11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Authors: Alan Cox <iiitac@pyr.swan.ac.uk> 51da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Version: $Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $ 81da177e4SLinus Torvalds * 91da177e4SLinus Torvalds * Fixes: 101da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 111da177e4SLinus Torvalds * balancer bugs. 121da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 131da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 141da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 151da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 161da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 171da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 181da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 191da177e4SLinus Torvalds * only put in the headers 201da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 211da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 221da177e4SLinus Torvalds * Andi Kleen : slabified it. 231da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 241da177e4SLinus Torvalds * 251da177e4SLinus Torvalds * NOTE: 261da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 271da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 281da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 291da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 301da177e4SLinus Torvalds * 311da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 321da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 331da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 341da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 351da177e4SLinus Torvalds */ 361da177e4SLinus Torvalds 371da177e4SLinus Torvalds /* 381da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 391da177e4SLinus Torvalds */ 401da177e4SLinus Torvalds 411da177e4SLinus Torvalds #include <linux/module.h> 421da177e4SLinus Torvalds #include <linux/types.h> 431da177e4SLinus Torvalds #include <linux/kernel.h> 441da177e4SLinus Torvalds #include <linux/sched.h> 451da177e4SLinus Torvalds #include <linux/mm.h> 461da177e4SLinus Torvalds #include <linux/interrupt.h> 471da177e4SLinus Torvalds #include <linux/in.h> 481da177e4SLinus Torvalds #include <linux/inet.h> 491da177e4SLinus Torvalds #include <linux/slab.h> 501da177e4SLinus Torvalds #include <linux/netdevice.h> 511da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 521da177e4SLinus Torvalds #include <net/pkt_sched.h> 531da177e4SLinus Torvalds #endif 541da177e4SLinus Torvalds #include <linux/string.h> 551da177e4SLinus Torvalds #include <linux/skbuff.h> 561da177e4SLinus Torvalds #include <linux/cache.h> 571da177e4SLinus Torvalds #include <linux/rtnetlink.h> 581da177e4SLinus Torvalds #include <linux/init.h> 591da177e4SLinus Torvalds 601da177e4SLinus Torvalds #include <net/protocol.h> 611da177e4SLinus Torvalds #include <net/dst.h> 621da177e4SLinus Torvalds #include <net/sock.h> 631da177e4SLinus Torvalds #include <net/checksum.h> 641da177e4SLinus Torvalds #include <net/xfrm.h> 651da177e4SLinus Torvalds 661da177e4SLinus Torvalds #include <asm/uaccess.h> 671da177e4SLinus Torvalds #include <asm/system.h> 681da177e4SLinus Torvalds 69a1f8e7f7SAl Viro #include "kmap_skb.h" 70a1f8e7f7SAl Viro 71ba89966cSEric Dumazet static kmem_cache_t *skbuff_head_cache __read_mostly; 72ba89966cSEric Dumazet static kmem_cache_t *skbuff_fclone_cache __read_mostly; 731da177e4SLinus Torvalds 741da177e4SLinus Torvalds /* 751da177e4SLinus Torvalds * Keep out-of-line to prevent kernel bloat. 761da177e4SLinus Torvalds * __builtin_return_address is not used because it is not always 771da177e4SLinus Torvalds * reliable. 781da177e4SLinus Torvalds */ 791da177e4SLinus Torvalds 801da177e4SLinus Torvalds /** 811da177e4SLinus Torvalds * skb_over_panic - private function 821da177e4SLinus Torvalds * @skb: buffer 831da177e4SLinus Torvalds * @sz: size 841da177e4SLinus Torvalds * @here: address 851da177e4SLinus Torvalds * 861da177e4SLinus Torvalds * Out of line support code for skb_put(). Not user callable. 871da177e4SLinus Torvalds */ 881da177e4SLinus Torvalds void skb_over_panic(struct sk_buff *skb, int sz, void *here) 891da177e4SLinus Torvalds { 9026095455SPatrick McHardy printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " 9126095455SPatrick McHardy "data:%p tail:%p end:%p dev:%s\n", 9226095455SPatrick McHardy here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end, 9326095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 941da177e4SLinus Torvalds BUG(); 951da177e4SLinus Torvalds } 961da177e4SLinus Torvalds 971da177e4SLinus Torvalds /** 981da177e4SLinus Torvalds * skb_under_panic - private function 991da177e4SLinus Torvalds * @skb: buffer 1001da177e4SLinus Torvalds * @sz: size 1011da177e4SLinus Torvalds * @here: address 1021da177e4SLinus Torvalds * 1031da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1041da177e4SLinus Torvalds */ 1051da177e4SLinus Torvalds 1061da177e4SLinus Torvalds void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1071da177e4SLinus Torvalds { 10826095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 10926095455SPatrick McHardy "data:%p tail:%p end:%p dev:%s\n", 11026095455SPatrick McHardy here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end, 11126095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1121da177e4SLinus Torvalds BUG(); 1131da177e4SLinus Torvalds } 1141da177e4SLinus Torvalds 115dc6de336SDavid S. Miller void skb_truesize_bug(struct sk_buff *skb) 116dc6de336SDavid S. Miller { 117dc6de336SDavid S. Miller printk(KERN_ERR "SKB BUG: Invalid truesize (%u) " 118dc6de336SDavid S. Miller "len=%u, sizeof(sk_buff)=%Zd\n", 119dc6de336SDavid S. Miller skb->truesize, skb->len, sizeof(struct sk_buff)); 120dc6de336SDavid S. Miller } 121dc6de336SDavid S. Miller EXPORT_SYMBOL(skb_truesize_bug); 122dc6de336SDavid S. Miller 1231da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1241da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1251da177e4SLinus Torvalds * [BEEP] leaks. 1261da177e4SLinus Torvalds * 1271da177e4SLinus Torvalds */ 1281da177e4SLinus Torvalds 1291da177e4SLinus Torvalds /** 130d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1311da177e4SLinus Torvalds * @size: size to allocate 1321da177e4SLinus Torvalds * @gfp_mask: allocation mask 133c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 134c83c2486SRandy Dunlap * and allocate a cloned (child) skb 135b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1361da177e4SLinus Torvalds * 1371da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1381da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1391da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1401da177e4SLinus Torvalds * 1411da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1421da177e4SLinus Torvalds * %GFP_ATOMIC. 1431da177e4SLinus Torvalds */ 144dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 145b30973f8SChristoph Hellwig int fclone, int node) 1461da177e4SLinus Torvalds { 1478798b3fbSHerbert Xu kmem_cache_t *cache; 1484947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1491da177e4SLinus Torvalds struct sk_buff *skb; 1501da177e4SLinus Torvalds u8 *data; 1511da177e4SLinus Torvalds 1528798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1538798b3fbSHerbert Xu 1541da177e4SLinus Torvalds /* Get the HEAD */ 155b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1561da177e4SLinus Torvalds if (!skb) 1571da177e4SLinus Torvalds goto out; 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds /* Get the DATA. Size must match skb_add_mtu(). */ 1601da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 161b30973f8SChristoph Hellwig data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), 162b30973f8SChristoph Hellwig gfp_mask, node); 1631da177e4SLinus Torvalds if (!data) 1641da177e4SLinus Torvalds goto nodata; 1651da177e4SLinus Torvalds 1661da177e4SLinus Torvalds memset(skb, 0, offsetof(struct sk_buff, truesize)); 1671da177e4SLinus Torvalds skb->truesize = size + sizeof(struct sk_buff); 1681da177e4SLinus Torvalds atomic_set(&skb->users, 1); 1691da177e4SLinus Torvalds skb->head = data; 1701da177e4SLinus Torvalds skb->data = data; 1711da177e4SLinus Torvalds skb->tail = data; 1721da177e4SLinus Torvalds skb->end = data + size; 1734947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 1744947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 1754947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 1764947d3efSBenjamin LaHaise shinfo->nr_frags = 0; 1777967168cSHerbert Xu shinfo->gso_size = 0; 1787967168cSHerbert Xu shinfo->gso_segs = 0; 1797967168cSHerbert Xu shinfo->gso_type = 0; 1804947d3efSBenjamin LaHaise shinfo->ip6_frag_id = 0; 1814947d3efSBenjamin LaHaise shinfo->frag_list = NULL; 1824947d3efSBenjamin LaHaise 183d179cd12SDavid S. Miller if (fclone) { 184d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 185d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 1861da177e4SLinus Torvalds 187d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 188d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 189d179cd12SDavid S. Miller 190d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 191d179cd12SDavid S. Miller } 1921da177e4SLinus Torvalds out: 1931da177e4SLinus Torvalds return skb; 1941da177e4SLinus Torvalds nodata: 1958798b3fbSHerbert Xu kmem_cache_free(cache, skb); 1961da177e4SLinus Torvalds skb = NULL; 1971da177e4SLinus Torvalds goto out; 1981da177e4SLinus Torvalds } 1991da177e4SLinus Torvalds 2001da177e4SLinus Torvalds /** 2011da177e4SLinus Torvalds * alloc_skb_from_cache - allocate a network buffer 2021da177e4SLinus Torvalds * @cp: kmem_cache from which to allocate the data area 2031da177e4SLinus Torvalds * (object size must be big enough for @size bytes + skb overheads) 2041da177e4SLinus Torvalds * @size: size to allocate 2051da177e4SLinus Torvalds * @gfp_mask: allocation mask 2061da177e4SLinus Torvalds * 2071da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and 2081da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 2091da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 2101da177e4SLinus Torvalds * 2111da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 2121da177e4SLinus Torvalds * %GFP_ATOMIC. 2131da177e4SLinus Torvalds */ 2141da177e4SLinus Torvalds struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp, 21586a76cafSVictor Fusco unsigned int size, 216dd0fc66fSAl Viro gfp_t gfp_mask) 2171da177e4SLinus Torvalds { 2181da177e4SLinus Torvalds struct sk_buff *skb; 2191da177e4SLinus Torvalds u8 *data; 2201da177e4SLinus Torvalds 2211da177e4SLinus Torvalds /* Get the HEAD */ 2221da177e4SLinus Torvalds skb = kmem_cache_alloc(skbuff_head_cache, 2231da177e4SLinus Torvalds gfp_mask & ~__GFP_DMA); 2241da177e4SLinus Torvalds if (!skb) 2251da177e4SLinus Torvalds goto out; 2261da177e4SLinus Torvalds 2271da177e4SLinus Torvalds /* Get the DATA. */ 2281da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 2291da177e4SLinus Torvalds data = kmem_cache_alloc(cp, gfp_mask); 2301da177e4SLinus Torvalds if (!data) 2311da177e4SLinus Torvalds goto nodata; 2321da177e4SLinus Torvalds 2331da177e4SLinus Torvalds memset(skb, 0, offsetof(struct sk_buff, truesize)); 2341da177e4SLinus Torvalds skb->truesize = size + sizeof(struct sk_buff); 2351da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2361da177e4SLinus Torvalds skb->head = data; 2371da177e4SLinus Torvalds skb->data = data; 2381da177e4SLinus Torvalds skb->tail = data; 2391da177e4SLinus Torvalds skb->end = data + size; 2401da177e4SLinus Torvalds 2411da177e4SLinus Torvalds atomic_set(&(skb_shinfo(skb)->dataref), 1); 2421da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 2437967168cSHerbert Xu skb_shinfo(skb)->gso_size = 0; 2447967168cSHerbert Xu skb_shinfo(skb)->gso_segs = 0; 2457967168cSHerbert Xu skb_shinfo(skb)->gso_type = 0; 2461da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = NULL; 2471da177e4SLinus Torvalds out: 2481da177e4SLinus Torvalds return skb; 2491da177e4SLinus Torvalds nodata: 2501da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 2511da177e4SLinus Torvalds skb = NULL; 2521da177e4SLinus Torvalds goto out; 2531da177e4SLinus Torvalds } 2541da177e4SLinus Torvalds 2558af27456SChristoph Hellwig /** 2568af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2578af27456SChristoph Hellwig * @dev: network device to receive on 2588af27456SChristoph Hellwig * @length: length to allocate 2598af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 2608af27456SChristoph Hellwig * 2618af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 2628af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 2638af27456SChristoph Hellwig * the headroom they think they need without accounting for the 2648af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 2658af27456SChristoph Hellwig * 2668af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 2678af27456SChristoph Hellwig */ 2688af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 2698af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 2708af27456SChristoph Hellwig { 271b30973f8SChristoph Hellwig int node = dev->class_dev.dev ? dev_to_node(dev->class_dev.dev) : -1; 2728af27456SChristoph Hellwig struct sk_buff *skb; 2738af27456SChristoph Hellwig 274b30973f8SChristoph Hellwig skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node); 2757b2e497aSChristoph Hellwig if (likely(skb)) { 2768af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 2777b2e497aSChristoph Hellwig skb->dev = dev; 2787b2e497aSChristoph Hellwig } 2798af27456SChristoph Hellwig return skb; 2808af27456SChristoph Hellwig } 2811da177e4SLinus Torvalds 28227b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 2831da177e4SLinus Torvalds { 28427b437c8SHerbert Xu struct sk_buff *list = *listp; 2851da177e4SLinus Torvalds 28627b437c8SHerbert Xu *listp = NULL; 2871da177e4SLinus Torvalds 2881da177e4SLinus Torvalds do { 2891da177e4SLinus Torvalds struct sk_buff *this = list; 2901da177e4SLinus Torvalds list = list->next; 2911da177e4SLinus Torvalds kfree_skb(this); 2921da177e4SLinus Torvalds } while (list); 2931da177e4SLinus Torvalds } 2941da177e4SLinus Torvalds 29527b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 29627b437c8SHerbert Xu { 29727b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 29827b437c8SHerbert Xu } 29927b437c8SHerbert Xu 3001da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3011da177e4SLinus Torvalds { 3021da177e4SLinus Torvalds struct sk_buff *list; 3031da177e4SLinus Torvalds 3041da177e4SLinus Torvalds for (list = skb_shinfo(skb)->frag_list; list; list = list->next) 3051da177e4SLinus Torvalds skb_get(list); 3061da177e4SLinus Torvalds } 3071da177e4SLinus Torvalds 3085bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3091da177e4SLinus Torvalds { 3101da177e4SLinus Torvalds if (!skb->cloned || 3111da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3121da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3131da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3141da177e4SLinus Torvalds int i; 3151da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 3161da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds 3191da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 3201da177e4SLinus Torvalds skb_drop_fraglist(skb); 3211da177e4SLinus Torvalds 3221da177e4SLinus Torvalds kfree(skb->head); 3231da177e4SLinus Torvalds } 3241da177e4SLinus Torvalds } 3251da177e4SLinus Torvalds 3261da177e4SLinus Torvalds /* 3271da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 3281da177e4SLinus Torvalds */ 3291da177e4SLinus Torvalds void kfree_skbmem(struct sk_buff *skb) 3301da177e4SLinus Torvalds { 331d179cd12SDavid S. Miller struct sk_buff *other; 332d179cd12SDavid S. Miller atomic_t *fclone_ref; 333d179cd12SDavid S. Miller 3341da177e4SLinus Torvalds skb_release_data(skb); 335d179cd12SDavid S. Miller switch (skb->fclone) { 336d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 3371da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 338d179cd12SDavid S. Miller break; 339d179cd12SDavid S. Miller 340d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 341d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 342d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 343d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 344d179cd12SDavid S. Miller break; 345d179cd12SDavid S. Miller 346d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 347d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 348d179cd12SDavid S. Miller other = skb - 1; 349d179cd12SDavid S. Miller 350d179cd12SDavid S. Miller /* The clone portion is available for 351d179cd12SDavid S. Miller * fast-cloning again. 352d179cd12SDavid S. Miller */ 353d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 354d179cd12SDavid S. Miller 355d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 356d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 357d179cd12SDavid S. Miller break; 358d179cd12SDavid S. Miller }; 3591da177e4SLinus Torvalds } 3601da177e4SLinus Torvalds 3611da177e4SLinus Torvalds /** 3621da177e4SLinus Torvalds * __kfree_skb - private function 3631da177e4SLinus Torvalds * @skb: buffer 3641da177e4SLinus Torvalds * 3651da177e4SLinus Torvalds * Free an sk_buff. Release anything attached to the buffer. 3661da177e4SLinus Torvalds * Clean the state. This is an internal helper function. Users should 3671da177e4SLinus Torvalds * always call kfree_skb 3681da177e4SLinus Torvalds */ 3691da177e4SLinus Torvalds 3701da177e4SLinus Torvalds void __kfree_skb(struct sk_buff *skb) 3711da177e4SLinus Torvalds { 3721da177e4SLinus Torvalds dst_release(skb->dst); 3731da177e4SLinus Torvalds #ifdef CONFIG_XFRM 3741da177e4SLinus Torvalds secpath_put(skb->sp); 3751da177e4SLinus Torvalds #endif 3761da177e4SLinus Torvalds if (skb->destructor) { 3779c2b3328SStephen Hemminger WARN_ON(in_irq()); 3781da177e4SLinus Torvalds skb->destructor(skb); 3791da177e4SLinus Torvalds } 3801da177e4SLinus Torvalds #ifdef CONFIG_NETFILTER 3811da177e4SLinus Torvalds nf_conntrack_put(skb->nfct); 3829fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3839fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 3849fb9cbb1SYasuyuki Kozakai #endif 3851da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 3861da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 3871da177e4SLinus Torvalds #endif 3881da177e4SLinus Torvalds #endif 3891da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 3901da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 3911da177e4SLinus Torvalds skb->tc_index = 0; 3921da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 3931da177e4SLinus Torvalds skb->tc_verd = 0; 3941da177e4SLinus Torvalds #endif 3951da177e4SLinus Torvalds #endif 3961da177e4SLinus Torvalds 3971da177e4SLinus Torvalds kfree_skbmem(skb); 3981da177e4SLinus Torvalds } 3991da177e4SLinus Torvalds 4001da177e4SLinus Torvalds /** 401231d06aeSJörn Engel * kfree_skb - free an sk_buff 402231d06aeSJörn Engel * @skb: buffer to free 403231d06aeSJörn Engel * 404231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 405231d06aeSJörn Engel * hit zero. 406231d06aeSJörn Engel */ 407231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 408231d06aeSJörn Engel { 409231d06aeSJörn Engel if (unlikely(!skb)) 410231d06aeSJörn Engel return; 411231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 412231d06aeSJörn Engel smp_rmb(); 413231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 414231d06aeSJörn Engel return; 415231d06aeSJörn Engel __kfree_skb(skb); 416231d06aeSJörn Engel } 417231d06aeSJörn Engel 418231d06aeSJörn Engel /** 4191da177e4SLinus Torvalds * skb_clone - duplicate an sk_buff 4201da177e4SLinus Torvalds * @skb: buffer to clone 4211da177e4SLinus Torvalds * @gfp_mask: allocation priority 4221da177e4SLinus Torvalds * 4231da177e4SLinus Torvalds * Duplicate an &sk_buff. The new one is not owned by a socket. Both 4241da177e4SLinus Torvalds * copies share the same packet data but not structure. The new 4251da177e4SLinus Torvalds * buffer has a reference count of 1. If the allocation fails the 4261da177e4SLinus Torvalds * function returns %NULL otherwise the new buffer is returned. 4271da177e4SLinus Torvalds * 4281da177e4SLinus Torvalds * If this function is called from an interrupt gfp_mask() must be 4291da177e4SLinus Torvalds * %GFP_ATOMIC. 4301da177e4SLinus Torvalds */ 4311da177e4SLinus Torvalds 432dd0fc66fSAl Viro struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 4331da177e4SLinus Torvalds { 434d179cd12SDavid S. Miller struct sk_buff *n; 4351da177e4SLinus Torvalds 436d179cd12SDavid S. Miller n = skb + 1; 437d179cd12SDavid S. Miller if (skb->fclone == SKB_FCLONE_ORIG && 438d179cd12SDavid S. Miller n->fclone == SKB_FCLONE_UNAVAILABLE) { 439d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (n + 1); 440d179cd12SDavid S. Miller n->fclone = SKB_FCLONE_CLONE; 441d179cd12SDavid S. Miller atomic_inc(fclone_ref); 442d179cd12SDavid S. Miller } else { 443d179cd12SDavid S. Miller n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 4441da177e4SLinus Torvalds if (!n) 4451da177e4SLinus Torvalds return NULL; 446d179cd12SDavid S. Miller n->fclone = SKB_FCLONE_UNAVAILABLE; 447d179cd12SDavid S. Miller } 4481da177e4SLinus Torvalds 4491da177e4SLinus Torvalds #define C(x) n->x = skb->x 4501da177e4SLinus Torvalds 4511da177e4SLinus Torvalds n->next = n->prev = NULL; 4521da177e4SLinus Torvalds n->sk = NULL; 453a61bbcf2SPatrick McHardy C(tstamp); 4541da177e4SLinus Torvalds C(dev); 4551da177e4SLinus Torvalds C(h); 4561da177e4SLinus Torvalds C(nh); 4571da177e4SLinus Torvalds C(mac); 4581da177e4SLinus Torvalds C(dst); 4591da177e4SLinus Torvalds dst_clone(skb->dst); 4601da177e4SLinus Torvalds C(sp); 4611da177e4SLinus Torvalds #ifdef CONFIG_INET 4621da177e4SLinus Torvalds secpath_get(skb->sp); 4631da177e4SLinus Torvalds #endif 4641da177e4SLinus Torvalds memcpy(n->cb, skb->cb, sizeof(skb->cb)); 4651da177e4SLinus Torvalds C(len); 4661da177e4SLinus Torvalds C(data_len); 4671da177e4SLinus Torvalds C(csum); 4681da177e4SLinus Torvalds C(local_df); 4691da177e4SLinus Torvalds n->cloned = 1; 4701da177e4SLinus Torvalds n->nohdr = 0; 4711da177e4SLinus Torvalds C(pkt_type); 4721da177e4SLinus Torvalds C(ip_summed); 4731da177e4SLinus Torvalds C(priority); 474a8372f03SYOSHIFUJI Hideaki #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 475a8372f03SYOSHIFUJI Hideaki C(ipvs_property); 476a8372f03SYOSHIFUJI Hideaki #endif 4771da177e4SLinus Torvalds C(protocol); 4781da177e4SLinus Torvalds n->destructor = NULL; 47982e91ffeSThomas Graf C(mark); 4801da177e4SLinus Torvalds #ifdef CONFIG_NETFILTER 4811da177e4SLinus Torvalds C(nfct); 4821da177e4SLinus Torvalds nf_conntrack_get(skb->nfct); 4831da177e4SLinus Torvalds C(nfctinfo); 4849fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4859fb9cbb1SYasuyuki Kozakai C(nfct_reasm); 4869fb9cbb1SYasuyuki Kozakai nf_conntrack_get_reasm(skb->nfct_reasm); 4879fb9cbb1SYasuyuki Kozakai #endif 4881da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 4891da177e4SLinus Torvalds C(nf_bridge); 4901da177e4SLinus Torvalds nf_bridge_get(skb->nf_bridge); 4911da177e4SLinus Torvalds #endif 4921da177e4SLinus Torvalds #endif /*CONFIG_NETFILTER*/ 4931da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4941da177e4SLinus Torvalds C(tc_index); 4951da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 4961da177e4SLinus Torvalds n->tc_verd = SET_TC_VERD(skb->tc_verd,0); 497b72f6eccSDavid S. Miller n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd); 498b72f6eccSDavid S. Miller n->tc_verd = CLR_TC_MUNGED(n->tc_verd); 4991da177e4SLinus Torvalds C(input_dev); 5001da177e4SLinus Torvalds #endif 501984bc16cSJames Morris skb_copy_secmark(n, skb); 5021da177e4SLinus Torvalds #endif 5031da177e4SLinus Torvalds C(truesize); 5041da177e4SLinus Torvalds atomic_set(&n->users, 1); 5051da177e4SLinus Torvalds C(head); 5061da177e4SLinus Torvalds C(data); 5071da177e4SLinus Torvalds C(tail); 5081da177e4SLinus Torvalds C(end); 5091da177e4SLinus Torvalds 5101da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 5111da177e4SLinus Torvalds skb->cloned = 1; 5121da177e4SLinus Torvalds 5131da177e4SLinus Torvalds return n; 5141da177e4SLinus Torvalds } 5151da177e4SLinus Torvalds 5161da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 5171da177e4SLinus Torvalds { 5181da177e4SLinus Torvalds /* 5191da177e4SLinus Torvalds * Shift between the two data areas in bytes 5201da177e4SLinus Torvalds */ 5211da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 5221da177e4SLinus Torvalds 5231da177e4SLinus Torvalds new->sk = NULL; 5241da177e4SLinus Torvalds new->dev = old->dev; 5251da177e4SLinus Torvalds new->priority = old->priority; 5261da177e4SLinus Torvalds new->protocol = old->protocol; 5271da177e4SLinus Torvalds new->dst = dst_clone(old->dst); 5281da177e4SLinus Torvalds #ifdef CONFIG_INET 5291da177e4SLinus Torvalds new->sp = secpath_get(old->sp); 5301da177e4SLinus Torvalds #endif 5311da177e4SLinus Torvalds new->h.raw = old->h.raw + offset; 5321da177e4SLinus Torvalds new->nh.raw = old->nh.raw + offset; 5331da177e4SLinus Torvalds new->mac.raw = old->mac.raw + offset; 5341da177e4SLinus Torvalds memcpy(new->cb, old->cb, sizeof(old->cb)); 5351da177e4SLinus Torvalds new->local_df = old->local_df; 536d179cd12SDavid S. Miller new->fclone = SKB_FCLONE_UNAVAILABLE; 5371da177e4SLinus Torvalds new->pkt_type = old->pkt_type; 538a61bbcf2SPatrick McHardy new->tstamp = old->tstamp; 5391da177e4SLinus Torvalds new->destructor = NULL; 54082e91ffeSThomas Graf new->mark = old->mark; 5411da177e4SLinus Torvalds #ifdef CONFIG_NETFILTER 5421da177e4SLinus Torvalds new->nfct = old->nfct; 5431da177e4SLinus Torvalds nf_conntrack_get(old->nfct); 5441da177e4SLinus Torvalds new->nfctinfo = old->nfctinfo; 5459fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 5469fb9cbb1SYasuyuki Kozakai new->nfct_reasm = old->nfct_reasm; 5479fb9cbb1SYasuyuki Kozakai nf_conntrack_get_reasm(old->nfct_reasm); 5489fb9cbb1SYasuyuki Kozakai #endif 549c98d80edSJulian Anastasov #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 550c98d80edSJulian Anastasov new->ipvs_property = old->ipvs_property; 551c98d80edSJulian Anastasov #endif 5521da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 5531da177e4SLinus Torvalds new->nf_bridge = old->nf_bridge; 5541da177e4SLinus Torvalds nf_bridge_get(old->nf_bridge); 5551da177e4SLinus Torvalds #endif 5561da177e4SLinus Torvalds #endif 5571da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 5581da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 5591da177e4SLinus Torvalds new->tc_verd = old->tc_verd; 5601da177e4SLinus Torvalds #endif 5611da177e4SLinus Torvalds new->tc_index = old->tc_index; 5621da177e4SLinus Torvalds #endif 563984bc16cSJames Morris skb_copy_secmark(new, old); 5641da177e4SLinus Torvalds atomic_set(&new->users, 1); 5657967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 5667967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 5677967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 5681da177e4SLinus Torvalds } 5691da177e4SLinus Torvalds 5701da177e4SLinus Torvalds /** 5711da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 5721da177e4SLinus Torvalds * @skb: buffer to copy 5731da177e4SLinus Torvalds * @gfp_mask: allocation priority 5741da177e4SLinus Torvalds * 5751da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 5761da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 5771da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 5781da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 5791da177e4SLinus Torvalds * 5801da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 5811da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 5821da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 5831da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 5841da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 5851da177e4SLinus Torvalds */ 5861da177e4SLinus Torvalds 587dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 5881da177e4SLinus Torvalds { 5891da177e4SLinus Torvalds int headerlen = skb->data - skb->head; 5901da177e4SLinus Torvalds /* 5911da177e4SLinus Torvalds * Allocate the copy buffer 5921da177e4SLinus Torvalds */ 5931da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len, 5941da177e4SLinus Torvalds gfp_mask); 5951da177e4SLinus Torvalds if (!n) 5961da177e4SLinus Torvalds return NULL; 5971da177e4SLinus Torvalds 5981da177e4SLinus Torvalds /* Set the data pointer */ 5991da177e4SLinus Torvalds skb_reserve(n, headerlen); 6001da177e4SLinus Torvalds /* Set the tail pointer and length */ 6011da177e4SLinus Torvalds skb_put(n, skb->len); 6021da177e4SLinus Torvalds n->csum = skb->csum; 6031da177e4SLinus Torvalds n->ip_summed = skb->ip_summed; 6041da177e4SLinus Torvalds 6051da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 6061da177e4SLinus Torvalds BUG(); 6071da177e4SLinus Torvalds 6081da177e4SLinus Torvalds copy_skb_header(n, skb); 6091da177e4SLinus Torvalds return n; 6101da177e4SLinus Torvalds } 6111da177e4SLinus Torvalds 6121da177e4SLinus Torvalds 6131da177e4SLinus Torvalds /** 6141da177e4SLinus Torvalds * pskb_copy - create copy of an sk_buff with private head. 6151da177e4SLinus Torvalds * @skb: buffer to copy 6161da177e4SLinus Torvalds * @gfp_mask: allocation priority 6171da177e4SLinus Torvalds * 6181da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 6191da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 6201da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 6211da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 6221da177e4SLinus Torvalds * or the pointer to the buffer on success. 6231da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 6241da177e4SLinus Torvalds */ 6251da177e4SLinus Torvalds 626dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) 6271da177e4SLinus Torvalds { 6281da177e4SLinus Torvalds /* 6291da177e4SLinus Torvalds * Allocate the copy buffer 6301da177e4SLinus Torvalds */ 6311da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask); 6321da177e4SLinus Torvalds 6331da177e4SLinus Torvalds if (!n) 6341da177e4SLinus Torvalds goto out; 6351da177e4SLinus Torvalds 6361da177e4SLinus Torvalds /* Set the data pointer */ 6371da177e4SLinus Torvalds skb_reserve(n, skb->data - skb->head); 6381da177e4SLinus Torvalds /* Set the tail pointer and length */ 6391da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 6401da177e4SLinus Torvalds /* Copy the bytes */ 6411da177e4SLinus Torvalds memcpy(n->data, skb->data, n->len); 6421da177e4SLinus Torvalds n->csum = skb->csum; 6431da177e4SLinus Torvalds n->ip_summed = skb->ip_summed; 6441da177e4SLinus Torvalds 64525f484a6SHerbert Xu n->truesize += skb->data_len; 6461da177e4SLinus Torvalds n->data_len = skb->data_len; 6471da177e4SLinus Torvalds n->len = skb->len; 6481da177e4SLinus Torvalds 6491da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 6501da177e4SLinus Torvalds int i; 6511da177e4SLinus Torvalds 6521da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 6531da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 6541da177e4SLinus Torvalds get_page(skb_shinfo(n)->frags[i].page); 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 6571da177e4SLinus Torvalds } 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 6601da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 6611da177e4SLinus Torvalds skb_clone_fraglist(n); 6621da177e4SLinus Torvalds } 6631da177e4SLinus Torvalds 6641da177e4SLinus Torvalds copy_skb_header(n, skb); 6651da177e4SLinus Torvalds out: 6661da177e4SLinus Torvalds return n; 6671da177e4SLinus Torvalds } 6681da177e4SLinus Torvalds 6691da177e4SLinus Torvalds /** 6701da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 6711da177e4SLinus Torvalds * @skb: buffer to reallocate 6721da177e4SLinus Torvalds * @nhead: room to add at head 6731da177e4SLinus Torvalds * @ntail: room to add at tail 6741da177e4SLinus Torvalds * @gfp_mask: allocation priority 6751da177e4SLinus Torvalds * 6761da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 6771da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 6781da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 6791da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 6801da177e4SLinus Torvalds * 6811da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 6821da177e4SLinus Torvalds * reloaded after call to this function. 6831da177e4SLinus Torvalds */ 6841da177e4SLinus Torvalds 68586a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 686dd0fc66fSAl Viro gfp_t gfp_mask) 6871da177e4SLinus Torvalds { 6881da177e4SLinus Torvalds int i; 6891da177e4SLinus Torvalds u8 *data; 6901da177e4SLinus Torvalds int size = nhead + (skb->end - skb->head) + ntail; 6911da177e4SLinus Torvalds long off; 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds if (skb_shared(skb)) 6941da177e4SLinus Torvalds BUG(); 6951da177e4SLinus Torvalds 6961da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 6991da177e4SLinus Torvalds if (!data) 7001da177e4SLinus Torvalds goto nodata; 7011da177e4SLinus Torvalds 7021da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 7031da177e4SLinus Torvalds * optimized for the cases when header is void. */ 7041da177e4SLinus Torvalds memcpy(data + nhead, skb->head, skb->tail - skb->head); 7051da177e4SLinus Torvalds memcpy(data + size, skb->end, sizeof(struct skb_shared_info)); 7061da177e4SLinus Torvalds 7071da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 7081da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 7091da177e4SLinus Torvalds 7101da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 7111da177e4SLinus Torvalds skb_clone_fraglist(skb); 7121da177e4SLinus Torvalds 7131da177e4SLinus Torvalds skb_release_data(skb); 7141da177e4SLinus Torvalds 7151da177e4SLinus Torvalds off = (data + nhead) - skb->head; 7161da177e4SLinus Torvalds 7171da177e4SLinus Torvalds skb->head = data; 7181da177e4SLinus Torvalds skb->end = data + size; 7191da177e4SLinus Torvalds skb->data += off; 7201da177e4SLinus Torvalds skb->tail += off; 7211da177e4SLinus Torvalds skb->mac.raw += off; 7221da177e4SLinus Torvalds skb->h.raw += off; 7231da177e4SLinus Torvalds skb->nh.raw += off; 7241da177e4SLinus Torvalds skb->cloned = 0; 7251da177e4SLinus Torvalds skb->nohdr = 0; 7261da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 7271da177e4SLinus Torvalds return 0; 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds nodata: 7301da177e4SLinus Torvalds return -ENOMEM; 7311da177e4SLinus Torvalds } 7321da177e4SLinus Torvalds 7331da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 7341da177e4SLinus Torvalds 7351da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 7361da177e4SLinus Torvalds { 7371da177e4SLinus Torvalds struct sk_buff *skb2; 7381da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 7391da177e4SLinus Torvalds 7401da177e4SLinus Torvalds if (delta <= 0) 7411da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 7421da177e4SLinus Torvalds else { 7431da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 7441da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 7451da177e4SLinus Torvalds GFP_ATOMIC)) { 7461da177e4SLinus Torvalds kfree_skb(skb2); 7471da177e4SLinus Torvalds skb2 = NULL; 7481da177e4SLinus Torvalds } 7491da177e4SLinus Torvalds } 7501da177e4SLinus Torvalds return skb2; 7511da177e4SLinus Torvalds } 7521da177e4SLinus Torvalds 7531da177e4SLinus Torvalds 7541da177e4SLinus Torvalds /** 7551da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 7561da177e4SLinus Torvalds * @skb: buffer to copy 7571da177e4SLinus Torvalds * @newheadroom: new free bytes at head 7581da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 7591da177e4SLinus Torvalds * @gfp_mask: allocation priority 7601da177e4SLinus Torvalds * 7611da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 7621da177e4SLinus Torvalds * allocate additional space. 7631da177e4SLinus Torvalds * 7641da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 7651da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 7661da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 7671da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 7681da177e4SLinus Torvalds * 7691da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 7701da177e4SLinus Torvalds * is called from an interrupt. 7711da177e4SLinus Torvalds * 7721da177e4SLinus Torvalds * BUG ALERT: ip_summed is not copied. Why does this work? Is it used 7731da177e4SLinus Torvalds * only by netfilter in the cases when checksum is recalculated? --ANK 7741da177e4SLinus Torvalds */ 7751da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 77686a76cafSVictor Fusco int newheadroom, int newtailroom, 777dd0fc66fSAl Viro gfp_t gfp_mask) 7781da177e4SLinus Torvalds { 7791da177e4SLinus Torvalds /* 7801da177e4SLinus Torvalds * Allocate the copy buffer 7811da177e4SLinus Torvalds */ 7821da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 7831da177e4SLinus Torvalds gfp_mask); 7841da177e4SLinus Torvalds int head_copy_len, head_copy_off; 7851da177e4SLinus Torvalds 7861da177e4SLinus Torvalds if (!n) 7871da177e4SLinus Torvalds return NULL; 7881da177e4SLinus Torvalds 7891da177e4SLinus Torvalds skb_reserve(n, newheadroom); 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds /* Set the tail pointer and length */ 7921da177e4SLinus Torvalds skb_put(n, skb->len); 7931da177e4SLinus Torvalds 7941da177e4SLinus Torvalds head_copy_len = skb_headroom(skb); 7951da177e4SLinus Torvalds head_copy_off = 0; 7961da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 7971da177e4SLinus Torvalds head_copy_len = newheadroom; 7981da177e4SLinus Torvalds else 7991da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds /* Copy the linear header and data. */ 8021da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 8031da177e4SLinus Torvalds skb->len + head_copy_len)) 8041da177e4SLinus Torvalds BUG(); 8051da177e4SLinus Torvalds 8061da177e4SLinus Torvalds copy_skb_header(n, skb); 8071da177e4SLinus Torvalds 8081da177e4SLinus Torvalds return n; 8091da177e4SLinus Torvalds } 8101da177e4SLinus Torvalds 8111da177e4SLinus Torvalds /** 8121da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 8131da177e4SLinus Torvalds * @skb: buffer to pad 8141da177e4SLinus Torvalds * @pad: space to pad 8151da177e4SLinus Torvalds * 8161da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 8171da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 8181da177e4SLinus Torvalds * beyond the buffer end onto the wire. 8191da177e4SLinus Torvalds * 8205b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 8211da177e4SLinus Torvalds */ 8221da177e4SLinus Torvalds 8235b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 8241da177e4SLinus Torvalds { 8255b057c6bSHerbert Xu int err; 8265b057c6bSHerbert Xu int ntail; 8271da177e4SLinus Torvalds 8281da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 8295b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 8301da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 8315b057c6bSHerbert Xu return 0; 8321da177e4SLinus Torvalds } 8331da177e4SLinus Torvalds 8345b057c6bSHerbert Xu ntail = skb->data_len + pad - (skb->end - skb->tail); 8355b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 8365b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 8375b057c6bSHerbert Xu if (unlikely(err)) 8385b057c6bSHerbert Xu goto free_skb; 8395b057c6bSHerbert Xu } 8405b057c6bSHerbert Xu 8415b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 8425b057c6bSHerbert Xu * to be audited. 8435b057c6bSHerbert Xu */ 8445b057c6bSHerbert Xu err = skb_linearize(skb); 8455b057c6bSHerbert Xu if (unlikely(err)) 8465b057c6bSHerbert Xu goto free_skb; 8475b057c6bSHerbert Xu 8485b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 8495b057c6bSHerbert Xu return 0; 8505b057c6bSHerbert Xu 8515b057c6bSHerbert Xu free_skb: 8521da177e4SLinus Torvalds kfree_skb(skb); 8535b057c6bSHerbert Xu return err; 8541da177e4SLinus Torvalds } 8551da177e4SLinus Torvalds 8563cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 8571da177e4SLinus Torvalds */ 8581da177e4SLinus Torvalds 8593cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 8601da177e4SLinus Torvalds { 86127b437c8SHerbert Xu struct sk_buff **fragp; 86227b437c8SHerbert Xu struct sk_buff *frag; 8631da177e4SLinus Torvalds int offset = skb_headlen(skb); 8641da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 8651da177e4SLinus Torvalds int i; 86627b437c8SHerbert Xu int err; 86727b437c8SHerbert Xu 86827b437c8SHerbert Xu if (skb_cloned(skb) && 86927b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 87027b437c8SHerbert Xu return err; 8711da177e4SLinus Torvalds 872f4d26fb3SHerbert Xu i = 0; 873f4d26fb3SHerbert Xu if (offset >= len) 874f4d26fb3SHerbert Xu goto drop_pages; 875f4d26fb3SHerbert Xu 876f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 8771da177e4SLinus Torvalds int end = offset + skb_shinfo(skb)->frags[i].size; 87827b437c8SHerbert Xu 87927b437c8SHerbert Xu if (end < len) { 8801da177e4SLinus Torvalds offset = end; 88127b437c8SHerbert Xu continue; 8821da177e4SLinus Torvalds } 8831da177e4SLinus Torvalds 88427b437c8SHerbert Xu skb_shinfo(skb)->frags[i++].size = len - offset; 88527b437c8SHerbert Xu 886f4d26fb3SHerbert Xu drop_pages: 88727b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 88827b437c8SHerbert Xu 88927b437c8SHerbert Xu for (; i < nfrags; i++) 89027b437c8SHerbert Xu put_page(skb_shinfo(skb)->frags[i].page); 89127b437c8SHerbert Xu 89227b437c8SHerbert Xu if (skb_shinfo(skb)->frag_list) 89327b437c8SHerbert Xu skb_drop_fraglist(skb); 894f4d26fb3SHerbert Xu goto done; 89527b437c8SHerbert Xu } 89627b437c8SHerbert Xu 89727b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 89827b437c8SHerbert Xu fragp = &frag->next) { 89927b437c8SHerbert Xu int end = offset + frag->len; 90027b437c8SHerbert Xu 90127b437c8SHerbert Xu if (skb_shared(frag)) { 90227b437c8SHerbert Xu struct sk_buff *nfrag; 90327b437c8SHerbert Xu 90427b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 90527b437c8SHerbert Xu if (unlikely(!nfrag)) 90627b437c8SHerbert Xu return -ENOMEM; 90727b437c8SHerbert Xu 90827b437c8SHerbert Xu nfrag->next = frag->next; 909f4d26fb3SHerbert Xu kfree_skb(frag); 91027b437c8SHerbert Xu frag = nfrag; 91127b437c8SHerbert Xu *fragp = frag; 91227b437c8SHerbert Xu } 91327b437c8SHerbert Xu 91427b437c8SHerbert Xu if (end < len) { 91527b437c8SHerbert Xu offset = end; 91627b437c8SHerbert Xu continue; 91727b437c8SHerbert Xu } 91827b437c8SHerbert Xu 91927b437c8SHerbert Xu if (end > len && 92027b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 92127b437c8SHerbert Xu return err; 92227b437c8SHerbert Xu 92327b437c8SHerbert Xu if (frag->next) 92427b437c8SHerbert Xu skb_drop_list(&frag->next); 92527b437c8SHerbert Xu break; 92627b437c8SHerbert Xu } 92727b437c8SHerbert Xu 928f4d26fb3SHerbert Xu done: 92927b437c8SHerbert Xu if (len > skb_headlen(skb)) { 9301da177e4SLinus Torvalds skb->data_len -= skb->len - len; 9311da177e4SLinus Torvalds skb->len = len; 9321da177e4SLinus Torvalds } else { 9331da177e4SLinus Torvalds skb->len = len; 9341da177e4SLinus Torvalds skb->data_len = 0; 9351da177e4SLinus Torvalds skb->tail = skb->data + len; 9361da177e4SLinus Torvalds } 9371da177e4SLinus Torvalds 9381da177e4SLinus Torvalds return 0; 9391da177e4SLinus Torvalds } 9401da177e4SLinus Torvalds 9411da177e4SLinus Torvalds /** 9421da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 9431da177e4SLinus Torvalds * @skb: buffer to reallocate 9441da177e4SLinus Torvalds * @delta: number of bytes to advance tail 9451da177e4SLinus Torvalds * 9461da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 9471da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 9481da177e4SLinus Torvalds * data from fragmented part. 9491da177e4SLinus Torvalds * 9501da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 9511da177e4SLinus Torvalds * 9521da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 9531da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 9541da177e4SLinus Torvalds * 9551da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 9561da177e4SLinus Torvalds * reloaded after call to this function. 9571da177e4SLinus Torvalds */ 9581da177e4SLinus Torvalds 9591da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 9601da177e4SLinus Torvalds * when it is necessary. 9611da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 9621da177e4SLinus Torvalds * 2. It may change skb pointers. 9631da177e4SLinus Torvalds * 9641da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 9651da177e4SLinus Torvalds */ 9661da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 9671da177e4SLinus Torvalds { 9681da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 9691da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 9701da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 9711da177e4SLinus Torvalds */ 9721da177e4SLinus Torvalds int i, k, eat = (skb->tail + delta) - skb->end; 9731da177e4SLinus Torvalds 9741da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 9751da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 9761da177e4SLinus Torvalds GFP_ATOMIC)) 9771da177e4SLinus Torvalds return NULL; 9781da177e4SLinus Torvalds } 9791da177e4SLinus Torvalds 9801da177e4SLinus Torvalds if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta)) 9811da177e4SLinus Torvalds BUG(); 9821da177e4SLinus Torvalds 9831da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 9841da177e4SLinus Torvalds * size of pulled pages. Superb. 9851da177e4SLinus Torvalds */ 9861da177e4SLinus Torvalds if (!skb_shinfo(skb)->frag_list) 9871da177e4SLinus Torvalds goto pull_pages; 9881da177e4SLinus Torvalds 9891da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 9901da177e4SLinus Torvalds eat = delta; 9911da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 9921da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size >= eat) 9931da177e4SLinus Torvalds goto pull_pages; 9941da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 9951da177e4SLinus Torvalds } 9961da177e4SLinus Torvalds 9971da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 9981da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 9991da177e4SLinus Torvalds * but taking into account that pulling is expected to 10001da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 10011da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 10021da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 10031da177e4SLinus Torvalds */ 10041da177e4SLinus Torvalds if (eat) { 10051da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 10061da177e4SLinus Torvalds struct sk_buff *clone = NULL; 10071da177e4SLinus Torvalds struct sk_buff *insp = NULL; 10081da177e4SLinus Torvalds 10091da177e4SLinus Torvalds do { 101009a62660SKris Katterjohn BUG_ON(!list); 10111da177e4SLinus Torvalds 10121da177e4SLinus Torvalds if (list->len <= eat) { 10131da177e4SLinus Torvalds /* Eaten as whole. */ 10141da177e4SLinus Torvalds eat -= list->len; 10151da177e4SLinus Torvalds list = list->next; 10161da177e4SLinus Torvalds insp = list; 10171da177e4SLinus Torvalds } else { 10181da177e4SLinus Torvalds /* Eaten partially. */ 10191da177e4SLinus Torvalds 10201da177e4SLinus Torvalds if (skb_shared(list)) { 10211da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 10221da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 10231da177e4SLinus Torvalds if (!clone) 10241da177e4SLinus Torvalds return NULL; 10251da177e4SLinus Torvalds insp = list->next; 10261da177e4SLinus Torvalds list = clone; 10271da177e4SLinus Torvalds } else { 10281da177e4SLinus Torvalds /* This may be pulled without 10291da177e4SLinus Torvalds * problems. */ 10301da177e4SLinus Torvalds insp = list; 10311da177e4SLinus Torvalds } 10321da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 10331da177e4SLinus Torvalds if (clone) 10341da177e4SLinus Torvalds kfree_skb(clone); 10351da177e4SLinus Torvalds return NULL; 10361da177e4SLinus Torvalds } 10371da177e4SLinus Torvalds break; 10381da177e4SLinus Torvalds } 10391da177e4SLinus Torvalds } while (eat); 10401da177e4SLinus Torvalds 10411da177e4SLinus Torvalds /* Free pulled out fragments. */ 10421da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 10431da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 10441da177e4SLinus Torvalds kfree_skb(list); 10451da177e4SLinus Torvalds } 10461da177e4SLinus Torvalds /* And insert new clone at head. */ 10471da177e4SLinus Torvalds if (clone) { 10481da177e4SLinus Torvalds clone->next = list; 10491da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 10501da177e4SLinus Torvalds } 10511da177e4SLinus Torvalds } 10521da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 10531da177e4SLinus Torvalds 10541da177e4SLinus Torvalds pull_pages: 10551da177e4SLinus Torvalds eat = delta; 10561da177e4SLinus Torvalds k = 0; 10571da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 10581da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size <= eat) { 10591da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 10601da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 10611da177e4SLinus Torvalds } else { 10621da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 10631da177e4SLinus Torvalds if (eat) { 10641da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 10651da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].size -= eat; 10661da177e4SLinus Torvalds eat = 0; 10671da177e4SLinus Torvalds } 10681da177e4SLinus Torvalds k++; 10691da177e4SLinus Torvalds } 10701da177e4SLinus Torvalds } 10711da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 10721da177e4SLinus Torvalds 10731da177e4SLinus Torvalds skb->tail += delta; 10741da177e4SLinus Torvalds skb->data_len -= delta; 10751da177e4SLinus Torvalds 10761da177e4SLinus Torvalds return skb->tail; 10771da177e4SLinus Torvalds } 10781da177e4SLinus Torvalds 10791da177e4SLinus Torvalds /* Copy some data bits from skb to kernel buffer. */ 10801da177e4SLinus Torvalds 10811da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 10821da177e4SLinus Torvalds { 10831da177e4SLinus Torvalds int i, copy; 10841da177e4SLinus Torvalds int start = skb_headlen(skb); 10851da177e4SLinus Torvalds 10861da177e4SLinus Torvalds if (offset > (int)skb->len - len) 10871da177e4SLinus Torvalds goto fault; 10881da177e4SLinus Torvalds 10891da177e4SLinus Torvalds /* Copy header. */ 10901da177e4SLinus Torvalds if ((copy = start - offset) > 0) { 10911da177e4SLinus Torvalds if (copy > len) 10921da177e4SLinus Torvalds copy = len; 10931da177e4SLinus Torvalds memcpy(to, skb->data + offset, copy); 10941da177e4SLinus Torvalds if ((len -= copy) == 0) 10951da177e4SLinus Torvalds return 0; 10961da177e4SLinus Torvalds offset += copy; 10971da177e4SLinus Torvalds to += copy; 10981da177e4SLinus Torvalds } 10991da177e4SLinus Torvalds 11001da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11011da177e4SLinus Torvalds int end; 11021da177e4SLinus Torvalds 11031da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 11041da177e4SLinus Torvalds 11051da177e4SLinus Torvalds end = start + skb_shinfo(skb)->frags[i].size; 11061da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 11071da177e4SLinus Torvalds u8 *vaddr; 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds if (copy > len) 11101da177e4SLinus Torvalds copy = len; 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 11131da177e4SLinus Torvalds memcpy(to, 11141da177e4SLinus Torvalds vaddr + skb_shinfo(skb)->frags[i].page_offset+ 11151da177e4SLinus Torvalds offset - start, copy); 11161da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 11171da177e4SLinus Torvalds 11181da177e4SLinus Torvalds if ((len -= copy) == 0) 11191da177e4SLinus Torvalds return 0; 11201da177e4SLinus Torvalds offset += copy; 11211da177e4SLinus Torvalds to += copy; 11221da177e4SLinus Torvalds } 11231da177e4SLinus Torvalds start = end; 11241da177e4SLinus Torvalds } 11251da177e4SLinus Torvalds 11261da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 11271da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 11281da177e4SLinus Torvalds 11291da177e4SLinus Torvalds for (; list; list = list->next) { 11301da177e4SLinus Torvalds int end; 11311da177e4SLinus Torvalds 11321da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 11331da177e4SLinus Torvalds 11341da177e4SLinus Torvalds end = start + list->len; 11351da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 11361da177e4SLinus Torvalds if (copy > len) 11371da177e4SLinus Torvalds copy = len; 11381da177e4SLinus Torvalds if (skb_copy_bits(list, offset - start, 11391da177e4SLinus Torvalds to, copy)) 11401da177e4SLinus Torvalds goto fault; 11411da177e4SLinus Torvalds if ((len -= copy) == 0) 11421da177e4SLinus Torvalds return 0; 11431da177e4SLinus Torvalds offset += copy; 11441da177e4SLinus Torvalds to += copy; 11451da177e4SLinus Torvalds } 11461da177e4SLinus Torvalds start = end; 11471da177e4SLinus Torvalds } 11481da177e4SLinus Torvalds } 11491da177e4SLinus Torvalds if (!len) 11501da177e4SLinus Torvalds return 0; 11511da177e4SLinus Torvalds 11521da177e4SLinus Torvalds fault: 11531da177e4SLinus Torvalds return -EFAULT; 11541da177e4SLinus Torvalds } 11551da177e4SLinus Torvalds 1156357b40a1SHerbert Xu /** 1157357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1158357b40a1SHerbert Xu * @skb: destination buffer 1159357b40a1SHerbert Xu * @offset: offset in destination 1160357b40a1SHerbert Xu * @from: source buffer 1161357b40a1SHerbert Xu * @len: number of bytes to copy 1162357b40a1SHerbert Xu * 1163357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1164357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1165357b40a1SHerbert Xu * traversing fragment lists and such. 1166357b40a1SHerbert Xu */ 1167357b40a1SHerbert Xu 1168357b40a1SHerbert Xu int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len) 1169357b40a1SHerbert Xu { 1170357b40a1SHerbert Xu int i, copy; 1171357b40a1SHerbert Xu int start = skb_headlen(skb); 1172357b40a1SHerbert Xu 1173357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1174357b40a1SHerbert Xu goto fault; 1175357b40a1SHerbert Xu 1176357b40a1SHerbert Xu if ((copy = start - offset) > 0) { 1177357b40a1SHerbert Xu if (copy > len) 1178357b40a1SHerbert Xu copy = len; 1179357b40a1SHerbert Xu memcpy(skb->data + offset, from, copy); 1180357b40a1SHerbert Xu if ((len -= copy) == 0) 1181357b40a1SHerbert Xu return 0; 1182357b40a1SHerbert Xu offset += copy; 1183357b40a1SHerbert Xu from += copy; 1184357b40a1SHerbert Xu } 1185357b40a1SHerbert Xu 1186357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1187357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1188357b40a1SHerbert Xu int end; 1189357b40a1SHerbert Xu 1190357b40a1SHerbert Xu BUG_TRAP(start <= offset + len); 1191357b40a1SHerbert Xu 1192357b40a1SHerbert Xu end = start + frag->size; 1193357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1194357b40a1SHerbert Xu u8 *vaddr; 1195357b40a1SHerbert Xu 1196357b40a1SHerbert Xu if (copy > len) 1197357b40a1SHerbert Xu copy = len; 1198357b40a1SHerbert Xu 1199357b40a1SHerbert Xu vaddr = kmap_skb_frag(frag); 1200357b40a1SHerbert Xu memcpy(vaddr + frag->page_offset + offset - start, 1201357b40a1SHerbert Xu from, copy); 1202357b40a1SHerbert Xu kunmap_skb_frag(vaddr); 1203357b40a1SHerbert Xu 1204357b40a1SHerbert Xu if ((len -= copy) == 0) 1205357b40a1SHerbert Xu return 0; 1206357b40a1SHerbert Xu offset += copy; 1207357b40a1SHerbert Xu from += copy; 1208357b40a1SHerbert Xu } 1209357b40a1SHerbert Xu start = end; 1210357b40a1SHerbert Xu } 1211357b40a1SHerbert Xu 1212357b40a1SHerbert Xu if (skb_shinfo(skb)->frag_list) { 1213357b40a1SHerbert Xu struct sk_buff *list = skb_shinfo(skb)->frag_list; 1214357b40a1SHerbert Xu 1215357b40a1SHerbert Xu for (; list; list = list->next) { 1216357b40a1SHerbert Xu int end; 1217357b40a1SHerbert Xu 1218357b40a1SHerbert Xu BUG_TRAP(start <= offset + len); 1219357b40a1SHerbert Xu 1220357b40a1SHerbert Xu end = start + list->len; 1221357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1222357b40a1SHerbert Xu if (copy > len) 1223357b40a1SHerbert Xu copy = len; 1224357b40a1SHerbert Xu if (skb_store_bits(list, offset - start, 1225357b40a1SHerbert Xu from, copy)) 1226357b40a1SHerbert Xu goto fault; 1227357b40a1SHerbert Xu if ((len -= copy) == 0) 1228357b40a1SHerbert Xu return 0; 1229357b40a1SHerbert Xu offset += copy; 1230357b40a1SHerbert Xu from += copy; 1231357b40a1SHerbert Xu } 1232357b40a1SHerbert Xu start = end; 1233357b40a1SHerbert Xu } 1234357b40a1SHerbert Xu } 1235357b40a1SHerbert Xu if (!len) 1236357b40a1SHerbert Xu return 0; 1237357b40a1SHerbert Xu 1238357b40a1SHerbert Xu fault: 1239357b40a1SHerbert Xu return -EFAULT; 1240357b40a1SHerbert Xu } 1241357b40a1SHerbert Xu 1242357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1243357b40a1SHerbert Xu 12441da177e4SLinus Torvalds /* Checksum skb data. */ 12451da177e4SLinus Torvalds 12462bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 12472bbbc868SAl Viro int len, __wsum csum) 12481da177e4SLinus Torvalds { 12491da177e4SLinus Torvalds int start = skb_headlen(skb); 12501da177e4SLinus Torvalds int i, copy = start - offset; 12511da177e4SLinus Torvalds int pos = 0; 12521da177e4SLinus Torvalds 12531da177e4SLinus Torvalds /* Checksum header. */ 12541da177e4SLinus Torvalds if (copy > 0) { 12551da177e4SLinus Torvalds if (copy > len) 12561da177e4SLinus Torvalds copy = len; 12571da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 12581da177e4SLinus Torvalds if ((len -= copy) == 0) 12591da177e4SLinus Torvalds return csum; 12601da177e4SLinus Torvalds offset += copy; 12611da177e4SLinus Torvalds pos = copy; 12621da177e4SLinus Torvalds } 12631da177e4SLinus Torvalds 12641da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12651da177e4SLinus Torvalds int end; 12661da177e4SLinus Torvalds 12671da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 12681da177e4SLinus Torvalds 12691da177e4SLinus Torvalds end = start + skb_shinfo(skb)->frags[i].size; 12701da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 127144bb9363SAl Viro __wsum csum2; 12721da177e4SLinus Torvalds u8 *vaddr; 12731da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 12741da177e4SLinus Torvalds 12751da177e4SLinus Torvalds if (copy > len) 12761da177e4SLinus Torvalds copy = len; 12771da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 12781da177e4SLinus Torvalds csum2 = csum_partial(vaddr + frag->page_offset + 12791da177e4SLinus Torvalds offset - start, copy, 0); 12801da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 12811da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 12821da177e4SLinus Torvalds if (!(len -= copy)) 12831da177e4SLinus Torvalds return csum; 12841da177e4SLinus Torvalds offset += copy; 12851da177e4SLinus Torvalds pos += copy; 12861da177e4SLinus Torvalds } 12871da177e4SLinus Torvalds start = end; 12881da177e4SLinus Torvalds } 12891da177e4SLinus Torvalds 12901da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 12911da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds for (; list; list = list->next) { 12941da177e4SLinus Torvalds int end; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 12971da177e4SLinus Torvalds 12981da177e4SLinus Torvalds end = start + list->len; 12991da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13005f92a738SAl Viro __wsum csum2; 13011da177e4SLinus Torvalds if (copy > len) 13021da177e4SLinus Torvalds copy = len; 13031da177e4SLinus Torvalds csum2 = skb_checksum(list, offset - start, 13041da177e4SLinus Torvalds copy, 0); 13051da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 13061da177e4SLinus Torvalds if ((len -= copy) == 0) 13071da177e4SLinus Torvalds return csum; 13081da177e4SLinus Torvalds offset += copy; 13091da177e4SLinus Torvalds pos += copy; 13101da177e4SLinus Torvalds } 13111da177e4SLinus Torvalds start = end; 13121da177e4SLinus Torvalds } 13131da177e4SLinus Torvalds } 131409a62660SKris Katterjohn BUG_ON(len); 13151da177e4SLinus Torvalds 13161da177e4SLinus Torvalds return csum; 13171da177e4SLinus Torvalds } 13181da177e4SLinus Torvalds 13191da177e4SLinus Torvalds /* Both of above in one bottle. */ 13201da177e4SLinus Torvalds 132181d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 132281d77662SAl Viro u8 *to, int len, __wsum csum) 13231da177e4SLinus Torvalds { 13241da177e4SLinus Torvalds int start = skb_headlen(skb); 13251da177e4SLinus Torvalds int i, copy = start - offset; 13261da177e4SLinus Torvalds int pos = 0; 13271da177e4SLinus Torvalds 13281da177e4SLinus Torvalds /* Copy header. */ 13291da177e4SLinus Torvalds if (copy > 0) { 13301da177e4SLinus Torvalds if (copy > len) 13311da177e4SLinus Torvalds copy = len; 13321da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 13331da177e4SLinus Torvalds copy, csum); 13341da177e4SLinus Torvalds if ((len -= copy) == 0) 13351da177e4SLinus Torvalds return csum; 13361da177e4SLinus Torvalds offset += copy; 13371da177e4SLinus Torvalds to += copy; 13381da177e4SLinus Torvalds pos = copy; 13391da177e4SLinus Torvalds } 13401da177e4SLinus Torvalds 13411da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 13421da177e4SLinus Torvalds int end; 13431da177e4SLinus Torvalds 13441da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 13451da177e4SLinus Torvalds 13461da177e4SLinus Torvalds end = start + skb_shinfo(skb)->frags[i].size; 13471da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13485084205fSAl Viro __wsum csum2; 13491da177e4SLinus Torvalds u8 *vaddr; 13501da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 13511da177e4SLinus Torvalds 13521da177e4SLinus Torvalds if (copy > len) 13531da177e4SLinus Torvalds copy = len; 13541da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 13551da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 13561da177e4SLinus Torvalds frag->page_offset + 13571da177e4SLinus Torvalds offset - start, to, 13581da177e4SLinus Torvalds copy, 0); 13591da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 13601da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 13611da177e4SLinus Torvalds if (!(len -= copy)) 13621da177e4SLinus Torvalds return csum; 13631da177e4SLinus Torvalds offset += copy; 13641da177e4SLinus Torvalds to += copy; 13651da177e4SLinus Torvalds pos += copy; 13661da177e4SLinus Torvalds } 13671da177e4SLinus Torvalds start = end; 13681da177e4SLinus Torvalds } 13691da177e4SLinus Torvalds 13701da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 13711da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 13721da177e4SLinus Torvalds 13731da177e4SLinus Torvalds for (; list; list = list->next) { 137481d77662SAl Viro __wsum csum2; 13751da177e4SLinus Torvalds int end; 13761da177e4SLinus Torvalds 13771da177e4SLinus Torvalds BUG_TRAP(start <= offset + len); 13781da177e4SLinus Torvalds 13791da177e4SLinus Torvalds end = start + list->len; 13801da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13811da177e4SLinus Torvalds if (copy > len) 13821da177e4SLinus Torvalds copy = len; 13831da177e4SLinus Torvalds csum2 = skb_copy_and_csum_bits(list, 13841da177e4SLinus Torvalds offset - start, 13851da177e4SLinus Torvalds to, copy, 0); 13861da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 13871da177e4SLinus Torvalds if ((len -= copy) == 0) 13881da177e4SLinus Torvalds return csum; 13891da177e4SLinus Torvalds offset += copy; 13901da177e4SLinus Torvalds to += copy; 13911da177e4SLinus Torvalds pos += copy; 13921da177e4SLinus Torvalds } 13931da177e4SLinus Torvalds start = end; 13941da177e4SLinus Torvalds } 13951da177e4SLinus Torvalds } 139609a62660SKris Katterjohn BUG_ON(len); 13971da177e4SLinus Torvalds return csum; 13981da177e4SLinus Torvalds } 13991da177e4SLinus Torvalds 14001da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 14011da177e4SLinus Torvalds { 1402d3bc23e7SAl Viro __wsum csum; 14031da177e4SLinus Torvalds long csstart; 14041da177e4SLinus Torvalds 140584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 14061da177e4SLinus Torvalds csstart = skb->h.raw - skb->data; 14071da177e4SLinus Torvalds else 14081da177e4SLinus Torvalds csstart = skb_headlen(skb); 14091da177e4SLinus Torvalds 141009a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 14111da177e4SLinus Torvalds 14121da177e4SLinus Torvalds memcpy(to, skb->data, csstart); 14131da177e4SLinus Torvalds 14141da177e4SLinus Torvalds csum = 0; 14151da177e4SLinus Torvalds if (csstart != skb->len) 14161da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 14171da177e4SLinus Torvalds skb->len - csstart, 0); 14181da177e4SLinus Torvalds 141984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 1420ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 14211da177e4SLinus Torvalds 1422d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 14231da177e4SLinus Torvalds } 14241da177e4SLinus Torvalds } 14251da177e4SLinus Torvalds 14261da177e4SLinus Torvalds /** 14271da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 14281da177e4SLinus Torvalds * @list: list to dequeue from 14291da177e4SLinus Torvalds * 14301da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 14311da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 14321da177e4SLinus Torvalds * returned or %NULL if the list is empty. 14331da177e4SLinus Torvalds */ 14341da177e4SLinus Torvalds 14351da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 14361da177e4SLinus Torvalds { 14371da177e4SLinus Torvalds unsigned long flags; 14381da177e4SLinus Torvalds struct sk_buff *result; 14391da177e4SLinus Torvalds 14401da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 14411da177e4SLinus Torvalds result = __skb_dequeue(list); 14421da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 14431da177e4SLinus Torvalds return result; 14441da177e4SLinus Torvalds } 14451da177e4SLinus Torvalds 14461da177e4SLinus Torvalds /** 14471da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 14481da177e4SLinus Torvalds * @list: list to dequeue from 14491da177e4SLinus Torvalds * 14501da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 14511da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 14521da177e4SLinus Torvalds * returned or %NULL if the list is empty. 14531da177e4SLinus Torvalds */ 14541da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 14551da177e4SLinus Torvalds { 14561da177e4SLinus Torvalds unsigned long flags; 14571da177e4SLinus Torvalds struct sk_buff *result; 14581da177e4SLinus Torvalds 14591da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 14601da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 14611da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 14621da177e4SLinus Torvalds return result; 14631da177e4SLinus Torvalds } 14641da177e4SLinus Torvalds 14651da177e4SLinus Torvalds /** 14661da177e4SLinus Torvalds * skb_queue_purge - empty a list 14671da177e4SLinus Torvalds * @list: list to empty 14681da177e4SLinus Torvalds * 14691da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 14701da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 14711da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 14721da177e4SLinus Torvalds */ 14731da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 14741da177e4SLinus Torvalds { 14751da177e4SLinus Torvalds struct sk_buff *skb; 14761da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 14771da177e4SLinus Torvalds kfree_skb(skb); 14781da177e4SLinus Torvalds } 14791da177e4SLinus Torvalds 14801da177e4SLinus Torvalds /** 14811da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 14821da177e4SLinus Torvalds * @list: list to use 14831da177e4SLinus Torvalds * @newsk: buffer to queue 14841da177e4SLinus Torvalds * 14851da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 14861da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 14871da177e4SLinus Torvalds * safely. 14881da177e4SLinus Torvalds * 14891da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 14901da177e4SLinus Torvalds */ 14911da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 14921da177e4SLinus Torvalds { 14931da177e4SLinus Torvalds unsigned long flags; 14941da177e4SLinus Torvalds 14951da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 14961da177e4SLinus Torvalds __skb_queue_head(list, newsk); 14971da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 14981da177e4SLinus Torvalds } 14991da177e4SLinus Torvalds 15001da177e4SLinus Torvalds /** 15011da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 15021da177e4SLinus Torvalds * @list: list to use 15031da177e4SLinus Torvalds * @newsk: buffer to queue 15041da177e4SLinus Torvalds * 15051da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 15061da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 15071da177e4SLinus Torvalds * safely. 15081da177e4SLinus Torvalds * 15091da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 15101da177e4SLinus Torvalds */ 15111da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 15121da177e4SLinus Torvalds { 15131da177e4SLinus Torvalds unsigned long flags; 15141da177e4SLinus Torvalds 15151da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 15161da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 15171da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 15181da177e4SLinus Torvalds } 15198728b834SDavid S. Miller 15201da177e4SLinus Torvalds /** 15211da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 15221da177e4SLinus Torvalds * @skb: buffer to remove 15238728b834SDavid S. Miller * @list: list to use 15241da177e4SLinus Torvalds * 15258728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 15268728b834SDavid S. Miller * function is atomic with respect to other list locked calls 15271da177e4SLinus Torvalds * 15288728b834SDavid S. Miller * You must know what list the SKB is on. 15291da177e4SLinus Torvalds */ 15308728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 15311da177e4SLinus Torvalds { 15321da177e4SLinus Torvalds unsigned long flags; 15331da177e4SLinus Torvalds 15341da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 15358728b834SDavid S. Miller __skb_unlink(skb, list); 15361da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 15371da177e4SLinus Torvalds } 15381da177e4SLinus Torvalds 15391da177e4SLinus Torvalds /** 15401da177e4SLinus Torvalds * skb_append - append a buffer 15411da177e4SLinus Torvalds * @old: buffer to insert after 15421da177e4SLinus Torvalds * @newsk: buffer to insert 15438728b834SDavid S. Miller * @list: list to use 15441da177e4SLinus Torvalds * 15451da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 15461da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 15471da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 15481da177e4SLinus Torvalds */ 15498728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 15501da177e4SLinus Torvalds { 15511da177e4SLinus Torvalds unsigned long flags; 15521da177e4SLinus Torvalds 15538728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 15548728b834SDavid S. Miller __skb_append(old, newsk, list); 15558728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 15561da177e4SLinus Torvalds } 15571da177e4SLinus Torvalds 15581da177e4SLinus Torvalds 15591da177e4SLinus Torvalds /** 15601da177e4SLinus Torvalds * skb_insert - insert a buffer 15611da177e4SLinus Torvalds * @old: buffer to insert before 15621da177e4SLinus Torvalds * @newsk: buffer to insert 15638728b834SDavid S. Miller * @list: list to use 15641da177e4SLinus Torvalds * 15658728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 15668728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 15678728b834SDavid S. Miller * calls. 15688728b834SDavid S. Miller * 15691da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 15701da177e4SLinus Torvalds */ 15718728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 15721da177e4SLinus Torvalds { 15731da177e4SLinus Torvalds unsigned long flags; 15741da177e4SLinus Torvalds 15758728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 15768728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 15778728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 15781da177e4SLinus Torvalds } 15791da177e4SLinus Torvalds 15801da177e4SLinus Torvalds #if 0 15811da177e4SLinus Torvalds /* 15821da177e4SLinus Torvalds * Tune the memory allocator for a new MTU size. 15831da177e4SLinus Torvalds */ 15841da177e4SLinus Torvalds void skb_add_mtu(int mtu) 15851da177e4SLinus Torvalds { 15861da177e4SLinus Torvalds /* Must match allocation in alloc_skb */ 15871da177e4SLinus Torvalds mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info); 15881da177e4SLinus Torvalds 15891da177e4SLinus Torvalds kmem_add_cache_size(mtu); 15901da177e4SLinus Torvalds } 15911da177e4SLinus Torvalds #endif 15921da177e4SLinus Torvalds 15931da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 15941da177e4SLinus Torvalds struct sk_buff* skb1, 15951da177e4SLinus Torvalds const u32 len, const int pos) 15961da177e4SLinus Torvalds { 15971da177e4SLinus Torvalds int i; 15981da177e4SLinus Torvalds 15991da177e4SLinus Torvalds memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len); 16001da177e4SLinus Torvalds 16011da177e4SLinus Torvalds /* And move data appendix as is. */ 16021da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 16031da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 16041da177e4SLinus Torvalds 16051da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 16061da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 16071da177e4SLinus Torvalds skb1->data_len = skb->data_len; 16081da177e4SLinus Torvalds skb1->len += skb1->data_len; 16091da177e4SLinus Torvalds skb->data_len = 0; 16101da177e4SLinus Torvalds skb->len = len; 16111da177e4SLinus Torvalds skb->tail = skb->data + len; 16121da177e4SLinus Torvalds } 16131da177e4SLinus Torvalds 16141da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 16151da177e4SLinus Torvalds struct sk_buff* skb1, 16161da177e4SLinus Torvalds const u32 len, int pos) 16171da177e4SLinus Torvalds { 16181da177e4SLinus Torvalds int i, k = 0; 16191da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 16201da177e4SLinus Torvalds 16211da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 16221da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 16231da177e4SLinus Torvalds skb->len = len; 16241da177e4SLinus Torvalds skb->data_len = len - pos; 16251da177e4SLinus Torvalds 16261da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 16271da177e4SLinus Torvalds int size = skb_shinfo(skb)->frags[i].size; 16281da177e4SLinus Torvalds 16291da177e4SLinus Torvalds if (pos + size > len) { 16301da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 16311da177e4SLinus Torvalds 16321da177e4SLinus Torvalds if (pos < len) { 16331da177e4SLinus Torvalds /* Split frag. 16341da177e4SLinus Torvalds * We have two variants in this case: 16351da177e4SLinus Torvalds * 1. Move all the frag to the second 16361da177e4SLinus Torvalds * part, if it is possible. F.e. 16371da177e4SLinus Torvalds * this approach is mandatory for TUX, 16381da177e4SLinus Torvalds * where splitting is expensive. 16391da177e4SLinus Torvalds * 2. Split is accurately. We make this. 16401da177e4SLinus Torvalds */ 16411da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 16421da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 16431da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].size -= len - pos; 16441da177e4SLinus Torvalds skb_shinfo(skb)->frags[i].size = len - pos; 16451da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 16461da177e4SLinus Torvalds } 16471da177e4SLinus Torvalds k++; 16481da177e4SLinus Torvalds } else 16491da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 16501da177e4SLinus Torvalds pos += size; 16511da177e4SLinus Torvalds } 16521da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 16531da177e4SLinus Torvalds } 16541da177e4SLinus Torvalds 16551da177e4SLinus Torvalds /** 16561da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 16571da177e4SLinus Torvalds * @skb: the buffer to split 16581da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 16591da177e4SLinus Torvalds * @len: new length for skb 16601da177e4SLinus Torvalds */ 16611da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 16621da177e4SLinus Torvalds { 16631da177e4SLinus Torvalds int pos = skb_headlen(skb); 16641da177e4SLinus Torvalds 16651da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 16661da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 16671da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 16681da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 16691da177e4SLinus Torvalds } 16701da177e4SLinus Torvalds 1671677e90edSThomas Graf /** 1672677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 1673677e90edSThomas Graf * @skb: the buffer to read 1674677e90edSThomas Graf * @from: lower offset of data to be read 1675677e90edSThomas Graf * @to: upper offset of data to be read 1676677e90edSThomas Graf * @st: state variable 1677677e90edSThomas Graf * 1678677e90edSThomas Graf * Initializes the specified state variable. Must be called before 1679677e90edSThomas Graf * invoking skb_seq_read() for the first time. 1680677e90edSThomas Graf */ 1681677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1682677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 1683677e90edSThomas Graf { 1684677e90edSThomas Graf st->lower_offset = from; 1685677e90edSThomas Graf st->upper_offset = to; 1686677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 1687677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 1688677e90edSThomas Graf st->frag_data = NULL; 1689677e90edSThomas Graf } 1690677e90edSThomas Graf 1691677e90edSThomas Graf /** 1692677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 1693677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 1694677e90edSThomas Graf * @data: destination pointer for data to be returned 1695677e90edSThomas Graf * @st: state variable 1696677e90edSThomas Graf * 1697677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 1698677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 1699677e90edSThomas Graf * the head of the data block to &data and returns the length 1700677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 1701677e90edSThomas Graf * offset has been reached. 1702677e90edSThomas Graf * 1703677e90edSThomas Graf * The caller is not required to consume all of the data 1704677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 1705677e90edSThomas Graf * of bytes already consumed and the next call to 1706677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 1707677e90edSThomas Graf * 1708677e90edSThomas Graf * Note: The size of each block of data returned can be arbitary, 1709677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 1710677e90edSThomas Graf * reads of potentially non linear data. 1711677e90edSThomas Graf * 1712677e90edSThomas Graf * Note: Fragment lists within fragments are not implemented 1713677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 1714677e90edSThomas Graf * a stack for this purpose. 1715677e90edSThomas Graf */ 1716677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1717677e90edSThomas Graf struct skb_seq_state *st) 1718677e90edSThomas Graf { 1719677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 1720677e90edSThomas Graf skb_frag_t *frag; 1721677e90edSThomas Graf 1722677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 1723677e90edSThomas Graf return 0; 1724677e90edSThomas Graf 1725677e90edSThomas Graf next_skb: 1726677e90edSThomas Graf block_limit = skb_headlen(st->cur_skb); 1727677e90edSThomas Graf 1728677e90edSThomas Graf if (abs_offset < block_limit) { 1729677e90edSThomas Graf *data = st->cur_skb->data + abs_offset; 1730677e90edSThomas Graf return block_limit - abs_offset; 1731677e90edSThomas Graf } 1732677e90edSThomas Graf 1733677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 1734677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 1735677e90edSThomas Graf 1736677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 1737677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 1738677e90edSThomas Graf block_limit = frag->size + st->stepped_offset; 1739677e90edSThomas Graf 1740677e90edSThomas Graf if (abs_offset < block_limit) { 1741677e90edSThomas Graf if (!st->frag_data) 1742677e90edSThomas Graf st->frag_data = kmap_skb_frag(frag); 1743677e90edSThomas Graf 1744677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 1745677e90edSThomas Graf (abs_offset - st->stepped_offset); 1746677e90edSThomas Graf 1747677e90edSThomas Graf return block_limit - abs_offset; 1748677e90edSThomas Graf } 1749677e90edSThomas Graf 1750677e90edSThomas Graf if (st->frag_data) { 1751677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 1752677e90edSThomas Graf st->frag_data = NULL; 1753677e90edSThomas Graf } 1754677e90edSThomas Graf 1755677e90edSThomas Graf st->frag_idx++; 1756677e90edSThomas Graf st->stepped_offset += frag->size; 1757677e90edSThomas Graf } 1758677e90edSThomas Graf 1759677e90edSThomas Graf if (st->cur_skb->next) { 1760677e90edSThomas Graf st->cur_skb = st->cur_skb->next; 1761677e90edSThomas Graf st->frag_idx = 0; 1762677e90edSThomas Graf goto next_skb; 1763677e90edSThomas Graf } else if (st->root_skb == st->cur_skb && 1764677e90edSThomas Graf skb_shinfo(st->root_skb)->frag_list) { 1765677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 1766677e90edSThomas Graf goto next_skb; 1767677e90edSThomas Graf } 1768677e90edSThomas Graf 1769677e90edSThomas Graf return 0; 1770677e90edSThomas Graf } 1771677e90edSThomas Graf 1772677e90edSThomas Graf /** 1773677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 1774677e90edSThomas Graf * @st: state variable 1775677e90edSThomas Graf * 1776677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 1777677e90edSThomas Graf * returned 0. 1778677e90edSThomas Graf */ 1779677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 1780677e90edSThomas Graf { 1781677e90edSThomas Graf if (st->frag_data) 1782677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 1783677e90edSThomas Graf } 1784677e90edSThomas Graf 17853fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 17863fc7e8a6SThomas Graf 17873fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 17883fc7e8a6SThomas Graf struct ts_config *conf, 17893fc7e8a6SThomas Graf struct ts_state *state) 17903fc7e8a6SThomas Graf { 17913fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 17923fc7e8a6SThomas Graf } 17933fc7e8a6SThomas Graf 17943fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 17953fc7e8a6SThomas Graf { 17963fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 17973fc7e8a6SThomas Graf } 17983fc7e8a6SThomas Graf 17993fc7e8a6SThomas Graf /** 18003fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 18013fc7e8a6SThomas Graf * @skb: the buffer to look in 18023fc7e8a6SThomas Graf * @from: search offset 18033fc7e8a6SThomas Graf * @to: search limit 18043fc7e8a6SThomas Graf * @config: textsearch configuration 18053fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 18063fc7e8a6SThomas Graf * 18073fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 18083fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 18093fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 18103fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 18113fc7e8a6SThomas Graf */ 18123fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 18133fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 18143fc7e8a6SThomas Graf struct ts_state *state) 18153fc7e8a6SThomas Graf { 1816f72b948dSPhil Oester unsigned int ret; 1817f72b948dSPhil Oester 18183fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 18193fc7e8a6SThomas Graf config->finish = skb_ts_finish; 18203fc7e8a6SThomas Graf 18213fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 18223fc7e8a6SThomas Graf 1823f72b948dSPhil Oester ret = textsearch_find(config, state); 1824f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 18253fc7e8a6SThomas Graf } 18263fc7e8a6SThomas Graf 1827e89e9cf5SAnanda Raju /** 1828e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 1829e89e9cf5SAnanda Raju * @sk: sock structure 1830e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 1831e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 1832e89e9cf5SAnanda Raju * @from: pointer to user message iov 1833e89e9cf5SAnanda Raju * @length: length of the iov message 1834e89e9cf5SAnanda Raju * 1835e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 1836e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 1837e89e9cf5SAnanda Raju */ 1838e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1839dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 1840e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 1841e89e9cf5SAnanda Raju void *from, int length) 1842e89e9cf5SAnanda Raju { 1843e89e9cf5SAnanda Raju int frg_cnt = 0; 1844e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 1845e89e9cf5SAnanda Raju struct page *page = NULL; 1846e89e9cf5SAnanda Raju int copy, left; 1847e89e9cf5SAnanda Raju int offset = 0; 1848e89e9cf5SAnanda Raju int ret; 1849e89e9cf5SAnanda Raju 1850e89e9cf5SAnanda Raju do { 1851e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 1852e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 1853e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 1854e89e9cf5SAnanda Raju return -EFAULT; 1855e89e9cf5SAnanda Raju 1856e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 1857e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 1858e89e9cf5SAnanda Raju 1859e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 1860e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 1861e89e9cf5SAnanda Raju */ 1862e89e9cf5SAnanda Raju if (page == NULL) 1863e89e9cf5SAnanda Raju return -ENOMEM; 1864e89e9cf5SAnanda Raju 1865e89e9cf5SAnanda Raju /* initialize the next frag */ 1866e89e9cf5SAnanda Raju sk->sk_sndmsg_page = page; 1867e89e9cf5SAnanda Raju sk->sk_sndmsg_off = 0; 1868e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 1869e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 1870e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 1871e89e9cf5SAnanda Raju 1872e89e9cf5SAnanda Raju /* get the new initialized frag */ 1873e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 1874e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 1875e89e9cf5SAnanda Raju 1876e89e9cf5SAnanda Raju /* copy the user data to page */ 1877e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 1878e89e9cf5SAnanda Raju copy = (length > left)? left : length; 1879e89e9cf5SAnanda Raju 1880e89e9cf5SAnanda Raju ret = getfrag(from, (page_address(frag->page) + 1881e89e9cf5SAnanda Raju frag->page_offset + frag->size), 1882e89e9cf5SAnanda Raju offset, copy, 0, skb); 1883e89e9cf5SAnanda Raju if (ret < 0) 1884e89e9cf5SAnanda Raju return -EFAULT; 1885e89e9cf5SAnanda Raju 1886e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 1887e89e9cf5SAnanda Raju sk->sk_sndmsg_off += copy; 1888e89e9cf5SAnanda Raju frag->size += copy; 1889e89e9cf5SAnanda Raju skb->len += copy; 1890e89e9cf5SAnanda Raju skb->data_len += copy; 1891e89e9cf5SAnanda Raju offset += copy; 1892e89e9cf5SAnanda Raju length -= copy; 1893e89e9cf5SAnanda Raju 1894e89e9cf5SAnanda Raju } while (length > 0); 1895e89e9cf5SAnanda Raju 1896e89e9cf5SAnanda Raju return 0; 1897e89e9cf5SAnanda Raju } 1898e89e9cf5SAnanda Raju 1899cbb042f9SHerbert Xu /** 1900cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 1901cbb042f9SHerbert Xu * @skb: buffer to update 1902cbb042f9SHerbert Xu * @start: start of data before pull 1903cbb042f9SHerbert Xu * @len: length of data pulled 1904cbb042f9SHerbert Xu * 1905cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 190684fa7933SPatrick McHardy * update the CHECKSUM_COMPLETE checksum. It should be used on 190784fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 190884fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 190984fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 1910cbb042f9SHerbert Xu */ 1911cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 1912cbb042f9SHerbert Xu { 1913cbb042f9SHerbert Xu BUG_ON(len > skb->len); 1914cbb042f9SHerbert Xu skb->len -= len; 1915cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 1916cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 1917cbb042f9SHerbert Xu return skb->data += len; 1918cbb042f9SHerbert Xu } 1919cbb042f9SHerbert Xu 1920f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 1921f94691acSArnaldo Carvalho de Melo 1922f4c50d99SHerbert Xu /** 1923f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 1924f4c50d99SHerbert Xu * @skb: buffer to segment 1925576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 1926f4c50d99SHerbert Xu * 1927f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 1928f4c50d99SHerbert Xu * the segment at the given position. It returns NULL if there are 1929f4c50d99SHerbert Xu * no more segments to generate, or when an error is encountered. 1930f4c50d99SHerbert Xu */ 1931576a30ebSHerbert Xu struct sk_buff *skb_segment(struct sk_buff *skb, int features) 1932f4c50d99SHerbert Xu { 1933f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 1934f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 1935f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 1936f4c50d99SHerbert Xu unsigned int doffset = skb->data - skb->mac.raw; 1937f4c50d99SHerbert Xu unsigned int offset = doffset; 1938f4c50d99SHerbert Xu unsigned int headroom; 1939f4c50d99SHerbert Xu unsigned int len; 1940576a30ebSHerbert Xu int sg = features & NETIF_F_SG; 1941f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 1942f4c50d99SHerbert Xu int err = -ENOMEM; 1943f4c50d99SHerbert Xu int i = 0; 1944f4c50d99SHerbert Xu int pos; 1945f4c50d99SHerbert Xu 1946f4c50d99SHerbert Xu __skb_push(skb, doffset); 1947f4c50d99SHerbert Xu headroom = skb_headroom(skb); 1948f4c50d99SHerbert Xu pos = skb_headlen(skb); 1949f4c50d99SHerbert Xu 1950f4c50d99SHerbert Xu do { 1951f4c50d99SHerbert Xu struct sk_buff *nskb; 1952f4c50d99SHerbert Xu skb_frag_t *frag; 1953c8884eddSHerbert Xu int hsize; 1954f4c50d99SHerbert Xu int k; 1955f4c50d99SHerbert Xu int size; 1956f4c50d99SHerbert Xu 1957f4c50d99SHerbert Xu len = skb->len - offset; 1958f4c50d99SHerbert Xu if (len > mss) 1959f4c50d99SHerbert Xu len = mss; 1960f4c50d99SHerbert Xu 1961f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 1962f4c50d99SHerbert Xu if (hsize < 0) 1963f4c50d99SHerbert Xu hsize = 0; 1964c8884eddSHerbert Xu if (hsize > len || !sg) 1965c8884eddSHerbert Xu hsize = len; 1966f4c50d99SHerbert Xu 1967c8884eddSHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC); 1968f4c50d99SHerbert Xu if (unlikely(!nskb)) 1969f4c50d99SHerbert Xu goto err; 1970f4c50d99SHerbert Xu 1971f4c50d99SHerbert Xu if (segs) 1972f4c50d99SHerbert Xu tail->next = nskb; 1973f4c50d99SHerbert Xu else 1974f4c50d99SHerbert Xu segs = nskb; 1975f4c50d99SHerbert Xu tail = nskb; 1976f4c50d99SHerbert Xu 1977f4c50d99SHerbert Xu nskb->dev = skb->dev; 1978f4c50d99SHerbert Xu nskb->priority = skb->priority; 1979f4c50d99SHerbert Xu nskb->protocol = skb->protocol; 1980f4c50d99SHerbert Xu nskb->dst = dst_clone(skb->dst); 1981f4c50d99SHerbert Xu memcpy(nskb->cb, skb->cb, sizeof(skb->cb)); 1982f4c50d99SHerbert Xu nskb->pkt_type = skb->pkt_type; 1983f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 1984f4c50d99SHerbert Xu 1985f4c50d99SHerbert Xu skb_reserve(nskb, headroom); 1986f4c50d99SHerbert Xu nskb->mac.raw = nskb->data; 1987f4c50d99SHerbert Xu nskb->nh.raw = nskb->data + skb->mac_len; 1988f4c50d99SHerbert Xu nskb->h.raw = nskb->nh.raw + (skb->h.raw - skb->nh.raw); 1989f4c50d99SHerbert Xu memcpy(skb_put(nskb, doffset), skb->data, doffset); 1990f4c50d99SHerbert Xu 1991f4c50d99SHerbert Xu if (!sg) { 1992f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 1993f4c50d99SHerbert Xu skb_put(nskb, len), 1994f4c50d99SHerbert Xu len, 0); 1995f4c50d99SHerbert Xu continue; 1996f4c50d99SHerbert Xu } 1997f4c50d99SHerbert Xu 1998f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 1999f4c50d99SHerbert Xu k = 0; 2000f4c50d99SHerbert Xu 200184fa7933SPatrick McHardy nskb->ip_summed = CHECKSUM_PARTIAL; 2002f4c50d99SHerbert Xu nskb->csum = skb->csum; 2003f4c50d99SHerbert Xu memcpy(skb_put(nskb, hsize), skb->data + offset, hsize); 2004f4c50d99SHerbert Xu 2005f4c50d99SHerbert Xu while (pos < offset + len) { 2006f4c50d99SHerbert Xu BUG_ON(i >= nfrags); 2007f4c50d99SHerbert Xu 2008f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2009f4c50d99SHerbert Xu get_page(frag->page); 2010f4c50d99SHerbert Xu size = frag->size; 2011f4c50d99SHerbert Xu 2012f4c50d99SHerbert Xu if (pos < offset) { 2013f4c50d99SHerbert Xu frag->page_offset += offset - pos; 2014f4c50d99SHerbert Xu frag->size -= offset - pos; 2015f4c50d99SHerbert Xu } 2016f4c50d99SHerbert Xu 2017f4c50d99SHerbert Xu k++; 2018f4c50d99SHerbert Xu 2019f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2020f4c50d99SHerbert Xu i++; 2021f4c50d99SHerbert Xu pos += size; 2022f4c50d99SHerbert Xu } else { 2023f4c50d99SHerbert Xu frag->size -= pos + size - (offset + len); 2024f4c50d99SHerbert Xu break; 2025f4c50d99SHerbert Xu } 2026f4c50d99SHerbert Xu 2027f4c50d99SHerbert Xu frag++; 2028f4c50d99SHerbert Xu } 2029f4c50d99SHerbert Xu 2030f4c50d99SHerbert Xu skb_shinfo(nskb)->nr_frags = k; 2031f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2032f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2033f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2034f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2035f4c50d99SHerbert Xu 2036f4c50d99SHerbert Xu return segs; 2037f4c50d99SHerbert Xu 2038f4c50d99SHerbert Xu err: 2039f4c50d99SHerbert Xu while ((skb = segs)) { 2040f4c50d99SHerbert Xu segs = skb->next; 2041f4c50d99SHerbert Xu kfree(skb); 2042f4c50d99SHerbert Xu } 2043f4c50d99SHerbert Xu return ERR_PTR(err); 2044f4c50d99SHerbert Xu } 2045f4c50d99SHerbert Xu 2046f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2047f4c50d99SHerbert Xu 20481da177e4SLinus Torvalds void __init skb_init(void) 20491da177e4SLinus Torvalds { 20501da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 20511da177e4SLinus Torvalds sizeof(struct sk_buff), 20521da177e4SLinus Torvalds 0, 2053e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 20541da177e4SLinus Torvalds NULL, NULL); 2055d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2056d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2057d179cd12SDavid S. Miller sizeof(atomic_t), 2058d179cd12SDavid S. Miller 0, 2059e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 2060d179cd12SDavid S. Miller NULL, NULL); 20611da177e4SLinus Torvalds } 20621da177e4SLinus Torvalds 20631da177e4SLinus Torvalds EXPORT_SYMBOL(___pskb_trim); 20641da177e4SLinus Torvalds EXPORT_SYMBOL(__kfree_skb); 2065231d06aeSJörn Engel EXPORT_SYMBOL(kfree_skb); 20661da177e4SLinus Torvalds EXPORT_SYMBOL(__pskb_pull_tail); 2067d179cd12SDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 20688af27456SChristoph Hellwig EXPORT_SYMBOL(__netdev_alloc_skb); 20691da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_copy); 20701da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_expand_head); 20711da177e4SLinus Torvalds EXPORT_SYMBOL(skb_checksum); 20721da177e4SLinus Torvalds EXPORT_SYMBOL(skb_clone); 20731da177e4SLinus Torvalds EXPORT_SYMBOL(skb_clone_fraglist); 20741da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy); 20751da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_bits); 20761da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_dev); 20771da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_bits); 20781da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_expand); 20791da177e4SLinus Torvalds EXPORT_SYMBOL(skb_over_panic); 20801da177e4SLinus Torvalds EXPORT_SYMBOL(skb_pad); 20811da177e4SLinus Torvalds EXPORT_SYMBOL(skb_realloc_headroom); 20821da177e4SLinus Torvalds EXPORT_SYMBOL(skb_under_panic); 20831da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue); 20841da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue_tail); 20851da177e4SLinus Torvalds EXPORT_SYMBOL(skb_insert); 20861da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_purge); 20871da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_head); 20881da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_tail); 20891da177e4SLinus Torvalds EXPORT_SYMBOL(skb_unlink); 20901da177e4SLinus Torvalds EXPORT_SYMBOL(skb_append); 20911da177e4SLinus Torvalds EXPORT_SYMBOL(skb_split); 2092677e90edSThomas Graf EXPORT_SYMBOL(skb_prepare_seq_read); 2093677e90edSThomas Graf EXPORT_SYMBOL(skb_seq_read); 2094677e90edSThomas Graf EXPORT_SYMBOL(skb_abort_seq_read); 20953fc7e8a6SThomas Graf EXPORT_SYMBOL(skb_find_text); 2096e89e9cf5SAnanda Raju EXPORT_SYMBOL(skb_append_datato_frags); 2097