIONS
4999 #define SKB_EXT_ALIGN_VALUE	8
5000 #define SKB_EXT_CHUNKSIZEOF(x)	(ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
5001 
5002 static const u8 skb_ext_type_len[] = {
5003 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
5004 	[SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
5005 #endif
5006 #ifdef CONFIG_XFRM
5007 	[SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
5008 #endif
5009 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5010 	[TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
5011 #endif
5012 #if IS_ENABLED(CONFIG_MPTCP)
5013 	[SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
5014 #endif
5015 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
5016 	[SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
5017 #endif
5018 };
5019 
5020 static __always_inline unsigned int skb_ext_total_length(void)
5021 {
5022 	unsigned int l = SKB_EXT_CHUNKSIZEOF(struct skb_ext);
5023 	int i;
5024 
5025 	for (i = 0; i < ARRAY_SIZE(skb_ext_type_len); i++)
5026 		l += skb_ext_type_len[i];
5027 
5028 	return l;
5029 }
5030 
5031 static void skb_extensions_init(void)
5032 {
5033 	BUILD_BUG_ON(SKB_EXT_NUM >= 8);
5034 #if !IS_ENABLED(CONFIG_KCOV_INSTRUMENT_ALL)
5035 	BUILD_BUG_ON(skb_ext_total_length() > 255);
5036 #endif
5037 
5038 	skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
5039 					     SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
5040 					     0,
5041 					     SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5042 					     NULL);
5043 }
5044 #else
5045 static void skb_extensions_init(void) {}
5046 #endif
5047 
5048 /* The SKB kmem_cache slab is critical for network performance.  Never
5049  * merge/alias the slab with similar sized objects.  This avoids fragmentation
5050  * that hurts performance of kmem_cache_{alloc,free}_bulk APIs.
5051  */
5052 #ifndef CONFIG_SLUB_TINY
5053 #define FLAG_SKB_NO_MERGE	SLAB_NO_MERGE
5054 #else /* CONFIG_SLUB_TINY - simple loop in kmem_cache_alloc_bulk */
5055 #define FLAG_SKB_NO_MERGE	0
5056 #endif
5057 
5058 void __init skb_init(void)
5059 {
5060 	net_hotdata.skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
5061 					      sizeof(struct sk_buff),
5062 					      0,
5063 					      SLAB_HWCACHE_ALIGN|SLAB_PANIC|
5064 						FLAG_SKB_NO_MERGE,
5065 					      offsetof(struct sk_buff, cb),
5066 					      sizeof_field(struct sk_buff, cb),
5067 					      NULL);
5068 	net_hotdata.skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
5069 						sizeof(struct sk_buff_fclones),
5070 						0,
5071 						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5072 						NULL);
5073 	/* usercopy should only access first SKB_SMALL_HEAD_HEADROOM bytes.
5074 	 * struct skb_shared_info is located at the end of skb->head,
5075 	 * and should not be copied to/from user.
5076 	 */
5077 	net_hotdata.skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
5078 						SKB_SMALL_HEAD_CACHE_SIZE,
5079 						0,
5080 						SLAB_HWCACHE_ALIGN | SLAB_PANIC,
5081 						0,
5082 						SKB_SMALL_HEAD_HEADROOM,
5083 						NULL);
5084 	skb_extensions_init();
5085 }
5086 
5087 static int
5088 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
5089 	       unsigned int recursion_level)
5090 {
5091 	int start = skb_headlen(skb);
5092 	int i, copy = start - offset;
5093 	struct sk_buff *frag_iter;
5094 	int elt = 0;
5095 
5096 	if (unlikely(recursion_level >= 24))
5097 		return -EMSGSIZE;
5098 
5099 	if (copy > 0) {
5100 		if (copy > len)
5101 			copy = len;
5102 		sg_set_buf(sg, skb->data + offset, copy);
5103 		elt++;
5104 		if ((len -= copy) == 0)
5105 			return elt;
5106 		offset += copy;
5107 	}
5108 
5109 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
5110 		int end;
5111 
5112 		WARN_ON(start > offset + len);
5113 
5114 		end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
5115 		if ((copy = end - offset) > 0) {
5116 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
5117 			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5118 				return -EMSGSIZE;
5119 
5120 			if (copy > len)
5121 				copy = len;
5122 			sg_set_page(&sg[elt], skb_frag_page(frag), copy,
5123 				    skb_frag_off(frag) + offset - start);
5124 			elt++;
5125 			if (!(len -= copy))
5126 				return elt;
5127 			offset += copy;
5128 		}
5129 		start = end;
5130 	}
5131 
5132 	skb_walk_frags(skb, frag_iter) {
5133 		int end, ret;
5134 
5135 		WARN_ON(start > offset + len);
5136 
5137 		end = start + frag_iter->len;
5138 		if ((copy = end - offset) > 0) {
5139 			if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5140 				return -EMSGSIZE;
5141 
5142 			if (copy > len)
5143 				copy = len;
5144 			ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
5145 					      copy, recursion_level + 1);
5146 			if (unlikely(ret < 0))
5147 				return ret;
5148 			elt += ret;
5149 			if ((len -= copy) == 0)
5150 				return elt;
5151 			offset += copy;
5152 		}
5153 		start = end;
5154 	}
5155 	BUG_ON(len);
5156 	return elt;
5157 }
5158 
5159 /**
5160  *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
5161  *	@skb: Socket buffer containing the buffers to be mapped
5162  *	@sg: The scatter-gather list to map into
5163  *	@offset: The offset into the buffer's contents to start mapping
5164  *	@len: Length of buffer space to be mapped
5165  *
5166  *	Fill the specified scatter-gather list with mappings/pointers into a
5167  *	region of the buffer space attached to a socket buffer. Returns either
5168  *	the number of scatterlist items used, or -EMSGSIZE if the contents
5169  *	could not fit.
5170  */
5171 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
5172 {
5173 	int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
5174 
5175 	if (nsg <= 0)
5176 		return nsg;
5177 
5178 	sg_mark_end(&sg[nsg - 1]);
5179 
5180 	return nsg;
5181 }
5182 EXPORT_SYMBOL_GPL(skb_to_sgvec);
5183 
5184 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
5185  * sglist without mark the sg which contain last skb data as the end.
5186  * So the caller can mannipulate sg list as will when padding new data after
5187  * the first call without calling sg_unmark_end to expend sg list.
5188  *
5189  * Scenario to use skb_to_sgvec_nomark:
5190  * 1. sg_init_table
5191  * 2. skb_to_sgvec_nomark(payload1)
5192  * 3. skb_to_sgvec_nomark(payload2)
5193  *
5194  * This is equivalent to:
5195  * 1. sg_init_table
5196  * 2. skb_to_sgvec(payload1)
5197  * 3. sg_unmark_end
5198  * 4. skb_to_sgvec(payload2)
5199  *
5200  * When mapping multiple payload conditionally, skb_to_sgvec_nomark
5201  * is more preferable.
5202  */
5203 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
5204 			int offset, int len)
5205 {
5206 	return __skb_to_sgvec(skb, sg, offset, len, 0);
5207 }
5208 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
5209 
5210 
5211 
5212 /**
5213  *	skb_cow_data - Check that a socket buffer's data buffers are writable
5214  *	@skb: The socket buffer to check.
5215  *	@tailbits: Amount of trailing space to be added
5216  *	@trailer: Returned pointer to the skb where the @tailbits space begins
5217  *
5218  *	Make sure that the data buffers attached to a socket buffer are
5219  *	writable. If they are not, private copies are made of the data buffers
5220  *	and the socket buffer is set to use these instead.
5221  *
5222  *	If @tailbits is given, make sure that there is space to write @tailbits
5223  *	bytes of data beyond current end of socket buffer.  @trailer will be
5224  *	set to point to the skb in which this space begins.
5225  *
5226  *	The number of scatterlist elements required to completely map the
5227  *	COW'd and extended socket buffer will be returned.
5228  */
5229 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
5230 {
5231 	int copyflag;
5232 	int elt;
5233 	struct sk_buff *skb1, **skb_p;
5234 
5235 	/* If skb is cloned or its head is paged, reallocate
5236 	 * head pulling out all the pages (pages are considered not writable
5237 	 * at the moment even if they are anonymous).
5238 	 */
5239 	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
5240 	    !__pskb_pull_tail(skb, __skb_pagelen(skb)))
5241 		return -ENOMEM;
5242 
5243 	/* Easy case. Most of packets will go this way. */
5244 	if (!skb_has_frag_list(skb)) {
5245 		/* A little of trouble, not enough of space for trailer.
5246 		 * This should not happen, when stack is tuned to generate
5247 		 * good frames. OK, on miss we reallocate and reserve even more
5248 		 * space, 128 bytes is fair. */
5249 
5250 		if (skb_tailroom(skb) < tailbits &&
5251 		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
5252 			return -ENOMEM;
5253 
5254 		/* Voila! */
5255 		*trailer = skb;
5256 		return 1;
5257 	}
5258 
5259 	/* Misery. We are in troubles, going to mincer fragments... */
5260 
5261 	elt = 1;
5262 	skb_p = &skb_shinfo(skb)->frag_list;
5263 	copyflag = 0;
5264 
5265 	while ((skb1 = *skb_p) != NULL) {
5266 		int ntail = 0;
5267 
5268 		/* The fragment is partially pulled by someone,
5269 		 * this can happen on input. Copy it and everything
5270 		 * after it. */
5271 
5272 		if (skb_shared(skb1))
5273 			copyflag = 1;
5274 
5275 		/* If the skb is the last, worry about trailer. */
5276 
5277 		if (skb1->next == NULL && tailbits) {
5278 			if (skb_shinfo(skb1)->nr_frags ||
5279 			    skb_has_frag_list(skb1) ||
5280 			    skb_tailroom(skb1) < tailbits)
5281 				ntail = tailbits + 128;
5282 		}
5283 
5284 		if (copyflag ||
5285 		    skb_cloned(skb1) ||
5286 		    ntail ||
5287 		    skb_shinfo(skb1)->nr_frags ||
5288 		    skb_has_frag_list(skb1)) {
5289 			struct sk_buff *skb2;
5290 
5291 			/* Fuck, we are miserable poor guys... */
5292 			if (ntail == 0)
5293 				skb2 = skb_copy(skb1, GFP_ATOMIC);
5294 			else
5295 				skb2 = skb_copy_expand(skb1,
5296 						       skb_headroom(skb1),
5297 						       ntail,
5298 						       GFP_ATOMIC);
5299 			if (unlikely(skb2 == NULL))
5300 				return -ENOMEM;
5301 
5302 			if (skb1->sk)
5303 				skb_set_owner_w(skb2, skb1->sk);
5304 
5305 			/* Looking around. Are we still alive?
5306 			 * OK, link new skb, drop old one */
5307 
5308 			skb2->next = skb1->next;
5309 			*skb_p = skb2;
5310 			kfree_skb(skb1);
5311 			skb1 = skb2;
5312 		}
5313 		elt++;
5314 		*trailer = skb1;
5315 		skb_p = &skb1->next;
5316 	}
5317 
5318 	return elt;
5319 }
5320 EXPORT_SYMBOL_GPL(skb_cow_data);
5321 
5322 static void sock_rmem_free(struct sk_buff *skb)
5323 {
5324 	struct sock *sk = skb->sk;
5325 
5326 	atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
5327 }
5328 
5329 static void skb_set_err_queue(struct sk_buff *skb)
5330 {
5331 	/* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
5332 	 * So, it is safe to (mis)use it to mark skbs on the error queue.
5333 	 */
5334 	skb->pkt_type = PACKET_OUTGOING;
5335 	BUILD_BUG_ON(PACKET_OUTGOING == 0);
5336 }
5337 
5338 /*
5339  * Note: We dont mem charge error packets (no sk_forward_alloc changes)
5340  */
5341 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
5342 {
5343 	if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
5344 	    (unsigned int)READ_ONCE(sk->sk_rcvbuf))
5345 		return -ENOMEM;
5346 
5347 	skb_orphan(skb);
5348 	skb->sk = sk;
5349 	skb->destructor = sock_rmem_free;
5350 	atomic_add(skb->truesize, &sk->sk_rmem_alloc);
5351 	skb_set_err_queue(skb);
5352 
5353 	/* before exiting rcu section, make sure dst is refcounted */
5354 	skb_dst_force(skb);
5355 
5356 	skb_queue_tail(&sk->sk_error_queue, skb);
5357 	if (!sock_flag(sk, SOCK_DEAD))
5358 		sk_error_report(sk);
5359 	return 0;
5360 }
5361 EXPORT_SYMBOL(sock_queue_err_skb);
5362 
5363 static bool is_icmp_err_skb(const struct sk_buff *skb)
5364 {
5365 	return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
5366 		       SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
5367 }
5368 
5369 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
5370 {
5371 	struct sk_buff_head *q = &sk->sk_error_queue;
5372 	struct sk_buff *skb, *skb_next = NULL;
5373 	bool icmp_next = false;
5374 	unsigned long flags;
5375 
5376 	if (skb_queue_empty_lockless(q))
5377 		return NULL;
5378 
5379 	spin_lock_irqsave(&q->lock, flags);
5380 	skb = __skb_dequeue(q);
5381 	if (skb && (skb_next = skb_peek(q))) {
5382 		icmp_next = is_icmp_err_skb(skb_next);
5383 		if (icmp_next)
5384 			sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
5385 	}
5386 	spin_unlock_irqrestore(&q->lock, flags);
5387 
5388 	if (is_icmp_err_skb(skb) && !icmp_next)
5389 		sk->sk_err = 0;
5390 
5391 	if (skb_next)
5392 		sk_error_report(sk);
5393 
5394 	return skb;
5395 }
5396 EXPORT_SYMBOL(sock_dequeue_err_skb);
5397 
5398 /**
5399  * skb_clone_sk - create clone of skb, and take reference to socket
5400  * @skb: the skb to clone
5401  *
5402  * This function creates a clone of a buffer that holds a reference on
5403  * sk_refcnt.  Buffers created via this function are meant to be
5404  * returned using sock_queue_err_skb, or free via kfree_skb.
5405  *
5406  * When passing buffers allocated with this function to sock_queue_err_skb
5407  * it is necessary to wrap the call with sock_hold/sock_put in order to
5408  * prevent the socket from being released prior to being enqueued on
5409  * the sk_error_queue.
5410  */
5411 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
5412 {
5413 	struct sock *sk = skb->sk;
5414 	struct sk_buff *clone;
5415 
5416 	if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
5417 		return NULL;
5418 
5419 	clone = skb_clone(skb, GFP_ATOMIC);
5420 	if (!clone) {
5421 		sock_put(sk);
5422 		return NULL;
5423 	}
5424 
5425 	clone->sk = sk;
5426 	clone->destructor = sock_efree;
5427 
5428 	return clone;
5429 }
5430 EXPORT_SYMBOL(skb_clone_sk);
5431 
5432 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
5433 					struct sock *sk,
5434 					int tstype,
5435 					bool opt_stats)
5436 {
5437 	struct sock_exterr_skb *serr;
5438 	int err;
5439 
5440 	BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
5441 
5442 	serr = SKB_EXT_ERR(skb);
5443 	memset(serr, 0, sizeof(*serr));
5444 	serr->ee.ee_errno = ENOMSG;
5445 	serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
5446 	serr->ee.ee_info = tstype;
5447 	serr->opt_stats = opt_stats;
5448 	serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
5449 	if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
5450 		serr->ee.ee_data = skb_shinfo(skb)->tskey;
5451 		if (sk_is_tcp(sk))
5452 			serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
5453 	}
5454 
5455 	err = sock_queue_err_skb(sk, skb);
5456 
5457 	if (err)
5458 		kfree_skb(skb);
5459 }
5460 
5461 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
5462 {
5463 	bool ret;
5464 
5465 	if (likely(READ_ONCE(sysctl_tstamp_allow_data) || tsonly))
5466 		return true;
5467 
5468 	read_lock_bh(&sk->sk_callback_lock);
5469 	ret = sk->sk_socket && sk->sk_socket->file &&
5470 	      file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
5471 	read_unlock_bh(&sk->sk_callback_lock);
5472 	return ret;
5473 }
5474 
5475 void skb_complete_tx_timestamp(struct sk_buff *skb,
5476 			       struct skb_shared_hwtstamps *hwtstamps)
5477 {
5478 	struct sock *sk = skb->sk;
5479 
5480 	if (!skb_may_tx_timestamp(sk, false))
5481 		goto err;
5482 
5483 	/* Take a reference to prevent skb_orphan() from freeing the socket,
5484 	 * but only if the socket refcount is not zero.
5485 	 */
5486 	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5487 		*skb_hwtstamps(skb) = *hwtstamps;
5488 		__skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5489 		sock_put(sk);
5490 		return;
5491 	}
5492 
5493 err:
5494 	kfree_skb(skb);
5495 }
5496 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5497 
5498 void __skb_tstamp_tx(struct sk_buff *orig_skb,
5499 		     const struct sk_buff *ack_skb,
5500 		     struct skb_shared_hwtstamps *hwtstamps,
5501 		     struct sock *sk, int tstype)
5502 {
5503 	struct sk_buff *skb;
5504 	bool tsonly, opt_stats = false;
5505 	u32 tsflags;
5506 
5507 	if (!sk)
5508 		return;
5509 
5510 	tsflags = READ_ONCE(sk->sk_tsflags);
5511 	if (!hwtstamps && !(tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5512 	    skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5513 		return;
5514 
5515 	tsonly = tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5516 	if (!skb_may_tx_timestamp(sk, tsonly))
5517 		return;
5518 
5519 	if (tsonly) {
5520 #ifdef CONFIG_INET
5521 		if ((tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5522 		    sk_is_tcp(sk)) {
5523 			skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5524 							     ack_skb);
5525 			opt_stats = true;
5526 		} else
5527 #endif
5528 			skb = alloc_skb(0, GFP_ATOMIC);
5529 	} else {
5530 		skb = skb_clone(orig_skb, GFP_ATOMIC);
5531 
5532 		if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
5533 			kfree_skb(skb);
5534 			return;
5535 		}
5536 	}
5537 	if (!skb)
5538 		return;
5539 
5540 	if (tsonly) {
5541 		skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5542 					     SKBTX_ANY_TSTAMP;
5543 		skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5544 	}
5545 
5546 	if (hwtstamps)
5547 		*skb_hwtstamps(skb) = *hwtstamps;
5548 	else
5549 		__net_timestamp(skb);
5550 
5551 	__skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5552 }
5553 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5554 
5555 void skb_tstamp_tx(struct sk_buff *orig_skb,
5556 		   struct skb_shared_hwtstamps *hwtstamps)
5557 {
5558 	return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5559 			       SCM_TSTAMP_SND);
5560 }
5561 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5562 
5563 #ifdef CONFIG_WIRELESS
5564 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5565 {
5566 	struct sock *sk = skb->sk;
5567 	struct sock_exterr_skb *serr;
5568 	int err = 1;
5569 
5570 	skb->wifi_acked_valid = 1;
5571 	skb->wifi_acked = acked;
5572 
5573 	serr = SKB_EXT_ERR(skb);
5574 	memset(serr, 0, sizeof(*serr));
5575 	serr->ee.ee_errno = ENOMSG;
5576 	serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5577 
5578 	/* Take a reference to prevent skb_orphan() from freeing the socket,
5579 	 * but only if the socket refcount is not zero.
5580 	 */
5581 	if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5582 		err = sock_queue_err_skb(sk, skb);
5583 		sock_put(sk);
5584 	}
5585 	if (err)
5586 		kfree_skb(skb);
5587 }
5588 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5589 #endif /* CONFIG_WIRELESS */
5590 
5591 /**
5592  * skb_partial_csum_set - set up and verify partial csum values for packet
5593  * @skb: the skb to set
5594  * @start: the number of bytes after skb->data to start checksumming.
5595  * @off: the offset from start to place the checksum.
5596  *
5597  * For untrusted partially-checksummed packets, we need to make sure the values
5598  * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5599  *
5600  * This function checks and sets those values and skb->ip_summed: if this
5601  * returns false you should drop the packet.
5602  */
5603 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5604 {
5605 	u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5606 	u32 csum_start = skb_headroom(skb) + (u32)start;
5607 
5608 	if (unlikely(csum_start >= U16_MAX || csum_end > skb_headlen(skb))) {
5609 		net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5610 				     start, off, skb_headroom(skb), skb_headlen(skb));
5611 		return false;
5612 	}
5613 	skb->ip_summed = CHECKSUM_PARTIAL;
5614 	skb->csum_start = csum_start;
5615 	skb->csum_offset = off;
5616 	skb->transport_header = csum_start;
5617 	return true;
5618 }
5619 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5620 
5621 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5622 			       unsigned int max)
5623 {
5624 	if (skb_headlen(skb) >= len)
5625 		return 0;
5626 
5627 	/* If we need to pullup then pullup to the max, so we
5628 	 * won't need to do it again.
5629 	 */
5630 	if (max > skb->len)
5631 		max = skb->len;
5632 
5633 	if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5634 		return -ENOMEM;
5635 
5636 	if (skb_headlen(skb) < len)
5637 		return -EPROTO;
5638 
5639 	return 0;
5640 }
5641 
5642 #define MAX_TCP_HDR_LEN (15 * 4)
5643 
5644 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5645 				      typeof(IPPROTO_IP) proto,
5646 				      unsigned int off)
5647 {
5648 	int err;
5649 
5650 	switch (proto) {
5651 	case IPPROTO_TCP:
5652 		err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5653 					  off + MAX_TCP_HDR_LEN);
5654 		if (!err && !skb_partial_csum_set(skb, off,
5655 						  offsetof(struct tcphdr,
5656 							   check)))
5657 			err = -EPROTO;
5658 		return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5659 
5660 	case IPPROTO_UDP:
5661 		err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5662 					  off + sizeof(struct udphdr));
5663 		if (!err && !skb_partial_csum_set(skb, off,
5664 						  offsetof(struct udphdr,
5665 							   check)))
5666 			err = -EPROTO;
5667 		return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5668 	}
5669 
5670 	return ERR_PTR(-EPROTO);
5671 }
5672 
5673 /* This value should be large enough to cover a tagged ethernet header plus
5674  * maximally sized IP and TCP or UDP headers.
5675  */
5676 #define MAX_IP_HDR_LEN 128
5677 
5678 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5679 {
5680 	unsigned int off;
5681 	bool fragment;
5682 	__sum16 *csum;
5683 	int err;
5684 
5685 	fragment = false;
5686 
5687 	err = skb_maybe_pull_tail(skb,
5688 				  sizeof(struct iphdr),
5689 				  MAX_IP_HDR_LEN);
5690 	if (err < 0)
5691 		goto out;
5692 
5693 	if (ip_is_fragment(ip_hdr(skb)))
5694 		fragment = true;
5695 
5696 	off = ip_hdrlen(skb);
5697 
5698 	err = -EPROTO;
5699 
5700 	if (fragment)
5701 		goto out;
5702 
5703 	csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5704 	if (IS_ERR(csum))
5705 		return PTR_ERR(csum);
5706 
5707 	if (recalculate)
5708 		*csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5709 					   ip_hdr(skb)->daddr,
5710 					   skb->len - off,
5711 					   ip_hdr(skb)->protocol, 0);
5712 	err = 0;
5713 
5714 out:
5715 	return err;
5716 }
5717 
5718 /* This value should be large enough to cover a tagged ethernet header plus
5719  * an IPv6 header, all options, and a maximal TCP or UDP header.
5720  */
5721 #define MAX_IPV6_HDR_LEN 256
5722 
5723 #define OPT_HDR(type, skb, off) \
5724 	(type *)(skb_network_header(skb) + (off))
5725 
5726 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5727 {
5728 	int err;
5729 	u8 nexthdr;
5730 	unsigned int off;
5731 	unsigned int len;
5732 	bool fragment;
5733 	bool done;
5734 	__sum16 *csum;
5735 
5736 	fragment = false;
5737 	done = false;
5738 
5739 	off = sizeof(struct ipv6hdr);
5740 
5741 	err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5742 	if (err < 0)
5743 		goto out;
5744 
5745 	nexthdr = ipv6_hdr(skb)->nexthdr;
5746 
5747 	len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5748 	while (off <= len && !done) {
5749 		switch (nexthdr) {
5750 		case IPPROTO_DSTOPTS:
5751 		case IPPROTO_HOPOPTS:
5752 		case IPPROTO_ROUTING: {
5753 			struct ipv6_opt_hdr *hp;
5754 
5755 			err = skb_maybe_pull_tail(skb,
5756 						  off +
5757 						  sizeof(struct ipv6_opt_hdr),
5758 						  MAX_IPV6_HDR_LEN);
5759 			if (err < 0)
5760 				goto out;
5761 
5762 			hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5763 			nexthdr = hp->nexthdr;
5764 			off += ipv6_optlen(hp);
5765 			break;
5766 		}
5767 		case IPPROTO_AH: {
5768 			struct ip_auth_hdr *hp;
5769 
5770 			err = skb_maybe_pull_tail(skb,
5771 						  off +
5772 						  sizeof(struct ip_auth_hdr),
5773 						  MAX_IPV6_HDR_LEN);
5774 			if (err < 0)
5775 				goto out;
5776 
5777 			hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5778 			nexthdr = hp->nexthdr;
5779 			off += ipv6_authlen(hp);
5780 			break;
5781 		}
5782 		case IPPROTO_FRAGMENT: {
5783 			struct frag_hdr *hp;
5784 
5785 			err = skb_maybe_pull_tail(skb,
5786 						  off +
5787 						  sizeof(struct frag_hdr),
5788 						  MAX_IPV6_HDR_LEN);
5789 			if (err < 0)
5790 				goto out;
5791 
5792 			hp = OPT_HDR(struct frag_hdr, skb, off);
5793 
5794 			if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5795 				fragment = true;
5796 
5797 			nexthdr = hp->nexthdr;
5798 			off += sizeof(struct frag_hdr);
5799 			break;
5800 		}
5801 		default:
5802 			done = true;
5803 			break;
5804 		}
5805 	}
5806 
5807 	err = -EPROTO;
5808 
5809 	if (!done || fragment)
5810 		goto out;
5811 
5812 	csum = skb_checksum_setup_ip(skb, nexthdr, off);
5813 	if (IS_ERR(csum))
5814 		return PTR_ERR(csum);
5815 
5816 	if (recalculate)
5817 		*csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5818 					 &ipv6_hdr(skb)->daddr,
5819 					 skb->len - off, nexthdr, 0);
5820 	err = 0;
5821 
5822 out:
5823 	return err;
5824 }
5825 
5826 /**
5827  * skb_checksum_setup - set up partial checksum offset
5828  * @skb: the skb to set up
5829  * @recalculate: if true the pseudo-header checksum will be recalculated
5830  */
5831 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5832 {
5833 	int err;
5834 
5835 	switch (skb->protocol) {
5836 	case htons(ETH_P_IP):
5837 		err = skb_checksum_setup_ipv4(skb, recalculate);
5838 		break;
5839 
5840 	case htons(ETH_P_IPV6):
5841 		err = skb_checksum_setup_ipv6(skb, recalculate);
5842 		break;
5843 
5844 	default:
5845 		err = -EPROTO;
5846 		break;
5847 	}
5848 
5849 	return err;
5850 }
5851 EXPORT_SYMBOL(skb_checksum_setup);
5852 
5853 /**
5854  * skb_checksum_maybe_trim - maybe trims the given skb
5855  * @skb: the skb to check
5856  * @transport_len: the data length beyond the network header
5857  *
5858  * Checks whether the given skb has data beyond the given transport length.
5859  * If so, returns a cloned skb trimmed to this transport length.
5860  * Otherwise returns the provided skb. Returns NULL in error cases
5861  * (e.g. transport_len exceeds skb length or out-of-memory).
5862  *
5863  * Caller needs to set the skb transport header and free any returned skb if it
5864  * differs from the provided skb.
5865  */
5866 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5867 					       unsigned int transport_len)
5868 {
5869 	struct sk_buff *skb_chk;
5870 	unsigned int len = skb_transport_offset(skb) + transport_len;
5871 	int ret;
5872 
5873 	if (skb->len < len)
5874 		return NULL;
5875 	else if (skb->len == len)
5876 		return skb;
5877 
5878 	skb_chk = skb_clone(skb, GFP_ATOMIC);
5879 	if (!skb_chk)
5880 		return NULL;
5881 
5882 	ret = pskb_trim_rcsum(skb_chk, len);
5883 	if (ret) {
5884 		kfree_skb(skb_chk);
5885 		return NULL;
5886 	}
5887 
5888 	return skb_chk;
5889 }
5890 
5891 /**
5892  * skb_checksum_trimmed - validate checksum of an skb
5893  * @skb: the skb to check
5894  * @transport_len: the data length beyond the network header
5895  * @skb_chkf: checksum function to use
5896  *
5897  * Applies the given checksum function skb_chkf to the provided skb.
5898  * Returns a checked and maybe trimmed skb. Returns NULL on error.
5899  *
5900  * If the skb has data beyond the given transport length, then a
5901  * trimmed & cloned skb is checked and returned.
5902  *
5903  * Caller needs to set the skb transport header and free any returned skb if it
5904  * differs from the provided skb.
5905  */
5906 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5907 				     unsigned int transport_len,
5908 				     __sum16(*skb_chkf)(struct sk_buff *skb))
5909 {
5910 	struct sk_buff *skb_chk;
5911 	unsigned int offset = skb_transport_offset(skb);
5912 	__sum16 ret;
5913 
5914 	skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5915 	if (!skb_chk)
5916 		goto err;
5917 
5918 	if (!pskb_may_pull(skb_chk, offset))
5919 		goto err;
5920 
5921 	skb_pull_rcsum(skb_chk, offset);
5922 	ret = skb_chkf(skb_chk);
5923 	skb_push_rcsum(skb_chk, offset);
5924 
5925 	if (ret)
5926 		goto err;
5927 
5928 	return skb_chk;
5929 
5930 err:
5931 	if (skb_chk && skb_chk != skb)
5932 		kfree_skb(skb_chk);
5933 
5934 	return NULL;
5935 
5936 }
5937 EXPORT_SYMBOL(skb_checksum_trimmed);
5938 
5939 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5940 {
5941 	net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5942 			     skb->dev->name);
5943 }
5944 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5945 
5946 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5947 {
5948 	if (head_stolen) {
5949 		skb_release_head_state(skb);
5950 		kmem_cache_free(net_hotdata.skbuff_cache, skb);
5951 	} else {
5952 		__kfree_skb(skb);
5953 	}
5954 }
5955 EXPORT_SYMBOL(kfree_skb_partial);
5956 
5957 /**
5958  * skb_try_coalesce - try to merge skb to prior one
5959  * @to: prior buffer
5960  * @from: buffer to add
5961  * @fragstolen: pointer to boolean
5962  * @delta_truesize: how much more was allocated than was requested
5963  */
5964 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5965 		      bool *fragstolen, int *delta_truesize)
5966 {
5967 	struct skb_shared_info *to_shinfo, *from_shinfo;
5968 	int i, delta, len = from->len;
5969 
5970 	*fragstolen = false;
5971 
5972 	if (skb_cloned(to))
5973 		return false;
5974 
5975 	/* In general, avoid mixing page_pool and non-page_pool allocated
5976 	 * pages within the same SKB. In theory we could take full
5977 	 * references if @from is cloned and !@to->pp_recycle but its
5978 	 * tricky (due to potential race with the clone disappearing) and
5979 	 * rare, so not worth dealing with.
5980 	 */
5981 	if (to->pp_recycle != from->pp_recycle)
5982 		return false;
5983 
5984 	if (len <= skb_tailroom(to)) {
5985 		if (len)
5986 			BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5987 		*delta_truesize = 0;
5988 		return true;
5989 	}
5990 
5991 	to_shinfo = skb_shinfo(to);
5992 	from_shinfo = skb_shinfo(from);
5993 	if (to_shinfo->frag_list || from_shinfo->frag_list)
5994 		return false;
5995 	if (skb_zcopy(to) || skb_zcopy(from))
5996 		return false;
5997 
5998 	if (skb_headlen(from) != 0) {
5999 		struct page *page;
6000 		unsigned int offset;
6001 
6002 		if (to_shinfo->nr_frags +
6003 		    from_shinfo->nr_frags >= MAX_SKB_FRAGS)
6004 			return false;
6005 
6006 		if (skb_head_is_locked(from))
6007 			return false;
6008 
6009 		delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
6010 
6011 		page = virt_to_head_page(from->head);
6012 		offset = from->data - (unsigned char *)page_address(page);
6013 
6014 		skb_fill_page_desc(to, to_shinfo->nr_frags,
6015 				   page, offset, skb_headlen(from));
6016 		*fragstolen = true;
6017 	} else {
6018 		if (to_shinfo->nr_frags +
6019 		    from_shinfo->nr_frags > MAX_SKB_FRAGS)
6020 			return false;
6021 
6022 		delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
6023 	}
6024 
6025 	WARN_ON_ONCE(delta < len);
6026 
6027 	memcpy(to_shinfo->frags + to_shinfo->nr_frags,
6028 	       from_shinfo->frags,
6029 	       from_shinfo->nr_frags * sizeof(skb_frag_t));
6030 	to_shinfo->nr_frags += from_shinfo->nr_frags;
6031 
6032 	if (!skb_cloned(from))
6033 		from_shinfo->nr_frags = 0;
6034 
6035 	/* if the skb is not cloned this does nothing
6036 	 * since we set nr_frags to 0.
6037 	 */
6038 	if (skb_pp_frag_ref(from)) {
6039 		for (i = 0; i < from_shinfo->nr_frags; i++)
6040 			__skb_frag_ref(&from_shinfo->frags[i]);
6041 	}
6042 
6043 	to->truesize += delta;
6044 	to->len += len;
6045 	to->data_len += len;
6046 
6047 	*delta_truesize = delta;
6048 	return true;
6049 }
6050 EXPORT_SYMBOL(skb_try_coalesce);
6051 
6052 /**
6053  * skb_scrub_packet - scrub an skb
6054  *
6055  * @skb: buffer to clean
6056  * @xnet: packet is crossing netns
6057  *
6058  * skb_scrub_packet can be used after encapsulating or decapsulating a packet
6059  * into/from a tunnel. Some information have to be cleared during these
6060  * operations.
6061  * skb_scrub_packet can also be used to clean a skb before injecting it in
6062  * another namespace (@xnet == true). We have to clear all information in the
6063  * skb that could impact namespace isolation.
6064  */
6065 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
6066 {
6067 	skb->pkt_type = PACKET_HOST;
6068 	skb->skb_iif = 0;
6069 	skb->ignore_df = 0;
6070 	skb_dst_drop(skb);
6071 	skb_ext_reset(skb);
6072 	nf_reset_ct(skb);
6073 	nf_reset_trace(skb);
6074 
6075 #ifdef CONFIG_NET_SWITCHDEV
6076 	skb->offload_fwd_mark = 0;
6077 	skb->offload_l3_fwd_mark = 0;
6078 #endif
6079 
6080 	if (!xnet)
6081 		return;
6082 
6083 	ipvs_reset(skb);
6084 	skb->mark = 0;
6085 	skb_clear_tstamp(skb);
6086 }
6087 EXPORT_SYMBOL_GPL(skb_scrub_packet);
6088 
6089 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
6090 {
6091 	int mac_len, meta_len;
6092 	void *meta;
6093 
6094 	if (skb_cow(skb, skb_headroom(skb)) < 0) {
6095 		kfree_skb(skb);
6096 		return NULL;
6097 	}
6098 
6099 	mac_len = skb->data - skb_mac_header(skb);
6100 	if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
6101 		memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
6102 			mac_len - VLAN_HLEN - ETH_TLEN);
6103 	}
6104 
6105 	meta_len = skb_metadata_len(skb);
6106 	if (meta_len) {
6107 		meta = skb_metadata_end(skb) - meta_len;
6108 		memmove(meta + VLAN_HLEN, meta, meta_len);
6109 	}
6110 
6111 	skb->mac_header += VLAN_HLEN;
6112 	return skb;
6113 }
6114 
6115 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
6116 {
6117 	struct vlan_hdr *vhdr;
6118 	u16 vlan_tci;
6119 
6120 	if (unlikely(skb_vlan_tag_present(skb))) {
6121 		/* vlan_tci is already set-up so leave this for another time */
6122 		return skb;
6123 	}
6124 
6125 	skb = skb_share_check(skb, GFP_ATOMIC);
6126 	if (unlikely(!skb))
6127 		goto err_free;
6128 	/* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
6129 	if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
6130 		goto err_free;
6131 
6132 	vhdr = (struct vlan_hdr *)skb->data;
6133 	vlan_tci = ntohs(vhdr->h_vlan_TCI);
6134 	__vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
6135 
6136 	skb_pull_rcsum(skb, VLAN_HLEN);
6137 	vlan_set_encap_proto(skb, vhdr);
6138 
6139 	skb = skb_reorder_vlan_header(skb);
6140 	if (unlikely(!skb))
6141 		goto err_free;
6142 
6143 	skb_reset_network_header(skb);
6144 	if (!skb_transport_header_was_set(skb))
6145 		skb_reset_transport_header(skb);
6146 	skb_reset_mac_len(skb);
6147 
6148 	return skb;
6149 
6150 err_free:
6151 	kfree_skb(skb);
6152 	return NULL;
6153 }
6154 EXPORT_SYMBOL(skb_vlan_untag);
6155 
6156 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
6157 {
6158 	if (!pskb_may_pull(skb, write_len))
6159 		return -ENOMEM;
6160 
6161 	if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
6162 		return 0;
6163 
6164 	return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6165 }
6166 EXPORT_SYMBOL(skb_ensure_writable);
6167 
6168 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev)
6169 {
6170 	int needed_headroom = dev->needed_headroom;
6171 	int needed_tailroom = dev->needed_tailroom;
6172 
6173 	/* For tail taggers, we need to pad short frames ourselves, to ensure
6174 	 * that the tail tag does not fail at its role of being at the end of
6175 	 * the packet, once the conduit interface pads the frame. Account for
6176 	 * that pad length here, and pad later.
6177 	 */
6178 	if (unlikely(needed_tailroom && skb->len < ETH_ZLEN))
6179 		needed_tailroom += ETH_ZLEN - skb->len;
6180 	/* skb_headroom() returns unsigned int... */
6181 	needed_headroom = max_t(int, needed_headroom - skb_headroom(skb), 0);
6182 	needed_tailroom = max_t(int, needed_tailroom - skb_tailroom(skb), 0);
6183 
6184 	if (likely(!needed_headroom && !needed_tailroom && !skb_cloned(skb)))
6185 		/* No reallocation needed, yay! */
6186 		return 0;
6187 
6188 	return pskb_expand_head(skb, needed_headroom, needed_tailroom,
6189 				GFP_ATOMIC);
6190 }
6191 EXPORT_SYMBOL(skb_ensure_writable_head_tail);
6192 
6193 /* remove VLAN header from packet and update csum accordingly.
6194  * expects a non skb_vlan_tag_present skb with a vlan tag payload
6195  */
6196 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
6197 {
6198 	int offset = skb->data - skb_mac_header(skb);
6199 	int err;
6200 
6201 	if (WARN_ONCE(offset,
6202 		      "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
6203 		      offset)) {
6204 		return -EINVAL;
6205 	}
6206 
6207 	err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
6208 	if (unlikely(err))
6209 		return err;
6210 
6211 	skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6212 
6213 	vlan_remove_tag(skb, vlan_tci);
6214 
6215 	skb->mac_header += VLAN_HLEN;
6216 
6217 	if (skb_network_offset(skb) < ETH_HLEN)
6218 		skb_set_network_header(skb, ETH_HLEN);
6219 
6220 	skb_reset_mac_len(skb);
6221 
6222 	return err;
6223 }
6224 EXPORT_SYMBOL(__skb_vlan_pop);
6225 
6226 /* Pop a vlan tag either from hwaccel or from payload.
6227  * Expects skb->data at mac header.
6228  */
6229 int skb_vlan_pop(struct sk_buff *skb)
6230 {
6231 	u16 vlan_tci;
6232 	__be16 vlan_proto;
6233 	int err;
6234 
6235 	if (likely(skb_vlan_tag_present(skb))) {
6236 		__vlan_hwaccel_clear_tag(skb);
6237 	} else {
6238 		if (unlikely(!eth_type_vlan(skb->protocol)))
6239 			return 0;
6240 
6241 		err = __skb_vlan_pop(skb, &vlan_tci);
6242 		if (err)
6243 			return err;
6244 	}
6245 	/* move next vlan tag to hw accel tag */
6246 	if (likely(!eth_type_vlan(skb->protocol)))
6247 		return 0;
6248 
6249 	vlan_proto = skb->protocol;
6250 	err = __skb_vlan_pop(skb, &vlan_tci);
6251 	if (unlikely(err))
6252 		return err;
6253 
6254 	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6255 	return 0;
6256 }
6257 EXPORT_SYMBOL(skb_vlan_pop);
6258 
6259 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
6260  * Expects skb->data at mac header.
6261  */
6262 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
6263 {
6264 	if (skb_vlan_tag_present(skb)) {
6265 		int offset = skb->data - skb_mac_header(skb);
6266 		int err;
6267 
6268 		if (WARN_ONCE(offset,
6269 			      "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
6270 			      offset)) {
6271 			return -EINVAL;
6272 		}
6273 
6274 		err = __vlan_insert_tag(skb, skb->vlan_proto,
6275 					skb_vlan_tag_get(skb));
6276 		if (err)
6277 			return err;
6278 
6279 		skb->protocol = skb->vlan_proto;
6280 		skb->network_header -= VLAN_HLEN;
6281 
6282 		skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6283 	}
6284 	__vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6285 	return 0;
6286 }
6287 EXPORT_SYMBOL(skb_vlan_push);
6288 
6289 /**
6290  * skb_eth_pop() - Drop the Ethernet header at the head of a packet
6291  *
6292  * @skb: Socket buffer to modify
6293  *
6294  * Drop the Ethernet header of @skb.
6295  *
6296  * Expects that skb->data points to the mac header and that no VLAN tags are
6297  * present.
6298  *
6299  * Returns 0 on success, -errno otherwise.
6300  */
6301 int skb_eth_pop(struct sk_buff *skb)
6302 {
6303 	if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
6304 	    skb_network_offset(skb) < ETH_HLEN)
6305 		return -EPROTO;
6306 
6307 	skb_pull_rcsum(skb, ETH_HLEN);
6308 	skb_reset_mac_header(skb);
6309 	skb_reset_mac_len(skb);
6310 
6311 	return 0;
6312 }
6313 EXPORT_SYMBOL(skb_eth_pop);
6314 
6315 /**
6316  * skb_eth_push() - Add a new Ethernet header at the head of a packet
6317  *
6318  * @skb: Socket buffer to modify
6319  * @dst: Destination MAC address of the new header
6320  * @src: Source MAC address of the new header
6321  *
6322  * Prepend @skb with a new Ethernet header.
6323  *
6324  * Expects that skb->data points to the mac header, which must be empty.
6325  *
6326  * Returns 0 on success, -errno otherwise.
6327  */
6328 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
6329 		 const unsigned char *src)
6330 {
6331 	struct ethhdr *eth;
6332 	int err;
6333 
6334 	if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
6335 		return -EPROTO;
6336 
6337 	err = skb_cow_head(skb, sizeof(*eth));
6338 	if (err < 0)
6339 		return err;
6340 
6341 	skb_push(skb, sizeof(*eth));
6342 	skb_reset_mac_header(skb);
6343 	skb_reset_mac_len(skb);
6344 
6345 	eth = eth_hdr(skb);
6346 	ether_addr_copy(eth->h_dest, dst);
6347 	ether_addr_copy(eth->h_source, src);
6348 	eth->h_proto = skb->protocol;
6349 
6350 	skb_postpush_rcsum(skb, eth, sizeof(*eth));
6351 
6352 	return 0;
6353 }
6354 EXPORT_SYMBOL(skb_eth_push);
6355 
6356 /* Update the ethertype of hdr and the skb csum value if required. */
6357 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
6358 			     __be16 ethertype)
6359 {
6360 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6361 		__be16 diff[] = { ~hdr->h_proto, ethertype };
6362 
6363 		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6364 	}
6365 
6366 	hdr->h_proto = ethertype;
6367 }
6368 
6369 /**
6370  * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6371  *                   the packet
6372  *
6373  * @skb: buffer
6374  * @mpls_lse: MPLS label stack entry to push
6375  * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6376  * @mac_len: length of the MAC header
6377  * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6378  *            ethernet
6379  *
6380  * Expects skb->data at mac header.
6381  *
6382  * Returns 0 on success, -errno otherwise.
6383  */
6384 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6385 		  int mac_len, bool ethernet)
6386 {
6387 	struct mpls_shim_hdr *lse;
6388 	int err;
6389 
6390 	if (unlikely(!eth_p_mpls(mpls_proto)))
6391 		return -EINVAL;
6392 
6393 	/* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6394 	if (skb->encapsulation)
6395 		return -EINVAL;
6396 
6397 	err = skb_cow_head(skb, MPLS_HLEN);
6398 	if (unlikely(err))
6399 		return err;
6400 
6401 	if (!skb->inner_protocol) {
6402 		skb_set_inner_network_header(skb, skb_network_offset(skb));
6403 		skb_set_inner_protocol(skb, skb->protocol);
6404 	}
6405 
6406 	skb_push(skb, MPLS_HLEN);
6407 	memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6408 		mac_len);
6409 	skb_reset_mac_header(skb);
6410 	skb_set_network_header(skb, mac_len);
6411 	skb_reset_mac_len(skb);
6412 
6413 	lse = mpls_hdr(skb);
6414 	lse->label_stack_entry = mpls_lse;
6415 	skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6416 
6417 	if (ethernet && mac_len >= ETH_HLEN)
6418 		skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6419 	skb->protocol = mpls_proto;
6420 
6421 	return 0;
6422 }
6423 EXPORT_SYMBOL_GPL(skb_mpls_push);
6424 
6425 /**
6426  * skb_mpls_pop() - pop the outermost MPLS header
6427  *
6428  * @skb: buffer
6429  * @next_proto: ethertype of header after popped MPLS header
6430  * @mac_len: length of the MAC header
6431  * @ethernet: flag to indicate if the packet is ethernet
6432  *
6433  * Expects skb->data at mac header.
6434  *
6435  * Returns 0 on success, -errno otherwise.
6436  */
6437 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6438 		 bool ethernet)
6439 {
6440 	int err;
6441 
6442 	if (unlikely(!eth_p_mpls(skb->protocol)))
6443 		return 0;
6444 
6445 	err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6446 	if (unlikely(err))
6447 		return err;
6448 
6449 	skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6450 	memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6451 		mac_len);
6452 
6453 	__skb_pull(skb, MPLS_HLEN);
6454 	skb_reset_mac_header(skb);
6455 	skb_set_network_header(skb, mac_len);
6456 
6457 	if (ethernet && mac_len >= ETH_HLEN) {
6458 		struct ethhdr *hdr;
6459 
6460 		/* use mpls_hdr() to get ethertype to account for VLANs. */
6461 		hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6462 		skb_mod_eth_type(skb, hdr, next_proto);
6463 	}
6464 	skb->protocol = next_proto;
6465 
6466 	return 0;
6467 }
6468 EXPORT_SYMBOL_GPL(skb_mpls_pop);
6469 
6470 /**
6471  * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6472  *
6473  * @skb: buffer
6474  * @mpls_lse: new MPLS label stack entry to update to
6475  *
6476  * Expects skb->data at mac header.
6477  *
6478  * Returns 0 on success, -errno otherwise.
6479  */
6480 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6481 {
6482 	int err;
6483 
6484 	if (unlikely(!eth_p_mpls(skb->protocol)))
6485 		return -EINVAL;
6486 
6487 	err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6488 	if (unlikely(err))
6489 		return err;
6490 
6491 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
6492 		__be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6493 
6494 		skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6495 	}
6496 
6497 	mpls_hdr(skb)->label_stack_entry = mpls_lse;
6498 
6499 	return 0;
6500 }
6501 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6502 
6503 /**
6504  * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6505  *
6506  * @skb: buffer
6507  *
6508  * Expects skb->data at mac header.
6509  *
6510  * Returns 0 on success, -errno otherwise.
6511  */
6512 int skb_mpls_dec_ttl(struct sk_buff *skb)
6513 {
6514 	u32 lse;
6515 	u8 ttl;
6516 
6517 	if (unlikely(!eth_p_mpls(skb->protocol)))
6518 		return -EINVAL;
6519 
6520 	if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6521 		return -ENOMEM;
6522 
6523 	lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6524 	ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6525 	if (!--ttl)
6526 		return -EINVAL;
6527 
6528 	lse &= ~MPLS_LS_TTL_MASK;
6529 	lse |= ttl << MPLS_LS_TTL_SHIFT;
6530 
6531 	return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6532 }
6533 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6534 
6535 /**
6536  * alloc_skb_with_frags - allocate skb with page frags
6537  *
6538  * @header_len: size of linear part
6539  * @data_len: needed length in frags
6540  * @order: max page order desired.
6541  * @errcode: pointer to error code if any
6542  * @gfp_mask: allocation mask
6543  *
6544  * This can be used to allocate a paged skb, given a maximal order for frags.
6545  */
6546 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6547 				     unsigned long data_len,
6548 				     int order,
6549 				     int *errcode,
6550 				     gfp_t gfp_mask)
6551 {
6552 	unsigned long chunk;
6553 	struct sk_buff *skb;
6554 	struct page *page;
6555 	int nr_frags = 0;
6556 
6557 	*errcode = -EMSGSIZE;
6558 	if (unlikely(data_len > MAX_SKB_FRAGS * (PAGE_SIZE << order)))
6559 		return NULL;
6560 
6561 	*errcode = -ENOBUFS;
6562 	skb = alloc_skb(header_len, gfp_mask);
6563 	if (!skb)
6564 		return NULL;
6565 
6566 	while (data_len) {
6567 		if (nr_frags == MAX_SKB_FRAGS - 1)
6568 			goto failure;
6569 		while (order && PAGE_ALIGN(data_len) < (PAGE_SIZE << order))
6570 			order--;
6571 
6572 		if (order) {
6573 			page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6574 					   __GFP_COMP |
6575 					   __GFP_NOWARN,
6576 					   order);
6577 			if (!page) {
6578 				order--;
6579 				continue;
6580 			}
6581 		} else {
6582 			page = alloc_page(gfp_mask);
6583 			if (!page)
6584 				goto failure;
6585 		}
6586 		chunk = min_t(unsigned long, data_len,
6587 			      PAGE_SIZE << order);
6588 		skb_fill_page_desc(skb, nr_frags, page, 0, chunk);
6589 		nr_frags++;
6590 		skb->truesize += (PAGE_SIZE << order);
6591 		data_len -= chunk;
6592 	}
6593 	return skb;
6594 
6595 failure:
6596 	kfree_skb(skb);
6597 	return NULL;
6598 }
6599 EXPORT_SYMBOL(alloc_skb_with_frags);
6600 
6601 /* carve out the first off bytes from skb when off < headlen */
6602 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6603 				    const int headlen, gfp_t gfp_mask)
6604 {
6605 	int i;
6606 	unsigned int size = skb_end_offset(skb);
6607 	int new_hlen = headlen - off;
6608 	u8 *data;
6609 
6610 	if (skb_pfmemalloc(skb))
6611 		gfp_mask |= __GFP_MEMALLOC;
6612 
6613 	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6614 	if (!data)
6615 		return -ENOMEM;
6616 	size = SKB_WITH_OVERHEAD(size);
6617 
6618 	/* Copy real data, and all frags */
6619 	skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6620 	skb->len -= off;
6621 
6622 	memcpy((struct skb_shared_info *)(data + size),
6623 	       skb_shinfo(skb),
6624 	       offsetof(struct skb_shared_info,
6625 			frags[skb_shinfo(skb)->nr_frags]));
6626 	if (skb_cloned(skb)) {
6627 		/* drop the old head gracefully */
6628 		if (skb_orphan_frags(skb, gfp_mask)) {
6629 			skb_kfree_head(data, size);
6630 			return -ENOMEM;
6631 		}
6632 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6633 			skb_frag_ref(skb, i);
6634 		if (skb_has_frag_list(skb))
6635 			skb_clone_fraglist(skb);
6636 		skb_release_data(skb, SKB_CONSUMED);
6637 	} else {
6638 		/* we can reuse existing recount- all we did was
6639 		 * relocate values
6640 		 */
6641 		skb_free_head(skb);
6642 	}
6643 
6644 	skb->head = data;
6645 	skb->data = data;
6646 	skb->head_frag = 0;
6647 	skb_set_end_offset(skb, size);
6648 	skb_set_tail_pointer(skb, skb_headlen(skb));
6649 	skb_headers_offset_update(skb, 0);
6650 	skb->cloned = 0;
6651 	skb->hdr_len = 0;
6652 	skb->nohdr = 0;
6653 	atomic_set(&skb_shinfo(skb)->dataref, 1);
6654 
6655 	return 0;
6656 }
6657 
6658 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6659 
6660 /* carve out the first eat bytes from skb's frag_list. May recurse into
6661  * pskb_carve()
6662  */
6663 static int pskb_carve_frag_list(struct sk_buff *skb,
6664 				struct skb_shared_info *shinfo, int eat,
6665 				gfp_t gfp_mask)
6666 {
6667 	struct sk_buff *list = shinfo->frag_list;
6668 	struct sk_buff *clone = NULL;
6669 	struct sk_buff *insp = NULL;
6670 
6671 	do {
6672 		if (!list) {
6673 			pr_err("Not enough bytes to eat. Want %d\n", eat);
6674 			return -EFAULT;
6675 		}
6676 		if (list->len <= eat) {
6677 			/* Eaten as whole. */
6678 			eat -= list->len;
6679 			list = list->next;
6680 			insp = list;
6681 		} else {
6682 			/* Eaten partially. */
6683 			if (skb_shared(list)) {
6684 				clone = skb_clone(list, gfp_mask);
6685 				if (!clone)
6686 					return -ENOMEM;
6687 				insp = list->next;
6688 				list = clone;
6689 			} else {
6690 				/* This may be pulled without problems. */
6691 				insp = list;
6692 			}
6693 			if (pskb_carve(list, eat, gfp_mask) < 0) {
6694 				kfree_skb(clone);
6695 				return -ENOMEM;
6696 			}
6697 			break;
6698 		}
6699 	} while (eat);
6700 
6701 	/* Free pulled out fragments. */
6702 	while ((list = shinfo->frag_list) != insp) {
6703 		shinfo->frag_list = list->next;
6704 		consume_skb(list);
6705 	}
6706 	/* And insert new clone at head. */
6707 	if (clone) {
6708 		clone->next = list;
6709 		shinfo->frag_list = clone;
6710 	}
6711 	return 0;
6712 }
6713 
6714 /* carve off first len bytes from skb. Split line (off) is in the
6715  * non-linear part of skb
6716  */
6717 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6718 				       int pos, gfp_t gfp_mask)
6719 {
6720 	int i, k = 0;
6721 	unsigned int size = skb_end_offset(skb);
6722 	u8 *data;
6723 	const int nfrags = skb_shinfo(skb)->nr_frags;
6724 	struct skb_shared_info *shinfo;
6725 
6726 	if (skb_pfmemalloc(skb))
6727 		gfp_mask |= __GFP_MEMALLOC;
6728 
6729 	data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6730 	if (!data)
6731 		return -ENOMEM;
6732 	size = SKB_WITH_OVERHEAD(size);
6733 
6734 	memcpy((struct skb_shared_info *)(data + size),
6735 	       skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6736 	if (skb_orphan_frags(skb, gfp_mask)) {
6737 		skb_kfree_head(data, size);
6738 		return -ENOMEM;
6739 	}
6740 	shinfo = (struct skb_shared_info *)(data + size);
6741 	for (i = 0; i < nfrags; i++) {
6742 		int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6743 
6744 		if (pos + fsize > off) {
6745 			shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6746 
6747 			if (pos < off) {
6748 				/* Split frag.
6749 				 * We have two variants in this case:
6750 				 * 1. Move all the frag to the second
6751 				 *    part, if it is possible. F.e.
6752 				 *    this approach is mandatory for TUX,
6753 				 *    where splitting is expensive.
6754 				 * 2. Split is accurately. We make this.
6755 				 */
6756 				skb_frag_off_add(&shinfo->frags[0], off - pos);
6757 				skb_frag_size_sub(&shinfo->frags[0], off - pos);
6758 			}
6759 			skb_frag_ref(skb, i);
6760 			k++;
6761 		}
6762 		pos += fsize;
6763 	}
6764 	shinfo->nr_frags = k;
6765 	if (skb_has_frag_list(skb))
6766 		skb_clone_fraglist(skb);
6767 
6768 	/* split line is in frag list */
6769 	if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6770 		/* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6771 		if (skb_has_frag_list(skb))
6772 			kfree_skb_list(skb_shinfo(skb)->frag_list);
6773 		skb_kfree_head(data, size);
6774 		return -ENOMEM;
6775 	}
6776 	skb_release_data(skb, SKB_CONSUMED);
6777 
6778 	skb->head = data;
6779 	skb->head_frag = 0;
6780 	skb->data = data;
6781 	skb_set_end_offset(skb, size);
6782 	skb_reset_tail_pointer(skb);
6783 	skb_headers_offset_update(skb, 0);
6784 	skb->cloned   = 0;
6785 	skb->hdr_len  = 0;
6786 	skb->nohdr    = 0;
6787 	skb->len -= off;
6788 	skb->data_len = skb->len;
6789 	atomic_set(&skb_shinfo(skb)->dataref, 1);
6790 	return 0;
6791 }
6792 
6793 /* remove len bytes from the beginning of the skb */
6794 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6795 {
6796 	int headlen = skb_headlen(skb);
6797 
6798 	if (len < headlen)
6799 		return pskb_carve_inside_header(skb, len, headlen, gfp);
6800 	else
6801 		return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6802 }
6803 
6804 /* Extract to_copy bytes starting at off from skb, and return this in
6805  * a new skb
6806  */
6807 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6808 			     int to_copy, gfp_t gfp)
6809 {
6810 	struct sk_buff  *clone = skb_clone(skb, gfp);
6811 
6812 	if (!clone)
6813 		return NULL;
6814 
6815 	if (pskb_carve(clone, off, gfp) < 0 ||
6816 	    pskb_trim(clone, to_copy)) {
6817 		kfree_skb(clone);
6818 		return NULL;
6819 	}
6820 	return clone;
6821 }
6822 EXPORT_SYMBOL(pskb_extract);
6823 
6824 /**
6825  * skb_condense - try to get rid of fragments/frag_list if possible
6826  * @skb: buffer
6827  *
6828  * Can be used to save memory before skb is added to a busy queue.
6829  * If packet has bytes in frags and enough tail room in skb->head,
6830  * pull all of them, so that we can free the frags right now and adjust
6831  * truesize.
6832  * Notes:
6833  *	We do not reallocate skb->head thus can not fail.
6834  *	Caller must re-evaluate skb->truesize if needed.
6835  */
6836 void skb_condense(struct sk_buff *skb)
6837 {
6838 	if (skb->data_len) {
6839 		if (skb->data_len > skb->end - skb->tail ||
6840 		    skb_cloned(skb))
6841 			return;
6842 
6843 		/* Nice, we can free page frag(s) right now */
6844 		__pskb_pull_tail(skb, skb->data_len);
6845 	}
6846 	/* At this point, skb->truesize might be over estimated,
6847 	 * because skb had a fragment, and fragments do not tell
6848 	 * their truesize.
6849 	 * When we pulled its content into skb->head, fragment
6850 	 * was freed, but __pskb_pull_tail() could not possibly
6851 	 * adjust skb->truesize, not knowing the frag truesize.
6852 	 */
6853 	skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6854 }
6855 EXPORT_SYMBOL(skb_condense);
6856 
6857 #ifdef CONFIG_SKB_EXTENSIONS
6858 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6859 {
6860 	return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6861 }
6862 
6863 /**
6864  * __skb_ext_alloc - allocate a new skb extensions storage
6865  *
6866  * @flags: See kmalloc().
6867  *
6868  * Returns the newly allocated pointer. The pointer can later attached to a
6869  * skb via __skb_ext_set().
6870  * Note: caller must handle the skb_ext as an opaque data.
6871  */
6872 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6873 {
6874 	struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6875 
6876 	if (new) {
6877 		memset(new->offset, 0, sizeof(new->offset));
6878 		refcount_set(&new->refcnt, 1);
6879 	}
6880 
6881 	return new;
6882 }
6883 
6884 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6885 					 unsigned int old_active)
6886 {
6887 	struct skb_ext *new;
6888 
6889 	if (refcount_read(&old->refcnt) == 1)
6890 		return old;
6891 
6892 	new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6893 	if (!new)
6894 		return NULL;
6895 
6896 	memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6897 	refcount_set(&new->refcnt, 1);
6898 
6899 #ifdef CONFIG_XFRM
6900 	if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6901 		struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6902 		unsigned int i;
6903 
6904 		for (i = 0; i < sp->len; i++)
6905 			xfrm_state_hold(sp->xvec[i]);
6906 	}
6907 #endif
6908 #ifdef CONFIG_MCTP_FLOWS
6909 	if (old_active & (1 << SKB_EXT_MCTP)) {
6910 		struct mctp_flow *flow = skb_ext_get_ptr(old, SKB_EXT_MCTP);
6911 
6912 		if (flow->key)
6913 			refcount_inc(&flow->key->refs);
6914 	}
6915 #endif
6916 	__skb_ext_put(old);
6917 	return new;
6918 }
6919 
6920 /**
6921  * __skb_ext_set - attach the specified extension storage to this skb
6922  * @skb: buffer
6923  * @id: extension id
6924  * @ext: extension storage previously allocated via __skb_ext_alloc()
6925  *
6926  * Existing extensions, if any, are cleared.
6927  *
6928  * Returns the pointer to the extension.
6929  */
6930 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6931 		    struct skb_ext *ext)
6932 {
6933 	unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6934 
6935 	skb_ext_put(skb);
6936 	newlen = newoff + skb_ext_type_len[id];
6937 	ext->chunks = newlen;
6938 	ext->offset[id] = newoff;
6939 	skb->extensions = ext;
6940 	skb->active_extensions = 1 << id;
6941 	return skb_ext_get_ptr(ext, id);
6942 }
6943 
6944 /**
6945  * skb_ext_add - allocate space for given extension, COW if needed
6946  * @skb: buffer
6947  * @id: extension to allocate space for
6948  *
6949  * Allocates enough space for the given extension.
6950  * If the extension is already present, a pointer to that extension
6951  * is returned.
6952  *
6953  * If the skb was cloned, COW applies and the returned memory can be
6954  * modified without changing the extension space of clones buffers.
6955  *
6956  * Returns pointer to the extension or NULL on allocation failure.
6957  */
6958 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6959 {
6960 	struct skb_ext *new, *old = NULL;
6961 	unsigned int newlen, newoff;
6962 
6963 	if (skb->active_extensions) {
6964 		old = skb->extensions;
6965 
6966 		new = skb_ext_maybe_cow(old, skb->active_extensions);
6967 		if (!new)
6968 			return NULL;
6969 
6970 		if (__skb_ext_exist(new, id))
6971 			goto set_active;
6972 
6973 		newoff = new->chunks;
6974 	} else {
6975 		newoff = SKB_EXT_CHUNKSIZEOF(*new);
6976 
6977 		new = __skb_ext_alloc(GFP_ATOMIC);
6978 		if (!new)
6979 			return NULL;
6980 	}
6981 
6982 	newlen = newoff + skb_ext_type_len[id];
6983 	new->chunks = newlen;
6984 	new->offset[id] = newoff;
6985 set_active:
6986 	skb->slow_gro = 1;
6987 	skb->extensions = new;
6988 	skb->active_extensions |= 1 << id;
6989 	return skb_ext_get_ptr(new, id);
6990 }
6991 EXPORT_SYMBOL(skb_ext_add);
6992 
6993 #ifdef CONFIG_XFRM
6994 static void skb_ext_put_sp(struct sec_path *sp)
6995 {
6996 	unsigned int i;
6997 
6998 	for (i = 0; i < sp->len; i++)
6999 		xfrm_state_put(sp->xvec[i]);
7000 }
7001 #endif
7002 
7003 #ifdef CONFIG_MCTP_FLOWS
7004 static void skb_ext_put_mctp(struct mctp_flow *flow)
7005 {
7006 	if (flow->key)
7007 		mctp_key_unref(flow->key);
7008 }
7009 #endif
7010 
7011 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
7012 {
7013 	struct skb_ext *ext = skb->extensions;
7014 
7015 	skb->active_extensions &= ~(1 << id);
7016 	if (skb->active_extensions == 0) {
7017 		skb->extensions = NULL;
7018 		__skb_ext_put(ext);
7019 #ifdef CONFIG_XFRM
7020 	} else if (id == SKB_EXT_SEC_PATH &&
7021 		   refcount_read(&ext->refcnt) == 1) {
7022 		struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
7023 
7024 		skb_ext_put_sp(sp);
7025 		sp->len = 0;
7026 #endif
7027 	}
7028 }
7029 EXPORT_SYMBOL(__skb_ext_del);
7030 
7031 void __skb_ext_put(struct skb_ext *ext)
7032 {
7033 	/* If this is last clone, nothing can increment
7034 	 * it after check passes.  Avoids one atomic op.
7035 	 */
7036 	if (refcount_read(&ext->refcnt) == 1)
7037 		goto free_now;
7038 
7039 	if (!refcount_dec_and_test(&ext->refcnt))
7040 		return;
7041 free_now:
7042 #ifdef CONFIG_XFRM
7043 	if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
7044 		skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
7045 #endif
7046 #ifdef CONFIG_MCTP_FLOWS
7047 	if (__skb_ext_exist(ext, SKB_EXT_MCTP))
7048 		skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
7049 #endif
7050 
7051 	kmem_cache_free(skbuff_ext_cache, ext);
7052 }
7053 EXPORT_SYMBOL(__skb_ext_put);
7054 #endif /* CONFIG_SKB_EXTENSIONS */
7055 
7056 static void kfree_skb_napi_cache(struct sk_buff *skb)
7057 {
7058 	/* if SKB is a clone, don't handle this case */
7059 	if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
7060 		__kfree_skb(skb);
7061 		return;
7062 	}
7063 
7064 	local_bh_disable();
7065 	__napi_kfree_skb(skb, SKB_CONSUMED);
7066 	local_bh_enable();
7067 }
7068 
7069 /**
7070  * skb_attempt_defer_free - queue skb for remote freeing
7071  * @skb: buffer
7072  *
7073  * Put @skb in a per-cpu list, using the cpu which
7074  * allocated the skb/pages to reduce false sharing
7075  * and memory zone spinlock contention.
7076  */
7077 void skb_attempt_defer_free(struct sk_buff *skb)
7078 {
7079 	int cpu = skb->alloc_cpu;
7080 	struct softnet_data *sd;
7081 	unsigned int defer_max;
7082 	bool kick;
7083 
7084 	if (cpu == raw_smp_processor_id() ||
7085 	    WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
7086 	    !cpu_online(cpu)) {
7087 nodefer:	kfree_skb_napi_cache(skb);
7088 		return;
7089 	}
7090 
7091 	DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
7092 	DEBUG_NET_WARN_ON_ONCE(skb->destructor);
7093 
7094 	sd = &per_cpu(softnet_data, cpu);
7095 	defer_max = READ_ONCE(net_hotdata.sysctl_skb_defer_max);
7096 	if (READ_ONCE(sd->defer_count) >= defer_max)
7097 		goto nodefer;
7098 
7099 	spin_lock_bh(&sd->defer_lock);
7100 	/* Send an IPI every time queue reaches half capacity. */
7101 	kick = sd->defer_count == (defer_max >> 1);
7102 	/* Paired with the READ_ONCE() few lines above */
7103 	WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
7104 
7105 	skb->next = sd->defer_list;
7106 	/* Paired with READ_ONCE() in skb_defer_free_flush() */
7107 	WRITE_ONCE(sd->defer_list, skb);
7108 	spin_unlock_bh(&sd->defer_lock);
7109 
7110 	/* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
7111 	 * if we are unlucky enough (this seems very unlikely).
7112 	 */
7113 	if (unlikely(kick))
7114 		kick_defer_list_purge(sd, cpu);
7115 }
7116 
7117 static void skb_splice_csum_page(struct sk_buff *skb, struct page *page,
7118 				 size_t offset, size_t len)
7119 {
7120 	const char *kaddr;
7121 	__wsum csum;
7122 
7123 	kaddr = kmap_local_page(page);
7124 	csum = csum_partial(kaddr + offset, len, 0);
7125 	kunmap_local(kaddr);
7126 	skb->csum = csum_block_add(skb->csum, csum, skb->len);
7127 }
7128 
7129 /**
7130  * skb_splice_from_iter - Splice (or copy) pages to skbuff
7131  * @skb: The buffer to add pages to
7132  * @iter: Iterator representing the pages to be added
7133  * @maxsize: Maximum amount of pages to be added
7134  * @gfp: Allocation flags
7135  *
7136  * This is a common helper function for supporting MSG_SPLICE_PAGES.  It
7137  * extracts pages from an iterator and adds them to the socket buffer if
7138  * possible, copying them to fragments if not possible (such as if they're slab
7139  * pages).
7140  *
7141  * Returns the amount of data spliced/copied or -EMSGSIZE if there's
7142  * insufficient space in the buffer to transfer anything.
7143  */
7144 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
7145 			     ssize_t maxsize, gfp_t gfp)
7146 {
7147 	size_t frag_limit = READ_ONCE(net_hotdata.sysctl_max_skb_frags);
7148 	struct page *pages[8], **ppages = pages;
7149 	ssize_t spliced = 0, ret = 0;
7150 	unsigned int i;
7151 
7152 	while (iter->count > 0) {
7153 		ssize_t space, nr, len;
7154 		size_t off;
7155 
7156 		ret = -EMSGSIZE;
7157 		space = frag_limit - skb_shinfo(skb)->nr_frags;
7158 		if (space < 0)
7159 			break;
7160 
7161 		/* We might be able to coalesce without increasing nr_frags */
7162 		nr = clamp_t(size_t, space, 1, ARRAY_SIZE(pages));
7163 
7164 		len = iov_iter_extract_pages(iter, &ppages, maxsize, nr, 0, &off);
7165 		if (len <= 0) {
7166 			ret = len ?: -EIO;
7167 			break;
7168 		}
7169 
7170 		i = 0;
7171 		do {
7172 			struct page *page = pages[i++];
7173 			size_t part = min_t(size_t, PAGE_SIZE - off, len);
7174 
7175 			ret = -EIO;
7176 			if (WARN_ON_ONCE(!sendpage_ok(page)))
7177 				goto out;
7178 
7179 			ret = skb_append_pagefrags(skb, page, off, part,
7180 						   frag_limit);
7181 			if (ret < 0) {
7182 				iov_iter_revert(iter, len);
7183 				goto out;
7184 			}
7185 
7186 			if (skb->ip_summed == CHECKSUM_NONE)
7187 				skb_splice_csum_page(skb, page, off, part);
7188 
7189 			off = 0;
7190 			spliced += part;
7191 			maxsize -= part;
7192 			len -= part;
7193 		} while (len > 0);
7194 
7195 		if (maxsize <= 0)
7196 			break;
7197 	}
7198 
7199 out:
7200 	skb_len_add(skb, spliced);
7201 	return spliced ?: ret;
7202 }
7203 EXPORT_SYMBOL(skb_splice_from_iter);
7204 
7205 static __always_inline
7206 size_t memcpy_from_iter_csum(void *iter_from, size_t progress,
7207 			     size_t len, void *to, void *priv2)
7208 {
7209 	__wsum *csum = priv2;
7210 	__wsum next = csum_partial_copy_nocheck(iter_from, to + progress, len);
7211 
7212 	*csum = csum_block_add(*csum, next, progress);
7213 	return 0;
7214 }
7215 
7216 static __always_inline
7217 size_t copy_from_user_iter_csum(void __user *iter_from, size_t progress,
7218 				size_t len, void *to, void *priv2)
7219 {
7220 	__wsum next, *csum = priv2;
7221 
7222 	next = csum_and_copy_from_user(iter_from, to + progress, len);
7223 	*csum = csum_block_add(*csum, next, progress);
7224 	return next ? 0 : len;
7225 }
7226 
7227 bool csum_and_copy_from_iter_full(void *addr, size_t bytes,
7228 				  __wsum *csum, struct iov_iter *i)
7229 {
7230 	size_t copied;
7231 
7232 	if (WARN_ON_ONCE(!i->data_source))
7233 		return false;
7234 	copied = iterate_and_advance2(i, bytes, addr, csum,
7235 				      copy_from_user_iter_csum,
7236 				      memcpy_from_iter_csum);
7237 	if (likely(copied == bytes))
7238 		return true;
7239 	iov_iter_revert(i, copied);
7240 	return false;
7241 }
7242 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
7243