aboutsummaryrefslogtreecommitdiffstats
path: root/net/core/skbuff.c
diff options
context:
space:
mode:
Diffstat (limited to 'net/core/skbuff.c')
-rw-r--r--net/core/skbuff.c1460
1 files changed, 1460 insertions, 0 deletions
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
new file mode 100644
index 000000000000..bf02ca9f80ac
--- /dev/null
+++ b/net/core/skbuff.c
@@ -0,0 +1,1460 @@
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <iiitac@pyr.swan.ac.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
8 *
9 * Fixes:
10 * Alan Cox : Fixed the worst of the load
11 * balancer bugs.
12 * Dave Platt : Interrupt stacking fix.
13 * Richard Kooijman : Timestamp fixes.
14 * Alan Cox : Changed buffer format.
15 * Alan Cox : destructor hook for AF_UNIX etc.
16 * Linus Torvalds : Better skb_clone.
17 * Alan Cox : Added skb_copy.
18 * Alan Cox : Added all the changed routines Linus
19 * only put in the headers
20 * Ray VanTassle : Fixed --skb->lock in free
21 * Alan Cox : skb_copy copy arp field
22 * Andi Kleen : slabified it.
23 * Robert Olsson : Removed skb_head_pool
24 *
25 * NOTE:
26 * The __skb_ routines should be called with interrupts
27 * disabled, or you better be *real* sure that the operation is atomic
28 * with respect to whatever list is being frobbed (e.g. via lock_sock()
29 * or via disabling bottom half handlers, etc).
30 *
31 * This program is free software; you can redistribute it and/or
32 * modify it under the terms of the GNU General Public License
33 * as published by the Free Software Foundation; either version
34 * 2 of the License, or (at your option) any later version.
35 */
36
37/*
38 * The functions in this file will not compile correctly with gcc 2.4.x
39 */
40
41#include <linux/config.h>
42#include <linux/module.h>
43#include <linux/types.h>
44#include <linux/kernel.h>
45#include <linux/sched.h>
46#include <linux/mm.h>
47#include <linux/interrupt.h>
48#include <linux/in.h>
49#include <linux/inet.h>
50#include <linux/slab.h>
51#include <linux/netdevice.h>
52#ifdef CONFIG_NET_CLS_ACT
53#include <net/pkt_sched.h>
54#endif
55#include <linux/string.h>
56#include <linux/skbuff.h>
57#include <linux/cache.h>
58#include <linux/rtnetlink.h>
59#include <linux/init.h>
60#include <linux/highmem.h>
61
62#include <net/protocol.h>
63#include <net/dst.h>
64#include <net/sock.h>
65#include <net/checksum.h>
66#include <net/xfrm.h>
67
68#include <asm/uaccess.h>
69#include <asm/system.h>
70
71static kmem_cache_t *skbuff_head_cache;
72
73/*
74 * Keep out-of-line to prevent kernel bloat.
75 * __builtin_return_address is not used because it is not always
76 * reliable.
77 */
78
79/**
80 * skb_over_panic - private function
81 * @skb: buffer
82 * @sz: size
83 * @here: address
84 *
85 * Out of line support code for skb_put(). Not user callable.
86 */
87void skb_over_panic(struct sk_buff *skb, int sz, void *here)
88{
89 printk(KERN_INFO "skput:over: %p:%d put:%d dev:%s",
90 here, skb->len, sz, skb->dev ? skb->dev->name : "<NULL>");
91 BUG();
92}
93
94/**
95 * skb_under_panic - private function
96 * @skb: buffer
97 * @sz: size
98 * @here: address
99 *
100 * Out of line support code for skb_push(). Not user callable.
101 */
102
103void skb_under_panic(struct sk_buff *skb, int sz, void *here)
104{
105 printk(KERN_INFO "skput:under: %p:%d put:%d dev:%s",
106 here, skb->len, sz, skb->dev ? skb->dev->name : "<NULL>");
107 BUG();
108}
109
110/* Allocate a new skbuff. We do this ourselves so we can fill in a few
111 * 'private' fields and also do memory statistics to find all the
112 * [BEEP] leaks.
113 *
114 */
115
116/**
117 * alloc_skb - allocate a network buffer
118 * @size: size to allocate
119 * @gfp_mask: allocation mask
120 *
121 * Allocate a new &sk_buff. The returned buffer has no headroom and a
122 * tail room of size bytes. The object has a reference count of one.
123 * The return is the buffer. On a failure the return is %NULL.
124 *
125 * Buffers may only be allocated from interrupts using a @gfp_mask of
126 * %GFP_ATOMIC.
127 */
128struct sk_buff *alloc_skb(unsigned int size, int gfp_mask)
129{
130 struct sk_buff *skb;
131 u8 *data;
132
133 /* Get the HEAD */
134 skb = kmem_cache_alloc(skbuff_head_cache,
135 gfp_mask & ~__GFP_DMA);
136 if (!skb)
137 goto out;
138
139 /* Get the DATA. Size must match skb_add_mtu(). */
140 size = SKB_DATA_ALIGN(size);
141 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
142 if (!data)
143 goto nodata;
144
145 memset(skb, 0, offsetof(struct sk_buff, truesize));
146 skb->truesize = size + sizeof(struct sk_buff);
147 atomic_set(&skb->users, 1);
148 skb->head = data;
149 skb->data = data;
150 skb->tail = data;
151 skb->end = data + size;
152
153 atomic_set(&(skb_shinfo(skb)->dataref), 1);
154 skb_shinfo(skb)->nr_frags = 0;
155 skb_shinfo(skb)->tso_size = 0;
156 skb_shinfo(skb)->tso_segs = 0;
157 skb_shinfo(skb)->frag_list = NULL;
158out:
159 return skb;
160nodata:
161 kmem_cache_free(skbuff_head_cache, skb);
162 skb = NULL;
163 goto out;
164}
165
166/**
167 * alloc_skb_from_cache - allocate a network buffer
168 * @cp: kmem_cache from which to allocate the data area
169 * (object size must be big enough for @size bytes + skb overheads)
170 * @size: size to allocate
171 * @gfp_mask: allocation mask
172 *
173 * Allocate a new &sk_buff. The returned buffer has no headroom and
174 * tail room of size bytes. The object has a reference count of one.
175 * The return is the buffer. On a failure the return is %NULL.
176 *
177 * Buffers may only be allocated from interrupts using a @gfp_mask of
178 * %GFP_ATOMIC.
179 */
180struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
181 unsigned int size, int gfp_mask)
182{
183 struct sk_buff *skb;
184 u8 *data;
185
186 /* Get the HEAD */
187 skb = kmem_cache_alloc(skbuff_head_cache,
188 gfp_mask & ~__GFP_DMA);
189 if (!skb)
190 goto out;
191
192 /* Get the DATA. */
193 size = SKB_DATA_ALIGN(size);
194 data = kmem_cache_alloc(cp, gfp_mask);
195 if (!data)
196 goto nodata;
197
198 memset(skb, 0, offsetof(struct sk_buff, truesize));
199 skb->truesize = size + sizeof(struct sk_buff);
200 atomic_set(&skb->users, 1);
201 skb->head = data;
202 skb->data = data;
203 skb->tail = data;
204 skb->end = data + size;
205
206 atomic_set(&(skb_shinfo(skb)->dataref), 1);
207 skb_shinfo(skb)->nr_frags = 0;
208 skb_shinfo(skb)->tso_size = 0;
209 skb_shinfo(skb)->tso_segs = 0;
210 skb_shinfo(skb)->frag_list = NULL;
211out:
212 return skb;
213nodata:
214 kmem_cache_free(skbuff_head_cache, skb);
215 skb = NULL;
216 goto out;
217}
218
219
220static void skb_drop_fraglist(struct sk_buff *skb)
221{
222 struct sk_buff *list = skb_shinfo(skb)->frag_list;
223
224 skb_shinfo(skb)->frag_list = NULL;
225
226 do {
227 struct sk_buff *this = list;
228 list = list->next;
229 kfree_skb(this);
230 } while (list);
231}
232
233static void skb_clone_fraglist(struct sk_buff *skb)
234{
235 struct sk_buff *list;
236
237 for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
238 skb_get(list);
239}
240
241void skb_release_data(struct sk_buff *skb)
242{
243 if (!skb->cloned ||
244 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
245 &skb_shinfo(skb)->dataref)) {
246 if (skb_shinfo(skb)->nr_frags) {
247 int i;
248 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
249 put_page(skb_shinfo(skb)->frags[i].page);
250 }
251
252 if (skb_shinfo(skb)->frag_list)
253 skb_drop_fraglist(skb);
254
255 kfree(skb->head);
256 }
257}
258
259/*
260 * Free an skbuff by memory without cleaning the state.
261 */
262void kfree_skbmem(struct sk_buff *skb)
263{
264 skb_release_data(skb);
265 kmem_cache_free(skbuff_head_cache, skb);
266}
267
268/**
269 * __kfree_skb - private function
270 * @skb: buffer
271 *
272 * Free an sk_buff. Release anything attached to the buffer.
273 * Clean the state. This is an internal helper function. Users should
274 * always call kfree_skb
275 */
276
277void __kfree_skb(struct sk_buff *skb)
278{
279 if (skb->list) {
280 printk(KERN_WARNING "Warning: kfree_skb passed an skb still "
281 "on a list (from %p).\n", NET_CALLER(skb));
282 BUG();
283 }
284
285 dst_release(skb->dst);
286#ifdef CONFIG_XFRM
287 secpath_put(skb->sp);
288#endif
289 if(skb->destructor) {
290 if (in_irq())
291 printk(KERN_WARNING "Warning: kfree_skb on "
292 "hard IRQ %p\n", NET_CALLER(skb));
293 skb->destructor(skb);
294 }
295#ifdef CONFIG_NETFILTER
296 nf_conntrack_put(skb->nfct);
297#ifdef CONFIG_BRIDGE_NETFILTER
298 nf_bridge_put(skb->nf_bridge);
299#endif
300#endif
301/* XXX: IS this still necessary? - JHS */
302#ifdef CONFIG_NET_SCHED
303 skb->tc_index = 0;
304#ifdef CONFIG_NET_CLS_ACT
305 skb->tc_verd = 0;
306 skb->tc_classid = 0;
307#endif
308#endif
309
310 kfree_skbmem(skb);
311}
312
313/**
314 * skb_clone - duplicate an sk_buff
315 * @skb: buffer to clone
316 * @gfp_mask: allocation priority
317 *
318 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
319 * copies share the same packet data but not structure. The new
320 * buffer has a reference count of 1. If the allocation fails the
321 * function returns %NULL otherwise the new buffer is returned.
322 *
323 * If this function is called from an interrupt gfp_mask() must be
324 * %GFP_ATOMIC.
325 */
326
327struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask)
328{
329 struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
330
331 if (!n)
332 return NULL;
333
334#define C(x) n->x = skb->x
335
336 n->next = n->prev = NULL;
337 n->list = NULL;
338 n->sk = NULL;
339 C(stamp);
340 C(dev);
341 C(real_dev);
342 C(h);
343 C(nh);
344 C(mac);
345 C(dst);
346 dst_clone(skb->dst);
347 C(sp);
348#ifdef CONFIG_INET
349 secpath_get(skb->sp);
350#endif
351 memcpy(n->cb, skb->cb, sizeof(skb->cb));
352 C(len);
353 C(data_len);
354 C(csum);
355 C(local_df);
356 n->cloned = 1;
357 n->nohdr = 0;
358 C(pkt_type);
359 C(ip_summed);
360 C(priority);
361 C(protocol);
362 C(security);
363 n->destructor = NULL;
364#ifdef CONFIG_NETFILTER
365 C(nfmark);
366 C(nfcache);
367 C(nfct);
368 nf_conntrack_get(skb->nfct);
369 C(nfctinfo);
370#ifdef CONFIG_NETFILTER_DEBUG
371 C(nf_debug);
372#endif
373#ifdef CONFIG_BRIDGE_NETFILTER
374 C(nf_bridge);
375 nf_bridge_get(skb->nf_bridge);
376#endif
377#endif /*CONFIG_NETFILTER*/
378#if defined(CONFIG_HIPPI)
379 C(private);
380#endif
381#ifdef CONFIG_NET_SCHED
382 C(tc_index);
383#ifdef CONFIG_NET_CLS_ACT
384 n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
385 n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
386 n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
387 C(input_dev);
388 C(tc_classid);
389#endif
390
391#endif
392 C(truesize);
393 atomic_set(&n->users, 1);
394 C(head);
395 C(data);
396 C(tail);
397 C(end);
398
399 atomic_inc(&(skb_shinfo(skb)->dataref));
400 skb->cloned = 1;
401
402 return n;
403}
404
405static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
406{
407 /*
408 * Shift between the two data areas in bytes
409 */
410 unsigned long offset = new->data - old->data;
411
412 new->list = NULL;
413 new->sk = NULL;
414 new->dev = old->dev;
415 new->real_dev = old->real_dev;
416 new->priority = old->priority;
417 new->protocol = old->protocol;
418 new->dst = dst_clone(old->dst);
419#ifdef CONFIG_INET
420 new->sp = secpath_get(old->sp);
421#endif
422 new->h.raw = old->h.raw + offset;
423 new->nh.raw = old->nh.raw + offset;
424 new->mac.raw = old->mac.raw + offset;
425 memcpy(new->cb, old->cb, sizeof(old->cb));
426 new->local_df = old->local_df;
427 new->pkt_type = old->pkt_type;
428 new->stamp = old->stamp;
429 new->destructor = NULL;
430 new->security = old->security;
431#ifdef CONFIG_NETFILTER
432 new->nfmark = old->nfmark;
433 new->nfcache = old->nfcache;
434 new->nfct = old->nfct;
435 nf_conntrack_get(old->nfct);
436 new->nfctinfo = old->nfctinfo;
437#ifdef CONFIG_NETFILTER_DEBUG
438 new->nf_debug = old->nf_debug;
439#endif
440#ifdef CONFIG_BRIDGE_NETFILTER
441 new->nf_bridge = old->nf_bridge;
442 nf_bridge_get(old->nf_bridge);
443#endif
444#endif
445#ifdef CONFIG_NET_SCHED
446#ifdef CONFIG_NET_CLS_ACT
447 new->tc_verd = old->tc_verd;
448#endif
449 new->tc_index = old->tc_index;
450#endif
451 atomic_set(&new->users, 1);
452 skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
453 skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
454}
455
456/**
457 * skb_copy - create private copy of an sk_buff
458 * @skb: buffer to copy
459 * @gfp_mask: allocation priority
460 *
461 * Make a copy of both an &sk_buff and its data. This is used when the
462 * caller wishes to modify the data and needs a private copy of the
463 * data to alter. Returns %NULL on failure or the pointer to the buffer
464 * on success. The returned buffer has a reference count of 1.
465 *
466 * As by-product this function converts non-linear &sk_buff to linear
467 * one, so that &sk_buff becomes completely private and caller is allowed
468 * to modify all the data of returned buffer. This means that this
469 * function is not recommended for use in circumstances when only
470 * header is going to be modified. Use pskb_copy() instead.
471 */
472
473struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
474{
475 int headerlen = skb->data - skb->head;
476 /*
477 * Allocate the copy buffer
478 */
479 struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
480 gfp_mask);
481 if (!n)
482 return NULL;
483
484 /* Set the data pointer */
485 skb_reserve(n, headerlen);
486 /* Set the tail pointer and length */
487 skb_put(n, skb->len);
488 n->csum = skb->csum;
489 n->ip_summed = skb->ip_summed;
490
491 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
492 BUG();
493
494 copy_skb_header(n, skb);
495 return n;
496}
497
498
499/**
500 * pskb_copy - create copy of an sk_buff with private head.
501 * @skb: buffer to copy
502 * @gfp_mask: allocation priority
503 *
504 * Make a copy of both an &sk_buff and part of its data, located
505 * in header. Fragmented data remain shared. This is used when
506 * the caller wishes to modify only header of &sk_buff and needs
507 * private copy of the header to alter. Returns %NULL on failure
508 * or the pointer to the buffer on success.
509 * The returned buffer has a reference count of 1.
510 */
511
512struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
513{
514 /*
515 * Allocate the copy buffer
516 */
517 struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
518
519 if (!n)
520 goto out;
521
522 /* Set the data pointer */
523 skb_reserve(n, skb->data - skb->head);
524 /* Set the tail pointer and length */
525 skb_put(n, skb_headlen(skb));
526 /* Copy the bytes */
527 memcpy(n->data, skb->data, n->len);
528 n->csum = skb->csum;
529 n->ip_summed = skb->ip_summed;
530
531 n->data_len = skb->data_len;
532 n->len = skb->len;
533
534 if (skb_shinfo(skb)->nr_frags) {
535 int i;
536
537 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
538 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
539 get_page(skb_shinfo(n)->frags[i].page);
540 }
541 skb_shinfo(n)->nr_frags = i;
542 }
543
544 if (skb_shinfo(skb)->frag_list) {
545 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
546 skb_clone_fraglist(n);
547 }
548
549 copy_skb_header(n, skb);
550out:
551 return n;
552}
553
554/**
555 * pskb_expand_head - reallocate header of &sk_buff
556 * @skb: buffer to reallocate
557 * @nhead: room to add at head
558 * @ntail: room to add at tail
559 * @gfp_mask: allocation priority
560 *
561 * Expands (or creates identical copy, if &nhead and &ntail are zero)
562 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
563 * reference count of 1. Returns zero in the case of success or error,
564 * if expansion failed. In the last case, &sk_buff is not changed.
565 *
566 * All the pointers pointing into skb header may change and must be
567 * reloaded after call to this function.
568 */
569
570int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
571{
572 int i;
573 u8 *data;
574 int size = nhead + (skb->end - skb->head) + ntail;
575 long off;
576
577 if (skb_shared(skb))
578 BUG();
579
580 size = SKB_DATA_ALIGN(size);
581
582 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
583 if (!data)
584 goto nodata;
585
586 /* Copy only real data... and, alas, header. This should be
587 * optimized for the cases when header is void. */
588 memcpy(data + nhead, skb->head, skb->tail - skb->head);
589 memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
590
591 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
592 get_page(skb_shinfo(skb)->frags[i].page);
593
594 if (skb_shinfo(skb)->frag_list)
595 skb_clone_fraglist(skb);
596
597 skb_release_data(skb);
598
599 off = (data + nhead) - skb->head;
600
601 skb->head = data;
602 skb->end = data + size;
603 skb->data += off;
604 skb->tail += off;
605 skb->mac.raw += off;
606 skb->h.raw += off;
607 skb->nh.raw += off;
608 skb->cloned = 0;
609 skb->nohdr = 0;
610 atomic_set(&skb_shinfo(skb)->dataref, 1);
611 return 0;
612
613nodata:
614 return -ENOMEM;
615}
616
617/* Make private copy of skb with writable head and some headroom */
618
619struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
620{
621 struct sk_buff *skb2;
622 int delta = headroom - skb_headroom(skb);
623
624 if (delta <= 0)
625 skb2 = pskb_copy(skb, GFP_ATOMIC);
626 else {
627 skb2 = skb_clone(skb, GFP_ATOMIC);
628 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
629 GFP_ATOMIC)) {
630 kfree_skb(skb2);
631 skb2 = NULL;
632 }
633 }
634 return skb2;
635}
636
637
638/**
639 * skb_copy_expand - copy and expand sk_buff
640 * @skb: buffer to copy
641 * @newheadroom: new free bytes at head
642 * @newtailroom: new free bytes at tail
643 * @gfp_mask: allocation priority
644 *
645 * Make a copy of both an &sk_buff and its data and while doing so
646 * allocate additional space.
647 *
648 * This is used when the caller wishes to modify the data and needs a
649 * private copy of the data to alter as well as more space for new fields.
650 * Returns %NULL on failure or the pointer to the buffer
651 * on success. The returned buffer has a reference count of 1.
652 *
653 * You must pass %GFP_ATOMIC as the allocation priority if this function
654 * is called from an interrupt.
655 *
656 * BUG ALERT: ip_summed is not copied. Why does this work? Is it used
657 * only by netfilter in the cases when checksum is recalculated? --ANK
658 */
659struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
660 int newheadroom, int newtailroom, int gfp_mask)
661{
662 /*
663 * Allocate the copy buffer
664 */
665 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
666 gfp_mask);
667 int head_copy_len, head_copy_off;
668
669 if (!n)
670 return NULL;
671
672 skb_reserve(n, newheadroom);
673
674 /* Set the tail pointer and length */
675 skb_put(n, skb->len);
676
677 head_copy_len = skb_headroom(skb);
678 head_copy_off = 0;
679 if (newheadroom <= head_copy_len)
680 head_copy_len = newheadroom;
681 else
682 head_copy_off = newheadroom - head_copy_len;
683
684 /* Copy the linear header and data. */
685 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
686 skb->len + head_copy_len))
687 BUG();
688
689 copy_skb_header(n, skb);
690
691 return n;
692}
693
694/**
695 * skb_pad - zero pad the tail of an skb
696 * @skb: buffer to pad
697 * @pad: space to pad
698 *
699 * Ensure that a buffer is followed by a padding area that is zero
700 * filled. Used by network drivers which may DMA or transfer data
701 * beyond the buffer end onto the wire.
702 *
703 * May return NULL in out of memory cases.
704 */
705
706struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
707{
708 struct sk_buff *nskb;
709
710 /* If the skbuff is non linear tailroom is always zero.. */
711 if (skb_tailroom(skb) >= pad) {
712 memset(skb->data+skb->len, 0, pad);
713 return skb;
714 }
715
716 nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
717 kfree_skb(skb);
718 if (nskb)
719 memset(nskb->data+nskb->len, 0, pad);
720 return nskb;
721}
722
723/* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
724 * If realloc==0 and trimming is impossible without change of data,
725 * it is BUG().
726 */
727
728int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
729{
730 int offset = skb_headlen(skb);
731 int nfrags = skb_shinfo(skb)->nr_frags;
732 int i;
733
734 for (i = 0; i < nfrags; i++) {
735 int end = offset + skb_shinfo(skb)->frags[i].size;
736 if (end > len) {
737 if (skb_cloned(skb)) {
738 if (!realloc)
739 BUG();
740 if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
741 return -ENOMEM;
742 }
743 if (len <= offset) {
744 put_page(skb_shinfo(skb)->frags[i].page);
745 skb_shinfo(skb)->nr_frags--;
746 } else {
747 skb_shinfo(skb)->frags[i].size = len - offset;
748 }
749 }
750 offset = end;
751 }
752
753 if (offset < len) {
754 skb->data_len -= skb->len - len;
755 skb->len = len;
756 } else {
757 if (len <= skb_headlen(skb)) {
758 skb->len = len;
759 skb->data_len = 0;
760 skb->tail = skb->data + len;
761 if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
762 skb_drop_fraglist(skb);
763 } else {
764 skb->data_len -= skb->len - len;
765 skb->len = len;
766 }
767 }
768
769 return 0;
770}
771
772/**
773 * __pskb_pull_tail - advance tail of skb header
774 * @skb: buffer to reallocate
775 * @delta: number of bytes to advance tail
776 *
777 * The function makes a sense only on a fragmented &sk_buff,
778 * it expands header moving its tail forward and copying necessary
779 * data from fragmented part.
780 *
781 * &sk_buff MUST have reference count of 1.
782 *
783 * Returns %NULL (and &sk_buff does not change) if pull failed
784 * or value of new tail of skb in the case of success.
785 *
786 * All the pointers pointing into skb header may change and must be
787 * reloaded after call to this function.
788 */
789
790/* Moves tail of skb head forward, copying data from fragmented part,
791 * when it is necessary.
792 * 1. It may fail due to malloc failure.
793 * 2. It may change skb pointers.
794 *
795 * It is pretty complicated. Luckily, it is called only in exceptional cases.
796 */
797unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
798{
799 /* If skb has not enough free space at tail, get new one
800 * plus 128 bytes for future expansions. If we have enough
801 * room at tail, reallocate without expansion only if skb is cloned.
802 */
803 int i, k, eat = (skb->tail + delta) - skb->end;
804
805 if (eat > 0 || skb_cloned(skb)) {
806 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
807 GFP_ATOMIC))
808 return NULL;
809 }
810
811 if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
812 BUG();
813
814 /* Optimization: no fragments, no reasons to preestimate
815 * size of pulled pages. Superb.
816 */
817 if (!skb_shinfo(skb)->frag_list)
818 goto pull_pages;
819
820 /* Estimate size of pulled pages. */
821 eat = delta;
822 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
823 if (skb_shinfo(skb)->frags[i].size >= eat)
824 goto pull_pages;
825 eat -= skb_shinfo(skb)->frags[i].size;
826 }
827
828 /* If we need update frag list, we are in troubles.
829 * Certainly, it possible to add an offset to skb data,
830 * but taking into account that pulling is expected to
831 * be very rare operation, it is worth to fight against
832 * further bloating skb head and crucify ourselves here instead.
833 * Pure masohism, indeed. 8)8)
834 */
835 if (eat) {
836 struct sk_buff *list = skb_shinfo(skb)->frag_list;
837 struct sk_buff *clone = NULL;
838 struct sk_buff *insp = NULL;
839
840 do {
841 if (!list)
842 BUG();
843
844 if (list->len <= eat) {
845 /* Eaten as whole. */
846 eat -= list->len;
847 list = list->next;
848 insp = list;
849 } else {
850 /* Eaten partially. */
851
852 if (skb_shared(list)) {
853 /* Sucks! We need to fork list. :-( */
854 clone = skb_clone(list, GFP_ATOMIC);
855 if (!clone)
856 return NULL;
857 insp = list->next;
858 list = clone;
859 } else {
860 /* This may be pulled without
861 * problems. */
862 insp = list;
863 }
864 if (!pskb_pull(list, eat)) {
865 if (clone)
866 kfree_skb(clone);
867 return NULL;
868 }
869 break;
870 }
871 } while (eat);
872
873 /* Free pulled out fragments. */
874 while ((list = skb_shinfo(skb)->frag_list) != insp) {
875 skb_shinfo(skb)->frag_list = list->next;
876 kfree_skb(list);
877 }
878 /* And insert new clone at head. */
879 if (clone) {
880 clone->next = list;
881 skb_shinfo(skb)->frag_list = clone;
882 }
883 }
884 /* Success! Now we may commit changes to skb data. */
885
886pull_pages:
887 eat = delta;
888 k = 0;
889 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
890 if (skb_shinfo(skb)->frags[i].size <= eat) {
891 put_page(skb_shinfo(skb)->frags[i].page);
892 eat -= skb_shinfo(skb)->frags[i].size;
893 } else {
894 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
895 if (eat) {
896 skb_shinfo(skb)->frags[k].page_offset += eat;
897 skb_shinfo(skb)->frags[k].size -= eat;
898 eat = 0;
899 }
900 k++;
901 }
902 }
903 skb_shinfo(skb)->nr_frags = k;
904
905 skb->tail += delta;
906 skb->data_len -= delta;
907
908 return skb->tail;
909}
910
911/* Copy some data bits from skb to kernel buffer. */
912
913int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
914{
915 int i, copy;
916 int start = skb_headlen(skb);
917
918 if (offset > (int)skb->len - len)
919 goto fault;
920
921 /* Copy header. */
922 if ((copy = start - offset) > 0) {
923 if (copy > len)
924 copy = len;
925 memcpy(to, skb->data + offset, copy);
926 if ((len -= copy) == 0)
927 return 0;
928 offset += copy;
929 to += copy;
930 }
931
932 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
933 int end;
934
935 BUG_TRAP(start <= offset + len);
936
937 end = start + skb_shinfo(skb)->frags[i].size;
938 if ((copy = end - offset) > 0) {
939 u8 *vaddr;
940
941 if (copy > len)
942 copy = len;
943
944 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
945 memcpy(to,
946 vaddr + skb_shinfo(skb)->frags[i].page_offset+
947 offset - start, copy);
948 kunmap_skb_frag(vaddr);
949
950 if ((len -= copy) == 0)
951 return 0;
952 offset += copy;
953 to += copy;
954 }
955 start = end;
956 }
957
958 if (skb_shinfo(skb)->frag_list) {
959 struct sk_buff *list = skb_shinfo(skb)->frag_list;
960
961 for (; list; list = list->next) {
962 int end;
963
964 BUG_TRAP(start <= offset + len);
965
966 end = start + list->len;
967 if ((copy = end - offset) > 0) {
968 if (copy > len)
969 copy = len;
970 if (skb_copy_bits(list, offset - start,
971 to, copy))
972 goto fault;
973 if ((len -= copy) == 0)
974 return 0;
975 offset += copy;
976 to += copy;
977 }
978 start = end;
979 }
980 }
981 if (!len)
982 return 0;
983
984fault:
985 return -EFAULT;
986}
987
988/* Checksum skb data. */
989
990unsigned int skb_checksum(const struct sk_buff *skb, int offset,
991 int len, unsigned int csum)
992{
993 int start = skb_headlen(skb);
994 int i, copy = start - offset;
995 int pos = 0;
996
997 /* Checksum header. */
998 if (copy > 0) {
999 if (copy > len)
1000 copy = len;
1001 csum = csum_partial(skb->data + offset, copy, csum);
1002 if ((len -= copy) == 0)
1003 return csum;
1004 offset += copy;
1005 pos = copy;
1006 }
1007
1008 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1009 int end;
1010
1011 BUG_TRAP(start <= offset + len);
1012
1013 end = start + skb_shinfo(skb)->frags[i].size;
1014 if ((copy = end - offset) > 0) {
1015 unsigned int csum2;
1016 u8 *vaddr;
1017 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1018
1019 if (copy > len)
1020 copy = len;
1021 vaddr = kmap_skb_frag(frag);
1022 csum2 = csum_partial(vaddr + frag->page_offset +
1023 offset - start, copy, 0);
1024 kunmap_skb_frag(vaddr);
1025 csum = csum_block_add(csum, csum2, pos);
1026 if (!(len -= copy))
1027 return csum;
1028 offset += copy;
1029 pos += copy;
1030 }
1031 start = end;
1032 }
1033
1034 if (skb_shinfo(skb)->frag_list) {
1035 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1036
1037 for (; list; list = list->next) {
1038 int end;
1039
1040 BUG_TRAP(start <= offset + len);
1041
1042 end = start + list->len;
1043 if ((copy = end - offset) > 0) {
1044 unsigned int csum2;
1045 if (copy > len)
1046 copy = len;
1047 csum2 = skb_checksum(list, offset - start,
1048 copy, 0);
1049 csum = csum_block_add(csum, csum2, pos);
1050 if ((len -= copy) == 0)
1051 return csum;
1052 offset += copy;
1053 pos += copy;
1054 }
1055 start = end;
1056 }
1057 }
1058 if (len)
1059 BUG();
1060
1061 return csum;
1062}
1063
1064/* Both of above in one bottle. */
1065
1066unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1067 u8 *to, int len, unsigned int csum)
1068{
1069 int start = skb_headlen(skb);
1070 int i, copy = start - offset;
1071 int pos = 0;
1072
1073 /* Copy header. */
1074 if (copy > 0) {
1075 if (copy > len)
1076 copy = len;
1077 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1078 copy, csum);
1079 if ((len -= copy) == 0)
1080 return csum;
1081 offset += copy;
1082 to += copy;
1083 pos = copy;
1084 }
1085
1086 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1087 int end;
1088
1089 BUG_TRAP(start <= offset + len);
1090
1091 end = start + skb_shinfo(skb)->frags[i].size;
1092 if ((copy = end - offset) > 0) {
1093 unsigned int csum2;
1094 u8 *vaddr;
1095 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1096
1097 if (copy > len)
1098 copy = len;
1099 vaddr = kmap_skb_frag(frag);
1100 csum2 = csum_partial_copy_nocheck(vaddr +
1101 frag->page_offset +
1102 offset - start, to,
1103 copy, 0);
1104 kunmap_skb_frag(vaddr);
1105 csum = csum_block_add(csum, csum2, pos);
1106 if (!(len -= copy))
1107 return csum;
1108 offset += copy;
1109 to += copy;
1110 pos += copy;
1111 }
1112 start = end;
1113 }
1114
1115 if (skb_shinfo(skb)->frag_list) {
1116 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1117
1118 for (; list; list = list->next) {
1119 unsigned int csum2;
1120 int end;
1121
1122 BUG_TRAP(start <= offset + len);
1123
1124 end = start + list->len;
1125 if ((copy = end - offset) > 0) {
1126 if (copy > len)
1127 copy = len;
1128 csum2 = skb_copy_and_csum_bits(list,
1129 offset - start,
1130 to, copy, 0);
1131 csum = csum_block_add(csum, csum2, pos);
1132 if ((len -= copy) == 0)
1133 return csum;
1134 offset += copy;
1135 to += copy;
1136 pos += copy;
1137 }
1138 start = end;
1139 }
1140 }
1141 if (len)
1142 BUG();
1143 return csum;
1144}
1145
1146void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1147{
1148 unsigned int csum;
1149 long csstart;
1150
1151 if (skb->ip_summed == CHECKSUM_HW)
1152 csstart = skb->h.raw - skb->data;
1153 else
1154 csstart = skb_headlen(skb);
1155
1156 if (csstart > skb_headlen(skb))
1157 BUG();
1158
1159 memcpy(to, skb->data, csstart);
1160
1161 csum = 0;
1162 if (csstart != skb->len)
1163 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1164 skb->len - csstart, 0);
1165
1166 if (skb->ip_summed == CHECKSUM_HW) {
1167 long csstuff = csstart + skb->csum;
1168
1169 *((unsigned short *)(to + csstuff)) = csum_fold(csum);
1170 }
1171}
1172
1173/**
1174 * skb_dequeue - remove from the head of the queue
1175 * @list: list to dequeue from
1176 *
1177 * Remove the head of the list. The list lock is taken so the function
1178 * may be used safely with other locking list functions. The head item is
1179 * returned or %NULL if the list is empty.
1180 */
1181
1182struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1183{
1184 unsigned long flags;
1185 struct sk_buff *result;
1186
1187 spin_lock_irqsave(&list->lock, flags);
1188 result = __skb_dequeue(list);
1189 spin_unlock_irqrestore(&list->lock, flags);
1190 return result;
1191}
1192
1193/**
1194 * skb_dequeue_tail - remove from the tail of the queue
1195 * @list: list to dequeue from
1196 *
1197 * Remove the tail of the list. The list lock is taken so the function
1198 * may be used safely with other locking list functions. The tail item is
1199 * returned or %NULL if the list is empty.
1200 */
1201struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1202{
1203 unsigned long flags;
1204 struct sk_buff *result;
1205
1206 spin_lock_irqsave(&list->lock, flags);
1207 result = __skb_dequeue_tail(list);
1208 spin_unlock_irqrestore(&list->lock, flags);
1209 return result;
1210}
1211
1212/**
1213 * skb_queue_purge - empty a list
1214 * @list: list to empty
1215 *
1216 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1217 * the list and one reference dropped. This function takes the list
1218 * lock and is atomic with respect to other list locking functions.
1219 */
1220void skb_queue_purge(struct sk_buff_head *list)
1221{
1222 struct sk_buff *skb;
1223 while ((skb = skb_dequeue(list)) != NULL)
1224 kfree_skb(skb);
1225}
1226
1227/**
1228 * skb_queue_head - queue a buffer at the list head
1229 * @list: list to use
1230 * @newsk: buffer to queue
1231 *
1232 * Queue a buffer at the start of the list. This function takes the
1233 * list lock and can be used safely with other locking &sk_buff functions
1234 * safely.
1235 *
1236 * A buffer cannot be placed on two lists at the same time.
1237 */
1238void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1239{
1240 unsigned long flags;
1241
1242 spin_lock_irqsave(&list->lock, flags);
1243 __skb_queue_head(list, newsk);
1244 spin_unlock_irqrestore(&list->lock, flags);
1245}
1246
1247/**
1248 * skb_queue_tail - queue a buffer at the list tail
1249 * @list: list to use
1250 * @newsk: buffer to queue
1251 *
1252 * Queue a buffer at the tail of the list. This function takes the
1253 * list lock and can be used safely with other locking &sk_buff functions
1254 * safely.
1255 *
1256 * A buffer cannot be placed on two lists at the same time.
1257 */
1258void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1259{
1260 unsigned long flags;
1261
1262 spin_lock_irqsave(&list->lock, flags);
1263 __skb_queue_tail(list, newsk);
1264 spin_unlock_irqrestore(&list->lock, flags);
1265}
1266/**
1267 * skb_unlink - remove a buffer from a list
1268 * @skb: buffer to remove
1269 *
1270 * Place a packet after a given packet in a list. The list locks are taken
1271 * and this function is atomic with respect to other list locked calls
1272 *
1273 * Works even without knowing the list it is sitting on, which can be
1274 * handy at times. It also means that THE LIST MUST EXIST when you
1275 * unlink. Thus a list must have its contents unlinked before it is
1276 * destroyed.
1277 */
1278void skb_unlink(struct sk_buff *skb)
1279{
1280 struct sk_buff_head *list = skb->list;
1281
1282 if (list) {
1283 unsigned long flags;
1284
1285 spin_lock_irqsave(&list->lock, flags);
1286 if (skb->list == list)
1287 __skb_unlink(skb, skb->list);
1288 spin_unlock_irqrestore(&list->lock, flags);
1289 }
1290}
1291
1292
1293/**
1294 * skb_append - append a buffer
1295 * @old: buffer to insert after
1296 * @newsk: buffer to insert
1297 *
1298 * Place a packet after a given packet in a list. The list locks are taken
1299 * and this function is atomic with respect to other list locked calls.
1300 * A buffer cannot be placed on two lists at the same time.
1301 */
1302
1303void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1304{
1305 unsigned long flags;
1306
1307 spin_lock_irqsave(&old->list->lock, flags);
1308 __skb_append(old, newsk);
1309 spin_unlock_irqrestore(&old->list->lock, flags);
1310}
1311
1312
1313/**
1314 * skb_insert - insert a buffer
1315 * @old: buffer to insert before
1316 * @newsk: buffer to insert
1317 *
1318 * Place a packet before a given packet in a list. The list locks are taken
1319 * and this function is atomic with respect to other list locked calls
1320 * A buffer cannot be placed on two lists at the same time.
1321 */
1322
1323void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1324{
1325 unsigned long flags;
1326
1327 spin_lock_irqsave(&old->list->lock, flags);
1328 __skb_insert(newsk, old->prev, old, old->list);
1329 spin_unlock_irqrestore(&old->list->lock, flags);
1330}
1331
1332#if 0
1333/*
1334 * Tune the memory allocator for a new MTU size.
1335 */
1336void skb_add_mtu(int mtu)
1337{
1338 /* Must match allocation in alloc_skb */
1339 mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1340
1341 kmem_add_cache_size(mtu);
1342}
1343#endif
1344
1345static inline void skb_split_inside_header(struct sk_buff *skb,
1346 struct sk_buff* skb1,
1347 const u32 len, const int pos)
1348{
1349 int i;
1350
1351 memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1352
1353 /* And move data appendix as is. */
1354 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1355 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1356
1357 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1358 skb_shinfo(skb)->nr_frags = 0;
1359 skb1->data_len = skb->data_len;
1360 skb1->len += skb1->data_len;
1361 skb->data_len = 0;
1362 skb->len = len;
1363 skb->tail = skb->data + len;
1364}
1365
1366static inline void skb_split_no_header(struct sk_buff *skb,
1367 struct sk_buff* skb1,
1368 const u32 len, int pos)
1369{
1370 int i, k = 0;
1371 const int nfrags = skb_shinfo(skb)->nr_frags;
1372
1373 skb_shinfo(skb)->nr_frags = 0;
1374 skb1->len = skb1->data_len = skb->len - len;
1375 skb->len = len;
1376 skb->data_len = len - pos;
1377
1378 for (i = 0; i < nfrags; i++) {
1379 int size = skb_shinfo(skb)->frags[i].size;
1380
1381 if (pos + size > len) {
1382 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1383
1384 if (pos < len) {
1385 /* Split frag.
1386 * We have two variants in this case:
1387 * 1. Move all the frag to the second
1388 * part, if it is possible. F.e.
1389 * this approach is mandatory for TUX,
1390 * where splitting is expensive.
1391 * 2. Split is accurately. We make this.
1392 */
1393 get_page(skb_shinfo(skb)->frags[i].page);
1394 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1395 skb_shinfo(skb1)->frags[0].size -= len - pos;
1396 skb_shinfo(skb)->frags[i].size = len - pos;
1397 skb_shinfo(skb)->nr_frags++;
1398 }
1399 k++;
1400 } else
1401 skb_shinfo(skb)->nr_frags++;
1402 pos += size;
1403 }
1404 skb_shinfo(skb1)->nr_frags = k;
1405}
1406
1407/**
1408 * skb_split - Split fragmented skb to two parts at length len.
1409 * @skb: the buffer to split
1410 * @skb1: the buffer to receive the second part
1411 * @len: new length for skb
1412 */
1413void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1414{
1415 int pos = skb_headlen(skb);
1416
1417 if (len < pos) /* Split line is inside header. */
1418 skb_split_inside_header(skb, skb1, len, pos);
1419 else /* Second chunk has no header, nothing to copy. */
1420 skb_split_no_header(skb, skb1, len, pos);
1421}
1422
1423void __init skb_init(void)
1424{
1425 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1426 sizeof(struct sk_buff),
1427 0,
1428 SLAB_HWCACHE_ALIGN,
1429 NULL, NULL);
1430 if (!skbuff_head_cache)
1431 panic("cannot create skbuff cache");
1432}
1433
1434EXPORT_SYMBOL(___pskb_trim);
1435EXPORT_SYMBOL(__kfree_skb);
1436EXPORT_SYMBOL(__pskb_pull_tail);
1437EXPORT_SYMBOL(alloc_skb);
1438EXPORT_SYMBOL(pskb_copy);
1439EXPORT_SYMBOL(pskb_expand_head);
1440EXPORT_SYMBOL(skb_checksum);
1441EXPORT_SYMBOL(skb_clone);
1442EXPORT_SYMBOL(skb_clone_fraglist);
1443EXPORT_SYMBOL(skb_copy);
1444EXPORT_SYMBOL(skb_copy_and_csum_bits);
1445EXPORT_SYMBOL(skb_copy_and_csum_dev);
1446EXPORT_SYMBOL(skb_copy_bits);
1447EXPORT_SYMBOL(skb_copy_expand);
1448EXPORT_SYMBOL(skb_over_panic);
1449EXPORT_SYMBOL(skb_pad);
1450EXPORT_SYMBOL(skb_realloc_headroom);
1451EXPORT_SYMBOL(skb_under_panic);
1452EXPORT_SYMBOL(skb_dequeue);
1453EXPORT_SYMBOL(skb_dequeue_tail);
1454EXPORT_SYMBOL(skb_insert);
1455EXPORT_SYMBOL(skb_queue_purge);
1456EXPORT_SYMBOL(skb_queue_head);
1457EXPORT_SYMBOL(skb_queue_tail);
1458EXPORT_SYMBOL(skb_unlink);
1459EXPORT_SYMBOL(skb_append);
1460EXPORT_SYMBOL(skb_split);
generated by cgit v1.2.2 (git 2.25.0) at 2024-12-22 19:41:17 -0500