1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /*
3 * Routines having to do with the 'struct sk_buff' memory handlers.
4 *
5 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
6 * Florian La Roche <rzsfl@rz.uni-sb.de>
7 *
8 * Fixes:
9 * Alan Cox : Fixed the worst of the load
10 * balancer bugs.
11 * Dave Platt : Interrupt stacking fix.
12 * Richard Kooijman : Timestamp fixes.
13 * Alan Cox : Changed buffer format.
14 * Alan Cox : destructor hook for AF_UNIX etc.
15 * Linus Torvalds : Better skb_clone.
16 * Alan Cox : Added skb_copy.
17 * Alan Cox : Added all the changed routines Linus
18 * only put in the headers
19 * Ray VanTassle : Fixed --skb->lock in free
20 * Alan Cox : skb_copy copy arp field
21 * Andi Kleen : slabified it.
22 * Robert Olsson : Removed skb_head_pool
23 *
24 * NOTE:
25 * The __skb_ routines should be called with interrupts
26 * disabled, or you better be *real* sure that the operation is atomic
27 * with respect to whatever list is being frobbed (e.g. via lock_sock()
28 * or via disabling bottom half handlers, etc).
29 */
30
31 /*
32 * The functions in this file will not compile correctly with gcc 2.4.x
33 */
34
35 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36
37 #include <linux/module.h>
38 #include <linux/types.h>
39 #include <linux/kernel.h>
40 #include <linux/mm.h>
41 #include <linux/interrupt.h>
42 #include <linux/in.h>
43 #include <linux/inet.h>
44 #include <linux/slab.h>
45 #include <linux/tcp.h>
46 #include <linux/udp.h>
47 #include <linux/sctp.h>
48 #include <linux/netdevice.h>
49 #ifdef CONFIG_NET_CLS_ACT
50 #include <net/pkt_sched.h>
51 #endif
52 #include <linux/string.h>
53 #include <linux/skbuff.h>
54 #include <linux/skbuff_ref.h>
55 #include <linux/splice.h>
56 #include <linux/cache.h>
57 #include <linux/rtnetlink.h>
58 #include <linux/init.h>
59 #include <linux/scatterlist.h>
60 #include <linux/errqueue.h>
61 #include <linux/prefetch.h>
62 #include <linux/bitfield.h>
63 #include <linux/if_vlan.h>
64 #include <linux/mpls.h>
65 #include <linux/kcov.h>
66 #include <linux/iov_iter.h>
67
68 #include <net/protocol.h>
69 #include <net/dst.h>
70 #include <net/sock.h>
71 #include <net/checksum.h>
72 #include <net/gso.h>
73 #include <net/hotdata.h>
74 #include <net/ip6_checksum.h>
75 #include <net/xfrm.h>
76 #include <net/mpls.h>
77 #include <net/mptcp.h>
78 #include <net/mctp.h>
79 #include <net/page_pool/helpers.h>
80 #include <net/dropreason.h>
81
82 #include <linux/uaccess.h>
83 #include <trace/events/skb.h>
84 #include <linux/highmem.h>
85 #include <linux/capability.h>
86 #include <linux/user_namespace.h>
87 #include <linux/indirect_call_wrapper.h>
88 #include <linux/textsearch.h>
89
90 #include "dev.h"
91 #include "netmem_priv.h"
92 #include "sock_destructor.h"
93
94 #ifdef CONFIG_SKB_EXTENSIONS
95 static struct kmem_cache *skbuff_ext_cache __ro_after_init;
96 #endif
97
98 #define SKB_SMALL_HEAD_SIZE SKB_HEAD_ALIGN(MAX_TCP_HEADER)
99
100 /* We want SKB_SMALL_HEAD_CACHE_SIZE to not be a power of two.
101 * This should ensure that SKB_SMALL_HEAD_HEADROOM is a unique
102 * size, and we can differentiate heads from skb_small_head_cache
103 * vs system slabs by looking at their size (skb_end_offset()).
104 */
105 #define SKB_SMALL_HEAD_CACHE_SIZE \
106 (is_power_of_2(SKB_SMALL_HEAD_SIZE) ? \
107 (SKB_SMALL_HEAD_SIZE + L1_CACHE_BYTES) : \
108 SKB_SMALL_HEAD_SIZE)
109
110 #define SKB_SMALL_HEAD_HEADROOM \
111 SKB_WITH_OVERHEAD(SKB_SMALL_HEAD_CACHE_SIZE)
112
113 /* kcm_write_msgs() relies on casting paged frags to bio_vec to use
114 * iov_iter_bvec(). These static asserts ensure the cast is valid is long as the
115 * netmem is a page.
116 */
117 static_assert(offsetof(struct bio_vec, bv_page) ==
118 offsetof(skb_frag_t, netmem));
119 static_assert(sizeof_field(struct bio_vec, bv_page) ==
120 sizeof_field(skb_frag_t, netmem));
121
122 static_assert(offsetof(struct bio_vec, bv_len) == offsetof(skb_frag_t, len));
123 static_assert(sizeof_field(struct bio_vec, bv_len) ==
124 sizeof_field(skb_frag_t, len));
125
126 static_assert(offsetof(struct bio_vec, bv_offset) ==
127 offsetof(skb_frag_t, offset));
128 static_assert(sizeof_field(struct bio_vec, bv_offset) ==
129 sizeof_field(skb_frag_t, offset));
130
131 #undef FN
132 #define FN(reason) [SKB_DROP_REASON_##reason] = #reason,
133 static const char * const drop_reasons[] = {
134 [SKB_CONSUMED] = "CONSUMED",
135 DEFINE_DROP_REASON(FN, FN)
136 };
137
138 static const struct drop_reason_list drop_reasons_core = {
139 .reasons = drop_reasons,
140 .n_reasons = ARRAY_SIZE(drop_reasons),
141 };
142
143 const struct drop_reason_list __rcu *
144 drop_reasons_by_subsys[SKB_DROP_REASON_SUBSYS_NUM] = {
145 [SKB_DROP_REASON_SUBSYS_CORE] = RCU_INITIALIZER(&drop_reasons_core),
146 };
147 EXPORT_SYMBOL(drop_reasons_by_subsys);
148
149 /**
150 * drop_reasons_register_subsys - register another drop reason subsystem
151 * @subsys: the subsystem to register, must not be the core
152 * @list: the list of drop reasons within the subsystem, must point to
153 * a statically initialized list
154 */
drop_reasons_register_subsys(enum skb_drop_reason_subsys subsys,const struct drop_reason_list * list)155 void drop_reasons_register_subsys(enum skb_drop_reason_subsys subsys,
156 const struct drop_reason_list *list)
157 {
158 if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
159 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
160 "invalid subsystem %d\n", subsys))
161 return;
162
163 /* must point to statically allocated memory, so INIT is OK */
164 RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], list);
165 }
166 EXPORT_SYMBOL_GPL(drop_reasons_register_subsys);
167
168 /**
169 * drop_reasons_unregister_subsys - unregister a drop reason subsystem
170 * @subsys: the subsystem to remove, must not be the core
171 *
172 * Note: This will synchronize_rcu() to ensure no users when it returns.
173 */
drop_reasons_unregister_subsys(enum skb_drop_reason_subsys subsys)174 void drop_reasons_unregister_subsys(enum skb_drop_reason_subsys subsys)
175 {
176 if (WARN(subsys <= SKB_DROP_REASON_SUBSYS_CORE ||
177 subsys >= ARRAY_SIZE(drop_reasons_by_subsys),
178 "invalid subsystem %d\n", subsys))
179 return;
180
181 RCU_INIT_POINTER(drop_reasons_by_subsys[subsys], NULL);
182
183 synchronize_rcu();
184 }
185 EXPORT_SYMBOL_GPL(drop_reasons_unregister_subsys);
186
187 /**
188 * skb_panic - private function for out-of-line support
189 * @skb: buffer
190 * @sz: size
191 * @addr: address
192 * @msg: skb_over_panic or skb_under_panic
193 *
194 * Out-of-line support for skb_put() and skb_push().
195 * Called via the wrapper skb_over_panic() or skb_under_panic().
196 * Keep out of line to prevent kernel bloat.
197 * __builtin_return_address is not used because it is not always reliable.
198 */
skb_panic(struct sk_buff * skb,unsigned int sz,void * addr,const char msg[])199 static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
200 const char msg[])
201 {
202 pr_emerg("%s: text:%px len:%d put:%d head:%px data:%px tail:%#lx end:%#lx dev:%s\n",
203 msg, addr, skb->len, sz, skb->head, skb->data,
204 (unsigned long)skb->tail, (unsigned long)skb->end,
205 skb->dev ? skb->dev->name : "<NULL>");
206 BUG();
207 }
208
skb_over_panic(struct sk_buff * skb,unsigned int sz,void * addr)209 static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
210 {
211 skb_panic(skb, sz, addr, __func__);
212 }
213
skb_under_panic(struct sk_buff * skb,unsigned int sz,void * addr)214 static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
215 {
216 skb_panic(skb, sz, addr, __func__);
217 }
218
219 #define NAPI_SKB_CACHE_SIZE 64
220 #define NAPI_SKB_CACHE_BULK 16
221 #define NAPI_SKB_CACHE_HALF (NAPI_SKB_CACHE_SIZE / 2)
222
223 #if PAGE_SIZE == SZ_4K
224
225 #define NAPI_HAS_SMALL_PAGE_FRAG 1
226 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) ((nc).pfmemalloc)
227
228 /* specialized page frag allocator using a single order 0 page
229 * and slicing it into 1K sized fragment. Constrained to systems
230 * with a very limited amount of 1K fragments fitting a single
231 * page - to avoid excessive truesize underestimation
232 */
233
234 struct page_frag_1k {
235 void *va;
236 u16 offset;
237 bool pfmemalloc;
238 };
239
page_frag_alloc_1k(struct page_frag_1k * nc,gfp_t gfp)240 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp)
241 {
242 struct page *page;
243 int offset;
244
245 offset = nc->offset - SZ_1K;
246 if (likely(offset >= 0))
247 goto use_frag;
248
249 page = alloc_pages_node(NUMA_NO_NODE, gfp, 0);
250 if (!page)
251 return NULL;
252
253 nc->va = page_address(page);
254 nc->pfmemalloc = page_is_pfmemalloc(page);
255 offset = PAGE_SIZE - SZ_1K;
256 page_ref_add(page, offset / SZ_1K);
257
258 use_frag:
259 nc->offset = offset;
260 return nc->va + offset;
261 }
262 #else
263
264 /* the small page is actually unused in this build; add dummy helpers
265 * to please the compiler and avoid later preprocessor's conditionals
266 */
267 #define NAPI_HAS_SMALL_PAGE_FRAG 0
268 #define NAPI_SMALL_PAGE_PFMEMALLOC(nc) false
269
270 struct page_frag_1k {
271 };
272
page_frag_alloc_1k(struct page_frag_1k * nc,gfp_t gfp_mask)273 static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
274 {
275 return NULL;
276 }
277
278 #endif
279
280 struct napi_alloc_cache {
281 local_lock_t bh_lock;
282 struct page_frag_cache page;
283 struct page_frag_1k page_small;
284 unsigned int skb_count;
285 void *skb_cache[NAPI_SKB_CACHE_SIZE];
286 };
287
288 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
289 static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = {
290 .bh_lock = INIT_LOCAL_LOCK(bh_lock),
291 };
292
293 /* Double check that napi_get_frags() allocates skbs with
294 * skb->head being backed by slab, not a page fragment.
295 * This is to make sure bug fixed in 3226b158e67c
296 * ("net: avoid 32 x truesize under-estimation for tiny skbs")
297 * does not accidentally come back.
298 */
napi_get_frags_check(struct napi_struct * napi)299 void napi_get_frags_check(struct napi_struct *napi)
300 {
301 struct sk_buff *skb;
302
303 local_bh_disable();
304 skb = napi_get_frags(napi);
305 WARN_ON_ONCE(!NAPI_HAS_SMALL_PAGE_FRAG && skb && skb->head_frag);
306 napi_free_frags(napi);
307 local_bh_enable();
308 }
309
__napi_alloc_frag_align(unsigned int fragsz,unsigned int align_mask)310 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
311 {
312 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
313 void *data;
314
315 fragsz = SKB_DATA_ALIGN(fragsz);
316
317 local_lock_nested_bh(&napi_alloc_cache.bh_lock);
318 data = __page_frag_alloc_align(&nc->page, fragsz,
319 GFP_ATOMIC | __GFP_NOWARN, align_mask);
320 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
321 return data;
322
323 }
324 EXPORT_SYMBOL(__napi_alloc_frag_align);
325
__netdev_alloc_frag_align(unsigned int fragsz,unsigned int align_mask)326 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
327 {
328 void *data;
329
330 if (in_hardirq() || irqs_disabled()) {
331 struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
332
333 fragsz = SKB_DATA_ALIGN(fragsz);
334 data = __page_frag_alloc_align(nc, fragsz,
335 GFP_ATOMIC | __GFP_NOWARN,
336 align_mask);
337 } else {
338 local_bh_disable();
339 data = __napi_alloc_frag_align(fragsz, align_mask);
340 local_bh_enable();
341 }
342 return data;
343 }
344 EXPORT_SYMBOL(__netdev_alloc_frag_align);
345
napi_skb_cache_get(void)346 static struct sk_buff *napi_skb_cache_get(void)
347 {
348 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
349 struct sk_buff *skb;
350
351 local_lock_nested_bh(&napi_alloc_cache.bh_lock);
352 if (unlikely(!nc->skb_count)) {
353 nc->skb_count = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
354 GFP_ATOMIC | __GFP_NOWARN,
355 NAPI_SKB_CACHE_BULK,
356 nc->skb_cache);
357 if (unlikely(!nc->skb_count)) {
358 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
359 return NULL;
360 }
361 }
362
363 skb = nc->skb_cache[--nc->skb_count];
364 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
365 kasan_mempool_unpoison_object(skb, kmem_cache_size(net_hotdata.skbuff_cache));
366
367 return skb;
368 }
369
__finalize_skb_around(struct sk_buff * skb,void * data,unsigned int size)370 static inline void __finalize_skb_around(struct sk_buff *skb, void *data,
371 unsigned int size)
372 {
373 struct skb_shared_info *shinfo;
374
375 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
376
377 /* Assumes caller memset cleared SKB */
378 skb->truesize = SKB_TRUESIZE(size);
379 refcount_set(&skb->users, 1);
380 skb->head = data;
381 skb->data = data;
382 skb_reset_tail_pointer(skb);
383 skb_set_end_offset(skb, size);
384 skb->mac_header = (typeof(skb->mac_header))~0U;
385 skb->transport_header = (typeof(skb->transport_header))~0U;
386 skb->alloc_cpu = raw_smp_processor_id();
387 /* make sure we initialize shinfo sequentially */
388 shinfo = skb_shinfo(skb);
389 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
390 atomic_set(&shinfo->dataref, 1);
391
392 skb_set_kcov_handle(skb, kcov_common_handle());
393 }
394
__slab_build_skb(struct sk_buff * skb,void * data,unsigned int * size)395 static inline void *__slab_build_skb(struct sk_buff *skb, void *data,
396 unsigned int *size)
397 {
398 void *resized;
399
400 /* Must find the allocation size (and grow it to match). */
401 *size = ksize(data);
402 /* krealloc() will immediately return "data" when
403 * "ksize(data)" is requested: it is the existing upper
404 * bounds. As a result, GFP_ATOMIC will be ignored. Note
405 * that this "new" pointer needs to be passed back to the
406 * caller for use so the __alloc_size hinting will be
407 * tracked correctly.
408 */
409 resized = krealloc(data, *size, GFP_ATOMIC);
410 WARN_ON_ONCE(resized != data);
411 return resized;
412 }
413
414 /* build_skb() variant which can operate on slab buffers.
415 * Note that this should be used sparingly as slab buffers
416 * cannot be combined efficiently by GRO!
417 */
slab_build_skb(void * data)418 struct sk_buff *slab_build_skb(void *data)
419 {
420 struct sk_buff *skb;
421 unsigned int size;
422
423 skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
424 GFP_ATOMIC | __GFP_NOWARN);
425 if (unlikely(!skb))
426 return NULL;
427
428 memset(skb, 0, offsetof(struct sk_buff, tail));
429 data = __slab_build_skb(skb, data, &size);
430 __finalize_skb_around(skb, data, size);
431
432 return skb;
433 }
434 EXPORT_SYMBOL(slab_build_skb);
435
436 /* Caller must provide SKB that is memset cleared */
__build_skb_around(struct sk_buff * skb,void * data,unsigned int frag_size)437 static void __build_skb_around(struct sk_buff *skb, void *data,
438 unsigned int frag_size)
439 {
440 unsigned int size = frag_size;
441
442 /* frag_size == 0 is considered deprecated now. Callers
443 * using slab buffer should use slab_build_skb() instead.
444 */
445 if (WARN_ONCE(size == 0, "Use slab_build_skb() instead"))
446 data = __slab_build_skb(skb, data, &size);
447
448 __finalize_skb_around(skb, data, size);
449 }
450
451 /**
452 * __build_skb - build a network buffer
453 * @data: data buffer provided by caller
454 * @frag_size: size of data (must not be 0)
455 *
456 * Allocate a new &sk_buff. Caller provides space holding head and
457 * skb_shared_info. @data must have been allocated from the page
458 * allocator or vmalloc(). (A @frag_size of 0 to indicate a kmalloc()
459 * allocation is deprecated, and callers should use slab_build_skb()
460 * instead.)
461 * The return is the new skb buffer.
462 * On a failure the return is %NULL, and @data is not freed.
463 * Notes :
464 * Before IO, driver allocates only data buffer where NIC put incoming frame
465 * Driver should add room at head (NET_SKB_PAD) and
466 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
467 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
468 * before giving packet to stack.
469 * RX rings only contains data buffers, not full skbs.
470 */
__build_skb(void * data,unsigned int frag_size)471 struct sk_buff *__build_skb(void *data, unsigned int frag_size)
472 {
473 struct sk_buff *skb;
474
475 skb = kmem_cache_alloc(net_hotdata.skbuff_cache,
476 GFP_ATOMIC | __GFP_NOWARN);
477 if (unlikely(!skb))
478 return NULL;
479
480 memset(skb, 0, offsetof(struct sk_buff, tail));
481 __build_skb_around(skb, data, frag_size);
482
483 return skb;
484 }
485
486 /* build_skb() is wrapper over __build_skb(), that specifically
487 * takes care of skb->head and skb->pfmemalloc
488 */
build_skb(void * data,unsigned int frag_size)489 struct sk_buff *build_skb(void *data, unsigned int frag_size)
490 {
491 struct sk_buff *skb = __build_skb(data, frag_size);
492
493 if (likely(skb && frag_size)) {
494 skb->head_frag = 1;
495 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
496 }
497 return skb;
498 }
499 EXPORT_SYMBOL(build_skb);
500
501 /**
502 * build_skb_around - build a network buffer around provided skb
503 * @skb: sk_buff provide by caller, must be memset cleared
504 * @data: data buffer provided by caller
505 * @frag_size: size of data
506 */
build_skb_around(struct sk_buff * skb,void * data,unsigned int frag_size)507 struct sk_buff *build_skb_around(struct sk_buff *skb,
508 void *data, unsigned int frag_size)
509 {
510 if (unlikely(!skb))
511 return NULL;
512
513 __build_skb_around(skb, data, frag_size);
514
515 if (frag_size) {
516 skb->head_frag = 1;
517 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
518 }
519 return skb;
520 }
521 EXPORT_SYMBOL(build_skb_around);
522
523 /**
524 * __napi_build_skb - build a network buffer
525 * @data: data buffer provided by caller
526 * @frag_size: size of data
527 *
528 * Version of __build_skb() that uses NAPI percpu caches to obtain
529 * skbuff_head instead of inplace allocation.
530 *
531 * Returns a new &sk_buff on success, %NULL on allocation failure.
532 */
__napi_build_skb(void * data,unsigned int frag_size)533 static struct sk_buff *__napi_build_skb(void *data, unsigned int frag_size)
534 {
535 struct sk_buff *skb;
536
537 skb = napi_skb_cache_get();
538 if (unlikely(!skb))
539 return NULL;
540
541 memset(skb, 0, offsetof(struct sk_buff, tail));
542 __build_skb_around(skb, data, frag_size);
543
544 return skb;
545 }
546
547 /**
548 * napi_build_skb - build a network buffer
549 * @data: data buffer provided by caller
550 * @frag_size: size of data
551 *
552 * Version of __napi_build_skb() that takes care of skb->head_frag
553 * and skb->pfmemalloc when the data is a page or page fragment.
554 *
555 * Returns a new &sk_buff on success, %NULL on allocation failure.
556 */
napi_build_skb(void * data,unsigned int frag_size)557 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size)
558 {
559 struct sk_buff *skb = __napi_build_skb(data, frag_size);
560
561 if (likely(skb) && frag_size) {
562 skb->head_frag = 1;
563 skb_propagate_pfmemalloc(virt_to_head_page(data), skb);
564 }
565
566 return skb;
567 }
568 EXPORT_SYMBOL(napi_build_skb);
569
570 /*
571 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
572 * the caller if emergency pfmemalloc reserves are being used. If it is and
573 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
574 * may be used. Otherwise, the packet data may be discarded until enough
575 * memory is free
576 */
kmalloc_reserve(unsigned int * size,gfp_t flags,int node,bool * pfmemalloc)577 static void *kmalloc_reserve(unsigned int *size, gfp_t flags, int node,
578 bool *pfmemalloc)
579 {
580 bool ret_pfmemalloc = false;
581 size_t obj_size;
582 void *obj;
583
584 obj_size = SKB_HEAD_ALIGN(*size);
585 if (obj_size <= SKB_SMALL_HEAD_CACHE_SIZE &&
586 !(flags & KMALLOC_NOT_NORMAL_BITS)) {
587 obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache,
588 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
589 node);
590 *size = SKB_SMALL_HEAD_CACHE_SIZE;
591 if (obj || !(gfp_pfmemalloc_allowed(flags)))
592 goto out;
593 /* Try again but now we are using pfmemalloc reserves */
594 ret_pfmemalloc = true;
595 obj = kmem_cache_alloc_node(net_hotdata.skb_small_head_cache, flags, node);
596 goto out;
597 }
598
599 obj_size = kmalloc_size_roundup(obj_size);
600 /* The following cast might truncate high-order bits of obj_size, this
601 * is harmless because kmalloc(obj_size >= 2^32) will fail anyway.
602 */
603 *size = (unsigned int)obj_size;
604
605 /*
606 * Try a regular allocation, when that fails and we're not entitled
607 * to the reserves, fail.
608 */
609 obj = kmalloc_node_track_caller(obj_size,
610 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
611 node);
612 if (obj || !(gfp_pfmemalloc_allowed(flags)))
613 goto out;
614
615 /* Try again but now we are using pfmemalloc reserves */
616 ret_pfmemalloc = true;
617 obj = kmalloc_node_track_caller(obj_size, flags, node);
618
619 out:
620 if (pfmemalloc)
621 *pfmemalloc = ret_pfmemalloc;
622
623 return obj;
624 }
625
626 /* Allocate a new skbuff. We do this ourselves so we can fill in a few
627 * 'private' fields and also do memory statistics to find all the
628 * [BEEP] leaks.
629 *
630 */
631
632 /**
633 * __alloc_skb - allocate a network buffer
634 * @size: size to allocate
635 * @gfp_mask: allocation mask
636 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
637 * instead of head cache and allocate a cloned (child) skb.
638 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
639 * allocations in case the data is required for writeback
640 * @node: numa node to allocate memory on
641 *
642 * Allocate a new &sk_buff. The returned buffer has no headroom and a
643 * tail room of at least size bytes. The object has a reference count
644 * of one. The return is the buffer. On a failure the return is %NULL.
645 *
646 * Buffers may only be allocated from interrupts using a @gfp_mask of
647 * %GFP_ATOMIC.
648 */
__alloc_skb(unsigned int size,gfp_t gfp_mask,int flags,int node)649 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
650 int flags, int node)
651 {
652 struct kmem_cache *cache;
653 struct sk_buff *skb;
654 bool pfmemalloc;
655 u8 *data;
656
657 cache = (flags & SKB_ALLOC_FCLONE)
658 ? net_hotdata.skbuff_fclone_cache : net_hotdata.skbuff_cache;
659
660 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
661 gfp_mask |= __GFP_MEMALLOC;
662
663 /* Get the HEAD */
664 if ((flags & (SKB_ALLOC_FCLONE | SKB_ALLOC_NAPI)) == SKB_ALLOC_NAPI &&
665 likely(node == NUMA_NO_NODE || node == numa_mem_id()))
666 skb = napi_skb_cache_get();
667 else
668 skb = kmem_cache_alloc_node(cache, gfp_mask & ~GFP_DMA, node);
669 if (unlikely(!skb))
670 return NULL;
671 prefetchw(skb);
672
673 /* We do our best to align skb_shared_info on a separate cache
674 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
675 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
676 * Both skb->head and skb_shared_info are cache line aligned.
677 */
678 data = kmalloc_reserve(&size, gfp_mask, node, &pfmemalloc);
679 if (unlikely(!data))
680 goto nodata;
681 /* kmalloc_size_roundup() might give us more room than requested.
682 * Put skb_shared_info exactly at the end of allocated zone,
683 * to allow max possible filling before reallocation.
684 */
685 prefetchw(data + SKB_WITH_OVERHEAD(size));
686
687 /*
688 * Only clear those fields we need to clear, not those that we will
689 * actually initialise below. Hence, don't put any more fields after
690 * the tail pointer in struct sk_buff!
691 */
692 memset(skb, 0, offsetof(struct sk_buff, tail));
693 __build_skb_around(skb, data, size);
694 skb->pfmemalloc = pfmemalloc;
695
696 if (flags & SKB_ALLOC_FCLONE) {
697 struct sk_buff_fclones *fclones;
698
699 fclones = container_of(skb, struct sk_buff_fclones, skb1);
700
701 skb->fclone = SKB_FCLONE_ORIG;
702 refcount_set(&fclones->fclone_ref, 1);
703 }
704
705 return skb;
706
707 nodata:
708 kmem_cache_free(cache, skb);
709 return NULL;
710 }
711 EXPORT_SYMBOL(__alloc_skb);
712
713 /**
714 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
715 * @dev: network device to receive on
716 * @len: length to allocate
717 * @gfp_mask: get_free_pages mask, passed to alloc_skb
718 *
719 * Allocate a new &sk_buff and assign it a usage count of one. The
720 * buffer has NET_SKB_PAD headroom built in. Users should allocate
721 * the headroom they think they need without accounting for the
722 * built in space. The built in space is used for optimisations.
723 *
724 * %NULL is returned if there is no free memory.
725 */
__netdev_alloc_skb(struct net_device * dev,unsigned int len,gfp_t gfp_mask)726 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
727 gfp_t gfp_mask)
728 {
729 struct page_frag_cache *nc;
730 struct sk_buff *skb;
731 bool pfmemalloc;
732 void *data;
733
734 len += NET_SKB_PAD;
735
736 /* If requested length is either too small or too big,
737 * we use kmalloc() for skb->head allocation.
738 */
739 if (len <= SKB_WITH_OVERHEAD(1024) ||
740 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
741 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
742 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
743 if (!skb)
744 goto skb_fail;
745 goto skb_success;
746 }
747
748 len = SKB_HEAD_ALIGN(len);
749
750 if (sk_memalloc_socks())
751 gfp_mask |= __GFP_MEMALLOC;
752
753 if (in_hardirq() || irqs_disabled()) {
754 nc = this_cpu_ptr(&netdev_alloc_cache);
755 data = page_frag_alloc(nc, len, gfp_mask);
756 pfmemalloc = page_frag_cache_is_pfmemalloc(nc);
757 } else {
758 local_bh_disable();
759 local_lock_nested_bh(&napi_alloc_cache.bh_lock);
760
761 nc = this_cpu_ptr(&napi_alloc_cache.page);
762 data = page_frag_alloc(nc, len, gfp_mask);
763 pfmemalloc = page_frag_cache_is_pfmemalloc(nc);
764
765 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
766 local_bh_enable();
767 }
768
769 if (unlikely(!data))
770 return NULL;
771
772 skb = __build_skb(data, len);
773 if (unlikely(!skb)) {
774 skb_free_frag(data);
775 return NULL;
776 }
777
778 if (pfmemalloc)
779 skb->pfmemalloc = 1;
780 skb->head_frag = 1;
781
782 skb_success:
783 skb_reserve(skb, NET_SKB_PAD);
784 skb->dev = dev;
785
786 skb_fail:
787 return skb;
788 }
789 EXPORT_SYMBOL(__netdev_alloc_skb);
790
791 /**
792 * napi_alloc_skb - allocate skbuff for rx in a specific NAPI instance
793 * @napi: napi instance this buffer was allocated for
794 * @len: length to allocate
795 *
796 * Allocate a new sk_buff for use in NAPI receive. This buffer will
797 * attempt to allocate the head from a special reserved region used
798 * only for NAPI Rx allocation. By doing this we can save several
799 * CPU cycles by avoiding having to disable and re-enable IRQs.
800 *
801 * %NULL is returned if there is no free memory.
802 */
napi_alloc_skb(struct napi_struct * napi,unsigned int len)803 struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len)
804 {
805 gfp_t gfp_mask = GFP_ATOMIC | __GFP_NOWARN;
806 struct napi_alloc_cache *nc;
807 struct sk_buff *skb;
808 bool pfmemalloc;
809 void *data;
810
811 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
812 len += NET_SKB_PAD + NET_IP_ALIGN;
813
814 /* If requested length is either too small or too big,
815 * we use kmalloc() for skb->head allocation.
816 * When the small frag allocator is available, prefer it over kmalloc
817 * for small fragments
818 */
819 if ((!NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) ||
820 len > SKB_WITH_OVERHEAD(PAGE_SIZE) ||
821 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
822 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX | SKB_ALLOC_NAPI,
823 NUMA_NO_NODE);
824 if (!skb)
825 goto skb_fail;
826 goto skb_success;
827 }
828
829 if (sk_memalloc_socks())
830 gfp_mask |= __GFP_MEMALLOC;
831
832 local_lock_nested_bh(&napi_alloc_cache.bh_lock);
833 nc = this_cpu_ptr(&napi_alloc_cache);
834 if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
835 /* we are artificially inflating the allocation size, but
836 * that is not as bad as it may look like, as:
837 * - 'len' less than GRO_MAX_HEAD makes little sense
838 * - On most systems, larger 'len' values lead to fragment
839 * size above 512 bytes
840 * - kmalloc would use the kmalloc-1k slab for such values
841 * - Builds with smaller GRO_MAX_HEAD will very likely do
842 * little networking, as that implies no WiFi and no
843 * tunnels support, and 32 bits arches.
844 */
845 len = SZ_1K;
846
847 data = page_frag_alloc_1k(&nc->page_small, gfp_mask);
848 pfmemalloc = NAPI_SMALL_PAGE_PFMEMALLOC(nc->page_small);
849 } else {
850 len = SKB_HEAD_ALIGN(len);
851
852 data = page_frag_alloc(&nc->page, len, gfp_mask);
853 pfmemalloc = page_frag_cache_is_pfmemalloc(&nc->page);
854 }
855 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
856
857 if (unlikely(!data))
858 return NULL;
859
860 skb = __napi_build_skb(data, len);
861 if (unlikely(!skb)) {
862 skb_free_frag(data);
863 return NULL;
864 }
865
866 if (pfmemalloc)
867 skb->pfmemalloc = 1;
868 skb->head_frag = 1;
869
870 skb_success:
871 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
872 skb->dev = napi->dev;
873
874 skb_fail:
875 return skb;
876 }
877 EXPORT_SYMBOL(napi_alloc_skb);
878
skb_add_rx_frag_netmem(struct sk_buff * skb,int i,netmem_ref netmem,int off,int size,unsigned int truesize)879 void skb_add_rx_frag_netmem(struct sk_buff *skb, int i, netmem_ref netmem,
880 int off, int size, unsigned int truesize)
881 {
882 DEBUG_NET_WARN_ON_ONCE(size > truesize);
883
884 skb_fill_netmem_desc(skb, i, netmem, off, size);
885 skb->len += size;
886 skb->data_len += size;
887 skb->truesize += truesize;
888 }
889 EXPORT_SYMBOL(skb_add_rx_frag_netmem);
890
skb_coalesce_rx_frag(struct sk_buff * skb,int i,int size,unsigned int truesize)891 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
892 unsigned int truesize)
893 {
894 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
895
896 DEBUG_NET_WARN_ON_ONCE(size > truesize);
897
898 skb_frag_size_add(frag, size);
899 skb->len += size;
900 skb->data_len += size;
901 skb->truesize += truesize;
902 }
903 EXPORT_SYMBOL(skb_coalesce_rx_frag);
904
skb_drop_list(struct sk_buff ** listp)905 static void skb_drop_list(struct sk_buff **listp)
906 {
907 kfree_skb_list(*listp);
908 *listp = NULL;
909 }
910
skb_drop_fraglist(struct sk_buff * skb)911 static inline void skb_drop_fraglist(struct sk_buff *skb)
912 {
913 skb_drop_list(&skb_shinfo(skb)->frag_list);
914 }
915
skb_clone_fraglist(struct sk_buff * skb)916 static void skb_clone_fraglist(struct sk_buff *skb)
917 {
918 struct sk_buff *list;
919
920 skb_walk_frags(skb, list)
921 skb_get(list);
922 }
923
is_pp_netmem(netmem_ref netmem)924 static bool is_pp_netmem(netmem_ref netmem)
925 {
926 return (netmem_get_pp_magic(netmem) & ~0x3UL) == PP_SIGNATURE;
927 }
928
skb_pp_cow_data(struct page_pool * pool,struct sk_buff ** pskb,unsigned int headroom)929 int skb_pp_cow_data(struct page_pool *pool, struct sk_buff **pskb,
930 unsigned int headroom)
931 {
932 #if IS_ENABLED(CONFIG_PAGE_POOL)
933 u32 size, truesize, len, max_head_size, off;
934 struct sk_buff *skb = *pskb, *nskb;
935 int err, i, head_off;
936 void *data;
937
938 /* XDP does not support fraglist so we need to linearize
939 * the skb.
940 */
941 if (skb_has_frag_list(skb))
942 return -EOPNOTSUPP;
943
944 max_head_size = SKB_WITH_OVERHEAD(PAGE_SIZE - headroom);
945 if (skb->len > max_head_size + MAX_SKB_FRAGS * PAGE_SIZE)
946 return -ENOMEM;
947
948 size = min_t(u32, skb->len, max_head_size);
949 truesize = SKB_HEAD_ALIGN(size) + headroom;
950 data = page_pool_dev_alloc_va(pool, &truesize);
951 if (!data)
952 return -ENOMEM;
953
954 nskb = napi_build_skb(data, truesize);
955 if (!nskb) {
956 page_pool_free_va(pool, data, true);
957 return -ENOMEM;
958 }
959
960 skb_reserve(nskb, headroom);
961 skb_copy_header(nskb, skb);
962 skb_mark_for_recycle(nskb);
963
964 err = skb_copy_bits(skb, 0, nskb->data, size);
965 if (err) {
966 consume_skb(nskb);
967 return err;
968 }
969 skb_put(nskb, size);
970
971 head_off = skb_headroom(nskb) - skb_headroom(skb);
972 skb_headers_offset_update(nskb, head_off);
973
974 off = size;
975 len = skb->len - off;
976 for (i = 0; i < MAX_SKB_FRAGS && off < skb->len; i++) {
977 struct page *page;
978 u32 page_off;
979
980 size = min_t(u32, len, PAGE_SIZE);
981 truesize = size;
982
983 page = page_pool_dev_alloc(pool, &page_off, &truesize);
984 if (!page) {
985 consume_skb(nskb);
986 return -ENOMEM;
987 }
988
989 skb_add_rx_frag(nskb, i, page, page_off, size, truesize);
990 err = skb_copy_bits(skb, off, page_address(page) + page_off,
991 size);
992 if (err) {
993 consume_skb(nskb);
994 return err;
995 }
996
997 len -= size;
998 off += size;
999 }
1000
1001 consume_skb(skb);
1002 *pskb = nskb;
1003
1004 return 0;
1005 #else
1006 return -EOPNOTSUPP;
1007 #endif
1008 }
1009 EXPORT_SYMBOL(skb_pp_cow_data);
1010
skb_cow_data_for_xdp(struct page_pool * pool,struct sk_buff ** pskb,struct bpf_prog * prog)1011 int skb_cow_data_for_xdp(struct page_pool *pool, struct sk_buff **pskb,
1012 struct bpf_prog *prog)
1013 {
1014 if (!prog->aux->xdp_has_frags)
1015 return -EINVAL;
1016
1017 return skb_pp_cow_data(pool, pskb, XDP_PACKET_HEADROOM);
1018 }
1019 EXPORT_SYMBOL(skb_cow_data_for_xdp);
1020
1021 #if IS_ENABLED(CONFIG_PAGE_POOL)
napi_pp_put_page(netmem_ref netmem)1022 bool napi_pp_put_page(netmem_ref netmem)
1023 {
1024 netmem = netmem_compound_head(netmem);
1025
1026 /* page->pp_magic is OR'ed with PP_SIGNATURE after the allocation
1027 * in order to preserve any existing bits, such as bit 0 for the
1028 * head page of compound page and bit 1 for pfmemalloc page, so
1029 * mask those bits for freeing side when doing below checking,
1030 * and page_is_pfmemalloc() is checked in __page_pool_put_page()
1031 * to avoid recycling the pfmemalloc page.
1032 */
1033 if (unlikely(!is_pp_netmem(netmem)))
1034 return false;
1035
1036 page_pool_put_full_netmem(netmem_get_pp(netmem), netmem, false);
1037
1038 return true;
1039 }
1040 EXPORT_SYMBOL(napi_pp_put_page);
1041 #endif
1042
skb_pp_recycle(struct sk_buff * skb,void * data)1043 static bool skb_pp_recycle(struct sk_buff *skb, void *data)
1044 {
1045 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle)
1046 return false;
1047 return napi_pp_put_page(page_to_netmem(virt_to_page(data)));
1048 }
1049
1050 /**
1051 * skb_pp_frag_ref() - Increase fragment references of a page pool aware skb
1052 * @skb: page pool aware skb
1053 *
1054 * Increase the fragment reference count (pp_ref_count) of a skb. This is
1055 * intended to gain fragment references only for page pool aware skbs,
1056 * i.e. when skb->pp_recycle is true, and not for fragments in a
1057 * non-pp-recycling skb. It has a fallback to increase references on normal
1058 * pages, as page pool aware skbs may also have normal page fragments.
1059 */
skb_pp_frag_ref(struct sk_buff * skb)1060 static int skb_pp_frag_ref(struct sk_buff *skb)
1061 {
1062 struct skb_shared_info *shinfo;
1063 netmem_ref head_netmem;
1064 int i;
1065
1066 if (!skb->pp_recycle)
1067 return -EINVAL;
1068
1069 shinfo = skb_shinfo(skb);
1070
1071 for (i = 0; i < shinfo->nr_frags; i++) {
1072 head_netmem = netmem_compound_head(shinfo->frags[i].netmem);
1073 if (likely(is_pp_netmem(head_netmem)))
1074 page_pool_ref_netmem(head_netmem);
1075 else
1076 page_ref_inc(netmem_to_page(head_netmem));
1077 }
1078 return 0;
1079 }
1080
skb_kfree_head(void * head,unsigned int end_offset)1081 static void skb_kfree_head(void *head, unsigned int end_offset)
1082 {
1083 if (end_offset == SKB_SMALL_HEAD_HEADROOM)
1084 kmem_cache_free(net_hotdata.skb_small_head_cache, head);
1085 else
1086 kfree(head);
1087 }
1088
skb_free_head(struct sk_buff * skb)1089 static void skb_free_head(struct sk_buff *skb)
1090 {
1091 unsigned char *head = skb->head;
1092
1093 if (skb->head_frag) {
1094 if (skb_pp_recycle(skb, head))
1095 return;
1096 skb_free_frag(head);
1097 } else {
1098 skb_kfree_head(head, skb_end_offset(skb));
1099 }
1100 }
1101
skb_release_data(struct sk_buff * skb,enum skb_drop_reason reason)1102 static void skb_release_data(struct sk_buff *skb, enum skb_drop_reason reason)
1103 {
1104 struct skb_shared_info *shinfo = skb_shinfo(skb);
1105 int i;
1106
1107 if (!skb_data_unref(skb, shinfo))
1108 goto exit;
1109
1110 if (skb_zcopy(skb)) {
1111 bool skip_unref = shinfo->flags & SKBFL_MANAGED_FRAG_REFS;
1112
1113 skb_zcopy_clear(skb, true);
1114 if (skip_unref)
1115 goto free_head;
1116 }
1117
1118 for (i = 0; i < shinfo->nr_frags; i++)
1119 __skb_frag_unref(&shinfo->frags[i], skb->pp_recycle);
1120
1121 free_head:
1122 if (shinfo->frag_list)
1123 kfree_skb_list_reason(shinfo->frag_list, reason);
1124
1125 skb_free_head(skb);
1126 exit:
1127 /* When we clone an SKB we copy the reycling bit. The pp_recycle
1128 * bit is only set on the head though, so in order to avoid races
1129 * while trying to recycle fragments on __skb_frag_unref() we need
1130 * to make one SKB responsible for triggering the recycle path.
1131 * So disable the recycling bit if an SKB is cloned and we have
1132 * additional references to the fragmented part of the SKB.
1133 * Eventually the last SKB will have the recycling bit set and it's
1134 * dataref set to 0, which will trigger the recycling
1135 */
1136 skb->pp_recycle = 0;
1137 }
1138
1139 /*
1140 * Free an skbuff by memory without cleaning the state.
1141 */
kfree_skbmem(struct sk_buff * skb)1142 static void kfree_skbmem(struct sk_buff *skb)
1143 {
1144 struct sk_buff_fclones *fclones;
1145
1146 switch (skb->fclone) {
1147 case SKB_FCLONE_UNAVAILABLE:
1148 kmem_cache_free(net_hotdata.skbuff_cache, skb);
1149 return;
1150
1151 case SKB_FCLONE_ORIG:
1152 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1153
1154 /* We usually free the clone (TX completion) before original skb
1155 * This test would have no chance to be true for the clone,
1156 * while here, branch prediction will be good.
1157 */
1158 if (refcount_read(&fclones->fclone_ref) == 1)
1159 goto fastpath;
1160 break;
1161
1162 default: /* SKB_FCLONE_CLONE */
1163 fclones = container_of(skb, struct sk_buff_fclones, skb2);
1164 break;
1165 }
1166 if (!refcount_dec_and_test(&fclones->fclone_ref))
1167 return;
1168 fastpath:
1169 kmem_cache_free(net_hotdata.skbuff_fclone_cache, fclones);
1170 }
1171
skb_release_head_state(struct sk_buff * skb)1172 void skb_release_head_state(struct sk_buff *skb)
1173 {
1174 skb_dst_drop(skb);
1175 if (skb->destructor) {
1176 DEBUG_NET_WARN_ON_ONCE(in_hardirq());
1177 skb->destructor(skb);
1178 }
1179 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
1180 nf_conntrack_put(skb_nfct(skb));
1181 #endif
1182 skb_ext_put(skb);
1183 }
1184
1185 /* Free everything but the sk_buff shell. */
skb_release_all(struct sk_buff * skb,enum skb_drop_reason reason)1186 static void skb_release_all(struct sk_buff *skb, enum skb_drop_reason reason)
1187 {
1188 skb_release_head_state(skb);
1189 if (likely(skb->head))
1190 skb_release_data(skb, reason);
1191 }
1192
1193 /**
1194 * __kfree_skb - private function
1195 * @skb: buffer
1196 *
1197 * Free an sk_buff. Release anything attached to the buffer.
1198 * Clean the state. This is an internal helper function. Users should
1199 * always call kfree_skb
1200 */
1201
__kfree_skb(struct sk_buff * skb)1202 void __kfree_skb(struct sk_buff *skb)
1203 {
1204 skb_release_all(skb, SKB_DROP_REASON_NOT_SPECIFIED);
1205 kfree_skbmem(skb);
1206 }
1207 EXPORT_SYMBOL(__kfree_skb);
1208
1209 static __always_inline
__sk_skb_reason_drop(struct sock * sk,struct sk_buff * skb,enum skb_drop_reason reason)1210 bool __sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb,
1211 enum skb_drop_reason reason)
1212 {
1213 if (unlikely(!skb_unref(skb)))
1214 return false;
1215
1216 DEBUG_NET_WARN_ON_ONCE(reason == SKB_NOT_DROPPED_YET ||
1217 u32_get_bits(reason,
1218 SKB_DROP_REASON_SUBSYS_MASK) >=
1219 SKB_DROP_REASON_SUBSYS_NUM);
1220
1221 if (reason == SKB_CONSUMED)
1222 trace_consume_skb(skb, __builtin_return_address(0));
1223 else
1224 trace_kfree_skb(skb, __builtin_return_address(0), reason, sk);
1225 return true;
1226 }
1227
1228 /**
1229 * sk_skb_reason_drop - free an sk_buff with special reason
1230 * @sk: the socket to receive @skb, or NULL if not applicable
1231 * @skb: buffer to free
1232 * @reason: reason why this skb is dropped
1233 *
1234 * Drop a reference to the buffer and free it if the usage count has hit
1235 * zero. Meanwhile, pass the receiving socket and drop reason to
1236 * 'kfree_skb' tracepoint.
1237 */
1238 void __fix_address
sk_skb_reason_drop(struct sock * sk,struct sk_buff * skb,enum skb_drop_reason reason)1239 sk_skb_reason_drop(struct sock *sk, struct sk_buff *skb, enum skb_drop_reason reason)
1240 {
1241 if (__sk_skb_reason_drop(sk, skb, reason))
1242 __kfree_skb(skb);
1243 }
1244 EXPORT_SYMBOL(sk_skb_reason_drop);
1245
1246 #define KFREE_SKB_BULK_SIZE 16
1247
1248 struct skb_free_array {
1249 unsigned int skb_count;
1250 void *skb_array[KFREE_SKB_BULK_SIZE];
1251 };
1252
kfree_skb_add_bulk(struct sk_buff * skb,struct skb_free_array * sa,enum skb_drop_reason reason)1253 static void kfree_skb_add_bulk(struct sk_buff *skb,
1254 struct skb_free_array *sa,
1255 enum skb_drop_reason reason)
1256 {
1257 /* if SKB is a clone, don't handle this case */
1258 if (unlikely(skb->fclone != SKB_FCLONE_UNAVAILABLE)) {
1259 __kfree_skb(skb);
1260 return;
1261 }
1262
1263 skb_release_all(skb, reason);
1264 sa->skb_array[sa->skb_count++] = skb;
1265
1266 if (unlikely(sa->skb_count == KFREE_SKB_BULK_SIZE)) {
1267 kmem_cache_free_bulk(net_hotdata.skbuff_cache, KFREE_SKB_BULK_SIZE,
1268 sa->skb_array);
1269 sa->skb_count = 0;
1270 }
1271 }
1272
1273 void __fix_address
kfree_skb_list_reason(struct sk_buff * segs,enum skb_drop_reason reason)1274 kfree_skb_list_reason(struct sk_buff *segs, enum skb_drop_reason reason)
1275 {
1276 struct skb_free_array sa;
1277
1278 sa.skb_count = 0;
1279
1280 while (segs) {
1281 struct sk_buff *next = segs->next;
1282
1283 if (__sk_skb_reason_drop(NULL, segs, reason)) {
1284 skb_poison_list(segs);
1285 kfree_skb_add_bulk(segs, &sa, reason);
1286 }
1287
1288 segs = next;
1289 }
1290
1291 if (sa.skb_count)
1292 kmem_cache_free_bulk(net_hotdata.skbuff_cache, sa.skb_count, sa.skb_array);
1293 }
1294 EXPORT_SYMBOL(kfree_skb_list_reason);
1295
1296 /* Dump skb information and contents.
1297 *
1298 * Must only be called from net_ratelimit()-ed paths.
1299 *
1300 * Dumps whole packets if full_pkt, only headers otherwise.
1301 */
skb_dump(const char * level,const struct sk_buff * skb,bool full_pkt)1302 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
1303 {
1304 struct skb_shared_info *sh = skb_shinfo(skb);
1305 struct net_device *dev = skb->dev;
1306 struct sock *sk = skb->sk;
1307 struct sk_buff *list_skb;
1308 bool has_mac, has_trans;
1309 int headroom, tailroom;
1310 int i, len, seg_len;
1311
1312 if (full_pkt)
1313 len = skb->len;
1314 else
1315 len = min_t(int, skb->len, MAX_HEADER + 128);
1316
1317 headroom = skb_headroom(skb);
1318 tailroom = skb_tailroom(skb);
1319
1320 has_mac = skb_mac_header_was_set(skb);
1321 has_trans = skb_transport_header_was_set(skb);
1322
1323 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
1324 "mac=(%d,%d) mac_len=%u net=(%d,%d) trans=%d\n"
1325 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
1326 "csum(0x%x start=%u offset=%u ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
1327 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n"
1328 "priority=0x%x mark=0x%x alloc_cpu=%u vlan_all=0x%x\n"
1329 "encapsulation=%d inner(proto=0x%04x, mac=%u, net=%u, trans=%u)\n",
1330 level, skb->len, headroom, skb_headlen(skb), tailroom,
1331 has_mac ? skb->mac_header : -1,
1332 has_mac ? skb_mac_header_len(skb) : -1,
1333 skb->mac_len,
1334 skb->network_header,
1335 has_trans ? skb_network_header_len(skb) : -1,
1336 has_trans ? skb->transport_header : -1,
1337 sh->tx_flags, sh->nr_frags,
1338 sh->gso_size, sh->gso_type, sh->gso_segs,
1339 skb->csum, skb->csum_start, skb->csum_offset, skb->ip_summed,
1340 skb->csum_complete_sw, skb->csum_valid, skb->csum_level,
1341 skb->hash, skb->sw_hash, skb->l4_hash,
1342 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif,
1343 skb->priority, skb->mark, skb->alloc_cpu, skb->vlan_all,
1344 skb->encapsulation, skb->inner_protocol, skb->inner_mac_header,
1345 skb->inner_network_header, skb->inner_transport_header);
1346
1347 if (dev)
1348 printk("%sdev name=%s feat=%pNF\n",
1349 level, dev->name, &dev->features);
1350 if (sk)
1351 printk("%ssk family=%hu type=%u proto=%u\n",
1352 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
1353
1354 if (full_pkt && headroom)
1355 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
1356 16, 1, skb->head, headroom, false);
1357
1358 seg_len = min_t(int, skb_headlen(skb), len);
1359 if (seg_len)
1360 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
1361 16, 1, skb->data, seg_len, false);
1362 len -= seg_len;
1363
1364 if (full_pkt && tailroom)
1365 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
1366 16, 1, skb_tail_pointer(skb), tailroom, false);
1367
1368 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
1369 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1370 u32 p_off, p_len, copied;
1371 struct page *p;
1372 u8 *vaddr;
1373
1374 if (skb_frag_is_net_iov(frag)) {
1375 printk("%sskb frag %d: not readable\n", level, i);
1376 len -= skb_frag_size(frag);
1377 if (!len)
1378 break;
1379 continue;
1380 }
1381
1382 skb_frag_foreach_page(frag, skb_frag_off(frag),
1383 skb_frag_size(frag), p, p_off, p_len,
1384 copied) {
1385 seg_len = min_t(int, p_len, len);
1386 vaddr = kmap_atomic(p);
1387 print_hex_dump(level, "skb frag: ",
1388 DUMP_PREFIX_OFFSET,
1389 16, 1, vaddr + p_off, seg_len, false);
1390 kunmap_atomic(vaddr);
1391 len -= seg_len;
1392 if (!len)
1393 break;
1394 }
1395 }
1396
1397 if (full_pkt && skb_has_frag_list(skb)) {
1398 printk("skb fraglist:\n");
1399 skb_walk_frags(skb, list_skb)
1400 skb_dump(level, list_skb, true);
1401 }
1402 }
1403 EXPORT_SYMBOL(skb_dump);
1404
1405 /**
1406 * skb_tx_error - report an sk_buff xmit error
1407 * @skb: buffer that triggered an error
1408 *
1409 * Report xmit error if a device callback is tracking this skb.
1410 * skb must be freed afterwards.
1411 */
skb_tx_error(struct sk_buff * skb)1412 void skb_tx_error(struct sk_buff *skb)
1413 {
1414 if (skb) {
1415 skb_zcopy_downgrade_managed(skb);
1416 skb_zcopy_clear(skb, true);
1417 }
1418 }
1419 EXPORT_SYMBOL(skb_tx_error);
1420
1421 #ifdef CONFIG_TRACEPOINTS
1422 /**
1423 * consume_skb - free an skbuff
1424 * @skb: buffer to free
1425 *
1426 * Drop a ref to the buffer and free it if the usage count has hit zero
1427 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
1428 * is being dropped after a failure and notes that
1429 */
consume_skb(struct sk_buff * skb)1430 void consume_skb(struct sk_buff *skb)
1431 {
1432 if (!skb_unref(skb))
1433 return;
1434
1435 trace_consume_skb(skb, __builtin_return_address(0));
1436 __kfree_skb(skb);
1437 }
1438 EXPORT_SYMBOL(consume_skb);
1439 #endif
1440
1441 /**
1442 * __consume_stateless_skb - free an skbuff, assuming it is stateless
1443 * @skb: buffer to free
1444 *
1445 * Alike consume_skb(), but this variant assumes that this is the last
1446 * skb reference and all the head states have been already dropped
1447 */
__consume_stateless_skb(struct sk_buff * skb)1448 void __consume_stateless_skb(struct sk_buff *skb)
1449 {
1450 trace_consume_skb(skb, __builtin_return_address(0));
1451 skb_release_data(skb, SKB_CONSUMED);
1452 kfree_skbmem(skb);
1453 }
1454
napi_skb_cache_put(struct sk_buff * skb)1455 static void napi_skb_cache_put(struct sk_buff *skb)
1456 {
1457 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
1458 u32 i;
1459
1460 if (!kasan_mempool_poison_object(skb))
1461 return;
1462
1463 local_lock_nested_bh(&napi_alloc_cache.bh_lock);
1464 nc->skb_cache[nc->skb_count++] = skb;
1465
1466 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
1467 for (i = NAPI_SKB_CACHE_HALF; i < NAPI_SKB_CACHE_SIZE; i++)
1468 kasan_mempool_unpoison_object(nc->skb_cache[i],
1469 kmem_cache_size(net_hotdata.skbuff_cache));
1470
1471 kmem_cache_free_bulk(net_hotdata.skbuff_cache, NAPI_SKB_CACHE_HALF,
1472 nc->skb_cache + NAPI_SKB_CACHE_HALF);
1473 nc->skb_count = NAPI_SKB_CACHE_HALF;
1474 }
1475 local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
1476 }
1477
__napi_kfree_skb(struct sk_buff * skb,enum skb_drop_reason reason)1478 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
1479 {
1480 skb_release_all(skb, reason);
1481 napi_skb_cache_put(skb);
1482 }
1483
napi_skb_free_stolen_head(struct sk_buff * skb)1484 void napi_skb_free_stolen_head(struct sk_buff *skb)
1485 {
1486 if (unlikely(skb->slow_gro)) {
1487 nf_reset_ct(skb);
1488 skb_dst_drop(skb);
1489 skb_ext_put(skb);
1490 skb_orphan(skb);
1491 skb->slow_gro = 0;
1492 }
1493 napi_skb_cache_put(skb);
1494 }
1495
napi_consume_skb(struct sk_buff * skb,int budget)1496 void napi_consume_skb(struct sk_buff *skb, int budget)
1497 {
1498 /* Zero budget indicate non-NAPI context called us, like netpoll */
1499 if (unlikely(!budget)) {
1500 dev_consume_skb_any(skb);
1501 return;
1502 }
1503
1504 DEBUG_NET_WARN_ON_ONCE(!in_softirq());
1505
1506 if (!skb_unref(skb))
1507 return;
1508
1509 /* if reaching here SKB is ready to free */
1510 trace_consume_skb(skb, __builtin_return_address(0));
1511
1512 /* if SKB is a clone, don't handle this case */
1513 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
1514 __kfree_skb(skb);
1515 return;
1516 }
1517
1518 skb_release_all(skb, SKB_CONSUMED);
1519 napi_skb_cache_put(skb);
1520 }
1521 EXPORT_SYMBOL(napi_consume_skb);
1522
1523 /* Make sure a field is contained by headers group */
1524 #define CHECK_SKB_FIELD(field) \
1525 BUILD_BUG_ON(offsetof(struct sk_buff, field) != \
1526 offsetof(struct sk_buff, headers.field)); \
1527
__copy_skb_header(struct sk_buff * new,const struct sk_buff * old)1528 static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1529 {
1530 new->tstamp = old->tstamp;
1531 /* We do not copy old->sk */
1532 new->dev = old->dev;
1533 memcpy(new->cb, old->cb, sizeof(old->cb));
1534 skb_dst_copy(new, old);
1535 __skb_ext_copy(new, old);
1536 __nf_copy(new, old, false);
1537
1538 /* Note : this field could be in the headers group.
1539 * It is not yet because we do not want to have a 16 bit hole
1540 */
1541 new->queue_mapping = old->queue_mapping;
1542
1543 memcpy(&new->headers, &old->headers, sizeof(new->headers));
1544 CHECK_SKB_FIELD(protocol);
1545 CHECK_SKB_FIELD(csum);
1546 CHECK_SKB_FIELD(hash);
1547 CHECK_SKB_FIELD(priority);
1548 CHECK_SKB_FIELD(skb_iif);
1549 CHECK_SKB_FIELD(vlan_proto);
1550 CHECK_SKB_FIELD(vlan_tci);
1551 CHECK_SKB_FIELD(transport_header);
1552 CHECK_SKB_FIELD(network_header);
1553 CHECK_SKB_FIELD(mac_header);
1554 CHECK_SKB_FIELD(inner_protocol);
1555 CHECK_SKB_FIELD(inner_transport_header);
1556 CHECK_SKB_FIELD(inner_network_header);
1557 CHECK_SKB_FIELD(inner_mac_header);
1558 CHECK_SKB_FIELD(mark);
1559 #ifdef CONFIG_NETWORK_SECMARK
1560 CHECK_SKB_FIELD(secmark);
1561 #endif
1562 #ifdef CONFIG_NET_RX_BUSY_POLL
1563 CHECK_SKB_FIELD(napi_id);
1564 #endif
1565 CHECK_SKB_FIELD(alloc_cpu);
1566 #ifdef CONFIG_XPS
1567 CHECK_SKB_FIELD(sender_cpu);
1568 #endif
1569 #ifdef CONFIG_NET_SCHED
1570 CHECK_SKB_FIELD(tc_index);
1571 #endif
1572
1573 }
1574
1575 /*
1576 * You should not add any new code to this function. Add it to
1577 * __copy_skb_header above instead.
1578 */
__skb_clone(struct sk_buff * n,struct sk_buff * skb)1579 static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
1580 {
1581 #define C(x) n->x = skb->x
1582
1583 n->next = n->prev = NULL;
1584 n->sk = NULL;
1585 __copy_skb_header(n, skb);
1586
1587 C(len);
1588 C(data_len);
1589 C(mac_len);
1590 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
1591 n->cloned = 1;
1592 n->nohdr = 0;
1593 n->peeked = 0;
1594 C(pfmemalloc);
1595 C(pp_recycle);
1596 n->destructor = NULL;
1597 C(tail);
1598 C(end);
1599 C(head);
1600 C(head_frag);
1601 C(data);
1602 C(truesize);
1603 refcount_set(&n->users, 1);
1604
1605 atomic_inc(&(skb_shinfo(skb)->dataref));
1606 skb->cloned = 1;
1607
1608 return n;
1609 #undef C
1610 }
1611
1612 /**
1613 * alloc_skb_for_msg() - allocate sk_buff to wrap frag list forming a msg
1614 * @first: first sk_buff of the msg
1615 */
alloc_skb_for_msg(struct sk_buff * first)1616 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1617 {
1618 struct sk_buff *n;
1619
1620 n = alloc_skb(0, GFP_ATOMIC);
1621 if (!n)
1622 return NULL;
1623
1624 n->len = first->len;
1625 n->data_len = first->len;
1626 n->truesize = first->truesize;
1627
1628 skb_shinfo(n)->frag_list = first;
1629
1630 __copy_skb_header(n, first);
1631 n->destructor = NULL;
1632
1633 return n;
1634 }
1635 EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1636
1637 /**
1638 * skb_morph - morph one skb into another
1639 * @dst: the skb to receive the contents
1640 * @src: the skb to supply the contents
1641 *
1642 * This is identical to skb_clone except that the target skb is
1643 * supplied by the user.
1644 *
1645 * The target skb is returned upon exit.
1646 */
skb_morph(struct sk_buff * dst,struct sk_buff * src)1647 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1648 {
1649 skb_release_all(dst, SKB_CONSUMED);
1650 return __skb_clone(dst, src);
1651 }
1652 EXPORT_SYMBOL_GPL(skb_morph);
1653
mm_account_pinned_pages(struct mmpin * mmp,size_t size)1654 int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1655 {
1656 unsigned long max_pg, num_pg, new_pg, old_pg, rlim;
1657 struct user_struct *user;
1658
1659 if (capable(CAP_IPC_LOCK) || !size)
1660 return 0;
1661
1662 rlim = rlimit(RLIMIT_MEMLOCK);
1663 if (rlim == RLIM_INFINITY)
1664 return 0;
1665
1666 num_pg = (size >> PAGE_SHIFT) + 2; /* worst case */
1667 max_pg = rlim >> PAGE_SHIFT;
1668 user = mmp->user ? : current_user();
1669
1670 old_pg = atomic_long_read(&user->locked_vm);
1671 do {
1672 new_pg = old_pg + num_pg;
1673 if (new_pg > max_pg)
1674 return -ENOBUFS;
1675 } while (!atomic_long_try_cmpxchg(&user->locked_vm, &old_pg, new_pg));
1676
1677 if (!mmp->user) {
1678 mmp->user = get_uid(user);
1679 mmp->num_pg = num_pg;
1680 } else {
1681 mmp->num_pg += num_pg;
1682 }
1683
1684 return 0;
1685 }
1686 EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1687
mm_unaccount_pinned_pages(struct mmpin * mmp)1688 void mm_unaccount_pinned_pages(struct mmpin *mmp)
1689 {
1690 if (mmp->user) {
1691 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1692 free_uid(mmp->user);
1693 }
1694 }
1695 EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1696
msg_zerocopy_alloc(struct sock * sk,size_t size)1697 static struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size)
1698 {
1699 struct ubuf_info_msgzc *uarg;
1700 struct sk_buff *skb;
1701
1702 WARN_ON_ONCE(!in_task());
1703
1704 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1705 if (!skb)
1706 return NULL;
1707
1708 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1709 uarg = (void *)skb->cb;
1710 uarg->mmp.user = NULL;
1711
1712 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1713 kfree_skb(skb);
1714 return NULL;
1715 }
1716
1717 uarg->ubuf.ops = &msg_zerocopy_ubuf_ops;
1718 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1719 uarg->len = 1;
1720 uarg->bytelen = size;
1721 uarg->zerocopy = 1;
1722 uarg->ubuf.flags = SKBFL_ZEROCOPY_FRAG | SKBFL_DONT_ORPHAN;
1723 refcount_set(&uarg->ubuf.refcnt, 1);
1724 sock_hold(sk);
1725
1726 return &uarg->ubuf;
1727 }
1728
skb_from_uarg(struct ubuf_info_msgzc * uarg)1729 static inline struct sk_buff *skb_from_uarg(struct ubuf_info_msgzc *uarg)
1730 {
1731 return container_of((void *)uarg, struct sk_buff, cb);
1732 }
1733
msg_zerocopy_realloc(struct sock * sk,size_t size,struct ubuf_info * uarg)1734 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size,
1735 struct ubuf_info *uarg)
1736 {
1737 if (uarg) {
1738 struct ubuf_info_msgzc *uarg_zc;
1739 const u32 byte_limit = 1 << 19; /* limit to a few TSO */
1740 u32 bytelen, next;
1741
1742 /* there might be non MSG_ZEROCOPY users */
1743 if (uarg->ops != &msg_zerocopy_ubuf_ops)
1744 return NULL;
1745
1746 /* realloc only when socket is locked (TCP, UDP cork),
1747 * so uarg->len and sk_zckey access is serialized
1748 */
1749 if (!sock_owned_by_user(sk)) {
1750 WARN_ON_ONCE(1);
1751 return NULL;
1752 }
1753
1754 uarg_zc = uarg_to_msgzc(uarg);
1755 bytelen = uarg_zc->bytelen + size;
1756 if (uarg_zc->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1757 /* TCP can create new skb to attach new uarg */
1758 if (sk->sk_type == SOCK_STREAM)
1759 goto new_alloc;
1760 return NULL;
1761 }
1762
1763 next = (u32)atomic_read(&sk->sk_zckey);
1764 if ((u32)(uarg_zc->id + uarg_zc->len) == next) {
1765 if (mm_account_pinned_pages(&uarg_zc->mmp, size))
1766 return NULL;
1767 uarg_zc->len++;
1768 uarg_zc->bytelen = bytelen;
1769 atomic_set(&sk->sk_zckey, ++next);
1770
1771 /* no extra ref when appending to datagram (MSG_MORE) */
1772 if (sk->sk_type == SOCK_STREAM)
1773 net_zcopy_get(uarg);
1774
1775 return uarg;
1776 }
1777 }
1778
1779 new_alloc:
1780 return msg_zerocopy_alloc(sk, size);
1781 }
1782 EXPORT_SYMBOL_GPL(msg_zerocopy_realloc);
1783
skb_zerocopy_notify_extend(struct sk_buff * skb,u32 lo,u16 len)1784 static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1785 {
1786 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1787 u32 old_lo, old_hi;
1788 u64 sum_len;
1789
1790 old_lo = serr->ee.ee_info;
1791 old_hi = serr->ee.ee_data;
1792 sum_len = old_hi - old_lo + 1ULL + len;
1793
1794 if (sum_len >= (1ULL << 32))
1795 return false;
1796
1797 if (lo != old_hi + 1)
1798 return false;
1799
1800 serr->ee.ee_data += len;
1801 return true;
1802 }
1803
__msg_zerocopy_callback(struct ubuf_info_msgzc * uarg)1804 static void __msg_zerocopy_callback(struct ubuf_info_msgzc *uarg)
1805 {
1806 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1807 struct sock_exterr_skb *serr;
1808 struct sock *sk = skb->sk;
1809 struct sk_buff_head *q;
1810 unsigned long flags;
1811 bool is_zerocopy;
1812 u32 lo, hi;
1813 u16 len;
1814
1815 mm_unaccount_pinned_pages(&uarg->mmp);
1816
1817 /* if !len, there was only 1 call, and it was aborted
1818 * so do not queue a completion notification
1819 */
1820 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1821 goto release;
1822
1823 len = uarg->len;
1824 lo = uarg->id;
1825 hi = uarg->id + len - 1;
1826 is_zerocopy = uarg->zerocopy;
1827
1828 serr = SKB_EXT_ERR(skb);
1829 memset(serr, 0, sizeof(*serr));
1830 serr->ee.ee_errno = 0;
1831 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1832 serr->ee.ee_data = hi;
1833 serr->ee.ee_info = lo;
1834 if (!is_zerocopy)
1835 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1836
1837 q = &sk->sk_error_queue;
1838 spin_lock_irqsave(&q->lock, flags);
1839 tail = skb_peek_tail(q);
1840 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1841 !skb_zerocopy_notify_extend(tail, lo, len)) {
1842 __skb_queue_tail(q, skb);
1843 skb = NULL;
1844 }
1845 spin_unlock_irqrestore(&q->lock, flags);
1846
1847 sk_error_report(sk);
1848
1849 release:
1850 consume_skb(skb);
1851 sock_put(sk);
1852 }
1853
msg_zerocopy_complete(struct sk_buff * skb,struct ubuf_info * uarg,bool success)1854 static void msg_zerocopy_complete(struct sk_buff *skb, struct ubuf_info *uarg,
1855 bool success)
1856 {
1857 struct ubuf_info_msgzc *uarg_zc = uarg_to_msgzc(uarg);
1858
1859 uarg_zc->zerocopy = uarg_zc->zerocopy & success;
1860
1861 if (refcount_dec_and_test(&uarg->refcnt))
1862 __msg_zerocopy_callback(uarg_zc);
1863 }
1864
msg_zerocopy_put_abort(struct ubuf_info * uarg,bool have_uref)1865 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1866 {
1867 struct sock *sk = skb_from_uarg(uarg_to_msgzc(uarg))->sk;
1868
1869 atomic_dec(&sk->sk_zckey);
1870 uarg_to_msgzc(uarg)->len--;
1871
1872 if (have_uref)
1873 msg_zerocopy_complete(NULL, uarg, true);
1874 }
1875 EXPORT_SYMBOL_GPL(msg_zerocopy_put_abort);
1876
1877 const struct ubuf_info_ops msg_zerocopy_ubuf_ops = {
1878 .complete = msg_zerocopy_complete,
1879 };
1880 EXPORT_SYMBOL_GPL(msg_zerocopy_ubuf_ops);
1881
skb_zerocopy_iter_stream(struct sock * sk,struct sk_buff * skb,struct msghdr * msg,int len,struct ubuf_info * uarg)1882 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1883 struct msghdr *msg, int len,
1884 struct ubuf_info *uarg)
1885 {
1886 int err, orig_len = skb->len;
1887
1888 if (uarg->ops->link_skb) {
1889 err = uarg->ops->link_skb(skb, uarg);
1890 if (err)
1891 return err;
1892 } else {
1893 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1894
1895 /* An skb can only point to one uarg. This edge case happens
1896 * when TCP appends to an skb, but zerocopy_realloc triggered
1897 * a new alloc.
1898 */
1899 if (orig_uarg && uarg != orig_uarg)
1900 return -EEXIST;
1901 }
1902
1903 err = __zerocopy_sg_from_iter(msg, sk, skb, &msg->msg_iter, len);
1904 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1905 struct sock *save_sk = skb->sk;
1906
1907 /* Streams do not free skb on error. Reset to prev state. */
1908 iov_iter_revert(&msg->msg_iter, skb->len - orig_len);
1909 skb->sk = sk;
1910 ___pskb_trim(skb, orig_len);
1911 skb->sk = save_sk;
1912 return err;
1913 }
1914
1915 skb_zcopy_set(skb, uarg, NULL);
1916 return skb->len - orig_len;
1917 }
1918 EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1919
__skb_zcopy_downgrade_managed(struct sk_buff * skb)1920 void __skb_zcopy_downgrade_managed(struct sk_buff *skb)
1921 {
1922 int i;
1923
1924 skb_shinfo(skb)->flags &= ~SKBFL_MANAGED_FRAG_REFS;
1925 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1926 skb_frag_ref(skb, i);
1927 }
1928 EXPORT_SYMBOL_GPL(__skb_zcopy_downgrade_managed);
1929
skb_zerocopy_clone(struct sk_buff * nskb,struct sk_buff * orig,gfp_t gfp_mask)1930 static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1931 gfp_t gfp_mask)
1932 {
1933 if (skb_zcopy(orig)) {
1934 if (skb_zcopy(nskb)) {
1935 /* !gfp_mask callers are verified to !skb_zcopy(nskb) */
1936 if (!gfp_mask) {
1937 WARN_ON_ONCE(1);
1938 return -ENOMEM;
1939 }
1940 if (skb_uarg(nskb) == skb_uarg(orig))
1941 return 0;
1942 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1943 return -EIO;
1944 }
1945 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1946 }
1947 return 0;
1948 }
1949
1950 /**
1951 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
1952 * @skb: the skb to modify
1953 * @gfp_mask: allocation priority
1954 *
1955 * This must be called on skb with SKBFL_ZEROCOPY_ENABLE.
1956 * It will copy all frags into kernel and drop the reference
1957 * to userspace pages.
1958 *
1959 * If this function is called from an interrupt gfp_mask() must be
1960 * %GFP_ATOMIC.
1961 *
1962 * Returns 0 on success or a negative error code on failure
1963 * to allocate kernel memory to copy to.
1964 */
skb_copy_ubufs(struct sk_buff * skb,gfp_t gfp_mask)1965 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1966 {
1967 int num_frags = skb_shinfo(skb)->nr_frags;
1968 struct page *page, *head = NULL;
1969 int i, order, psize, new_frags;
1970 u32 d_off;
1971
1972 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1973 return -EINVAL;
1974
1975 if (!skb_frags_readable(skb))
1976 return -EFAULT;
1977
1978 if (!num_frags)
1979 goto release;
1980
1981 /* We might have to allocate high order pages, so compute what minimum
1982 * page order is needed.
1983 */
1984 order = 0;
1985 while ((PAGE_SIZE << order) * MAX_SKB_FRAGS < __skb_pagelen(skb))
1986 order++;
1987 psize = (PAGE_SIZE << order);
1988
1989 new_frags = (__skb_pagelen(skb) + psize - 1) >> (PAGE_SHIFT + order);
1990 for (i = 0; i < new_frags; i++) {
1991 page = alloc_pages(gfp_mask | __GFP_COMP, order);
1992 if (!page) {
1993 while (head) {
1994 struct page *next = (struct page *)page_private(head);
1995 put_page(head);
1996 head = next;
1997 }
1998 return -ENOMEM;
1999 }
2000 set_page_private(page, (unsigned long)head);
2001 head = page;
2002 }
2003
2004 page = head;
2005 d_off = 0;
2006 for (i = 0; i < num_frags; i++) {
2007 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2008 u32 p_off, p_len, copied;
2009 struct page *p;
2010 u8 *vaddr;
2011
2012 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
2013 p, p_off, p_len, copied) {
2014 u32 copy, done = 0;
2015 vaddr = kmap_atomic(p);
2016
2017 while (done < p_len) {
2018 if (d_off == psize) {
2019 d_off = 0;
2020 page = (struct page *)page_private(page);
2021 }
2022 copy = min_t(u32, psize - d_off, p_len - done);
2023 memcpy(page_address(page) + d_off,
2024 vaddr + p_off + done, copy);
2025 done += copy;
2026 d_off += copy;
2027 }
2028 kunmap_atomic(vaddr);
2029 }
2030 }
2031
2032 /* skb frags release userspace buffers */
2033 for (i = 0; i < num_frags; i++)
2034 skb_frag_unref(skb, i);
2035
2036 /* skb frags point to kernel buffers */
2037 for (i = 0; i < new_frags - 1; i++) {
2038 __skb_fill_netmem_desc(skb, i, page_to_netmem(head), 0, psize);
2039 head = (struct page *)page_private(head);
2040 }
2041 __skb_fill_netmem_desc(skb, new_frags - 1, page_to_netmem(head), 0,
2042 d_off);
2043 skb_shinfo(skb)->nr_frags = new_frags;
2044
2045 release:
2046 skb_zcopy_clear(skb, false);
2047 return 0;
2048 }
2049 EXPORT_SYMBOL_GPL(skb_copy_ubufs);
2050
2051 /**
2052 * skb_clone - duplicate an sk_buff
2053 * @skb: buffer to clone
2054 * @gfp_mask: allocation priority
2055 *
2056 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
2057 * copies share the same packet data but not structure. The new
2058 * buffer has a reference count of 1. If the allocation fails the
2059 * function returns %NULL otherwise the new buffer is returned.
2060 *
2061 * If this function is called from an interrupt gfp_mask() must be
2062 * %GFP_ATOMIC.
2063 */
2064
skb_clone(struct sk_buff * skb,gfp_t gfp_mask)2065 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
2066 {
2067 struct sk_buff_fclones *fclones = container_of(skb,
2068 struct sk_buff_fclones,
2069 skb1);
2070 struct sk_buff *n;
2071
2072 if (skb_orphan_frags(skb, gfp_mask))
2073 return NULL;
2074
2075 if (skb->fclone == SKB_FCLONE_ORIG &&
2076 refcount_read(&fclones->fclone_ref) == 1) {
2077 n = &fclones->skb2;
2078 refcount_set(&fclones->fclone_ref, 2);
2079 n->fclone = SKB_FCLONE_CLONE;
2080 } else {
2081 if (skb_pfmemalloc(skb))
2082 gfp_mask |= __GFP_MEMALLOC;
2083
2084 n = kmem_cache_alloc(net_hotdata.skbuff_cache, gfp_mask);
2085 if (!n)
2086 return NULL;
2087
2088 n->fclone = SKB_FCLONE_UNAVAILABLE;
2089 }
2090
2091 return __skb_clone(n, skb);
2092 }
2093 EXPORT_SYMBOL(skb_clone);
2094
skb_headers_offset_update(struct sk_buff * skb,int off)2095 void skb_headers_offset_update(struct sk_buff *skb, int off)
2096 {
2097 /* Only adjust this if it actually is csum_start rather than csum */
2098 if (skb->ip_summed == CHECKSUM_PARTIAL)
2099 skb->csum_start += off;
2100 /* {transport,network,mac}_header and tail are relative to skb->head */
2101 skb->transport_header += off;
2102 skb->network_header += off;
2103 if (skb_mac_header_was_set(skb))
2104 skb->mac_header += off;
2105 skb->inner_transport_header += off;
2106 skb->inner_network_header += off;
2107 skb->inner_mac_header += off;
2108 }
2109 EXPORT_SYMBOL(skb_headers_offset_update);
2110
skb_copy_header(struct sk_buff * new,const struct sk_buff * old)2111 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
2112 {
2113 __copy_skb_header(new, old);
2114
2115 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
2116 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
2117 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
2118 }
2119 EXPORT_SYMBOL(skb_copy_header);
2120
skb_alloc_rx_flag(const struct sk_buff * skb)2121 static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
2122 {
2123 if (skb_pfmemalloc(skb))
2124 return SKB_ALLOC_RX;
2125 return 0;
2126 }
2127
2128 /**
2129 * skb_copy - create private copy of an sk_buff
2130 * @skb: buffer to copy
2131 * @gfp_mask: allocation priority
2132 *
2133 * Make a copy of both an &sk_buff and its data. This is used when the
2134 * caller wishes to modify the data and needs a private copy of the
2135 * data to alter. Returns %NULL on failure or the pointer to the buffer
2136 * on success. The returned buffer has a reference count of 1.
2137 *
2138 * As by-product this function converts non-linear &sk_buff to linear
2139 * one, so that &sk_buff becomes completely private and caller is allowed
2140 * to modify all the data of returned buffer. This means that this
2141 * function is not recommended for use in circumstances when only
2142 * header is going to be modified. Use pskb_copy() instead.
2143 */
2144
skb_copy(const struct sk_buff * skb,gfp_t gfp_mask)2145 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
2146 {
2147 struct sk_buff *n;
2148 unsigned int size;
2149 int headerlen;
2150
2151 if (!skb_frags_readable(skb))
2152 return NULL;
2153
2154 if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2155 return NULL;
2156
2157 headerlen = skb_headroom(skb);
2158 size = skb_end_offset(skb) + skb->data_len;
2159 n = __alloc_skb(size, gfp_mask,
2160 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
2161 if (!n)
2162 return NULL;
2163
2164 /* Set the data pointer */
2165 skb_reserve(n, headerlen);
2166 /* Set the tail pointer and length */
2167 skb_put(n, skb->len);
2168
2169 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
2170
2171 skb_copy_header(n, skb);
2172 return n;
2173 }
2174 EXPORT_SYMBOL(skb_copy);
2175
2176 /**
2177 * __pskb_copy_fclone - create copy of an sk_buff with private head.
2178 * @skb: buffer to copy
2179 * @headroom: headroom of new skb
2180 * @gfp_mask: allocation priority
2181 * @fclone: if true allocate the copy of the skb from the fclone
2182 * cache instead of the head cache; it is recommended to set this
2183 * to true for the cases where the copy will likely be cloned
2184 *
2185 * Make a copy of both an &sk_buff and part of its data, located
2186 * in header. Fragmented data remain shared. This is used when
2187 * the caller wishes to modify only header of &sk_buff and needs
2188 * private copy of the header to alter. Returns %NULL on failure
2189 * or the pointer to the buffer on success.
2190 * The returned buffer has a reference count of 1.
2191 */
2192
__pskb_copy_fclone(struct sk_buff * skb,int headroom,gfp_t gfp_mask,bool fclone)2193 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
2194 gfp_t gfp_mask, bool fclone)
2195 {
2196 unsigned int size = skb_headlen(skb) + headroom;
2197 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
2198 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
2199
2200 if (!n)
2201 goto out;
2202
2203 /* Set the data pointer */
2204 skb_reserve(n, headroom);
2205 /* Set the tail pointer and length */
2206 skb_put(n, skb_headlen(skb));
2207 /* Copy the bytes */
2208 skb_copy_from_linear_data(skb, n->data, n->len);
2209
2210 n->truesize += skb->data_len;
2211 n->data_len = skb->data_len;
2212 n->len = skb->len;
2213
2214 if (skb_shinfo(skb)->nr_frags) {
2215 int i;
2216
2217 if (skb_orphan_frags(skb, gfp_mask) ||
2218 skb_zerocopy_clone(n, skb, gfp_mask)) {
2219 kfree_skb(n);
2220 n = NULL;
2221 goto out;
2222 }
2223 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2224 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
2225 skb_frag_ref(skb, i);
2226 }
2227 skb_shinfo(n)->nr_frags = i;
2228 }
2229
2230 if (skb_has_frag_list(skb)) {
2231 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
2232 skb_clone_fraglist(n);
2233 }
2234
2235 skb_copy_header(n, skb);
2236 out:
2237 return n;
2238 }
2239 EXPORT_SYMBOL(__pskb_copy_fclone);
2240
2241 /**
2242 * pskb_expand_head - reallocate header of &sk_buff
2243 * @skb: buffer to reallocate
2244 * @nhead: room to add at head
2245 * @ntail: room to add at tail
2246 * @gfp_mask: allocation priority
2247 *
2248 * Expands (or creates identical copy, if @nhead and @ntail are zero)
2249 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
2250 * reference count of 1. Returns zero in the case of success or error,
2251 * if expansion failed. In the last case, &sk_buff is not changed.
2252 *
2253 * All the pointers pointing into skb header may change and must be
2254 * reloaded after call to this function.
2255 */
2256
pskb_expand_head(struct sk_buff * skb,int nhead,int ntail,gfp_t gfp_mask)2257 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
2258 gfp_t gfp_mask)
2259 {
2260 unsigned int osize = skb_end_offset(skb);
2261 unsigned int size = osize + nhead + ntail;
2262 long off;
2263 u8 *data;
2264 int i;
2265
2266 BUG_ON(nhead < 0);
2267
2268 BUG_ON(skb_shared(skb));
2269
2270 skb_zcopy_downgrade_managed(skb);
2271
2272 if (skb_pfmemalloc(skb))
2273 gfp_mask |= __GFP_MEMALLOC;
2274
2275 data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
2276 if (!data)
2277 goto nodata;
2278 size = SKB_WITH_OVERHEAD(size);
2279
2280 /* Copy only real data... and, alas, header. This should be
2281 * optimized for the cases when header is void.
2282 */
2283 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
2284
2285 memcpy((struct skb_shared_info *)(data + size),
2286 skb_shinfo(skb),
2287 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
2288
2289 /*
2290 * if shinfo is shared we must drop the old head gracefully, but if it
2291 * is not we can just drop the old head and let the existing refcount
2292 * be since all we did is relocate the values
2293 */
2294 if (skb_cloned(skb)) {
2295 if (skb_orphan_frags(skb, gfp_mask))
2296 goto nofrags;
2297 if (skb_zcopy(skb))
2298 refcount_inc(&skb_uarg(skb)->refcnt);
2299 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2300 skb_frag_ref(skb, i);
2301
2302 if (skb_has_frag_list(skb))
2303 skb_clone_fraglist(skb);
2304
2305 skb_release_data(skb, SKB_CONSUMED);
2306 } else {
2307 skb_free_head(skb);
2308 }
2309 off = (data + nhead) - skb->head;
2310
2311 skb->head = data;
2312 skb->head_frag = 0;
2313 skb->data += off;
2314
2315 skb_set_end_offset(skb, size);
2316 #ifdef NET_SKBUFF_DATA_USES_OFFSET
2317 off = nhead;
2318 #endif
2319 skb->tail += off;
2320 skb_headers_offset_update(skb, nhead);
2321 skb->cloned = 0;
2322 skb->hdr_len = 0;
2323 skb->nohdr = 0;
2324 atomic_set(&skb_shinfo(skb)->dataref, 1);
2325
2326 skb_metadata_clear(skb);
2327
2328 /* It is not generally safe to change skb->truesize.
2329 * For the moment, we really care of rx path, or
2330 * when skb is orphaned (not attached to a socket).
2331 */
2332 if (!skb->sk || skb->destructor == sock_edemux)
2333 skb->truesize += size - osize;
2334
2335 return 0;
2336
2337 nofrags:
2338 skb_kfree_head(data, size);
2339 nodata:
2340 return -ENOMEM;
2341 }
2342 EXPORT_SYMBOL(pskb_expand_head);
2343
2344 /* Make private copy of skb with writable head and some headroom */
2345
skb_realloc_headroom(struct sk_buff * skb,unsigned int headroom)2346 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
2347 {
2348 struct sk_buff *skb2;
2349 int delta = headroom - skb_headroom(skb);
2350
2351 if (delta <= 0)
2352 skb2 = pskb_copy(skb, GFP_ATOMIC);
2353 else {
2354 skb2 = skb_clone(skb, GFP_ATOMIC);
2355 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
2356 GFP_ATOMIC)) {
2357 kfree_skb(skb2);
2358 skb2 = NULL;
2359 }
2360 }
2361 return skb2;
2362 }
2363 EXPORT_SYMBOL(skb_realloc_headroom);
2364
2365 /* Note: We plan to rework this in linux-6.4 */
__skb_unclone_keeptruesize(struct sk_buff * skb,gfp_t pri)2366 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri)
2367 {
2368 unsigned int saved_end_offset, saved_truesize;
2369 struct skb_shared_info *shinfo;
2370 int res;
2371
2372 saved_end_offset = skb_end_offset(skb);
2373 saved_truesize = skb->truesize;
2374
2375 res = pskb_expand_head(skb, 0, 0, pri);
2376 if (res)
2377 return res;
2378
2379 skb->truesize = saved_truesize;
2380
2381 if (likely(skb_end_offset(skb) == saved_end_offset))
2382 return 0;
2383
2384 /* We can not change skb->end if the original or new value
2385 * is SKB_SMALL_HEAD_HEADROOM, as it might break skb_kfree_head().
2386 */
2387 if (saved_end_offset == SKB_SMALL_HEAD_HEADROOM ||
2388 skb_end_offset(skb) == SKB_SMALL_HEAD_HEADROOM) {
2389 /* We think this path should not be taken.
2390 * Add a temporary trace to warn us just in case.
2391 */
2392 pr_err_once("__skb_unclone_keeptruesize() skb_end_offset() %u -> %u\n",
2393 saved_end_offset, skb_end_offset(skb));
2394 WARN_ON_ONCE(1);
2395 return 0;
2396 }
2397
2398 shinfo = skb_shinfo(skb);
2399
2400 /* We are about to change back skb->end,
2401 * we need to move skb_shinfo() to its new location.
2402 */
2403 memmove(skb->head + saved_end_offset,
2404 shinfo,
2405 offsetof(struct skb_shared_info, frags[shinfo->nr_frags]));
2406
2407 skb_set_end_offset(skb, saved_end_offset);
2408
2409 return 0;
2410 }
2411
2412 /**
2413 * skb_expand_head - reallocate header of &sk_buff
2414 * @skb: buffer to reallocate
2415 * @headroom: needed headroom
2416 *
2417 * Unlike skb_realloc_headroom, this one does not allocate a new skb
2418 * if possible; copies skb->sk to new skb as needed
2419 * and frees original skb in case of failures.
2420 *
2421 * It expect increased headroom and generates warning otherwise.
2422 */
2423
skb_expand_head(struct sk_buff * skb,unsigned int headroom)2424 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom)
2425 {
2426 int delta = headroom - skb_headroom(skb);
2427 int osize = skb_end_offset(skb);
2428 struct sock *sk = skb->sk;
2429
2430 if (WARN_ONCE(delta <= 0,
2431 "%s is expecting an increase in the headroom", __func__))
2432 return skb;
2433
2434 delta = SKB_DATA_ALIGN(delta);
2435 /* pskb_expand_head() might crash, if skb is shared. */
2436 if (skb_shared(skb) || !is_skb_wmem(skb)) {
2437 struct sk_buff *nskb = skb_clone(skb, GFP_ATOMIC);
2438
2439 if (unlikely(!nskb))
2440 goto fail;
2441
2442 if (sk)
2443 skb_set_owner_w(nskb, sk);
2444 consume_skb(skb);
2445 skb = nskb;
2446 }
2447 if (pskb_expand_head(skb, delta, 0, GFP_ATOMIC))
2448 goto fail;
2449
2450 if (sk && is_skb_wmem(skb)) {
2451 delta = skb_end_offset(skb) - osize;
2452 refcount_add(delta, &sk->sk_wmem_alloc);
2453 skb->truesize += delta;
2454 }
2455 return skb;
2456
2457 fail:
2458 kfree_skb(skb);
2459 return NULL;
2460 }
2461 EXPORT_SYMBOL(skb_expand_head);
2462
2463 /**
2464 * skb_copy_expand - copy and expand sk_buff
2465 * @skb: buffer to copy
2466 * @newheadroom: new free bytes at head
2467 * @newtailroom: new free bytes at tail
2468 * @gfp_mask: allocation priority
2469 *
2470 * Make a copy of both an &sk_buff and its data and while doing so
2471 * allocate additional space.
2472 *
2473 * This is used when the caller wishes to modify the data and needs a
2474 * private copy of the data to alter as well as more space for new fields.
2475 * Returns %NULL on failure or the pointer to the buffer
2476 * on success. The returned buffer has a reference count of 1.
2477 *
2478 * You must pass %GFP_ATOMIC as the allocation priority if this function
2479 * is called from an interrupt.
2480 */
skb_copy_expand(const struct sk_buff * skb,int newheadroom,int newtailroom,gfp_t gfp_mask)2481 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
2482 int newheadroom, int newtailroom,
2483 gfp_t gfp_mask)
2484 {
2485 /*
2486 * Allocate the copy buffer
2487 */
2488 int head_copy_len, head_copy_off;
2489 struct sk_buff *n;
2490 int oldheadroom;
2491
2492 if (!skb_frags_readable(skb))
2493 return NULL;
2494
2495 if (WARN_ON_ONCE(skb_shinfo(skb)->gso_type & SKB_GSO_FRAGLIST))
2496 return NULL;
2497
2498 oldheadroom = skb_headroom(skb);
2499 n = __alloc_skb(newheadroom + skb->len + newtailroom,
2500 gfp_mask, skb_alloc_rx_flag(skb),
2501 NUMA_NO_NODE);
2502 if (!n)
2503 return NULL;
2504
2505 skb_reserve(n, newheadroom);
2506
2507 /* Set the tail pointer and length */
2508 skb_put(n, skb->len);
2509
2510 head_copy_len = oldheadroom;
2511 head_copy_off = 0;
2512 if (newheadroom <= head_copy_len)
2513 head_copy_len = newheadroom;
2514 else
2515 head_copy_off = newheadroom - head_copy_len;
2516
2517 /* Copy the linear header and data. */
2518 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
2519 skb->len + head_copy_len));
2520
2521 skb_copy_header(n, skb);
2522
2523 skb_headers_offset_update(n, newheadroom - oldheadroom);
2524
2525 return n;
2526 }
2527 EXPORT_SYMBOL(skb_copy_expand);
2528
2529 /**
2530 * __skb_pad - zero pad the tail of an skb
2531 * @skb: buffer to pad
2532 * @pad: space to pad
2533 * @free_on_error: free buffer on error
2534 *
2535 * Ensure that a buffer is followed by a padding area that is zero
2536 * filled. Used by network drivers which may DMA or transfer data
2537 * beyond the buffer end onto the wire.
2538 *
2539 * May return error in out of memory cases. The skb is freed on error
2540 * if @free_on_error is true.
2541 */
2542
__skb_pad(struct sk_buff * skb,int pad,bool free_on_error)2543 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
2544 {
2545 int err;
2546 int ntail;
2547
2548 /* If the skbuff is non linear tailroom is always zero.. */
2549 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
2550 memset(skb->data+skb->len, 0, pad);
2551 return 0;
2552 }
2553
2554 ntail = skb->data_len + pad - (skb->end - skb->tail);
2555 if (likely(skb_cloned(skb) || ntail > 0)) {
2556 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
2557 if (unlikely(err))
2558 goto free_skb;
2559 }
2560
2561 /* FIXME: The use of this function with non-linear skb's really needs
2562 * to be audited.
2563 */
2564 err = skb_linearize(skb);
2565 if (unlikely(err))
2566 goto free_skb;
2567
2568 memset(skb->data + skb->len, 0, pad);
2569 return 0;
2570
2571 free_skb:
2572 if (free_on_error)
2573 kfree_skb(skb);
2574 return err;
2575 }
2576 EXPORT_SYMBOL(__skb_pad);
2577
2578 /**
2579 * pskb_put - add data to the tail of a potentially fragmented buffer
2580 * @skb: start of the buffer to use
2581 * @tail: tail fragment of the buffer to use
2582 * @len: amount of data to add
2583 *
2584 * This function extends the used data area of the potentially
2585 * fragmented buffer. @tail must be the last fragment of @skb -- or
2586 * @skb itself. If this would exceed the total buffer size the kernel
2587 * will panic. A pointer to the first byte of the extra data is
2588 * returned.
2589 */
2590
pskb_put(struct sk_buff * skb,struct sk_buff * tail,int len)2591 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
2592 {
2593 if (tail != skb) {
2594 skb->data_len += len;
2595 skb->len += len;
2596 }
2597 return skb_put(tail, len);
2598 }
2599 EXPORT_SYMBOL_GPL(pskb_put);
2600
2601 /**
2602 * skb_put - add data to a buffer
2603 * @skb: buffer to use
2604 * @len: amount of data to add
2605 *
2606 * This function extends the used data area of the buffer. If this would
2607 * exceed the total buffer size the kernel will panic. A pointer to the
2608 * first byte of the extra data is returned.
2609 */
skb_put(struct sk_buff * skb,unsigned int len)2610 void *skb_put(struct sk_buff *skb, unsigned int len)
2611 {
2612 void *tmp = skb_tail_pointer(skb);
2613 SKB_LINEAR_ASSERT(skb);
2614 skb->tail += len;
2615 skb->len += len;
2616 if (unlikely(skb->tail > skb->end))
2617 skb_over_panic(skb, len, __builtin_return_address(0));
2618 return tmp;
2619 }
2620 EXPORT_SYMBOL(skb_put);
2621
2622 /**
2623 * skb_push - add data to the start of a buffer
2624 * @skb: buffer to use
2625 * @len: amount of data to add
2626 *
2627 * This function extends the used data area of the buffer at the buffer
2628 * start. If this would exceed the total buffer headroom the kernel will
2629 * panic. A pointer to the first byte of the extra data is returned.
2630 */
skb_push(struct sk_buff * skb,unsigned int len)2631 void *skb_push(struct sk_buff *skb, unsigned int len)
2632 {
2633 skb->data -= len;
2634 skb->len += len;
2635 if (unlikely(skb->data < skb->head))
2636 skb_under_panic(skb, len, __builtin_return_address(0));
2637 return skb->data;
2638 }
2639 EXPORT_SYMBOL(skb_push);
2640
2641 /**
2642 * skb_pull - remove data from the start of a buffer
2643 * @skb: buffer to use
2644 * @len: amount of data to remove
2645 *
2646 * This function removes data from the start of a buffer, returning
2647 * the memory to the headroom. A pointer to the next data in the buffer
2648 * is returned. Once the data has been pulled future pushes will overwrite
2649 * the old data.
2650 */
skb_pull(struct sk_buff * skb,unsigned int len)2651 void *skb_pull(struct sk_buff *skb, unsigned int len)
2652 {
2653 return skb_pull_inline(skb, len);
2654 }
2655 EXPORT_SYMBOL(skb_pull);
2656
2657 /**
2658 * skb_pull_data - remove data from the start of a buffer returning its
2659 * original position.
2660 * @skb: buffer to use
2661 * @len: amount of data to remove
2662 *
2663 * This function removes data from the start of a buffer, returning
2664 * the memory to the headroom. A pointer to the original data in the buffer
2665 * is returned after checking if there is enough data to pull. Once the
2666 * data has been pulled future pushes will overwrite the old data.
2667 */
skb_pull_data(struct sk_buff * skb,size_t len)2668 void *skb_pull_data(struct sk_buff *skb, size_t len)
2669 {
2670 void *data = skb->data;
2671
2672 if (skb->len < len)
2673 return NULL;
2674
2675 skb_pull(skb, len);
2676
2677 return data;
2678 }
2679 EXPORT_SYMBOL(skb_pull_data);
2680
2681 /**
2682 * skb_trim - remove end from a buffer
2683 * @skb: buffer to alter
2684 * @len: new length
2685 *
2686 * Cut the length of a buffer down by removing data from the tail. If
2687 * the buffer is already under the length specified it is not modified.
2688 * The skb must be linear.
2689 */
skb_trim(struct sk_buff * skb,unsigned int len)2690 void skb_trim(struct sk_buff *skb, unsigned int len)
2691 {
2692 if (skb->len > len)
2693 __skb_trim(skb, len);
2694 }
2695 EXPORT_SYMBOL(skb_trim);
2696
2697 /* Trims skb to length len. It can change skb pointers.
2698 */
2699
___pskb_trim(struct sk_buff * skb,unsigned int len)2700 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
2701 {
2702 struct sk_buff **fragp;
2703 struct sk_buff *frag;
2704 int offset = skb_headlen(skb);
2705 int nfrags = skb_shinfo(skb)->nr_frags;
2706 int i;
2707 int err;
2708
2709 if (skb_cloned(skb) &&
2710 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
2711 return err;
2712
2713 i = 0;
2714 if (offset >= len)
2715 goto drop_pages;
2716
2717 for (; i < nfrags; i++) {
2718 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2719
2720 if (end < len) {
2721 offset = end;
2722 continue;
2723 }
2724
2725 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
2726
2727 drop_pages:
2728 skb_shinfo(skb)->nr_frags = i;
2729
2730 for (; i < nfrags; i++)
2731 skb_frag_unref(skb, i);
2732
2733 if (skb_has_frag_list(skb))
2734 skb_drop_fraglist(skb);
2735 goto done;
2736 }
2737
2738 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
2739 fragp = &frag->next) {
2740 int end = offset + frag->len;
2741
2742 if (skb_shared(frag)) {
2743 struct sk_buff *nfrag;
2744
2745 nfrag = skb_clone(frag, GFP_ATOMIC);
2746 if (unlikely(!nfrag))
2747 return -ENOMEM;
2748
2749 nfrag->next = frag->next;
2750 consume_skb(frag);
2751 frag = nfrag;
2752 *fragp = frag;
2753 }
2754
2755 if (end < len) {
2756 offset = end;
2757 continue;
2758 }
2759
2760 if (end > len &&
2761 unlikely((err = pskb_trim(frag, len - offset))))
2762 return err;
2763
2764 if (frag->next)
2765 skb_drop_list(&frag->next);
2766 break;
2767 }
2768
2769 done:
2770 if (len > skb_headlen(skb)) {
2771 skb->data_len -= skb->len - len;
2772 skb->len = len;
2773 } else {
2774 skb->len = len;
2775 skb->data_len = 0;
2776 skb_set_tail_pointer(skb, len);
2777 }
2778
2779 if (!skb->sk || skb->destructor == sock_edemux)
2780 skb_condense(skb);
2781 return 0;
2782 }
2783 EXPORT_SYMBOL(___pskb_trim);
2784
2785 /* Note : use pskb_trim_rcsum() instead of calling this directly
2786 */
pskb_trim_rcsum_slow(struct sk_buff * skb,unsigned int len)2787 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2788 {
2789 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2790 int delta = skb->len - len;
2791
2792 skb->csum = csum_block_sub(skb->csum,
2793 skb_checksum(skb, len, delta, 0),
2794 len);
2795 } else if (skb->ip_summed == CHECKSUM_PARTIAL) {
2796 int hdlen = (len > skb_headlen(skb)) ? skb_headlen(skb) : len;
2797 int offset = skb_checksum_start_offset(skb) + skb->csum_offset;
2798
2799 if (offset + sizeof(__sum16) > hdlen)
2800 return -EINVAL;
2801 }
2802 return __pskb_trim(skb, len);
2803 }
2804 EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2805
2806 /**
2807 * __pskb_pull_tail - advance tail of skb header
2808 * @skb: buffer to reallocate
2809 * @delta: number of bytes to advance tail
2810 *
2811 * The function makes a sense only on a fragmented &sk_buff,
2812 * it expands header moving its tail forward and copying necessary
2813 * data from fragmented part.
2814 *
2815 * &sk_buff MUST have reference count of 1.
2816 *
2817 * Returns %NULL (and &sk_buff does not change) if pull failed
2818 * or value of new tail of skb in the case of success.
2819 *
2820 * All the pointers pointing into skb header may change and must be
2821 * reloaded after call to this function.
2822 */
2823
2824 /* Moves tail of skb head forward, copying data from fragmented part,
2825 * when it is necessary.
2826 * 1. It may fail due to malloc failure.
2827 * 2. It may change skb pointers.
2828 *
2829 * It is pretty complicated. Luckily, it is called only in exceptional cases.
2830 */
__pskb_pull_tail(struct sk_buff * skb,int delta)2831 void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2832 {
2833 /* If skb has not enough free space at tail, get new one
2834 * plus 128 bytes for future expansions. If we have enough
2835 * room at tail, reallocate without expansion only if skb is cloned.
2836 */
2837 int i, k, eat = (skb->tail + delta) - skb->end;
2838
2839 if (!skb_frags_readable(skb))
2840 return NULL;
2841
2842 if (eat > 0 || skb_cloned(skb)) {
2843 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2844 GFP_ATOMIC))
2845 return NULL;
2846 }
2847
2848 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2849 skb_tail_pointer(skb), delta));
2850
2851 /* Optimization: no fragments, no reasons to preestimate
2852 * size of pulled pages. Superb.
2853 */
2854 if (!skb_has_frag_list(skb))
2855 goto pull_pages;
2856
2857 /* Estimate size of pulled pages. */
2858 eat = delta;
2859 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2860 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2861
2862 if (size >= eat)
2863 goto pull_pages;
2864 eat -= size;
2865 }
2866
2867 /* If we need update frag list, we are in troubles.
2868 * Certainly, it is possible to add an offset to skb data,
2869 * but taking into account that pulling is expected to
2870 * be very rare operation, it is worth to fight against
2871 * further bloating skb head and crucify ourselves here instead.
2872 * Pure masohism, indeed. 8)8)
2873 */
2874 if (eat) {
2875 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2876 struct sk_buff *clone = NULL;
2877 struct sk_buff *insp = NULL;
2878
2879 do {
2880 if (list->len <= eat) {
2881 /* Eaten as whole. */
2882 eat -= list->len;
2883 list = list->next;
2884 insp = list;
2885 } else {
2886 /* Eaten partially. */
2887 if (skb_is_gso(skb) && !list->head_frag &&
2888 skb_headlen(list))
2889 skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
2890
2891 if (skb_shared(list)) {
2892 /* Sucks! We need to fork list. :-( */
2893 clone = skb_clone(list, GFP_ATOMIC);
2894 if (!clone)
2895 return NULL;
2896 insp = list->next;
2897 list = clone;
2898 } else {
2899 /* This may be pulled without
2900 * problems. */
2901 insp = list;
2902 }
2903 if (!pskb_pull(list, eat)) {
2904 kfree_skb(clone);
2905 return NULL;
2906 }
2907 break;
2908 }
2909 } while (eat);
2910
2911 /* Free pulled out fragments. */
2912 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2913 skb_shinfo(skb)->frag_list = list->next;
2914 consume_skb(list);
2915 }
2916 /* And insert new clone at head. */
2917 if (clone) {
2918 clone->next = list;
2919 skb_shinfo(skb)->frag_list = clone;
2920 }
2921 }
2922 /* Success! Now we may commit changes to skb data. */
2923
2924 pull_pages:
2925 eat = delta;
2926 k = 0;
2927 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2928 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2929
2930 if (size <= eat) {
2931 skb_frag_unref(skb, i);
2932 eat -= size;
2933 } else {
2934 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2935
2936 *frag = skb_shinfo(skb)->frags[i];
2937 if (eat) {
2938 skb_frag_off_add(frag, eat);
2939 skb_frag_size_sub(frag, eat);
2940 if (!i)
2941 goto end;
2942 eat = 0;
2943 }
2944 k++;
2945 }
2946 }
2947 skb_shinfo(skb)->nr_frags = k;
2948
2949 end:
2950 skb->tail += delta;
2951 skb->data_len -= delta;
2952
2953 if (!skb->data_len)
2954 skb_zcopy_clear(skb, false);
2955
2956 return skb_tail_pointer(skb);
2957 }
2958 EXPORT_SYMBOL(__pskb_pull_tail);
2959
2960 /**
2961 * skb_copy_bits - copy bits from skb to kernel buffer
2962 * @skb: source skb
2963 * @offset: offset in source
2964 * @to: destination buffer
2965 * @len: number of bytes to copy
2966 *
2967 * Copy the specified number of bytes from the source skb to the
2968 * destination buffer.
2969 *
2970 * CAUTION ! :
2971 * If its prototype is ever changed,
2972 * check arch/{*}/net/{*}.S files,
2973 * since it is called from BPF assembly code.
2974 */
skb_copy_bits(const struct sk_buff * skb,int offset,void * to,int len)2975 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2976 {
2977 int start = skb_headlen(skb);
2978 struct sk_buff *frag_iter;
2979 int i, copy;
2980
2981 if (offset > (int)skb->len - len)
2982 goto fault;
2983
2984 /* Copy header. */
2985 if ((copy = start - offset) > 0) {
2986 if (copy > len)
2987 copy = len;
2988 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2989 if ((len -= copy) == 0)
2990 return 0;
2991 offset += copy;
2992 to += copy;
2993 }
2994
2995 if (!skb_frags_readable(skb))
2996 goto fault;
2997
2998 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2999 int end;
3000 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
3001
3002 WARN_ON(start > offset + len);
3003
3004 end = start + skb_frag_size(f);
3005 if ((copy = end - offset) > 0) {
3006 u32 p_off, p_len, copied;
3007 struct page *p;
3008 u8 *vaddr;
3009
3010 if (copy > len)
3011 copy = len;
3012
3013 skb_frag_foreach_page(f,
3014 skb_frag_off(f) + offset - start,
3015 copy, p, p_off, p_len, copied) {
3016 vaddr = kmap_atomic(p);
3017 memcpy(to + copied, vaddr + p_off, p_len);
3018 kunmap_atomic(vaddr);
3019 }
3020
3021 if ((len -= copy) == 0)
3022 return 0;
3023 offset += copy;
3024 to += copy;
3025 }
3026 start = end;
3027 }
3028
3029 skb_walk_frags(skb, frag_iter) {
3030 int end;
3031
3032 WARN_ON(start > offset + len);
3033
3034 end = start + frag_iter->len;
3035 if ((copy = end - offset) > 0) {
3036 if (copy > len)
3037 copy = len;
3038 if (skb_copy_bits(frag_iter, offset - start, to, copy))
3039 goto fault;
3040 if ((len -= copy) == 0)
3041 return 0;
3042 offset += copy;
3043 to += copy;
3044 }
3045 start = end;
3046 }
3047
3048 if (!len)
3049 return 0;
3050
3051 fault:
3052 return -EFAULT;
3053 }
3054 EXPORT_SYMBOL(skb_copy_bits);
3055
3056 /*
3057 * Callback from splice_to_pipe(), if we need to release some pages
3058 * at the end of the spd in case we error'ed out in filling the pipe.
3059 */
sock_spd_release(struct splice_pipe_desc * spd,unsigned int i)3060 static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3061 {
3062 put_page(spd->pages[i]);
3063 }
3064
linear_to_page(struct page * page,unsigned int * len,unsigned int * offset,struct sock * sk)3065 static struct page *linear_to_page(struct page *page, unsigned int *len,
3066 unsigned int *offset,
3067 struct sock *sk)
3068 {
3069 struct page_frag *pfrag = sk_page_frag(sk);
3070
3071 if (!sk_page_frag_refill(sk, pfrag))
3072 return NULL;
3073
3074 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
3075
3076 memcpy(page_address(pfrag->page) + pfrag->offset,
3077 page_address(page) + *offset, *len);
3078 *offset = pfrag->offset;
3079 pfrag->offset += *len;
3080
3081 return pfrag->page;
3082 }
3083
spd_can_coalesce(const struct splice_pipe_desc * spd,struct page * page,unsigned int offset)3084 static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
3085 struct page *page,
3086 unsigned int offset)
3087 {
3088 return spd->nr_pages &&
3089 spd->pages[spd->nr_pages - 1] == page &&
3090 (spd->partial[spd->nr_pages - 1].offset +
3091 spd->partial[spd->nr_pages - 1].len == offset);
3092 }
3093
3094 /*
3095 * Fill page/offset/length into spd, if it can hold more pages.
3096 */
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)3097 static bool spd_fill_page(struct splice_pipe_desc *spd,
3098 struct pipe_inode_info *pipe, struct page *page,
3099 unsigned int *len, unsigned int offset,
3100 bool linear,
3101 struct sock *sk)
3102 {
3103 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
3104 return true;
3105
3106 if (linear) {
3107 page = linear_to_page(page, len, &offset, sk);
3108 if (!page)
3109 return true;
3110 }
3111 if (spd_can_coalesce(spd, page, offset)) {
3112 spd->partial[spd->nr_pages - 1].len += *len;
3113 return false;
3114 }
3115 get_page(page);
3116 spd->pages[spd->nr_pages] = page;
3117 spd->partial[spd->nr_pages].len = *len;
3118 spd->partial[spd->nr_pages].offset = offset;
3119 spd->nr_pages++;
3120
3121 return false;
3122 }
3123
__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)3124 static bool __splice_segment(struct page *page, unsigned int poff,
3125 unsigned int plen, unsigned int *off,
3126 unsigned int *len,
3127 struct splice_pipe_desc *spd, bool linear,
3128 struct sock *sk,
3129 struct pipe_inode_info *pipe)
3130 {
3131 if (!*len)
3132 return true;
3133
3134 /* skip this segment if already processed */
3135 if (*off >= plen) {
3136 *off -= plen;
3137 return false;
3138 }
3139
3140 /* ignore any bits we already processed */
3141 poff += *off;
3142 plen -= *off;
3143 *off = 0;
3144
3145 do {
3146 unsigned int flen = min(*len, plen);
3147
3148 if (spd_fill_page(spd, pipe, page, &flen, poff,
3149 linear, sk))
3150 return true;
3151 poff += flen;
3152 plen -= flen;
3153 *len -= flen;
3154 } while (*len && plen);
3155
3156 return false;
3157 }
3158
3159 /*
3160 * Map linear and fragment data from the skb to spd. It reports true if the
3161 * pipe is full or if we already spliced the requested length.
3162 */
__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)3163 static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
3164 unsigned int *offset, unsigned int *len,
3165 struct splice_pipe_desc *spd, struct sock *sk)
3166 {
3167 int seg;
3168 struct sk_buff *iter;
3169
3170 /* map the linear part :
3171 * If skb->head_frag is set, this 'linear' part is backed by a
3172 * fragment, and if the head is not shared with any clones then
3173 * we can avoid a copy since we own the head portion of this page.
3174 */
3175 if (__splice_segment(virt_to_page(skb->data),
3176 (unsigned long) skb->data & (PAGE_SIZE - 1),
3177 skb_headlen(skb),
3178 offset, len, spd,
3179 skb_head_is_locked(skb),
3180 sk, pipe))
3181 return true;
3182
3183 /*
3184 * then map the fragments
3185 */
3186 if (!skb_frags_readable(skb))
3187 return false;
3188
3189 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
3190 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
3191
3192 if (WARN_ON_ONCE(!skb_frag_page(f)))
3193 return false;
3194
3195 if (__splice_segment(skb_frag_page(f),
3196 skb_frag_off(f), skb_frag_size(f),
3197 offset, len, spd, false, sk, pipe))
3198 return true;
3199 }
3200
3201 skb_walk_frags(skb, iter) {
3202 if (*offset >= iter->len) {
3203 *offset -= iter->len;
3204 continue;
3205 }
3206 /* __skb_splice_bits() only fails if the output has no room
3207 * left, so no point in going over the frag_list for the error
3208 * case.
3209 */
3210 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
3211 return true;
3212 }
3213
3214 return false;
3215 }
3216
3217 /*
3218 * Map data from the skb to a pipe. Should handle both the linear part,
3219 * the fragments, and the frag list.
3220 */
skb_splice_bits(struct sk_buff * skb,struct sock * sk,unsigned int offset,struct pipe_inode_info * pipe,unsigned int tlen,unsigned int flags)3221 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3222 struct pipe_inode_info *pipe, unsigned int tlen,
3223 unsigned int flags)
3224 {
3225 struct partial_page partial[MAX_SKB_FRAGS];
3226 struct page *pages[MAX_SKB_FRAGS];
3227 struct splice_pipe_desc spd = {
3228 .pages = pages,
3229 .partial = partial,
3230 .nr_pages_max = MAX_SKB_FRAGS,
3231 .ops = &nosteal_pipe_buf_ops,
3232 .spd_release = sock_spd_release,
3233 };
3234 int ret = 0;
3235
3236 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
3237
3238 if (spd.nr_pages)
3239 ret = splice_to_pipe(pipe, &spd);
3240
3241 return ret;
3242 }
3243 EXPORT_SYMBOL_GPL(skb_splice_bits);
3244
sendmsg_locked(struct sock * sk,struct msghdr * msg)3245 static int sendmsg_locked(struct sock *sk, struct msghdr *msg)
3246 {
3247 struct socket *sock = sk->sk_socket;
3248 size_t size = msg_data_left(msg);
3249
3250 if (!sock)
3251 return -EINVAL;
3252
3253 if (!sock->ops->sendmsg_locked)
3254 return sock_no_sendmsg_locked(sk, msg, size);
3255
3256 return sock->ops->sendmsg_locked(sk, msg, size);
3257 }
3258
sendmsg_unlocked(struct sock * sk,struct msghdr * msg)3259 static int sendmsg_unlocked(struct sock *sk, struct msghdr *msg)
3260 {
3261 struct socket *sock = sk->sk_socket;
3262
3263 if (!sock)
3264 return -EINVAL;
3265 return sock_sendmsg(sock, msg);
3266 }
3267
3268 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)3269 static int __skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset,
3270 int len, sendmsg_func sendmsg)
3271 {
3272 unsigned int orig_len = len;
3273 struct sk_buff *head = skb;
3274 unsigned short fragidx;
3275 int slen, ret;
3276
3277 do_frag_list:
3278
3279 /* Deal with head data */
3280 while (offset < skb_headlen(skb) && len) {
3281 struct kvec kv;
3282 struct msghdr msg;
3283
3284 slen = min_t(int, len, skb_headlen(skb) - offset);
3285 kv.iov_base = skb->data + offset;
3286 kv.iov_len = slen;
3287 memset(&msg, 0, sizeof(msg));
3288 msg.msg_flags = MSG_DONTWAIT;
3289
3290 iov_iter_kvec(&msg.msg_iter, ITER_SOURCE, &kv, 1, slen);
3291 ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3292 sendmsg_unlocked, sk, &msg);
3293 if (ret <= 0)
3294 goto error;
3295
3296 offset += ret;
3297 len -= ret;
3298 }
3299
3300 /* All the data was skb head? */
3301 if (!len)
3302 goto out;
3303
3304 /* Make offset relative to start of frags */
3305 offset -= skb_headlen(skb);
3306
3307 /* Find where we are in frag list */
3308 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3309 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
3310
3311 if (offset < skb_frag_size(frag))
3312 break;
3313
3314 offset -= skb_frag_size(frag);
3315 }
3316
3317 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
3318 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
3319
3320 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
3321
3322 while (slen) {
3323 struct bio_vec bvec;
3324 struct msghdr msg = {
3325 .msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT,
3326 };
3327
3328 bvec_set_page(&bvec, skb_frag_page(frag), slen,
3329 skb_frag_off(frag) + offset);
3330 iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1,
3331 slen);
3332
3333 ret = INDIRECT_CALL_2(sendmsg, sendmsg_locked,
3334 sendmsg_unlocked, sk, &msg);
3335 if (ret <= 0)
3336 goto error;
3337
3338 len -= ret;
3339 offset += ret;
3340 slen -= ret;
3341 }
3342
3343 offset = 0;
3344 }
3345
3346 if (len) {
3347 /* Process any frag lists */
3348
3349 if (skb == head) {
3350 if (skb_has_frag_list(skb)) {
3351 skb = skb_shinfo(skb)->frag_list;
3352 goto do_frag_list;
3353 }
3354 } else if (skb->next) {
3355 skb = skb->next;
3356 goto do_frag_list;
3357 }
3358 }
3359
3360 out:
3361 return orig_len - len;
3362
3363 error:
3364 return orig_len == len ? ret : orig_len - len;
3365 }
3366
3367 /* 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)3368 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3369 int len)
3370 {
3371 return __skb_send_sock(sk, skb, offset, len, sendmsg_locked);
3372 }
3373 EXPORT_SYMBOL_GPL(skb_send_sock_locked);
3374
3375 /* Send skb data on a socket. Socket must be unlocked. */
skb_send_sock(struct sock * sk,struct sk_buff * skb,int offset,int len)3376 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len)
3377 {
3378 return __skb_send_sock(sk, skb, offset, len, sendmsg_unlocked);
3379 }
3380
3381 /**
3382 * skb_store_bits - store bits from kernel buffer to skb
3383 * @skb: destination buffer
3384 * @offset: offset in destination
3385 * @from: source buffer
3386 * @len: number of bytes to copy
3387 *
3388 * Copy the specified number of bytes from the source buffer to the
3389 * destination skb. This function handles all the messy bits of
3390 * traversing fragment lists and such.
3391 */
3392
skb_store_bits(struct sk_buff * skb,int offset,const void * from,int len)3393 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
3394 {
3395 int start = skb_headlen(skb);
3396 struct sk_buff *frag_iter;
3397 int i, copy;
3398
3399 if (offset > (int)skb->len - len)
3400 goto fault;
3401
3402 if ((copy = start - offset) > 0) {
3403 if (copy > len)
3404 copy = len;
3405 skb_copy_to_linear_data_offset(skb, offset, from, copy);
3406 if ((len -= copy) == 0)
3407 return 0;
3408 offset += copy;
3409 from += copy;
3410 }
3411
3412 if (!skb_frags_readable(skb))
3413 goto fault;
3414
3415 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3416 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3417 int end;
3418
3419 WARN_ON(start > offset + len);
3420
3421 end = start + skb_frag_size(frag);
3422 if ((copy = end - offset) > 0) {
3423 u32 p_off, p_len, copied;
3424 struct page *p;
3425 u8 *vaddr;
3426
3427 if (copy > len)
3428 copy = len;
3429
3430 skb_frag_foreach_page(frag,
3431 skb_frag_off(frag) + offset - start,
3432 copy, p, p_off, p_len, copied) {
3433 vaddr = kmap_atomic(p);
3434 memcpy(vaddr + p_off, from + copied, p_len);
3435 kunmap_atomic(vaddr);
3436 }
3437
3438 if ((len -= copy) == 0)
3439 return 0;
3440 offset += copy;
3441 from += copy;
3442 }
3443 start = end;
3444 }
3445
3446 skb_walk_frags(skb, frag_iter) {
3447 int end;
3448
3449 WARN_ON(start > offset + len);
3450
3451 end = start + frag_iter->len;
3452 if ((copy = end - offset) > 0) {
3453 if (copy > len)
3454 copy = len;
3455 if (skb_store_bits(frag_iter, offset - start,
3456 from, copy))
3457 goto fault;
3458 if ((len -= copy) == 0)
3459 return 0;
3460 offset += copy;
3461 from += copy;
3462 }
3463 start = end;
3464 }
3465 if (!len)
3466 return 0;
3467
3468 fault:
3469 return -EFAULT;
3470 }
3471 EXPORT_SYMBOL(skb_store_bits);
3472
3473 /* Checksum skb data. */
__skb_checksum(const struct sk_buff * skb,int offset,int len,__wsum csum,const struct skb_checksum_ops * ops)3474 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3475 __wsum csum, const struct skb_checksum_ops *ops)
3476 {
3477 int start = skb_headlen(skb);
3478 int i, copy = start - offset;
3479 struct sk_buff *frag_iter;
3480 int pos = 0;
3481
3482 /* Checksum header. */
3483 if (copy > 0) {
3484 if (copy > len)
3485 copy = len;
3486 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
3487 skb->data + offset, copy, csum);
3488 if ((len -= copy) == 0)
3489 return csum;
3490 offset += copy;
3491 pos = copy;
3492 }
3493
3494 if (WARN_ON_ONCE(!skb_frags_readable(skb)))
3495 return 0;
3496
3497 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3498 int end;
3499 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3500
3501 WARN_ON(start > offset + len);
3502
3503 end = start + skb_frag_size(frag);
3504 if ((copy = end - offset) > 0) {
3505 u32 p_off, p_len, copied;
3506 struct page *p;
3507 __wsum csum2;
3508 u8 *vaddr;
3509
3510 if (copy > len)
3511 copy = len;
3512
3513 skb_frag_foreach_page(frag,
3514 skb_frag_off(frag) + offset - start,
3515 copy, p, p_off, p_len, copied) {
3516 vaddr = kmap_atomic(p);
3517 csum2 = INDIRECT_CALL_1(ops->update,
3518 csum_partial_ext,
3519 vaddr + p_off, p_len, 0);
3520 kunmap_atomic(vaddr);
3521 csum = INDIRECT_CALL_1(ops->combine,
3522 csum_block_add_ext, csum,
3523 csum2, pos, p_len);
3524 pos += p_len;
3525 }
3526
3527 if (!(len -= copy))
3528 return csum;
3529 offset += copy;
3530 }
3531 start = end;
3532 }
3533
3534 skb_walk_frags(skb, frag_iter) {
3535 int end;
3536
3537 WARN_ON(start > offset + len);
3538
3539 end = start + frag_iter->len;
3540 if ((copy = end - offset) > 0) {
3541 __wsum csum2;
3542 if (copy > len)
3543 copy = len;
3544 csum2 = __skb_checksum(frag_iter, offset - start,
3545 copy, 0, ops);
3546 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
3547 csum, csum2, pos, copy);
3548 if ((len -= copy) == 0)
3549 return csum;
3550 offset += copy;
3551 pos += copy;
3552 }
3553 start = end;
3554 }
3555 BUG_ON(len);
3556
3557 return csum;
3558 }
3559 EXPORT_SYMBOL(__skb_checksum);
3560
skb_checksum(const struct sk_buff * skb,int offset,int len,__wsum csum)3561 __wsum skb_checksum(const struct sk_buff *skb, int offset,
3562 int len, __wsum csum)
3563 {
3564 const struct skb_checksum_ops ops = {
3565 .update = csum_partial_ext,
3566 .combine = csum_block_add_ext,
3567 };
3568
3569 return __skb_checksum(skb, offset, len, csum, &ops);
3570 }
3571 EXPORT_SYMBOL(skb_checksum);
3572
3573 /* Both of above in one bottle. */
3574
skb_copy_and_csum_bits(const struct sk_buff * skb,int offset,u8 * to,int len)3575 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
3576 u8 *to, int len)
3577 {
3578 int start = skb_headlen(skb);
3579 int i, copy = start - offset;
3580 struct sk_buff *frag_iter;
3581 int pos = 0;
3582 __wsum csum = 0;
3583
3584 /* Copy header. */
3585 if (copy > 0) {
3586 if (copy > len)
3587 copy = len;
3588 csum = csum_partial_copy_nocheck(skb->data + offset, to,
3589 copy);
3590 if ((len -= copy) == 0)
3591 return csum;
3592 offset += copy;
3593 to += copy;
3594 pos = copy;
3595 }
3596
3597 if (!skb_frags_readable(skb))
3598 return 0;
3599
3600 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3601 int end;
3602
3603 WARN_ON(start > offset + len);
3604
3605 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3606 if ((copy = end - offset) > 0) {
3607 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3608 u32 p_off, p_len, copied;
3609 struct page *p;
3610 __wsum csum2;
3611 u8 *vaddr;
3612
3613 if (copy > len)
3614 copy = len;
3615
3616 skb_frag_foreach_page(frag,
3617 skb_frag_off(frag) + offset - start,
3618 copy, p, p_off, p_len, copied) {
3619 vaddr = kmap_atomic(p);
3620 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
3621 to + copied,
3622 p_len);
3623 kunmap_atomic(vaddr);
3624 csum = csum_block_add(csum, csum2, pos);
3625 pos += p_len;
3626 }
3627
3628 if (!(len -= copy))
3629 return csum;
3630 offset += copy;
3631 to += copy;
3632 }
3633 start = end;
3634 }
3635
3636 skb_walk_frags(skb, frag_iter) {
3637 __wsum csum2;
3638 int end;
3639
3640 WARN_ON(start > offset + len);
3641
3642 end = start + frag_iter->len;
3643 if ((copy = end - offset) > 0) {
3644 if (copy > len)
3645 copy = len;
3646 csum2 = skb_copy_and_csum_bits(frag_iter,
3647 offset - start,
3648 to, copy);
3649 csum = csum_block_add(csum, csum2, pos);
3650 if ((len -= copy) == 0)
3651 return csum;
3652 offset += copy;
3653 to += copy;
3654 pos += copy;
3655 }
3656 start = end;
3657 }
3658 BUG_ON(len);
3659 return csum;
3660 }
3661 EXPORT_SYMBOL(skb_copy_and_csum_bits);
3662
__skb_checksum_complete_head(struct sk_buff * skb,int len)3663 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
3664 {
3665 __sum16 sum;
3666
3667 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
3668 /* See comments in __skb_checksum_complete(). */
3669 if (likely(!sum)) {
3670 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3671 !skb->csum_complete_sw)
3672 netdev_rx_csum_fault(skb->dev, skb);
3673 }
3674 if (!skb_shared(skb))
3675 skb->csum_valid = !sum;
3676 return sum;
3677 }
3678 EXPORT_SYMBOL(__skb_checksum_complete_head);
3679
3680 /* This function assumes skb->csum already holds pseudo header's checksum,
3681 * which has been changed from the hardware checksum, for example, by
3682 * __skb_checksum_validate_complete(). And, the original skb->csum must
3683 * have been validated unsuccessfully for CHECKSUM_COMPLETE case.
3684 *
3685 * It returns non-zero if the recomputed checksum is still invalid, otherwise
3686 * zero. The new checksum is stored back into skb->csum unless the skb is
3687 * shared.
3688 */
__skb_checksum_complete(struct sk_buff * skb)3689 __sum16 __skb_checksum_complete(struct sk_buff *skb)
3690 {
3691 __wsum csum;
3692 __sum16 sum;
3693
3694 csum = skb_checksum(skb, 0, skb->len, 0);
3695
3696 sum = csum_fold(csum_add(skb->csum, csum));
3697 /* This check is inverted, because we already knew the hardware
3698 * checksum is invalid before calling this function. So, if the
3699 * re-computed checksum is valid instead, then we have a mismatch
3700 * between the original skb->csum and skb_checksum(). This means either
3701 * the original hardware checksum is incorrect or we screw up skb->csum
3702 * when moving skb->data around.
3703 */
3704 if (likely(!sum)) {
3705 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
3706 !skb->csum_complete_sw)
3707 netdev_rx_csum_fault(skb->dev, skb);
3708 }
3709
3710 if (!skb_shared(skb)) {
3711 /* Save full packet checksum */
3712 skb->csum = csum;
3713 skb->ip_summed = CHECKSUM_COMPLETE;
3714 skb->csum_complete_sw = 1;
3715 skb->csum_valid = !sum;
3716 }
3717
3718 return sum;
3719 }
3720 EXPORT_SYMBOL(__skb_checksum_complete);
3721
warn_crc32c_csum_update(const void * buff,int len,__wsum sum)3722 static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
3723 {
3724 net_warn_ratelimited(
3725 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3726 __func__);
3727 return 0;
3728 }
3729
warn_crc32c_csum_combine(__wsum csum,__wsum csum2,int offset,int len)3730 static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
3731 int offset, int len)
3732 {
3733 net_warn_ratelimited(
3734 "%s: attempt to compute crc32c without libcrc32c.ko\n",
3735 __func__);
3736 return 0;
3737 }
3738
3739 static const struct skb_checksum_ops default_crc32c_ops = {
3740 .update = warn_crc32c_csum_update,
3741 .combine = warn_crc32c_csum_combine,
3742 };
3743
3744 const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
3745 &default_crc32c_ops;
3746 EXPORT_SYMBOL(crc32c_csum_stub);
3747
3748 /**
3749 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
3750 * @from: source buffer
3751 *
3752 * Calculates the amount of linear headroom needed in the 'to' skb passed
3753 * into skb_zerocopy().
3754 */
3755 unsigned int
skb_zerocopy_headlen(const struct sk_buff * from)3756 skb_zerocopy_headlen(const struct sk_buff *from)
3757 {
3758 unsigned int hlen = 0;
3759
3760 if (!from->head_frag ||
3761 skb_headlen(from) < L1_CACHE_BYTES ||
3762 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS) {
3763 hlen = skb_headlen(from);
3764 if (!hlen)
3765 hlen = from->len;
3766 }
3767
3768 if (skb_has_frag_list(from))
3769 hlen = from->len;
3770
3771 return hlen;
3772 }
3773 EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
3774
3775 /**
3776 * skb_zerocopy - Zero copy skb to skb
3777 * @to: destination buffer
3778 * @from: source buffer
3779 * @len: number of bytes to copy from source buffer
3780 * @hlen: size of linear headroom in destination buffer
3781 *
3782 * Copies up to `len` bytes from `from` to `to` by creating references
3783 * to the frags in the source buffer.
3784 *
3785 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
3786 * headroom in the `to` buffer.
3787 *
3788 * Return value:
3789 * 0: everything is OK
3790 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
3791 * -EFAULT: skb_copy_bits() found some problem with skb geometry
3792 */
3793 int
skb_zerocopy(struct sk_buff * to,struct sk_buff * from,int len,int hlen)3794 skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
3795 {
3796 int i, j = 0;
3797 int plen = 0; /* length of skb->head fragment */
3798 int ret;
3799 struct page *page;
3800 unsigned int offset;
3801
3802 BUG_ON(!from->head_frag && !hlen);
3803
3804 /* dont bother with small payloads */
3805 if (len <= skb_tailroom(to))
3806 return skb_copy_bits(from, 0, skb_put(to, len), len);
3807
3808 if (hlen) {
3809 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
3810 if (unlikely(ret))
3811 return ret;
3812 len -= hlen;
3813 } else {
3814 plen = min_t(int, skb_headlen(from), len);
3815 if (plen) {
3816 page = virt_to_head_page(from->head);
3817 offset = from->data - (unsigned char *)page_address(page);
3818 __skb_fill_netmem_desc(to, 0, page_to_netmem(page),
3819 offset, plen);
3820 get_page(page);
3821 j = 1;
3822 len -= plen;
3823 }
3824 }
3825
3826 skb_len_add(to, len + plen);
3827
3828 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
3829 skb_tx_error(from);
3830 return -ENOMEM;
3831 }
3832 skb_zerocopy_clone(to, from, GFP_ATOMIC);
3833
3834 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
3835 int size;
3836
3837 if (!len)
3838 break;
3839 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
3840 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
3841 len);
3842 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
3843 len -= size;
3844 skb_frag_ref(to, j);
3845 j++;
3846 }
3847 skb_shinfo(to)->nr_frags = j;
3848
3849 return 0;
3850 }
3851 EXPORT_SYMBOL_GPL(skb_zerocopy);
3852
skb_copy_and_csum_dev(const struct sk_buff * skb,u8 * to)3853 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
3854 {
3855 __wsum csum;
3856 long csstart;
3857
3858 if (skb->ip_summed == CHECKSUM_PARTIAL)
3859 csstart = skb_checksum_start_offset(skb);
3860 else
3861 csstart = skb_headlen(skb);
3862
3863 BUG_ON(csstart > skb_headlen(skb));
3864
3865 skb_copy_from_linear_data(skb, to, csstart);
3866
3867 csum = 0;
3868 if (csstart != skb->len)
3869 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3870 skb->len - csstart);
3871
3872 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3873 long csstuff = csstart + skb->csum_offset;
3874
3875 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3876 }
3877 }
3878 EXPORT_SYMBOL(skb_copy_and_csum_dev);
3879
3880 /**
3881 * skb_dequeue - remove from the head of the queue
3882 * @list: list to dequeue from
3883 *
3884 * Remove the head of the list. The list lock is taken so the function
3885 * may be used safely with other locking list functions. The head item is
3886 * returned or %NULL if the list is empty.
3887 */
3888
skb_dequeue(struct sk_buff_head * list)3889 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3890 {
3891 unsigned long flags;
3892 struct sk_buff *result;
3893
3894 spin_lock_irqsave(&list->lock, flags);
3895 result = __skb_dequeue(list);
3896 spin_unlock_irqrestore(&list->lock, flags);
3897 return result;
3898 }
3899 EXPORT_SYMBOL(skb_dequeue);
3900
3901 /**
3902 * skb_dequeue_tail - remove from the tail of the queue
3903 * @list: list to dequeue from
3904 *
3905 * Remove the tail of the list. The list lock is taken so the function
3906 * may be used safely with other locking list functions. The tail item is
3907 * returned or %NULL if the list is empty.
3908 */
skb_dequeue_tail(struct sk_buff_head * list)3909 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3910 {
3911 unsigned long flags;
3912 struct sk_buff *result;
3913
3914 spin_lock_irqsave(&list->lock, flags);
3915 result = __skb_dequeue_tail(list);
3916 spin_unlock_irqrestore(&list->lock, flags);
3917 return result;
3918 }
3919 EXPORT_SYMBOL(skb_dequeue_tail);
3920
3921 /**
3922 * skb_queue_purge_reason - empty a list
3923 * @list: list to empty
3924 * @reason: drop reason
3925 *
3926 * Delete all buffers on an &sk_buff list. Each buffer is removed from
3927 * the list and one reference dropped. This function takes the list
3928 * lock and is atomic with respect to other list locking functions.
3929 */
skb_queue_purge_reason(struct sk_buff_head * list,enum skb_drop_reason reason)3930 void skb_queue_purge_reason(struct sk_buff_head *list,
3931 enum skb_drop_reason reason)
3932 {
3933 struct sk_buff_head tmp;
3934 unsigned long flags;
3935
3936 if (skb_queue_empty_lockless(list))
3937 return;
3938
3939 __skb_queue_head_init(&tmp);
3940
3941 spin_lock_irqsave(&list->lock, flags);
3942 skb_queue_splice_init(list, &tmp);
3943 spin_unlock_irqrestore(&list->lock, flags);
3944
3945 __skb_queue_purge_reason(&tmp, reason);
3946 }
3947 EXPORT_SYMBOL(skb_queue_purge_reason);
3948
3949 /**
3950 * skb_rbtree_purge - empty a skb rbtree
3951 * @root: root of the rbtree to empty
3952 * Return value: the sum of truesizes of all purged skbs.
3953 *
3954 * Delete all buffers on an &sk_buff rbtree. Each buffer is removed from
3955 * the list and one reference dropped. This function does not take
3956 * any lock. Synchronization should be handled by the caller (e.g., TCP
3957 * out-of-order queue is protected by the socket lock).
3958 */
skb_rbtree_purge(struct rb_root * root)3959 unsigned int skb_rbtree_purge(struct rb_root *root)
3960 {
3961 struct rb_node *p = rb_first(root);
3962 unsigned int sum = 0;
3963
3964 while (p) {
3965 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3966
3967 p = rb_next(p);
3968 rb_erase(&skb->rbnode, root);
3969 sum += skb->truesize;
3970 kfree_skb(skb);
3971 }
3972 return sum;
3973 }
3974
skb_errqueue_purge(struct sk_buff_head * list)3975 void skb_errqueue_purge(struct sk_buff_head *list)
3976 {
3977 struct sk_buff *skb, *next;
3978 struct sk_buff_head kill;
3979 unsigned long flags;
3980
3981 __skb_queue_head_init(&kill);
3982
3983 spin_lock_irqsave(&list->lock, flags);
3984 skb_queue_walk_safe(list, skb, next) {
3985 if (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ZEROCOPY ||
3986 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_TIMESTAMPING)
3987 continue;
3988 __skb_unlink(skb, list);
3989 __skb_queue_tail(&kill, skb);
3990 }
3991 spin_unlock_irqrestore(&list->lock, flags);
3992 __skb_queue_purge(&kill);
3993 }
3994 EXPORT_SYMBOL(skb_errqueue_purge);
3995
3996 /**
3997 * skb_queue_head - queue a buffer at the list head
3998 * @list: list to use
3999 * @newsk: buffer to queue
4000 *
4001 * Queue a buffer at the start of the list. This function takes the
4002 * list lock and can be used safely with other locking &sk_buff functions
4003 * safely.
4004 *
4005 * A buffer cannot be placed on two lists at the same time.
4006 */
skb_queue_head(struct sk_buff_head * list,struct sk_buff * newsk)4007 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
4008 {
4009 unsigned long flags;
4010
4011 spin_lock_irqsave(&list->lock, flags);
4012 __skb_queue_head(list, newsk);
4013 spin_unlock_irqrestore(&list->lock, flags);
4014 }
4015 EXPORT_SYMBOL(skb_queue_head);
4016
4017 /**
4018 * skb_queue_tail - queue a buffer at the list tail
4019 * @list: list to use
4020 * @newsk: buffer to queue
4021 *
4022 * Queue a buffer at the tail of the list. This function takes the
4023 * list lock and can be used safely with other locking &sk_buff functions
4024 * safely.
4025 *
4026 * A buffer cannot be placed on two lists at the same time.
4027 */
skb_queue_tail(struct sk_buff_head * list,struct sk_buff * newsk)4028 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
4029 {
4030 unsigned long flags;
4031
4032 spin_lock_irqsave(&list->lock, flags);
4033 __skb_queue_tail(list, newsk);
4034 spin_unlock_irqrestore(&list->lock, flags);
4035 }
4036 EXPORT_SYMBOL(skb_queue_tail);
4037
4038 /**
4039 * skb_unlink - remove a buffer from a list
4040 * @skb: buffer to remove
4041 * @list: list to use
4042 *
4043 * Remove a packet from a list. The list locks are taken and this
4044 * function is atomic with respect to other list locked calls
4045 *
4046 * You must know what list the SKB is on.
4047 */
skb_unlink(struct sk_buff * skb,struct sk_buff_head * list)4048 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
4049 {
4050 unsigned long flags;
4051
4052 spin_lock_irqsave(&list->lock, flags);
4053 __skb_unlink(skb, list);
4054 spin_unlock_irqrestore(&list->lock, flags);
4055 }
4056 EXPORT_SYMBOL(skb_unlink);
4057
4058 /**
4059 * skb_append - append a buffer
4060 * @old: buffer to insert after
4061 * @newsk: buffer to insert
4062 * @list: list to use
4063 *
4064 * Place a packet after a given packet in a list. The list locks are taken
4065 * and this function is atomic with respect to other list locked calls.
4066 * A buffer cannot be placed on two lists at the same time.
4067 */
skb_append(struct sk_buff * old,struct sk_buff * newsk,struct sk_buff_head * list)4068 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
4069 {
4070 unsigned long flags;
4071
4072 spin_lock_irqsave(&list->lock, flags);
4073 __skb_queue_after(list, old, newsk);
4074 spin_unlock_irqrestore(&list->lock, flags);
4075 }
4076 EXPORT_SYMBOL(skb_append);
4077
skb_split_inside_header(struct sk_buff * skb,struct sk_buff * skb1,const u32 len,const int pos)4078 static inline void skb_split_inside_header(struct sk_buff *skb,
4079 struct sk_buff* skb1,
4080 const u32 len, const int pos)
4081 {
4082 int i;
4083
4084 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
4085 pos - len);
4086 /* And move data appendix as is. */
4087 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4088 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
4089
4090 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
4091 skb1->unreadable = skb->unreadable;
4092 skb_shinfo(skb)->nr_frags = 0;
4093 skb1->data_len = skb->data_len;
4094 skb1->len += skb1->data_len;
4095 skb->data_len = 0;
4096 skb->len = len;
4097 skb_set_tail_pointer(skb, len);
4098 }
4099
skb_split_no_header(struct sk_buff * skb,struct sk_buff * skb1,const u32 len,int pos)4100 static inline void skb_split_no_header(struct sk_buff *skb,
4101 struct sk_buff* skb1,
4102 const u32 len, int pos)
4103 {
4104 int i, k = 0;
4105 const int nfrags = skb_shinfo(skb)->nr_frags;
4106
4107 skb_shinfo(skb)->nr_frags = 0;
4108 skb1->len = skb1->data_len = skb->len - len;
4109 skb->len = len;
4110 skb->data_len = len - pos;
4111
4112 for (i = 0; i < nfrags; i++) {
4113 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4114
4115 if (pos + size > len) {
4116 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
4117
4118 if (pos < len) {
4119 /* Split frag.
4120 * We have two variants in this case:
4121 * 1. Move all the frag to the second
4122 * part, if it is possible. F.e.
4123 * this approach is mandatory for TUX,
4124 * where splitting is expensive.
4125 * 2. Split is accurately. We make this.
4126 */
4127 skb_frag_ref(skb, i);
4128 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
4129 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
4130 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
4131 skb_shinfo(skb)->nr_frags++;
4132 }
4133 k++;
4134 } else
4135 skb_shinfo(skb)->nr_frags++;
4136 pos += size;
4137 }
4138 skb_shinfo(skb1)->nr_frags = k;
4139
4140 skb1->unreadable = skb->unreadable;
4141 }
4142
4143 /**
4144 * skb_split - Split fragmented skb to two parts at length len.
4145 * @skb: the buffer to split
4146 * @skb1: the buffer to receive the second part
4147 * @len: new length for skb
4148 */
skb_split(struct sk_buff * skb,struct sk_buff * skb1,const u32 len)4149 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
4150 {
4151 int pos = skb_headlen(skb);
4152 const int zc_flags = SKBFL_SHARED_FRAG | SKBFL_PURE_ZEROCOPY;
4153
4154 skb_zcopy_downgrade_managed(skb);
4155
4156 skb_shinfo(skb1)->flags |= skb_shinfo(skb)->flags & zc_flags;
4157 skb_zerocopy_clone(skb1, skb, 0);
4158 if (len < pos) /* Split line is inside header. */
4159 skb_split_inside_header(skb, skb1, len, pos);
4160 else /* Second chunk has no header, nothing to copy. */
4161 skb_split_no_header(skb, skb1, len, pos);
4162 }
4163 EXPORT_SYMBOL(skb_split);
4164
4165 /* Shifting from/to a cloned skb is a no-go.
4166 *
4167 * Caller cannot keep skb_shinfo related pointers past calling here!
4168 */
skb_prepare_for_shift(struct sk_buff * skb)4169 static int skb_prepare_for_shift(struct sk_buff *skb)
4170 {
4171 return skb_unclone_keeptruesize(skb, GFP_ATOMIC);
4172 }
4173
4174 /**
4175 * skb_shift - Shifts paged data partially from skb to another
4176 * @tgt: buffer into which tail data gets added
4177 * @skb: buffer from which the paged data comes from
4178 * @shiftlen: shift up to this many bytes
4179 *
4180 * Attempts to shift up to shiftlen worth of bytes, which may be less than
4181 * the length of the skb, from skb to tgt. Returns number bytes shifted.
4182 * It's up to caller to free skb if everything was shifted.
4183 *
4184 * If @tgt runs out of frags, the whole operation is aborted.
4185 *
4186 * Skb cannot include anything else but paged data while tgt is allowed
4187 * to have non-paged data as well.
4188 *
4189 * TODO: full sized shift could be optimized but that would need
4190 * specialized skb free'er to handle frags without up-to-date nr_frags.
4191 */
skb_shift(struct sk_buff * tgt,struct sk_buff * skb,int shiftlen)4192 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
4193 {
4194 int from, to, merge, todo;
4195 skb_frag_t *fragfrom, *fragto;
4196
4197 BUG_ON(shiftlen > skb->len);
4198
4199 if (skb_headlen(skb))
4200 return 0;
4201 if (skb_zcopy(tgt) || skb_zcopy(skb))
4202 return 0;
4203
4204 DEBUG_NET_WARN_ON_ONCE(tgt->pp_recycle != skb->pp_recycle);
4205 DEBUG_NET_WARN_ON_ONCE(skb_cmp_decrypted(tgt, skb));
4206
4207 todo = shiftlen;
4208 from = 0;
4209 to = skb_shinfo(tgt)->nr_frags;
4210 fragfrom = &skb_shinfo(skb)->frags[from];
4211
4212 /* Actual merge is delayed until the point when we know we can
4213 * commit all, so that we don't have to undo partial changes
4214 */
4215 if (!skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
4216 skb_frag_off(fragfrom))) {
4217 merge = -1;
4218 } else {
4219 merge = to - 1;
4220
4221 todo -= skb_frag_size(fragfrom);
4222 if (todo < 0) {
4223 if (skb_prepare_for_shift(skb) ||
4224 skb_prepare_for_shift(tgt))
4225 return 0;
4226
4227 /* All previous frag pointers might be stale! */
4228 fragfrom = &skb_shinfo(skb)->frags[from];
4229 fragto = &skb_shinfo(tgt)->frags[merge];
4230
4231 skb_frag_size_add(fragto, shiftlen);
4232 skb_frag_size_sub(fragfrom, shiftlen);
4233 skb_frag_off_add(fragfrom, shiftlen);
4234
4235 goto onlymerged;
4236 }
4237
4238 from++;
4239 }
4240
4241 /* Skip full, not-fitting skb to avoid expensive operations */
4242 if ((shiftlen == skb->len) &&
4243 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
4244 return 0;
4245
4246 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
4247 return 0;
4248
4249 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
4250 if (to == MAX_SKB_FRAGS)
4251 return 0;
4252
4253 fragfrom = &skb_shinfo(skb)->frags[from];
4254 fragto = &skb_shinfo(tgt)->frags[to];
4255
4256 if (todo >= skb_frag_size(fragfrom)) {
4257 *fragto = *fragfrom;
4258 todo -= skb_frag_size(fragfrom);
4259 from++;
4260 to++;
4261
4262 } else {
4263 __skb_frag_ref(fragfrom);
4264 skb_frag_page_copy(fragto, fragfrom);
4265 skb_frag_off_copy(fragto, fragfrom);
4266 skb_frag_size_set(fragto, todo);
4267
4268 skb_frag_off_add(fragfrom, todo);
4269 skb_frag_size_sub(fragfrom, todo);
4270 todo = 0;
4271
4272 to++;
4273 break;
4274 }
4275 }
4276
4277 /* Ready to "commit" this state change to tgt */
4278 skb_shinfo(tgt)->nr_frags = to;
4279
4280 if (merge >= 0) {
4281 fragfrom = &skb_shinfo(skb)->frags[0];
4282 fragto = &skb_shinfo(tgt)->frags[merge];
4283
4284 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
4285 __skb_frag_unref(fragfrom, skb->pp_recycle);
4286 }
4287
4288 /* Reposition in the original skb */
4289 to = 0;
4290 while (from < skb_shinfo(skb)->nr_frags)
4291 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
4292 skb_shinfo(skb)->nr_frags = to;
4293
4294 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
4295
4296 onlymerged:
4297 /* Most likely the tgt won't ever need its checksum anymore, skb on
4298 * the other hand might need it if it needs to be resent
4299 */
4300 tgt->ip_summed = CHECKSUM_PARTIAL;
4301 skb->ip_summed = CHECKSUM_PARTIAL;
4302
4303 skb_len_add(skb, -shiftlen);
4304 skb_len_add(tgt, shiftlen);
4305
4306 return shiftlen;
4307 }
4308
4309 /**
4310 * skb_prepare_seq_read - Prepare a sequential read of skb data
4311 * @skb: the buffer to read
4312 * @from: lower offset of data to be read
4313 * @to: upper offset of data to be read
4314 * @st: state variable
4315 *
4316 * Initializes the specified state variable. Must be called before
4317 * invoking skb_seq_read() for the first time.
4318 */
skb_prepare_seq_read(struct sk_buff * skb,unsigned int from,unsigned int to,struct skb_seq_state * st)4319 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
4320 unsigned int to, struct skb_seq_state *st)
4321 {
4322 st->lower_offset = from;
4323 st->upper_offset = to;
4324 st->root_skb = st->cur_skb = skb;
4325 st->frag_idx = st->stepped_offset = 0;
4326 st->frag_data = NULL;
4327 st->frag_off = 0;
4328 }
4329 EXPORT_SYMBOL(skb_prepare_seq_read);
4330
4331 /**
4332 * skb_seq_read - Sequentially read skb data
4333 * @consumed: number of bytes consumed by the caller so far
4334 * @data: destination pointer for data to be returned
4335 * @st: state variable
4336 *
4337 * Reads a block of skb data at @consumed relative to the
4338 * lower offset specified to skb_prepare_seq_read(). Assigns
4339 * the head of the data block to @data and returns the length
4340 * of the block or 0 if the end of the skb data or the upper
4341 * offset has been reached.
4342 *
4343 * The caller is not required to consume all of the data
4344 * returned, i.e. @consumed is typically set to the number
4345 * of bytes already consumed and the next call to
4346 * skb_seq_read() will return the remaining part of the block.
4347 *
4348 * Note 1: The size of each block of data returned can be arbitrary,
4349 * this limitation is the cost for zerocopy sequential
4350 * reads of potentially non linear data.
4351 *
4352 * Note 2: Fragment lists within fragments are not implemented
4353 * at the moment, state->root_skb could be replaced with
4354 * a stack for this purpose.
4355 */
skb_seq_read(unsigned int consumed,const u8 ** data,struct skb_seq_state * st)4356 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
4357 struct skb_seq_state *st)
4358 {
4359 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
4360 skb_frag_t *frag;
4361
4362 if (unlikely(abs_offset >= st->upper_offset)) {
4363 if (st->frag_data) {
4364 kunmap_atomic(st->frag_data);
4365 st->frag_data = NULL;
4366 }
4367 return 0;
4368 }
4369
4370 next_skb:
4371 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
4372
4373 if (abs_offset < block_limit && !st->frag_data) {
4374 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
4375 return block_limit - abs_offset;
4376 }
4377
4378 if (!skb_frags_readable(st->cur_skb))
4379 return 0;
4380
4381 if (st->frag_idx == 0 && !st->frag_data)
4382 st->stepped_offset += skb_headlen(st->cur_skb);
4383
4384 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
4385 unsigned int pg_idx, pg_off, pg_sz;
4386
4387 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
4388
4389 pg_idx = 0;
4390 pg_off = skb_frag_off(frag);
4391 pg_sz = skb_frag_size(frag);
4392
4393 if (skb_frag_must_loop(skb_frag_page(frag))) {
4394 pg_idx = (pg_off + st->frag_off) >> PAGE_SHIFT;
4395 pg_off = offset_in_page(pg_off + st->frag_off);
4396 pg_sz = min_t(unsigned int, pg_sz - st->frag_off,
4397 PAGE_SIZE - pg_off);
4398 }
4399
4400 block_limit = pg_sz + st->stepped_offset;
4401 if (abs_offset < block_limit) {
4402 if (!st->frag_data)
4403 st->frag_data = kmap_atomic(skb_frag_page(frag) + pg_idx);
4404
4405 *data = (u8 *)st->frag_data + pg_off +
4406 (abs_offset - st->stepped_offset);
4407
4408 return block_limit - abs_offset;
4409 }
4410
4411 if (st->frag_data) {
4412 kunmap_atomic(st->frag_data);
4413 st->frag_data = NULL;
4414 }
4415
4416 st->stepped_offset += pg_sz;
4417 st->frag_off += pg_sz;
4418 if (st->frag_off == skb_frag_size(frag)) {
4419 st->frag_off = 0;
4420 st->frag_idx++;
4421 }
4422 }
4423
4424 if (st->frag_data) {
4425 kunmap_atomic(st->frag_data);
4426 st->frag_data = NULL;
4427 }
4428
4429 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
4430 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
4431 st->frag_idx = 0;
4432 goto next_skb;
4433 } else if (st->cur_skb->next) {
4434 st->cur_skb = st->cur_skb->next;
4435 st->frag_idx = 0;
4436 goto next_skb;
4437 }
4438
4439 return 0;
4440 }
4441 EXPORT_SYMBOL(skb_seq_read);
4442
4443 /**
4444 * skb_abort_seq_read - Abort a sequential read of skb data
4445 * @st: state variable
4446 *
4447 * Must be called if skb_seq_read() was not called until it
4448 * returned 0.
4449 */
skb_abort_seq_read(struct skb_seq_state * st)4450 void skb_abort_seq_read(struct skb_seq_state *st)
4451 {
4452 if (st->frag_data)
4453 kunmap_atomic(st->frag_data);
4454 }
4455 EXPORT_SYMBOL(skb_abort_seq_read);
4456
4457 /**
4458 * skb_copy_seq_read() - copy from a skb_seq_state to a buffer
4459 * @st: source skb_seq_state
4460 * @offset: offset in source
4461 * @to: destination buffer
4462 * @len: number of bytes to copy
4463 *
4464 * Copy @len bytes from @offset bytes into the source @st to the destination
4465 * buffer @to. `offset` should increase (or be unchanged) with each subsequent
4466 * call to this function. If offset needs to decrease from the previous use `st`
4467 * should be reset first.
4468 *
4469 * Return: 0 on success or -EINVAL if the copy ended early
4470 */
skb_copy_seq_read(struct skb_seq_state * st,int offset,void * to,int len)4471 int skb_copy_seq_read(struct skb_seq_state *st, int offset, void *to, int len)
4472 {
4473 const u8 *data;
4474 u32 sqlen;
4475
4476 for (;;) {
4477 sqlen = skb_seq_read(offset, &data, st);
4478 if (sqlen == 0)
4479 return -EINVAL;
4480 if (sqlen >= len) {
4481 memcpy(to, data, len);
4482 return 0;
4483 }
4484 memcpy(to, data, sqlen);
4485 to += sqlen;
4486 offset += sqlen;
4487 len -= sqlen;
4488 }
4489 }
4490 EXPORT_SYMBOL(skb_copy_seq_read);
4491
4492 #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
4493
skb_ts_get_next_block(unsigned int offset,const u8 ** text,struct ts_config * conf,struct ts_state * state)4494 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
4495 struct ts_config *conf,
4496 struct ts_state *state)
4497 {
4498 return skb_seq_read(offset, text, TS_SKB_CB(state));
4499 }
4500
skb_ts_finish(struct ts_config * conf,struct ts_state * state)4501 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
4502 {
4503 skb_abort_seq_read(TS_SKB_CB(state));
4504 }
4505
4506 /**
4507 * skb_find_text - Find a text pattern in skb data
4508 * @skb: the buffer to look in
4509 * @from: search offset
4510 * @to: search limit
4511 * @config: textsearch configuration
4512 *
4513 * Finds a pattern in the skb data according to the specified
4514 * textsearch configuration. Use textsearch_next() to retrieve
4515 * subsequent occurrences of the pattern. Returns the offset
4516 * to the first occurrence or UINT_MAX if no match was found.
4517 */
skb_find_text(struct sk_buff * skb,unsigned int from,unsigned int to,struct ts_config * config)4518 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
4519 unsigned int to, struct ts_config *config)
4520 {
4521 unsigned int patlen = config->ops->get_pattern_len(config);
4522 struct ts_state state;
4523 unsigned int ret;
4524
4525 BUILD_BUG_ON(sizeof(struct skb_seq_state) > sizeof(state.cb));
4526
4527 config->get_next_block = skb_ts_get_next_block;
4528 config->finish = skb_ts_finish;
4529
4530 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
4531
4532 ret = textsearch_find(config, &state);
4533 return (ret + patlen <= to - from ? ret : UINT_MAX);
4534 }
4535 EXPORT_SYMBOL(skb_find_text);
4536
skb_append_pagefrags(struct sk_buff * skb,struct page * page,int offset,size_t size,size_t max_frags)4537 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
4538 int offset, size_t size, size_t max_frags)
4539 {
4540 int i = skb_shinfo(skb)->nr_frags;
4541
4542 if (skb_can_coalesce(skb, i, page, offset)) {
4543 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
4544 } else if (i < max_frags) {
4545 skb_zcopy_downgrade_managed(skb);
4546 get_page(page);
4547 skb_fill_page_desc_noacc(skb, i, page, offset, size);
4548 } else {
4549 return -EMSGSIZE;
4550 }
4551
4552 return 0;
4553 }
4554 EXPORT_SYMBOL_GPL(skb_append_pagefrags);
4555
4556 /**
4557 * skb_pull_rcsum - pull skb and update receive checksum
4558 * @skb: buffer to update
4559 * @len: length of data pulled
4560 *
4561 * This function performs an skb_pull on the packet and updates
4562 * the CHECKSUM_COMPLETE checksum. It should be used on
4563 * receive path processing instead of skb_pull unless you know
4564 * that the checksum difference is zero (e.g., a valid IP header)
4565 * or you are setting ip_summed to CHECKSUM_NONE.
4566 */
skb_pull_rcsum(struct sk_buff * skb,unsigned int len)4567 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
4568 {
4569 unsigned char *data = skb->data;
4570
4571 BUG_ON(len > skb->len);
4572 __skb_pull(skb, len);
4573 skb_postpull_rcsum(skb, data, len);
4574 return skb->data;
4575 }
4576 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
4577
skb_head_frag_to_page_desc(struct sk_buff * frag_skb)4578 static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
4579 {
4580 skb_frag_t head_frag;
4581 struct page *page;
4582
4583 page = virt_to_head_page(frag_skb->head);
4584 skb_frag_fill_page_desc(&head_frag, page, frag_skb->data -
4585 (unsigned char *)page_address(page),
4586 skb_headlen(frag_skb));
4587 return head_frag;
4588 }
4589
skb_segment_list(struct sk_buff * skb,netdev_features_t features,unsigned int offset)4590 struct sk_buff *skb_segment_list(struct sk_buff *skb,
4591 netdev_features_t features,
4592 unsigned int offset)
4593 {
4594 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
4595 unsigned int tnl_hlen = skb_tnl_header_len(skb);
4596 unsigned int delta_truesize = 0;
4597 unsigned int delta_len = 0;
4598 struct sk_buff *tail = NULL;
4599 struct sk_buff *nskb, *tmp;
4600 int len_diff, err;
4601
4602 skb_push(skb, -skb_network_offset(skb) + offset);
4603
4604 /* Ensure the head is writeable before touching the shared info */
4605 err = skb_unclone(skb, GFP_ATOMIC);
4606 if (err)
4607 goto err_linearize;
4608
4609 skb_shinfo(skb)->frag_list = NULL;
4610
4611 while (list_skb) {
4612 nskb = list_skb;
4613 list_skb = list_skb->next;
4614
4615 err = 0;
4616 delta_truesize += nskb->truesize;
4617 if (skb_shared(nskb)) {
4618 tmp = skb_clone(nskb, GFP_ATOMIC);
4619 if (tmp) {
4620 consume_skb(nskb);
4621 nskb = tmp;
4622 err = skb_unclone(nskb, GFP_ATOMIC);
4623 } else {
4624 err = -ENOMEM;
4625 }
4626 }
4627
4628 if (!tail)
4629 skb->next = nskb;
4630 else
4631 tail->next = nskb;
4632
4633 if (unlikely(err)) {
4634 nskb->next = list_skb;
4635 goto err_linearize;
4636 }
4637
4638 tail = nskb;
4639
4640 delta_len += nskb->len;
4641
4642 skb_push(nskb, -skb_network_offset(nskb) + offset);
4643
4644 skb_release_head_state(nskb);
4645 len_diff = skb_network_header_len(nskb) - skb_network_header_len(skb);
4646 __copy_skb_header(nskb, skb);
4647
4648 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
4649 nskb->transport_header += len_diff;
4650 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
4651 nskb->data - tnl_hlen,
4652 offset + tnl_hlen);
4653
4654 if (skb_needs_linearize(nskb, features) &&
4655 __skb_linearize(nskb))
4656 goto err_linearize;
4657 }
4658
4659 skb->truesize = skb->truesize - delta_truesize;
4660 skb->data_len = skb->data_len - delta_len;
4661 skb->len = skb->len - delta_len;
4662
4663 skb_gso_reset(skb);
4664
4665 skb->prev = tail;
4666
4667 if (skb_needs_linearize(skb, features) &&
4668 __skb_linearize(skb))
4669 goto err_linearize;
4670
4671 skb_get(skb);
4672
4673 return skb;
4674
4675 err_linearize:
4676 kfree_skb_list(skb->next);
4677 skb->next = NULL;
4678 return ERR_PTR(-ENOMEM);
4679 }
4680 EXPORT_SYMBOL_GPL(skb_segment_list);
4681
4682 /**
4683 * skb_segment - Perform protocol segmentation on skb.
4684 * @head_skb: buffer to segment
4685 * @features: features for the output path (see dev->features)
4686 *
4687 * This function performs segmentation on the given skb. It returns
4688 * a pointer to the first in a list of new skbs for the segments.
4689 * In case of error it returns ERR_PTR(err).
4690 */
skb_segment(struct sk_buff * head_skb,netdev_features_t features)4691 struct sk_buff *skb_segment(struct sk_buff *head_skb,
4692 netdev_features_t features)
4693 {
4694 struct sk_buff *segs = NULL;
4695 struct sk_buff *tail = NULL;
4696 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
4697 unsigned int mss = skb_shinfo(head_skb)->gso_size;
4698 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
4699 unsigned int offset = doffset;
4700 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
4701 unsigned int partial_segs = 0;
4702 unsigned int headroom;
4703 unsigned int len = head_skb->len;
4704 struct sk_buff *frag_skb;
4705 skb_frag_t *frag;
4706 __be16 proto;
4707 bool csum, sg;
4708 int err = -ENOMEM;
4709 int i = 0;
4710 int nfrags, pos;
4711
4712 if ((skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY) &&
4713 mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb)) {
4714 struct sk_buff *check_skb;
4715
4716 for (check_skb = list_skb; check_skb; check_skb = check_skb->next) {
4717 if (skb_headlen(check_skb) && !check_skb->head_frag) {
4718 /* gso_size is untrusted, and we have a frag_list with
4719 * a linear non head_frag item.
4720 *
4721 * If head_skb's headlen does not fit requested gso_size,
4722 * it means that the frag_list members do NOT terminate
4723 * on exact gso_size boundaries. Hence we cannot perform
4724 * skb_frag_t page sharing. Therefore we must fallback to
4725 * copying the frag_list skbs; we do so by disabling SG.
4726 */
4727 features &= ~NETIF_F_SG;
4728 break;
4729 }
4730 }
4731 }
4732
4733 __skb_push(head_skb, doffset);
4734 proto = skb_network_protocol(head_skb, NULL);
4735 if (unlikely(!proto))
4736 return ERR_PTR(-EINVAL);
4737
4738 sg = !!(features & NETIF_F_SG);
4739 csum = !!can_checksum_protocol(features, proto);
4740
4741 if (sg && csum && (mss != GSO_BY_FRAGS)) {
4742 if (!(features & NETIF_F_GSO_PARTIAL)) {
4743 struct sk_buff *iter;
4744 unsigned int frag_len;
4745
4746 if (!list_skb ||
4747 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
4748 goto normal;
4749
4750 /* If we get here then all the required
4751 * GSO features except frag_list are supported.
4752 * Try to split the SKB to multiple GSO SKBs
4753 * with no frag_list.
4754 * Currently we can do that only when the buffers don't
4755 * have a linear part and all the buffers except
4756 * the last are of the same length.
4757 */
4758 frag_len = list_skb->len;
4759 skb_walk_frags(head_skb, iter) {
4760 if (frag_len != iter->len && iter->next)
4761 goto normal;
4762 if (skb_headlen(iter) && !iter->head_frag)
4763 goto normal;
4764
4765 len -= iter->len;
4766 }
4767
4768 if (len != frag_len)
4769 goto normal;
4770 }
4771
4772 /* GSO partial only requires that we trim off any excess that
4773 * doesn't fit into an MSS sized block, so take care of that
4774 * now.
4775 * Cap len to not accidentally hit GSO_BY_FRAGS.
4776 */
4777 partial_segs = min(len, GSO_BY_FRAGS - 1) / mss;
4778 if (partial_segs > 1)
4779 mss *= partial_segs;
4780 else
4781 partial_segs = 0;
4782 }
4783
4784 normal:
4785 headroom = skb_headroom(head_skb);
4786 pos = skb_headlen(head_skb);
4787
4788 if (skb_orphan_frags(head_skb, GFP_ATOMIC))
4789 return ERR_PTR(-ENOMEM);
4790
4791 nfrags = skb_shinfo(head_skb)->nr_frags;
4792 frag = skb_shinfo(head_skb)->frags;
4793 frag_skb = head_skb;
4794
4795 do {
4796 struct sk_buff *nskb;
4797 skb_frag_t *nskb_frag;
4798 int hsize;
4799 int size;
4800
4801 if (unlikely(mss == GSO_BY_FRAGS)) {
4802 len = list_skb->len;
4803 } else {
4804 len = head_skb->len - offset;
4805 if (len > mss)
4806 len = mss;
4807 }
4808
4809 hsize = skb_headlen(head_skb) - offset;
4810
4811 if (hsize <= 0 && i >= nfrags && skb_headlen(list_skb) &&
4812 (skb_headlen(list_skb) == len || sg)) {
4813 BUG_ON(skb_headlen(list_skb) > len);
4814
4815 nskb = skb_clone(list_skb, GFP_ATOMIC);
4816 if (unlikely(!nskb))
4817 goto err;
4818
4819 i = 0;
4820 nfrags = skb_shinfo(list_skb)->nr_frags;
4821 frag = skb_shinfo(list_skb)->frags;
4822 frag_skb = list_skb;
4823 pos += skb_headlen(list_skb);
4824
4825 while (pos < offset + len) {
4826 BUG_ON(i >= nfrags);
4827
4828 size = skb_frag_size(frag);
4829 if (pos + size > offset + len)
4830 break;
4831
4832 i++;
4833 pos += size;
4834 frag++;
4835 }
4836
4837 list_skb = list_skb->next;
4838
4839 if (unlikely(pskb_trim(nskb, len))) {
4840 kfree_skb(nskb);
4841 goto err;
4842 }
4843
4844 hsize = skb_end_offset(nskb);
4845 if (skb_cow_head(nskb, doffset + headroom)) {
4846 kfree_skb(nskb);
4847 goto err;
4848 }
4849
4850 nskb->truesize += skb_end_offset(nskb) - hsize;
4851 skb_release_head_state(nskb);
4852 __skb_push(nskb, doffset);
4853 } else {
4854 if (hsize < 0)
4855 hsize = 0;
4856 if (hsize > len || !sg)
4857 hsize = len;
4858
4859 nskb = __alloc_skb(hsize + doffset + headroom,
4860 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
4861 NUMA_NO_NODE);
4862
4863 if (unlikely(!nskb))
4864 goto err;
4865
4866 skb_reserve(nskb, headroom);
4867 __skb_put(nskb, doffset);
4868 }
4869
4870 if (segs)
4871 tail->next = nskb;
4872 else
4873 segs = nskb;
4874 tail = nskb;
4875
4876 __copy_skb_header(nskb, head_skb);
4877
4878 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
4879 skb_reset_mac_len(nskb);
4880
4881 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
4882 nskb->data - tnl_hlen,
4883 doffset + tnl_hlen);
4884
4885 if (nskb->len == len + doffset)
4886 goto perform_csum_check;
4887
4888 if (!sg) {
4889 if (!csum) {
4890 if (!nskb->remcsum_offload)
4891 nskb->ip_summed = CHECKSUM_NONE;
4892 SKB_GSO_CB(nskb)->csum =
4893 skb_copy_and_csum_bits(head_skb, offset,
4894 skb_put(nskb,
4895 len),
4896 len);
4897 SKB_GSO_CB(nskb)->csum_start =
4898 skb_headroom(nskb) + doffset;
4899 } else {
4900 if (skb_copy_bits(head_skb, offset, skb_put(nskb, len), len))
4901 goto err;
4902 }
4903 continue;
4904 }
4905
4906 nskb_frag = skb_shinfo(nskb)->frags;
4907
4908 skb_copy_from_linear_data_offset(head_skb, offset,
4909 skb_put(nskb, hsize), hsize);
4910
4911 skb_shinfo(nskb)->flags |= skb_shinfo(head_skb)->flags &
4912 SKBFL_SHARED_FRAG;
4913
4914 if (skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
4915 goto err;
4916
4917 while (pos < offset + len) {
4918 if (i >= nfrags) {
4919 if (skb_orphan_frags(list_skb, GFP_ATOMIC) ||
4920 skb_zerocopy_clone(nskb, list_skb,
4921 GFP_ATOMIC))
4922 goto err;
4923
4924 i = 0;
4925 nfrags = skb_shinfo(list_skb)->nr_frags;
4926 frag = skb_shinfo(list_skb)->frags;
4927 frag_skb = list_skb;
4928 if (!skb_headlen(list_skb)) {
4929 BUG_ON(!nfrags);
4930 } else {
4931 BUG_ON(!list_skb->head_frag);
4932
4933 /* to make room for head_frag. */
4934 i--;
4935 frag--;
4936 }
4937
4938 list_skb = list_skb->next;
4939 }
4940
4941 if (unlikely(skb_shinfo(nskb)->nr_frags >=
4942 MAX_SKB_FRAGS)) {
4943 net_warn_ratelimited(
4944 "skb_segment: too many frags: %u %u\n",
4945 pos, mss);
4946 err = -EINVAL;
4947 goto err;
4948 }
4949
4950 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
4951 __skb_frag_ref(nskb_frag);
4952 size = skb_frag_size(nskb_frag);
4953
4954 if (pos < offset) {
4955 skb_frag_off_add(nskb_frag, offset - pos);
4956 skb_frag_size_sub(nskb_frag, offset - pos);
4957 }
4958
4959 skb_shinfo(nskb)->nr_frags++;
4960
4961 if (pos + size <= offset + len) {
4962 i++;
4963 frag++;
4964 pos += size;
4965 } else {
4966 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4967 goto skip_fraglist;
4968 }
4969
4970 nskb_frag++;
4971 }
4972
4973 skip_fraglist:
4974 nskb->data_len = len - hsize;
4975 nskb->len += nskb->data_len;
4976 nskb->truesize += nskb->data_len;
4977
4978 perform_csum_check:
4979 if (!csum) {
4980 if (skb_has_shared_frag(nskb) &&
4981 __skb_linearize(nskb))
4982 goto err;
4983
4984 if (!nskb->remcsum_offload)
4985 nskb->ip_summed = CHECKSUM_NONE;
4986 SKB_GSO_CB(nskb)->csum =
4987 skb_checksum(nskb, doffset,
4988 nskb->len - doffset, 0);
4989 SKB_GSO_CB(nskb)->csum_start =
4990 skb_headroom(nskb) + doffset;
4991 }
4992 } while ((offset += len) < head_skb->len);
4993
4994 /* Some callers want to get the end of the list.
4995 * Put it in segs->prev to avoid walking the list.
4996 * (see validate_xmit_skb_list() for example)
4997 */
4998 segs->prev = tail;
4999
5000 if (partial_segs) {
5001 struct sk_buff *iter;
5002 int type = skb_shinfo(head_skb)->gso_type;
5003 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
5004
5005 /* Update type to add partial and then remove dodgy if set */
5006 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
5007 type &= ~SKB_GSO_DODGY;
5008
5009 /* Update GSO info and prepare to start updating headers on
5010 * our way back down the stack of protocols.
5011 */
5012 for (iter = segs; iter; iter = iter->next) {
5013 skb_shinfo(iter)->gso_size = gso_size;
5014 skb_shinfo(iter)->gso_segs = partial_segs;
5015 skb_shinfo(iter)->gso_type = type;
5016 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
5017 }
5018
5019 if (tail->len - doffset <= gso_size)
5020 skb_shinfo(tail)->gso_size = 0;
5021 else if (tail != segs)
5022 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
5023 }
5024
5025 /* Following permits correct backpressure, for protocols
5026 * using skb_set_owner_w().
5027 * Idea is to tranfert ownership from head_skb to last segment.
5028 */
5029 if (head_skb->destructor == sock_wfree) {
5030 swap(tail->truesize, head_skb->truesize);
5031 swap(tail->destructor, head_skb->destructor);
5032 swap(tail->sk, head_skb->sk);
5033 }
5034 return segs;
5035
5036 err:
5037 kfree_skb_list(segs);
5038 return ERR_PTR(err);
5039 }
5040 EXPORT_SYMBOL_GPL(skb_segment);
5041
5042 #ifdef CONFIG_SKB_EXTENSIONS
5043 #define SKB_EXT_ALIGN_VALUE 8
5044 #define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
5045
5046 static const u8 skb_ext_type_len[] = {
5047 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
5048 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
5049 #endif
5050 #ifdef CONFIG_XFRM
5051 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
5052 #endif
5053 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
5054 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
5055 #endif
5056 #if IS_ENABLED(CONFIG_MPTCP)
5057 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
5058 #endif
5059 #if IS_ENABLED(CONFIG_MCTP_FLOWS)
5060 [SKB_EXT_MCTP] = SKB_EXT_CHUNKSIZEOF(struct mctp_flow),
5061 #endif
5062 };
5063
skb_ext_total_length(void)5064 static __always_inline unsigned int skb_ext_total_length(void)
5065 {
5066 unsigned int l = SKB_EXT_CHUNKSIZEOF(struct skb_ext);
5067 int i;
5068
5069 for (i = 0; i < ARRAY_SIZE(skb_ext_type_len); i++)
5070 l += skb_ext_type_len[i];
5071
5072 return l;
5073 }
5074
skb_extensions_init(void)5075 static void skb_extensions_init(void)
5076 {
5077 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
5078 #if !IS_ENABLED(CONFIG_KCOV_INSTRUMENT_ALL)
5079 BUILD_BUG_ON(skb_ext_total_length() > 255);
5080 #endif
5081
5082 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
5083 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
5084 0,
5085 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5086 NULL);
5087 }
5088 #else
skb_extensions_init(void)5089 static void skb_extensions_init(void) {}
5090 #endif
5091
5092 /* The SKB kmem_cache slab is critical for network performance. Never
5093 * merge/alias the slab with similar sized objects. This avoids fragmentation
5094 * that hurts performance of kmem_cache_{alloc,free}_bulk APIs.
5095 */
5096 #ifndef CONFIG_SLUB_TINY
5097 #define FLAG_SKB_NO_MERGE SLAB_NO_MERGE
5098 #else /* CONFIG_SLUB_TINY - simple loop in kmem_cache_alloc_bulk */
5099 #define FLAG_SKB_NO_MERGE 0
5100 #endif
5101
skb_init(void)5102 void __init skb_init(void)
5103 {
5104 net_hotdata.skbuff_cache = kmem_cache_create_usercopy("skbuff_head_cache",
5105 sizeof(struct sk_buff),
5106 0,
5107 SLAB_HWCACHE_ALIGN|SLAB_PANIC|
5108 FLAG_SKB_NO_MERGE,
5109 offsetof(struct sk_buff, cb),
5110 sizeof_field(struct sk_buff, cb),
5111 NULL);
5112 net_hotdata.skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
5113 sizeof(struct sk_buff_fclones),
5114 0,
5115 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
5116 NULL);
5117 /* usercopy should only access first SKB_SMALL_HEAD_HEADROOM bytes.
5118 * struct skb_shared_info is located at the end of skb->head,
5119 * and should not be copied to/from user.
5120 */
5121 net_hotdata.skb_small_head_cache = kmem_cache_create_usercopy("skbuff_small_head",
5122 SKB_SMALL_HEAD_CACHE_SIZE,
5123 0,
5124 SLAB_HWCACHE_ALIGN | SLAB_PANIC,
5125 0,
5126 SKB_SMALL_HEAD_HEADROOM,
5127 NULL);
5128 skb_extensions_init();
5129 }
5130
5131 static int
__skb_to_sgvec(struct sk_buff * skb,struct scatterlist * sg,int offset,int len,unsigned int recursion_level)5132 __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
5133 unsigned int recursion_level)
5134 {
5135 int start = skb_headlen(skb);
5136 int i, copy = start - offset;
5137 struct sk_buff *frag_iter;
5138 int elt = 0;
5139
5140 if (unlikely(recursion_level >= 24))
5141 return -EMSGSIZE;
5142
5143 if (copy > 0) {
5144 if (copy > len)
5145 copy = len;
5146 sg_set_buf(sg, skb->data + offset, copy);
5147 elt++;
5148 if ((len -= copy) == 0)
5149 return elt;
5150 offset += copy;
5151 }
5152
5153 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
5154 int end;
5155
5156 WARN_ON(start > offset + len);
5157
5158 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
5159 if ((copy = end - offset) > 0) {
5160 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
5161 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5162 return -EMSGSIZE;
5163
5164 if (copy > len)
5165 copy = len;
5166 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
5167 skb_frag_off(frag) + offset - start);
5168 elt++;
5169 if (!(len -= copy))
5170 return elt;
5171 offset += copy;
5172 }
5173 start = end;
5174 }
5175
5176 skb_walk_frags(skb, frag_iter) {
5177 int end, ret;
5178
5179 WARN_ON(start > offset + len);
5180
5181 end = start + frag_iter->len;
5182 if ((copy = end - offset) > 0) {
5183 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
5184 return -EMSGSIZE;
5185
5186 if (copy > len)
5187 copy = len;
5188 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
5189 copy, recursion_level + 1);
5190 if (unlikely(ret < 0))
5191 return ret;
5192 elt += ret;
5193 if ((len -= copy) == 0)
5194 return elt;
5195 offset += copy;
5196 }
5197 start = end;
5198 }
5199 BUG_ON(len);
5200 return elt;
5201 }
5202
5203 /**
5204 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
5205 * @skb: Socket buffer containing the buffers to be mapped
5206 * @sg: The scatter-gather list to map into
5207 * @offset: The offset into the buffer's contents to start mapping
5208 * @len: Length of buffer space to be mapped
5209 *
5210 * Fill the specified scatter-gather list with mappings/pointers into a
5211 * region of the buffer space attached to a socket buffer. Returns either
5212 * the number of scatterlist items used, or -EMSGSIZE if the contents
5213 * could not fit.
5214 */
skb_to_sgvec(struct sk_buff * skb,struct scatterlist * sg,int offset,int len)5215 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
5216 {
5217 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
5218
5219 if (nsg <= 0)
5220 return nsg;
5221
5222 sg_mark_end(&sg[nsg - 1]);
5223
5224 return nsg;
5225 }
5226 EXPORT_SYMBOL_GPL(skb_to_sgvec);
5227
5228 /* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
5229 * sglist without mark the sg which contain last skb data as the end.
5230 * So the caller can mannipulate sg list as will when padding new data after
5231 * the first call without calling sg_unmark_end to expend sg list.
5232 *
5233 * Scenario to use skb_to_sgvec_nomark:
5234 * 1. sg_init_table
5235 * 2. skb_to_sgvec_nomark(payload1)
5236 * 3. skb_to_sgvec_nomark(payload2)
5237 *
5238 * This is equivalent to:
5239 * 1. sg_init_table
5240 * 2. skb_to_sgvec(payload1)
5241 * 3. sg_unmark_end
5242 * 4. skb_to_sgvec(payload2)
5243 *
5244 * When mapping multiple payload conditionally, skb_to_sgvec_nomark
5245 * is more preferable.
5246 */
skb_to_sgvec_nomark(struct sk_buff * skb,struct scatterlist * sg,int offset,int len)5247 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
5248 int offset, int len)
5249 {
5250 return __skb_to_sgvec(skb, sg, offset, len, 0);
5251 }
5252 EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
5253
5254
5255
5256 /**
5257 * skb_cow_data - Check that a socket buffer's data buffers are writable
5258 * @skb: The socket buffer to check.
5259 * @tailbits: Amount of trailing space to be added
5260 * @trailer: Returned pointer to the skb where the @tailbits space begins
5261 *
5262 * Make sure that the data buffers attached to a socket buffer are
5263 * writable. If they are not, private copies are made of the data buffers
5264 * and the socket buffer is set to use these instead.
5265 *
5266 * If @tailbits is given, make sure that there is space to write @tailbits
5267 * bytes of data beyond current end of socket buffer. @trailer will be
5268 * set to point to the skb in which this space begins.
5269 *
5270 * The number of scatterlist elements required to completely map the
5271 * COW'd and extended socket buffer will be returned.
5272 */
skb_cow_data(struct sk_buff * skb,int tailbits,struct sk_buff ** trailer)5273 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
5274 {
5275 int copyflag;
5276 int elt;
5277 struct sk_buff *skb1, **skb_p;
5278
5279 /* If skb is cloned or its head is paged, reallocate
5280 * head pulling out all the pages (pages are considered not writable
5281 * at the moment even if they are anonymous).
5282 */
5283 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
5284 !__pskb_pull_tail(skb, __skb_pagelen(skb)))
5285 return -ENOMEM;
5286
5287 /* Easy case. Most of packets will go this way. */
5288 if (!skb_has_frag_list(skb)) {
5289 /* A little of trouble, not enough of space for trailer.
5290 * This should not happen, when stack is tuned to generate
5291 * good frames. OK, on miss we reallocate and reserve even more
5292 * space, 128 bytes is fair. */
5293
5294 if (skb_tailroom(skb) < tailbits &&
5295 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
5296 return -ENOMEM;
5297
5298 /* Voila! */
5299 *trailer = skb;
5300 return 1;
5301 }
5302
5303 /* Misery. We are in troubles, going to mincer fragments... */
5304
5305 elt = 1;
5306 skb_p = &skb_shinfo(skb)->frag_list;
5307 copyflag = 0;
5308
5309 while ((skb1 = *skb_p) != NULL) {
5310 int ntail = 0;
5311
5312 /* The fragment is partially pulled by someone,
5313 * this can happen on input. Copy it and everything
5314 * after it. */
5315
5316 if (skb_shared(skb1))
5317 copyflag = 1;
5318
5319 /* If the skb is the last, worry about trailer. */
5320
5321 if (skb1->next == NULL && tailbits) {
5322 if (skb_shinfo(skb1)->nr_frags ||
5323 skb_has_frag_list(skb1) ||
5324 skb_tailroom(skb1) < tailbits)
5325 ntail = tailbits + 128;
5326 }
5327
5328 if (copyflag ||
5329 skb_cloned(skb1) ||
5330 ntail ||
5331 skb_shinfo(skb1)->nr_frags ||
5332 skb_has_frag_list(skb1)) {
5333 struct sk_buff *skb2;
5334
5335 /* Fuck, we are miserable poor guys... */
5336 if (ntail == 0)
5337 skb2 = skb_copy(skb1, GFP_ATOMIC);
5338 else
5339 skb2 = skb_copy_expand(skb1,
5340 skb_headroom(skb1),
5341 ntail,
5342 GFP_ATOMIC);
5343 if (unlikely(skb2 == NULL))
5344 return -ENOMEM;
5345
5346 if (skb1->sk)
5347 skb_set_owner_w(skb2, skb1->sk);
5348
5349 /* Looking around. Are we still alive?
5350 * OK, link new skb, drop old one */
5351
5352 skb2->next = skb1->next;
5353 *skb_p = skb2;
5354 kfree_skb(skb1);
5355 skb1 = skb2;
5356 }
5357 elt++;
5358 *trailer = skb1;
5359 skb_p = &skb1->next;
5360 }
5361
5362 return elt;
5363 }
5364 EXPORT_SYMBOL_GPL(skb_cow_data);
5365
sock_rmem_free(struct sk_buff * skb)5366 static void sock_rmem_free(struct sk_buff *skb)
5367 {
5368 struct sock *sk = skb->sk;
5369
5370 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
5371 }
5372
skb_set_err_queue(struct sk_buff * skb)5373 static void skb_set_err_queue(struct sk_buff *skb)
5374 {
5375 /* pkt_type of skbs received on local sockets is never PACKET_OUTGOING.
5376 * So, it is safe to (mis)use it to mark skbs on the error queue.
5377 */
5378 skb->pkt_type = PACKET_OUTGOING;
5379 BUILD_BUG_ON(PACKET_OUTGOING == 0);
5380 }
5381
5382 /*
5383 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
5384 */
sock_queue_err_skb(struct sock * sk,struct sk_buff * skb)5385 int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
5386 {
5387 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
5388 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
5389 return -ENOMEM;
5390
5391 skb_orphan(skb);
5392 skb->sk = sk;
5393 skb->destructor = sock_rmem_free;
5394 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
5395 skb_set_err_queue(skb);
5396
5397 /* before exiting rcu section, make sure dst is refcounted */
5398 skb_dst_force(skb);
5399
5400 skb_queue_tail(&sk->sk_error_queue, skb);
5401 if (!sock_flag(sk, SOCK_DEAD))
5402 sk_error_report(sk);
5403 return 0;
5404 }
5405 EXPORT_SYMBOL(sock_queue_err_skb);
5406
is_icmp_err_skb(const struct sk_buff * skb)5407 static bool is_icmp_err_skb(const struct sk_buff *skb)
5408 {
5409 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
5410 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
5411 }
5412
sock_dequeue_err_skb(struct sock * sk)5413 struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
5414 {
5415 struct sk_buff_head *q = &sk->sk_error_queue;
5416 struct sk_buff *skb, *skb_next = NULL;
5417 bool icmp_next = false;
5418 unsigned long flags;
5419
5420 if (skb_queue_empty_lockless(q))
5421 return NULL;
5422
5423 spin_lock_irqsave(&q->lock, flags);
5424 skb = __skb_dequeue(q);
5425 if (skb && (skb_next = skb_peek(q))) {
5426 icmp_next = is_icmp_err_skb(skb_next);
5427 if (icmp_next)
5428 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
5429 }
5430 spin_unlock_irqrestore(&q->lock, flags);
5431
5432 if (is_icmp_err_skb(skb) && !icmp_next)
5433 sk->sk_err = 0;
5434
5435 if (skb_next)
5436 sk_error_report(sk);
5437
5438 return skb;
5439 }
5440 EXPORT_SYMBOL(sock_dequeue_err_skb);
5441
5442 /**
5443 * skb_clone_sk - create clone of skb, and take reference to socket
5444 * @skb: the skb to clone
5445 *
5446 * This function creates a clone of a buffer that holds a reference on
5447 * sk_refcnt. Buffers created via this function are meant to be
5448 * returned using sock_queue_err_skb, or free via kfree_skb.
5449 *
5450 * When passing buffers allocated with this function to sock_queue_err_skb
5451 * it is necessary to wrap the call with sock_hold/sock_put in order to
5452 * prevent the socket from being released prior to being enqueued on
5453 * the sk_error_queue.
5454 */
skb_clone_sk(struct sk_buff * skb)5455 struct sk_buff *skb_clone_sk(struct sk_buff *skb)
5456 {
5457 struct sock *sk = skb->sk;
5458 struct sk_buff *clone;
5459
5460 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
5461 return NULL;
5462
5463 clone = skb_clone(skb, GFP_ATOMIC);
5464 if (!clone) {
5465 sock_put(sk);
5466 return NULL;
5467 }
5468
5469 clone->sk = sk;
5470 clone->destructor = sock_efree;
5471
5472 return clone;
5473 }
5474 EXPORT_SYMBOL(skb_clone_sk);
5475
__skb_complete_tx_timestamp(struct sk_buff * skb,struct sock * sk,int tstype,bool opt_stats)5476 static void __skb_complete_tx_timestamp(struct sk_buff *skb,
5477 struct sock *sk,
5478 int tstype,
5479 bool opt_stats)
5480 {
5481 struct sock_exterr_skb *serr;
5482 int err;
5483
5484 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
5485
5486 serr = SKB_EXT_ERR(skb);
5487 memset(serr, 0, sizeof(*serr));
5488 serr->ee.ee_errno = ENOMSG;
5489 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
5490 serr->ee.ee_info = tstype;
5491 serr->opt_stats = opt_stats;
5492 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
5493 if (READ_ONCE(sk->sk_tsflags) & SOF_TIMESTAMPING_OPT_ID) {
5494 serr->ee.ee_data = skb_shinfo(skb)->tskey;
5495 if (sk_is_tcp(sk))
5496 serr->ee.ee_data -= atomic_read(&sk->sk_tskey);
5497 }
5498
5499 err = sock_queue_err_skb(sk, skb);
5500
5501 if (err)
5502 kfree_skb(skb);
5503 }
5504
skb_may_tx_timestamp(struct sock * sk,bool tsonly)5505 static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
5506 {
5507 bool ret;
5508
5509 if (likely(tsonly || READ_ONCE(sock_net(sk)->core.sysctl_tstamp_allow_data)))
5510 return true;
5511
5512 read_lock_bh(&sk->sk_callback_lock);
5513 ret = sk->sk_socket && sk->sk_socket->file &&
5514 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
5515 read_unlock_bh(&sk->sk_callback_lock);
5516 return ret;
5517 }
5518
skb_complete_tx_timestamp(struct sk_buff * skb,struct skb_shared_hwtstamps * hwtstamps)5519 void skb_complete_tx_timestamp(struct sk_buff *skb,
5520 struct skb_shared_hwtstamps *hwtstamps)
5521 {
5522 struct sock *sk = skb->sk;
5523
5524 if (!skb_may_tx_timestamp(sk, false))
5525 goto err;
5526
5527 /* Take a reference to prevent skb_orphan() from freeing the socket,
5528 * but only if the socket refcount is not zero.
5529 */
5530 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5531 *skb_hwtstamps(skb) = *hwtstamps;
5532 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
5533 sock_put(sk);
5534 return;
5535 }
5536
5537 err:
5538 kfree_skb(skb);
5539 }
5540 EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
5541
__skb_tstamp_tx(struct sk_buff * orig_skb,const struct sk_buff * ack_skb,struct skb_shared_hwtstamps * hwtstamps,struct sock * sk,int tstype)5542 void __skb_tstamp_tx(struct sk_buff *orig_skb,
5543 const struct sk_buff *ack_skb,
5544 struct skb_shared_hwtstamps *hwtstamps,
5545 struct sock *sk, int tstype)
5546 {
5547 struct sk_buff *skb;
5548 bool tsonly, opt_stats = false;
5549 u32 tsflags;
5550
5551 if (!sk)
5552 return;
5553
5554 tsflags = READ_ONCE(sk->sk_tsflags);
5555 if (!hwtstamps && !(tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
5556 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
5557 return;
5558
5559 tsonly = tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
5560 if (!skb_may_tx_timestamp(sk, tsonly))
5561 return;
5562
5563 if (tsonly) {
5564 #ifdef CONFIG_INET
5565 if ((tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
5566 sk_is_tcp(sk)) {
5567 skb = tcp_get_timestamping_opt_stats(sk, orig_skb,
5568 ack_skb);
5569 opt_stats = true;
5570 } else
5571 #endif
5572 skb = alloc_skb(0, GFP_ATOMIC);
5573 } else {
5574 skb = skb_clone(orig_skb, GFP_ATOMIC);
5575
5576 if (skb_orphan_frags_rx(skb, GFP_ATOMIC)) {
5577 kfree_skb(skb);
5578 return;
5579 }
5580 }
5581 if (!skb)
5582 return;
5583
5584 if (tsonly) {
5585 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
5586 SKBTX_ANY_TSTAMP;
5587 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
5588 }
5589
5590 if (hwtstamps)
5591 *skb_hwtstamps(skb) = *hwtstamps;
5592 else
5593 __net_timestamp(skb);
5594
5595 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
5596 }
5597 EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
5598
skb_tstamp_tx(struct sk_buff * orig_skb,struct skb_shared_hwtstamps * hwtstamps)5599 void skb_tstamp_tx(struct sk_buff *orig_skb,
5600 struct skb_shared_hwtstamps *hwtstamps)
5601 {
5602 return __skb_tstamp_tx(orig_skb, NULL, hwtstamps, orig_skb->sk,
5603 SCM_TSTAMP_SND);
5604 }
5605 EXPORT_SYMBOL_GPL(skb_tstamp_tx);
5606
5607 #ifdef CONFIG_WIRELESS
skb_complete_wifi_ack(struct sk_buff * skb,bool acked)5608 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
5609 {
5610 struct sock *sk = skb->sk;
5611 struct sock_exterr_skb *serr;
5612 int err = 1;
5613
5614 skb->wifi_acked_valid = 1;
5615 skb->wifi_acked = acked;
5616
5617 serr = SKB_EXT_ERR(skb);
5618 memset(serr, 0, sizeof(*serr));
5619 serr->ee.ee_errno = ENOMSG;
5620 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
5621
5622 /* Take a reference to prevent skb_orphan() from freeing the socket,
5623 * but only if the socket refcount is not zero.
5624 */
5625 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
5626 err = sock_queue_err_skb(sk, skb);
5627 sock_put(sk);
5628 }
5629 if (err)
5630 kfree_skb(skb);
5631 }
5632 EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
5633 #endif /* CONFIG_WIRELESS */
5634
5635 /**
5636 * skb_partial_csum_set - set up and verify partial csum values for packet
5637 * @skb: the skb to set
5638 * @start: the number of bytes after skb->data to start checksumming.
5639 * @off: the offset from start to place the checksum.
5640 *
5641 * For untrusted partially-checksummed packets, we need to make sure the values
5642 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
5643 *
5644 * This function checks and sets those values and skb->ip_summed: if this
5645 * returns false you should drop the packet.
5646 */
skb_partial_csum_set(struct sk_buff * skb,u16 start,u16 off)5647 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
5648 {
5649 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
5650 u32 csum_start = skb_headroom(skb) + (u32)start;
5651
5652 if (unlikely(csum_start >= U16_MAX || csum_end > skb_headlen(skb))) {
5653 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
5654 start, off, skb_headroom(skb), skb_headlen(skb));
5655 return false;
5656 }
5657 skb->ip_summed = CHECKSUM_PARTIAL;
5658 skb->csum_start = csum_start;
5659 skb->csum_offset = off;
5660 skb->transport_header = csum_start;
5661 return true;
5662 }
5663 EXPORT_SYMBOL_GPL(skb_partial_csum_set);
5664
skb_maybe_pull_tail(struct sk_buff * skb,unsigned int len,unsigned int max)5665 static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
5666 unsigned int max)
5667 {
5668 if (skb_headlen(skb) >= len)
5669 return 0;
5670
5671 /* If we need to pullup then pullup to the max, so we
5672 * won't need to do it again.
5673 */
5674 if (max > skb->len)
5675 max = skb->len;
5676
5677 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
5678 return -ENOMEM;
5679
5680 if (skb_headlen(skb) < len)
5681 return -EPROTO;
5682
5683 return 0;
5684 }
5685
5686 #define MAX_TCP_HDR_LEN (15 * 4)
5687
skb_checksum_setup_ip(struct sk_buff * skb,typeof(IPPROTO_IP) proto,unsigned int off)5688 static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
5689 typeof(IPPROTO_IP) proto,
5690 unsigned int off)
5691 {
5692 int err;
5693
5694 switch (proto) {
5695 case IPPROTO_TCP:
5696 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
5697 off + MAX_TCP_HDR_LEN);
5698 if (!err && !skb_partial_csum_set(skb, off,
5699 offsetof(struct tcphdr,
5700 check)))
5701 err = -EPROTO;
5702 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
5703
5704 case IPPROTO_UDP:
5705 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
5706 off + sizeof(struct udphdr));
5707 if (!err && !skb_partial_csum_set(skb, off,
5708 offsetof(struct udphdr,
5709 check)))
5710 err = -EPROTO;
5711 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
5712 }
5713
5714 return ERR_PTR(-EPROTO);
5715 }
5716
5717 /* This value should be large enough to cover a tagged ethernet header plus
5718 * maximally sized IP and TCP or UDP headers.
5719 */
5720 #define MAX_IP_HDR_LEN 128
5721
skb_checksum_setup_ipv4(struct sk_buff * skb,bool recalculate)5722 static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
5723 {
5724 unsigned int off;
5725 bool fragment;
5726 __sum16 *csum;
5727 int err;
5728
5729 fragment = false;
5730
5731 err = skb_maybe_pull_tail(skb,
5732 sizeof(struct iphdr),
5733 MAX_IP_HDR_LEN);
5734 if (err < 0)
5735 goto out;
5736
5737 if (ip_is_fragment(ip_hdr(skb)))
5738 fragment = true;
5739
5740 off = ip_hdrlen(skb);
5741
5742 err = -EPROTO;
5743
5744 if (fragment)
5745 goto out;
5746
5747 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
5748 if (IS_ERR(csum))
5749 return PTR_ERR(csum);
5750
5751 if (recalculate)
5752 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
5753 ip_hdr(skb)->daddr,
5754 skb->len - off,
5755 ip_hdr(skb)->protocol, 0);
5756 err = 0;
5757
5758 out:
5759 return err;
5760 }
5761
5762 /* This value should be large enough to cover a tagged ethernet header plus
5763 * an IPv6 header, all options, and a maximal TCP or UDP header.
5764 */
5765 #define MAX_IPV6_HDR_LEN 256
5766
5767 #define OPT_HDR(type, skb, off) \
5768 (type *)(skb_network_header(skb) + (off))
5769
skb_checksum_setup_ipv6(struct sk_buff * skb,bool recalculate)5770 static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
5771 {
5772 int err;
5773 u8 nexthdr;
5774 unsigned int off;
5775 unsigned int len;
5776 bool fragment;
5777 bool done;
5778 __sum16 *csum;
5779
5780 fragment = false;
5781 done = false;
5782
5783 off = sizeof(struct ipv6hdr);
5784
5785 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
5786 if (err < 0)
5787 goto out;
5788
5789 nexthdr = ipv6_hdr(skb)->nexthdr;
5790
5791 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
5792 while (off <= len && !done) {
5793 switch (nexthdr) {
5794 case IPPROTO_DSTOPTS:
5795 case IPPROTO_HOPOPTS:
5796 case IPPROTO_ROUTING: {
5797 struct ipv6_opt_hdr *hp;
5798
5799 err = skb_maybe_pull_tail(skb,
5800 off +
5801 sizeof(struct ipv6_opt_hdr),
5802 MAX_IPV6_HDR_LEN);
5803 if (err < 0)
5804 goto out;
5805
5806 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
5807 nexthdr = hp->nexthdr;
5808 off += ipv6_optlen(hp);
5809 break;
5810 }
5811 case IPPROTO_AH: {
5812 struct ip_auth_hdr *hp;
5813
5814 err = skb_maybe_pull_tail(skb,
5815 off +
5816 sizeof(struct ip_auth_hdr),
5817 MAX_IPV6_HDR_LEN);
5818 if (err < 0)
5819 goto out;
5820
5821 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
5822 nexthdr = hp->nexthdr;
5823 off += ipv6_authlen(hp);
5824 break;
5825 }
5826 case IPPROTO_FRAGMENT: {
5827 struct frag_hdr *hp;
5828
5829 err = skb_maybe_pull_tail(skb,
5830 off +
5831 sizeof(struct frag_hdr),
5832 MAX_IPV6_HDR_LEN);
5833 if (err < 0)
5834 goto out;
5835
5836 hp = OPT_HDR(struct frag_hdr, skb, off);
5837
5838 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
5839 fragment = true;
5840
5841 nexthdr = hp->nexthdr;
5842 off += sizeof(struct frag_hdr);
5843 break;
5844 }
5845 default:
5846 done = true;
5847 break;
5848 }
5849 }
5850
5851 err = -EPROTO;
5852
5853 if (!done || fragment)
5854 goto out;
5855
5856 csum = skb_checksum_setup_ip(skb, nexthdr, off);
5857 if (IS_ERR(csum))
5858 return PTR_ERR(csum);
5859
5860 if (recalculate)
5861 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
5862 &ipv6_hdr(skb)->daddr,
5863 skb->len - off, nexthdr, 0);
5864 err = 0;
5865
5866 out:
5867 return err;
5868 }
5869
5870 /**
5871 * skb_checksum_setup - set up partial checksum offset
5872 * @skb: the skb to set up
5873 * @recalculate: if true the pseudo-header checksum will be recalculated
5874 */
skb_checksum_setup(struct sk_buff * skb,bool recalculate)5875 int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
5876 {
5877 int err;
5878
5879 switch (skb->protocol) {
5880 case htons(ETH_P_IP):
5881 err = skb_checksum_setup_ipv4(skb, recalculate);
5882 break;
5883
5884 case htons(ETH_P_IPV6):
5885 err = skb_checksum_setup_ipv6(skb, recalculate);
5886 break;
5887
5888 default:
5889 err = -EPROTO;
5890 break;
5891 }
5892
5893 return err;
5894 }
5895 EXPORT_SYMBOL(skb_checksum_setup);
5896
5897 /**
5898 * skb_checksum_maybe_trim - maybe trims the given skb
5899 * @skb: the skb to check
5900 * @transport_len: the data length beyond the network header
5901 *
5902 * Checks whether the given skb has data beyond the given transport length.
5903 * If so, returns a cloned skb trimmed to this transport length.
5904 * Otherwise returns the provided skb. Returns NULL in error cases
5905 * (e.g. transport_len exceeds skb length or out-of-memory).
5906 *
5907 * Caller needs to set the skb transport header and free any returned skb if it
5908 * differs from the provided skb.
5909 */
skb_checksum_maybe_trim(struct sk_buff * skb,unsigned int transport_len)5910 static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5911 unsigned int transport_len)
5912 {
5913 struct sk_buff *skb_chk;
5914 unsigned int len = skb_transport_offset(skb) + transport_len;
5915 int ret;
5916
5917 if (skb->len < len)
5918 return NULL;
5919 else if (skb->len == len)
5920 return skb;
5921
5922 skb_chk = skb_clone(skb, GFP_ATOMIC);
5923 if (!skb_chk)
5924 return NULL;
5925
5926 ret = pskb_trim_rcsum(skb_chk, len);
5927 if (ret) {
5928 kfree_skb(skb_chk);
5929 return NULL;
5930 }
5931
5932 return skb_chk;
5933 }
5934
5935 /**
5936 * skb_checksum_trimmed - validate checksum of an skb
5937 * @skb: the skb to check
5938 * @transport_len: the data length beyond the network header
5939 * @skb_chkf: checksum function to use
5940 *
5941 * Applies the given checksum function skb_chkf to the provided skb.
5942 * Returns a checked and maybe trimmed skb. Returns NULL on error.
5943 *
5944 * If the skb has data beyond the given transport length, then a
5945 * trimmed & cloned skb is checked and returned.
5946 *
5947 * Caller needs to set the skb transport header and free any returned skb if it
5948 * differs from the provided skb.
5949 */
skb_checksum_trimmed(struct sk_buff * skb,unsigned int transport_len,__sum16 (* skb_chkf)(struct sk_buff * skb))5950 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5951 unsigned int transport_len,
5952 __sum16(*skb_chkf)(struct sk_buff *skb))
5953 {
5954 struct sk_buff *skb_chk;
5955 unsigned int offset = skb_transport_offset(skb);
5956 __sum16 ret;
5957
5958 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5959 if (!skb_chk)
5960 goto err;
5961
5962 if (!pskb_may_pull(skb_chk, offset))
5963 goto err;
5964
5965 skb_pull_rcsum(skb_chk, offset);
5966 ret = skb_chkf(skb_chk);
5967 skb_push_rcsum(skb_chk, offset);
5968
5969 if (ret)
5970 goto err;
5971
5972 return skb_chk;
5973
5974 err:
5975 if (skb_chk && skb_chk != skb)
5976 kfree_skb(skb_chk);
5977
5978 return NULL;
5979
5980 }
5981 EXPORT_SYMBOL(skb_checksum_trimmed);
5982
__skb_warn_lro_forwarding(const struct sk_buff * skb)5983 void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5984 {
5985 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5986 skb->dev->name);
5987 }
5988 EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5989
kfree_skb_partial(struct sk_buff * skb,bool head_stolen)5990 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5991 {
5992 if (head_stolen) {
5993 skb_release_head_state(skb);
5994 kmem_cache_free(net_hotdata.skbuff_cache, skb);
5995 } else {
5996 __kfree_skb(skb);
5997 }
5998 }
5999 EXPORT_SYMBOL(kfree_skb_partial);
6000
6001 /**
6002 * skb_try_coalesce - try to merge skb to prior one
6003 * @to: prior buffer
6004 * @from: buffer to add
6005 * @fragstolen: pointer to boolean
6006 * @delta_truesize: how much more was allocated than was requested
6007 */
skb_try_coalesce(struct sk_buff * to,struct sk_buff * from,bool * fragstolen,int * delta_truesize)6008 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
6009 bool *fragstolen, int *delta_truesize)
6010 {
6011 struct skb_shared_info *to_shinfo, *from_shinfo;
6012 int i, delta, len = from->len;
6013
6014 *fragstolen = false;
6015
6016 if (skb_cloned(to))
6017 return false;
6018
6019 /* In general, avoid mixing page_pool and non-page_pool allocated
6020 * pages within the same SKB. In theory we could take full
6021 * references if @from is cloned and !@to->pp_recycle but its
6022 * tricky (due to potential race with the clone disappearing) and
6023 * rare, so not worth dealing with.
6024 */
6025 if (to->pp_recycle != from->pp_recycle)
6026 return false;
6027
6028 if (skb_frags_readable(from) != skb_frags_readable(to))
6029 return false;
6030
6031 if (len <= skb_tailroom(to) && skb_frags_readable(from)) {
6032 if (len)
6033 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
6034 *delta_truesize = 0;
6035 return true;
6036 }
6037
6038 to_shinfo = skb_shinfo(to);
6039 from_shinfo = skb_shinfo(from);
6040 if (to_shinfo->frag_list || from_shinfo->frag_list)
6041 return false;
6042 if (skb_zcopy(to) || skb_zcopy(from))
6043 return false;
6044
6045 if (skb_headlen(from) != 0) {
6046 struct page *page;
6047 unsigned int offset;
6048
6049 if (to_shinfo->nr_frags +
6050 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
6051 return false;
6052
6053 if (skb_head_is_locked(from))
6054 return false;
6055
6056 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
6057
6058 page = virt_to_head_page(from->head);
6059 offset = from->data - (unsigned char *)page_address(page);
6060
6061 skb_fill_page_desc(to, to_shinfo->nr_frags,
6062 page, offset, skb_headlen(from));
6063 *fragstolen = true;
6064 } else {
6065 if (to_shinfo->nr_frags +
6066 from_shinfo->nr_frags > MAX_SKB_FRAGS)
6067 return false;
6068
6069 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
6070 }
6071
6072 WARN_ON_ONCE(delta < len);
6073
6074 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
6075 from_shinfo->frags,
6076 from_shinfo->nr_frags * sizeof(skb_frag_t));
6077 to_shinfo->nr_frags += from_shinfo->nr_frags;
6078
6079 if (!skb_cloned(from))
6080 from_shinfo->nr_frags = 0;
6081
6082 /* if the skb is not cloned this does nothing
6083 * since we set nr_frags to 0.
6084 */
6085 if (skb_pp_frag_ref(from)) {
6086 for (i = 0; i < from_shinfo->nr_frags; i++)
6087 __skb_frag_ref(&from_shinfo->frags[i]);
6088 }
6089
6090 to->truesize += delta;
6091 to->len += len;
6092 to->data_len += len;
6093
6094 *delta_truesize = delta;
6095 return true;
6096 }
6097 EXPORT_SYMBOL(skb_try_coalesce);
6098
6099 /**
6100 * skb_scrub_packet - scrub an skb
6101 *
6102 * @skb: buffer to clean
6103 * @xnet: packet is crossing netns
6104 *
6105 * skb_scrub_packet can be used after encapsulating or decapsulating a packet
6106 * into/from a tunnel. Some information have to be cleared during these
6107 * operations.
6108 * skb_scrub_packet can also be used to clean a skb before injecting it in
6109 * another namespace (@xnet == true). We have to clear all information in the
6110 * skb that could impact namespace isolation.
6111 */
skb_scrub_packet(struct sk_buff * skb,bool xnet)6112 void skb_scrub_packet(struct sk_buff *skb, bool xnet)
6113 {
6114 skb->pkt_type = PACKET_HOST;
6115 skb->skb_iif = 0;
6116 skb->ignore_df = 0;
6117 skb_dst_drop(skb);
6118 skb_ext_reset(skb);
6119 nf_reset_ct(skb);
6120 nf_reset_trace(skb);
6121
6122 #ifdef CONFIG_NET_SWITCHDEV
6123 skb->offload_fwd_mark = 0;
6124 skb->offload_l3_fwd_mark = 0;
6125 #endif
6126
6127 if (!xnet)
6128 return;
6129
6130 ipvs_reset(skb);
6131 skb->mark = 0;
6132 skb_clear_tstamp(skb);
6133 }
6134 EXPORT_SYMBOL_GPL(skb_scrub_packet);
6135
skb_reorder_vlan_header(struct sk_buff * skb)6136 static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
6137 {
6138 int mac_len, meta_len;
6139 void *meta;
6140
6141 if (skb_cow(skb, skb_headroom(skb)) < 0) {
6142 kfree_skb(skb);
6143 return NULL;
6144 }
6145
6146 mac_len = skb->data - skb_mac_header(skb);
6147 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
6148 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
6149 mac_len - VLAN_HLEN - ETH_TLEN);
6150 }
6151
6152 meta_len = skb_metadata_len(skb);
6153 if (meta_len) {
6154 meta = skb_metadata_end(skb) - meta_len;
6155 memmove(meta + VLAN_HLEN, meta, meta_len);
6156 }
6157
6158 skb->mac_header += VLAN_HLEN;
6159 return skb;
6160 }
6161
skb_vlan_untag(struct sk_buff * skb)6162 struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
6163 {
6164 struct vlan_hdr *vhdr;
6165 u16 vlan_tci;
6166
6167 if (unlikely(skb_vlan_tag_present(skb))) {
6168 /* vlan_tci is already set-up so leave this for another time */
6169 return skb;
6170 }
6171
6172 skb = skb_share_check(skb, GFP_ATOMIC);
6173 if (unlikely(!skb))
6174 goto err_free;
6175 /* We may access the two bytes after vlan_hdr in vlan_set_encap_proto(). */
6176 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN + sizeof(unsigned short))))
6177 goto err_free;
6178
6179 vhdr = (struct vlan_hdr *)skb->data;
6180 vlan_tci = ntohs(vhdr->h_vlan_TCI);
6181 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
6182
6183 skb_pull_rcsum(skb, VLAN_HLEN);
6184 vlan_set_encap_proto(skb, vhdr);
6185
6186 skb = skb_reorder_vlan_header(skb);
6187 if (unlikely(!skb))
6188 goto err_free;
6189
6190 skb_reset_network_header(skb);
6191 if (!skb_transport_header_was_set(skb))
6192 skb_reset_transport_header(skb);
6193 skb_reset_mac_len(skb);
6194
6195 return skb;
6196
6197 err_free:
6198 kfree_skb(skb);
6199 return NULL;
6200 }
6201 EXPORT_SYMBOL(skb_vlan_untag);
6202
skb_ensure_writable(struct sk_buff * skb,unsigned int write_len)6203 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len)
6204 {
6205 if (!pskb_may_pull(skb, write_len))
6206 return -ENOMEM;
6207
6208 if (!skb_frags_readable(skb))
6209 return -EFAULT;
6210
6211 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
6212 return 0;
6213
6214 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
6215 }
6216 EXPORT_SYMBOL(skb_ensure_writable);
6217
skb_ensure_writable_head_tail(struct sk_buff * skb,struct net_device * dev)6218 int skb_ensure_writable_head_tail(struct sk_buff *skb, struct net_device *dev)
6219 {
6220 int needed_headroom = dev->needed_headroom;
6221 int needed_tailroom = dev->needed_tailroom;
6222
6223 /* For tail taggers, we need to pad short frames ourselves, to ensure
6224 * that the tail tag does not fail at its role of being at the end of
6225 * the packet, once the conduit interface pads the frame. Account for
6226 * that pad length here, and pad later.
6227 */
6228 if (unlikely(needed_tailroom && skb->len < ETH_ZLEN))
6229 needed_tailroom += ETH_ZLEN - skb->len;
6230 /* skb_headroom() returns unsigned int... */
6231 needed_headroom = max_t(int, needed_headroom - skb_headroom(skb), 0);
6232 needed_tailroom = max_t(int, needed_tailroom - skb_tailroom(skb), 0);
6233
6234 if (likely(!needed_headroom && !needed_tailroom && !skb_cloned(skb)))
6235 /* No reallocation needed, yay! */
6236 return 0;
6237
6238 return pskb_expand_head(skb, needed_headroom, needed_tailroom,
6239 GFP_ATOMIC);
6240 }
6241 EXPORT_SYMBOL(skb_ensure_writable_head_tail);
6242
6243 /* remove VLAN header from packet and update csum accordingly.
6244 * expects a non skb_vlan_tag_present skb with a vlan tag payload
6245 */
__skb_vlan_pop(struct sk_buff * skb,u16 * vlan_tci)6246 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
6247 {
6248 int offset = skb->data - skb_mac_header(skb);
6249 int err;
6250
6251 if (WARN_ONCE(offset,
6252 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
6253 offset)) {
6254 return -EINVAL;
6255 }
6256
6257 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
6258 if (unlikely(err))
6259 return err;
6260
6261 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6262
6263 vlan_remove_tag(skb, vlan_tci);
6264
6265 skb->mac_header += VLAN_HLEN;
6266
6267 if (skb_network_offset(skb) < ETH_HLEN)
6268 skb_set_network_header(skb, ETH_HLEN);
6269
6270 skb_reset_mac_len(skb);
6271
6272 return err;
6273 }
6274 EXPORT_SYMBOL(__skb_vlan_pop);
6275
6276 /* Pop a vlan tag either from hwaccel or from payload.
6277 * Expects skb->data at mac header.
6278 */
skb_vlan_pop(struct sk_buff * skb)6279 int skb_vlan_pop(struct sk_buff *skb)
6280 {
6281 u16 vlan_tci;
6282 __be16 vlan_proto;
6283 int err;
6284
6285 if (likely(skb_vlan_tag_present(skb))) {
6286 __vlan_hwaccel_clear_tag(skb);
6287 } else {
6288 if (unlikely(!eth_type_vlan(skb->protocol)))
6289 return 0;
6290
6291 err = __skb_vlan_pop(skb, &vlan_tci);
6292 if (err)
6293 return err;
6294 }
6295 /* move next vlan tag to hw accel tag */
6296 if (likely(!eth_type_vlan(skb->protocol)))
6297 return 0;
6298
6299 vlan_proto = skb->protocol;
6300 err = __skb_vlan_pop(skb, &vlan_tci);
6301 if (unlikely(err))
6302 return err;
6303
6304 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6305 return 0;
6306 }
6307 EXPORT_SYMBOL(skb_vlan_pop);
6308
6309 /* Push a vlan tag either into hwaccel or into payload (if hwaccel tag present).
6310 * Expects skb->data at mac header.
6311 */
skb_vlan_push(struct sk_buff * skb,__be16 vlan_proto,u16 vlan_tci)6312 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
6313 {
6314 if (skb_vlan_tag_present(skb)) {
6315 int offset = skb->data - skb_mac_header(skb);
6316 int err;
6317
6318 if (WARN_ONCE(offset,
6319 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
6320 offset)) {
6321 return -EINVAL;
6322 }
6323
6324 err = __vlan_insert_tag(skb, skb->vlan_proto,
6325 skb_vlan_tag_get(skb));
6326 if (err)
6327 return err;
6328
6329 skb->protocol = skb->vlan_proto;
6330 skb->network_header -= VLAN_HLEN;
6331
6332 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
6333 }
6334 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
6335 return 0;
6336 }
6337 EXPORT_SYMBOL(skb_vlan_push);
6338
6339 /**
6340 * skb_eth_pop() - Drop the Ethernet header at the head of a packet
6341 *
6342 * @skb: Socket buffer to modify
6343 *
6344 * Drop the Ethernet header of @skb.
6345 *
6346 * Expects that skb->data points to the mac header and that no VLAN tags are
6347 * present.
6348 *
6349 * Returns 0 on success, -errno otherwise.
6350 */
skb_eth_pop(struct sk_buff * skb)6351 int skb_eth_pop(struct sk_buff *skb)
6352 {
6353 if (!pskb_may_pull(skb, ETH_HLEN) || skb_vlan_tagged(skb) ||
6354 skb_network_offset(skb) < ETH_HLEN)
6355 return -EPROTO;
6356
6357 skb_pull_rcsum(skb, ETH_HLEN);
6358 skb_reset_mac_header(skb);
6359 skb_reset_mac_len(skb);
6360
6361 return 0;
6362 }
6363 EXPORT_SYMBOL(skb_eth_pop);
6364
6365 /**
6366 * skb_eth_push() - Add a new Ethernet header at the head of a packet
6367 *
6368 * @skb: Socket buffer to modify
6369 * @dst: Destination MAC address of the new header
6370 * @src: Source MAC address of the new header
6371 *
6372 * Prepend @skb with a new Ethernet header.
6373 *
6374 * Expects that skb->data points to the mac header, which must be empty.
6375 *
6376 * Returns 0 on success, -errno otherwise.
6377 */
skb_eth_push(struct sk_buff * skb,const unsigned char * dst,const unsigned char * src)6378 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst,
6379 const unsigned char *src)
6380 {
6381 struct ethhdr *eth;
6382 int err;
6383
6384 if (skb_network_offset(skb) || skb_vlan_tag_present(skb))
6385 return -EPROTO;
6386
6387 err = skb_cow_head(skb, sizeof(*eth));
6388 if (err < 0)
6389 return err;
6390
6391 skb_push(skb, sizeof(*eth));
6392 skb_reset_mac_header(skb);
6393 skb_reset_mac_len(skb);
6394
6395 eth = eth_hdr(skb);
6396 ether_addr_copy(eth->h_dest, dst);
6397 ether_addr_copy(eth->h_source, src);
6398 eth->h_proto = skb->protocol;
6399
6400 skb_postpush_rcsum(skb, eth, sizeof(*eth));
6401
6402 return 0;
6403 }
6404 EXPORT_SYMBOL(skb_eth_push);
6405
6406 /* 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)6407 static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
6408 __be16 ethertype)
6409 {
6410 if (skb->ip_summed == CHECKSUM_COMPLETE) {
6411 __be16 diff[] = { ~hdr->h_proto, ethertype };
6412
6413 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6414 }
6415
6416 hdr->h_proto = ethertype;
6417 }
6418
6419 /**
6420 * skb_mpls_push() - push a new MPLS header after mac_len bytes from start of
6421 * the packet
6422 *
6423 * @skb: buffer
6424 * @mpls_lse: MPLS label stack entry to push
6425 * @mpls_proto: ethertype of the new MPLS header (expects 0x8847 or 0x8848)
6426 * @mac_len: length of the MAC header
6427 * @ethernet: flag to indicate if the resulting packet after skb_mpls_push is
6428 * ethernet
6429 *
6430 * Expects skb->data at mac header.
6431 *
6432 * Returns 0 on success, -errno otherwise.
6433 */
skb_mpls_push(struct sk_buff * skb,__be32 mpls_lse,__be16 mpls_proto,int mac_len,bool ethernet)6434 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
6435 int mac_len, bool ethernet)
6436 {
6437 struct mpls_shim_hdr *lse;
6438 int err;
6439
6440 if (unlikely(!eth_p_mpls(mpls_proto)))
6441 return -EINVAL;
6442
6443 /* Networking stack does not allow simultaneous Tunnel and MPLS GSO. */
6444 if (skb->encapsulation)
6445 return -EINVAL;
6446
6447 err = skb_cow_head(skb, MPLS_HLEN);
6448 if (unlikely(err))
6449 return err;
6450
6451 if (!skb->inner_protocol) {
6452 skb_set_inner_network_header(skb, skb_network_offset(skb));
6453 skb_set_inner_protocol(skb, skb->protocol);
6454 }
6455
6456 skb_push(skb, MPLS_HLEN);
6457 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
6458 mac_len);
6459 skb_reset_mac_header(skb);
6460 skb_set_network_header(skb, mac_len);
6461 skb_reset_mac_len(skb);
6462
6463 lse = mpls_hdr(skb);
6464 lse->label_stack_entry = mpls_lse;
6465 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
6466
6467 if (ethernet && mac_len >= ETH_HLEN)
6468 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
6469 skb->protocol = mpls_proto;
6470
6471 return 0;
6472 }
6473 EXPORT_SYMBOL_GPL(skb_mpls_push);
6474
6475 /**
6476 * skb_mpls_pop() - pop the outermost MPLS header
6477 *
6478 * @skb: buffer
6479 * @next_proto: ethertype of header after popped MPLS header
6480 * @mac_len: length of the MAC header
6481 * @ethernet: flag to indicate if the packet is ethernet
6482 *
6483 * Expects skb->data at mac header.
6484 *
6485 * Returns 0 on success, -errno otherwise.
6486 */
skb_mpls_pop(struct sk_buff * skb,__be16 next_proto,int mac_len,bool ethernet)6487 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
6488 bool ethernet)
6489 {
6490 int err;
6491
6492 if (unlikely(!eth_p_mpls(skb->protocol)))
6493 return 0;
6494
6495 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
6496 if (unlikely(err))
6497 return err;
6498
6499 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
6500 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
6501 mac_len);
6502
6503 __skb_pull(skb, MPLS_HLEN);
6504 skb_reset_mac_header(skb);
6505 skb_set_network_header(skb, mac_len);
6506
6507 if (ethernet && mac_len >= ETH_HLEN) {
6508 struct ethhdr *hdr;
6509
6510 /* use mpls_hdr() to get ethertype to account for VLANs. */
6511 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
6512 skb_mod_eth_type(skb, hdr, next_proto);
6513 }
6514 skb->protocol = next_proto;
6515
6516 return 0;
6517 }
6518 EXPORT_SYMBOL_GPL(skb_mpls_pop);
6519
6520 /**
6521 * skb_mpls_update_lse() - modify outermost MPLS header and update csum
6522 *
6523 * @skb: buffer
6524 * @mpls_lse: new MPLS label stack entry to update to
6525 *
6526 * Expects skb->data at mac header.
6527 *
6528 * Returns 0 on success, -errno otherwise.
6529 */
skb_mpls_update_lse(struct sk_buff * skb,__be32 mpls_lse)6530 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
6531 {
6532 int err;
6533
6534 if (unlikely(!eth_p_mpls(skb->protocol)))
6535 return -EINVAL;
6536
6537 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
6538 if (unlikely(err))
6539 return err;
6540
6541 if (skb->ip_summed == CHECKSUM_COMPLETE) {
6542 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
6543
6544 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
6545 }
6546
6547 mpls_hdr(skb)->label_stack_entry = mpls_lse;
6548
6549 return 0;
6550 }
6551 EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
6552
6553 /**
6554 * skb_mpls_dec_ttl() - decrement the TTL of the outermost MPLS header
6555 *
6556 * @skb: buffer
6557 *
6558 * Expects skb->data at mac header.
6559 *
6560 * Returns 0 on success, -errno otherwise.
6561 */
skb_mpls_dec_ttl(struct sk_buff * skb)6562 int skb_mpls_dec_ttl(struct sk_buff *skb)
6563 {
6564 u32 lse;
6565 u8 ttl;
6566
6567 if (unlikely(!eth_p_mpls(skb->protocol)))
6568 return -EINVAL;
6569
6570 if (!pskb_may_pull(skb, skb_network_offset(skb) + MPLS_HLEN))
6571 return -ENOMEM;
6572
6573 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
6574 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
6575 if (!--ttl)
6576 return -EINVAL;
6577
6578 lse &= ~MPLS_LS_TTL_MASK;
6579 lse |= ttl << MPLS_LS_TTL_SHIFT;
6580
6581 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
6582 }
6583 EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
6584
6585 /**
6586 * alloc_skb_with_frags - allocate skb with page frags
6587 *
6588 * @header_len: size of linear part
6589 * @data_len: needed length in frags
6590 * @order: max page order desired.
6591 * @errcode: pointer to error code if any
6592 * @gfp_mask: allocation mask
6593 *
6594 * This can be used to allocate a paged skb, given a maximal order for frags.
6595 */
alloc_skb_with_frags(unsigned long header_len,unsigned long data_len,int order,int * errcode,gfp_t gfp_mask)6596 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
6597 unsigned long data_len,
6598 int order,
6599 int *errcode,
6600 gfp_t gfp_mask)
6601 {
6602 unsigned long chunk;
6603 struct sk_buff *skb;
6604 struct page *page;
6605 int nr_frags = 0;
6606
6607 *errcode = -EMSGSIZE;
6608 if (unlikely(data_len > MAX_SKB_FRAGS * (PAGE_SIZE << order)))
6609 return NULL;
6610
6611 *errcode = -ENOBUFS;
6612 skb = alloc_skb(header_len, gfp_mask);
6613 if (!skb)
6614 return NULL;
6615
6616 while (data_len) {
6617 if (nr_frags == MAX_SKB_FRAGS - 1)
6618 goto failure;
6619 while (order && PAGE_ALIGN(data_len) < (PAGE_SIZE << order))
6620 order--;
6621
6622 if (order) {
6623 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
6624 __GFP_COMP |
6625 __GFP_NOWARN,
6626 order);
6627 if (!page) {
6628 order--;
6629 continue;
6630 }
6631 } else {
6632 page = alloc_page(gfp_mask);
6633 if (!page)
6634 goto failure;
6635 }
6636 chunk = min_t(unsigned long, data_len,
6637 PAGE_SIZE << order);
6638 skb_fill_page_desc(skb, nr_frags, page, 0, chunk);
6639 nr_frags++;
6640 skb->truesize += (PAGE_SIZE << order);
6641 data_len -= chunk;
6642 }
6643 return skb;
6644
6645 failure:
6646 kfree_skb(skb);
6647 return NULL;
6648 }
6649 EXPORT_SYMBOL(alloc_skb_with_frags);
6650
6651 /* 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)6652 static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
6653 const int headlen, gfp_t gfp_mask)
6654 {
6655 int i;
6656 unsigned int size = skb_end_offset(skb);
6657 int new_hlen = headlen - off;
6658 u8 *data;
6659
6660 if (skb_pfmemalloc(skb))
6661 gfp_mask |= __GFP_MEMALLOC;
6662
6663 data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6664 if (!data)
6665 return -ENOMEM;
6666 size = SKB_WITH_OVERHEAD(size);
6667
6668 /* Copy real data, and all frags */
6669 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
6670 skb->len -= off;
6671
6672 memcpy((struct skb_shared_info *)(data + size),
6673 skb_shinfo(skb),
6674 offsetof(struct skb_shared_info,
6675 frags[skb_shinfo(skb)->nr_frags]));
6676 if (skb_cloned(skb)) {
6677 /* drop the old head gracefully */
6678 if (skb_orphan_frags(skb, gfp_mask)) {
6679 skb_kfree_head(data, size);
6680 return -ENOMEM;
6681 }
6682 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
6683 skb_frag_ref(skb, i);
6684 if (skb_has_frag_list(skb))
6685 skb_clone_fraglist(skb);
6686 skb_release_data(skb, SKB_CONSUMED);
6687 } else {
6688 /* we can reuse existing recount- all we did was
6689 * relocate values
6690 */
6691 skb_free_head(skb);
6692 }
6693
6694 skb->head = data;
6695 skb->data = data;
6696 skb->head_frag = 0;
6697 skb_set_end_offset(skb, size);
6698 skb_set_tail_pointer(skb, skb_headlen(skb));
6699 skb_headers_offset_update(skb, 0);
6700 skb->cloned = 0;
6701 skb->hdr_len = 0;
6702 skb->nohdr = 0;
6703 atomic_set(&skb_shinfo(skb)->dataref, 1);
6704
6705 return 0;
6706 }
6707
6708 static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
6709
6710 /* carve out the first eat bytes from skb's frag_list. May recurse into
6711 * pskb_carve()
6712 */
pskb_carve_frag_list(struct sk_buff * skb,struct skb_shared_info * shinfo,int eat,gfp_t gfp_mask)6713 static int pskb_carve_frag_list(struct sk_buff *skb,
6714 struct skb_shared_info *shinfo, int eat,
6715 gfp_t gfp_mask)
6716 {
6717 struct sk_buff *list = shinfo->frag_list;
6718 struct sk_buff *clone = NULL;
6719 struct sk_buff *insp = NULL;
6720
6721 do {
6722 if (!list) {
6723 pr_err("Not enough bytes to eat. Want %d\n", eat);
6724 return -EFAULT;
6725 }
6726 if (list->len <= eat) {
6727 /* Eaten as whole. */
6728 eat -= list->len;
6729 list = list->next;
6730 insp = list;
6731 } else {
6732 /* Eaten partially. */
6733 if (skb_shared(list)) {
6734 clone = skb_clone(list, gfp_mask);
6735 if (!clone)
6736 return -ENOMEM;
6737 insp = list->next;
6738 list = clone;
6739 } else {
6740 /* This may be pulled without problems. */
6741 insp = list;
6742 }
6743 if (pskb_carve(list, eat, gfp_mask) < 0) {
6744 kfree_skb(clone);
6745 return -ENOMEM;
6746 }
6747 break;
6748 }
6749 } while (eat);
6750
6751 /* Free pulled out fragments. */
6752 while ((list = shinfo->frag_list) != insp) {
6753 shinfo->frag_list = list->next;
6754 consume_skb(list);
6755 }
6756 /* And insert new clone at head. */
6757 if (clone) {
6758 clone->next = list;
6759 shinfo->frag_list = clone;
6760 }
6761 return 0;
6762 }
6763
6764 /* carve off first len bytes from skb. Split line (off) is in the
6765 * non-linear part of skb
6766 */
pskb_carve_inside_nonlinear(struct sk_buff * skb,const u32 off,int pos,gfp_t gfp_mask)6767 static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
6768 int pos, gfp_t gfp_mask)
6769 {
6770 int i, k = 0;
6771 unsigned int size = skb_end_offset(skb);
6772 u8 *data;
6773 const int nfrags = skb_shinfo(skb)->nr_frags;
6774 struct skb_shared_info *shinfo;
6775
6776 if (skb_pfmemalloc(skb))
6777 gfp_mask |= __GFP_MEMALLOC;
6778
6779 data = kmalloc_reserve(&size, gfp_mask, NUMA_NO_NODE, NULL);
6780 if (!data)
6781 return -ENOMEM;
6782 size = SKB_WITH_OVERHEAD(size);
6783
6784 memcpy((struct skb_shared_info *)(data + size),
6785 skb_shinfo(skb), offsetof(struct skb_shared_info, frags[0]));
6786 if (skb_orphan_frags(skb, gfp_mask)) {
6787 skb_kfree_head(data, size);
6788 return -ENOMEM;
6789 }
6790 shinfo = (struct skb_shared_info *)(data + size);
6791 for (i = 0; i < nfrags; i++) {
6792 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
6793
6794 if (pos + fsize > off) {
6795 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
6796
6797 if (pos < off) {
6798 /* Split frag.
6799 * We have two variants in this case:
6800 * 1. Move all the frag to the second
6801 * part, if it is possible. F.e.
6802 * this approach is mandatory for TUX,
6803 * where splitting is expensive.
6804 * 2. Split is accurately. We make this.
6805 */
6806 skb_frag_off_add(&shinfo->frags[0], off - pos);
6807 skb_frag_size_sub(&shinfo->frags[0], off - pos);
6808 }
6809 skb_frag_ref(skb, i);
6810 k++;
6811 }
6812 pos += fsize;
6813 }
6814 shinfo->nr_frags = k;
6815 if (skb_has_frag_list(skb))
6816 skb_clone_fraglist(skb);
6817
6818 /* split line is in frag list */
6819 if (k == 0 && pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask)) {
6820 /* skb_frag_unref() is not needed here as shinfo->nr_frags = 0. */
6821 if (skb_has_frag_list(skb))
6822 kfree_skb_list(skb_shinfo(skb)->frag_list);
6823 skb_kfree_head(data, size);
6824 return -ENOMEM;
6825 }
6826 skb_release_data(skb, SKB_CONSUMED);
6827
6828 skb->head = data;
6829 skb->head_frag = 0;
6830 skb->data = data;
6831 skb_set_end_offset(skb, size);
6832 skb_reset_tail_pointer(skb);
6833 skb_headers_offset_update(skb, 0);
6834 skb->cloned = 0;
6835 skb->hdr_len = 0;
6836 skb->nohdr = 0;
6837 skb->len -= off;
6838 skb->data_len = skb->len;
6839 atomic_set(&skb_shinfo(skb)->dataref, 1);
6840 return 0;
6841 }
6842
6843 /* remove len bytes from the beginning of the skb */
pskb_carve(struct sk_buff * skb,const u32 len,gfp_t gfp)6844 static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6845 {
6846 int headlen = skb_headlen(skb);
6847
6848 if (len < headlen)
6849 return pskb_carve_inside_header(skb, len, headlen, gfp);
6850 else
6851 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6852 }
6853
6854 /* Extract to_copy bytes starting at off from skb, and return this in
6855 * a new skb
6856 */
pskb_extract(struct sk_buff * skb,int off,int to_copy,gfp_t gfp)6857 struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6858 int to_copy, gfp_t gfp)
6859 {
6860 struct sk_buff *clone = skb_clone(skb, gfp);
6861
6862 if (!clone)
6863 return NULL;
6864
6865 if (pskb_carve(clone, off, gfp) < 0 ||
6866 pskb_trim(clone, to_copy)) {
6867 kfree_skb(clone);
6868 return NULL;
6869 }
6870 return clone;
6871 }
6872 EXPORT_SYMBOL(pskb_extract);
6873
6874 /**
6875 * skb_condense - try to get rid of fragments/frag_list if possible
6876 * @skb: buffer
6877 *
6878 * Can be used to save memory before skb is added to a busy queue.
6879 * If packet has bytes in frags and enough tail room in skb->head,
6880 * pull all of them, so that we can free the frags right now and adjust
6881 * truesize.
6882 * Notes:
6883 * We do not reallocate skb->head thus can not fail.
6884 * Caller must re-evaluate skb->truesize if needed.
6885 */
skb_condense(struct sk_buff * skb)6886 void skb_condense(struct sk_buff *skb)
6887 {
6888 if (skb->data_len) {
6889 if (skb->data_len > skb->end - skb->tail ||
6890 skb_cloned(skb) || !skb_frags_readable(skb))
6891 return;
6892
6893 /* Nice, we can free page frag(s) right now */
6894 __pskb_pull_tail(skb, skb->data_len);
6895 }
6896 /* At this point, skb->truesize might be over estimated,
6897 * because skb had a fragment, and fragments do not tell
6898 * their truesize.
6899 * When we pulled its content into skb->head, fragment
6900 * was freed, but __pskb_pull_tail() could not possibly
6901 * adjust skb->truesize, not knowing the frag truesize.
6902 */
6903 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6904 }
6905 EXPORT_SYMBOL(skb_condense);
6906
6907 #ifdef CONFIG_SKB_EXTENSIONS
skb_ext_get_ptr(struct skb_ext * ext,enum skb_ext_id id)6908 static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6909 {
6910 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6911 }
6912
6913 /**
6914 * __skb_ext_alloc - allocate a new skb extensions storage
6915 *
6916 * @flags: See kmalloc().
6917 *
6918 * Returns the newly allocated pointer. The pointer can later attached to a
6919 * skb via __skb_ext_set().
6920 * Note: caller must handle the skb_ext as an opaque data.
6921 */
__skb_ext_alloc(gfp_t flags)6922 struct skb_ext *__skb_ext_alloc(gfp_t flags)
6923 {
6924 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, flags);
6925
6926 if (new) {
6927 memset(new->offset, 0, sizeof(new->offset));
6928 refcount_set(&new->refcnt, 1);
6929 }
6930
6931 return new;
6932 }
6933
skb_ext_maybe_cow(struct skb_ext * old,unsigned int old_active)6934 static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6935 unsigned int old_active)
6936 {
6937 struct skb_ext *new;
6938
6939 if (refcount_read(&old->refcnt) == 1)
6940 return old;
6941
6942 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6943 if (!new)
6944 return NULL;
6945
6946 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6947 refcount_set(&new->refcnt, 1);
6948
6949 #ifdef CONFIG_XFRM
6950 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6951 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6952 unsigned int i;
6953
6954 for (i = 0; i < sp->len; i++)
6955 xfrm_state_hold(sp->xvec[i]);
6956 }
6957 #endif
6958 #ifdef CONFIG_MCTP_FLOWS
6959 if (old_active & (1 << SKB_EXT_MCTP)) {
6960 struct mctp_flow *flow = skb_ext_get_ptr(old, SKB_EXT_MCTP);
6961
6962 if (flow->key)
6963 refcount_inc(&flow->key->refs);
6964 }
6965 #endif
6966 __skb_ext_put(old);
6967 return new;
6968 }
6969
6970 /**
6971 * __skb_ext_set - attach the specified extension storage to this skb
6972 * @skb: buffer
6973 * @id: extension id
6974 * @ext: extension storage previously allocated via __skb_ext_alloc()
6975 *
6976 * Existing extensions, if any, are cleared.
6977 *
6978 * Returns the pointer to the extension.
6979 */
__skb_ext_set(struct sk_buff * skb,enum skb_ext_id id,struct skb_ext * ext)6980 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6981 struct skb_ext *ext)
6982 {
6983 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6984
6985 skb_ext_put(skb);
6986 newlen = newoff + skb_ext_type_len[id];
6987 ext->chunks = newlen;
6988 ext->offset[id] = newoff;
6989 skb->extensions = ext;
6990 skb->active_extensions = 1 << id;
6991 return skb_ext_get_ptr(ext, id);
6992 }
6993
6994 /**
6995 * skb_ext_add - allocate space for given extension, COW if needed
6996 * @skb: buffer
6997 * @id: extension to allocate space for
6998 *
6999 * Allocates enough space for the given extension.
7000 * If the extension is already present, a pointer to that extension
7001 * is returned.
7002 *
7003 * If the skb was cloned, COW applies and the returned memory can be
7004 * modified without changing the extension space of clones buffers.
7005 *
7006 * Returns pointer to the extension or NULL on allocation failure.
7007 */
skb_ext_add(struct sk_buff * skb,enum skb_ext_id id)7008 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
7009 {
7010 struct skb_ext *new, *old = NULL;
7011 unsigned int newlen, newoff;
7012
7013 if (skb->active_extensions) {
7014 old = skb->extensions;
7015
7016 new = skb_ext_maybe_cow(old, skb->active_extensions);
7017 if (!new)
7018 return NULL;
7019
7020 if (__skb_ext_exist(new, id))
7021 goto set_active;
7022
7023 newoff = new->chunks;
7024 } else {
7025 newoff = SKB_EXT_CHUNKSIZEOF(*new);
7026
7027 new = __skb_ext_alloc(GFP_ATOMIC);
7028 if (!new)
7029 return NULL;
7030 }
7031
7032 newlen = newoff + skb_ext_type_len[id];
7033 new->chunks = newlen;
7034 new->offset[id] = newoff;
7035 set_active:
7036 skb->slow_gro = 1;
7037 skb->extensions = new;
7038 skb->active_extensions |= 1 << id;
7039 return skb_ext_get_ptr(new, id);
7040 }
7041 EXPORT_SYMBOL(skb_ext_add);
7042
7043 #ifdef CONFIG_XFRM
skb_ext_put_sp(struct sec_path * sp)7044 static void skb_ext_put_sp(struct sec_path *sp)
7045 {
7046 unsigned int i;
7047
7048 for (i = 0; i < sp->len; i++)
7049 xfrm_state_put(sp->xvec[i]);
7050 }
7051 #endif
7052
7053 #ifdef CONFIG_MCTP_FLOWS
skb_ext_put_mctp(struct mctp_flow * flow)7054 static void skb_ext_put_mctp(struct mctp_flow *flow)
7055 {
7056 if (flow->key)
7057 mctp_key_unref(flow->key);
7058 }
7059 #endif
7060
__skb_ext_del(struct sk_buff * skb,enum skb_ext_id id)7061 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
7062 {
7063 struct skb_ext *ext = skb->extensions;
7064
7065 skb->active_extensions &= ~(1 << id);
7066 if (skb->active_extensions == 0) {
7067 skb->extensions = NULL;
7068 __skb_ext_put(ext);
7069 #ifdef CONFIG_XFRM
7070 } else if (id == SKB_EXT_SEC_PATH &&
7071 refcount_read(&ext->refcnt) == 1) {
7072 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
7073
7074 skb_ext_put_sp(sp);
7075 sp->len = 0;
7076 #endif
7077 }
7078 }
7079 EXPORT_SYMBOL(__skb_ext_del);
7080
__skb_ext_put(struct skb_ext * ext)7081 void __skb_ext_put(struct skb_ext *ext)
7082 {
7083 /* If this is last clone, nothing can increment
7084 * it after check passes. Avoids one atomic op.
7085 */
7086 if (refcount_read(&ext->refcnt) == 1)
7087 goto free_now;
7088
7089 if (!refcount_dec_and_test(&ext->refcnt))
7090 return;
7091 free_now:
7092 #ifdef CONFIG_XFRM
7093 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
7094 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
7095 #endif
7096 #ifdef CONFIG_MCTP_FLOWS
7097 if (__skb_ext_exist(ext, SKB_EXT_MCTP))
7098 skb_ext_put_mctp(skb_ext_get_ptr(ext, SKB_EXT_MCTP));
7099 #endif
7100
7101 kmem_cache_free(skbuff_ext_cache, ext);
7102 }
7103 EXPORT_SYMBOL(__skb_ext_put);
7104 #endif /* CONFIG_SKB_EXTENSIONS */
7105
kfree_skb_napi_cache(struct sk_buff * skb)7106 static void kfree_skb_napi_cache(struct sk_buff *skb)
7107 {
7108 /* if SKB is a clone, don't handle this case */
7109 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
7110 __kfree_skb(skb);
7111 return;
7112 }
7113
7114 local_bh_disable();
7115 __napi_kfree_skb(skb, SKB_CONSUMED);
7116 local_bh_enable();
7117 }
7118
7119 /**
7120 * skb_attempt_defer_free - queue skb for remote freeing
7121 * @skb: buffer
7122 *
7123 * Put @skb in a per-cpu list, using the cpu which
7124 * allocated the skb/pages to reduce false sharing
7125 * and memory zone spinlock contention.
7126 */
skb_attempt_defer_free(struct sk_buff * skb)7127 void skb_attempt_defer_free(struct sk_buff *skb)
7128 {
7129 int cpu = skb->alloc_cpu;
7130 struct softnet_data *sd;
7131 unsigned int defer_max;
7132 bool kick;
7133
7134 if (cpu == raw_smp_processor_id() ||
7135 WARN_ON_ONCE(cpu >= nr_cpu_ids) ||
7136 !cpu_online(cpu)) {
7137 nodefer: kfree_skb_napi_cache(skb);
7138 return;
7139 }
7140
7141 DEBUG_NET_WARN_ON_ONCE(skb_dst(skb));
7142 DEBUG_NET_WARN_ON_ONCE(skb->destructor);
7143
7144 sd = &per_cpu(softnet_data, cpu);
7145 defer_max = READ_ONCE(net_hotdata.sysctl_skb_defer_max);
7146 if (READ_ONCE(sd->defer_count) >= defer_max)
7147 goto nodefer;
7148
7149 spin_lock_bh(&sd->defer_lock);
7150 /* Send an IPI every time queue reaches half capacity. */
7151 kick = sd->defer_count == (defer_max >> 1);
7152 /* Paired with the READ_ONCE() few lines above */
7153 WRITE_ONCE(sd->defer_count, sd->defer_count + 1);
7154
7155 skb->next = sd->defer_list;
7156 /* Paired with READ_ONCE() in skb_defer_free_flush() */
7157 WRITE_ONCE(sd->defer_list, skb);
7158 spin_unlock_bh(&sd->defer_lock);
7159
7160 /* Make sure to trigger NET_RX_SOFTIRQ on the remote CPU
7161 * if we are unlucky enough (this seems very unlikely).
7162 */
7163 if (unlikely(kick))
7164 kick_defer_list_purge(sd, cpu);
7165 }
7166
skb_splice_csum_page(struct sk_buff * skb,struct page * page,size_t offset,size_t len)7167 static void skb_splice_csum_page(struct sk_buff *skb, struct page *page,
7168 size_t offset, size_t len)
7169 {
7170 const char *kaddr;
7171 __wsum csum;
7172
7173 kaddr = kmap_local_page(page);
7174 csum = csum_partial(kaddr + offset, len, 0);
7175 kunmap_local(kaddr);
7176 skb->csum = csum_block_add(skb->csum, csum, skb->len);
7177 }
7178
7179 /**
7180 * skb_splice_from_iter - Splice (or copy) pages to skbuff
7181 * @skb: The buffer to add pages to
7182 * @iter: Iterator representing the pages to be added
7183 * @maxsize: Maximum amount of pages to be added
7184 * @gfp: Allocation flags
7185 *
7186 * This is a common helper function for supporting MSG_SPLICE_PAGES. It
7187 * extracts pages from an iterator and adds them to the socket buffer if
7188 * possible, copying them to fragments if not possible (such as if they're slab
7189 * pages).
7190 *
7191 * Returns the amount of data spliced/copied or -EMSGSIZE if there's
7192 * insufficient space in the buffer to transfer anything.
7193 */
skb_splice_from_iter(struct sk_buff * skb,struct iov_iter * iter,ssize_t maxsize,gfp_t gfp)7194 ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter,
7195 ssize_t maxsize, gfp_t gfp)
7196 {
7197 size_t frag_limit = READ_ONCE(net_hotdata.sysctl_max_skb_frags);
7198 struct page *pages[8], **ppages = pages;
7199 ssize_t spliced = 0, ret = 0;
7200 unsigned int i;
7201
7202 while (iter->count > 0) {
7203 ssize_t space, nr, len;
7204 size_t off;
7205
7206 ret = -EMSGSIZE;
7207 space = frag_limit - skb_shinfo(skb)->nr_frags;
7208 if (space < 0)
7209 break;
7210
7211 /* We might be able to coalesce without increasing nr_frags */
7212 nr = clamp_t(size_t, space, 1, ARRAY_SIZE(pages));
7213
7214 len = iov_iter_extract_pages(iter, &ppages, maxsize, nr, 0, &off);
7215 if (len <= 0) {
7216 ret = len ?: -EIO;
7217 break;
7218 }
7219
7220 i = 0;
7221 do {
7222 struct page *page = pages[i++];
7223 size_t part = min_t(size_t, PAGE_SIZE - off, len);
7224
7225 ret = -EIO;
7226 if (WARN_ON_ONCE(!sendpage_ok(page)))
7227 goto out;
7228
7229 ret = skb_append_pagefrags(skb, page, off, part,
7230 frag_limit);
7231 if (ret < 0) {
7232 iov_iter_revert(iter, len);
7233 goto out;
7234 }
7235
7236 if (skb->ip_summed == CHECKSUM_NONE)
7237 skb_splice_csum_page(skb, page, off, part);
7238
7239 off = 0;
7240 spliced += part;
7241 maxsize -= part;
7242 len -= part;
7243 } while (len > 0);
7244
7245 if (maxsize <= 0)
7246 break;
7247 }
7248
7249 out:
7250 skb_len_add(skb, spliced);
7251 return spliced ?: ret;
7252 }
7253 EXPORT_SYMBOL(skb_splice_from_iter);
7254
7255 static __always_inline
memcpy_from_iter_csum(void * iter_from,size_t progress,size_t len,void * to,void * priv2)7256 size_t memcpy_from_iter_csum(void *iter_from, size_t progress,
7257 size_t len, void *to, void *priv2)
7258 {
7259 __wsum *csum = priv2;
7260 __wsum next = csum_partial_copy_nocheck(iter_from, to + progress, len);
7261
7262 *csum = csum_block_add(*csum, next, progress);
7263 return 0;
7264 }
7265
7266 static __always_inline
copy_from_user_iter_csum(void __user * iter_from,size_t progress,size_t len,void * to,void * priv2)7267 size_t copy_from_user_iter_csum(void __user *iter_from, size_t progress,
7268 size_t len, void *to, void *priv2)
7269 {
7270 __wsum next, *csum = priv2;
7271
7272 next = csum_and_copy_from_user(iter_from, to + progress, len);
7273 *csum = csum_block_add(*csum, next, progress);
7274 return next ? 0 : len;
7275 }
7276
csum_and_copy_from_iter_full(void * addr,size_t bytes,__wsum * csum,struct iov_iter * i)7277 bool csum_and_copy_from_iter_full(void *addr, size_t bytes,
7278 __wsum *csum, struct iov_iter *i)
7279 {
7280 size_t copied;
7281
7282 if (WARN_ON_ONCE(!i->data_source))
7283 return false;
7284 copied = iterate_and_advance2(i, bytes, addr, csum,
7285 copy_from_user_iter_csum,
7286 memcpy_from_iter_csum);
7287 if (likely(copied == bytes))
7288 return true;
7289 iov_iter_revert(i, copied);
7290 return false;
7291 }
7292 EXPORT_SYMBOL(csum_and_copy_from_iter_full);
7293