11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * Routines having to do with the 'struct sk_buff' memory handlers. 31da177e4SLinus Torvalds * 4113aa838SAlan Cox * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk> 51da177e4SLinus Torvalds * Florian La Roche <rzsfl@rz.uni-sb.de> 61da177e4SLinus Torvalds * 71da177e4SLinus Torvalds * Fixes: 81da177e4SLinus Torvalds * Alan Cox : Fixed the worst of the load 91da177e4SLinus Torvalds * balancer bugs. 101da177e4SLinus Torvalds * Dave Platt : Interrupt stacking fix. 111da177e4SLinus Torvalds * Richard Kooijman : Timestamp fixes. 121da177e4SLinus Torvalds * Alan Cox : Changed buffer format. 131da177e4SLinus Torvalds * Alan Cox : destructor hook for AF_UNIX etc. 141da177e4SLinus Torvalds * Linus Torvalds : Better skb_clone. 151da177e4SLinus Torvalds * Alan Cox : Added skb_copy. 161da177e4SLinus Torvalds * Alan Cox : Added all the changed routines Linus 171da177e4SLinus Torvalds * only put in the headers 181da177e4SLinus Torvalds * Ray VanTassle : Fixed --skb->lock in free 191da177e4SLinus Torvalds * Alan Cox : skb_copy copy arp field 201da177e4SLinus Torvalds * Andi Kleen : slabified it. 211da177e4SLinus Torvalds * Robert Olsson : Removed skb_head_pool 221da177e4SLinus Torvalds * 231da177e4SLinus Torvalds * NOTE: 241da177e4SLinus Torvalds * The __skb_ routines should be called with interrupts 251da177e4SLinus Torvalds * disabled, or you better be *real* sure that the operation is atomic 261da177e4SLinus Torvalds * with respect to whatever list is being frobbed (e.g. via lock_sock() 271da177e4SLinus Torvalds * or via disabling bottom half handlers, etc). 281da177e4SLinus Torvalds * 291da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 301da177e4SLinus Torvalds * modify it under the terms of the GNU General Public License 311da177e4SLinus Torvalds * as published by the Free Software Foundation; either version 321da177e4SLinus Torvalds * 2 of the License, or (at your option) any later version. 331da177e4SLinus Torvalds */ 341da177e4SLinus Torvalds 351da177e4SLinus Torvalds /* 361da177e4SLinus Torvalds * The functions in this file will not compile correctly with gcc 2.4.x 371da177e4SLinus Torvalds */ 381da177e4SLinus Torvalds 391da177e4SLinus Torvalds #include <linux/module.h> 401da177e4SLinus Torvalds #include <linux/types.h> 411da177e4SLinus Torvalds #include <linux/kernel.h> 421da177e4SLinus Torvalds #include <linux/mm.h> 431da177e4SLinus Torvalds #include <linux/interrupt.h> 441da177e4SLinus Torvalds #include <linux/in.h> 451da177e4SLinus Torvalds #include <linux/inet.h> 461da177e4SLinus Torvalds #include <linux/slab.h> 471da177e4SLinus Torvalds #include <linux/netdevice.h> 481da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 491da177e4SLinus Torvalds #include <net/pkt_sched.h> 501da177e4SLinus Torvalds #endif 511da177e4SLinus Torvalds #include <linux/string.h> 521da177e4SLinus Torvalds #include <linux/skbuff.h> 539c55e01cSJens Axboe #include <linux/splice.h> 541da177e4SLinus Torvalds #include <linux/cache.h> 551da177e4SLinus Torvalds #include <linux/rtnetlink.h> 561da177e4SLinus Torvalds #include <linux/init.h> 57716ea3a7SDavid Howells #include <linux/scatterlist.h> 581da177e4SLinus Torvalds 591da177e4SLinus Torvalds #include <net/protocol.h> 601da177e4SLinus Torvalds #include <net/dst.h> 611da177e4SLinus Torvalds #include <net/sock.h> 621da177e4SLinus Torvalds #include <net/checksum.h> 631da177e4SLinus Torvalds #include <net/xfrm.h> 641da177e4SLinus Torvalds 651da177e4SLinus Torvalds #include <asm/uaccess.h> 661da177e4SLinus Torvalds #include <asm/system.h> 671da177e4SLinus Torvalds 68a1f8e7f7SAl Viro #include "kmap_skb.h" 69a1f8e7f7SAl Viro 70e18b890bSChristoph Lameter static struct kmem_cache *skbuff_head_cache __read_mostly; 71e18b890bSChristoph Lameter static struct kmem_cache *skbuff_fclone_cache __read_mostly; 721da177e4SLinus Torvalds 739c55e01cSJens Axboe static void sock_pipe_buf_release(struct pipe_inode_info *pipe, 749c55e01cSJens Axboe struct pipe_buffer *buf) 759c55e01cSJens Axboe { 768b9d3728SJarek Poplawski put_page(buf->page); 779c55e01cSJens Axboe } 789c55e01cSJens Axboe 799c55e01cSJens Axboe static void sock_pipe_buf_get(struct pipe_inode_info *pipe, 809c55e01cSJens Axboe struct pipe_buffer *buf) 819c55e01cSJens Axboe { 828b9d3728SJarek Poplawski get_page(buf->page); 839c55e01cSJens Axboe } 849c55e01cSJens Axboe 859c55e01cSJens Axboe static int sock_pipe_buf_steal(struct pipe_inode_info *pipe, 869c55e01cSJens Axboe struct pipe_buffer *buf) 879c55e01cSJens Axboe { 889c55e01cSJens Axboe return 1; 899c55e01cSJens Axboe } 909c55e01cSJens Axboe 919c55e01cSJens Axboe 929c55e01cSJens Axboe /* Pipe buffer operations for a socket. */ 939c55e01cSJens Axboe static struct pipe_buf_operations sock_pipe_buf_ops = { 949c55e01cSJens Axboe .can_merge = 0, 959c55e01cSJens Axboe .map = generic_pipe_buf_map, 969c55e01cSJens Axboe .unmap = generic_pipe_buf_unmap, 979c55e01cSJens Axboe .confirm = generic_pipe_buf_confirm, 989c55e01cSJens Axboe .release = sock_pipe_buf_release, 999c55e01cSJens Axboe .steal = sock_pipe_buf_steal, 1009c55e01cSJens Axboe .get = sock_pipe_buf_get, 1019c55e01cSJens Axboe }; 1029c55e01cSJens Axboe 1031da177e4SLinus Torvalds /* 1041da177e4SLinus Torvalds * Keep out-of-line to prevent kernel bloat. 1051da177e4SLinus Torvalds * __builtin_return_address is not used because it is not always 1061da177e4SLinus Torvalds * reliable. 1071da177e4SLinus Torvalds */ 1081da177e4SLinus Torvalds 1091da177e4SLinus Torvalds /** 1101da177e4SLinus Torvalds * skb_over_panic - private function 1111da177e4SLinus Torvalds * @skb: buffer 1121da177e4SLinus Torvalds * @sz: size 1131da177e4SLinus Torvalds * @here: address 1141da177e4SLinus Torvalds * 1151da177e4SLinus Torvalds * Out of line support code for skb_put(). Not user callable. 1161da177e4SLinus Torvalds */ 1171da177e4SLinus Torvalds void skb_over_panic(struct sk_buff *skb, int sz, void *here) 1181da177e4SLinus Torvalds { 11926095455SPatrick McHardy printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p " 1204305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 12127a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1224305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 12326095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1241da177e4SLinus Torvalds BUG(); 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds /** 1281da177e4SLinus Torvalds * skb_under_panic - private function 1291da177e4SLinus Torvalds * @skb: buffer 1301da177e4SLinus Torvalds * @sz: size 1311da177e4SLinus Torvalds * @here: address 1321da177e4SLinus Torvalds * 1331da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1341da177e4SLinus Torvalds */ 1351da177e4SLinus Torvalds 1361da177e4SLinus Torvalds void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1371da177e4SLinus Torvalds { 13826095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 1394305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 14027a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1414305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 14226095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1431da177e4SLinus Torvalds BUG(); 1441da177e4SLinus Torvalds } 1451da177e4SLinus Torvalds 146dc6de336SDavid S. Miller void skb_truesize_bug(struct sk_buff *skb) 147dc6de336SDavid S. Miller { 1488f480c0eSArjan van de Ven WARN(net_ratelimit(), KERN_ERR "SKB BUG: Invalid truesize (%u) " 149dc6de336SDavid S. Miller "len=%u, sizeof(sk_buff)=%Zd\n", 150dc6de336SDavid S. Miller skb->truesize, skb->len, sizeof(struct sk_buff)); 151dc6de336SDavid S. Miller } 152dc6de336SDavid S. Miller EXPORT_SYMBOL(skb_truesize_bug); 153dc6de336SDavid S. Miller 1541da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1551da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1561da177e4SLinus Torvalds * [BEEP] leaks. 1571da177e4SLinus Torvalds * 1581da177e4SLinus Torvalds */ 1591da177e4SLinus Torvalds 1601da177e4SLinus Torvalds /** 161d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1621da177e4SLinus Torvalds * @size: size to allocate 1631da177e4SLinus Torvalds * @gfp_mask: allocation mask 164c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 165c83c2486SRandy Dunlap * and allocate a cloned (child) skb 166b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1671da177e4SLinus Torvalds * 1681da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1691da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1701da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1711da177e4SLinus Torvalds * 1721da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1731da177e4SLinus Torvalds * %GFP_ATOMIC. 1741da177e4SLinus Torvalds */ 175dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 176b30973f8SChristoph Hellwig int fclone, int node) 1771da177e4SLinus Torvalds { 178e18b890bSChristoph Lameter struct kmem_cache *cache; 1794947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1801da177e4SLinus Torvalds struct sk_buff *skb; 1811da177e4SLinus Torvalds u8 *data; 1821da177e4SLinus Torvalds 1838798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1848798b3fbSHerbert Xu 1851da177e4SLinus Torvalds /* Get the HEAD */ 186b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1871da177e4SLinus Torvalds if (!skb) 1881da177e4SLinus Torvalds goto out; 1891da177e4SLinus Torvalds 1901da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 191b30973f8SChristoph Hellwig data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), 192b30973f8SChristoph Hellwig gfp_mask, node); 1931da177e4SLinus Torvalds if (!data) 1941da177e4SLinus Torvalds goto nodata; 1951da177e4SLinus Torvalds 196ca0605a7SArnaldo Carvalho de Melo /* 197c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 198c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 199c8005785SJohannes Berg * the tail pointer in struct sk_buff! 200ca0605a7SArnaldo Carvalho de Melo */ 201ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 2021da177e4SLinus Torvalds skb->truesize = size + sizeof(struct sk_buff); 2031da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2041da177e4SLinus Torvalds skb->head = data; 2051da177e4SLinus Torvalds skb->data = data; 20627a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2074305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 2084947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2094947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 2104947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2114947d3efSBenjamin LaHaise shinfo->nr_frags = 0; 2127967168cSHerbert Xu shinfo->gso_size = 0; 2137967168cSHerbert Xu shinfo->gso_segs = 0; 2147967168cSHerbert Xu shinfo->gso_type = 0; 2154947d3efSBenjamin LaHaise shinfo->ip6_frag_id = 0; 2164947d3efSBenjamin LaHaise shinfo->frag_list = NULL; 2174947d3efSBenjamin LaHaise 218d179cd12SDavid S. Miller if (fclone) { 219d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 220d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2211da177e4SLinus Torvalds 222d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 223d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 224d179cd12SDavid S. Miller 225d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 226d179cd12SDavid S. Miller } 2271da177e4SLinus Torvalds out: 2281da177e4SLinus Torvalds return skb; 2291da177e4SLinus Torvalds nodata: 2308798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2311da177e4SLinus Torvalds skb = NULL; 2321da177e4SLinus Torvalds goto out; 2331da177e4SLinus Torvalds } 2341da177e4SLinus Torvalds 2351da177e4SLinus Torvalds /** 2368af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2378af27456SChristoph Hellwig * @dev: network device to receive on 2388af27456SChristoph Hellwig * @length: length to allocate 2398af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 2408af27456SChristoph Hellwig * 2418af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 2428af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 2438af27456SChristoph Hellwig * the headroom they think they need without accounting for the 2448af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 2458af27456SChristoph Hellwig * 2468af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 2478af27456SChristoph Hellwig */ 2488af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 2498af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 2508af27456SChristoph Hellwig { 25143cb76d9SGreg Kroah-Hartman int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; 2528af27456SChristoph Hellwig struct sk_buff *skb; 2538af27456SChristoph Hellwig 254b30973f8SChristoph Hellwig skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node); 2557b2e497aSChristoph Hellwig if (likely(skb)) { 2568af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 2577b2e497aSChristoph Hellwig skb->dev = dev; 2587b2e497aSChristoph Hellwig } 2598af27456SChristoph Hellwig return skb; 2608af27456SChristoph Hellwig } 2611da177e4SLinus Torvalds 262654bed16SPeter Zijlstra struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask) 263654bed16SPeter Zijlstra { 264654bed16SPeter Zijlstra int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; 265654bed16SPeter Zijlstra struct page *page; 266654bed16SPeter Zijlstra 267654bed16SPeter Zijlstra page = alloc_pages_node(node, gfp_mask, 0); 268654bed16SPeter Zijlstra return page; 269654bed16SPeter Zijlstra } 270654bed16SPeter Zijlstra EXPORT_SYMBOL(__netdev_alloc_page); 271654bed16SPeter Zijlstra 272654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 273654bed16SPeter Zijlstra int size) 274654bed16SPeter Zijlstra { 275654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 276654bed16SPeter Zijlstra skb->len += size; 277654bed16SPeter Zijlstra skb->data_len += size; 278654bed16SPeter Zijlstra skb->truesize += size; 279654bed16SPeter Zijlstra } 280654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 281654bed16SPeter Zijlstra 282f58518e6SIlpo Järvinen /** 283f58518e6SIlpo Järvinen * dev_alloc_skb - allocate an skbuff for receiving 284f58518e6SIlpo Järvinen * @length: length to allocate 285f58518e6SIlpo Järvinen * 286f58518e6SIlpo Järvinen * Allocate a new &sk_buff and assign it a usage count of one. The 287f58518e6SIlpo Järvinen * buffer has unspecified headroom built in. Users should allocate 288f58518e6SIlpo Järvinen * the headroom they think they need without accounting for the 289f58518e6SIlpo Järvinen * built in space. The built in space is used for optimisations. 290f58518e6SIlpo Järvinen * 291f58518e6SIlpo Järvinen * %NULL is returned if there is no free memory. Although this function 292f58518e6SIlpo Järvinen * allocates memory it can be called from an interrupt. 293f58518e6SIlpo Järvinen */ 294f58518e6SIlpo Järvinen struct sk_buff *dev_alloc_skb(unsigned int length) 295f58518e6SIlpo Järvinen { 2961483b874SDenys Vlasenko /* 2971483b874SDenys Vlasenko * There is more code here than it seems: 298a0f55e0eSDavid S. Miller * __dev_alloc_skb is an inline 2991483b874SDenys Vlasenko */ 300f58518e6SIlpo Järvinen return __dev_alloc_skb(length, GFP_ATOMIC); 301f58518e6SIlpo Järvinen } 302f58518e6SIlpo Järvinen EXPORT_SYMBOL(dev_alloc_skb); 303f58518e6SIlpo Järvinen 30427b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 3051da177e4SLinus Torvalds { 30627b437c8SHerbert Xu struct sk_buff *list = *listp; 3071da177e4SLinus Torvalds 30827b437c8SHerbert Xu *listp = NULL; 3091da177e4SLinus Torvalds 3101da177e4SLinus Torvalds do { 3111da177e4SLinus Torvalds struct sk_buff *this = list; 3121da177e4SLinus Torvalds list = list->next; 3131da177e4SLinus Torvalds kfree_skb(this); 3141da177e4SLinus Torvalds } while (list); 3151da177e4SLinus Torvalds } 3161da177e4SLinus Torvalds 31727b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 31827b437c8SHerbert Xu { 31927b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 32027b437c8SHerbert Xu } 32127b437c8SHerbert Xu 3221da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3231da177e4SLinus Torvalds { 3241da177e4SLinus Torvalds struct sk_buff *list; 3251da177e4SLinus Torvalds 3261da177e4SLinus Torvalds for (list = skb_shinfo(skb)->frag_list; list; list = list->next) 3271da177e4SLinus Torvalds skb_get(list); 3281da177e4SLinus Torvalds } 3291da177e4SLinus Torvalds 3305bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3311da177e4SLinus Torvalds { 3321da177e4SLinus Torvalds if (!skb->cloned || 3331da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3341da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3351da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3361da177e4SLinus Torvalds int i; 3371da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 3381da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 3391da177e4SLinus Torvalds } 3401da177e4SLinus Torvalds 3411da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 3421da177e4SLinus Torvalds skb_drop_fraglist(skb); 3431da177e4SLinus Torvalds 3441da177e4SLinus Torvalds kfree(skb->head); 3451da177e4SLinus Torvalds } 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds /* 3491da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 3501da177e4SLinus Torvalds */ 3512d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 3521da177e4SLinus Torvalds { 353d179cd12SDavid S. Miller struct sk_buff *other; 354d179cd12SDavid S. Miller atomic_t *fclone_ref; 355d179cd12SDavid S. Miller 356d179cd12SDavid S. Miller switch (skb->fclone) { 357d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 3581da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 359d179cd12SDavid S. Miller break; 360d179cd12SDavid S. Miller 361d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 362d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 363d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 364d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 365d179cd12SDavid S. Miller break; 366d179cd12SDavid S. Miller 367d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 368d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 369d179cd12SDavid S. Miller other = skb - 1; 370d179cd12SDavid S. Miller 371d179cd12SDavid S. Miller /* The clone portion is available for 372d179cd12SDavid S. Miller * fast-cloning again. 373d179cd12SDavid S. Miller */ 374d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 375d179cd12SDavid S. Miller 376d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 377d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 378d179cd12SDavid S. Miller break; 3793ff50b79SStephen Hemminger } 3801da177e4SLinus Torvalds } 3811da177e4SLinus Torvalds 38204a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 3831da177e4SLinus Torvalds { 3841da177e4SLinus Torvalds dst_release(skb->dst); 3851da177e4SLinus Torvalds #ifdef CONFIG_XFRM 3861da177e4SLinus Torvalds secpath_put(skb->sp); 3871da177e4SLinus Torvalds #endif 3881da177e4SLinus Torvalds if (skb->destructor) { 3899c2b3328SStephen Hemminger WARN_ON(in_irq()); 3901da177e4SLinus Torvalds skb->destructor(skb); 3911da177e4SLinus Torvalds } 3929fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3935f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 3949fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 3959fb9cbb1SYasuyuki Kozakai #endif 3961da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 3971da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 3981da177e4SLinus Torvalds #endif 3991da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 4001da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4011da177e4SLinus Torvalds skb->tc_index = 0; 4021da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 4031da177e4SLinus Torvalds skb->tc_verd = 0; 4041da177e4SLinus Torvalds #endif 4051da177e4SLinus Torvalds #endif 40604a4bb55SLennert Buytenhek } 40704a4bb55SLennert Buytenhek 40804a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 40904a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 41004a4bb55SLennert Buytenhek { 41104a4bb55SLennert Buytenhek skb_release_head_state(skb); 4122d4baff8SHerbert Xu skb_release_data(skb); 4132d4baff8SHerbert Xu } 4141da177e4SLinus Torvalds 4152d4baff8SHerbert Xu /** 4162d4baff8SHerbert Xu * __kfree_skb - private function 4172d4baff8SHerbert Xu * @skb: buffer 4182d4baff8SHerbert Xu * 4192d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 4202d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 4212d4baff8SHerbert Xu * always call kfree_skb 4222d4baff8SHerbert Xu */ 4232d4baff8SHerbert Xu 4242d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 4252d4baff8SHerbert Xu { 4262d4baff8SHerbert Xu skb_release_all(skb); 4271da177e4SLinus Torvalds kfree_skbmem(skb); 4281da177e4SLinus Torvalds } 4291da177e4SLinus Torvalds 4301da177e4SLinus Torvalds /** 431231d06aeSJörn Engel * kfree_skb - free an sk_buff 432231d06aeSJörn Engel * @skb: buffer to free 433231d06aeSJörn Engel * 434231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 435231d06aeSJörn Engel * hit zero. 436231d06aeSJörn Engel */ 437231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 438231d06aeSJörn Engel { 439231d06aeSJörn Engel if (unlikely(!skb)) 440231d06aeSJörn Engel return; 441231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 442231d06aeSJörn Engel smp_rmb(); 443231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 444231d06aeSJörn Engel return; 445231d06aeSJörn Engel __kfree_skb(skb); 446231d06aeSJörn Engel } 447231d06aeSJörn Engel 448d1a203eaSStephen Hemminger /** 449d1a203eaSStephen Hemminger * skb_recycle_check - check if skb can be reused for receive 450d1a203eaSStephen Hemminger * @skb: buffer 451d1a203eaSStephen Hemminger * @skb_size: minimum receive buffer size 452d1a203eaSStephen Hemminger * 453d1a203eaSStephen Hemminger * Checks that the skb passed in is not shared or cloned, and 454d1a203eaSStephen Hemminger * that it is linear and its head portion at least as large as 455d1a203eaSStephen Hemminger * skb_size so that it can be recycled as a receive buffer. 456d1a203eaSStephen Hemminger * If these conditions are met, this function does any necessary 457d1a203eaSStephen Hemminger * reference count dropping and cleans up the skbuff as if it 458d1a203eaSStephen Hemminger * just came from __alloc_skb(). 459d1a203eaSStephen Hemminger */ 46004a4bb55SLennert Buytenhek int skb_recycle_check(struct sk_buff *skb, int skb_size) 46104a4bb55SLennert Buytenhek { 46204a4bb55SLennert Buytenhek struct skb_shared_info *shinfo; 46304a4bb55SLennert Buytenhek 46404a4bb55SLennert Buytenhek if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) 46504a4bb55SLennert Buytenhek return 0; 46604a4bb55SLennert Buytenhek 46704a4bb55SLennert Buytenhek skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); 46804a4bb55SLennert Buytenhek if (skb_end_pointer(skb) - skb->head < skb_size) 46904a4bb55SLennert Buytenhek return 0; 47004a4bb55SLennert Buytenhek 47104a4bb55SLennert Buytenhek if (skb_shared(skb) || skb_cloned(skb)) 47204a4bb55SLennert Buytenhek return 0; 47304a4bb55SLennert Buytenhek 47404a4bb55SLennert Buytenhek skb_release_head_state(skb); 47504a4bb55SLennert Buytenhek shinfo = skb_shinfo(skb); 47604a4bb55SLennert Buytenhek atomic_set(&shinfo->dataref, 1); 47704a4bb55SLennert Buytenhek shinfo->nr_frags = 0; 47804a4bb55SLennert Buytenhek shinfo->gso_size = 0; 47904a4bb55SLennert Buytenhek shinfo->gso_segs = 0; 48004a4bb55SLennert Buytenhek shinfo->gso_type = 0; 48104a4bb55SLennert Buytenhek shinfo->ip6_frag_id = 0; 48204a4bb55SLennert Buytenhek shinfo->frag_list = NULL; 48304a4bb55SLennert Buytenhek 48404a4bb55SLennert Buytenhek memset(skb, 0, offsetof(struct sk_buff, tail)); 48504a4bb55SLennert Buytenhek skb->data = skb->head + NET_SKB_PAD; 4865cd33db2SLennert Buytenhek skb_reset_tail_pointer(skb); 48704a4bb55SLennert Buytenhek 48804a4bb55SLennert Buytenhek return 1; 48904a4bb55SLennert Buytenhek } 49004a4bb55SLennert Buytenhek EXPORT_SYMBOL(skb_recycle_check); 49104a4bb55SLennert Buytenhek 492dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 493dec18810SHerbert Xu { 494dec18810SHerbert Xu new->tstamp = old->tstamp; 495dec18810SHerbert Xu new->dev = old->dev; 496dec18810SHerbert Xu new->transport_header = old->transport_header; 497dec18810SHerbert Xu new->network_header = old->network_header; 498dec18810SHerbert Xu new->mac_header = old->mac_header; 499dec18810SHerbert Xu new->dst = dst_clone(old->dst); 500def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 501dec18810SHerbert Xu new->sp = secpath_get(old->sp); 502dec18810SHerbert Xu #endif 503dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 504dec18810SHerbert Xu new->csum_start = old->csum_start; 505dec18810SHerbert Xu new->csum_offset = old->csum_offset; 506dec18810SHerbert Xu new->local_df = old->local_df; 507dec18810SHerbert Xu new->pkt_type = old->pkt_type; 508dec18810SHerbert Xu new->ip_summed = old->ip_summed; 509dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 510dec18810SHerbert Xu new->priority = old->priority; 511dec18810SHerbert Xu #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 512dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 513dec18810SHerbert Xu #endif 514dec18810SHerbert Xu new->protocol = old->protocol; 515dec18810SHerbert Xu new->mark = old->mark; 516dec18810SHerbert Xu __nf_copy(new, old); 517dec18810SHerbert Xu #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ 518dec18810SHerbert Xu defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) 519dec18810SHerbert Xu new->nf_trace = old->nf_trace; 520dec18810SHerbert Xu #endif 521dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 522dec18810SHerbert Xu new->tc_index = old->tc_index; 523dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 524dec18810SHerbert Xu new->tc_verd = old->tc_verd; 525dec18810SHerbert Xu #endif 526dec18810SHerbert Xu #endif 5276aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 5286aa895b0SPatrick McHardy 529dec18810SHerbert Xu skb_copy_secmark(new, old); 530dec18810SHerbert Xu } 531dec18810SHerbert Xu 532e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 5331da177e4SLinus Torvalds { 5341da177e4SLinus Torvalds #define C(x) n->x = skb->x 5351da177e4SLinus Torvalds 5361da177e4SLinus Torvalds n->next = n->prev = NULL; 5371da177e4SLinus Torvalds n->sk = NULL; 538dec18810SHerbert Xu __copy_skb_header(n, skb); 539dec18810SHerbert Xu 5401da177e4SLinus Torvalds C(len); 5411da177e4SLinus Torvalds C(data_len); 5423e6b3b2eSAlexey Dobriyan C(mac_len); 543334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 54402f1c89dSPaul Moore n->cloned = 1; 5451da177e4SLinus Torvalds n->nohdr = 0; 5461da177e4SLinus Torvalds n->destructor = NULL; 54702f1c89dSPaul Moore C(iif); 5481da177e4SLinus Torvalds C(tail); 5491da177e4SLinus Torvalds C(end); 55002f1c89dSPaul Moore C(head); 55102f1c89dSPaul Moore C(data); 55202f1c89dSPaul Moore C(truesize); 553d0f09804SJohannes Berg #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE) 554d0f09804SJohannes Berg C(do_not_encrypt); 5558b30b1feSSujith C(requeue); 556d0f09804SJohannes Berg #endif 55702f1c89dSPaul Moore atomic_set(&n->users, 1); 5581da177e4SLinus Torvalds 5591da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 5601da177e4SLinus Torvalds skb->cloned = 1; 5611da177e4SLinus Torvalds 5621da177e4SLinus Torvalds return n; 563e0053ec0SHerbert Xu #undef C 564e0053ec0SHerbert Xu } 565e0053ec0SHerbert Xu 566e0053ec0SHerbert Xu /** 567e0053ec0SHerbert Xu * skb_morph - morph one skb into another 568e0053ec0SHerbert Xu * @dst: the skb to receive the contents 569e0053ec0SHerbert Xu * @src: the skb to supply the contents 570e0053ec0SHerbert Xu * 571e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 572e0053ec0SHerbert Xu * supplied by the user. 573e0053ec0SHerbert Xu * 574e0053ec0SHerbert Xu * The target skb is returned upon exit. 575e0053ec0SHerbert Xu */ 576e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 577e0053ec0SHerbert Xu { 5782d4baff8SHerbert Xu skb_release_all(dst); 579e0053ec0SHerbert Xu return __skb_clone(dst, src); 580e0053ec0SHerbert Xu } 581e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 582e0053ec0SHerbert Xu 583e0053ec0SHerbert Xu /** 584e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 585e0053ec0SHerbert Xu * @skb: buffer to clone 586e0053ec0SHerbert Xu * @gfp_mask: allocation priority 587e0053ec0SHerbert Xu * 588e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 589e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 590e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 591e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 592e0053ec0SHerbert Xu * 593e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 594e0053ec0SHerbert Xu * %GFP_ATOMIC. 595e0053ec0SHerbert Xu */ 596e0053ec0SHerbert Xu 597e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 598e0053ec0SHerbert Xu { 599e0053ec0SHerbert Xu struct sk_buff *n; 600e0053ec0SHerbert Xu 601e0053ec0SHerbert Xu n = skb + 1; 602e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 603e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 604e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 605e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 606e0053ec0SHerbert Xu atomic_inc(fclone_ref); 607e0053ec0SHerbert Xu } else { 608e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 609e0053ec0SHerbert Xu if (!n) 610e0053ec0SHerbert Xu return NULL; 611e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 612e0053ec0SHerbert Xu } 613e0053ec0SHerbert Xu 614e0053ec0SHerbert Xu return __skb_clone(n, skb); 6151da177e4SLinus Torvalds } 6161da177e4SLinus Torvalds 6171da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 6181da177e4SLinus Torvalds { 6192e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6201da177e4SLinus Torvalds /* 6211da177e4SLinus Torvalds * Shift between the two data areas in bytes 6221da177e4SLinus Torvalds */ 6231da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 6242e07fa9cSArnaldo Carvalho de Melo #endif 625dec18810SHerbert Xu 626dec18810SHerbert Xu __copy_skb_header(new, old); 627dec18810SHerbert Xu 6282e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6292e07fa9cSArnaldo Carvalho de Melo /* {transport,network,mac}_header are relative to skb->head */ 6302e07fa9cSArnaldo Carvalho de Melo new->transport_header += offset; 6312e07fa9cSArnaldo Carvalho de Melo new->network_header += offset; 6322e07fa9cSArnaldo Carvalho de Melo new->mac_header += offset; 6332e07fa9cSArnaldo Carvalho de Melo #endif 6347967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 6357967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 6367967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds 6391da177e4SLinus Torvalds /** 6401da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 6411da177e4SLinus Torvalds * @skb: buffer to copy 6421da177e4SLinus Torvalds * @gfp_mask: allocation priority 6431da177e4SLinus Torvalds * 6441da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 6451da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 6461da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 6471da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 6481da177e4SLinus Torvalds * 6491da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 6501da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 6511da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 6521da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 6531da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 6541da177e4SLinus Torvalds */ 6551da177e4SLinus Torvalds 656dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 6571da177e4SLinus Torvalds { 6581da177e4SLinus Torvalds int headerlen = skb->data - skb->head; 6591da177e4SLinus Torvalds /* 6601da177e4SLinus Torvalds * Allocate the copy buffer 6611da177e4SLinus Torvalds */ 6624305b541SArnaldo Carvalho de Melo struct sk_buff *n; 6634305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 6644305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end + skb->data_len, gfp_mask); 6654305b541SArnaldo Carvalho de Melo #else 6664305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask); 6674305b541SArnaldo Carvalho de Melo #endif 6681da177e4SLinus Torvalds if (!n) 6691da177e4SLinus Torvalds return NULL; 6701da177e4SLinus Torvalds 6711da177e4SLinus Torvalds /* Set the data pointer */ 6721da177e4SLinus Torvalds skb_reserve(n, headerlen); 6731da177e4SLinus Torvalds /* Set the tail pointer and length */ 6741da177e4SLinus Torvalds skb_put(n, skb->len); 6751da177e4SLinus Torvalds 6761da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 6771da177e4SLinus Torvalds BUG(); 6781da177e4SLinus Torvalds 6791da177e4SLinus Torvalds copy_skb_header(n, skb); 6801da177e4SLinus Torvalds return n; 6811da177e4SLinus Torvalds } 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds 6841da177e4SLinus Torvalds /** 6851da177e4SLinus Torvalds * pskb_copy - create copy of an sk_buff with private head. 6861da177e4SLinus Torvalds * @skb: buffer to copy 6871da177e4SLinus Torvalds * @gfp_mask: allocation priority 6881da177e4SLinus Torvalds * 6891da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 6901da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 6911da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 6921da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 6931da177e4SLinus Torvalds * or the pointer to the buffer on success. 6941da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 6951da177e4SLinus Torvalds */ 6961da177e4SLinus Torvalds 697dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) 6981da177e4SLinus Torvalds { 6991da177e4SLinus Torvalds /* 7001da177e4SLinus Torvalds * Allocate the copy buffer 7011da177e4SLinus Torvalds */ 7024305b541SArnaldo Carvalho de Melo struct sk_buff *n; 7034305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 7044305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end, gfp_mask); 7054305b541SArnaldo Carvalho de Melo #else 7064305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end - skb->head, gfp_mask); 7074305b541SArnaldo Carvalho de Melo #endif 7081da177e4SLinus Torvalds if (!n) 7091da177e4SLinus Torvalds goto out; 7101da177e4SLinus Torvalds 7111da177e4SLinus Torvalds /* Set the data pointer */ 7121da177e4SLinus Torvalds skb_reserve(n, skb->data - skb->head); 7131da177e4SLinus Torvalds /* Set the tail pointer and length */ 7141da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 7151da177e4SLinus Torvalds /* Copy the bytes */ 716d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 7171da177e4SLinus Torvalds 71825f484a6SHerbert Xu n->truesize += skb->data_len; 7191da177e4SLinus Torvalds n->data_len = skb->data_len; 7201da177e4SLinus Torvalds n->len = skb->len; 7211da177e4SLinus Torvalds 7221da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 7231da177e4SLinus Torvalds int i; 7241da177e4SLinus Torvalds 7251da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 7261da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 7271da177e4SLinus Torvalds get_page(skb_shinfo(n)->frags[i].page); 7281da177e4SLinus Torvalds } 7291da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 7301da177e4SLinus Torvalds } 7311da177e4SLinus Torvalds 7321da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 7331da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 7341da177e4SLinus Torvalds skb_clone_fraglist(n); 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds 7371da177e4SLinus Torvalds copy_skb_header(n, skb); 7381da177e4SLinus Torvalds out: 7391da177e4SLinus Torvalds return n; 7401da177e4SLinus Torvalds } 7411da177e4SLinus Torvalds 7421da177e4SLinus Torvalds /** 7431da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 7441da177e4SLinus Torvalds * @skb: buffer to reallocate 7451da177e4SLinus Torvalds * @nhead: room to add at head 7461da177e4SLinus Torvalds * @ntail: room to add at tail 7471da177e4SLinus Torvalds * @gfp_mask: allocation priority 7481da177e4SLinus Torvalds * 7491da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 7501da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 7511da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 7521da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 7531da177e4SLinus Torvalds * 7541da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 7551da177e4SLinus Torvalds * reloaded after call to this function. 7561da177e4SLinus Torvalds */ 7571da177e4SLinus Torvalds 75886a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 759dd0fc66fSAl Viro gfp_t gfp_mask) 7601da177e4SLinus Torvalds { 7611da177e4SLinus Torvalds int i; 7621da177e4SLinus Torvalds u8 *data; 7634305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 7644305b541SArnaldo Carvalho de Melo int size = nhead + skb->end + ntail; 7654305b541SArnaldo Carvalho de Melo #else 7661da177e4SLinus Torvalds int size = nhead + (skb->end - skb->head) + ntail; 7674305b541SArnaldo Carvalho de Melo #endif 7681da177e4SLinus Torvalds long off; 7691da177e4SLinus Torvalds 7704edd87adSHerbert Xu BUG_ON(nhead < 0); 7714edd87adSHerbert Xu 7721da177e4SLinus Torvalds if (skb_shared(skb)) 7731da177e4SLinus Torvalds BUG(); 7741da177e4SLinus Torvalds 7751da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 7761da177e4SLinus Torvalds 7771da177e4SLinus Torvalds data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 7781da177e4SLinus Torvalds if (!data) 7791da177e4SLinus Torvalds goto nodata; 7801da177e4SLinus Torvalds 7811da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 7821da177e4SLinus Torvalds * optimized for the cases when header is void. */ 7834305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 784b6ccc67dSMikael Pettersson memcpy(data + nhead, skb->head, skb->tail); 7854305b541SArnaldo Carvalho de Melo #else 786b6ccc67dSMikael Pettersson memcpy(data + nhead, skb->head, skb->tail - skb->head); 78727a884dcSArnaldo Carvalho de Melo #endif 7884305b541SArnaldo Carvalho de Melo memcpy(data + size, skb_end_pointer(skb), 7894305b541SArnaldo Carvalho de Melo sizeof(struct skb_shared_info)); 7901da177e4SLinus Torvalds 7911da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 7921da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 7931da177e4SLinus Torvalds 7941da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 7951da177e4SLinus Torvalds skb_clone_fraglist(skb); 7961da177e4SLinus Torvalds 7971da177e4SLinus Torvalds skb_release_data(skb); 7981da177e4SLinus Torvalds 7991da177e4SLinus Torvalds off = (data + nhead) - skb->head; 8001da177e4SLinus Torvalds 8011da177e4SLinus Torvalds skb->head = data; 8021da177e4SLinus Torvalds skb->data += off; 8034305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 8044305b541SArnaldo Carvalho de Melo skb->end = size; 80556eb8882SPatrick McHardy off = nhead; 8064305b541SArnaldo Carvalho de Melo #else 8074305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 80856eb8882SPatrick McHardy #endif 80927a884dcSArnaldo Carvalho de Melo /* {transport,network,mac}_header and tail are relative to skb->head */ 81027a884dcSArnaldo Carvalho de Melo skb->tail += off; 811b0e380b1SArnaldo Carvalho de Melo skb->transport_header += off; 812b0e380b1SArnaldo Carvalho de Melo skb->network_header += off; 813b0e380b1SArnaldo Carvalho de Melo skb->mac_header += off; 814172a863fSHerbert Xu skb->csum_start += nhead; 8151da177e4SLinus Torvalds skb->cloned = 0; 816334a8132SPatrick McHardy skb->hdr_len = 0; 8171da177e4SLinus Torvalds skb->nohdr = 0; 8181da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 8191da177e4SLinus Torvalds return 0; 8201da177e4SLinus Torvalds 8211da177e4SLinus Torvalds nodata: 8221da177e4SLinus Torvalds return -ENOMEM; 8231da177e4SLinus Torvalds } 8241da177e4SLinus Torvalds 8251da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 8261da177e4SLinus Torvalds 8271da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 8281da177e4SLinus Torvalds { 8291da177e4SLinus Torvalds struct sk_buff *skb2; 8301da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 8311da177e4SLinus Torvalds 8321da177e4SLinus Torvalds if (delta <= 0) 8331da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 8341da177e4SLinus Torvalds else { 8351da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 8361da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 8371da177e4SLinus Torvalds GFP_ATOMIC)) { 8381da177e4SLinus Torvalds kfree_skb(skb2); 8391da177e4SLinus Torvalds skb2 = NULL; 8401da177e4SLinus Torvalds } 8411da177e4SLinus Torvalds } 8421da177e4SLinus Torvalds return skb2; 8431da177e4SLinus Torvalds } 8441da177e4SLinus Torvalds 8451da177e4SLinus Torvalds 8461da177e4SLinus Torvalds /** 8471da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 8481da177e4SLinus Torvalds * @skb: buffer to copy 8491da177e4SLinus Torvalds * @newheadroom: new free bytes at head 8501da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 8511da177e4SLinus Torvalds * @gfp_mask: allocation priority 8521da177e4SLinus Torvalds * 8531da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 8541da177e4SLinus Torvalds * allocate additional space. 8551da177e4SLinus Torvalds * 8561da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 8571da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 8581da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 8591da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 8601da177e4SLinus Torvalds * 8611da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 8621da177e4SLinus Torvalds * is called from an interrupt. 8631da177e4SLinus Torvalds */ 8641da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 86586a76cafSVictor Fusco int newheadroom, int newtailroom, 866dd0fc66fSAl Viro gfp_t gfp_mask) 8671da177e4SLinus Torvalds { 8681da177e4SLinus Torvalds /* 8691da177e4SLinus Torvalds * Allocate the copy buffer 8701da177e4SLinus Torvalds */ 8711da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 8721da177e4SLinus Torvalds gfp_mask); 873efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 8741da177e4SLinus Torvalds int head_copy_len, head_copy_off; 87552886051SHerbert Xu int off; 8761da177e4SLinus Torvalds 8771da177e4SLinus Torvalds if (!n) 8781da177e4SLinus Torvalds return NULL; 8791da177e4SLinus Torvalds 8801da177e4SLinus Torvalds skb_reserve(n, newheadroom); 8811da177e4SLinus Torvalds 8821da177e4SLinus Torvalds /* Set the tail pointer and length */ 8831da177e4SLinus Torvalds skb_put(n, skb->len); 8841da177e4SLinus Torvalds 885efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 8861da177e4SLinus Torvalds head_copy_off = 0; 8871da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 8881da177e4SLinus Torvalds head_copy_len = newheadroom; 8891da177e4SLinus Torvalds else 8901da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 8911da177e4SLinus Torvalds 8921da177e4SLinus Torvalds /* Copy the linear header and data. */ 8931da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 8941da177e4SLinus Torvalds skb->len + head_copy_len)) 8951da177e4SLinus Torvalds BUG(); 8961da177e4SLinus Torvalds 8971da177e4SLinus Torvalds copy_skb_header(n, skb); 8981da177e4SLinus Torvalds 899efd1e8d5SPatrick McHardy off = newheadroom - oldheadroom; 90052886051SHerbert Xu n->csum_start += off; 90152886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET 902efd1e8d5SPatrick McHardy n->transport_header += off; 903efd1e8d5SPatrick McHardy n->network_header += off; 904efd1e8d5SPatrick McHardy n->mac_header += off; 90552886051SHerbert Xu #endif 906efd1e8d5SPatrick McHardy 9071da177e4SLinus Torvalds return n; 9081da177e4SLinus Torvalds } 9091da177e4SLinus Torvalds 9101da177e4SLinus Torvalds /** 9111da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 9121da177e4SLinus Torvalds * @skb: buffer to pad 9131da177e4SLinus Torvalds * @pad: space to pad 9141da177e4SLinus Torvalds * 9151da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 9161da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 9171da177e4SLinus Torvalds * beyond the buffer end onto the wire. 9181da177e4SLinus Torvalds * 9195b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 9201da177e4SLinus Torvalds */ 9211da177e4SLinus Torvalds 9225b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 9231da177e4SLinus Torvalds { 9245b057c6bSHerbert Xu int err; 9255b057c6bSHerbert Xu int ntail; 9261da177e4SLinus Torvalds 9271da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 9285b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 9291da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 9305b057c6bSHerbert Xu return 0; 9311da177e4SLinus Torvalds } 9321da177e4SLinus Torvalds 9334305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 9345b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 9355b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 9365b057c6bSHerbert Xu if (unlikely(err)) 9375b057c6bSHerbert Xu goto free_skb; 9385b057c6bSHerbert Xu } 9395b057c6bSHerbert Xu 9405b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 9415b057c6bSHerbert Xu * to be audited. 9425b057c6bSHerbert Xu */ 9435b057c6bSHerbert Xu err = skb_linearize(skb); 9445b057c6bSHerbert Xu if (unlikely(err)) 9455b057c6bSHerbert Xu goto free_skb; 9465b057c6bSHerbert Xu 9475b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 9485b057c6bSHerbert Xu return 0; 9495b057c6bSHerbert Xu 9505b057c6bSHerbert Xu free_skb: 9511da177e4SLinus Torvalds kfree_skb(skb); 9525b057c6bSHerbert Xu return err; 9531da177e4SLinus Torvalds } 9541da177e4SLinus Torvalds 9550dde3e16SIlpo Järvinen /** 9560dde3e16SIlpo Järvinen * skb_put - add data to a buffer 9570dde3e16SIlpo Järvinen * @skb: buffer to use 9580dde3e16SIlpo Järvinen * @len: amount of data to add 9590dde3e16SIlpo Järvinen * 9600dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 9610dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 9620dde3e16SIlpo Järvinen * first byte of the extra data is returned. 9630dde3e16SIlpo Järvinen */ 9640dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 9650dde3e16SIlpo Järvinen { 9660dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 9670dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 9680dde3e16SIlpo Järvinen skb->tail += len; 9690dde3e16SIlpo Järvinen skb->len += len; 9700dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 9710dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 9720dde3e16SIlpo Järvinen return tmp; 9730dde3e16SIlpo Järvinen } 9740dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 9750dde3e16SIlpo Järvinen 9766be8ac2fSIlpo Järvinen /** 977c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 978c2aa270aSIlpo Järvinen * @skb: buffer to use 979c2aa270aSIlpo Järvinen * @len: amount of data to add 980c2aa270aSIlpo Järvinen * 981c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 982c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 983c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 984c2aa270aSIlpo Järvinen */ 985c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 986c2aa270aSIlpo Järvinen { 987c2aa270aSIlpo Järvinen skb->data -= len; 988c2aa270aSIlpo Järvinen skb->len += len; 989c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 990c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 991c2aa270aSIlpo Järvinen return skb->data; 992c2aa270aSIlpo Järvinen } 993c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 994c2aa270aSIlpo Järvinen 995c2aa270aSIlpo Järvinen /** 9966be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 9976be8ac2fSIlpo Järvinen * @skb: buffer to use 9986be8ac2fSIlpo Järvinen * @len: amount of data to remove 9996be8ac2fSIlpo Järvinen * 10006be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 10016be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 10026be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 10036be8ac2fSIlpo Järvinen * the old data. 10046be8ac2fSIlpo Järvinen */ 10056be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 10066be8ac2fSIlpo Järvinen { 10076be8ac2fSIlpo Järvinen return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 10086be8ac2fSIlpo Järvinen } 10096be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 10106be8ac2fSIlpo Järvinen 1011419ae74eSIlpo Järvinen /** 1012419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1013419ae74eSIlpo Järvinen * @skb: buffer to alter 1014419ae74eSIlpo Järvinen * @len: new length 1015419ae74eSIlpo Järvinen * 1016419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1017419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1018419ae74eSIlpo Järvinen * The skb must be linear. 1019419ae74eSIlpo Järvinen */ 1020419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1021419ae74eSIlpo Järvinen { 1022419ae74eSIlpo Järvinen if (skb->len > len) 1023419ae74eSIlpo Järvinen __skb_trim(skb, len); 1024419ae74eSIlpo Järvinen } 1025419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1026419ae74eSIlpo Järvinen 10273cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 10281da177e4SLinus Torvalds */ 10291da177e4SLinus Torvalds 10303cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 10311da177e4SLinus Torvalds { 103227b437c8SHerbert Xu struct sk_buff **fragp; 103327b437c8SHerbert Xu struct sk_buff *frag; 10341da177e4SLinus Torvalds int offset = skb_headlen(skb); 10351da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 10361da177e4SLinus Torvalds int i; 103727b437c8SHerbert Xu int err; 103827b437c8SHerbert Xu 103927b437c8SHerbert Xu if (skb_cloned(skb) && 104027b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 104127b437c8SHerbert Xu return err; 10421da177e4SLinus Torvalds 1043f4d26fb3SHerbert Xu i = 0; 1044f4d26fb3SHerbert Xu if (offset >= len) 1045f4d26fb3SHerbert Xu goto drop_pages; 1046f4d26fb3SHerbert Xu 1047f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 10481da177e4SLinus Torvalds int end = offset + skb_shinfo(skb)->frags[i].size; 104927b437c8SHerbert Xu 105027b437c8SHerbert Xu if (end < len) { 10511da177e4SLinus Torvalds offset = end; 105227b437c8SHerbert Xu continue; 10531da177e4SLinus Torvalds } 10541da177e4SLinus Torvalds 105527b437c8SHerbert Xu skb_shinfo(skb)->frags[i++].size = len - offset; 105627b437c8SHerbert Xu 1057f4d26fb3SHerbert Xu drop_pages: 105827b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 105927b437c8SHerbert Xu 106027b437c8SHerbert Xu for (; i < nfrags; i++) 106127b437c8SHerbert Xu put_page(skb_shinfo(skb)->frags[i].page); 106227b437c8SHerbert Xu 106327b437c8SHerbert Xu if (skb_shinfo(skb)->frag_list) 106427b437c8SHerbert Xu skb_drop_fraglist(skb); 1065f4d26fb3SHerbert Xu goto done; 106627b437c8SHerbert Xu } 106727b437c8SHerbert Xu 106827b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 106927b437c8SHerbert Xu fragp = &frag->next) { 107027b437c8SHerbert Xu int end = offset + frag->len; 107127b437c8SHerbert Xu 107227b437c8SHerbert Xu if (skb_shared(frag)) { 107327b437c8SHerbert Xu struct sk_buff *nfrag; 107427b437c8SHerbert Xu 107527b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 107627b437c8SHerbert Xu if (unlikely(!nfrag)) 107727b437c8SHerbert Xu return -ENOMEM; 107827b437c8SHerbert Xu 107927b437c8SHerbert Xu nfrag->next = frag->next; 1080f4d26fb3SHerbert Xu kfree_skb(frag); 108127b437c8SHerbert Xu frag = nfrag; 108227b437c8SHerbert Xu *fragp = frag; 108327b437c8SHerbert Xu } 108427b437c8SHerbert Xu 108527b437c8SHerbert Xu if (end < len) { 108627b437c8SHerbert Xu offset = end; 108727b437c8SHerbert Xu continue; 108827b437c8SHerbert Xu } 108927b437c8SHerbert Xu 109027b437c8SHerbert Xu if (end > len && 109127b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 109227b437c8SHerbert Xu return err; 109327b437c8SHerbert Xu 109427b437c8SHerbert Xu if (frag->next) 109527b437c8SHerbert Xu skb_drop_list(&frag->next); 109627b437c8SHerbert Xu break; 109727b437c8SHerbert Xu } 109827b437c8SHerbert Xu 1099f4d26fb3SHerbert Xu done: 110027b437c8SHerbert Xu if (len > skb_headlen(skb)) { 11011da177e4SLinus Torvalds skb->data_len -= skb->len - len; 11021da177e4SLinus Torvalds skb->len = len; 11031da177e4SLinus Torvalds } else { 11041da177e4SLinus Torvalds skb->len = len; 11051da177e4SLinus Torvalds skb->data_len = 0; 110627a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 11071da177e4SLinus Torvalds } 11081da177e4SLinus Torvalds 11091da177e4SLinus Torvalds return 0; 11101da177e4SLinus Torvalds } 11111da177e4SLinus Torvalds 11121da177e4SLinus Torvalds /** 11131da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 11141da177e4SLinus Torvalds * @skb: buffer to reallocate 11151da177e4SLinus Torvalds * @delta: number of bytes to advance tail 11161da177e4SLinus Torvalds * 11171da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 11181da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 11191da177e4SLinus Torvalds * data from fragmented part. 11201da177e4SLinus Torvalds * 11211da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 11221da177e4SLinus Torvalds * 11231da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 11241da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 11251da177e4SLinus Torvalds * 11261da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11271da177e4SLinus Torvalds * reloaded after call to this function. 11281da177e4SLinus Torvalds */ 11291da177e4SLinus Torvalds 11301da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 11311da177e4SLinus Torvalds * when it is necessary. 11321da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 11331da177e4SLinus Torvalds * 2. It may change skb pointers. 11341da177e4SLinus Torvalds * 11351da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 11361da177e4SLinus Torvalds */ 11371da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 11381da177e4SLinus Torvalds { 11391da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 11401da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 11411da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 11421da177e4SLinus Torvalds */ 11434305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 11441da177e4SLinus Torvalds 11451da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 11461da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 11471da177e4SLinus Torvalds GFP_ATOMIC)) 11481da177e4SLinus Torvalds return NULL; 11491da177e4SLinus Torvalds } 11501da177e4SLinus Torvalds 115127a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 11521da177e4SLinus Torvalds BUG(); 11531da177e4SLinus Torvalds 11541da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 11551da177e4SLinus Torvalds * size of pulled pages. Superb. 11561da177e4SLinus Torvalds */ 11571da177e4SLinus Torvalds if (!skb_shinfo(skb)->frag_list) 11581da177e4SLinus Torvalds goto pull_pages; 11591da177e4SLinus Torvalds 11601da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 11611da177e4SLinus Torvalds eat = delta; 11621da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11631da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size >= eat) 11641da177e4SLinus Torvalds goto pull_pages; 11651da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 11661da177e4SLinus Torvalds } 11671da177e4SLinus Torvalds 11681da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 11691da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 11701da177e4SLinus Torvalds * but taking into account that pulling is expected to 11711da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 11721da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 11731da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 11741da177e4SLinus Torvalds */ 11751da177e4SLinus Torvalds if (eat) { 11761da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 11771da177e4SLinus Torvalds struct sk_buff *clone = NULL; 11781da177e4SLinus Torvalds struct sk_buff *insp = NULL; 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds do { 118109a62660SKris Katterjohn BUG_ON(!list); 11821da177e4SLinus Torvalds 11831da177e4SLinus Torvalds if (list->len <= eat) { 11841da177e4SLinus Torvalds /* Eaten as whole. */ 11851da177e4SLinus Torvalds eat -= list->len; 11861da177e4SLinus Torvalds list = list->next; 11871da177e4SLinus Torvalds insp = list; 11881da177e4SLinus Torvalds } else { 11891da177e4SLinus Torvalds /* Eaten partially. */ 11901da177e4SLinus Torvalds 11911da177e4SLinus Torvalds if (skb_shared(list)) { 11921da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 11931da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 11941da177e4SLinus Torvalds if (!clone) 11951da177e4SLinus Torvalds return NULL; 11961da177e4SLinus Torvalds insp = list->next; 11971da177e4SLinus Torvalds list = clone; 11981da177e4SLinus Torvalds } else { 11991da177e4SLinus Torvalds /* This may be pulled without 12001da177e4SLinus Torvalds * problems. */ 12011da177e4SLinus Torvalds insp = list; 12021da177e4SLinus Torvalds } 12031da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 12041da177e4SLinus Torvalds if (clone) 12051da177e4SLinus Torvalds kfree_skb(clone); 12061da177e4SLinus Torvalds return NULL; 12071da177e4SLinus Torvalds } 12081da177e4SLinus Torvalds break; 12091da177e4SLinus Torvalds } 12101da177e4SLinus Torvalds } while (eat); 12111da177e4SLinus Torvalds 12121da177e4SLinus Torvalds /* Free pulled out fragments. */ 12131da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 12141da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 12151da177e4SLinus Torvalds kfree_skb(list); 12161da177e4SLinus Torvalds } 12171da177e4SLinus Torvalds /* And insert new clone at head. */ 12181da177e4SLinus Torvalds if (clone) { 12191da177e4SLinus Torvalds clone->next = list; 12201da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds } 12231da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 12241da177e4SLinus Torvalds 12251da177e4SLinus Torvalds pull_pages: 12261da177e4SLinus Torvalds eat = delta; 12271da177e4SLinus Torvalds k = 0; 12281da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12291da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size <= eat) { 12301da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 12311da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 12321da177e4SLinus Torvalds } else { 12331da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 12341da177e4SLinus Torvalds if (eat) { 12351da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 12361da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].size -= eat; 12371da177e4SLinus Torvalds eat = 0; 12381da177e4SLinus Torvalds } 12391da177e4SLinus Torvalds k++; 12401da177e4SLinus Torvalds } 12411da177e4SLinus Torvalds } 12421da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 12431da177e4SLinus Torvalds 12441da177e4SLinus Torvalds skb->tail += delta; 12451da177e4SLinus Torvalds skb->data_len -= delta; 12461da177e4SLinus Torvalds 124727a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 12481da177e4SLinus Torvalds } 12491da177e4SLinus Torvalds 12501da177e4SLinus Torvalds /* Copy some data bits from skb to kernel buffer. */ 12511da177e4SLinus Torvalds 12521da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 12531da177e4SLinus Torvalds { 12541da177e4SLinus Torvalds int i, copy; 12551a028e50SDavid S. Miller int start = skb_headlen(skb); 12561da177e4SLinus Torvalds 12571da177e4SLinus Torvalds if (offset > (int)skb->len - len) 12581da177e4SLinus Torvalds goto fault; 12591da177e4SLinus Torvalds 12601da177e4SLinus Torvalds /* Copy header. */ 12611a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 12621da177e4SLinus Torvalds if (copy > len) 12631da177e4SLinus Torvalds copy = len; 1264d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 12651da177e4SLinus Torvalds if ((len -= copy) == 0) 12661da177e4SLinus Torvalds return 0; 12671da177e4SLinus Torvalds offset += copy; 12681da177e4SLinus Torvalds to += copy; 12691da177e4SLinus Torvalds } 12701da177e4SLinus Torvalds 12711da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12721a028e50SDavid S. Miller int end; 12731da177e4SLinus Torvalds 1274547b792cSIlpo Järvinen WARN_ON(start > offset + len); 12751a028e50SDavid S. Miller 12761a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 12771da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 12781da177e4SLinus Torvalds u8 *vaddr; 12791da177e4SLinus Torvalds 12801da177e4SLinus Torvalds if (copy > len) 12811da177e4SLinus Torvalds copy = len; 12821da177e4SLinus Torvalds 12831da177e4SLinus Torvalds vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 12841da177e4SLinus Torvalds memcpy(to, 12851a028e50SDavid S. Miller vaddr + skb_shinfo(skb)->frags[i].page_offset+ 12861a028e50SDavid S. Miller offset - start, copy); 12871da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 12881da177e4SLinus Torvalds 12891da177e4SLinus Torvalds if ((len -= copy) == 0) 12901da177e4SLinus Torvalds return 0; 12911da177e4SLinus Torvalds offset += copy; 12921da177e4SLinus Torvalds to += copy; 12931da177e4SLinus Torvalds } 12941a028e50SDavid S. Miller start = end; 12951da177e4SLinus Torvalds } 12961da177e4SLinus Torvalds 12971da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 12981da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 12991da177e4SLinus Torvalds 13001da177e4SLinus Torvalds for (; list; list = list->next) { 13011a028e50SDavid S. Miller int end; 13021da177e4SLinus Torvalds 1303547b792cSIlpo Järvinen WARN_ON(start > offset + len); 13041a028e50SDavid S. Miller 13051a028e50SDavid S. Miller end = start + list->len; 13061da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13071da177e4SLinus Torvalds if (copy > len) 13081da177e4SLinus Torvalds copy = len; 13091a028e50SDavid S. Miller if (skb_copy_bits(list, offset - start, 13101a028e50SDavid S. Miller to, copy)) 13111da177e4SLinus Torvalds goto fault; 13121da177e4SLinus Torvalds if ((len -= copy) == 0) 13131da177e4SLinus Torvalds return 0; 13141da177e4SLinus Torvalds offset += copy; 13151da177e4SLinus Torvalds to += copy; 13161da177e4SLinus Torvalds } 13171a028e50SDavid S. Miller start = end; 13181da177e4SLinus Torvalds } 13191da177e4SLinus Torvalds } 13201da177e4SLinus Torvalds if (!len) 13211da177e4SLinus Torvalds return 0; 13221da177e4SLinus Torvalds 13231da177e4SLinus Torvalds fault: 13241da177e4SLinus Torvalds return -EFAULT; 13251da177e4SLinus Torvalds } 13261da177e4SLinus Torvalds 13279c55e01cSJens Axboe /* 13289c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 13299c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 13309c55e01cSJens Axboe */ 13319c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 13329c55e01cSJens Axboe { 13338b9d3728SJarek Poplawski put_page(spd->pages[i]); 13348b9d3728SJarek Poplawski } 13359c55e01cSJens Axboe 13368b9d3728SJarek Poplawski static inline struct page *linear_to_page(struct page *page, unsigned int len, 13378b9d3728SJarek Poplawski unsigned int offset) 13388b9d3728SJarek Poplawski { 13398b9d3728SJarek Poplawski struct page *p = alloc_pages(GFP_KERNEL, 0); 13408b9d3728SJarek Poplawski 13418b9d3728SJarek Poplawski if (!p) 13428b9d3728SJarek Poplawski return NULL; 13438b9d3728SJarek Poplawski memcpy(page_address(p) + offset, page_address(page) + offset, len); 13448b9d3728SJarek Poplawski 13458b9d3728SJarek Poplawski return p; 13469c55e01cSJens Axboe } 13479c55e01cSJens Axboe 13489c55e01cSJens Axboe /* 13499c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 13509c55e01cSJens Axboe */ 13519c55e01cSJens Axboe static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page, 13529c55e01cSJens Axboe unsigned int len, unsigned int offset, 13538b9d3728SJarek Poplawski struct sk_buff *skb, int linear) 13549c55e01cSJens Axboe { 13559c55e01cSJens Axboe if (unlikely(spd->nr_pages == PIPE_BUFFERS)) 13569c55e01cSJens Axboe return 1; 13579c55e01cSJens Axboe 13588b9d3728SJarek Poplawski if (linear) { 13598b9d3728SJarek Poplawski page = linear_to_page(page, len, offset); 13608b9d3728SJarek Poplawski if (!page) 13618b9d3728SJarek Poplawski return 1; 13628b9d3728SJarek Poplawski } else 13638b9d3728SJarek Poplawski get_page(page); 13648b9d3728SJarek Poplawski 13659c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 13669c55e01cSJens Axboe spd->partial[spd->nr_pages].len = len; 13679c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 13689c55e01cSJens Axboe spd->nr_pages++; 13698b9d3728SJarek Poplawski 13709c55e01cSJens Axboe return 0; 13719c55e01cSJens Axboe } 13729c55e01cSJens Axboe 13732870c43dSOctavian Purdila static inline void __segment_seek(struct page **page, unsigned int *poff, 13742870c43dSOctavian Purdila unsigned int *plen, unsigned int off) 13752870c43dSOctavian Purdila { 13762870c43dSOctavian Purdila *poff += off; 13772870c43dSOctavian Purdila *page += *poff / PAGE_SIZE; 13782870c43dSOctavian Purdila *poff = *poff % PAGE_SIZE; 13792870c43dSOctavian Purdila *plen -= off; 13802870c43dSOctavian Purdila } 13812870c43dSOctavian Purdila 13822870c43dSOctavian Purdila static inline int __splice_segment(struct page *page, unsigned int poff, 13832870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 13842870c43dSOctavian Purdila unsigned int *len, struct sk_buff *skb, 13858b9d3728SJarek Poplawski struct splice_pipe_desc *spd, int linear) 13869c55e01cSJens Axboe { 13872870c43dSOctavian Purdila if (!*len) 13882870c43dSOctavian Purdila return 1; 13899c55e01cSJens Axboe 13902870c43dSOctavian Purdila /* skip this segment if already processed */ 13912870c43dSOctavian Purdila if (*off >= plen) { 13922870c43dSOctavian Purdila *off -= plen; 13932870c43dSOctavian Purdila return 0; 13942870c43dSOctavian Purdila } 13952870c43dSOctavian Purdila 13962870c43dSOctavian Purdila /* ignore any bits we already processed */ 13972870c43dSOctavian Purdila if (*off) { 13982870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, *off); 13992870c43dSOctavian Purdila *off = 0; 14002870c43dSOctavian Purdila } 14012870c43dSOctavian Purdila 14022870c43dSOctavian Purdila do { 14032870c43dSOctavian Purdila unsigned int flen = min(*len, plen); 14042870c43dSOctavian Purdila 14052870c43dSOctavian Purdila /* the linear region may spread across several pages */ 14062870c43dSOctavian Purdila flen = min_t(unsigned int, flen, PAGE_SIZE - poff); 14072870c43dSOctavian Purdila 14088b9d3728SJarek Poplawski if (spd_fill_page(spd, page, flen, poff, skb, linear)) 14092870c43dSOctavian Purdila return 1; 14102870c43dSOctavian Purdila 14112870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, flen); 14122870c43dSOctavian Purdila *len -= flen; 14132870c43dSOctavian Purdila 14142870c43dSOctavian Purdila } while (*len && plen); 14152870c43dSOctavian Purdila 14162870c43dSOctavian Purdila return 0; 1417db43a282SOctavian Purdila } 14189c55e01cSJens Axboe 14199c55e01cSJens Axboe /* 14202870c43dSOctavian Purdila * Map linear and fragment data from the skb to spd. It reports failure if the 14212870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 14229c55e01cSJens Axboe */ 14237b1c65faSHarvey Harrison static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset, 14242870c43dSOctavian Purdila unsigned int *len, 14252870c43dSOctavian Purdila struct splice_pipe_desc *spd) 14262870c43dSOctavian Purdila { 14272870c43dSOctavian Purdila int seg; 14289c55e01cSJens Axboe 14299c55e01cSJens Axboe /* 14302870c43dSOctavian Purdila * map the linear part 14319c55e01cSJens Axboe */ 14322870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 14332870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 14342870c43dSOctavian Purdila skb_headlen(skb), 14358b9d3728SJarek Poplawski offset, len, skb, spd, 1)) 14362870c43dSOctavian Purdila return 1; 14379c55e01cSJens Axboe 14389c55e01cSJens Axboe /* 14399c55e01cSJens Axboe * then map the fragments 14409c55e01cSJens Axboe */ 14419c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 14429c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 14439c55e01cSJens Axboe 14442870c43dSOctavian Purdila if (__splice_segment(f->page, f->page_offset, f->size, 14458b9d3728SJarek Poplawski offset, len, skb, spd, 0)) 14462870c43dSOctavian Purdila return 1; 14479c55e01cSJens Axboe } 14489c55e01cSJens Axboe 14499c55e01cSJens Axboe return 0; 14509c55e01cSJens Axboe } 14519c55e01cSJens Axboe 14529c55e01cSJens Axboe /* 14539c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 14549c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 14559c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 14569c55e01cSJens Axboe * handle that cleanly. 14579c55e01cSJens Axboe */ 14588b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 14599c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 14609c55e01cSJens Axboe unsigned int flags) 14619c55e01cSJens Axboe { 14629c55e01cSJens Axboe struct partial_page partial[PIPE_BUFFERS]; 14639c55e01cSJens Axboe struct page *pages[PIPE_BUFFERS]; 14649c55e01cSJens Axboe struct splice_pipe_desc spd = { 14659c55e01cSJens Axboe .pages = pages, 14669c55e01cSJens Axboe .partial = partial, 14679c55e01cSJens Axboe .flags = flags, 14689c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 14699c55e01cSJens Axboe .spd_release = sock_spd_release, 14709c55e01cSJens Axboe }; 14719c55e01cSJens Axboe 14729c55e01cSJens Axboe /* 14739c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 14749c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 14759c55e01cSJens Axboe */ 14769c55e01cSJens Axboe if (__skb_splice_bits(skb, &offset, &tlen, &spd)) 14779c55e01cSJens Axboe goto done; 14789c55e01cSJens Axboe else if (!tlen) 14799c55e01cSJens Axboe goto done; 14809c55e01cSJens Axboe 14819c55e01cSJens Axboe /* 14829c55e01cSJens Axboe * now see if we have a frag_list to map 14839c55e01cSJens Axboe */ 14849c55e01cSJens Axboe if (skb_shinfo(skb)->frag_list) { 14859c55e01cSJens Axboe struct sk_buff *list = skb_shinfo(skb)->frag_list; 14869c55e01cSJens Axboe 14879c55e01cSJens Axboe for (; list && tlen; list = list->next) { 14889c55e01cSJens Axboe if (__skb_splice_bits(list, &offset, &tlen, &spd)) 14899c55e01cSJens Axboe break; 14909c55e01cSJens Axboe } 14919c55e01cSJens Axboe } 14929c55e01cSJens Axboe 14939c55e01cSJens Axboe done: 14949c55e01cSJens Axboe if (spd.nr_pages) { 14958b9d3728SJarek Poplawski struct sock *sk = skb->sk; 14969c55e01cSJens Axboe int ret; 14979c55e01cSJens Axboe 14989c55e01cSJens Axboe /* 14999c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 15009c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 15019c55e01cSJens Axboe * here as compared to sendfile(). We enter here 15029c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 15039c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 15049c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 15059c55e01cSJens Axboe * and networking will grab the socket lock. 15069c55e01cSJens Axboe */ 1507293ad604SOctavian Purdila release_sock(sk); 15089c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1509293ad604SOctavian Purdila lock_sock(sk); 15109c55e01cSJens Axboe return ret; 15119c55e01cSJens Axboe } 15129c55e01cSJens Axboe 15139c55e01cSJens Axboe return 0; 15149c55e01cSJens Axboe } 15159c55e01cSJens Axboe 1516357b40a1SHerbert Xu /** 1517357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1518357b40a1SHerbert Xu * @skb: destination buffer 1519357b40a1SHerbert Xu * @offset: offset in destination 1520357b40a1SHerbert Xu * @from: source buffer 1521357b40a1SHerbert Xu * @len: number of bytes to copy 1522357b40a1SHerbert Xu * 1523357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1524357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1525357b40a1SHerbert Xu * traversing fragment lists and such. 1526357b40a1SHerbert Xu */ 1527357b40a1SHerbert Xu 15280c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1529357b40a1SHerbert Xu { 1530357b40a1SHerbert Xu int i, copy; 15311a028e50SDavid S. Miller int start = skb_headlen(skb); 1532357b40a1SHerbert Xu 1533357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1534357b40a1SHerbert Xu goto fault; 1535357b40a1SHerbert Xu 15361a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1537357b40a1SHerbert Xu if (copy > len) 1538357b40a1SHerbert Xu copy = len; 153927d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1540357b40a1SHerbert Xu if ((len -= copy) == 0) 1541357b40a1SHerbert Xu return 0; 1542357b40a1SHerbert Xu offset += copy; 1543357b40a1SHerbert Xu from += copy; 1544357b40a1SHerbert Xu } 1545357b40a1SHerbert Xu 1546357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1547357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 15481a028e50SDavid S. Miller int end; 1549357b40a1SHerbert Xu 1550547b792cSIlpo Järvinen WARN_ON(start > offset + len); 15511a028e50SDavid S. Miller 15521a028e50SDavid S. Miller end = start + frag->size; 1553357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1554357b40a1SHerbert Xu u8 *vaddr; 1555357b40a1SHerbert Xu 1556357b40a1SHerbert Xu if (copy > len) 1557357b40a1SHerbert Xu copy = len; 1558357b40a1SHerbert Xu 1559357b40a1SHerbert Xu vaddr = kmap_skb_frag(frag); 15601a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 15611a028e50SDavid S. Miller from, copy); 1562357b40a1SHerbert Xu kunmap_skb_frag(vaddr); 1563357b40a1SHerbert Xu 1564357b40a1SHerbert Xu if ((len -= copy) == 0) 1565357b40a1SHerbert Xu return 0; 1566357b40a1SHerbert Xu offset += copy; 1567357b40a1SHerbert Xu from += copy; 1568357b40a1SHerbert Xu } 15691a028e50SDavid S. Miller start = end; 1570357b40a1SHerbert Xu } 1571357b40a1SHerbert Xu 1572357b40a1SHerbert Xu if (skb_shinfo(skb)->frag_list) { 1573357b40a1SHerbert Xu struct sk_buff *list = skb_shinfo(skb)->frag_list; 1574357b40a1SHerbert Xu 1575357b40a1SHerbert Xu for (; list; list = list->next) { 15761a028e50SDavid S. Miller int end; 1577357b40a1SHerbert Xu 1578547b792cSIlpo Järvinen WARN_ON(start > offset + len); 15791a028e50SDavid S. Miller 15801a028e50SDavid S. Miller end = start + list->len; 1581357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1582357b40a1SHerbert Xu if (copy > len) 1583357b40a1SHerbert Xu copy = len; 15841a028e50SDavid S. Miller if (skb_store_bits(list, offset - start, 15851a028e50SDavid S. Miller from, copy)) 1586357b40a1SHerbert Xu goto fault; 1587357b40a1SHerbert Xu if ((len -= copy) == 0) 1588357b40a1SHerbert Xu return 0; 1589357b40a1SHerbert Xu offset += copy; 1590357b40a1SHerbert Xu from += copy; 1591357b40a1SHerbert Xu } 15921a028e50SDavid S. Miller start = end; 1593357b40a1SHerbert Xu } 1594357b40a1SHerbert Xu } 1595357b40a1SHerbert Xu if (!len) 1596357b40a1SHerbert Xu return 0; 1597357b40a1SHerbert Xu 1598357b40a1SHerbert Xu fault: 1599357b40a1SHerbert Xu return -EFAULT; 1600357b40a1SHerbert Xu } 1601357b40a1SHerbert Xu 1602357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1603357b40a1SHerbert Xu 16041da177e4SLinus Torvalds /* Checksum skb data. */ 16051da177e4SLinus Torvalds 16062bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 16072bbbc868SAl Viro int len, __wsum csum) 16081da177e4SLinus Torvalds { 16091a028e50SDavid S. Miller int start = skb_headlen(skb); 16101a028e50SDavid S. Miller int i, copy = start - offset; 16111da177e4SLinus Torvalds int pos = 0; 16121da177e4SLinus Torvalds 16131da177e4SLinus Torvalds /* Checksum header. */ 16141da177e4SLinus Torvalds if (copy > 0) { 16151da177e4SLinus Torvalds if (copy > len) 16161da177e4SLinus Torvalds copy = len; 16171da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 16181da177e4SLinus Torvalds if ((len -= copy) == 0) 16191da177e4SLinus Torvalds return csum; 16201da177e4SLinus Torvalds offset += copy; 16211da177e4SLinus Torvalds pos = copy; 16221da177e4SLinus Torvalds } 16231da177e4SLinus Torvalds 16241da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16251a028e50SDavid S. Miller int end; 16261da177e4SLinus Torvalds 1627547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16281a028e50SDavid S. Miller 16291a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 16301da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 163144bb9363SAl Viro __wsum csum2; 16321da177e4SLinus Torvalds u8 *vaddr; 16331da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 16341da177e4SLinus Torvalds 16351da177e4SLinus Torvalds if (copy > len) 16361da177e4SLinus Torvalds copy = len; 16371da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 16381a028e50SDavid S. Miller csum2 = csum_partial(vaddr + frag->page_offset + 16391a028e50SDavid S. Miller offset - start, copy, 0); 16401da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 16411da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 16421da177e4SLinus Torvalds if (!(len -= copy)) 16431da177e4SLinus Torvalds return csum; 16441da177e4SLinus Torvalds offset += copy; 16451da177e4SLinus Torvalds pos += copy; 16461da177e4SLinus Torvalds } 16471a028e50SDavid S. Miller start = end; 16481da177e4SLinus Torvalds } 16491da177e4SLinus Torvalds 16501da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 16511da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 16521da177e4SLinus Torvalds 16531da177e4SLinus Torvalds for (; list; list = list->next) { 16541a028e50SDavid S. Miller int end; 16551da177e4SLinus Torvalds 1656547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16571a028e50SDavid S. Miller 16581a028e50SDavid S. Miller end = start + list->len; 16591da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 16605f92a738SAl Viro __wsum csum2; 16611da177e4SLinus Torvalds if (copy > len) 16621da177e4SLinus Torvalds copy = len; 16631a028e50SDavid S. Miller csum2 = skb_checksum(list, offset - start, 16641a028e50SDavid S. Miller copy, 0); 16651da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 16661da177e4SLinus Torvalds if ((len -= copy) == 0) 16671da177e4SLinus Torvalds return csum; 16681da177e4SLinus Torvalds offset += copy; 16691da177e4SLinus Torvalds pos += copy; 16701da177e4SLinus Torvalds } 16711a028e50SDavid S. Miller start = end; 16721da177e4SLinus Torvalds } 16731da177e4SLinus Torvalds } 167409a62660SKris Katterjohn BUG_ON(len); 16751da177e4SLinus Torvalds 16761da177e4SLinus Torvalds return csum; 16771da177e4SLinus Torvalds } 16781da177e4SLinus Torvalds 16791da177e4SLinus Torvalds /* Both of above in one bottle. */ 16801da177e4SLinus Torvalds 168181d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 168281d77662SAl Viro u8 *to, int len, __wsum csum) 16831da177e4SLinus Torvalds { 16841a028e50SDavid S. Miller int start = skb_headlen(skb); 16851a028e50SDavid S. Miller int i, copy = start - offset; 16861da177e4SLinus Torvalds int pos = 0; 16871da177e4SLinus Torvalds 16881da177e4SLinus Torvalds /* Copy header. */ 16891da177e4SLinus Torvalds if (copy > 0) { 16901da177e4SLinus Torvalds if (copy > len) 16911da177e4SLinus Torvalds copy = len; 16921da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 16931da177e4SLinus Torvalds copy, csum); 16941da177e4SLinus Torvalds if ((len -= copy) == 0) 16951da177e4SLinus Torvalds return csum; 16961da177e4SLinus Torvalds offset += copy; 16971da177e4SLinus Torvalds to += copy; 16981da177e4SLinus Torvalds pos = copy; 16991da177e4SLinus Torvalds } 17001da177e4SLinus Torvalds 17011da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17021a028e50SDavid S. Miller int end; 17031da177e4SLinus Torvalds 1704547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17051a028e50SDavid S. Miller 17061a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 17071da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17085084205fSAl Viro __wsum csum2; 17091da177e4SLinus Torvalds u8 *vaddr; 17101da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 17111da177e4SLinus Torvalds 17121da177e4SLinus Torvalds if (copy > len) 17131da177e4SLinus Torvalds copy = len; 17141da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 17151da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 17161a028e50SDavid S. Miller frag->page_offset + 17171a028e50SDavid S. Miller offset - start, to, 17181a028e50SDavid S. Miller copy, 0); 17191da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 17201da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17211da177e4SLinus Torvalds if (!(len -= copy)) 17221da177e4SLinus Torvalds return csum; 17231da177e4SLinus Torvalds offset += copy; 17241da177e4SLinus Torvalds to += copy; 17251da177e4SLinus Torvalds pos += copy; 17261da177e4SLinus Torvalds } 17271a028e50SDavid S. Miller start = end; 17281da177e4SLinus Torvalds } 17291da177e4SLinus Torvalds 17301da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 17311da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 17321da177e4SLinus Torvalds 17331da177e4SLinus Torvalds for (; list; list = list->next) { 173481d77662SAl Viro __wsum csum2; 17351a028e50SDavid S. Miller int end; 17361da177e4SLinus Torvalds 1737547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17381a028e50SDavid S. Miller 17391a028e50SDavid S. Miller end = start + list->len; 17401da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17411da177e4SLinus Torvalds if (copy > len) 17421da177e4SLinus Torvalds copy = len; 17431a028e50SDavid S. Miller csum2 = skb_copy_and_csum_bits(list, 17441a028e50SDavid S. Miller offset - start, 17451da177e4SLinus Torvalds to, copy, 0); 17461da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17471da177e4SLinus Torvalds if ((len -= copy) == 0) 17481da177e4SLinus Torvalds return csum; 17491da177e4SLinus Torvalds offset += copy; 17501da177e4SLinus Torvalds to += copy; 17511da177e4SLinus Torvalds pos += copy; 17521da177e4SLinus Torvalds } 17531a028e50SDavid S. Miller start = end; 17541da177e4SLinus Torvalds } 17551da177e4SLinus Torvalds } 175609a62660SKris Katterjohn BUG_ON(len); 17571da177e4SLinus Torvalds return csum; 17581da177e4SLinus Torvalds } 17591da177e4SLinus Torvalds 17601da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 17611da177e4SLinus Torvalds { 1762d3bc23e7SAl Viro __wsum csum; 17631da177e4SLinus Torvalds long csstart; 17641da177e4SLinus Torvalds 176584fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 1766663ead3bSHerbert Xu csstart = skb->csum_start - skb_headroom(skb); 17671da177e4SLinus Torvalds else 17681da177e4SLinus Torvalds csstart = skb_headlen(skb); 17691da177e4SLinus Torvalds 177009a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 17711da177e4SLinus Torvalds 1772d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds csum = 0; 17751da177e4SLinus Torvalds if (csstart != skb->len) 17761da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 17771da177e4SLinus Torvalds skb->len - csstart, 0); 17781da177e4SLinus Torvalds 177984fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 1780ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 17811da177e4SLinus Torvalds 1782d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 17831da177e4SLinus Torvalds } 17841da177e4SLinus Torvalds } 17851da177e4SLinus Torvalds 17861da177e4SLinus Torvalds /** 17871da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 17881da177e4SLinus Torvalds * @list: list to dequeue from 17891da177e4SLinus Torvalds * 17901da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 17911da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 17921da177e4SLinus Torvalds * returned or %NULL if the list is empty. 17931da177e4SLinus Torvalds */ 17941da177e4SLinus Torvalds 17951da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 17961da177e4SLinus Torvalds { 17971da177e4SLinus Torvalds unsigned long flags; 17981da177e4SLinus Torvalds struct sk_buff *result; 17991da177e4SLinus Torvalds 18001da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18011da177e4SLinus Torvalds result = __skb_dequeue(list); 18021da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18031da177e4SLinus Torvalds return result; 18041da177e4SLinus Torvalds } 18051da177e4SLinus Torvalds 18061da177e4SLinus Torvalds /** 18071da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 18081da177e4SLinus Torvalds * @list: list to dequeue from 18091da177e4SLinus Torvalds * 18101da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 18111da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 18121da177e4SLinus Torvalds * returned or %NULL if the list is empty. 18131da177e4SLinus Torvalds */ 18141da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 18151da177e4SLinus Torvalds { 18161da177e4SLinus Torvalds unsigned long flags; 18171da177e4SLinus Torvalds struct sk_buff *result; 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18201da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 18211da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18221da177e4SLinus Torvalds return result; 18231da177e4SLinus Torvalds } 18241da177e4SLinus Torvalds 18251da177e4SLinus Torvalds /** 18261da177e4SLinus Torvalds * skb_queue_purge - empty a list 18271da177e4SLinus Torvalds * @list: list to empty 18281da177e4SLinus Torvalds * 18291da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 18301da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 18311da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 18321da177e4SLinus Torvalds */ 18331da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 18341da177e4SLinus Torvalds { 18351da177e4SLinus Torvalds struct sk_buff *skb; 18361da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 18371da177e4SLinus Torvalds kfree_skb(skb); 18381da177e4SLinus Torvalds } 18391da177e4SLinus Torvalds 18401da177e4SLinus Torvalds /** 18411da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 18421da177e4SLinus Torvalds * @list: list to use 18431da177e4SLinus Torvalds * @newsk: buffer to queue 18441da177e4SLinus Torvalds * 18451da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 18461da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 18471da177e4SLinus Torvalds * safely. 18481da177e4SLinus Torvalds * 18491da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 18501da177e4SLinus Torvalds */ 18511da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 18521da177e4SLinus Torvalds { 18531da177e4SLinus Torvalds unsigned long flags; 18541da177e4SLinus Torvalds 18551da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18561da177e4SLinus Torvalds __skb_queue_head(list, newsk); 18571da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18581da177e4SLinus Torvalds } 18591da177e4SLinus Torvalds 18601da177e4SLinus Torvalds /** 18611da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 18621da177e4SLinus Torvalds * @list: list to use 18631da177e4SLinus Torvalds * @newsk: buffer to queue 18641da177e4SLinus Torvalds * 18651da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 18661da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 18671da177e4SLinus Torvalds * safely. 18681da177e4SLinus Torvalds * 18691da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 18701da177e4SLinus Torvalds */ 18711da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 18721da177e4SLinus Torvalds { 18731da177e4SLinus Torvalds unsigned long flags; 18741da177e4SLinus Torvalds 18751da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18761da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 18771da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18781da177e4SLinus Torvalds } 18798728b834SDavid S. Miller 18801da177e4SLinus Torvalds /** 18811da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 18821da177e4SLinus Torvalds * @skb: buffer to remove 18838728b834SDavid S. Miller * @list: list to use 18841da177e4SLinus Torvalds * 18858728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 18868728b834SDavid S. Miller * function is atomic with respect to other list locked calls 18871da177e4SLinus Torvalds * 18888728b834SDavid S. Miller * You must know what list the SKB is on. 18891da177e4SLinus Torvalds */ 18908728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 18911da177e4SLinus Torvalds { 18921da177e4SLinus Torvalds unsigned long flags; 18931da177e4SLinus Torvalds 18941da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18958728b834SDavid S. Miller __skb_unlink(skb, list); 18961da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18971da177e4SLinus Torvalds } 18981da177e4SLinus Torvalds 18991da177e4SLinus Torvalds /** 19001da177e4SLinus Torvalds * skb_append - append a buffer 19011da177e4SLinus Torvalds * @old: buffer to insert after 19021da177e4SLinus Torvalds * @newsk: buffer to insert 19038728b834SDavid S. Miller * @list: list to use 19041da177e4SLinus Torvalds * 19051da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 19061da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 19071da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19081da177e4SLinus Torvalds */ 19098728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 19101da177e4SLinus Torvalds { 19111da177e4SLinus Torvalds unsigned long flags; 19121da177e4SLinus Torvalds 19138728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 19147de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 19158728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 19161da177e4SLinus Torvalds } 19171da177e4SLinus Torvalds 19181da177e4SLinus Torvalds 19191da177e4SLinus Torvalds /** 19201da177e4SLinus Torvalds * skb_insert - insert a buffer 19211da177e4SLinus Torvalds * @old: buffer to insert before 19221da177e4SLinus Torvalds * @newsk: buffer to insert 19238728b834SDavid S. Miller * @list: list to use 19241da177e4SLinus Torvalds * 19258728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 19268728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 19278728b834SDavid S. Miller * calls. 19288728b834SDavid S. Miller * 19291da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19301da177e4SLinus Torvalds */ 19318728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 19321da177e4SLinus Torvalds { 19331da177e4SLinus Torvalds unsigned long flags; 19341da177e4SLinus Torvalds 19358728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 19368728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 19378728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 19381da177e4SLinus Torvalds } 19391da177e4SLinus Torvalds 19401da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 19411da177e4SLinus Torvalds struct sk_buff* skb1, 19421da177e4SLinus Torvalds const u32 len, const int pos) 19431da177e4SLinus Torvalds { 19441da177e4SLinus Torvalds int i; 19451da177e4SLinus Torvalds 1946d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 1947d626f62bSArnaldo Carvalho de Melo pos - len); 19481da177e4SLinus Torvalds /* And move data appendix as is. */ 19491da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 19501da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 19511da177e4SLinus Torvalds 19521da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 19531da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 19541da177e4SLinus Torvalds skb1->data_len = skb->data_len; 19551da177e4SLinus Torvalds skb1->len += skb1->data_len; 19561da177e4SLinus Torvalds skb->data_len = 0; 19571da177e4SLinus Torvalds skb->len = len; 195827a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 19591da177e4SLinus Torvalds } 19601da177e4SLinus Torvalds 19611da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 19621da177e4SLinus Torvalds struct sk_buff* skb1, 19631da177e4SLinus Torvalds const u32 len, int pos) 19641da177e4SLinus Torvalds { 19651da177e4SLinus Torvalds int i, k = 0; 19661da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 19671da177e4SLinus Torvalds 19681da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 19691da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 19701da177e4SLinus Torvalds skb->len = len; 19711da177e4SLinus Torvalds skb->data_len = len - pos; 19721da177e4SLinus Torvalds 19731da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 19741da177e4SLinus Torvalds int size = skb_shinfo(skb)->frags[i].size; 19751da177e4SLinus Torvalds 19761da177e4SLinus Torvalds if (pos + size > len) { 19771da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 19781da177e4SLinus Torvalds 19791da177e4SLinus Torvalds if (pos < len) { 19801da177e4SLinus Torvalds /* Split frag. 19811da177e4SLinus Torvalds * We have two variants in this case: 19821da177e4SLinus Torvalds * 1. Move all the frag to the second 19831da177e4SLinus Torvalds * part, if it is possible. F.e. 19841da177e4SLinus Torvalds * this approach is mandatory for TUX, 19851da177e4SLinus Torvalds * where splitting is expensive. 19861da177e4SLinus Torvalds * 2. Split is accurately. We make this. 19871da177e4SLinus Torvalds */ 19881da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 19891da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 19901da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].size -= len - pos; 19911da177e4SLinus Torvalds skb_shinfo(skb)->frags[i].size = len - pos; 19921da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 19931da177e4SLinus Torvalds } 19941da177e4SLinus Torvalds k++; 19951da177e4SLinus Torvalds } else 19961da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 19971da177e4SLinus Torvalds pos += size; 19981da177e4SLinus Torvalds } 19991da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 20001da177e4SLinus Torvalds } 20011da177e4SLinus Torvalds 20021da177e4SLinus Torvalds /** 20031da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 20041da177e4SLinus Torvalds * @skb: the buffer to split 20051da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 20061da177e4SLinus Torvalds * @len: new length for skb 20071da177e4SLinus Torvalds */ 20081da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 20091da177e4SLinus Torvalds { 20101da177e4SLinus Torvalds int pos = skb_headlen(skb); 20111da177e4SLinus Torvalds 20121da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 20131da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 20141da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 20151da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 20161da177e4SLinus Torvalds } 20171da177e4SLinus Torvalds 20189f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 20199f782db3SIlpo Järvinen * 20209f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 20219f782db3SIlpo Järvinen */ 2022832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2023832d11c5SIlpo Järvinen { 20240ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2025832d11c5SIlpo Järvinen } 2026832d11c5SIlpo Järvinen 2027832d11c5SIlpo Järvinen /** 2028832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2029832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2030832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2031832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2032832d11c5SIlpo Järvinen * 2033832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 2034832d11c5SIlpo Järvinen * the length of the skb, from tgt to skb. Returns number bytes shifted. 2035832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2036832d11c5SIlpo Järvinen * 2037832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2038832d11c5SIlpo Järvinen * 2039832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2040832d11c5SIlpo Järvinen * to have non-paged data as well. 2041832d11c5SIlpo Järvinen * 2042832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2043832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2044832d11c5SIlpo Järvinen */ 2045832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2046832d11c5SIlpo Järvinen { 2047832d11c5SIlpo Järvinen int from, to, merge, todo; 2048832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2049832d11c5SIlpo Järvinen 2050832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2051832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2052832d11c5SIlpo Järvinen 2053832d11c5SIlpo Järvinen todo = shiftlen; 2054832d11c5SIlpo Järvinen from = 0; 2055832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2056832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2057832d11c5SIlpo Järvinen 2058832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2059832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2060832d11c5SIlpo Järvinen */ 2061832d11c5SIlpo Järvinen if (!to || 2062832d11c5SIlpo Järvinen !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) { 2063832d11c5SIlpo Järvinen merge = -1; 2064832d11c5SIlpo Järvinen } else { 2065832d11c5SIlpo Järvinen merge = to - 1; 2066832d11c5SIlpo Järvinen 2067832d11c5SIlpo Järvinen todo -= fragfrom->size; 2068832d11c5SIlpo Järvinen if (todo < 0) { 2069832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2070832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2071832d11c5SIlpo Järvinen return 0; 2072832d11c5SIlpo Järvinen 20739f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 20749f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2075832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2076832d11c5SIlpo Järvinen 2077832d11c5SIlpo Järvinen fragto->size += shiftlen; 2078832d11c5SIlpo Järvinen fragfrom->size -= shiftlen; 2079832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2080832d11c5SIlpo Järvinen 2081832d11c5SIlpo Järvinen goto onlymerged; 2082832d11c5SIlpo Järvinen } 2083832d11c5SIlpo Järvinen 2084832d11c5SIlpo Järvinen from++; 2085832d11c5SIlpo Järvinen } 2086832d11c5SIlpo Järvinen 2087832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2088832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2089832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2090832d11c5SIlpo Järvinen return 0; 2091832d11c5SIlpo Järvinen 2092832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2093832d11c5SIlpo Järvinen return 0; 2094832d11c5SIlpo Järvinen 2095832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2096832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2097832d11c5SIlpo Järvinen return 0; 2098832d11c5SIlpo Järvinen 2099832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2100832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2101832d11c5SIlpo Järvinen 2102832d11c5SIlpo Järvinen if (todo >= fragfrom->size) { 2103832d11c5SIlpo Järvinen *fragto = *fragfrom; 2104832d11c5SIlpo Järvinen todo -= fragfrom->size; 2105832d11c5SIlpo Järvinen from++; 2106832d11c5SIlpo Järvinen to++; 2107832d11c5SIlpo Järvinen 2108832d11c5SIlpo Järvinen } else { 2109832d11c5SIlpo Järvinen get_page(fragfrom->page); 2110832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2111832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 2112832d11c5SIlpo Järvinen fragto->size = todo; 2113832d11c5SIlpo Järvinen 2114832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 2115832d11c5SIlpo Järvinen fragfrom->size -= todo; 2116832d11c5SIlpo Järvinen todo = 0; 2117832d11c5SIlpo Järvinen 2118832d11c5SIlpo Järvinen to++; 2119832d11c5SIlpo Järvinen break; 2120832d11c5SIlpo Järvinen } 2121832d11c5SIlpo Järvinen } 2122832d11c5SIlpo Järvinen 2123832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2124832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2125832d11c5SIlpo Järvinen 2126832d11c5SIlpo Järvinen if (merge >= 0) { 2127832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2128832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2129832d11c5SIlpo Järvinen 2130832d11c5SIlpo Järvinen fragto->size += fragfrom->size; 2131832d11c5SIlpo Järvinen put_page(fragfrom->page); 2132832d11c5SIlpo Järvinen } 2133832d11c5SIlpo Järvinen 2134832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2135832d11c5SIlpo Järvinen to = 0; 2136832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2137832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2138832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2139832d11c5SIlpo Järvinen 2140832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2141832d11c5SIlpo Järvinen 2142832d11c5SIlpo Järvinen onlymerged: 2143832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2144832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2145832d11c5SIlpo Järvinen */ 2146832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2147832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2148832d11c5SIlpo Järvinen 2149832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2150832d11c5SIlpo Järvinen skb->len -= shiftlen; 2151832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2152832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2153832d11c5SIlpo Järvinen tgt->len += shiftlen; 2154832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2155832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2156832d11c5SIlpo Järvinen 2157832d11c5SIlpo Järvinen return shiftlen; 2158832d11c5SIlpo Järvinen } 2159832d11c5SIlpo Järvinen 2160677e90edSThomas Graf /** 2161677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2162677e90edSThomas Graf * @skb: the buffer to read 2163677e90edSThomas Graf * @from: lower offset of data to be read 2164677e90edSThomas Graf * @to: upper offset of data to be read 2165677e90edSThomas Graf * @st: state variable 2166677e90edSThomas Graf * 2167677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2168677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2169677e90edSThomas Graf */ 2170677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2171677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2172677e90edSThomas Graf { 2173677e90edSThomas Graf st->lower_offset = from; 2174677e90edSThomas Graf st->upper_offset = to; 2175677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2176677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2177677e90edSThomas Graf st->frag_data = NULL; 2178677e90edSThomas Graf } 2179677e90edSThomas Graf 2180677e90edSThomas Graf /** 2181677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2182677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2183677e90edSThomas Graf * @data: destination pointer for data to be returned 2184677e90edSThomas Graf * @st: state variable 2185677e90edSThomas Graf * 2186677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 2187677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2188677e90edSThomas Graf * the head of the data block to &data and returns the length 2189677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2190677e90edSThomas Graf * offset has been reached. 2191677e90edSThomas Graf * 2192677e90edSThomas Graf * The caller is not required to consume all of the data 2193677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 2194677e90edSThomas Graf * of bytes already consumed and the next call to 2195677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2196677e90edSThomas Graf * 2197bc2cda1eSRandy Dunlap * Note 1: The size of each block of data returned can be arbitary, 2198677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2199677e90edSThomas Graf * reads of potentially non linear data. 2200677e90edSThomas Graf * 2201bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2202677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2203677e90edSThomas Graf * a stack for this purpose. 2204677e90edSThomas Graf */ 2205677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2206677e90edSThomas Graf struct skb_seq_state *st) 2207677e90edSThomas Graf { 2208677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2209677e90edSThomas Graf skb_frag_t *frag; 2210677e90edSThomas Graf 2211677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 2212677e90edSThomas Graf return 0; 2213677e90edSThomas Graf 2214677e90edSThomas Graf next_skb: 2215677e90edSThomas Graf block_limit = skb_headlen(st->cur_skb); 2216677e90edSThomas Graf 2217677e90edSThomas Graf if (abs_offset < block_limit) { 2218677e90edSThomas Graf *data = st->cur_skb->data + abs_offset; 2219677e90edSThomas Graf return block_limit - abs_offset; 2220677e90edSThomas Graf } 2221677e90edSThomas Graf 2222677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2223677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2224677e90edSThomas Graf 2225677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2226677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 2227677e90edSThomas Graf block_limit = frag->size + st->stepped_offset; 2228677e90edSThomas Graf 2229677e90edSThomas Graf if (abs_offset < block_limit) { 2230677e90edSThomas Graf if (!st->frag_data) 2231677e90edSThomas Graf st->frag_data = kmap_skb_frag(frag); 2232677e90edSThomas Graf 2233677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2234677e90edSThomas Graf (abs_offset - st->stepped_offset); 2235677e90edSThomas Graf 2236677e90edSThomas Graf return block_limit - abs_offset; 2237677e90edSThomas Graf } 2238677e90edSThomas Graf 2239677e90edSThomas Graf if (st->frag_data) { 2240677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2241677e90edSThomas Graf st->frag_data = NULL; 2242677e90edSThomas Graf } 2243677e90edSThomas Graf 2244677e90edSThomas Graf st->frag_idx++; 2245677e90edSThomas Graf st->stepped_offset += frag->size; 2246677e90edSThomas Graf } 2247677e90edSThomas Graf 22485b5a60daSOlaf Kirch if (st->frag_data) { 22495b5a60daSOlaf Kirch kunmap_skb_frag(st->frag_data); 22505b5a60daSOlaf Kirch st->frag_data = NULL; 22515b5a60daSOlaf Kirch } 22525b5a60daSOlaf Kirch 2253677e90edSThomas Graf if (st->cur_skb->next) { 2254677e90edSThomas Graf st->cur_skb = st->cur_skb->next; 2255677e90edSThomas Graf st->frag_idx = 0; 2256677e90edSThomas Graf goto next_skb; 2257677e90edSThomas Graf } else if (st->root_skb == st->cur_skb && 2258677e90edSThomas Graf skb_shinfo(st->root_skb)->frag_list) { 2259677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 2260677e90edSThomas Graf goto next_skb; 2261677e90edSThomas Graf } 2262677e90edSThomas Graf 2263677e90edSThomas Graf return 0; 2264677e90edSThomas Graf } 2265677e90edSThomas Graf 2266677e90edSThomas Graf /** 2267677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2268677e90edSThomas Graf * @st: state variable 2269677e90edSThomas Graf * 2270677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2271677e90edSThomas Graf * returned 0. 2272677e90edSThomas Graf */ 2273677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2274677e90edSThomas Graf { 2275677e90edSThomas Graf if (st->frag_data) 2276677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2277677e90edSThomas Graf } 2278677e90edSThomas Graf 22793fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 22803fc7e8a6SThomas Graf 22813fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 22823fc7e8a6SThomas Graf struct ts_config *conf, 22833fc7e8a6SThomas Graf struct ts_state *state) 22843fc7e8a6SThomas Graf { 22853fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 22863fc7e8a6SThomas Graf } 22873fc7e8a6SThomas Graf 22883fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 22893fc7e8a6SThomas Graf { 22903fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 22913fc7e8a6SThomas Graf } 22923fc7e8a6SThomas Graf 22933fc7e8a6SThomas Graf /** 22943fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 22953fc7e8a6SThomas Graf * @skb: the buffer to look in 22963fc7e8a6SThomas Graf * @from: search offset 22973fc7e8a6SThomas Graf * @to: search limit 22983fc7e8a6SThomas Graf * @config: textsearch configuration 22993fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 23003fc7e8a6SThomas Graf * 23013fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 23023fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 23033fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 23043fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 23053fc7e8a6SThomas Graf */ 23063fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 23073fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 23083fc7e8a6SThomas Graf struct ts_state *state) 23093fc7e8a6SThomas Graf { 2310f72b948dSPhil Oester unsigned int ret; 2311f72b948dSPhil Oester 23123fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 23133fc7e8a6SThomas Graf config->finish = skb_ts_finish; 23143fc7e8a6SThomas Graf 23153fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 23163fc7e8a6SThomas Graf 2317f72b948dSPhil Oester ret = textsearch_find(config, state); 2318f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 23193fc7e8a6SThomas Graf } 23203fc7e8a6SThomas Graf 2321e89e9cf5SAnanda Raju /** 2322e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 2323e89e9cf5SAnanda Raju * @sk: sock structure 2324e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2325e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2326e89e9cf5SAnanda Raju * @from: pointer to user message iov 2327e89e9cf5SAnanda Raju * @length: length of the iov message 2328e89e9cf5SAnanda Raju * 2329e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2330e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2331e89e9cf5SAnanda Raju */ 2332e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2333dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2334e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2335e89e9cf5SAnanda Raju void *from, int length) 2336e89e9cf5SAnanda Raju { 2337e89e9cf5SAnanda Raju int frg_cnt = 0; 2338e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 2339e89e9cf5SAnanda Raju struct page *page = NULL; 2340e89e9cf5SAnanda Raju int copy, left; 2341e89e9cf5SAnanda Raju int offset = 0; 2342e89e9cf5SAnanda Raju int ret; 2343e89e9cf5SAnanda Raju 2344e89e9cf5SAnanda Raju do { 2345e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2346e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2347e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2348e89e9cf5SAnanda Raju return -EFAULT; 2349e89e9cf5SAnanda Raju 2350e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 2351e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 2352e89e9cf5SAnanda Raju 2353e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 2354e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 2355e89e9cf5SAnanda Raju */ 2356e89e9cf5SAnanda Raju if (page == NULL) 2357e89e9cf5SAnanda Raju return -ENOMEM; 2358e89e9cf5SAnanda Raju 2359e89e9cf5SAnanda Raju /* initialize the next frag */ 2360e89e9cf5SAnanda Raju sk->sk_sndmsg_page = page; 2361e89e9cf5SAnanda Raju sk->sk_sndmsg_off = 0; 2362e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2363e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 2364e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2365e89e9cf5SAnanda Raju 2366e89e9cf5SAnanda Raju /* get the new initialized frag */ 2367e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2368e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2369e89e9cf5SAnanda Raju 2370e89e9cf5SAnanda Raju /* copy the user data to page */ 2371e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 2372e89e9cf5SAnanda Raju copy = (length > left)? left : length; 2373e89e9cf5SAnanda Raju 2374e89e9cf5SAnanda Raju ret = getfrag(from, (page_address(frag->page) + 2375e89e9cf5SAnanda Raju frag->page_offset + frag->size), 2376e89e9cf5SAnanda Raju offset, copy, 0, skb); 2377e89e9cf5SAnanda Raju if (ret < 0) 2378e89e9cf5SAnanda Raju return -EFAULT; 2379e89e9cf5SAnanda Raju 2380e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2381e89e9cf5SAnanda Raju sk->sk_sndmsg_off += copy; 2382e89e9cf5SAnanda Raju frag->size += copy; 2383e89e9cf5SAnanda Raju skb->len += copy; 2384e89e9cf5SAnanda Raju skb->data_len += copy; 2385e89e9cf5SAnanda Raju offset += copy; 2386e89e9cf5SAnanda Raju length -= copy; 2387e89e9cf5SAnanda Raju 2388e89e9cf5SAnanda Raju } while (length > 0); 2389e89e9cf5SAnanda Raju 2390e89e9cf5SAnanda Raju return 0; 2391e89e9cf5SAnanda Raju } 2392e89e9cf5SAnanda Raju 2393cbb042f9SHerbert Xu /** 2394cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2395cbb042f9SHerbert Xu * @skb: buffer to update 2396cbb042f9SHerbert Xu * @len: length of data pulled 2397cbb042f9SHerbert Xu * 2398cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2399fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 240084fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 240184fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 240284fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2403cbb042f9SHerbert Xu */ 2404cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2405cbb042f9SHerbert Xu { 2406cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2407cbb042f9SHerbert Xu skb->len -= len; 2408cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2409cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2410cbb042f9SHerbert Xu return skb->data += len; 2411cbb042f9SHerbert Xu } 2412cbb042f9SHerbert Xu 2413f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2414f94691acSArnaldo Carvalho de Melo 2415f4c50d99SHerbert Xu /** 2416f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2417f4c50d99SHerbert Xu * @skb: buffer to segment 2418576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2419f4c50d99SHerbert Xu * 2420f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 24214c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 24224c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2423f4c50d99SHerbert Xu */ 2424576a30ebSHerbert Xu struct sk_buff *skb_segment(struct sk_buff *skb, int features) 2425f4c50d99SHerbert Xu { 2426f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2427f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 242889319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2429f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 243098e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2431f4c50d99SHerbert Xu unsigned int offset = doffset; 2432f4c50d99SHerbert Xu unsigned int headroom; 2433f4c50d99SHerbert Xu unsigned int len; 2434576a30ebSHerbert Xu int sg = features & NETIF_F_SG; 2435f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2436f4c50d99SHerbert Xu int err = -ENOMEM; 2437f4c50d99SHerbert Xu int i = 0; 2438f4c50d99SHerbert Xu int pos; 2439f4c50d99SHerbert Xu 2440f4c50d99SHerbert Xu __skb_push(skb, doffset); 2441f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2442f4c50d99SHerbert Xu pos = skb_headlen(skb); 2443f4c50d99SHerbert Xu 2444f4c50d99SHerbert Xu do { 2445f4c50d99SHerbert Xu struct sk_buff *nskb; 2446f4c50d99SHerbert Xu skb_frag_t *frag; 2447c8884eddSHerbert Xu int hsize; 2448f4c50d99SHerbert Xu int size; 2449f4c50d99SHerbert Xu 2450f4c50d99SHerbert Xu len = skb->len - offset; 2451f4c50d99SHerbert Xu if (len > mss) 2452f4c50d99SHerbert Xu len = mss; 2453f4c50d99SHerbert Xu 2454f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2455f4c50d99SHerbert Xu if (hsize < 0) 2456f4c50d99SHerbert Xu hsize = 0; 2457c8884eddSHerbert Xu if (hsize > len || !sg) 2458c8884eddSHerbert Xu hsize = len; 2459f4c50d99SHerbert Xu 246089319d38SHerbert Xu if (!hsize && i >= nfrags) { 246189319d38SHerbert Xu BUG_ON(fskb->len != len); 246289319d38SHerbert Xu 246389319d38SHerbert Xu pos += len; 246489319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 246589319d38SHerbert Xu fskb = fskb->next; 246689319d38SHerbert Xu 2467f4c50d99SHerbert Xu if (unlikely(!nskb)) 2468f4c50d99SHerbert Xu goto err; 2469f4c50d99SHerbert Xu 247089319d38SHerbert Xu hsize = skb_end_pointer(nskb) - nskb->head; 247189319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 247289319d38SHerbert Xu kfree_skb(nskb); 247389319d38SHerbert Xu goto err; 247489319d38SHerbert Xu } 247589319d38SHerbert Xu 247689319d38SHerbert Xu nskb->truesize += skb_end_pointer(nskb) - nskb->head - 247789319d38SHerbert Xu hsize; 247889319d38SHerbert Xu skb_release_head_state(nskb); 247989319d38SHerbert Xu __skb_push(nskb, doffset); 248089319d38SHerbert Xu } else { 248189319d38SHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, 248289319d38SHerbert Xu GFP_ATOMIC); 248389319d38SHerbert Xu 248489319d38SHerbert Xu if (unlikely(!nskb)) 248589319d38SHerbert Xu goto err; 248689319d38SHerbert Xu 248789319d38SHerbert Xu skb_reserve(nskb, headroom); 248889319d38SHerbert Xu __skb_put(nskb, doffset); 248989319d38SHerbert Xu } 249089319d38SHerbert Xu 2491f4c50d99SHerbert Xu if (segs) 2492f4c50d99SHerbert Xu tail->next = nskb; 2493f4c50d99SHerbert Xu else 2494f4c50d99SHerbert Xu segs = nskb; 2495f4c50d99SHerbert Xu tail = nskb; 2496f4c50d99SHerbert Xu 24976f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2498f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2499f4c50d99SHerbert Xu 2500459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(nskb); 2501ddc7b8e3SArnaldo Carvalho de Melo skb_set_network_header(nskb, skb->mac_len); 2502b0e380b1SArnaldo Carvalho de Melo nskb->transport_header = (nskb->network_header + 2503b0e380b1SArnaldo Carvalho de Melo skb_network_header_len(skb)); 250489319d38SHerbert Xu skb_copy_from_linear_data(skb, nskb->data, doffset); 250589319d38SHerbert Xu 250689319d38SHerbert Xu if (pos >= offset + len) 250789319d38SHerbert Xu continue; 250889319d38SHerbert Xu 2509f4c50d99SHerbert Xu if (!sg) { 25106f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2511f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2512f4c50d99SHerbert Xu skb_put(nskb, len), 2513f4c50d99SHerbert Xu len, 0); 2514f4c50d99SHerbert Xu continue; 2515f4c50d99SHerbert Xu } 2516f4c50d99SHerbert Xu 2517f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2518f4c50d99SHerbert Xu 2519d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2520d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2521f4c50d99SHerbert Xu 252289319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2523f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2524f4c50d99SHerbert Xu get_page(frag->page); 2525f4c50d99SHerbert Xu size = frag->size; 2526f4c50d99SHerbert Xu 2527f4c50d99SHerbert Xu if (pos < offset) { 2528f4c50d99SHerbert Xu frag->page_offset += offset - pos; 2529f4c50d99SHerbert Xu frag->size -= offset - pos; 2530f4c50d99SHerbert Xu } 2531f4c50d99SHerbert Xu 253289319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2533f4c50d99SHerbert Xu 2534f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2535f4c50d99SHerbert Xu i++; 2536f4c50d99SHerbert Xu pos += size; 2537f4c50d99SHerbert Xu } else { 2538f4c50d99SHerbert Xu frag->size -= pos + size - (offset + len); 253989319d38SHerbert Xu goto skip_fraglist; 2540f4c50d99SHerbert Xu } 2541f4c50d99SHerbert Xu 2542f4c50d99SHerbert Xu frag++; 2543f4c50d99SHerbert Xu } 2544f4c50d99SHerbert Xu 254589319d38SHerbert Xu if (pos < offset + len) { 254689319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 254789319d38SHerbert Xu 254889319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 254989319d38SHerbert Xu 255089319d38SHerbert Xu pos += fskb->len; 255189319d38SHerbert Xu fskb = fskb->next; 255289319d38SHerbert Xu 255389319d38SHerbert Xu if (fskb2->next) { 255489319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 255589319d38SHerbert Xu if (!fskb2) 255689319d38SHerbert Xu goto err; 255789319d38SHerbert Xu } else 255889319d38SHerbert Xu skb_get(fskb2); 255989319d38SHerbert Xu 256089319d38SHerbert Xu BUG_ON(skb_shinfo(nskb)->frag_list); 256189319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 256289319d38SHerbert Xu } 256389319d38SHerbert Xu 256489319d38SHerbert Xu skip_fraglist: 2565f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2566f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2567f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2568f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2569f4c50d99SHerbert Xu 2570f4c50d99SHerbert Xu return segs; 2571f4c50d99SHerbert Xu 2572f4c50d99SHerbert Xu err: 2573f4c50d99SHerbert Xu while ((skb = segs)) { 2574f4c50d99SHerbert Xu segs = skb->next; 2575b08d5840SPatrick McHardy kfree_skb(skb); 2576f4c50d99SHerbert Xu } 2577f4c50d99SHerbert Xu return ERR_PTR(err); 2578f4c50d99SHerbert Xu } 2579f4c50d99SHerbert Xu 2580f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2581f4c50d99SHerbert Xu 258271d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 258371d93b39SHerbert Xu { 258471d93b39SHerbert Xu struct sk_buff *p = *head; 258571d93b39SHerbert Xu struct sk_buff *nskb; 258671d93b39SHerbert Xu unsigned int headroom; 258771d93b39SHerbert Xu unsigned int hlen = p->data - skb_mac_header(p); 258871d93b39SHerbert Xu 258971d93b39SHerbert Xu if (hlen + p->len + skb->len >= 65536) 259071d93b39SHerbert Xu return -E2BIG; 259171d93b39SHerbert Xu 259271d93b39SHerbert Xu if (skb_shinfo(p)->frag_list) 259371d93b39SHerbert Xu goto merge; 25945d38a079SHerbert Xu else if (!skb_headlen(p) && !skb_headlen(skb) && 25955d38a079SHerbert Xu skb_shinfo(p)->nr_frags + skb_shinfo(skb)->nr_frags < 25965d38a079SHerbert Xu MAX_SKB_FRAGS) { 25975d38a079SHerbert Xu memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags, 25985d38a079SHerbert Xu skb_shinfo(skb)->frags, 25995d38a079SHerbert Xu skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); 26005d38a079SHerbert Xu 26015d38a079SHerbert Xu skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags; 2602f5572068SHerbert Xu skb_shinfo(skb)->nr_frags = 0; 2603f5572068SHerbert Xu 2604f5572068SHerbert Xu skb->truesize -= skb->data_len; 2605f5572068SHerbert Xu skb->len -= skb->data_len; 2606f5572068SHerbert Xu skb->data_len = 0; 2607f5572068SHerbert Xu 26085d38a079SHerbert Xu NAPI_GRO_CB(skb)->free = 1; 26095d38a079SHerbert Xu goto done; 26105d38a079SHerbert Xu } 261171d93b39SHerbert Xu 261271d93b39SHerbert Xu headroom = skb_headroom(p); 261371d93b39SHerbert Xu nskb = netdev_alloc_skb(p->dev, headroom); 261471d93b39SHerbert Xu if (unlikely(!nskb)) 261571d93b39SHerbert Xu return -ENOMEM; 261671d93b39SHerbert Xu 261771d93b39SHerbert Xu __copy_skb_header(nskb, p); 261871d93b39SHerbert Xu nskb->mac_len = p->mac_len; 261971d93b39SHerbert Xu 262071d93b39SHerbert Xu skb_reserve(nskb, headroom); 262171d93b39SHerbert Xu 262271d93b39SHerbert Xu skb_set_mac_header(nskb, -hlen); 262371d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 262471d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 262571d93b39SHerbert Xu 262671d93b39SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), hlen); 262771d93b39SHerbert Xu 262871d93b39SHerbert Xu *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); 262971d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 2630b530256dSHerbert Xu skb_shinfo(nskb)->gso_size = skb_shinfo(p)->gso_size; 263171d93b39SHerbert Xu skb_header_release(p); 263271d93b39SHerbert Xu nskb->prev = p; 263371d93b39SHerbert Xu 263471d93b39SHerbert Xu nskb->data_len += p->len; 263571d93b39SHerbert Xu nskb->truesize += p->len; 263671d93b39SHerbert Xu nskb->len += p->len; 263771d93b39SHerbert Xu 263871d93b39SHerbert Xu *head = nskb; 263971d93b39SHerbert Xu nskb->next = p->next; 264071d93b39SHerbert Xu p->next = NULL; 264171d93b39SHerbert Xu 264271d93b39SHerbert Xu p = nskb; 264371d93b39SHerbert Xu 264471d93b39SHerbert Xu merge: 264571d93b39SHerbert Xu p->prev->next = skb; 264671d93b39SHerbert Xu p->prev = skb; 264771d93b39SHerbert Xu skb_header_release(skb); 264871d93b39SHerbert Xu 26495d38a079SHerbert Xu done: 26505d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 265171d93b39SHerbert Xu p->data_len += skb->len; 265271d93b39SHerbert Xu p->truesize += skb->len; 265371d93b39SHerbert Xu p->len += skb->len; 265471d93b39SHerbert Xu 265571d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 265671d93b39SHerbert Xu return 0; 265771d93b39SHerbert Xu } 265871d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 265971d93b39SHerbert Xu 26601da177e4SLinus Torvalds void __init skb_init(void) 26611da177e4SLinus Torvalds { 26621da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 26631da177e4SLinus Torvalds sizeof(struct sk_buff), 26641da177e4SLinus Torvalds 0, 2665e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 266620c2df83SPaul Mundt NULL); 2667d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2668d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2669d179cd12SDavid S. Miller sizeof(atomic_t), 2670d179cd12SDavid S. Miller 0, 2671e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 267220c2df83SPaul Mundt NULL); 26731da177e4SLinus Torvalds } 26741da177e4SLinus Torvalds 2675716ea3a7SDavid Howells /** 2676716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 2677716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 2678716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 2679716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 2680716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 2681716ea3a7SDavid Howells * 2682716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 2683716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 2684716ea3a7SDavid Howells */ 268551c739d1SDavid S. Miller static int 268651c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 2687716ea3a7SDavid Howells { 26881a028e50SDavid S. Miller int start = skb_headlen(skb); 26891a028e50SDavid S. Miller int i, copy = start - offset; 2690716ea3a7SDavid Howells int elt = 0; 2691716ea3a7SDavid Howells 2692716ea3a7SDavid Howells if (copy > 0) { 2693716ea3a7SDavid Howells if (copy > len) 2694716ea3a7SDavid Howells copy = len; 2695642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 2696716ea3a7SDavid Howells elt++; 2697716ea3a7SDavid Howells if ((len -= copy) == 0) 2698716ea3a7SDavid Howells return elt; 2699716ea3a7SDavid Howells offset += copy; 2700716ea3a7SDavid Howells } 2701716ea3a7SDavid Howells 2702716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 27031a028e50SDavid S. Miller int end; 2704716ea3a7SDavid Howells 2705547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27061a028e50SDavid S. Miller 27071a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 2708716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2709716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2710716ea3a7SDavid Howells 2711716ea3a7SDavid Howells if (copy > len) 2712716ea3a7SDavid Howells copy = len; 2713642f1490SJens Axboe sg_set_page(&sg[elt], frag->page, copy, 2714642f1490SJens Axboe frag->page_offset+offset-start); 2715716ea3a7SDavid Howells elt++; 2716716ea3a7SDavid Howells if (!(len -= copy)) 2717716ea3a7SDavid Howells return elt; 2718716ea3a7SDavid Howells offset += copy; 2719716ea3a7SDavid Howells } 27201a028e50SDavid S. Miller start = end; 2721716ea3a7SDavid Howells } 2722716ea3a7SDavid Howells 2723716ea3a7SDavid Howells if (skb_shinfo(skb)->frag_list) { 2724716ea3a7SDavid Howells struct sk_buff *list = skb_shinfo(skb)->frag_list; 2725716ea3a7SDavid Howells 2726716ea3a7SDavid Howells for (; list; list = list->next) { 27271a028e50SDavid S. Miller int end; 2728716ea3a7SDavid Howells 2729547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27301a028e50SDavid S. Miller 27311a028e50SDavid S. Miller end = start + list->len; 2732716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2733716ea3a7SDavid Howells if (copy > len) 2734716ea3a7SDavid Howells copy = len; 273551c739d1SDavid S. Miller elt += __skb_to_sgvec(list, sg+elt, offset - start, 273651c739d1SDavid S. Miller copy); 2737716ea3a7SDavid Howells if ((len -= copy) == 0) 2738716ea3a7SDavid Howells return elt; 2739716ea3a7SDavid Howells offset += copy; 2740716ea3a7SDavid Howells } 27411a028e50SDavid S. Miller start = end; 2742716ea3a7SDavid Howells } 2743716ea3a7SDavid Howells } 2744716ea3a7SDavid Howells BUG_ON(len); 2745716ea3a7SDavid Howells return elt; 2746716ea3a7SDavid Howells } 2747716ea3a7SDavid Howells 274851c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 274951c739d1SDavid S. Miller { 275051c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 275151c739d1SDavid S. Miller 2752c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 275351c739d1SDavid S. Miller 275451c739d1SDavid S. Miller return nsg; 275551c739d1SDavid S. Miller } 275651c739d1SDavid S. Miller 2757716ea3a7SDavid Howells /** 2758716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 2759716ea3a7SDavid Howells * @skb: The socket buffer to check. 2760716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 2761716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 2762716ea3a7SDavid Howells * 2763716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 2764716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 2765716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 2766716ea3a7SDavid Howells * 2767716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 2768716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 2769716ea3a7SDavid Howells * set to point to the skb in which this space begins. 2770716ea3a7SDavid Howells * 2771716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 2772716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 2773716ea3a7SDavid Howells */ 2774716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 2775716ea3a7SDavid Howells { 2776716ea3a7SDavid Howells int copyflag; 2777716ea3a7SDavid Howells int elt; 2778716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 2779716ea3a7SDavid Howells 2780716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 2781716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 2782716ea3a7SDavid Howells * at the moment even if they are anonymous). 2783716ea3a7SDavid Howells */ 2784716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 2785716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 2786716ea3a7SDavid Howells return -ENOMEM; 2787716ea3a7SDavid Howells 2788716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 2789716ea3a7SDavid Howells if (!skb_shinfo(skb)->frag_list) { 2790716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 2791716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 2792716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 2793716ea3a7SDavid Howells * space, 128 bytes is fair. */ 2794716ea3a7SDavid Howells 2795716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 2796716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 2797716ea3a7SDavid Howells return -ENOMEM; 2798716ea3a7SDavid Howells 2799716ea3a7SDavid Howells /* Voila! */ 2800716ea3a7SDavid Howells *trailer = skb; 2801716ea3a7SDavid Howells return 1; 2802716ea3a7SDavid Howells } 2803716ea3a7SDavid Howells 2804716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 2805716ea3a7SDavid Howells 2806716ea3a7SDavid Howells elt = 1; 2807716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 2808716ea3a7SDavid Howells copyflag = 0; 2809716ea3a7SDavid Howells 2810716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 2811716ea3a7SDavid Howells int ntail = 0; 2812716ea3a7SDavid Howells 2813716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 2814716ea3a7SDavid Howells * this can happen on input. Copy it and everything 2815716ea3a7SDavid Howells * after it. */ 2816716ea3a7SDavid Howells 2817716ea3a7SDavid Howells if (skb_shared(skb1)) 2818716ea3a7SDavid Howells copyflag = 1; 2819716ea3a7SDavid Howells 2820716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 2821716ea3a7SDavid Howells 2822716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 2823716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 2824716ea3a7SDavid Howells skb_shinfo(skb1)->frag_list || 2825716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 2826716ea3a7SDavid Howells ntail = tailbits + 128; 2827716ea3a7SDavid Howells } 2828716ea3a7SDavid Howells 2829716ea3a7SDavid Howells if (copyflag || 2830716ea3a7SDavid Howells skb_cloned(skb1) || 2831716ea3a7SDavid Howells ntail || 2832716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 2833716ea3a7SDavid Howells skb_shinfo(skb1)->frag_list) { 2834716ea3a7SDavid Howells struct sk_buff *skb2; 2835716ea3a7SDavid Howells 2836716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 2837716ea3a7SDavid Howells if (ntail == 0) 2838716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 2839716ea3a7SDavid Howells else 2840716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 2841716ea3a7SDavid Howells skb_headroom(skb1), 2842716ea3a7SDavid Howells ntail, 2843716ea3a7SDavid Howells GFP_ATOMIC); 2844716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 2845716ea3a7SDavid Howells return -ENOMEM; 2846716ea3a7SDavid Howells 2847716ea3a7SDavid Howells if (skb1->sk) 2848716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 2849716ea3a7SDavid Howells 2850716ea3a7SDavid Howells /* Looking around. Are we still alive? 2851716ea3a7SDavid Howells * OK, link new skb, drop old one */ 2852716ea3a7SDavid Howells 2853716ea3a7SDavid Howells skb2->next = skb1->next; 2854716ea3a7SDavid Howells *skb_p = skb2; 2855716ea3a7SDavid Howells kfree_skb(skb1); 2856716ea3a7SDavid Howells skb1 = skb2; 2857716ea3a7SDavid Howells } 2858716ea3a7SDavid Howells elt++; 2859716ea3a7SDavid Howells *trailer = skb1; 2860716ea3a7SDavid Howells skb_p = &skb1->next; 2861716ea3a7SDavid Howells } 2862716ea3a7SDavid Howells 2863716ea3a7SDavid Howells return elt; 2864716ea3a7SDavid Howells } 2865716ea3a7SDavid Howells 2866f35d9d8aSRusty Russell /** 2867f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 2868f35d9d8aSRusty Russell * @skb: the skb to set 2869f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 2870f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 2871f35d9d8aSRusty Russell * 2872f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 2873f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 2874f35d9d8aSRusty Russell * 2875f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 2876f35d9d8aSRusty Russell * returns false you should drop the packet. 2877f35d9d8aSRusty Russell */ 2878f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 2879f35d9d8aSRusty Russell { 2880f35d9d8aSRusty Russell if (unlikely(start > skb->len - 2) || 2881f35d9d8aSRusty Russell unlikely((int)start + off > skb->len - 2)) { 2882f35d9d8aSRusty Russell if (net_ratelimit()) 2883f35d9d8aSRusty Russell printk(KERN_WARNING 2884f35d9d8aSRusty Russell "bad partial csum: csum=%u/%u len=%u\n", 2885f35d9d8aSRusty Russell start, off, skb->len); 2886f35d9d8aSRusty Russell return false; 2887f35d9d8aSRusty Russell } 2888f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 2889f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 2890f35d9d8aSRusty Russell skb->csum_offset = off; 2891f35d9d8aSRusty Russell return true; 2892f35d9d8aSRusty Russell } 2893f35d9d8aSRusty Russell 28944497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 28954497b076SBen Hutchings { 28964497b076SBen Hutchings if (net_ratelimit()) 28974497b076SBen Hutchings pr_warning("%s: received packets cannot be forwarded" 28984497b076SBen Hutchings " while LRO is enabled\n", skb->dev->name); 28994497b076SBen Hutchings } 29004497b076SBen Hutchings 29011da177e4SLinus Torvalds EXPORT_SYMBOL(___pskb_trim); 29021da177e4SLinus Torvalds EXPORT_SYMBOL(__kfree_skb); 2903231d06aeSJörn Engel EXPORT_SYMBOL(kfree_skb); 29041da177e4SLinus Torvalds EXPORT_SYMBOL(__pskb_pull_tail); 2905d179cd12SDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 29068af27456SChristoph Hellwig EXPORT_SYMBOL(__netdev_alloc_skb); 29071da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_copy); 29081da177e4SLinus Torvalds EXPORT_SYMBOL(pskb_expand_head); 29091da177e4SLinus Torvalds EXPORT_SYMBOL(skb_checksum); 29101da177e4SLinus Torvalds EXPORT_SYMBOL(skb_clone); 29111da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy); 29121da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_bits); 29131da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_and_csum_dev); 29141da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_bits); 29151da177e4SLinus Torvalds EXPORT_SYMBOL(skb_copy_expand); 29161da177e4SLinus Torvalds EXPORT_SYMBOL(skb_over_panic); 29171da177e4SLinus Torvalds EXPORT_SYMBOL(skb_pad); 29181da177e4SLinus Torvalds EXPORT_SYMBOL(skb_realloc_headroom); 29191da177e4SLinus Torvalds EXPORT_SYMBOL(skb_under_panic); 29201da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue); 29211da177e4SLinus Torvalds EXPORT_SYMBOL(skb_dequeue_tail); 29221da177e4SLinus Torvalds EXPORT_SYMBOL(skb_insert); 29231da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_purge); 29241da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_head); 29251da177e4SLinus Torvalds EXPORT_SYMBOL(skb_queue_tail); 29261da177e4SLinus Torvalds EXPORT_SYMBOL(skb_unlink); 29271da177e4SLinus Torvalds EXPORT_SYMBOL(skb_append); 29281da177e4SLinus Torvalds EXPORT_SYMBOL(skb_split); 2929677e90edSThomas Graf EXPORT_SYMBOL(skb_prepare_seq_read); 2930677e90edSThomas Graf EXPORT_SYMBOL(skb_seq_read); 2931677e90edSThomas Graf EXPORT_SYMBOL(skb_abort_seq_read); 29323fc7e8a6SThomas Graf EXPORT_SYMBOL(skb_find_text); 2933e89e9cf5SAnanda Raju EXPORT_SYMBOL(skb_append_datato_frags); 29344497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 2935716ea3a7SDavid Howells 2936716ea3a7SDavid Howells EXPORT_SYMBOL_GPL(skb_to_sgvec); 2937716ea3a7SDavid Howells EXPORT_SYMBOL_GPL(skb_cow_data); 2938f35d9d8aSRusty Russell EXPORT_SYMBOL_GPL(skb_partial_csum_set); 2939