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 } 126b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_over_panic); 1271da177e4SLinus Torvalds 1281da177e4SLinus Torvalds /** 1291da177e4SLinus Torvalds * skb_under_panic - private function 1301da177e4SLinus Torvalds * @skb: buffer 1311da177e4SLinus Torvalds * @sz: size 1321da177e4SLinus Torvalds * @here: address 1331da177e4SLinus Torvalds * 1341da177e4SLinus Torvalds * Out of line support code for skb_push(). Not user callable. 1351da177e4SLinus Torvalds */ 1361da177e4SLinus Torvalds 1371da177e4SLinus Torvalds void skb_under_panic(struct sk_buff *skb, int sz, void *here) 1381da177e4SLinus Torvalds { 13926095455SPatrick McHardy printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p " 1404305b541SArnaldo Carvalho de Melo "data:%p tail:%#lx end:%#lx dev:%s\n", 14127a884dcSArnaldo Carvalho de Melo here, skb->len, sz, skb->head, skb->data, 1424305b541SArnaldo Carvalho de Melo (unsigned long)skb->tail, (unsigned long)skb->end, 14326095455SPatrick McHardy skb->dev ? skb->dev->name : "<NULL>"); 1441da177e4SLinus Torvalds BUG(); 1451da177e4SLinus Torvalds } 146b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_under_panic); 1471da177e4SLinus Torvalds 148dc6de336SDavid S. Miller void skb_truesize_bug(struct sk_buff *skb) 149dc6de336SDavid S. Miller { 1508f480c0eSArjan van de Ven WARN(net_ratelimit(), KERN_ERR "SKB BUG: Invalid truesize (%u) " 151dc6de336SDavid S. Miller "len=%u, sizeof(sk_buff)=%Zd\n", 152dc6de336SDavid S. Miller skb->truesize, skb->len, sizeof(struct sk_buff)); 153dc6de336SDavid S. Miller } 154dc6de336SDavid S. Miller EXPORT_SYMBOL(skb_truesize_bug); 155dc6de336SDavid S. Miller 1561da177e4SLinus Torvalds /* Allocate a new skbuff. We do this ourselves so we can fill in a few 1571da177e4SLinus Torvalds * 'private' fields and also do memory statistics to find all the 1581da177e4SLinus Torvalds * [BEEP] leaks. 1591da177e4SLinus Torvalds * 1601da177e4SLinus Torvalds */ 1611da177e4SLinus Torvalds 1621da177e4SLinus Torvalds /** 163d179cd12SDavid S. Miller * __alloc_skb - allocate a network buffer 1641da177e4SLinus Torvalds * @size: size to allocate 1651da177e4SLinus Torvalds * @gfp_mask: allocation mask 166c83c2486SRandy Dunlap * @fclone: allocate from fclone cache instead of head cache 167c83c2486SRandy Dunlap * and allocate a cloned (child) skb 168b30973f8SChristoph Hellwig * @node: numa node to allocate memory on 1691da177e4SLinus Torvalds * 1701da177e4SLinus Torvalds * Allocate a new &sk_buff. The returned buffer has no headroom and a 1711da177e4SLinus Torvalds * tail room of size bytes. The object has a reference count of one. 1721da177e4SLinus Torvalds * The return is the buffer. On a failure the return is %NULL. 1731da177e4SLinus Torvalds * 1741da177e4SLinus Torvalds * Buffers may only be allocated from interrupts using a @gfp_mask of 1751da177e4SLinus Torvalds * %GFP_ATOMIC. 1761da177e4SLinus Torvalds */ 177dd0fc66fSAl Viro struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask, 178b30973f8SChristoph Hellwig int fclone, int node) 1791da177e4SLinus Torvalds { 180e18b890bSChristoph Lameter struct kmem_cache *cache; 1814947d3efSBenjamin LaHaise struct skb_shared_info *shinfo; 1821da177e4SLinus Torvalds struct sk_buff *skb; 1831da177e4SLinus Torvalds u8 *data; 1841da177e4SLinus Torvalds 1858798b3fbSHerbert Xu cache = fclone ? skbuff_fclone_cache : skbuff_head_cache; 1868798b3fbSHerbert Xu 1871da177e4SLinus Torvalds /* Get the HEAD */ 188b30973f8SChristoph Hellwig skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node); 1891da177e4SLinus Torvalds if (!skb) 1901da177e4SLinus Torvalds goto out; 1911da177e4SLinus Torvalds 1921da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 193b30973f8SChristoph Hellwig data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info), 194b30973f8SChristoph Hellwig gfp_mask, node); 1951da177e4SLinus Torvalds if (!data) 1961da177e4SLinus Torvalds goto nodata; 1971da177e4SLinus Torvalds 198ca0605a7SArnaldo Carvalho de Melo /* 199c8005785SJohannes Berg * Only clear those fields we need to clear, not those that we will 200c8005785SJohannes Berg * actually initialise below. Hence, don't put any more fields after 201c8005785SJohannes Berg * the tail pointer in struct sk_buff! 202ca0605a7SArnaldo Carvalho de Melo */ 203ca0605a7SArnaldo Carvalho de Melo memset(skb, 0, offsetof(struct sk_buff, tail)); 2041da177e4SLinus Torvalds skb->truesize = size + sizeof(struct sk_buff); 2051da177e4SLinus Torvalds atomic_set(&skb->users, 1); 2061da177e4SLinus Torvalds skb->head = data; 2071da177e4SLinus Torvalds skb->data = data; 20827a884dcSArnaldo Carvalho de Melo skb_reset_tail_pointer(skb); 2094305b541SArnaldo Carvalho de Melo skb->end = skb->tail + size; 2104947d3efSBenjamin LaHaise /* make sure we initialize shinfo sequentially */ 2114947d3efSBenjamin LaHaise shinfo = skb_shinfo(skb); 2124947d3efSBenjamin LaHaise atomic_set(&shinfo->dataref, 1); 2134947d3efSBenjamin LaHaise shinfo->nr_frags = 0; 2147967168cSHerbert Xu shinfo->gso_size = 0; 2157967168cSHerbert Xu shinfo->gso_segs = 0; 2167967168cSHerbert Xu shinfo->gso_type = 0; 2174947d3efSBenjamin LaHaise shinfo->ip6_frag_id = 0; 2184947d3efSBenjamin LaHaise shinfo->frag_list = NULL; 2194947d3efSBenjamin LaHaise 220d179cd12SDavid S. Miller if (fclone) { 221d179cd12SDavid S. Miller struct sk_buff *child = skb + 1; 222d179cd12SDavid S. Miller atomic_t *fclone_ref = (atomic_t *) (child + 1); 2231da177e4SLinus Torvalds 224d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_ORIG; 225d179cd12SDavid S. Miller atomic_set(fclone_ref, 1); 226d179cd12SDavid S. Miller 227d179cd12SDavid S. Miller child->fclone = SKB_FCLONE_UNAVAILABLE; 228d179cd12SDavid S. Miller } 2291da177e4SLinus Torvalds out: 2301da177e4SLinus Torvalds return skb; 2311da177e4SLinus Torvalds nodata: 2328798b3fbSHerbert Xu kmem_cache_free(cache, skb); 2331da177e4SLinus Torvalds skb = NULL; 2341da177e4SLinus Torvalds goto out; 2351da177e4SLinus Torvalds } 236b4ac530fSDavid S. Miller EXPORT_SYMBOL(__alloc_skb); 2371da177e4SLinus Torvalds 2381da177e4SLinus Torvalds /** 2398af27456SChristoph Hellwig * __netdev_alloc_skb - allocate an skbuff for rx on a specific device 2408af27456SChristoph Hellwig * @dev: network device to receive on 2418af27456SChristoph Hellwig * @length: length to allocate 2428af27456SChristoph Hellwig * @gfp_mask: get_free_pages mask, passed to alloc_skb 2438af27456SChristoph Hellwig * 2448af27456SChristoph Hellwig * Allocate a new &sk_buff and assign it a usage count of one. The 2458af27456SChristoph Hellwig * buffer has unspecified headroom built in. Users should allocate 2468af27456SChristoph Hellwig * the headroom they think they need without accounting for the 2478af27456SChristoph Hellwig * built in space. The built in space is used for optimisations. 2488af27456SChristoph Hellwig * 2498af27456SChristoph Hellwig * %NULL is returned if there is no free memory. 2508af27456SChristoph Hellwig */ 2518af27456SChristoph Hellwig struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 2528af27456SChristoph Hellwig unsigned int length, gfp_t gfp_mask) 2538af27456SChristoph Hellwig { 25443cb76d9SGreg Kroah-Hartman int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; 2558af27456SChristoph Hellwig struct sk_buff *skb; 2568af27456SChristoph Hellwig 257b30973f8SChristoph Hellwig skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node); 2587b2e497aSChristoph Hellwig if (likely(skb)) { 2598af27456SChristoph Hellwig skb_reserve(skb, NET_SKB_PAD); 2607b2e497aSChristoph Hellwig skb->dev = dev; 2617b2e497aSChristoph Hellwig } 2628af27456SChristoph Hellwig return skb; 2638af27456SChristoph Hellwig } 264b4ac530fSDavid S. Miller EXPORT_SYMBOL(__netdev_alloc_skb); 2651da177e4SLinus Torvalds 266654bed16SPeter Zijlstra struct page *__netdev_alloc_page(struct net_device *dev, gfp_t gfp_mask) 267654bed16SPeter Zijlstra { 268654bed16SPeter Zijlstra int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1; 269654bed16SPeter Zijlstra struct page *page; 270654bed16SPeter Zijlstra 271654bed16SPeter Zijlstra page = alloc_pages_node(node, gfp_mask, 0); 272654bed16SPeter Zijlstra return page; 273654bed16SPeter Zijlstra } 274654bed16SPeter Zijlstra EXPORT_SYMBOL(__netdev_alloc_page); 275654bed16SPeter Zijlstra 276654bed16SPeter Zijlstra void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 277654bed16SPeter Zijlstra int size) 278654bed16SPeter Zijlstra { 279654bed16SPeter Zijlstra skb_fill_page_desc(skb, i, page, off, size); 280654bed16SPeter Zijlstra skb->len += size; 281654bed16SPeter Zijlstra skb->data_len += size; 282654bed16SPeter Zijlstra skb->truesize += size; 283654bed16SPeter Zijlstra } 284654bed16SPeter Zijlstra EXPORT_SYMBOL(skb_add_rx_frag); 285654bed16SPeter Zijlstra 286f58518e6SIlpo Järvinen /** 287f58518e6SIlpo Järvinen * dev_alloc_skb - allocate an skbuff for receiving 288f58518e6SIlpo Järvinen * @length: length to allocate 289f58518e6SIlpo Järvinen * 290f58518e6SIlpo Järvinen * Allocate a new &sk_buff and assign it a usage count of one. The 291f58518e6SIlpo Järvinen * buffer has unspecified headroom built in. Users should allocate 292f58518e6SIlpo Järvinen * the headroom they think they need without accounting for the 293f58518e6SIlpo Järvinen * built in space. The built in space is used for optimisations. 294f58518e6SIlpo Järvinen * 295f58518e6SIlpo Järvinen * %NULL is returned if there is no free memory. Although this function 296f58518e6SIlpo Järvinen * allocates memory it can be called from an interrupt. 297f58518e6SIlpo Järvinen */ 298f58518e6SIlpo Järvinen struct sk_buff *dev_alloc_skb(unsigned int length) 299f58518e6SIlpo Järvinen { 3001483b874SDenys Vlasenko /* 3011483b874SDenys Vlasenko * There is more code here than it seems: 302a0f55e0eSDavid S. Miller * __dev_alloc_skb is an inline 3031483b874SDenys Vlasenko */ 304f58518e6SIlpo Järvinen return __dev_alloc_skb(length, GFP_ATOMIC); 305f58518e6SIlpo Järvinen } 306f58518e6SIlpo Järvinen EXPORT_SYMBOL(dev_alloc_skb); 307f58518e6SIlpo Järvinen 30827b437c8SHerbert Xu static void skb_drop_list(struct sk_buff **listp) 3091da177e4SLinus Torvalds { 31027b437c8SHerbert Xu struct sk_buff *list = *listp; 3111da177e4SLinus Torvalds 31227b437c8SHerbert Xu *listp = NULL; 3131da177e4SLinus Torvalds 3141da177e4SLinus Torvalds do { 3151da177e4SLinus Torvalds struct sk_buff *this = list; 3161da177e4SLinus Torvalds list = list->next; 3171da177e4SLinus Torvalds kfree_skb(this); 3181da177e4SLinus Torvalds } while (list); 3191da177e4SLinus Torvalds } 3201da177e4SLinus Torvalds 32127b437c8SHerbert Xu static inline void skb_drop_fraglist(struct sk_buff *skb) 32227b437c8SHerbert Xu { 32327b437c8SHerbert Xu skb_drop_list(&skb_shinfo(skb)->frag_list); 32427b437c8SHerbert Xu } 32527b437c8SHerbert Xu 3261da177e4SLinus Torvalds static void skb_clone_fraglist(struct sk_buff *skb) 3271da177e4SLinus Torvalds { 3281da177e4SLinus Torvalds struct sk_buff *list; 3291da177e4SLinus Torvalds 3301da177e4SLinus Torvalds for (list = skb_shinfo(skb)->frag_list; list; list = list->next) 3311da177e4SLinus Torvalds skb_get(list); 3321da177e4SLinus Torvalds } 3331da177e4SLinus Torvalds 3345bba1712SAdrian Bunk static void skb_release_data(struct sk_buff *skb) 3351da177e4SLinus Torvalds { 3361da177e4SLinus Torvalds if (!skb->cloned || 3371da177e4SLinus Torvalds !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1, 3381da177e4SLinus Torvalds &skb_shinfo(skb)->dataref)) { 3391da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 3401da177e4SLinus Torvalds int i; 3411da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 3421da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 3431da177e4SLinus Torvalds } 3441da177e4SLinus Torvalds 3451da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 3461da177e4SLinus Torvalds skb_drop_fraglist(skb); 3471da177e4SLinus Torvalds 3481da177e4SLinus Torvalds kfree(skb->head); 3491da177e4SLinus Torvalds } 3501da177e4SLinus Torvalds } 3511da177e4SLinus Torvalds 3521da177e4SLinus Torvalds /* 3531da177e4SLinus Torvalds * Free an skbuff by memory without cleaning the state. 3541da177e4SLinus Torvalds */ 3552d4baff8SHerbert Xu static void kfree_skbmem(struct sk_buff *skb) 3561da177e4SLinus Torvalds { 357d179cd12SDavid S. Miller struct sk_buff *other; 358d179cd12SDavid S. Miller atomic_t *fclone_ref; 359d179cd12SDavid S. Miller 360d179cd12SDavid S. Miller switch (skb->fclone) { 361d179cd12SDavid S. Miller case SKB_FCLONE_UNAVAILABLE: 3621da177e4SLinus Torvalds kmem_cache_free(skbuff_head_cache, skb); 363d179cd12SDavid S. Miller break; 364d179cd12SDavid S. Miller 365d179cd12SDavid S. Miller case SKB_FCLONE_ORIG: 366d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 2); 367d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 368d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, skb); 369d179cd12SDavid S. Miller break; 370d179cd12SDavid S. Miller 371d179cd12SDavid S. Miller case SKB_FCLONE_CLONE: 372d179cd12SDavid S. Miller fclone_ref = (atomic_t *) (skb + 1); 373d179cd12SDavid S. Miller other = skb - 1; 374d179cd12SDavid S. Miller 375d179cd12SDavid S. Miller /* The clone portion is available for 376d179cd12SDavid S. Miller * fast-cloning again. 377d179cd12SDavid S. Miller */ 378d179cd12SDavid S. Miller skb->fclone = SKB_FCLONE_UNAVAILABLE; 379d179cd12SDavid S. Miller 380d179cd12SDavid S. Miller if (atomic_dec_and_test(fclone_ref)) 381d179cd12SDavid S. Miller kmem_cache_free(skbuff_fclone_cache, other); 382d179cd12SDavid S. Miller break; 3833ff50b79SStephen Hemminger } 3841da177e4SLinus Torvalds } 3851da177e4SLinus Torvalds 38604a4bb55SLennert Buytenhek static void skb_release_head_state(struct sk_buff *skb) 3871da177e4SLinus Torvalds { 3881da177e4SLinus Torvalds dst_release(skb->dst); 3891da177e4SLinus Torvalds #ifdef CONFIG_XFRM 3901da177e4SLinus Torvalds secpath_put(skb->sp); 3911da177e4SLinus Torvalds #endif 3921da177e4SLinus Torvalds if (skb->destructor) { 3939c2b3328SStephen Hemminger WARN_ON(in_irq()); 3941da177e4SLinus Torvalds skb->destructor(skb); 3951da177e4SLinus Torvalds } 3969fb9cbb1SYasuyuki Kozakai #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3975f79e0f9SYasuyuki Kozakai nf_conntrack_put(skb->nfct); 3989fb9cbb1SYasuyuki Kozakai nf_conntrack_put_reasm(skb->nfct_reasm); 3999fb9cbb1SYasuyuki Kozakai #endif 4001da177e4SLinus Torvalds #ifdef CONFIG_BRIDGE_NETFILTER 4011da177e4SLinus Torvalds nf_bridge_put(skb->nf_bridge); 4021da177e4SLinus Torvalds #endif 4031da177e4SLinus Torvalds /* XXX: IS this still necessary? - JHS */ 4041da177e4SLinus Torvalds #ifdef CONFIG_NET_SCHED 4051da177e4SLinus Torvalds skb->tc_index = 0; 4061da177e4SLinus Torvalds #ifdef CONFIG_NET_CLS_ACT 4071da177e4SLinus Torvalds skb->tc_verd = 0; 4081da177e4SLinus Torvalds #endif 4091da177e4SLinus Torvalds #endif 41004a4bb55SLennert Buytenhek } 41104a4bb55SLennert Buytenhek 41204a4bb55SLennert Buytenhek /* Free everything but the sk_buff shell. */ 41304a4bb55SLennert Buytenhek static void skb_release_all(struct sk_buff *skb) 41404a4bb55SLennert Buytenhek { 41504a4bb55SLennert Buytenhek skb_release_head_state(skb); 4162d4baff8SHerbert Xu skb_release_data(skb); 4172d4baff8SHerbert Xu } 4181da177e4SLinus Torvalds 4192d4baff8SHerbert Xu /** 4202d4baff8SHerbert Xu * __kfree_skb - private function 4212d4baff8SHerbert Xu * @skb: buffer 4222d4baff8SHerbert Xu * 4232d4baff8SHerbert Xu * Free an sk_buff. Release anything attached to the buffer. 4242d4baff8SHerbert Xu * Clean the state. This is an internal helper function. Users should 4252d4baff8SHerbert Xu * always call kfree_skb 4262d4baff8SHerbert Xu */ 4272d4baff8SHerbert Xu 4282d4baff8SHerbert Xu void __kfree_skb(struct sk_buff *skb) 4292d4baff8SHerbert Xu { 4302d4baff8SHerbert Xu skb_release_all(skb); 4311da177e4SLinus Torvalds kfree_skbmem(skb); 4321da177e4SLinus Torvalds } 433b4ac530fSDavid S. Miller EXPORT_SYMBOL(__kfree_skb); 4341da177e4SLinus Torvalds 4351da177e4SLinus Torvalds /** 436231d06aeSJörn Engel * kfree_skb - free an sk_buff 437231d06aeSJörn Engel * @skb: buffer to free 438231d06aeSJörn Engel * 439231d06aeSJörn Engel * Drop a reference to the buffer and free it if the usage count has 440231d06aeSJörn Engel * hit zero. 441231d06aeSJörn Engel */ 442231d06aeSJörn Engel void kfree_skb(struct sk_buff *skb) 443231d06aeSJörn Engel { 444231d06aeSJörn Engel if (unlikely(!skb)) 445231d06aeSJörn Engel return; 446231d06aeSJörn Engel if (likely(atomic_read(&skb->users) == 1)) 447231d06aeSJörn Engel smp_rmb(); 448231d06aeSJörn Engel else if (likely(!atomic_dec_and_test(&skb->users))) 449231d06aeSJörn Engel return; 450231d06aeSJörn Engel __kfree_skb(skb); 451231d06aeSJörn Engel } 452b4ac530fSDavid S. Miller EXPORT_SYMBOL(kfree_skb); 453231d06aeSJörn Engel 454d1a203eaSStephen Hemminger /** 455d1a203eaSStephen Hemminger * skb_recycle_check - check if skb can be reused for receive 456d1a203eaSStephen Hemminger * @skb: buffer 457d1a203eaSStephen Hemminger * @skb_size: minimum receive buffer size 458d1a203eaSStephen Hemminger * 459d1a203eaSStephen Hemminger * Checks that the skb passed in is not shared or cloned, and 460d1a203eaSStephen Hemminger * that it is linear and its head portion at least as large as 461d1a203eaSStephen Hemminger * skb_size so that it can be recycled as a receive buffer. 462d1a203eaSStephen Hemminger * If these conditions are met, this function does any necessary 463d1a203eaSStephen Hemminger * reference count dropping and cleans up the skbuff as if it 464d1a203eaSStephen Hemminger * just came from __alloc_skb(). 465d1a203eaSStephen Hemminger */ 46604a4bb55SLennert Buytenhek int skb_recycle_check(struct sk_buff *skb, int skb_size) 46704a4bb55SLennert Buytenhek { 46804a4bb55SLennert Buytenhek struct skb_shared_info *shinfo; 46904a4bb55SLennert Buytenhek 47004a4bb55SLennert Buytenhek if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) 47104a4bb55SLennert Buytenhek return 0; 47204a4bb55SLennert Buytenhek 47304a4bb55SLennert Buytenhek skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); 47404a4bb55SLennert Buytenhek if (skb_end_pointer(skb) - skb->head < skb_size) 47504a4bb55SLennert Buytenhek return 0; 47604a4bb55SLennert Buytenhek 47704a4bb55SLennert Buytenhek if (skb_shared(skb) || skb_cloned(skb)) 47804a4bb55SLennert Buytenhek return 0; 47904a4bb55SLennert Buytenhek 48004a4bb55SLennert Buytenhek skb_release_head_state(skb); 48104a4bb55SLennert Buytenhek shinfo = skb_shinfo(skb); 48204a4bb55SLennert Buytenhek atomic_set(&shinfo->dataref, 1); 48304a4bb55SLennert Buytenhek shinfo->nr_frags = 0; 48404a4bb55SLennert Buytenhek shinfo->gso_size = 0; 48504a4bb55SLennert Buytenhek shinfo->gso_segs = 0; 48604a4bb55SLennert Buytenhek shinfo->gso_type = 0; 48704a4bb55SLennert Buytenhek shinfo->ip6_frag_id = 0; 48804a4bb55SLennert Buytenhek shinfo->frag_list = NULL; 48904a4bb55SLennert Buytenhek 49004a4bb55SLennert Buytenhek memset(skb, 0, offsetof(struct sk_buff, tail)); 49104a4bb55SLennert Buytenhek skb->data = skb->head + NET_SKB_PAD; 4925cd33db2SLennert Buytenhek skb_reset_tail_pointer(skb); 49304a4bb55SLennert Buytenhek 49404a4bb55SLennert Buytenhek return 1; 49504a4bb55SLennert Buytenhek } 49604a4bb55SLennert Buytenhek EXPORT_SYMBOL(skb_recycle_check); 49704a4bb55SLennert Buytenhek 498dec18810SHerbert Xu static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 499dec18810SHerbert Xu { 500dec18810SHerbert Xu new->tstamp = old->tstamp; 501dec18810SHerbert Xu new->dev = old->dev; 502dec18810SHerbert Xu new->transport_header = old->transport_header; 503dec18810SHerbert Xu new->network_header = old->network_header; 504dec18810SHerbert Xu new->mac_header = old->mac_header; 505dec18810SHerbert Xu new->dst = dst_clone(old->dst); 506def8b4faSAlexey Dobriyan #ifdef CONFIG_XFRM 507dec18810SHerbert Xu new->sp = secpath_get(old->sp); 508dec18810SHerbert Xu #endif 509dec18810SHerbert Xu memcpy(new->cb, old->cb, sizeof(old->cb)); 510dec18810SHerbert Xu new->csum_start = old->csum_start; 511dec18810SHerbert Xu new->csum_offset = old->csum_offset; 512dec18810SHerbert Xu new->local_df = old->local_df; 513dec18810SHerbert Xu new->pkt_type = old->pkt_type; 514dec18810SHerbert Xu new->ip_summed = old->ip_summed; 515dec18810SHerbert Xu skb_copy_queue_mapping(new, old); 516dec18810SHerbert Xu new->priority = old->priority; 517dec18810SHerbert Xu #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE) 518dec18810SHerbert Xu new->ipvs_property = old->ipvs_property; 519dec18810SHerbert Xu #endif 520dec18810SHerbert Xu new->protocol = old->protocol; 521dec18810SHerbert Xu new->mark = old->mark; 522dec18810SHerbert Xu __nf_copy(new, old); 523dec18810SHerbert Xu #if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \ 524dec18810SHerbert Xu defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE) 525dec18810SHerbert Xu new->nf_trace = old->nf_trace; 526dec18810SHerbert Xu #endif 527dec18810SHerbert Xu #ifdef CONFIG_NET_SCHED 528dec18810SHerbert Xu new->tc_index = old->tc_index; 529dec18810SHerbert Xu #ifdef CONFIG_NET_CLS_ACT 530dec18810SHerbert Xu new->tc_verd = old->tc_verd; 531dec18810SHerbert Xu #endif 532dec18810SHerbert Xu #endif 5336aa895b0SPatrick McHardy new->vlan_tci = old->vlan_tci; 5346aa895b0SPatrick McHardy 535dec18810SHerbert Xu skb_copy_secmark(new, old); 536dec18810SHerbert Xu } 537dec18810SHerbert Xu 538e0053ec0SHerbert Xu static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb) 5391da177e4SLinus Torvalds { 5401da177e4SLinus Torvalds #define C(x) n->x = skb->x 5411da177e4SLinus Torvalds 5421da177e4SLinus Torvalds n->next = n->prev = NULL; 5431da177e4SLinus Torvalds n->sk = NULL; 544dec18810SHerbert Xu __copy_skb_header(n, skb); 545dec18810SHerbert Xu 5461da177e4SLinus Torvalds C(len); 5471da177e4SLinus Torvalds C(data_len); 5483e6b3b2eSAlexey Dobriyan C(mac_len); 549334a8132SPatrick McHardy n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len; 55002f1c89dSPaul Moore n->cloned = 1; 5511da177e4SLinus Torvalds n->nohdr = 0; 5521da177e4SLinus Torvalds n->destructor = NULL; 55302f1c89dSPaul Moore C(iif); 5541da177e4SLinus Torvalds C(tail); 5551da177e4SLinus Torvalds C(end); 55602f1c89dSPaul Moore C(head); 55702f1c89dSPaul Moore C(data); 55802f1c89dSPaul Moore C(truesize); 559d0f09804SJohannes Berg #if defined(CONFIG_MAC80211) || defined(CONFIG_MAC80211_MODULE) 560d0f09804SJohannes Berg C(do_not_encrypt); 5618b30b1feSSujith C(requeue); 562d0f09804SJohannes Berg #endif 56302f1c89dSPaul Moore atomic_set(&n->users, 1); 5641da177e4SLinus Torvalds 5651da177e4SLinus Torvalds atomic_inc(&(skb_shinfo(skb)->dataref)); 5661da177e4SLinus Torvalds skb->cloned = 1; 5671da177e4SLinus Torvalds 5681da177e4SLinus Torvalds return n; 569e0053ec0SHerbert Xu #undef C 570e0053ec0SHerbert Xu } 571e0053ec0SHerbert Xu 572e0053ec0SHerbert Xu /** 573e0053ec0SHerbert Xu * skb_morph - morph one skb into another 574e0053ec0SHerbert Xu * @dst: the skb to receive the contents 575e0053ec0SHerbert Xu * @src: the skb to supply the contents 576e0053ec0SHerbert Xu * 577e0053ec0SHerbert Xu * This is identical to skb_clone except that the target skb is 578e0053ec0SHerbert Xu * supplied by the user. 579e0053ec0SHerbert Xu * 580e0053ec0SHerbert Xu * The target skb is returned upon exit. 581e0053ec0SHerbert Xu */ 582e0053ec0SHerbert Xu struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src) 583e0053ec0SHerbert Xu { 5842d4baff8SHerbert Xu skb_release_all(dst); 585e0053ec0SHerbert Xu return __skb_clone(dst, src); 586e0053ec0SHerbert Xu } 587e0053ec0SHerbert Xu EXPORT_SYMBOL_GPL(skb_morph); 588e0053ec0SHerbert Xu 589e0053ec0SHerbert Xu /** 590e0053ec0SHerbert Xu * skb_clone - duplicate an sk_buff 591e0053ec0SHerbert Xu * @skb: buffer to clone 592e0053ec0SHerbert Xu * @gfp_mask: allocation priority 593e0053ec0SHerbert Xu * 594e0053ec0SHerbert Xu * Duplicate an &sk_buff. The new one is not owned by a socket. Both 595e0053ec0SHerbert Xu * copies share the same packet data but not structure. The new 596e0053ec0SHerbert Xu * buffer has a reference count of 1. If the allocation fails the 597e0053ec0SHerbert Xu * function returns %NULL otherwise the new buffer is returned. 598e0053ec0SHerbert Xu * 599e0053ec0SHerbert Xu * If this function is called from an interrupt gfp_mask() must be 600e0053ec0SHerbert Xu * %GFP_ATOMIC. 601e0053ec0SHerbert Xu */ 602e0053ec0SHerbert Xu 603e0053ec0SHerbert Xu struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask) 604e0053ec0SHerbert Xu { 605e0053ec0SHerbert Xu struct sk_buff *n; 606e0053ec0SHerbert Xu 607e0053ec0SHerbert Xu n = skb + 1; 608e0053ec0SHerbert Xu if (skb->fclone == SKB_FCLONE_ORIG && 609e0053ec0SHerbert Xu n->fclone == SKB_FCLONE_UNAVAILABLE) { 610e0053ec0SHerbert Xu atomic_t *fclone_ref = (atomic_t *) (n + 1); 611e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_CLONE; 612e0053ec0SHerbert Xu atomic_inc(fclone_ref); 613e0053ec0SHerbert Xu } else { 614e0053ec0SHerbert Xu n = kmem_cache_alloc(skbuff_head_cache, gfp_mask); 615e0053ec0SHerbert Xu if (!n) 616e0053ec0SHerbert Xu return NULL; 617e0053ec0SHerbert Xu n->fclone = SKB_FCLONE_UNAVAILABLE; 618e0053ec0SHerbert Xu } 619e0053ec0SHerbert Xu 620e0053ec0SHerbert Xu return __skb_clone(n, skb); 6211da177e4SLinus Torvalds } 622b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_clone); 6231da177e4SLinus Torvalds 6241da177e4SLinus Torvalds static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old) 6251da177e4SLinus Torvalds { 6262e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6271da177e4SLinus Torvalds /* 6281da177e4SLinus Torvalds * Shift between the two data areas in bytes 6291da177e4SLinus Torvalds */ 6301da177e4SLinus Torvalds unsigned long offset = new->data - old->data; 6312e07fa9cSArnaldo Carvalho de Melo #endif 632dec18810SHerbert Xu 633dec18810SHerbert Xu __copy_skb_header(new, old); 634dec18810SHerbert Xu 6352e07fa9cSArnaldo Carvalho de Melo #ifndef NET_SKBUFF_DATA_USES_OFFSET 6362e07fa9cSArnaldo Carvalho de Melo /* {transport,network,mac}_header are relative to skb->head */ 6372e07fa9cSArnaldo Carvalho de Melo new->transport_header += offset; 6382e07fa9cSArnaldo Carvalho de Melo new->network_header += offset; 6392e07fa9cSArnaldo Carvalho de Melo new->mac_header += offset; 6402e07fa9cSArnaldo Carvalho de Melo #endif 6417967168cSHerbert Xu skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size; 6427967168cSHerbert Xu skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs; 6437967168cSHerbert Xu skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type; 6441da177e4SLinus Torvalds } 6451da177e4SLinus Torvalds 6461da177e4SLinus Torvalds /** 6471da177e4SLinus Torvalds * skb_copy - create private copy of an sk_buff 6481da177e4SLinus Torvalds * @skb: buffer to copy 6491da177e4SLinus Torvalds * @gfp_mask: allocation priority 6501da177e4SLinus Torvalds * 6511da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data. This is used when the 6521da177e4SLinus Torvalds * caller wishes to modify the data and needs a private copy of the 6531da177e4SLinus Torvalds * data to alter. Returns %NULL on failure or the pointer to the buffer 6541da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 6551da177e4SLinus Torvalds * 6561da177e4SLinus Torvalds * As by-product this function converts non-linear &sk_buff to linear 6571da177e4SLinus Torvalds * one, so that &sk_buff becomes completely private and caller is allowed 6581da177e4SLinus Torvalds * to modify all the data of returned buffer. This means that this 6591da177e4SLinus Torvalds * function is not recommended for use in circumstances when only 6601da177e4SLinus Torvalds * header is going to be modified. Use pskb_copy() instead. 6611da177e4SLinus Torvalds */ 6621da177e4SLinus Torvalds 663dd0fc66fSAl Viro struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask) 6641da177e4SLinus Torvalds { 6651da177e4SLinus Torvalds int headerlen = skb->data - skb->head; 6661da177e4SLinus Torvalds /* 6671da177e4SLinus Torvalds * Allocate the copy buffer 6681da177e4SLinus Torvalds */ 6694305b541SArnaldo Carvalho de Melo struct sk_buff *n; 6704305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 6714305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end + skb->data_len, gfp_mask); 6724305b541SArnaldo Carvalho de Melo #else 6734305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask); 6744305b541SArnaldo Carvalho de Melo #endif 6751da177e4SLinus Torvalds if (!n) 6761da177e4SLinus Torvalds return NULL; 6771da177e4SLinus Torvalds 6781da177e4SLinus Torvalds /* Set the data pointer */ 6791da177e4SLinus Torvalds skb_reserve(n, headerlen); 6801da177e4SLinus Torvalds /* Set the tail pointer and length */ 6811da177e4SLinus Torvalds skb_put(n, skb->len); 6821da177e4SLinus Torvalds 6831da177e4SLinus Torvalds if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len)) 6841da177e4SLinus Torvalds BUG(); 6851da177e4SLinus Torvalds 6861da177e4SLinus Torvalds copy_skb_header(n, skb); 6871da177e4SLinus Torvalds return n; 6881da177e4SLinus Torvalds } 689b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy); 6901da177e4SLinus Torvalds 6911da177e4SLinus Torvalds /** 6921da177e4SLinus Torvalds * pskb_copy - create copy of an sk_buff with private head. 6931da177e4SLinus Torvalds * @skb: buffer to copy 6941da177e4SLinus Torvalds * @gfp_mask: allocation priority 6951da177e4SLinus Torvalds * 6961da177e4SLinus Torvalds * Make a copy of both an &sk_buff and part of its data, located 6971da177e4SLinus Torvalds * in header. Fragmented data remain shared. This is used when 6981da177e4SLinus Torvalds * the caller wishes to modify only header of &sk_buff and needs 6991da177e4SLinus Torvalds * private copy of the header to alter. Returns %NULL on failure 7001da177e4SLinus Torvalds * or the pointer to the buffer on success. 7011da177e4SLinus Torvalds * The returned buffer has a reference count of 1. 7021da177e4SLinus Torvalds */ 7031da177e4SLinus Torvalds 704dd0fc66fSAl Viro struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask) 7051da177e4SLinus Torvalds { 7061da177e4SLinus Torvalds /* 7071da177e4SLinus Torvalds * Allocate the copy buffer 7081da177e4SLinus Torvalds */ 7094305b541SArnaldo Carvalho de Melo struct sk_buff *n; 7104305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 7114305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end, gfp_mask); 7124305b541SArnaldo Carvalho de Melo #else 7134305b541SArnaldo Carvalho de Melo n = alloc_skb(skb->end - skb->head, gfp_mask); 7144305b541SArnaldo Carvalho de Melo #endif 7151da177e4SLinus Torvalds if (!n) 7161da177e4SLinus Torvalds goto out; 7171da177e4SLinus Torvalds 7181da177e4SLinus Torvalds /* Set the data pointer */ 7191da177e4SLinus Torvalds skb_reserve(n, skb->data - skb->head); 7201da177e4SLinus Torvalds /* Set the tail pointer and length */ 7211da177e4SLinus Torvalds skb_put(n, skb_headlen(skb)); 7221da177e4SLinus Torvalds /* Copy the bytes */ 723d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, n->data, n->len); 7241da177e4SLinus Torvalds 72525f484a6SHerbert Xu n->truesize += skb->data_len; 7261da177e4SLinus Torvalds n->data_len = skb->data_len; 7271da177e4SLinus Torvalds n->len = skb->len; 7281da177e4SLinus Torvalds 7291da177e4SLinus Torvalds if (skb_shinfo(skb)->nr_frags) { 7301da177e4SLinus Torvalds int i; 7311da177e4SLinus Torvalds 7321da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 7331da177e4SLinus Torvalds skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i]; 7341da177e4SLinus Torvalds get_page(skb_shinfo(n)->frags[i].page); 7351da177e4SLinus Torvalds } 7361da177e4SLinus Torvalds skb_shinfo(n)->nr_frags = i; 7371da177e4SLinus Torvalds } 7381da177e4SLinus Torvalds 7391da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 7401da177e4SLinus Torvalds skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list; 7411da177e4SLinus Torvalds skb_clone_fraglist(n); 7421da177e4SLinus Torvalds } 7431da177e4SLinus Torvalds 7441da177e4SLinus Torvalds copy_skb_header(n, skb); 7451da177e4SLinus Torvalds out: 7461da177e4SLinus Torvalds return n; 7471da177e4SLinus Torvalds } 748b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_copy); 7491da177e4SLinus Torvalds 7501da177e4SLinus Torvalds /** 7511da177e4SLinus Torvalds * pskb_expand_head - reallocate header of &sk_buff 7521da177e4SLinus Torvalds * @skb: buffer to reallocate 7531da177e4SLinus Torvalds * @nhead: room to add at head 7541da177e4SLinus Torvalds * @ntail: room to add at tail 7551da177e4SLinus Torvalds * @gfp_mask: allocation priority 7561da177e4SLinus Torvalds * 7571da177e4SLinus Torvalds * Expands (or creates identical copy, if &nhead and &ntail are zero) 7581da177e4SLinus Torvalds * header of skb. &sk_buff itself is not changed. &sk_buff MUST have 7591da177e4SLinus Torvalds * reference count of 1. Returns zero in the case of success or error, 7601da177e4SLinus Torvalds * if expansion failed. In the last case, &sk_buff is not changed. 7611da177e4SLinus Torvalds * 7621da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 7631da177e4SLinus Torvalds * reloaded after call to this function. 7641da177e4SLinus Torvalds */ 7651da177e4SLinus Torvalds 76686a76cafSVictor Fusco int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, 767dd0fc66fSAl Viro gfp_t gfp_mask) 7681da177e4SLinus Torvalds { 7691da177e4SLinus Torvalds int i; 7701da177e4SLinus Torvalds u8 *data; 7714305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 7724305b541SArnaldo Carvalho de Melo int size = nhead + skb->end + ntail; 7734305b541SArnaldo Carvalho de Melo #else 7741da177e4SLinus Torvalds int size = nhead + (skb->end - skb->head) + ntail; 7754305b541SArnaldo Carvalho de Melo #endif 7761da177e4SLinus Torvalds long off; 7771da177e4SLinus Torvalds 7784edd87adSHerbert Xu BUG_ON(nhead < 0); 7794edd87adSHerbert Xu 7801da177e4SLinus Torvalds if (skb_shared(skb)) 7811da177e4SLinus Torvalds BUG(); 7821da177e4SLinus Torvalds 7831da177e4SLinus Torvalds size = SKB_DATA_ALIGN(size); 7841da177e4SLinus Torvalds 7851da177e4SLinus Torvalds data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask); 7861da177e4SLinus Torvalds if (!data) 7871da177e4SLinus Torvalds goto nodata; 7881da177e4SLinus Torvalds 7891da177e4SLinus Torvalds /* Copy only real data... and, alas, header. This should be 7901da177e4SLinus Torvalds * optimized for the cases when header is void. */ 7914305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 792b6ccc67dSMikael Pettersson memcpy(data + nhead, skb->head, skb->tail); 7934305b541SArnaldo Carvalho de Melo #else 794b6ccc67dSMikael Pettersson memcpy(data + nhead, skb->head, skb->tail - skb->head); 79527a884dcSArnaldo Carvalho de Melo #endif 7964305b541SArnaldo Carvalho de Melo memcpy(data + size, skb_end_pointer(skb), 7974305b541SArnaldo Carvalho de Melo sizeof(struct skb_shared_info)); 7981da177e4SLinus Torvalds 7991da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 8001da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 8011da177e4SLinus Torvalds 8021da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) 8031da177e4SLinus Torvalds skb_clone_fraglist(skb); 8041da177e4SLinus Torvalds 8051da177e4SLinus Torvalds skb_release_data(skb); 8061da177e4SLinus Torvalds 8071da177e4SLinus Torvalds off = (data + nhead) - skb->head; 8081da177e4SLinus Torvalds 8091da177e4SLinus Torvalds skb->head = data; 8101da177e4SLinus Torvalds skb->data += off; 8114305b541SArnaldo Carvalho de Melo #ifdef NET_SKBUFF_DATA_USES_OFFSET 8124305b541SArnaldo Carvalho de Melo skb->end = size; 81356eb8882SPatrick McHardy off = nhead; 8144305b541SArnaldo Carvalho de Melo #else 8154305b541SArnaldo Carvalho de Melo skb->end = skb->head + size; 81656eb8882SPatrick McHardy #endif 81727a884dcSArnaldo Carvalho de Melo /* {transport,network,mac}_header and tail are relative to skb->head */ 81827a884dcSArnaldo Carvalho de Melo skb->tail += off; 819b0e380b1SArnaldo Carvalho de Melo skb->transport_header += off; 820b0e380b1SArnaldo Carvalho de Melo skb->network_header += off; 821b0e380b1SArnaldo Carvalho de Melo skb->mac_header += off; 822172a863fSHerbert Xu skb->csum_start += nhead; 8231da177e4SLinus Torvalds skb->cloned = 0; 824334a8132SPatrick McHardy skb->hdr_len = 0; 8251da177e4SLinus Torvalds skb->nohdr = 0; 8261da177e4SLinus Torvalds atomic_set(&skb_shinfo(skb)->dataref, 1); 8271da177e4SLinus Torvalds return 0; 8281da177e4SLinus Torvalds 8291da177e4SLinus Torvalds nodata: 8301da177e4SLinus Torvalds return -ENOMEM; 8311da177e4SLinus Torvalds } 832b4ac530fSDavid S. Miller EXPORT_SYMBOL(pskb_expand_head); 8331da177e4SLinus Torvalds 8341da177e4SLinus Torvalds /* Make private copy of skb with writable head and some headroom */ 8351da177e4SLinus Torvalds 8361da177e4SLinus Torvalds struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom) 8371da177e4SLinus Torvalds { 8381da177e4SLinus Torvalds struct sk_buff *skb2; 8391da177e4SLinus Torvalds int delta = headroom - skb_headroom(skb); 8401da177e4SLinus Torvalds 8411da177e4SLinus Torvalds if (delta <= 0) 8421da177e4SLinus Torvalds skb2 = pskb_copy(skb, GFP_ATOMIC); 8431da177e4SLinus Torvalds else { 8441da177e4SLinus Torvalds skb2 = skb_clone(skb, GFP_ATOMIC); 8451da177e4SLinus Torvalds if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0, 8461da177e4SLinus Torvalds GFP_ATOMIC)) { 8471da177e4SLinus Torvalds kfree_skb(skb2); 8481da177e4SLinus Torvalds skb2 = NULL; 8491da177e4SLinus Torvalds } 8501da177e4SLinus Torvalds } 8511da177e4SLinus Torvalds return skb2; 8521da177e4SLinus Torvalds } 853b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_realloc_headroom); 8541da177e4SLinus Torvalds 8551da177e4SLinus Torvalds /** 8561da177e4SLinus Torvalds * skb_copy_expand - copy and expand sk_buff 8571da177e4SLinus Torvalds * @skb: buffer to copy 8581da177e4SLinus Torvalds * @newheadroom: new free bytes at head 8591da177e4SLinus Torvalds * @newtailroom: new free bytes at tail 8601da177e4SLinus Torvalds * @gfp_mask: allocation priority 8611da177e4SLinus Torvalds * 8621da177e4SLinus Torvalds * Make a copy of both an &sk_buff and its data and while doing so 8631da177e4SLinus Torvalds * allocate additional space. 8641da177e4SLinus Torvalds * 8651da177e4SLinus Torvalds * This is used when the caller wishes to modify the data and needs a 8661da177e4SLinus Torvalds * private copy of the data to alter as well as more space for new fields. 8671da177e4SLinus Torvalds * Returns %NULL on failure or the pointer to the buffer 8681da177e4SLinus Torvalds * on success. The returned buffer has a reference count of 1. 8691da177e4SLinus Torvalds * 8701da177e4SLinus Torvalds * You must pass %GFP_ATOMIC as the allocation priority if this function 8711da177e4SLinus Torvalds * is called from an interrupt. 8721da177e4SLinus Torvalds */ 8731da177e4SLinus Torvalds struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 87486a76cafSVictor Fusco int newheadroom, int newtailroom, 875dd0fc66fSAl Viro gfp_t gfp_mask) 8761da177e4SLinus Torvalds { 8771da177e4SLinus Torvalds /* 8781da177e4SLinus Torvalds * Allocate the copy buffer 8791da177e4SLinus Torvalds */ 8801da177e4SLinus Torvalds struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom, 8811da177e4SLinus Torvalds gfp_mask); 882efd1e8d5SPatrick McHardy int oldheadroom = skb_headroom(skb); 8831da177e4SLinus Torvalds int head_copy_len, head_copy_off; 88452886051SHerbert Xu int off; 8851da177e4SLinus Torvalds 8861da177e4SLinus Torvalds if (!n) 8871da177e4SLinus Torvalds return NULL; 8881da177e4SLinus Torvalds 8891da177e4SLinus Torvalds skb_reserve(n, newheadroom); 8901da177e4SLinus Torvalds 8911da177e4SLinus Torvalds /* Set the tail pointer and length */ 8921da177e4SLinus Torvalds skb_put(n, skb->len); 8931da177e4SLinus Torvalds 894efd1e8d5SPatrick McHardy head_copy_len = oldheadroom; 8951da177e4SLinus Torvalds head_copy_off = 0; 8961da177e4SLinus Torvalds if (newheadroom <= head_copy_len) 8971da177e4SLinus Torvalds head_copy_len = newheadroom; 8981da177e4SLinus Torvalds else 8991da177e4SLinus Torvalds head_copy_off = newheadroom - head_copy_len; 9001da177e4SLinus Torvalds 9011da177e4SLinus Torvalds /* Copy the linear header and data. */ 9021da177e4SLinus Torvalds if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off, 9031da177e4SLinus Torvalds skb->len + head_copy_len)) 9041da177e4SLinus Torvalds BUG(); 9051da177e4SLinus Torvalds 9061da177e4SLinus Torvalds copy_skb_header(n, skb); 9071da177e4SLinus Torvalds 908efd1e8d5SPatrick McHardy off = newheadroom - oldheadroom; 90952886051SHerbert Xu n->csum_start += off; 91052886051SHerbert Xu #ifdef NET_SKBUFF_DATA_USES_OFFSET 911efd1e8d5SPatrick McHardy n->transport_header += off; 912efd1e8d5SPatrick McHardy n->network_header += off; 913efd1e8d5SPatrick McHardy n->mac_header += off; 91452886051SHerbert Xu #endif 915efd1e8d5SPatrick McHardy 9161da177e4SLinus Torvalds return n; 9171da177e4SLinus Torvalds } 918b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_expand); 9191da177e4SLinus Torvalds 9201da177e4SLinus Torvalds /** 9211da177e4SLinus Torvalds * skb_pad - zero pad the tail of an skb 9221da177e4SLinus Torvalds * @skb: buffer to pad 9231da177e4SLinus Torvalds * @pad: space to pad 9241da177e4SLinus Torvalds * 9251da177e4SLinus Torvalds * Ensure that a buffer is followed by a padding area that is zero 9261da177e4SLinus Torvalds * filled. Used by network drivers which may DMA or transfer data 9271da177e4SLinus Torvalds * beyond the buffer end onto the wire. 9281da177e4SLinus Torvalds * 9295b057c6bSHerbert Xu * May return error in out of memory cases. The skb is freed on error. 9301da177e4SLinus Torvalds */ 9311da177e4SLinus Torvalds 9325b057c6bSHerbert Xu int skb_pad(struct sk_buff *skb, int pad) 9331da177e4SLinus Torvalds { 9345b057c6bSHerbert Xu int err; 9355b057c6bSHerbert Xu int ntail; 9361da177e4SLinus Torvalds 9371da177e4SLinus Torvalds /* If the skbuff is non linear tailroom is always zero.. */ 9385b057c6bSHerbert Xu if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) { 9391da177e4SLinus Torvalds memset(skb->data+skb->len, 0, pad); 9405b057c6bSHerbert Xu return 0; 9411da177e4SLinus Torvalds } 9421da177e4SLinus Torvalds 9434305b541SArnaldo Carvalho de Melo ntail = skb->data_len + pad - (skb->end - skb->tail); 9445b057c6bSHerbert Xu if (likely(skb_cloned(skb) || ntail > 0)) { 9455b057c6bSHerbert Xu err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC); 9465b057c6bSHerbert Xu if (unlikely(err)) 9475b057c6bSHerbert Xu goto free_skb; 9485b057c6bSHerbert Xu } 9495b057c6bSHerbert Xu 9505b057c6bSHerbert Xu /* FIXME: The use of this function with non-linear skb's really needs 9515b057c6bSHerbert Xu * to be audited. 9525b057c6bSHerbert Xu */ 9535b057c6bSHerbert Xu err = skb_linearize(skb); 9545b057c6bSHerbert Xu if (unlikely(err)) 9555b057c6bSHerbert Xu goto free_skb; 9565b057c6bSHerbert Xu 9575b057c6bSHerbert Xu memset(skb->data + skb->len, 0, pad); 9585b057c6bSHerbert Xu return 0; 9595b057c6bSHerbert Xu 9605b057c6bSHerbert Xu free_skb: 9611da177e4SLinus Torvalds kfree_skb(skb); 9625b057c6bSHerbert Xu return err; 9631da177e4SLinus Torvalds } 964b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_pad); 9651da177e4SLinus Torvalds 9660dde3e16SIlpo Järvinen /** 9670dde3e16SIlpo Järvinen * skb_put - add data to a buffer 9680dde3e16SIlpo Järvinen * @skb: buffer to use 9690dde3e16SIlpo Järvinen * @len: amount of data to add 9700dde3e16SIlpo Järvinen * 9710dde3e16SIlpo Järvinen * This function extends the used data area of the buffer. If this would 9720dde3e16SIlpo Järvinen * exceed the total buffer size the kernel will panic. A pointer to the 9730dde3e16SIlpo Järvinen * first byte of the extra data is returned. 9740dde3e16SIlpo Järvinen */ 9750dde3e16SIlpo Järvinen unsigned char *skb_put(struct sk_buff *skb, unsigned int len) 9760dde3e16SIlpo Järvinen { 9770dde3e16SIlpo Järvinen unsigned char *tmp = skb_tail_pointer(skb); 9780dde3e16SIlpo Järvinen SKB_LINEAR_ASSERT(skb); 9790dde3e16SIlpo Järvinen skb->tail += len; 9800dde3e16SIlpo Järvinen skb->len += len; 9810dde3e16SIlpo Järvinen if (unlikely(skb->tail > skb->end)) 9820dde3e16SIlpo Järvinen skb_over_panic(skb, len, __builtin_return_address(0)); 9830dde3e16SIlpo Järvinen return tmp; 9840dde3e16SIlpo Järvinen } 9850dde3e16SIlpo Järvinen EXPORT_SYMBOL(skb_put); 9860dde3e16SIlpo Järvinen 9876be8ac2fSIlpo Järvinen /** 988c2aa270aSIlpo Järvinen * skb_push - add data to the start of a buffer 989c2aa270aSIlpo Järvinen * @skb: buffer to use 990c2aa270aSIlpo Järvinen * @len: amount of data to add 991c2aa270aSIlpo Järvinen * 992c2aa270aSIlpo Järvinen * This function extends the used data area of the buffer at the buffer 993c2aa270aSIlpo Järvinen * start. If this would exceed the total buffer headroom the kernel will 994c2aa270aSIlpo Järvinen * panic. A pointer to the first byte of the extra data is returned. 995c2aa270aSIlpo Järvinen */ 996c2aa270aSIlpo Järvinen unsigned char *skb_push(struct sk_buff *skb, unsigned int len) 997c2aa270aSIlpo Järvinen { 998c2aa270aSIlpo Järvinen skb->data -= len; 999c2aa270aSIlpo Järvinen skb->len += len; 1000c2aa270aSIlpo Järvinen if (unlikely(skb->data<skb->head)) 1001c2aa270aSIlpo Järvinen skb_under_panic(skb, len, __builtin_return_address(0)); 1002c2aa270aSIlpo Järvinen return skb->data; 1003c2aa270aSIlpo Järvinen } 1004c2aa270aSIlpo Järvinen EXPORT_SYMBOL(skb_push); 1005c2aa270aSIlpo Järvinen 1006c2aa270aSIlpo Järvinen /** 10076be8ac2fSIlpo Järvinen * skb_pull - remove data from the start of a buffer 10086be8ac2fSIlpo Järvinen * @skb: buffer to use 10096be8ac2fSIlpo Järvinen * @len: amount of data to remove 10106be8ac2fSIlpo Järvinen * 10116be8ac2fSIlpo Järvinen * This function removes data from the start of a buffer, returning 10126be8ac2fSIlpo Järvinen * the memory to the headroom. A pointer to the next data in the buffer 10136be8ac2fSIlpo Järvinen * is returned. Once the data has been pulled future pushes will overwrite 10146be8ac2fSIlpo Järvinen * the old data. 10156be8ac2fSIlpo Järvinen */ 10166be8ac2fSIlpo Järvinen unsigned char *skb_pull(struct sk_buff *skb, unsigned int len) 10176be8ac2fSIlpo Järvinen { 10186be8ac2fSIlpo Järvinen return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 10196be8ac2fSIlpo Järvinen } 10206be8ac2fSIlpo Järvinen EXPORT_SYMBOL(skb_pull); 10216be8ac2fSIlpo Järvinen 1022419ae74eSIlpo Järvinen /** 1023419ae74eSIlpo Järvinen * skb_trim - remove end from a buffer 1024419ae74eSIlpo Järvinen * @skb: buffer to alter 1025419ae74eSIlpo Järvinen * @len: new length 1026419ae74eSIlpo Järvinen * 1027419ae74eSIlpo Järvinen * Cut the length of a buffer down by removing data from the tail. If 1028419ae74eSIlpo Järvinen * the buffer is already under the length specified it is not modified. 1029419ae74eSIlpo Järvinen * The skb must be linear. 1030419ae74eSIlpo Järvinen */ 1031419ae74eSIlpo Järvinen void skb_trim(struct sk_buff *skb, unsigned int len) 1032419ae74eSIlpo Järvinen { 1033419ae74eSIlpo Järvinen if (skb->len > len) 1034419ae74eSIlpo Järvinen __skb_trim(skb, len); 1035419ae74eSIlpo Järvinen } 1036419ae74eSIlpo Järvinen EXPORT_SYMBOL(skb_trim); 1037419ae74eSIlpo Järvinen 10383cc0e873SHerbert Xu /* Trims skb to length len. It can change skb pointers. 10391da177e4SLinus Torvalds */ 10401da177e4SLinus Torvalds 10413cc0e873SHerbert Xu int ___pskb_trim(struct sk_buff *skb, unsigned int len) 10421da177e4SLinus Torvalds { 104327b437c8SHerbert Xu struct sk_buff **fragp; 104427b437c8SHerbert Xu struct sk_buff *frag; 10451da177e4SLinus Torvalds int offset = skb_headlen(skb); 10461da177e4SLinus Torvalds int nfrags = skb_shinfo(skb)->nr_frags; 10471da177e4SLinus Torvalds int i; 104827b437c8SHerbert Xu int err; 104927b437c8SHerbert Xu 105027b437c8SHerbert Xu if (skb_cloned(skb) && 105127b437c8SHerbert Xu unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC)))) 105227b437c8SHerbert Xu return err; 10531da177e4SLinus Torvalds 1054f4d26fb3SHerbert Xu i = 0; 1055f4d26fb3SHerbert Xu if (offset >= len) 1056f4d26fb3SHerbert Xu goto drop_pages; 1057f4d26fb3SHerbert Xu 1058f4d26fb3SHerbert Xu for (; i < nfrags; i++) { 10591da177e4SLinus Torvalds int end = offset + skb_shinfo(skb)->frags[i].size; 106027b437c8SHerbert Xu 106127b437c8SHerbert Xu if (end < len) { 10621da177e4SLinus Torvalds offset = end; 106327b437c8SHerbert Xu continue; 10641da177e4SLinus Torvalds } 10651da177e4SLinus Torvalds 106627b437c8SHerbert Xu skb_shinfo(skb)->frags[i++].size = len - offset; 106727b437c8SHerbert Xu 1068f4d26fb3SHerbert Xu drop_pages: 106927b437c8SHerbert Xu skb_shinfo(skb)->nr_frags = i; 107027b437c8SHerbert Xu 107127b437c8SHerbert Xu for (; i < nfrags; i++) 107227b437c8SHerbert Xu put_page(skb_shinfo(skb)->frags[i].page); 107327b437c8SHerbert Xu 107427b437c8SHerbert Xu if (skb_shinfo(skb)->frag_list) 107527b437c8SHerbert Xu skb_drop_fraglist(skb); 1076f4d26fb3SHerbert Xu goto done; 107727b437c8SHerbert Xu } 107827b437c8SHerbert Xu 107927b437c8SHerbert Xu for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp); 108027b437c8SHerbert Xu fragp = &frag->next) { 108127b437c8SHerbert Xu int end = offset + frag->len; 108227b437c8SHerbert Xu 108327b437c8SHerbert Xu if (skb_shared(frag)) { 108427b437c8SHerbert Xu struct sk_buff *nfrag; 108527b437c8SHerbert Xu 108627b437c8SHerbert Xu nfrag = skb_clone(frag, GFP_ATOMIC); 108727b437c8SHerbert Xu if (unlikely(!nfrag)) 108827b437c8SHerbert Xu return -ENOMEM; 108927b437c8SHerbert Xu 109027b437c8SHerbert Xu nfrag->next = frag->next; 1091f4d26fb3SHerbert Xu kfree_skb(frag); 109227b437c8SHerbert Xu frag = nfrag; 109327b437c8SHerbert Xu *fragp = frag; 109427b437c8SHerbert Xu } 109527b437c8SHerbert Xu 109627b437c8SHerbert Xu if (end < len) { 109727b437c8SHerbert Xu offset = end; 109827b437c8SHerbert Xu continue; 109927b437c8SHerbert Xu } 110027b437c8SHerbert Xu 110127b437c8SHerbert Xu if (end > len && 110227b437c8SHerbert Xu unlikely((err = pskb_trim(frag, len - offset)))) 110327b437c8SHerbert Xu return err; 110427b437c8SHerbert Xu 110527b437c8SHerbert Xu if (frag->next) 110627b437c8SHerbert Xu skb_drop_list(&frag->next); 110727b437c8SHerbert Xu break; 110827b437c8SHerbert Xu } 110927b437c8SHerbert Xu 1110f4d26fb3SHerbert Xu done: 111127b437c8SHerbert Xu if (len > skb_headlen(skb)) { 11121da177e4SLinus Torvalds skb->data_len -= skb->len - len; 11131da177e4SLinus Torvalds skb->len = len; 11141da177e4SLinus Torvalds } else { 11151da177e4SLinus Torvalds skb->len = len; 11161da177e4SLinus Torvalds skb->data_len = 0; 111727a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 11181da177e4SLinus Torvalds } 11191da177e4SLinus Torvalds 11201da177e4SLinus Torvalds return 0; 11211da177e4SLinus Torvalds } 1122b4ac530fSDavid S. Miller EXPORT_SYMBOL(___pskb_trim); 11231da177e4SLinus Torvalds 11241da177e4SLinus Torvalds /** 11251da177e4SLinus Torvalds * __pskb_pull_tail - advance tail of skb header 11261da177e4SLinus Torvalds * @skb: buffer to reallocate 11271da177e4SLinus Torvalds * @delta: number of bytes to advance tail 11281da177e4SLinus Torvalds * 11291da177e4SLinus Torvalds * The function makes a sense only on a fragmented &sk_buff, 11301da177e4SLinus Torvalds * it expands header moving its tail forward and copying necessary 11311da177e4SLinus Torvalds * data from fragmented part. 11321da177e4SLinus Torvalds * 11331da177e4SLinus Torvalds * &sk_buff MUST have reference count of 1. 11341da177e4SLinus Torvalds * 11351da177e4SLinus Torvalds * Returns %NULL (and &sk_buff does not change) if pull failed 11361da177e4SLinus Torvalds * or value of new tail of skb in the case of success. 11371da177e4SLinus Torvalds * 11381da177e4SLinus Torvalds * All the pointers pointing into skb header may change and must be 11391da177e4SLinus Torvalds * reloaded after call to this function. 11401da177e4SLinus Torvalds */ 11411da177e4SLinus Torvalds 11421da177e4SLinus Torvalds /* Moves tail of skb head forward, copying data from fragmented part, 11431da177e4SLinus Torvalds * when it is necessary. 11441da177e4SLinus Torvalds * 1. It may fail due to malloc failure. 11451da177e4SLinus Torvalds * 2. It may change skb pointers. 11461da177e4SLinus Torvalds * 11471da177e4SLinus Torvalds * It is pretty complicated. Luckily, it is called only in exceptional cases. 11481da177e4SLinus Torvalds */ 11491da177e4SLinus Torvalds unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta) 11501da177e4SLinus Torvalds { 11511da177e4SLinus Torvalds /* If skb has not enough free space at tail, get new one 11521da177e4SLinus Torvalds * plus 128 bytes for future expansions. If we have enough 11531da177e4SLinus Torvalds * room at tail, reallocate without expansion only if skb is cloned. 11541da177e4SLinus Torvalds */ 11554305b541SArnaldo Carvalho de Melo int i, k, eat = (skb->tail + delta) - skb->end; 11561da177e4SLinus Torvalds 11571da177e4SLinus Torvalds if (eat > 0 || skb_cloned(skb)) { 11581da177e4SLinus Torvalds if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0, 11591da177e4SLinus Torvalds GFP_ATOMIC)) 11601da177e4SLinus Torvalds return NULL; 11611da177e4SLinus Torvalds } 11621da177e4SLinus Torvalds 116327a884dcSArnaldo Carvalho de Melo if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta)) 11641da177e4SLinus Torvalds BUG(); 11651da177e4SLinus Torvalds 11661da177e4SLinus Torvalds /* Optimization: no fragments, no reasons to preestimate 11671da177e4SLinus Torvalds * size of pulled pages. Superb. 11681da177e4SLinus Torvalds */ 11691da177e4SLinus Torvalds if (!skb_shinfo(skb)->frag_list) 11701da177e4SLinus Torvalds goto pull_pages; 11711da177e4SLinus Torvalds 11721da177e4SLinus Torvalds /* Estimate size of pulled pages. */ 11731da177e4SLinus Torvalds eat = delta; 11741da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 11751da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size >= eat) 11761da177e4SLinus Torvalds goto pull_pages; 11771da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 11781da177e4SLinus Torvalds } 11791da177e4SLinus Torvalds 11801da177e4SLinus Torvalds /* If we need update frag list, we are in troubles. 11811da177e4SLinus Torvalds * Certainly, it possible to add an offset to skb data, 11821da177e4SLinus Torvalds * but taking into account that pulling is expected to 11831da177e4SLinus Torvalds * be very rare operation, it is worth to fight against 11841da177e4SLinus Torvalds * further bloating skb head and crucify ourselves here instead. 11851da177e4SLinus Torvalds * Pure masohism, indeed. 8)8) 11861da177e4SLinus Torvalds */ 11871da177e4SLinus Torvalds if (eat) { 11881da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 11891da177e4SLinus Torvalds struct sk_buff *clone = NULL; 11901da177e4SLinus Torvalds struct sk_buff *insp = NULL; 11911da177e4SLinus Torvalds 11921da177e4SLinus Torvalds do { 119309a62660SKris Katterjohn BUG_ON(!list); 11941da177e4SLinus Torvalds 11951da177e4SLinus Torvalds if (list->len <= eat) { 11961da177e4SLinus Torvalds /* Eaten as whole. */ 11971da177e4SLinus Torvalds eat -= list->len; 11981da177e4SLinus Torvalds list = list->next; 11991da177e4SLinus Torvalds insp = list; 12001da177e4SLinus Torvalds } else { 12011da177e4SLinus Torvalds /* Eaten partially. */ 12021da177e4SLinus Torvalds 12031da177e4SLinus Torvalds if (skb_shared(list)) { 12041da177e4SLinus Torvalds /* Sucks! We need to fork list. :-( */ 12051da177e4SLinus Torvalds clone = skb_clone(list, GFP_ATOMIC); 12061da177e4SLinus Torvalds if (!clone) 12071da177e4SLinus Torvalds return NULL; 12081da177e4SLinus Torvalds insp = list->next; 12091da177e4SLinus Torvalds list = clone; 12101da177e4SLinus Torvalds } else { 12111da177e4SLinus Torvalds /* This may be pulled without 12121da177e4SLinus Torvalds * problems. */ 12131da177e4SLinus Torvalds insp = list; 12141da177e4SLinus Torvalds } 12151da177e4SLinus Torvalds if (!pskb_pull(list, eat)) { 12161da177e4SLinus Torvalds if (clone) 12171da177e4SLinus Torvalds kfree_skb(clone); 12181da177e4SLinus Torvalds return NULL; 12191da177e4SLinus Torvalds } 12201da177e4SLinus Torvalds break; 12211da177e4SLinus Torvalds } 12221da177e4SLinus Torvalds } while (eat); 12231da177e4SLinus Torvalds 12241da177e4SLinus Torvalds /* Free pulled out fragments. */ 12251da177e4SLinus Torvalds while ((list = skb_shinfo(skb)->frag_list) != insp) { 12261da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = list->next; 12271da177e4SLinus Torvalds kfree_skb(list); 12281da177e4SLinus Torvalds } 12291da177e4SLinus Torvalds /* And insert new clone at head. */ 12301da177e4SLinus Torvalds if (clone) { 12311da177e4SLinus Torvalds clone->next = list; 12321da177e4SLinus Torvalds skb_shinfo(skb)->frag_list = clone; 12331da177e4SLinus Torvalds } 12341da177e4SLinus Torvalds } 12351da177e4SLinus Torvalds /* Success! Now we may commit changes to skb data. */ 12361da177e4SLinus Torvalds 12371da177e4SLinus Torvalds pull_pages: 12381da177e4SLinus Torvalds eat = delta; 12391da177e4SLinus Torvalds k = 0; 12401da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12411da177e4SLinus Torvalds if (skb_shinfo(skb)->frags[i].size <= eat) { 12421da177e4SLinus Torvalds put_page(skb_shinfo(skb)->frags[i].page); 12431da177e4SLinus Torvalds eat -= skb_shinfo(skb)->frags[i].size; 12441da177e4SLinus Torvalds } else { 12451da177e4SLinus Torvalds skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i]; 12461da177e4SLinus Torvalds if (eat) { 12471da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].page_offset += eat; 12481da177e4SLinus Torvalds skb_shinfo(skb)->frags[k].size -= eat; 12491da177e4SLinus Torvalds eat = 0; 12501da177e4SLinus Torvalds } 12511da177e4SLinus Torvalds k++; 12521da177e4SLinus Torvalds } 12531da177e4SLinus Torvalds } 12541da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = k; 12551da177e4SLinus Torvalds 12561da177e4SLinus Torvalds skb->tail += delta; 12571da177e4SLinus Torvalds skb->data_len -= delta; 12581da177e4SLinus Torvalds 125927a884dcSArnaldo Carvalho de Melo return skb_tail_pointer(skb); 12601da177e4SLinus Torvalds } 1261b4ac530fSDavid S. Miller EXPORT_SYMBOL(__pskb_pull_tail); 12621da177e4SLinus Torvalds 12631da177e4SLinus Torvalds /* Copy some data bits from skb to kernel buffer. */ 12641da177e4SLinus Torvalds 12651da177e4SLinus Torvalds int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len) 12661da177e4SLinus Torvalds { 12671da177e4SLinus Torvalds int i, copy; 12681a028e50SDavid S. Miller int start = skb_headlen(skb); 12691da177e4SLinus Torvalds 12701da177e4SLinus Torvalds if (offset > (int)skb->len - len) 12711da177e4SLinus Torvalds goto fault; 12721da177e4SLinus Torvalds 12731da177e4SLinus Torvalds /* Copy header. */ 12741a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 12751da177e4SLinus Torvalds if (copy > len) 12761da177e4SLinus Torvalds copy = len; 1277d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, to, copy); 12781da177e4SLinus Torvalds if ((len -= copy) == 0) 12791da177e4SLinus Torvalds return 0; 12801da177e4SLinus Torvalds offset += copy; 12811da177e4SLinus Torvalds to += copy; 12821da177e4SLinus Torvalds } 12831da177e4SLinus Torvalds 12841da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 12851a028e50SDavid S. Miller int end; 12861da177e4SLinus Torvalds 1287547b792cSIlpo Järvinen WARN_ON(start > offset + len); 12881a028e50SDavid S. Miller 12891a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 12901da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 12911da177e4SLinus Torvalds u8 *vaddr; 12921da177e4SLinus Torvalds 12931da177e4SLinus Torvalds if (copy > len) 12941da177e4SLinus Torvalds copy = len; 12951da177e4SLinus Torvalds 12961da177e4SLinus Torvalds vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]); 12971da177e4SLinus Torvalds memcpy(to, 12981a028e50SDavid S. Miller vaddr + skb_shinfo(skb)->frags[i].page_offset+ 12991a028e50SDavid S. Miller offset - start, copy); 13001da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 13011da177e4SLinus Torvalds 13021da177e4SLinus Torvalds if ((len -= copy) == 0) 13031da177e4SLinus Torvalds return 0; 13041da177e4SLinus Torvalds offset += copy; 13051da177e4SLinus Torvalds to += copy; 13061da177e4SLinus Torvalds } 13071a028e50SDavid S. Miller start = end; 13081da177e4SLinus Torvalds } 13091da177e4SLinus Torvalds 13101da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 13111da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 13121da177e4SLinus Torvalds 13131da177e4SLinus Torvalds for (; list; list = list->next) { 13141a028e50SDavid S. Miller int end; 13151da177e4SLinus Torvalds 1316547b792cSIlpo Järvinen WARN_ON(start > offset + len); 13171a028e50SDavid S. Miller 13181a028e50SDavid S. Miller end = start + list->len; 13191da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 13201da177e4SLinus Torvalds if (copy > len) 13211da177e4SLinus Torvalds copy = len; 13221a028e50SDavid S. Miller if (skb_copy_bits(list, offset - start, 13231a028e50SDavid S. Miller to, copy)) 13241da177e4SLinus Torvalds goto fault; 13251da177e4SLinus Torvalds if ((len -= copy) == 0) 13261da177e4SLinus Torvalds return 0; 13271da177e4SLinus Torvalds offset += copy; 13281da177e4SLinus Torvalds to += copy; 13291da177e4SLinus Torvalds } 13301a028e50SDavid S. Miller start = end; 13311da177e4SLinus Torvalds } 13321da177e4SLinus Torvalds } 13331da177e4SLinus Torvalds if (!len) 13341da177e4SLinus Torvalds return 0; 13351da177e4SLinus Torvalds 13361da177e4SLinus Torvalds fault: 13371da177e4SLinus Torvalds return -EFAULT; 13381da177e4SLinus Torvalds } 1339b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_bits); 13401da177e4SLinus Torvalds 13419c55e01cSJens Axboe /* 13429c55e01cSJens Axboe * Callback from splice_to_pipe(), if we need to release some pages 13439c55e01cSJens Axboe * at the end of the spd in case we error'ed out in filling the pipe. 13449c55e01cSJens Axboe */ 13459c55e01cSJens Axboe static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i) 13469c55e01cSJens Axboe { 13478b9d3728SJarek Poplawski put_page(spd->pages[i]); 13488b9d3728SJarek Poplawski } 13499c55e01cSJens Axboe 13504fb66994SJarek Poplawski static inline struct page *linear_to_page(struct page *page, unsigned int *len, 13514fb66994SJarek Poplawski unsigned int *offset, 13524fb66994SJarek Poplawski struct sk_buff *skb) 13538b9d3728SJarek Poplawski { 13544fb66994SJarek Poplawski struct sock *sk = skb->sk; 13554fb66994SJarek Poplawski struct page *p = sk->sk_sndmsg_page; 13564fb66994SJarek Poplawski unsigned int off; 13578b9d3728SJarek Poplawski 13584fb66994SJarek Poplawski if (!p) { 13594fb66994SJarek Poplawski new_page: 13604fb66994SJarek Poplawski p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0); 13618b9d3728SJarek Poplawski if (!p) 13628b9d3728SJarek Poplawski return NULL; 13634fb66994SJarek Poplawski 13644fb66994SJarek Poplawski off = sk->sk_sndmsg_off = 0; 13654fb66994SJarek Poplawski /* hold one ref to this page until it's full */ 13664fb66994SJarek Poplawski } else { 13674fb66994SJarek Poplawski unsigned int mlen; 13684fb66994SJarek Poplawski 13694fb66994SJarek Poplawski off = sk->sk_sndmsg_off; 13704fb66994SJarek Poplawski mlen = PAGE_SIZE - off; 13714fb66994SJarek Poplawski if (mlen < 64 && mlen < *len) { 13724fb66994SJarek Poplawski put_page(p); 13734fb66994SJarek Poplawski goto new_page; 13744fb66994SJarek Poplawski } 13754fb66994SJarek Poplawski 13764fb66994SJarek Poplawski *len = min_t(unsigned int, *len, mlen); 13774fb66994SJarek Poplawski } 13784fb66994SJarek Poplawski 13794fb66994SJarek Poplawski memcpy(page_address(p) + off, page_address(page) + *offset, *len); 13804fb66994SJarek Poplawski sk->sk_sndmsg_off += *len; 13814fb66994SJarek Poplawski *offset = off; 13824fb66994SJarek Poplawski get_page(p); 13838b9d3728SJarek Poplawski 13848b9d3728SJarek Poplawski return p; 13859c55e01cSJens Axboe } 13869c55e01cSJens Axboe 13879c55e01cSJens Axboe /* 13889c55e01cSJens Axboe * Fill page/offset/length into spd, if it can hold more pages. 13899c55e01cSJens Axboe */ 13909c55e01cSJens Axboe static inline int spd_fill_page(struct splice_pipe_desc *spd, struct page *page, 13914fb66994SJarek Poplawski unsigned int *len, unsigned int offset, 13928b9d3728SJarek Poplawski struct sk_buff *skb, int linear) 13939c55e01cSJens Axboe { 13949c55e01cSJens Axboe if (unlikely(spd->nr_pages == PIPE_BUFFERS)) 13959c55e01cSJens Axboe return 1; 13969c55e01cSJens Axboe 13978b9d3728SJarek Poplawski if (linear) { 13984fb66994SJarek Poplawski page = linear_to_page(page, len, &offset, skb); 13998b9d3728SJarek Poplawski if (!page) 14008b9d3728SJarek Poplawski return 1; 14018b9d3728SJarek Poplawski } else 14028b9d3728SJarek Poplawski get_page(page); 14038b9d3728SJarek Poplawski 14049c55e01cSJens Axboe spd->pages[spd->nr_pages] = page; 14054fb66994SJarek Poplawski spd->partial[spd->nr_pages].len = *len; 14069c55e01cSJens Axboe spd->partial[spd->nr_pages].offset = offset; 14079c55e01cSJens Axboe spd->nr_pages++; 14088b9d3728SJarek Poplawski 14099c55e01cSJens Axboe return 0; 14109c55e01cSJens Axboe } 14119c55e01cSJens Axboe 14122870c43dSOctavian Purdila static inline void __segment_seek(struct page **page, unsigned int *poff, 14132870c43dSOctavian Purdila unsigned int *plen, unsigned int off) 14142870c43dSOctavian Purdila { 1415ce3dd395SJarek Poplawski unsigned long n; 1416ce3dd395SJarek Poplawski 14172870c43dSOctavian Purdila *poff += off; 1418ce3dd395SJarek Poplawski n = *poff / PAGE_SIZE; 1419ce3dd395SJarek Poplawski if (n) 1420ce3dd395SJarek Poplawski *page = nth_page(*page, n); 1421ce3dd395SJarek Poplawski 14222870c43dSOctavian Purdila *poff = *poff % PAGE_SIZE; 14232870c43dSOctavian Purdila *plen -= off; 14242870c43dSOctavian Purdila } 14252870c43dSOctavian Purdila 14262870c43dSOctavian Purdila static inline int __splice_segment(struct page *page, unsigned int poff, 14272870c43dSOctavian Purdila unsigned int plen, unsigned int *off, 14282870c43dSOctavian Purdila unsigned int *len, struct sk_buff *skb, 14298b9d3728SJarek Poplawski struct splice_pipe_desc *spd, int linear) 14309c55e01cSJens Axboe { 14312870c43dSOctavian Purdila if (!*len) 14322870c43dSOctavian Purdila return 1; 14339c55e01cSJens Axboe 14342870c43dSOctavian Purdila /* skip this segment if already processed */ 14352870c43dSOctavian Purdila if (*off >= plen) { 14362870c43dSOctavian Purdila *off -= plen; 14372870c43dSOctavian Purdila return 0; 14382870c43dSOctavian Purdila } 14392870c43dSOctavian Purdila 14402870c43dSOctavian Purdila /* ignore any bits we already processed */ 14412870c43dSOctavian Purdila if (*off) { 14422870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, *off); 14432870c43dSOctavian Purdila *off = 0; 14442870c43dSOctavian Purdila } 14452870c43dSOctavian Purdila 14462870c43dSOctavian Purdila do { 14472870c43dSOctavian Purdila unsigned int flen = min(*len, plen); 14482870c43dSOctavian Purdila 14492870c43dSOctavian Purdila /* the linear region may spread across several pages */ 14502870c43dSOctavian Purdila flen = min_t(unsigned int, flen, PAGE_SIZE - poff); 14512870c43dSOctavian Purdila 14524fb66994SJarek Poplawski if (spd_fill_page(spd, page, &flen, poff, skb, linear)) 14532870c43dSOctavian Purdila return 1; 14542870c43dSOctavian Purdila 14552870c43dSOctavian Purdila __segment_seek(&page, &poff, &plen, flen); 14562870c43dSOctavian Purdila *len -= flen; 14572870c43dSOctavian Purdila 14582870c43dSOctavian Purdila } while (*len && plen); 14592870c43dSOctavian Purdila 14602870c43dSOctavian Purdila return 0; 1461db43a282SOctavian Purdila } 14629c55e01cSJens Axboe 14639c55e01cSJens Axboe /* 14642870c43dSOctavian Purdila * Map linear and fragment data from the skb to spd. It reports failure if the 14652870c43dSOctavian Purdila * pipe is full or if we already spliced the requested length. 14669c55e01cSJens Axboe */ 14677b1c65faSHarvey Harrison static int __skb_splice_bits(struct sk_buff *skb, unsigned int *offset, 14682870c43dSOctavian Purdila unsigned int *len, 14692870c43dSOctavian Purdila struct splice_pipe_desc *spd) 14702870c43dSOctavian Purdila { 14712870c43dSOctavian Purdila int seg; 14729c55e01cSJens Axboe 14739c55e01cSJens Axboe /* 14742870c43dSOctavian Purdila * map the linear part 14759c55e01cSJens Axboe */ 14762870c43dSOctavian Purdila if (__splice_segment(virt_to_page(skb->data), 14772870c43dSOctavian Purdila (unsigned long) skb->data & (PAGE_SIZE - 1), 14782870c43dSOctavian Purdila skb_headlen(skb), 14798b9d3728SJarek Poplawski offset, len, skb, spd, 1)) 14802870c43dSOctavian Purdila return 1; 14819c55e01cSJens Axboe 14829c55e01cSJens Axboe /* 14839c55e01cSJens Axboe * then map the fragments 14849c55e01cSJens Axboe */ 14859c55e01cSJens Axboe for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) { 14869c55e01cSJens Axboe const skb_frag_t *f = &skb_shinfo(skb)->frags[seg]; 14879c55e01cSJens Axboe 14882870c43dSOctavian Purdila if (__splice_segment(f->page, f->page_offset, f->size, 14898b9d3728SJarek Poplawski offset, len, skb, spd, 0)) 14902870c43dSOctavian Purdila return 1; 14919c55e01cSJens Axboe } 14929c55e01cSJens Axboe 14939c55e01cSJens Axboe return 0; 14949c55e01cSJens Axboe } 14959c55e01cSJens Axboe 14969c55e01cSJens Axboe /* 14979c55e01cSJens Axboe * Map data from the skb to a pipe. Should handle both the linear part, 14989c55e01cSJens Axboe * the fragments, and the frag list. It does NOT handle frag lists within 14999c55e01cSJens Axboe * the frag list, if such a thing exists. We'd probably need to recurse to 15009c55e01cSJens Axboe * handle that cleanly. 15019c55e01cSJens Axboe */ 15028b9d3728SJarek Poplawski int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 15039c55e01cSJens Axboe struct pipe_inode_info *pipe, unsigned int tlen, 15049c55e01cSJens Axboe unsigned int flags) 15059c55e01cSJens Axboe { 15069c55e01cSJens Axboe struct partial_page partial[PIPE_BUFFERS]; 15079c55e01cSJens Axboe struct page *pages[PIPE_BUFFERS]; 15089c55e01cSJens Axboe struct splice_pipe_desc spd = { 15099c55e01cSJens Axboe .pages = pages, 15109c55e01cSJens Axboe .partial = partial, 15119c55e01cSJens Axboe .flags = flags, 15129c55e01cSJens Axboe .ops = &sock_pipe_buf_ops, 15139c55e01cSJens Axboe .spd_release = sock_spd_release, 15149c55e01cSJens Axboe }; 15159c55e01cSJens Axboe 15169c55e01cSJens Axboe /* 15179c55e01cSJens Axboe * __skb_splice_bits() only fails if the output has no room left, 15189c55e01cSJens Axboe * so no point in going over the frag_list for the error case. 15199c55e01cSJens Axboe */ 15209c55e01cSJens Axboe if (__skb_splice_bits(skb, &offset, &tlen, &spd)) 15219c55e01cSJens Axboe goto done; 15229c55e01cSJens Axboe else if (!tlen) 15239c55e01cSJens Axboe goto done; 15249c55e01cSJens Axboe 15259c55e01cSJens Axboe /* 15269c55e01cSJens Axboe * now see if we have a frag_list to map 15279c55e01cSJens Axboe */ 15289c55e01cSJens Axboe if (skb_shinfo(skb)->frag_list) { 15299c55e01cSJens Axboe struct sk_buff *list = skb_shinfo(skb)->frag_list; 15309c55e01cSJens Axboe 15319c55e01cSJens Axboe for (; list && tlen; list = list->next) { 15329c55e01cSJens Axboe if (__skb_splice_bits(list, &offset, &tlen, &spd)) 15339c55e01cSJens Axboe break; 15349c55e01cSJens Axboe } 15359c55e01cSJens Axboe } 15369c55e01cSJens Axboe 15379c55e01cSJens Axboe done: 15389c55e01cSJens Axboe if (spd.nr_pages) { 15398b9d3728SJarek Poplawski struct sock *sk = skb->sk; 15409c55e01cSJens Axboe int ret; 15419c55e01cSJens Axboe 15429c55e01cSJens Axboe /* 15439c55e01cSJens Axboe * Drop the socket lock, otherwise we have reverse 15449c55e01cSJens Axboe * locking dependencies between sk_lock and i_mutex 15459c55e01cSJens Axboe * here as compared to sendfile(). We enter here 15469c55e01cSJens Axboe * with the socket lock held, and splice_to_pipe() will 15479c55e01cSJens Axboe * grab the pipe inode lock. For sendfile() emulation, 15489c55e01cSJens Axboe * we call into ->sendpage() with the i_mutex lock held 15499c55e01cSJens Axboe * and networking will grab the socket lock. 15509c55e01cSJens Axboe */ 1551293ad604SOctavian Purdila release_sock(sk); 15529c55e01cSJens Axboe ret = splice_to_pipe(pipe, &spd); 1553293ad604SOctavian Purdila lock_sock(sk); 15549c55e01cSJens Axboe return ret; 15559c55e01cSJens Axboe } 15569c55e01cSJens Axboe 15579c55e01cSJens Axboe return 0; 15589c55e01cSJens Axboe } 15599c55e01cSJens Axboe 1560357b40a1SHerbert Xu /** 1561357b40a1SHerbert Xu * skb_store_bits - store bits from kernel buffer to skb 1562357b40a1SHerbert Xu * @skb: destination buffer 1563357b40a1SHerbert Xu * @offset: offset in destination 1564357b40a1SHerbert Xu * @from: source buffer 1565357b40a1SHerbert Xu * @len: number of bytes to copy 1566357b40a1SHerbert Xu * 1567357b40a1SHerbert Xu * Copy the specified number of bytes from the source buffer to the 1568357b40a1SHerbert Xu * destination skb. This function handles all the messy bits of 1569357b40a1SHerbert Xu * traversing fragment lists and such. 1570357b40a1SHerbert Xu */ 1571357b40a1SHerbert Xu 15720c6fcc8aSStephen Hemminger int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len) 1573357b40a1SHerbert Xu { 1574357b40a1SHerbert Xu int i, copy; 15751a028e50SDavid S. Miller int start = skb_headlen(skb); 1576357b40a1SHerbert Xu 1577357b40a1SHerbert Xu if (offset > (int)skb->len - len) 1578357b40a1SHerbert Xu goto fault; 1579357b40a1SHerbert Xu 15801a028e50SDavid S. Miller if ((copy = start - offset) > 0) { 1581357b40a1SHerbert Xu if (copy > len) 1582357b40a1SHerbert Xu copy = len; 158327d7ff46SArnaldo Carvalho de Melo skb_copy_to_linear_data_offset(skb, offset, from, copy); 1584357b40a1SHerbert Xu if ((len -= copy) == 0) 1585357b40a1SHerbert Xu return 0; 1586357b40a1SHerbert Xu offset += copy; 1587357b40a1SHerbert Xu from += copy; 1588357b40a1SHerbert Xu } 1589357b40a1SHerbert Xu 1590357b40a1SHerbert Xu for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 1591357b40a1SHerbert Xu skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 15921a028e50SDavid S. Miller int end; 1593357b40a1SHerbert Xu 1594547b792cSIlpo Järvinen WARN_ON(start > offset + len); 15951a028e50SDavid S. Miller 15961a028e50SDavid S. Miller end = start + frag->size; 1597357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1598357b40a1SHerbert Xu u8 *vaddr; 1599357b40a1SHerbert Xu 1600357b40a1SHerbert Xu if (copy > len) 1601357b40a1SHerbert Xu copy = len; 1602357b40a1SHerbert Xu 1603357b40a1SHerbert Xu vaddr = kmap_skb_frag(frag); 16041a028e50SDavid S. Miller memcpy(vaddr + frag->page_offset + offset - start, 16051a028e50SDavid S. Miller from, copy); 1606357b40a1SHerbert Xu kunmap_skb_frag(vaddr); 1607357b40a1SHerbert Xu 1608357b40a1SHerbert Xu if ((len -= copy) == 0) 1609357b40a1SHerbert Xu return 0; 1610357b40a1SHerbert Xu offset += copy; 1611357b40a1SHerbert Xu from += copy; 1612357b40a1SHerbert Xu } 16131a028e50SDavid S. Miller start = end; 1614357b40a1SHerbert Xu } 1615357b40a1SHerbert Xu 1616357b40a1SHerbert Xu if (skb_shinfo(skb)->frag_list) { 1617357b40a1SHerbert Xu struct sk_buff *list = skb_shinfo(skb)->frag_list; 1618357b40a1SHerbert Xu 1619357b40a1SHerbert Xu for (; list; list = list->next) { 16201a028e50SDavid S. Miller int end; 1621357b40a1SHerbert Xu 1622547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16231a028e50SDavid S. Miller 16241a028e50SDavid S. Miller end = start + list->len; 1625357b40a1SHerbert Xu if ((copy = end - offset) > 0) { 1626357b40a1SHerbert Xu if (copy > len) 1627357b40a1SHerbert Xu copy = len; 16281a028e50SDavid S. Miller if (skb_store_bits(list, offset - start, 16291a028e50SDavid S. Miller from, copy)) 1630357b40a1SHerbert Xu goto fault; 1631357b40a1SHerbert Xu if ((len -= copy) == 0) 1632357b40a1SHerbert Xu return 0; 1633357b40a1SHerbert Xu offset += copy; 1634357b40a1SHerbert Xu from += copy; 1635357b40a1SHerbert Xu } 16361a028e50SDavid S. Miller start = end; 1637357b40a1SHerbert Xu } 1638357b40a1SHerbert Xu } 1639357b40a1SHerbert Xu if (!len) 1640357b40a1SHerbert Xu return 0; 1641357b40a1SHerbert Xu 1642357b40a1SHerbert Xu fault: 1643357b40a1SHerbert Xu return -EFAULT; 1644357b40a1SHerbert Xu } 1645357b40a1SHerbert Xu EXPORT_SYMBOL(skb_store_bits); 1646357b40a1SHerbert Xu 16471da177e4SLinus Torvalds /* Checksum skb data. */ 16481da177e4SLinus Torvalds 16492bbbc868SAl Viro __wsum skb_checksum(const struct sk_buff *skb, int offset, 16502bbbc868SAl Viro int len, __wsum csum) 16511da177e4SLinus Torvalds { 16521a028e50SDavid S. Miller int start = skb_headlen(skb); 16531a028e50SDavid S. Miller int i, copy = start - offset; 16541da177e4SLinus Torvalds int pos = 0; 16551da177e4SLinus Torvalds 16561da177e4SLinus Torvalds /* Checksum header. */ 16571da177e4SLinus Torvalds if (copy > 0) { 16581da177e4SLinus Torvalds if (copy > len) 16591da177e4SLinus Torvalds copy = len; 16601da177e4SLinus Torvalds csum = csum_partial(skb->data + offset, copy, csum); 16611da177e4SLinus Torvalds if ((len -= copy) == 0) 16621da177e4SLinus Torvalds return csum; 16631da177e4SLinus Torvalds offset += copy; 16641da177e4SLinus Torvalds pos = copy; 16651da177e4SLinus Torvalds } 16661da177e4SLinus Torvalds 16671da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 16681a028e50SDavid S. Miller int end; 16691da177e4SLinus Torvalds 1670547b792cSIlpo Järvinen WARN_ON(start > offset + len); 16711a028e50SDavid S. Miller 16721a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 16731da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 167444bb9363SAl Viro __wsum csum2; 16751da177e4SLinus Torvalds u8 *vaddr; 16761da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 16771da177e4SLinus Torvalds 16781da177e4SLinus Torvalds if (copy > len) 16791da177e4SLinus Torvalds copy = len; 16801da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 16811a028e50SDavid S. Miller csum2 = csum_partial(vaddr + frag->page_offset + 16821a028e50SDavid S. Miller offset - start, copy, 0); 16831da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 16841da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 16851da177e4SLinus Torvalds if (!(len -= copy)) 16861da177e4SLinus Torvalds return csum; 16871da177e4SLinus Torvalds offset += copy; 16881da177e4SLinus Torvalds pos += copy; 16891da177e4SLinus Torvalds } 16901a028e50SDavid S. Miller start = end; 16911da177e4SLinus Torvalds } 16921da177e4SLinus Torvalds 16931da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 16941da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 16951da177e4SLinus Torvalds 16961da177e4SLinus Torvalds for (; list; list = list->next) { 16971a028e50SDavid S. Miller int end; 16981da177e4SLinus Torvalds 1699547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17001a028e50SDavid S. Miller 17011a028e50SDavid S. Miller end = start + list->len; 17021da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17035f92a738SAl Viro __wsum csum2; 17041da177e4SLinus Torvalds if (copy > len) 17051da177e4SLinus Torvalds copy = len; 17061a028e50SDavid S. Miller csum2 = skb_checksum(list, offset - start, 17071a028e50SDavid S. Miller copy, 0); 17081da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17091da177e4SLinus Torvalds if ((len -= copy) == 0) 17101da177e4SLinus Torvalds return csum; 17111da177e4SLinus Torvalds offset += copy; 17121da177e4SLinus Torvalds pos += copy; 17131da177e4SLinus Torvalds } 17141a028e50SDavid S. Miller start = end; 17151da177e4SLinus Torvalds } 17161da177e4SLinus Torvalds } 171709a62660SKris Katterjohn BUG_ON(len); 17181da177e4SLinus Torvalds 17191da177e4SLinus Torvalds return csum; 17201da177e4SLinus Torvalds } 1721b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_checksum); 17221da177e4SLinus Torvalds 17231da177e4SLinus Torvalds /* Both of above in one bottle. */ 17241da177e4SLinus Torvalds 172581d77662SAl Viro __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, 172681d77662SAl Viro u8 *to, int len, __wsum csum) 17271da177e4SLinus Torvalds { 17281a028e50SDavid S. Miller int start = skb_headlen(skb); 17291a028e50SDavid S. Miller int i, copy = start - offset; 17301da177e4SLinus Torvalds int pos = 0; 17311da177e4SLinus Torvalds 17321da177e4SLinus Torvalds /* Copy header. */ 17331da177e4SLinus Torvalds if (copy > 0) { 17341da177e4SLinus Torvalds if (copy > len) 17351da177e4SLinus Torvalds copy = len; 17361da177e4SLinus Torvalds csum = csum_partial_copy_nocheck(skb->data + offset, to, 17371da177e4SLinus Torvalds copy, csum); 17381da177e4SLinus Torvalds if ((len -= copy) == 0) 17391da177e4SLinus Torvalds return csum; 17401da177e4SLinus Torvalds offset += copy; 17411da177e4SLinus Torvalds to += copy; 17421da177e4SLinus Torvalds pos = copy; 17431da177e4SLinus Torvalds } 17441da177e4SLinus Torvalds 17451da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 17461a028e50SDavid S. Miller int end; 17471da177e4SLinus Torvalds 1748547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17491a028e50SDavid S. Miller 17501a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 17511da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17525084205fSAl Viro __wsum csum2; 17531da177e4SLinus Torvalds u8 *vaddr; 17541da177e4SLinus Torvalds skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 17551da177e4SLinus Torvalds 17561da177e4SLinus Torvalds if (copy > len) 17571da177e4SLinus Torvalds copy = len; 17581da177e4SLinus Torvalds vaddr = kmap_skb_frag(frag); 17591da177e4SLinus Torvalds csum2 = csum_partial_copy_nocheck(vaddr + 17601a028e50SDavid S. Miller frag->page_offset + 17611a028e50SDavid S. Miller offset - start, to, 17621a028e50SDavid S. Miller copy, 0); 17631da177e4SLinus Torvalds kunmap_skb_frag(vaddr); 17641da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17651da177e4SLinus Torvalds if (!(len -= copy)) 17661da177e4SLinus Torvalds return csum; 17671da177e4SLinus Torvalds offset += copy; 17681da177e4SLinus Torvalds to += copy; 17691da177e4SLinus Torvalds pos += copy; 17701da177e4SLinus Torvalds } 17711a028e50SDavid S. Miller start = end; 17721da177e4SLinus Torvalds } 17731da177e4SLinus Torvalds 17741da177e4SLinus Torvalds if (skb_shinfo(skb)->frag_list) { 17751da177e4SLinus Torvalds struct sk_buff *list = skb_shinfo(skb)->frag_list; 17761da177e4SLinus Torvalds 17771da177e4SLinus Torvalds for (; list; list = list->next) { 177881d77662SAl Viro __wsum csum2; 17791a028e50SDavid S. Miller int end; 17801da177e4SLinus Torvalds 1781547b792cSIlpo Järvinen WARN_ON(start > offset + len); 17821a028e50SDavid S. Miller 17831a028e50SDavid S. Miller end = start + list->len; 17841da177e4SLinus Torvalds if ((copy = end - offset) > 0) { 17851da177e4SLinus Torvalds if (copy > len) 17861da177e4SLinus Torvalds copy = len; 17871a028e50SDavid S. Miller csum2 = skb_copy_and_csum_bits(list, 17881a028e50SDavid S. Miller offset - start, 17891da177e4SLinus Torvalds to, copy, 0); 17901da177e4SLinus Torvalds csum = csum_block_add(csum, csum2, pos); 17911da177e4SLinus Torvalds if ((len -= copy) == 0) 17921da177e4SLinus Torvalds return csum; 17931da177e4SLinus Torvalds offset += copy; 17941da177e4SLinus Torvalds to += copy; 17951da177e4SLinus Torvalds pos += copy; 17961da177e4SLinus Torvalds } 17971a028e50SDavid S. Miller start = end; 17981da177e4SLinus Torvalds } 17991da177e4SLinus Torvalds } 180009a62660SKris Katterjohn BUG_ON(len); 18011da177e4SLinus Torvalds return csum; 18021da177e4SLinus Torvalds } 1803b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_bits); 18041da177e4SLinus Torvalds 18051da177e4SLinus Torvalds void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to) 18061da177e4SLinus Torvalds { 1807d3bc23e7SAl Viro __wsum csum; 18081da177e4SLinus Torvalds long csstart; 18091da177e4SLinus Torvalds 181084fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) 1811663ead3bSHerbert Xu csstart = skb->csum_start - skb_headroom(skb); 18121da177e4SLinus Torvalds else 18131da177e4SLinus Torvalds csstart = skb_headlen(skb); 18141da177e4SLinus Torvalds 181509a62660SKris Katterjohn BUG_ON(csstart > skb_headlen(skb)); 18161da177e4SLinus Torvalds 1817d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data(skb, to, csstart); 18181da177e4SLinus Torvalds 18191da177e4SLinus Torvalds csum = 0; 18201da177e4SLinus Torvalds if (csstart != skb->len) 18211da177e4SLinus Torvalds csum = skb_copy_and_csum_bits(skb, csstart, to + csstart, 18221da177e4SLinus Torvalds skb->len - csstart, 0); 18231da177e4SLinus Torvalds 182484fa7933SPatrick McHardy if (skb->ip_summed == CHECKSUM_PARTIAL) { 1825ff1dcadbSAl Viro long csstuff = csstart + skb->csum_offset; 18261da177e4SLinus Torvalds 1827d3bc23e7SAl Viro *((__sum16 *)(to + csstuff)) = csum_fold(csum); 18281da177e4SLinus Torvalds } 18291da177e4SLinus Torvalds } 1830b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_copy_and_csum_dev); 18311da177e4SLinus Torvalds 18321da177e4SLinus Torvalds /** 18331da177e4SLinus Torvalds * skb_dequeue - remove from the head of the queue 18341da177e4SLinus Torvalds * @list: list to dequeue from 18351da177e4SLinus Torvalds * 18361da177e4SLinus Torvalds * Remove the head of the list. The list lock is taken so the function 18371da177e4SLinus Torvalds * may be used safely with other locking list functions. The head item is 18381da177e4SLinus Torvalds * returned or %NULL if the list is empty. 18391da177e4SLinus Torvalds */ 18401da177e4SLinus Torvalds 18411da177e4SLinus Torvalds struct sk_buff *skb_dequeue(struct sk_buff_head *list) 18421da177e4SLinus Torvalds { 18431da177e4SLinus Torvalds unsigned long flags; 18441da177e4SLinus Torvalds struct sk_buff *result; 18451da177e4SLinus Torvalds 18461da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18471da177e4SLinus Torvalds result = __skb_dequeue(list); 18481da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18491da177e4SLinus Torvalds return result; 18501da177e4SLinus Torvalds } 1851b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue); 18521da177e4SLinus Torvalds 18531da177e4SLinus Torvalds /** 18541da177e4SLinus Torvalds * skb_dequeue_tail - remove from the tail of the queue 18551da177e4SLinus Torvalds * @list: list to dequeue from 18561da177e4SLinus Torvalds * 18571da177e4SLinus Torvalds * Remove the tail of the list. The list lock is taken so the function 18581da177e4SLinus Torvalds * may be used safely with other locking list functions. The tail item is 18591da177e4SLinus Torvalds * returned or %NULL if the list is empty. 18601da177e4SLinus Torvalds */ 18611da177e4SLinus Torvalds struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list) 18621da177e4SLinus Torvalds { 18631da177e4SLinus Torvalds unsigned long flags; 18641da177e4SLinus Torvalds struct sk_buff *result; 18651da177e4SLinus Torvalds 18661da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 18671da177e4SLinus Torvalds result = __skb_dequeue_tail(list); 18681da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 18691da177e4SLinus Torvalds return result; 18701da177e4SLinus Torvalds } 1871b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_dequeue_tail); 18721da177e4SLinus Torvalds 18731da177e4SLinus Torvalds /** 18741da177e4SLinus Torvalds * skb_queue_purge - empty a list 18751da177e4SLinus Torvalds * @list: list to empty 18761da177e4SLinus Torvalds * 18771da177e4SLinus Torvalds * Delete all buffers on an &sk_buff list. Each buffer is removed from 18781da177e4SLinus Torvalds * the list and one reference dropped. This function takes the list 18791da177e4SLinus Torvalds * lock and is atomic with respect to other list locking functions. 18801da177e4SLinus Torvalds */ 18811da177e4SLinus Torvalds void skb_queue_purge(struct sk_buff_head *list) 18821da177e4SLinus Torvalds { 18831da177e4SLinus Torvalds struct sk_buff *skb; 18841da177e4SLinus Torvalds while ((skb = skb_dequeue(list)) != NULL) 18851da177e4SLinus Torvalds kfree_skb(skb); 18861da177e4SLinus Torvalds } 1887b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_purge); 18881da177e4SLinus Torvalds 18891da177e4SLinus Torvalds /** 18901da177e4SLinus Torvalds * skb_queue_head - queue a buffer at the list head 18911da177e4SLinus Torvalds * @list: list to use 18921da177e4SLinus Torvalds * @newsk: buffer to queue 18931da177e4SLinus Torvalds * 18941da177e4SLinus Torvalds * Queue a buffer at the start of the list. This function takes the 18951da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 18961da177e4SLinus Torvalds * safely. 18971da177e4SLinus Torvalds * 18981da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 18991da177e4SLinus Torvalds */ 19001da177e4SLinus Torvalds void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk) 19011da177e4SLinus Torvalds { 19021da177e4SLinus Torvalds unsigned long flags; 19031da177e4SLinus Torvalds 19041da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19051da177e4SLinus Torvalds __skb_queue_head(list, newsk); 19061da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19071da177e4SLinus Torvalds } 1908b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_head); 19091da177e4SLinus Torvalds 19101da177e4SLinus Torvalds /** 19111da177e4SLinus Torvalds * skb_queue_tail - queue a buffer at the list tail 19121da177e4SLinus Torvalds * @list: list to use 19131da177e4SLinus Torvalds * @newsk: buffer to queue 19141da177e4SLinus Torvalds * 19151da177e4SLinus Torvalds * Queue a buffer at the tail of the list. This function takes the 19161da177e4SLinus Torvalds * list lock and can be used safely with other locking &sk_buff functions 19171da177e4SLinus Torvalds * safely. 19181da177e4SLinus Torvalds * 19191da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19201da177e4SLinus Torvalds */ 19211da177e4SLinus Torvalds void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk) 19221da177e4SLinus Torvalds { 19231da177e4SLinus Torvalds unsigned long flags; 19241da177e4SLinus Torvalds 19251da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19261da177e4SLinus Torvalds __skb_queue_tail(list, newsk); 19271da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19281da177e4SLinus Torvalds } 1929b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_queue_tail); 19308728b834SDavid S. Miller 19311da177e4SLinus Torvalds /** 19321da177e4SLinus Torvalds * skb_unlink - remove a buffer from a list 19331da177e4SLinus Torvalds * @skb: buffer to remove 19348728b834SDavid S. Miller * @list: list to use 19351da177e4SLinus Torvalds * 19368728b834SDavid S. Miller * Remove a packet from a list. The list locks are taken and this 19378728b834SDavid S. Miller * function is atomic with respect to other list locked calls 19381da177e4SLinus Torvalds * 19398728b834SDavid S. Miller * You must know what list the SKB is on. 19401da177e4SLinus Torvalds */ 19418728b834SDavid S. Miller void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 19421da177e4SLinus Torvalds { 19431da177e4SLinus Torvalds unsigned long flags; 19441da177e4SLinus Torvalds 19451da177e4SLinus Torvalds spin_lock_irqsave(&list->lock, flags); 19468728b834SDavid S. Miller __skb_unlink(skb, list); 19471da177e4SLinus Torvalds spin_unlock_irqrestore(&list->lock, flags); 19481da177e4SLinus Torvalds } 1949b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_unlink); 19501da177e4SLinus Torvalds 19511da177e4SLinus Torvalds /** 19521da177e4SLinus Torvalds * skb_append - append a buffer 19531da177e4SLinus Torvalds * @old: buffer to insert after 19541da177e4SLinus Torvalds * @newsk: buffer to insert 19558728b834SDavid S. Miller * @list: list to use 19561da177e4SLinus Torvalds * 19571da177e4SLinus Torvalds * Place a packet after a given packet in a list. The list locks are taken 19581da177e4SLinus Torvalds * and this function is atomic with respect to other list locked calls. 19591da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19601da177e4SLinus Torvalds */ 19618728b834SDavid S. Miller void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 19621da177e4SLinus Torvalds { 19631da177e4SLinus Torvalds unsigned long flags; 19641da177e4SLinus Torvalds 19658728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 19667de6c033SGerrit Renker __skb_queue_after(list, old, newsk); 19678728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 19681da177e4SLinus Torvalds } 1969b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append); 19701da177e4SLinus Torvalds 19711da177e4SLinus Torvalds /** 19721da177e4SLinus Torvalds * skb_insert - insert a buffer 19731da177e4SLinus Torvalds * @old: buffer to insert before 19741da177e4SLinus Torvalds * @newsk: buffer to insert 19758728b834SDavid S. Miller * @list: list to use 19761da177e4SLinus Torvalds * 19778728b834SDavid S. Miller * Place a packet before a given packet in a list. The list locks are 19788728b834SDavid S. Miller * taken and this function is atomic with respect to other list locked 19798728b834SDavid S. Miller * calls. 19808728b834SDavid S. Miller * 19811da177e4SLinus Torvalds * A buffer cannot be placed on two lists at the same time. 19821da177e4SLinus Torvalds */ 19838728b834SDavid S. Miller void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list) 19841da177e4SLinus Torvalds { 19851da177e4SLinus Torvalds unsigned long flags; 19861da177e4SLinus Torvalds 19878728b834SDavid S. Miller spin_lock_irqsave(&list->lock, flags); 19888728b834SDavid S. Miller __skb_insert(newsk, old->prev, old, list); 19898728b834SDavid S. Miller spin_unlock_irqrestore(&list->lock, flags); 19901da177e4SLinus Torvalds } 1991b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_insert); 19921da177e4SLinus Torvalds 19931da177e4SLinus Torvalds static inline void skb_split_inside_header(struct sk_buff *skb, 19941da177e4SLinus Torvalds struct sk_buff* skb1, 19951da177e4SLinus Torvalds const u32 len, const int pos) 19961da177e4SLinus Torvalds { 19971da177e4SLinus Torvalds int i; 19981da177e4SLinus Torvalds 1999d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len), 2000d626f62bSArnaldo Carvalho de Melo pos - len); 20011da177e4SLinus Torvalds /* And move data appendix as is. */ 20021da177e4SLinus Torvalds for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) 20031da177e4SLinus Torvalds skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i]; 20041da177e4SLinus Torvalds 20051da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags; 20061da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 20071da177e4SLinus Torvalds skb1->data_len = skb->data_len; 20081da177e4SLinus Torvalds skb1->len += skb1->data_len; 20091da177e4SLinus Torvalds skb->data_len = 0; 20101da177e4SLinus Torvalds skb->len = len; 201127a884dcSArnaldo Carvalho de Melo skb_set_tail_pointer(skb, len); 20121da177e4SLinus Torvalds } 20131da177e4SLinus Torvalds 20141da177e4SLinus Torvalds static inline void skb_split_no_header(struct sk_buff *skb, 20151da177e4SLinus Torvalds struct sk_buff* skb1, 20161da177e4SLinus Torvalds const u32 len, int pos) 20171da177e4SLinus Torvalds { 20181da177e4SLinus Torvalds int i, k = 0; 20191da177e4SLinus Torvalds const int nfrags = skb_shinfo(skb)->nr_frags; 20201da177e4SLinus Torvalds 20211da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags = 0; 20221da177e4SLinus Torvalds skb1->len = skb1->data_len = skb->len - len; 20231da177e4SLinus Torvalds skb->len = len; 20241da177e4SLinus Torvalds skb->data_len = len - pos; 20251da177e4SLinus Torvalds 20261da177e4SLinus Torvalds for (i = 0; i < nfrags; i++) { 20271da177e4SLinus Torvalds int size = skb_shinfo(skb)->frags[i].size; 20281da177e4SLinus Torvalds 20291da177e4SLinus Torvalds if (pos + size > len) { 20301da177e4SLinus Torvalds skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i]; 20311da177e4SLinus Torvalds 20321da177e4SLinus Torvalds if (pos < len) { 20331da177e4SLinus Torvalds /* Split frag. 20341da177e4SLinus Torvalds * We have two variants in this case: 20351da177e4SLinus Torvalds * 1. Move all the frag to the second 20361da177e4SLinus Torvalds * part, if it is possible. F.e. 20371da177e4SLinus Torvalds * this approach is mandatory for TUX, 20381da177e4SLinus Torvalds * where splitting is expensive. 20391da177e4SLinus Torvalds * 2. Split is accurately. We make this. 20401da177e4SLinus Torvalds */ 20411da177e4SLinus Torvalds get_page(skb_shinfo(skb)->frags[i].page); 20421da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].page_offset += len - pos; 20431da177e4SLinus Torvalds skb_shinfo(skb1)->frags[0].size -= len - pos; 20441da177e4SLinus Torvalds skb_shinfo(skb)->frags[i].size = len - pos; 20451da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 20461da177e4SLinus Torvalds } 20471da177e4SLinus Torvalds k++; 20481da177e4SLinus Torvalds } else 20491da177e4SLinus Torvalds skb_shinfo(skb)->nr_frags++; 20501da177e4SLinus Torvalds pos += size; 20511da177e4SLinus Torvalds } 20521da177e4SLinus Torvalds skb_shinfo(skb1)->nr_frags = k; 20531da177e4SLinus Torvalds } 20541da177e4SLinus Torvalds 20551da177e4SLinus Torvalds /** 20561da177e4SLinus Torvalds * skb_split - Split fragmented skb to two parts at length len. 20571da177e4SLinus Torvalds * @skb: the buffer to split 20581da177e4SLinus Torvalds * @skb1: the buffer to receive the second part 20591da177e4SLinus Torvalds * @len: new length for skb 20601da177e4SLinus Torvalds */ 20611da177e4SLinus Torvalds void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len) 20621da177e4SLinus Torvalds { 20631da177e4SLinus Torvalds int pos = skb_headlen(skb); 20641da177e4SLinus Torvalds 20651da177e4SLinus Torvalds if (len < pos) /* Split line is inside header. */ 20661da177e4SLinus Torvalds skb_split_inside_header(skb, skb1, len, pos); 20671da177e4SLinus Torvalds else /* Second chunk has no header, nothing to copy. */ 20681da177e4SLinus Torvalds skb_split_no_header(skb, skb1, len, pos); 20691da177e4SLinus Torvalds } 2070b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_split); 20711da177e4SLinus Torvalds 20729f782db3SIlpo Järvinen /* Shifting from/to a cloned skb is a no-go. 20739f782db3SIlpo Järvinen * 20749f782db3SIlpo Järvinen * Caller cannot keep skb_shinfo related pointers past calling here! 20759f782db3SIlpo Järvinen */ 2076832d11c5SIlpo Järvinen static int skb_prepare_for_shift(struct sk_buff *skb) 2077832d11c5SIlpo Järvinen { 20780ace2856SIlpo Järvinen return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2079832d11c5SIlpo Järvinen } 2080832d11c5SIlpo Järvinen 2081832d11c5SIlpo Järvinen /** 2082832d11c5SIlpo Järvinen * skb_shift - Shifts paged data partially from skb to another 2083832d11c5SIlpo Järvinen * @tgt: buffer into which tail data gets added 2084832d11c5SIlpo Järvinen * @skb: buffer from which the paged data comes from 2085832d11c5SIlpo Järvinen * @shiftlen: shift up to this many bytes 2086832d11c5SIlpo Järvinen * 2087832d11c5SIlpo Järvinen * Attempts to shift up to shiftlen worth of bytes, which may be less than 2088832d11c5SIlpo Järvinen * the length of the skb, from tgt to skb. Returns number bytes shifted. 2089832d11c5SIlpo Järvinen * It's up to caller to free skb if everything was shifted. 2090832d11c5SIlpo Järvinen * 2091832d11c5SIlpo Järvinen * If @tgt runs out of frags, the whole operation is aborted. 2092832d11c5SIlpo Järvinen * 2093832d11c5SIlpo Järvinen * Skb cannot include anything else but paged data while tgt is allowed 2094832d11c5SIlpo Järvinen * to have non-paged data as well. 2095832d11c5SIlpo Järvinen * 2096832d11c5SIlpo Järvinen * TODO: full sized shift could be optimized but that would need 2097832d11c5SIlpo Järvinen * specialized skb free'er to handle frags without up-to-date nr_frags. 2098832d11c5SIlpo Järvinen */ 2099832d11c5SIlpo Järvinen int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen) 2100832d11c5SIlpo Järvinen { 2101832d11c5SIlpo Järvinen int from, to, merge, todo; 2102832d11c5SIlpo Järvinen struct skb_frag_struct *fragfrom, *fragto; 2103832d11c5SIlpo Järvinen 2104832d11c5SIlpo Järvinen BUG_ON(shiftlen > skb->len); 2105832d11c5SIlpo Järvinen BUG_ON(skb_headlen(skb)); /* Would corrupt stream */ 2106832d11c5SIlpo Järvinen 2107832d11c5SIlpo Järvinen todo = shiftlen; 2108832d11c5SIlpo Järvinen from = 0; 2109832d11c5SIlpo Järvinen to = skb_shinfo(tgt)->nr_frags; 2110832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2111832d11c5SIlpo Järvinen 2112832d11c5SIlpo Järvinen /* Actual merge is delayed until the point when we know we can 2113832d11c5SIlpo Järvinen * commit all, so that we don't have to undo partial changes 2114832d11c5SIlpo Järvinen */ 2115832d11c5SIlpo Järvinen if (!to || 2116832d11c5SIlpo Järvinen !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) { 2117832d11c5SIlpo Järvinen merge = -1; 2118832d11c5SIlpo Järvinen } else { 2119832d11c5SIlpo Järvinen merge = to - 1; 2120832d11c5SIlpo Järvinen 2121832d11c5SIlpo Järvinen todo -= fragfrom->size; 2122832d11c5SIlpo Järvinen if (todo < 0) { 2123832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || 2124832d11c5SIlpo Järvinen skb_prepare_for_shift(tgt)) 2125832d11c5SIlpo Järvinen return 0; 2126832d11c5SIlpo Järvinen 21279f782db3SIlpo Järvinen /* All previous frag pointers might be stale! */ 21289f782db3SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2129832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2130832d11c5SIlpo Järvinen 2131832d11c5SIlpo Järvinen fragto->size += shiftlen; 2132832d11c5SIlpo Järvinen fragfrom->size -= shiftlen; 2133832d11c5SIlpo Järvinen fragfrom->page_offset += shiftlen; 2134832d11c5SIlpo Järvinen 2135832d11c5SIlpo Järvinen goto onlymerged; 2136832d11c5SIlpo Järvinen } 2137832d11c5SIlpo Järvinen 2138832d11c5SIlpo Järvinen from++; 2139832d11c5SIlpo Järvinen } 2140832d11c5SIlpo Järvinen 2141832d11c5SIlpo Järvinen /* Skip full, not-fitting skb to avoid expensive operations */ 2142832d11c5SIlpo Järvinen if ((shiftlen == skb->len) && 2143832d11c5SIlpo Järvinen (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to)) 2144832d11c5SIlpo Järvinen return 0; 2145832d11c5SIlpo Järvinen 2146832d11c5SIlpo Järvinen if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt)) 2147832d11c5SIlpo Järvinen return 0; 2148832d11c5SIlpo Järvinen 2149832d11c5SIlpo Järvinen while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) { 2150832d11c5SIlpo Järvinen if (to == MAX_SKB_FRAGS) 2151832d11c5SIlpo Järvinen return 0; 2152832d11c5SIlpo Järvinen 2153832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[from]; 2154832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[to]; 2155832d11c5SIlpo Järvinen 2156832d11c5SIlpo Järvinen if (todo >= fragfrom->size) { 2157832d11c5SIlpo Järvinen *fragto = *fragfrom; 2158832d11c5SIlpo Järvinen todo -= fragfrom->size; 2159832d11c5SIlpo Järvinen from++; 2160832d11c5SIlpo Järvinen to++; 2161832d11c5SIlpo Järvinen 2162832d11c5SIlpo Järvinen } else { 2163832d11c5SIlpo Järvinen get_page(fragfrom->page); 2164832d11c5SIlpo Järvinen fragto->page = fragfrom->page; 2165832d11c5SIlpo Järvinen fragto->page_offset = fragfrom->page_offset; 2166832d11c5SIlpo Järvinen fragto->size = todo; 2167832d11c5SIlpo Järvinen 2168832d11c5SIlpo Järvinen fragfrom->page_offset += todo; 2169832d11c5SIlpo Järvinen fragfrom->size -= todo; 2170832d11c5SIlpo Järvinen todo = 0; 2171832d11c5SIlpo Järvinen 2172832d11c5SIlpo Järvinen to++; 2173832d11c5SIlpo Järvinen break; 2174832d11c5SIlpo Järvinen } 2175832d11c5SIlpo Järvinen } 2176832d11c5SIlpo Järvinen 2177832d11c5SIlpo Järvinen /* Ready to "commit" this state change to tgt */ 2178832d11c5SIlpo Järvinen skb_shinfo(tgt)->nr_frags = to; 2179832d11c5SIlpo Järvinen 2180832d11c5SIlpo Järvinen if (merge >= 0) { 2181832d11c5SIlpo Järvinen fragfrom = &skb_shinfo(skb)->frags[0]; 2182832d11c5SIlpo Järvinen fragto = &skb_shinfo(tgt)->frags[merge]; 2183832d11c5SIlpo Järvinen 2184832d11c5SIlpo Järvinen fragto->size += fragfrom->size; 2185832d11c5SIlpo Järvinen put_page(fragfrom->page); 2186832d11c5SIlpo Järvinen } 2187832d11c5SIlpo Järvinen 2188832d11c5SIlpo Järvinen /* Reposition in the original skb */ 2189832d11c5SIlpo Järvinen to = 0; 2190832d11c5SIlpo Järvinen while (from < skb_shinfo(skb)->nr_frags) 2191832d11c5SIlpo Järvinen skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++]; 2192832d11c5SIlpo Järvinen skb_shinfo(skb)->nr_frags = to; 2193832d11c5SIlpo Järvinen 2194832d11c5SIlpo Järvinen BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags); 2195832d11c5SIlpo Järvinen 2196832d11c5SIlpo Järvinen onlymerged: 2197832d11c5SIlpo Järvinen /* Most likely the tgt won't ever need its checksum anymore, skb on 2198832d11c5SIlpo Järvinen * the other hand might need it if it needs to be resent 2199832d11c5SIlpo Järvinen */ 2200832d11c5SIlpo Järvinen tgt->ip_summed = CHECKSUM_PARTIAL; 2201832d11c5SIlpo Järvinen skb->ip_summed = CHECKSUM_PARTIAL; 2202832d11c5SIlpo Järvinen 2203832d11c5SIlpo Järvinen /* Yak, is it really working this way? Some helper please? */ 2204832d11c5SIlpo Järvinen skb->len -= shiftlen; 2205832d11c5SIlpo Järvinen skb->data_len -= shiftlen; 2206832d11c5SIlpo Järvinen skb->truesize -= shiftlen; 2207832d11c5SIlpo Järvinen tgt->len += shiftlen; 2208832d11c5SIlpo Järvinen tgt->data_len += shiftlen; 2209832d11c5SIlpo Järvinen tgt->truesize += shiftlen; 2210832d11c5SIlpo Järvinen 2211832d11c5SIlpo Järvinen return shiftlen; 2212832d11c5SIlpo Järvinen } 2213832d11c5SIlpo Järvinen 2214677e90edSThomas Graf /** 2215677e90edSThomas Graf * skb_prepare_seq_read - Prepare a sequential read of skb data 2216677e90edSThomas Graf * @skb: the buffer to read 2217677e90edSThomas Graf * @from: lower offset of data to be read 2218677e90edSThomas Graf * @to: upper offset of data to be read 2219677e90edSThomas Graf * @st: state variable 2220677e90edSThomas Graf * 2221677e90edSThomas Graf * Initializes the specified state variable. Must be called before 2222677e90edSThomas Graf * invoking skb_seq_read() for the first time. 2223677e90edSThomas Graf */ 2224677e90edSThomas Graf void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 2225677e90edSThomas Graf unsigned int to, struct skb_seq_state *st) 2226677e90edSThomas Graf { 2227677e90edSThomas Graf st->lower_offset = from; 2228677e90edSThomas Graf st->upper_offset = to; 2229677e90edSThomas Graf st->root_skb = st->cur_skb = skb; 2230677e90edSThomas Graf st->frag_idx = st->stepped_offset = 0; 2231677e90edSThomas Graf st->frag_data = NULL; 2232677e90edSThomas Graf } 2233b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_prepare_seq_read); 2234677e90edSThomas Graf 2235677e90edSThomas Graf /** 2236677e90edSThomas Graf * skb_seq_read - Sequentially read skb data 2237677e90edSThomas Graf * @consumed: number of bytes consumed by the caller so far 2238677e90edSThomas Graf * @data: destination pointer for data to be returned 2239677e90edSThomas Graf * @st: state variable 2240677e90edSThomas Graf * 2241677e90edSThomas Graf * Reads a block of skb data at &consumed relative to the 2242677e90edSThomas Graf * lower offset specified to skb_prepare_seq_read(). Assigns 2243677e90edSThomas Graf * the head of the data block to &data and returns the length 2244677e90edSThomas Graf * of the block or 0 if the end of the skb data or the upper 2245677e90edSThomas Graf * offset has been reached. 2246677e90edSThomas Graf * 2247677e90edSThomas Graf * The caller is not required to consume all of the data 2248677e90edSThomas Graf * returned, i.e. &consumed is typically set to the number 2249677e90edSThomas Graf * of bytes already consumed and the next call to 2250677e90edSThomas Graf * skb_seq_read() will return the remaining part of the block. 2251677e90edSThomas Graf * 2252bc2cda1eSRandy Dunlap * Note 1: The size of each block of data returned can be arbitary, 2253677e90edSThomas Graf * this limitation is the cost for zerocopy seqeuental 2254677e90edSThomas Graf * reads of potentially non linear data. 2255677e90edSThomas Graf * 2256bc2cda1eSRandy Dunlap * Note 2: Fragment lists within fragments are not implemented 2257677e90edSThomas Graf * at the moment, state->root_skb could be replaced with 2258677e90edSThomas Graf * a stack for this purpose. 2259677e90edSThomas Graf */ 2260677e90edSThomas Graf unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 2261677e90edSThomas Graf struct skb_seq_state *st) 2262677e90edSThomas Graf { 2263677e90edSThomas Graf unsigned int block_limit, abs_offset = consumed + st->lower_offset; 2264677e90edSThomas Graf skb_frag_t *frag; 2265677e90edSThomas Graf 2266677e90edSThomas Graf if (unlikely(abs_offset >= st->upper_offset)) 2267677e90edSThomas Graf return 0; 2268677e90edSThomas Graf 2269677e90edSThomas Graf next_skb: 227095e3b24cSHerbert Xu block_limit = skb_headlen(st->cur_skb) + st->stepped_offset; 2271677e90edSThomas Graf 2272677e90edSThomas Graf if (abs_offset < block_limit) { 227395e3b24cSHerbert Xu *data = st->cur_skb->data + (abs_offset - st->stepped_offset); 2274677e90edSThomas Graf return block_limit - abs_offset; 2275677e90edSThomas Graf } 2276677e90edSThomas Graf 2277677e90edSThomas Graf if (st->frag_idx == 0 && !st->frag_data) 2278677e90edSThomas Graf st->stepped_offset += skb_headlen(st->cur_skb); 2279677e90edSThomas Graf 2280677e90edSThomas Graf while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) { 2281677e90edSThomas Graf frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx]; 2282677e90edSThomas Graf block_limit = frag->size + st->stepped_offset; 2283677e90edSThomas Graf 2284677e90edSThomas Graf if (abs_offset < block_limit) { 2285677e90edSThomas Graf if (!st->frag_data) 2286677e90edSThomas Graf st->frag_data = kmap_skb_frag(frag); 2287677e90edSThomas Graf 2288677e90edSThomas Graf *data = (u8 *) st->frag_data + frag->page_offset + 2289677e90edSThomas Graf (abs_offset - st->stepped_offset); 2290677e90edSThomas Graf 2291677e90edSThomas Graf return block_limit - abs_offset; 2292677e90edSThomas Graf } 2293677e90edSThomas Graf 2294677e90edSThomas Graf if (st->frag_data) { 2295677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2296677e90edSThomas Graf st->frag_data = NULL; 2297677e90edSThomas Graf } 2298677e90edSThomas Graf 2299677e90edSThomas Graf st->frag_idx++; 2300677e90edSThomas Graf st->stepped_offset += frag->size; 2301677e90edSThomas Graf } 2302677e90edSThomas Graf 23035b5a60daSOlaf Kirch if (st->frag_data) { 23045b5a60daSOlaf Kirch kunmap_skb_frag(st->frag_data); 23055b5a60daSOlaf Kirch st->frag_data = NULL; 23065b5a60daSOlaf Kirch } 23075b5a60daSOlaf Kirch 230871b3346dSShyam Iyer if (st->root_skb == st->cur_skb && 2309677e90edSThomas Graf skb_shinfo(st->root_skb)->frag_list) { 2310677e90edSThomas Graf st->cur_skb = skb_shinfo(st->root_skb)->frag_list; 231195e3b24cSHerbert Xu st->frag_idx = 0; 2312677e90edSThomas Graf goto next_skb; 231371b3346dSShyam Iyer } else if (st->cur_skb->next) { 231471b3346dSShyam Iyer st->cur_skb = st->cur_skb->next; 231571b3346dSShyam Iyer st->frag_idx = 0; 2316677e90edSThomas Graf goto next_skb; 2317677e90edSThomas Graf } 2318677e90edSThomas Graf 2319677e90edSThomas Graf return 0; 2320677e90edSThomas Graf } 2321b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_seq_read); 2322677e90edSThomas Graf 2323677e90edSThomas Graf /** 2324677e90edSThomas Graf * skb_abort_seq_read - Abort a sequential read of skb data 2325677e90edSThomas Graf * @st: state variable 2326677e90edSThomas Graf * 2327677e90edSThomas Graf * Must be called if skb_seq_read() was not called until it 2328677e90edSThomas Graf * returned 0. 2329677e90edSThomas Graf */ 2330677e90edSThomas Graf void skb_abort_seq_read(struct skb_seq_state *st) 2331677e90edSThomas Graf { 2332677e90edSThomas Graf if (st->frag_data) 2333677e90edSThomas Graf kunmap_skb_frag(st->frag_data); 2334677e90edSThomas Graf } 2335b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_abort_seq_read); 2336677e90edSThomas Graf 23373fc7e8a6SThomas Graf #define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb)) 23383fc7e8a6SThomas Graf 23393fc7e8a6SThomas Graf static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text, 23403fc7e8a6SThomas Graf struct ts_config *conf, 23413fc7e8a6SThomas Graf struct ts_state *state) 23423fc7e8a6SThomas Graf { 23433fc7e8a6SThomas Graf return skb_seq_read(offset, text, TS_SKB_CB(state)); 23443fc7e8a6SThomas Graf } 23453fc7e8a6SThomas Graf 23463fc7e8a6SThomas Graf static void skb_ts_finish(struct ts_config *conf, struct ts_state *state) 23473fc7e8a6SThomas Graf { 23483fc7e8a6SThomas Graf skb_abort_seq_read(TS_SKB_CB(state)); 23493fc7e8a6SThomas Graf } 23503fc7e8a6SThomas Graf 23513fc7e8a6SThomas Graf /** 23523fc7e8a6SThomas Graf * skb_find_text - Find a text pattern in skb data 23533fc7e8a6SThomas Graf * @skb: the buffer to look in 23543fc7e8a6SThomas Graf * @from: search offset 23553fc7e8a6SThomas Graf * @to: search limit 23563fc7e8a6SThomas Graf * @config: textsearch configuration 23573fc7e8a6SThomas Graf * @state: uninitialized textsearch state variable 23583fc7e8a6SThomas Graf * 23593fc7e8a6SThomas Graf * Finds a pattern in the skb data according to the specified 23603fc7e8a6SThomas Graf * textsearch configuration. Use textsearch_next() to retrieve 23613fc7e8a6SThomas Graf * subsequent occurrences of the pattern. Returns the offset 23623fc7e8a6SThomas Graf * to the first occurrence or UINT_MAX if no match was found. 23633fc7e8a6SThomas Graf */ 23643fc7e8a6SThomas Graf unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 23653fc7e8a6SThomas Graf unsigned int to, struct ts_config *config, 23663fc7e8a6SThomas Graf struct ts_state *state) 23673fc7e8a6SThomas Graf { 2368f72b948dSPhil Oester unsigned int ret; 2369f72b948dSPhil Oester 23703fc7e8a6SThomas Graf config->get_next_block = skb_ts_get_next_block; 23713fc7e8a6SThomas Graf config->finish = skb_ts_finish; 23723fc7e8a6SThomas Graf 23733fc7e8a6SThomas Graf skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state)); 23743fc7e8a6SThomas Graf 2375f72b948dSPhil Oester ret = textsearch_find(config, state); 2376f72b948dSPhil Oester return (ret <= to - from ? ret : UINT_MAX); 23773fc7e8a6SThomas Graf } 2378b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_find_text); 23793fc7e8a6SThomas Graf 2380e89e9cf5SAnanda Raju /** 2381e89e9cf5SAnanda Raju * skb_append_datato_frags: - append the user data to a skb 2382e89e9cf5SAnanda Raju * @sk: sock structure 2383e89e9cf5SAnanda Raju * @skb: skb structure to be appened with user data. 2384e89e9cf5SAnanda Raju * @getfrag: call back function to be used for getting the user data 2385e89e9cf5SAnanda Raju * @from: pointer to user message iov 2386e89e9cf5SAnanda Raju * @length: length of the iov message 2387e89e9cf5SAnanda Raju * 2388e89e9cf5SAnanda Raju * Description: This procedure append the user data in the fragment part 2389e89e9cf5SAnanda Raju * of the skb if any page alloc fails user this procedure returns -ENOMEM 2390e89e9cf5SAnanda Raju */ 2391e89e9cf5SAnanda Raju int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 2392dab9630fSMartin Waitz int (*getfrag)(void *from, char *to, int offset, 2393e89e9cf5SAnanda Raju int len, int odd, struct sk_buff *skb), 2394e89e9cf5SAnanda Raju void *from, int length) 2395e89e9cf5SAnanda Raju { 2396e89e9cf5SAnanda Raju int frg_cnt = 0; 2397e89e9cf5SAnanda Raju skb_frag_t *frag = NULL; 2398e89e9cf5SAnanda Raju struct page *page = NULL; 2399e89e9cf5SAnanda Raju int copy, left; 2400e89e9cf5SAnanda Raju int offset = 0; 2401e89e9cf5SAnanda Raju int ret; 2402e89e9cf5SAnanda Raju 2403e89e9cf5SAnanda Raju do { 2404e89e9cf5SAnanda Raju /* Return error if we don't have space for new frag */ 2405e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2406e89e9cf5SAnanda Raju if (frg_cnt >= MAX_SKB_FRAGS) 2407e89e9cf5SAnanda Raju return -EFAULT; 2408e89e9cf5SAnanda Raju 2409e89e9cf5SAnanda Raju /* allocate a new page for next frag */ 2410e89e9cf5SAnanda Raju page = alloc_pages(sk->sk_allocation, 0); 2411e89e9cf5SAnanda Raju 2412e89e9cf5SAnanda Raju /* If alloc_page fails just return failure and caller will 2413e89e9cf5SAnanda Raju * free previous allocated pages by doing kfree_skb() 2414e89e9cf5SAnanda Raju */ 2415e89e9cf5SAnanda Raju if (page == NULL) 2416e89e9cf5SAnanda Raju return -ENOMEM; 2417e89e9cf5SAnanda Raju 2418e89e9cf5SAnanda Raju /* initialize the next frag */ 2419e89e9cf5SAnanda Raju sk->sk_sndmsg_page = page; 2420e89e9cf5SAnanda Raju sk->sk_sndmsg_off = 0; 2421e89e9cf5SAnanda Raju skb_fill_page_desc(skb, frg_cnt, page, 0, 0); 2422e89e9cf5SAnanda Raju skb->truesize += PAGE_SIZE; 2423e89e9cf5SAnanda Raju atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc); 2424e89e9cf5SAnanda Raju 2425e89e9cf5SAnanda Raju /* get the new initialized frag */ 2426e89e9cf5SAnanda Raju frg_cnt = skb_shinfo(skb)->nr_frags; 2427e89e9cf5SAnanda Raju frag = &skb_shinfo(skb)->frags[frg_cnt - 1]; 2428e89e9cf5SAnanda Raju 2429e89e9cf5SAnanda Raju /* copy the user data to page */ 2430e89e9cf5SAnanda Raju left = PAGE_SIZE - frag->page_offset; 2431e89e9cf5SAnanda Raju copy = (length > left)? left : length; 2432e89e9cf5SAnanda Raju 2433e89e9cf5SAnanda Raju ret = getfrag(from, (page_address(frag->page) + 2434e89e9cf5SAnanda Raju frag->page_offset + frag->size), 2435e89e9cf5SAnanda Raju offset, copy, 0, skb); 2436e89e9cf5SAnanda Raju if (ret < 0) 2437e89e9cf5SAnanda Raju return -EFAULT; 2438e89e9cf5SAnanda Raju 2439e89e9cf5SAnanda Raju /* copy was successful so update the size parameters */ 2440e89e9cf5SAnanda Raju sk->sk_sndmsg_off += copy; 2441e89e9cf5SAnanda Raju frag->size += copy; 2442e89e9cf5SAnanda Raju skb->len += copy; 2443e89e9cf5SAnanda Raju skb->data_len += copy; 2444e89e9cf5SAnanda Raju offset += copy; 2445e89e9cf5SAnanda Raju length -= copy; 2446e89e9cf5SAnanda Raju 2447e89e9cf5SAnanda Raju } while (length > 0); 2448e89e9cf5SAnanda Raju 2449e89e9cf5SAnanda Raju return 0; 2450e89e9cf5SAnanda Raju } 2451b4ac530fSDavid S. Miller EXPORT_SYMBOL(skb_append_datato_frags); 2452e89e9cf5SAnanda Raju 2453cbb042f9SHerbert Xu /** 2454cbb042f9SHerbert Xu * skb_pull_rcsum - pull skb and update receive checksum 2455cbb042f9SHerbert Xu * @skb: buffer to update 2456cbb042f9SHerbert Xu * @len: length of data pulled 2457cbb042f9SHerbert Xu * 2458cbb042f9SHerbert Xu * This function performs an skb_pull on the packet and updates 2459fee54fa5SUrs Thuermann * the CHECKSUM_COMPLETE checksum. It should be used on 246084fa7933SPatrick McHardy * receive path processing instead of skb_pull unless you know 246184fa7933SPatrick McHardy * that the checksum difference is zero (e.g., a valid IP header) 246284fa7933SPatrick McHardy * or you are setting ip_summed to CHECKSUM_NONE. 2463cbb042f9SHerbert Xu */ 2464cbb042f9SHerbert Xu unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len) 2465cbb042f9SHerbert Xu { 2466cbb042f9SHerbert Xu BUG_ON(len > skb->len); 2467cbb042f9SHerbert Xu skb->len -= len; 2468cbb042f9SHerbert Xu BUG_ON(skb->len < skb->data_len); 2469cbb042f9SHerbert Xu skb_postpull_rcsum(skb, skb->data, len); 2470cbb042f9SHerbert Xu return skb->data += len; 2471cbb042f9SHerbert Xu } 2472cbb042f9SHerbert Xu 2473f94691acSArnaldo Carvalho de Melo EXPORT_SYMBOL_GPL(skb_pull_rcsum); 2474f94691acSArnaldo Carvalho de Melo 2475f4c50d99SHerbert Xu /** 2476f4c50d99SHerbert Xu * skb_segment - Perform protocol segmentation on skb. 2477f4c50d99SHerbert Xu * @skb: buffer to segment 2478576a30ebSHerbert Xu * @features: features for the output path (see dev->features) 2479f4c50d99SHerbert Xu * 2480f4c50d99SHerbert Xu * This function performs segmentation on the given skb. It returns 24814c821d75SBen Hutchings * a pointer to the first in a list of new skbs for the segments. 24824c821d75SBen Hutchings * In case of error it returns ERR_PTR(err). 2483f4c50d99SHerbert Xu */ 2484576a30ebSHerbert Xu struct sk_buff *skb_segment(struct sk_buff *skb, int features) 2485f4c50d99SHerbert Xu { 2486f4c50d99SHerbert Xu struct sk_buff *segs = NULL; 2487f4c50d99SHerbert Xu struct sk_buff *tail = NULL; 248889319d38SHerbert Xu struct sk_buff *fskb = skb_shinfo(skb)->frag_list; 2489f4c50d99SHerbert Xu unsigned int mss = skb_shinfo(skb)->gso_size; 249098e399f8SArnaldo Carvalho de Melo unsigned int doffset = skb->data - skb_mac_header(skb); 2491f4c50d99SHerbert Xu unsigned int offset = doffset; 2492f4c50d99SHerbert Xu unsigned int headroom; 2493f4c50d99SHerbert Xu unsigned int len; 2494576a30ebSHerbert Xu int sg = features & NETIF_F_SG; 2495f4c50d99SHerbert Xu int nfrags = skb_shinfo(skb)->nr_frags; 2496f4c50d99SHerbert Xu int err = -ENOMEM; 2497f4c50d99SHerbert Xu int i = 0; 2498f4c50d99SHerbert Xu int pos; 2499f4c50d99SHerbert Xu 2500f4c50d99SHerbert Xu __skb_push(skb, doffset); 2501f4c50d99SHerbert Xu headroom = skb_headroom(skb); 2502f4c50d99SHerbert Xu pos = skb_headlen(skb); 2503f4c50d99SHerbert Xu 2504f4c50d99SHerbert Xu do { 2505f4c50d99SHerbert Xu struct sk_buff *nskb; 2506f4c50d99SHerbert Xu skb_frag_t *frag; 2507c8884eddSHerbert Xu int hsize; 2508f4c50d99SHerbert Xu int size; 2509f4c50d99SHerbert Xu 2510f4c50d99SHerbert Xu len = skb->len - offset; 2511f4c50d99SHerbert Xu if (len > mss) 2512f4c50d99SHerbert Xu len = mss; 2513f4c50d99SHerbert Xu 2514f4c50d99SHerbert Xu hsize = skb_headlen(skb) - offset; 2515f4c50d99SHerbert Xu if (hsize < 0) 2516f4c50d99SHerbert Xu hsize = 0; 2517c8884eddSHerbert Xu if (hsize > len || !sg) 2518c8884eddSHerbert Xu hsize = len; 2519f4c50d99SHerbert Xu 252089319d38SHerbert Xu if (!hsize && i >= nfrags) { 252189319d38SHerbert Xu BUG_ON(fskb->len != len); 252289319d38SHerbert Xu 252389319d38SHerbert Xu pos += len; 252489319d38SHerbert Xu nskb = skb_clone(fskb, GFP_ATOMIC); 252589319d38SHerbert Xu fskb = fskb->next; 252689319d38SHerbert Xu 2527f4c50d99SHerbert Xu if (unlikely(!nskb)) 2528f4c50d99SHerbert Xu goto err; 2529f4c50d99SHerbert Xu 253089319d38SHerbert Xu hsize = skb_end_pointer(nskb) - nskb->head; 253189319d38SHerbert Xu if (skb_cow_head(nskb, doffset + headroom)) { 253289319d38SHerbert Xu kfree_skb(nskb); 253389319d38SHerbert Xu goto err; 253489319d38SHerbert Xu } 253589319d38SHerbert Xu 253689319d38SHerbert Xu nskb->truesize += skb_end_pointer(nskb) - nskb->head - 253789319d38SHerbert Xu hsize; 253889319d38SHerbert Xu skb_release_head_state(nskb); 253989319d38SHerbert Xu __skb_push(nskb, doffset); 254089319d38SHerbert Xu } else { 254189319d38SHerbert Xu nskb = alloc_skb(hsize + doffset + headroom, 254289319d38SHerbert Xu GFP_ATOMIC); 254389319d38SHerbert Xu 254489319d38SHerbert Xu if (unlikely(!nskb)) 254589319d38SHerbert Xu goto err; 254689319d38SHerbert Xu 254789319d38SHerbert Xu skb_reserve(nskb, headroom); 254889319d38SHerbert Xu __skb_put(nskb, doffset); 254989319d38SHerbert Xu } 255089319d38SHerbert Xu 2551f4c50d99SHerbert Xu if (segs) 2552f4c50d99SHerbert Xu tail->next = nskb; 2553f4c50d99SHerbert Xu else 2554f4c50d99SHerbert Xu segs = nskb; 2555f4c50d99SHerbert Xu tail = nskb; 2556f4c50d99SHerbert Xu 25576f85a124SHerbert Xu __copy_skb_header(nskb, skb); 2558f4c50d99SHerbert Xu nskb->mac_len = skb->mac_len; 2559f4c50d99SHerbert Xu 2560459a98edSArnaldo Carvalho de Melo skb_reset_mac_header(nskb); 2561ddc7b8e3SArnaldo Carvalho de Melo skb_set_network_header(nskb, skb->mac_len); 2562b0e380b1SArnaldo Carvalho de Melo nskb->transport_header = (nskb->network_header + 2563b0e380b1SArnaldo Carvalho de Melo skb_network_header_len(skb)); 256489319d38SHerbert Xu skb_copy_from_linear_data(skb, nskb->data, doffset); 256589319d38SHerbert Xu 256689319d38SHerbert Xu if (pos >= offset + len) 256789319d38SHerbert Xu continue; 256889319d38SHerbert Xu 2569f4c50d99SHerbert Xu if (!sg) { 25706f85a124SHerbert Xu nskb->ip_summed = CHECKSUM_NONE; 2571f4c50d99SHerbert Xu nskb->csum = skb_copy_and_csum_bits(skb, offset, 2572f4c50d99SHerbert Xu skb_put(nskb, len), 2573f4c50d99SHerbert Xu len, 0); 2574f4c50d99SHerbert Xu continue; 2575f4c50d99SHerbert Xu } 2576f4c50d99SHerbert Xu 2577f4c50d99SHerbert Xu frag = skb_shinfo(nskb)->frags; 2578f4c50d99SHerbert Xu 2579d626f62bSArnaldo Carvalho de Melo skb_copy_from_linear_data_offset(skb, offset, 2580d626f62bSArnaldo Carvalho de Melo skb_put(nskb, hsize), hsize); 2581f4c50d99SHerbert Xu 258289319d38SHerbert Xu while (pos < offset + len && i < nfrags) { 2583f4c50d99SHerbert Xu *frag = skb_shinfo(skb)->frags[i]; 2584f4c50d99SHerbert Xu get_page(frag->page); 2585f4c50d99SHerbert Xu size = frag->size; 2586f4c50d99SHerbert Xu 2587f4c50d99SHerbert Xu if (pos < offset) { 2588f4c50d99SHerbert Xu frag->page_offset += offset - pos; 2589f4c50d99SHerbert Xu frag->size -= offset - pos; 2590f4c50d99SHerbert Xu } 2591f4c50d99SHerbert Xu 259289319d38SHerbert Xu skb_shinfo(nskb)->nr_frags++; 2593f4c50d99SHerbert Xu 2594f4c50d99SHerbert Xu if (pos + size <= offset + len) { 2595f4c50d99SHerbert Xu i++; 2596f4c50d99SHerbert Xu pos += size; 2597f4c50d99SHerbert Xu } else { 2598f4c50d99SHerbert Xu frag->size -= pos + size - (offset + len); 259989319d38SHerbert Xu goto skip_fraglist; 2600f4c50d99SHerbert Xu } 2601f4c50d99SHerbert Xu 2602f4c50d99SHerbert Xu frag++; 2603f4c50d99SHerbert Xu } 2604f4c50d99SHerbert Xu 260589319d38SHerbert Xu if (pos < offset + len) { 260689319d38SHerbert Xu struct sk_buff *fskb2 = fskb; 260789319d38SHerbert Xu 260889319d38SHerbert Xu BUG_ON(pos + fskb->len != offset + len); 260989319d38SHerbert Xu 261089319d38SHerbert Xu pos += fskb->len; 261189319d38SHerbert Xu fskb = fskb->next; 261289319d38SHerbert Xu 261389319d38SHerbert Xu if (fskb2->next) { 261489319d38SHerbert Xu fskb2 = skb_clone(fskb2, GFP_ATOMIC); 261589319d38SHerbert Xu if (!fskb2) 261689319d38SHerbert Xu goto err; 261789319d38SHerbert Xu } else 261889319d38SHerbert Xu skb_get(fskb2); 261989319d38SHerbert Xu 262089319d38SHerbert Xu BUG_ON(skb_shinfo(nskb)->frag_list); 262189319d38SHerbert Xu skb_shinfo(nskb)->frag_list = fskb2; 262289319d38SHerbert Xu } 262389319d38SHerbert Xu 262489319d38SHerbert Xu skip_fraglist: 2625f4c50d99SHerbert Xu nskb->data_len = len - hsize; 2626f4c50d99SHerbert Xu nskb->len += nskb->data_len; 2627f4c50d99SHerbert Xu nskb->truesize += nskb->data_len; 2628f4c50d99SHerbert Xu } while ((offset += len) < skb->len); 2629f4c50d99SHerbert Xu 2630f4c50d99SHerbert Xu return segs; 2631f4c50d99SHerbert Xu 2632f4c50d99SHerbert Xu err: 2633f4c50d99SHerbert Xu while ((skb = segs)) { 2634f4c50d99SHerbert Xu segs = skb->next; 2635b08d5840SPatrick McHardy kfree_skb(skb); 2636f4c50d99SHerbert Xu } 2637f4c50d99SHerbert Xu return ERR_PTR(err); 2638f4c50d99SHerbert Xu } 2639f4c50d99SHerbert Xu EXPORT_SYMBOL_GPL(skb_segment); 2640f4c50d99SHerbert Xu 264171d93b39SHerbert Xu int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb) 264271d93b39SHerbert Xu { 264371d93b39SHerbert Xu struct sk_buff *p = *head; 264471d93b39SHerbert Xu struct sk_buff *nskb; 264571d93b39SHerbert Xu unsigned int headroom; 264686911732SHerbert Xu unsigned int len = skb_gro_len(skb); 264771d93b39SHerbert Xu 264886911732SHerbert Xu if (p->len + len >= 65536) 264971d93b39SHerbert Xu return -E2BIG; 265071d93b39SHerbert Xu 265171d93b39SHerbert Xu if (skb_shinfo(p)->frag_list) 265271d93b39SHerbert Xu goto merge; 265381705ad1SHerbert Xu else if (skb_headlen(skb) <= skb_gro_offset(skb)) { 265481705ad1SHerbert Xu if (skb_shinfo(p)->nr_frags + skb_shinfo(skb)->nr_frags > 265581705ad1SHerbert Xu MAX_SKB_FRAGS) 265681705ad1SHerbert Xu return -E2BIG; 265781705ad1SHerbert Xu 265886911732SHerbert Xu skb_shinfo(skb)->frags[0].page_offset += 265986911732SHerbert Xu skb_gro_offset(skb) - skb_headlen(skb); 266086911732SHerbert Xu skb_shinfo(skb)->frags[0].size -= 266186911732SHerbert Xu skb_gro_offset(skb) - skb_headlen(skb); 266286911732SHerbert Xu 26635d38a079SHerbert Xu memcpy(skb_shinfo(p)->frags + skb_shinfo(p)->nr_frags, 26645d38a079SHerbert Xu skb_shinfo(skb)->frags, 26655d38a079SHerbert Xu skb_shinfo(skb)->nr_frags * sizeof(skb_frag_t)); 26665d38a079SHerbert Xu 26675d38a079SHerbert Xu skb_shinfo(p)->nr_frags += skb_shinfo(skb)->nr_frags; 2668f5572068SHerbert Xu skb_shinfo(skb)->nr_frags = 0; 2669f5572068SHerbert Xu 2670f5572068SHerbert Xu skb->truesize -= skb->data_len; 2671f5572068SHerbert Xu skb->len -= skb->data_len; 2672f5572068SHerbert Xu skb->data_len = 0; 2673f5572068SHerbert Xu 26745d38a079SHerbert Xu NAPI_GRO_CB(skb)->free = 1; 26755d38a079SHerbert Xu goto done; 26765d38a079SHerbert Xu } 267771d93b39SHerbert Xu 267871d93b39SHerbert Xu headroom = skb_headroom(p); 267986911732SHerbert Xu nskb = netdev_alloc_skb(p->dev, headroom + skb_gro_offset(p)); 268071d93b39SHerbert Xu if (unlikely(!nskb)) 268171d93b39SHerbert Xu return -ENOMEM; 268271d93b39SHerbert Xu 268371d93b39SHerbert Xu __copy_skb_header(nskb, p); 268471d93b39SHerbert Xu nskb->mac_len = p->mac_len; 268571d93b39SHerbert Xu 268671d93b39SHerbert Xu skb_reserve(nskb, headroom); 268786911732SHerbert Xu __skb_put(nskb, skb_gro_offset(p)); 268871d93b39SHerbert Xu 268986911732SHerbert Xu skb_set_mac_header(nskb, skb_mac_header(p) - p->data); 269071d93b39SHerbert Xu skb_set_network_header(nskb, skb_network_offset(p)); 269171d93b39SHerbert Xu skb_set_transport_header(nskb, skb_transport_offset(p)); 269271d93b39SHerbert Xu 269386911732SHerbert Xu __skb_pull(p, skb_gro_offset(p)); 269486911732SHerbert Xu memcpy(skb_mac_header(nskb), skb_mac_header(p), 269586911732SHerbert Xu p->data - skb_mac_header(p)); 269671d93b39SHerbert Xu 269771d93b39SHerbert Xu *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p); 269871d93b39SHerbert Xu skb_shinfo(nskb)->frag_list = p; 2699b530256dSHerbert Xu skb_shinfo(nskb)->gso_size = skb_shinfo(p)->gso_size; 270071d93b39SHerbert Xu skb_header_release(p); 270171d93b39SHerbert Xu nskb->prev = p; 270271d93b39SHerbert Xu 270371d93b39SHerbert Xu nskb->data_len += p->len; 270471d93b39SHerbert Xu nskb->truesize += p->len; 270571d93b39SHerbert Xu nskb->len += p->len; 270671d93b39SHerbert Xu 270771d93b39SHerbert Xu *head = nskb; 270871d93b39SHerbert Xu nskb->next = p->next; 270971d93b39SHerbert Xu p->next = NULL; 271071d93b39SHerbert Xu 271171d93b39SHerbert Xu p = nskb; 271271d93b39SHerbert Xu 271371d93b39SHerbert Xu merge: 271456035022SHerbert Xu if (skb_gro_offset(skb) > skb_headlen(skb)) { 271556035022SHerbert Xu skb_shinfo(skb)->frags[0].page_offset += 271656035022SHerbert Xu skb_gro_offset(skb) - skb_headlen(skb); 271756035022SHerbert Xu skb_shinfo(skb)->frags[0].size -= 271856035022SHerbert Xu skb_gro_offset(skb) - skb_headlen(skb); 271956035022SHerbert Xu skb_gro_reset_offset(skb); 272056035022SHerbert Xu skb_gro_pull(skb, skb_headlen(skb)); 272156035022SHerbert Xu } 272256035022SHerbert Xu 272356035022SHerbert Xu __skb_pull(skb, skb_gro_offset(skb)); 272456035022SHerbert Xu 272571d93b39SHerbert Xu p->prev->next = skb; 272671d93b39SHerbert Xu p->prev = skb; 272771d93b39SHerbert Xu skb_header_release(skb); 272871d93b39SHerbert Xu 27295d38a079SHerbert Xu done: 27305d38a079SHerbert Xu NAPI_GRO_CB(p)->count++; 273137fe4732SHerbert Xu p->data_len += len; 273237fe4732SHerbert Xu p->truesize += len; 273337fe4732SHerbert Xu p->len += len; 273471d93b39SHerbert Xu 273571d93b39SHerbert Xu NAPI_GRO_CB(skb)->same_flow = 1; 273671d93b39SHerbert Xu return 0; 273771d93b39SHerbert Xu } 273871d93b39SHerbert Xu EXPORT_SYMBOL_GPL(skb_gro_receive); 273971d93b39SHerbert Xu 27401da177e4SLinus Torvalds void __init skb_init(void) 27411da177e4SLinus Torvalds { 27421da177e4SLinus Torvalds skbuff_head_cache = kmem_cache_create("skbuff_head_cache", 27431da177e4SLinus Torvalds sizeof(struct sk_buff), 27441da177e4SLinus Torvalds 0, 2745e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 274620c2df83SPaul Mundt NULL); 2747d179cd12SDavid S. Miller skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache", 2748d179cd12SDavid S. Miller (2*sizeof(struct sk_buff)) + 2749d179cd12SDavid S. Miller sizeof(atomic_t), 2750d179cd12SDavid S. Miller 0, 2751e5d679f3SAlexey Dobriyan SLAB_HWCACHE_ALIGN|SLAB_PANIC, 275220c2df83SPaul Mundt NULL); 27531da177e4SLinus Torvalds } 27541da177e4SLinus Torvalds 2755716ea3a7SDavid Howells /** 2756716ea3a7SDavid Howells * skb_to_sgvec - Fill a scatter-gather list from a socket buffer 2757716ea3a7SDavid Howells * @skb: Socket buffer containing the buffers to be mapped 2758716ea3a7SDavid Howells * @sg: The scatter-gather list to map into 2759716ea3a7SDavid Howells * @offset: The offset into the buffer's contents to start mapping 2760716ea3a7SDavid Howells * @len: Length of buffer space to be mapped 2761716ea3a7SDavid Howells * 2762716ea3a7SDavid Howells * Fill the specified scatter-gather list with mappings/pointers into a 2763716ea3a7SDavid Howells * region of the buffer space attached to a socket buffer. 2764716ea3a7SDavid Howells */ 276551c739d1SDavid S. Miller static int 276651c739d1SDavid S. Miller __skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 2767716ea3a7SDavid Howells { 27681a028e50SDavid S. Miller int start = skb_headlen(skb); 27691a028e50SDavid S. Miller int i, copy = start - offset; 2770716ea3a7SDavid Howells int elt = 0; 2771716ea3a7SDavid Howells 2772716ea3a7SDavid Howells if (copy > 0) { 2773716ea3a7SDavid Howells if (copy > len) 2774716ea3a7SDavid Howells copy = len; 2775642f1490SJens Axboe sg_set_buf(sg, skb->data + offset, copy); 2776716ea3a7SDavid Howells elt++; 2777716ea3a7SDavid Howells if ((len -= copy) == 0) 2778716ea3a7SDavid Howells return elt; 2779716ea3a7SDavid Howells offset += copy; 2780716ea3a7SDavid Howells } 2781716ea3a7SDavid Howells 2782716ea3a7SDavid Howells for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) { 27831a028e50SDavid S. Miller int end; 2784716ea3a7SDavid Howells 2785547b792cSIlpo Järvinen WARN_ON(start > offset + len); 27861a028e50SDavid S. Miller 27871a028e50SDavid S. Miller end = start + skb_shinfo(skb)->frags[i].size; 2788716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2789716ea3a7SDavid Howells skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2790716ea3a7SDavid Howells 2791716ea3a7SDavid Howells if (copy > len) 2792716ea3a7SDavid Howells copy = len; 2793642f1490SJens Axboe sg_set_page(&sg[elt], frag->page, copy, 2794642f1490SJens Axboe frag->page_offset+offset-start); 2795716ea3a7SDavid Howells elt++; 2796716ea3a7SDavid Howells if (!(len -= copy)) 2797716ea3a7SDavid Howells return elt; 2798716ea3a7SDavid Howells offset += copy; 2799716ea3a7SDavid Howells } 28001a028e50SDavid S. Miller start = end; 2801716ea3a7SDavid Howells } 2802716ea3a7SDavid Howells 2803716ea3a7SDavid Howells if (skb_shinfo(skb)->frag_list) { 2804716ea3a7SDavid Howells struct sk_buff *list = skb_shinfo(skb)->frag_list; 2805716ea3a7SDavid Howells 2806716ea3a7SDavid Howells for (; list; list = list->next) { 28071a028e50SDavid S. Miller int end; 2808716ea3a7SDavid Howells 2809547b792cSIlpo Järvinen WARN_ON(start > offset + len); 28101a028e50SDavid S. Miller 28111a028e50SDavid S. Miller end = start + list->len; 2812716ea3a7SDavid Howells if ((copy = end - offset) > 0) { 2813716ea3a7SDavid Howells if (copy > len) 2814716ea3a7SDavid Howells copy = len; 281551c739d1SDavid S. Miller elt += __skb_to_sgvec(list, sg+elt, offset - start, 281651c739d1SDavid S. Miller copy); 2817716ea3a7SDavid Howells if ((len -= copy) == 0) 2818716ea3a7SDavid Howells return elt; 2819716ea3a7SDavid Howells offset += copy; 2820716ea3a7SDavid Howells } 28211a028e50SDavid S. Miller start = end; 2822716ea3a7SDavid Howells } 2823716ea3a7SDavid Howells } 2824716ea3a7SDavid Howells BUG_ON(len); 2825716ea3a7SDavid Howells return elt; 2826716ea3a7SDavid Howells } 2827716ea3a7SDavid Howells 282851c739d1SDavid S. Miller int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len) 282951c739d1SDavid S. Miller { 283051c739d1SDavid S. Miller int nsg = __skb_to_sgvec(skb, sg, offset, len); 283151c739d1SDavid S. Miller 2832c46f2334SJens Axboe sg_mark_end(&sg[nsg - 1]); 283351c739d1SDavid S. Miller 283451c739d1SDavid S. Miller return nsg; 283551c739d1SDavid S. Miller } 2836b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_to_sgvec); 283751c739d1SDavid S. Miller 2838716ea3a7SDavid Howells /** 2839716ea3a7SDavid Howells * skb_cow_data - Check that a socket buffer's data buffers are writable 2840716ea3a7SDavid Howells * @skb: The socket buffer to check. 2841716ea3a7SDavid Howells * @tailbits: Amount of trailing space to be added 2842716ea3a7SDavid Howells * @trailer: Returned pointer to the skb where the @tailbits space begins 2843716ea3a7SDavid Howells * 2844716ea3a7SDavid Howells * Make sure that the data buffers attached to a socket buffer are 2845716ea3a7SDavid Howells * writable. If they are not, private copies are made of the data buffers 2846716ea3a7SDavid Howells * and the socket buffer is set to use these instead. 2847716ea3a7SDavid Howells * 2848716ea3a7SDavid Howells * If @tailbits is given, make sure that there is space to write @tailbits 2849716ea3a7SDavid Howells * bytes of data beyond current end of socket buffer. @trailer will be 2850716ea3a7SDavid Howells * set to point to the skb in which this space begins. 2851716ea3a7SDavid Howells * 2852716ea3a7SDavid Howells * The number of scatterlist elements required to completely map the 2853716ea3a7SDavid Howells * COW'd and extended socket buffer will be returned. 2854716ea3a7SDavid Howells */ 2855716ea3a7SDavid Howells int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer) 2856716ea3a7SDavid Howells { 2857716ea3a7SDavid Howells int copyflag; 2858716ea3a7SDavid Howells int elt; 2859716ea3a7SDavid Howells struct sk_buff *skb1, **skb_p; 2860716ea3a7SDavid Howells 2861716ea3a7SDavid Howells /* If skb is cloned or its head is paged, reallocate 2862716ea3a7SDavid Howells * head pulling out all the pages (pages are considered not writable 2863716ea3a7SDavid Howells * at the moment even if they are anonymous). 2864716ea3a7SDavid Howells */ 2865716ea3a7SDavid Howells if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) && 2866716ea3a7SDavid Howells __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL) 2867716ea3a7SDavid Howells return -ENOMEM; 2868716ea3a7SDavid Howells 2869716ea3a7SDavid Howells /* Easy case. Most of packets will go this way. */ 2870716ea3a7SDavid Howells if (!skb_shinfo(skb)->frag_list) { 2871716ea3a7SDavid Howells /* A little of trouble, not enough of space for trailer. 2872716ea3a7SDavid Howells * This should not happen, when stack is tuned to generate 2873716ea3a7SDavid Howells * good frames. OK, on miss we reallocate and reserve even more 2874716ea3a7SDavid Howells * space, 128 bytes is fair. */ 2875716ea3a7SDavid Howells 2876716ea3a7SDavid Howells if (skb_tailroom(skb) < tailbits && 2877716ea3a7SDavid Howells pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC)) 2878716ea3a7SDavid Howells return -ENOMEM; 2879716ea3a7SDavid Howells 2880716ea3a7SDavid Howells /* Voila! */ 2881716ea3a7SDavid Howells *trailer = skb; 2882716ea3a7SDavid Howells return 1; 2883716ea3a7SDavid Howells } 2884716ea3a7SDavid Howells 2885716ea3a7SDavid Howells /* Misery. We are in troubles, going to mincer fragments... */ 2886716ea3a7SDavid Howells 2887716ea3a7SDavid Howells elt = 1; 2888716ea3a7SDavid Howells skb_p = &skb_shinfo(skb)->frag_list; 2889716ea3a7SDavid Howells copyflag = 0; 2890716ea3a7SDavid Howells 2891716ea3a7SDavid Howells while ((skb1 = *skb_p) != NULL) { 2892716ea3a7SDavid Howells int ntail = 0; 2893716ea3a7SDavid Howells 2894716ea3a7SDavid Howells /* The fragment is partially pulled by someone, 2895716ea3a7SDavid Howells * this can happen on input. Copy it and everything 2896716ea3a7SDavid Howells * after it. */ 2897716ea3a7SDavid Howells 2898716ea3a7SDavid Howells if (skb_shared(skb1)) 2899716ea3a7SDavid Howells copyflag = 1; 2900716ea3a7SDavid Howells 2901716ea3a7SDavid Howells /* If the skb is the last, worry about trailer. */ 2902716ea3a7SDavid Howells 2903716ea3a7SDavid Howells if (skb1->next == NULL && tailbits) { 2904716ea3a7SDavid Howells if (skb_shinfo(skb1)->nr_frags || 2905716ea3a7SDavid Howells skb_shinfo(skb1)->frag_list || 2906716ea3a7SDavid Howells skb_tailroom(skb1) < tailbits) 2907716ea3a7SDavid Howells ntail = tailbits + 128; 2908716ea3a7SDavid Howells } 2909716ea3a7SDavid Howells 2910716ea3a7SDavid Howells if (copyflag || 2911716ea3a7SDavid Howells skb_cloned(skb1) || 2912716ea3a7SDavid Howells ntail || 2913716ea3a7SDavid Howells skb_shinfo(skb1)->nr_frags || 2914716ea3a7SDavid Howells skb_shinfo(skb1)->frag_list) { 2915716ea3a7SDavid Howells struct sk_buff *skb2; 2916716ea3a7SDavid Howells 2917716ea3a7SDavid Howells /* Fuck, we are miserable poor guys... */ 2918716ea3a7SDavid Howells if (ntail == 0) 2919716ea3a7SDavid Howells skb2 = skb_copy(skb1, GFP_ATOMIC); 2920716ea3a7SDavid Howells else 2921716ea3a7SDavid Howells skb2 = skb_copy_expand(skb1, 2922716ea3a7SDavid Howells skb_headroom(skb1), 2923716ea3a7SDavid Howells ntail, 2924716ea3a7SDavid Howells GFP_ATOMIC); 2925716ea3a7SDavid Howells if (unlikely(skb2 == NULL)) 2926716ea3a7SDavid Howells return -ENOMEM; 2927716ea3a7SDavid Howells 2928716ea3a7SDavid Howells if (skb1->sk) 2929716ea3a7SDavid Howells skb_set_owner_w(skb2, skb1->sk); 2930716ea3a7SDavid Howells 2931716ea3a7SDavid Howells /* Looking around. Are we still alive? 2932716ea3a7SDavid Howells * OK, link new skb, drop old one */ 2933716ea3a7SDavid Howells 2934716ea3a7SDavid Howells skb2->next = skb1->next; 2935716ea3a7SDavid Howells *skb_p = skb2; 2936716ea3a7SDavid Howells kfree_skb(skb1); 2937716ea3a7SDavid Howells skb1 = skb2; 2938716ea3a7SDavid Howells } 2939716ea3a7SDavid Howells elt++; 2940716ea3a7SDavid Howells *trailer = skb1; 2941716ea3a7SDavid Howells skb_p = &skb1->next; 2942716ea3a7SDavid Howells } 2943716ea3a7SDavid Howells 2944716ea3a7SDavid Howells return elt; 2945716ea3a7SDavid Howells } 2946b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_cow_data); 2947716ea3a7SDavid Howells 2948f35d9d8aSRusty Russell /** 2949f35d9d8aSRusty Russell * skb_partial_csum_set - set up and verify partial csum values for packet 2950f35d9d8aSRusty Russell * @skb: the skb to set 2951f35d9d8aSRusty Russell * @start: the number of bytes after skb->data to start checksumming. 2952f35d9d8aSRusty Russell * @off: the offset from start to place the checksum. 2953f35d9d8aSRusty Russell * 2954f35d9d8aSRusty Russell * For untrusted partially-checksummed packets, we need to make sure the values 2955f35d9d8aSRusty Russell * for skb->csum_start and skb->csum_offset are valid so we don't oops. 2956f35d9d8aSRusty Russell * 2957f35d9d8aSRusty Russell * This function checks and sets those values and skb->ip_summed: if this 2958f35d9d8aSRusty Russell * returns false you should drop the packet. 2959f35d9d8aSRusty Russell */ 2960f35d9d8aSRusty Russell bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off) 2961f35d9d8aSRusty Russell { 2962f35d9d8aSRusty Russell if (unlikely(start > skb->len - 2) || 2963f35d9d8aSRusty Russell unlikely((int)start + off > skb->len - 2)) { 2964f35d9d8aSRusty Russell if (net_ratelimit()) 2965f35d9d8aSRusty Russell printk(KERN_WARNING 2966f35d9d8aSRusty Russell "bad partial csum: csum=%u/%u len=%u\n", 2967f35d9d8aSRusty Russell start, off, skb->len); 2968f35d9d8aSRusty Russell return false; 2969f35d9d8aSRusty Russell } 2970f35d9d8aSRusty Russell skb->ip_summed = CHECKSUM_PARTIAL; 2971f35d9d8aSRusty Russell skb->csum_start = skb_headroom(skb) + start; 2972f35d9d8aSRusty Russell skb->csum_offset = off; 2973f35d9d8aSRusty Russell return true; 2974f35d9d8aSRusty Russell } 2975b4ac530fSDavid S. Miller EXPORT_SYMBOL_GPL(skb_partial_csum_set); 2976f35d9d8aSRusty Russell 29774497b076SBen Hutchings void __skb_warn_lro_forwarding(const struct sk_buff *skb) 29784497b076SBen Hutchings { 29794497b076SBen Hutchings if (net_ratelimit()) 29804497b076SBen Hutchings pr_warning("%s: received packets cannot be forwarded" 29814497b076SBen Hutchings " while LRO is enabled\n", skb->dev->name); 29824497b076SBen Hutchings } 29834497b076SBen Hutchings EXPORT_SYMBOL(__skb_warn_lro_forwarding); 2984