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